| /* dwarf.c -- display DWARF contents of a BFD binary file | 
 |    Copyright (C) 2005-2025 Free Software Foundation, Inc. | 
 |  | 
 |    This file is part of GNU Binutils. | 
 |  | 
 |    This program is free software; you can redistribute it and/or modify | 
 |    it under the terms of the GNU General Public License as published by | 
 |    the Free Software Foundation; either version 3 of the License, or | 
 |    (at your option) any later version. | 
 |  | 
 |    This program is distributed in the hope that it will be useful, | 
 |    but WITHOUT ANY WARRANTY; without even the implied warranty of | 
 |    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the | 
 |    GNU General Public License for more details. | 
 |  | 
 |    You should have received a copy of the GNU General Public License | 
 |    along with this program; if not, write to the Free Software | 
 |    Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, MA | 
 |    02110-1301, USA.  */ | 
 |  | 
 | #include "sysdep.h" | 
 | #include "libiberty.h" | 
 | #include "bfd.h" | 
 | #include <stdint.h> | 
 | #include "bucomm.h" | 
 | #include "elfcomm.h" | 
 | #include "elf/common.h" | 
 | #include "dwarf2.h" | 
 | #include "dwarf.h" | 
 | #include "gdb/gdb-index.h" | 
 | #include "filenames.h" | 
 | #include "safe-ctype.h" | 
 | #include "sframe-api.h" | 
 | #include <assert.h> | 
 |  | 
 | #ifdef HAVE_LIBDEBUGINFOD | 
 | #include <elfutils/debuginfod.h> | 
 | #endif | 
 |  | 
 | #include <limits.h> | 
 | #ifndef CHAR_BIT | 
 | #define CHAR_BIT 8 | 
 | #endif | 
 |  | 
 | #ifndef ENABLE_CHECKING | 
 | #define ENABLE_CHECKING 0 | 
 | #endif | 
 |  | 
 | #undef MAX | 
 | #undef MIN | 
 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | 
 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | 
 |  | 
 | static const char *regname (unsigned int regno, int row); | 
 | static const char *regname_internal_by_table_only (unsigned int regno); | 
 |  | 
 | static int have_frame_base; | 
 | static int frame_base_level = -1; /* To support nested DW_TAG_subprogram's.  */ | 
 | static int need_base_address; | 
 |  | 
 | static unsigned int num_debug_info_entries = 0; | 
 | static unsigned int alloc_num_debug_info_entries = 0; | 
 | static debug_info *debug_information = NULL; | 
 | /* Special value for num_debug_info_entries to indicate | 
 |    that the .debug_info section could not be loaded/parsed.  */ | 
 | #define DEBUG_INFO_UNAVAILABLE  (unsigned int) -1 | 
 |  | 
 | /* A .debug_info section can contain multiple links to separate | 
 |    DWO object files.  We use these structures to record these links.  */ | 
 | typedef enum dwo_type | 
 | { | 
 |   DWO_NAME, | 
 |   DWO_DIR, | 
 |   DWO_ID | 
 | } dwo_type; | 
 |  | 
 | typedef struct dwo_info | 
 | { | 
 |   dwo_type          type; | 
 |   const char *      value; | 
 |   uint64_t          cu_offset; | 
 |   struct dwo_info * next; | 
 | } dwo_info; | 
 |  | 
 | static dwo_info *first_dwo_info = NULL; | 
 | static bool need_dwo_info; | 
 |  | 
 | separate_info * first_separate_info = NULL; | 
 |  | 
 | unsigned int eh_addr_size; | 
 |  | 
 | int do_debug_info; | 
 | int do_debug_abbrevs; | 
 | int do_debug_lines; | 
 | int do_debug_pubnames; | 
 | int do_debug_pubtypes; | 
 | int do_debug_aranges; | 
 | int do_debug_ranges; | 
 | int do_debug_frames; | 
 | int do_debug_frames_interp; | 
 | int do_debug_macinfo; | 
 | int do_debug_str; | 
 | int do_debug_str_offsets; | 
 | int do_debug_loc; | 
 | int do_gdb_index; | 
 | int do_sframe; | 
 | int do_trace_info; | 
 | int do_trace_abbrevs; | 
 | int do_trace_aranges; | 
 | int do_debug_addr; | 
 | int do_debug_cu_index; | 
 | int do_wide; | 
 | int do_debug_links; | 
 | int do_follow_links = DEFAULT_FOR_FOLLOW_LINKS; | 
 | #ifdef HAVE_LIBDEBUGINFOD | 
 | int use_debuginfod = 1; | 
 | #endif | 
 | bool do_checks; | 
 |  | 
 | int dwarf_cutoff_level = -1; | 
 | unsigned long dwarf_start_die; | 
 |  | 
 | int dwarf_check = 0; | 
 |  | 
 | /* Collection of CU/TU section sets from .debug_cu_index and .debug_tu_index | 
 |    sections.  For version 1 package files, each set is stored in SHNDX_POOL | 
 |    as a zero-terminated list of section indexes comprising one set of debug | 
 |    sections from a .dwo file.  */ | 
 |  | 
 | static unsigned int *shndx_pool = NULL; | 
 | static unsigned int shndx_pool_size = 0; | 
 | static unsigned int shndx_pool_used = 0; | 
 |  | 
 | /* For version 2 package files, each set contains an array of section offsets | 
 |    and an array of section sizes, giving the offset and size of the | 
 |    contribution from a CU or TU within one of the debug sections. | 
 |    When displaying debug info from a package file, we need to use these | 
 |    tables to locate the corresponding contributions to each section.  */ | 
 |  | 
 | struct cu_tu_set | 
 | { | 
 |   uint64_t signature; | 
 |   uint64_t section_offsets[DW_SECT_MAX]; | 
 |   size_t   section_sizes[DW_SECT_MAX]; | 
 | }; | 
 |  | 
 | static int cu_count = 0; | 
 | static int tu_count = 0; | 
 | static struct cu_tu_set *cu_sets = NULL; | 
 | static struct cu_tu_set *tu_sets = NULL; | 
 |  | 
 | static bool load_cu_tu_indexes (void *); | 
 |  | 
 | /* An array that indicates for a given level of CU nesting whether | 
 |    the latest DW_AT_type seen for that level was a signed type or | 
 |    an unsigned type.  */ | 
 | #define MAX_CU_NESTING (1 << 8) | 
 | static bool level_type_signed[MAX_CU_NESTING]; | 
 |  | 
 | /* Values for do_debug_lines.  */ | 
 | #define FLAG_DEBUG_LINES_RAW	 1 | 
 | #define FLAG_DEBUG_LINES_DECODED 2 | 
 |  | 
 | static unsigned int | 
 | size_of_encoded_value (int encoding) | 
 | { | 
 |   switch (encoding & 0x7) | 
 |     { | 
 |     default:	/* ??? */ | 
 |     case 0:	return eh_addr_size; | 
 |     case 2:	return 2; | 
 |     case 3:	return 4; | 
 |     case 4:	return 8; | 
 |     } | 
 | } | 
 |  | 
 | static uint64_t | 
 | get_encoded_value (unsigned char **pdata, | 
 | 		   int encoding, | 
 | 		   struct dwarf_section *section, | 
 | 		   unsigned char * end) | 
 | { | 
 |   unsigned char * data = * pdata; | 
 |   unsigned int size = size_of_encoded_value (encoding); | 
 |   uint64_t val; | 
 |  | 
 |   if (data >= end || size > (size_t) (end - data)) | 
 |     { | 
 |       warn (_("Encoded value extends past end of section\n")); | 
 |       * pdata = end; | 
 |       return 0; | 
 |     } | 
 |  | 
 |   /* PR 17512: file: 002-829853-0.004.  */ | 
 |   if (size > 8) | 
 |     { | 
 |       warn (_("Encoded size of %d is too large to read\n"), size); | 
 |       * pdata = end; | 
 |       return 0; | 
 |     } | 
 |  | 
 |   /* PR 17512: file: 1085-5603-0.004.  */ | 
 |   if (size == 0) | 
 |     { | 
 |       warn (_("Encoded size of 0 is too small to read\n")); | 
 |       * pdata = end; | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (encoding & DW_EH_PE_signed) | 
 |     val = byte_get_signed (data, size); | 
 |   else | 
 |     val = byte_get (data, size); | 
 |  | 
 |   if ((encoding & 0x70) == DW_EH_PE_pcrel) | 
 |     val += section->address + (data - section->start); | 
 |  | 
 |   * pdata = data + size; | 
 |   return val; | 
 | } | 
 |  | 
 | /* Print a uint64_t value (typically an address, offset or length) in | 
 |    hexadecimal format, followed by a space.  The precision displayed is | 
 |    determined by the NUM_BYTES parameter.  */ | 
 |  | 
 | static void | 
 | print_hex (uint64_t value, unsigned num_bytes) | 
 | { | 
 |   if (num_bytes == 0) | 
 |     num_bytes = 2; | 
 |  | 
 |   printf ("%0*" PRIx64 " ", num_bytes * 2, | 
 | 	  value & ~(~(uint64_t) 0 << num_bytes * 4 << num_bytes * 4)); | 
 | } | 
 |  | 
 | /* Like print_hex, but no trailing space.  */ | 
 |  | 
 | static void | 
 | print_hex_ns (uint64_t value, unsigned num_bytes) | 
 | { | 
 |   if (num_bytes == 0) | 
 |     num_bytes = 2; | 
 |  | 
 |   printf ("%0*" PRIx64, num_bytes * 2, | 
 | 	  value & ~(~(uint64_t) 0 << num_bytes * 4 << num_bytes * 4)); | 
 | } | 
 |  | 
 | /* Print a view number in hexadecimal value, with the same width as | 
 |    print_hex would have printed it.  */ | 
 |  | 
 | static void | 
 | print_view (uint64_t value, unsigned num_bytes) | 
 | { | 
 |   if (num_bytes == 0) | 
 |     num_bytes = 2; | 
 |  | 
 |   printf ("v%0*" PRIx64 " ", num_bytes * 2 - 1, | 
 | 	  value & ~(~(uint64_t) 0 << num_bytes * 4 << num_bytes * 4)); | 
 | } | 
 |  | 
 | static const char * | 
 | null_name (const char *p) | 
 | { | 
 |   if (p == NULL) | 
 |     p = _("unknown"); | 
 |   return p; | 
 | } | 
 |  | 
 | /* Read in a LEB128 encoded value starting at address DATA. | 
 |    If SIGN is true, return a signed LEB128 value. | 
 |    If LENGTH_RETURN is not NULL, return in it the number of bytes read. | 
 |    If STATUS_RETURN is not NULL, return with bit 0 (LSB) set if the | 
 |    terminating byte was not found and with bit 1 set if the value | 
 |    overflows a uint64_t. | 
 |    No bytes will be read at address END or beyond.  */ | 
 |  | 
 | uint64_t | 
 | read_leb128 (unsigned char *data, | 
 | 	     const unsigned char *const end, | 
 | 	     bool sign, | 
 | 	     unsigned int *length_return, | 
 | 	     int *status_return) | 
 | { | 
 |   uint64_t result = 0; | 
 |   unsigned int num_read = 0; | 
 |   unsigned int shift = 0; | 
 |   int status = 1; | 
 |  | 
 |   while (data < end) | 
 |     { | 
 |       unsigned char byte = *data++; | 
 |       unsigned char lost, mask; | 
 |  | 
 |       num_read++; | 
 |  | 
 |       if (shift < CHAR_BIT * sizeof (result)) | 
 | 	{ | 
 | 	  result |= ((uint64_t) (byte & 0x7f)) << shift; | 
 | 	  /* These bits overflowed.  */ | 
 | 	  lost = byte ^ (result >> shift); | 
 | 	  /* And this is the mask of possible overflow bits.  */ | 
 | 	  mask = 0x7f ^ ((uint64_t) 0x7f << shift >> shift); | 
 | 	  shift += 7; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  lost = byte; | 
 | 	  mask = 0x7f; | 
 | 	} | 
 |       if ((lost & mask) != (sign && (int64_t) result < 0 ? mask : 0)) | 
 | 	status |= 2; | 
 |  | 
 |       if ((byte & 0x80) == 0) | 
 | 	{ | 
 | 	  status &= ~1; | 
 | 	  if (sign && shift < CHAR_BIT * sizeof (result) && (byte & 0x40)) | 
 | 	    result |= -((uint64_t) 1 << shift); | 
 | 	  break; | 
 | 	} | 
 |     } | 
 |  | 
 |   if (length_return != NULL) | 
 |     *length_return = num_read; | 
 |   if (status_return != NULL) | 
 |     *status_return = status; | 
 |  | 
 |   return result; | 
 | } | 
 |  | 
 | /* Read AMOUNT bytes from PTR and store them in VAL. | 
 |    Checks to make sure that the read will not reach or pass END. | 
 |    FUNC chooses whether the value read is unsigned or signed, and may | 
 |    be either byte_get or byte_get_signed.  If INC is true, PTR is | 
 |    incremented after reading the value. | 
 |    This macro cannot protect against PTR values derived from user input. | 
 |    The C standard sections 6.5.6 and 6.5.8 say attempts to do so using | 
 |    pointers is undefined behaviour.  */ | 
 | #define SAFE_BYTE_GET_INTERNAL(VAL, PTR, AMOUNT, END, FUNC, INC)	\ | 
 |   do									\ | 
 |     {									\ | 
 |       size_t amount = (AMOUNT);						\ | 
 |       if (sizeof (VAL) < amount)					\ | 
 | 	{								\ | 
 | 	  error (ngettext ("internal error: attempt to read %d byte "	\ | 
 | 			   "of data in to %d sized variable",		\ | 
 | 			   "internal error: attempt to read %d bytes "	\ | 
 | 			   "of data in to %d sized variable",		\ | 
 | 			   amount),					\ | 
 | 		 (int) amount, (int) sizeof (VAL));			\ | 
 | 	  amount = sizeof (VAL);					\ | 
 | 	}								\ | 
 |       if (ENABLE_CHECKING)						\ | 
 | 	assert ((PTR) <= (END));					\ | 
 |       size_t avail = (END) - (PTR);					\ | 
 |       if ((PTR) > (END))						\ | 
 | 	avail = 0;							\ | 
 |       if (amount > avail)						\ | 
 | 	amount = avail;							\ | 
 |       if (amount == 0)							\ | 
 | 	(VAL) = 0;							\ | 
 |       else								\ | 
 | 	(VAL) = (FUNC) ((PTR), amount);					\ | 
 |       if (INC)								\ | 
 | 	(PTR) += amount;						\ | 
 |     }									\ | 
 |   while (0) | 
 |  | 
 | #define SAFE_BYTE_GET(VAL, PTR, AMOUNT, END)	\ | 
 |   SAFE_BYTE_GET_INTERNAL (VAL, PTR, AMOUNT, END, byte_get, false) | 
 |  | 
 | #define SAFE_BYTE_GET_AND_INC(VAL, PTR, AMOUNT, END)	\ | 
 |   SAFE_BYTE_GET_INTERNAL (VAL, PTR, AMOUNT, END, byte_get, true) | 
 |  | 
 | #define SAFE_SIGNED_BYTE_GET(VAL, PTR, AMOUNT, END)	\ | 
 |   SAFE_BYTE_GET_INTERNAL (VAL, PTR, AMOUNT, END, byte_get_signed, false) | 
 |  | 
 | #define SAFE_SIGNED_BYTE_GET_AND_INC(VAL, PTR, AMOUNT, END)	\ | 
 |   SAFE_BYTE_GET_INTERNAL (VAL, PTR, AMOUNT, END, byte_get_signed, true) | 
 |  | 
 | typedef struct State_Machine_Registers | 
 | { | 
 |   uint64_t address; | 
 |   unsigned int view; | 
 |   unsigned int file; | 
 |   unsigned int line; | 
 |   unsigned int column; | 
 |   int is_stmt; | 
 |   int basic_block; | 
 |   unsigned char op_index; | 
 |   unsigned char end_sequence; | 
 |   /* This variable hold the number of the last entry seen | 
 |      in the File Table.  */ | 
 |   unsigned int last_file_entry; | 
 | } SMR; | 
 |  | 
 | static SMR state_machine_regs; | 
 |  | 
 | static void | 
 | reset_state_machine (int is_stmt) | 
 | { | 
 |   state_machine_regs.address = 0; | 
 |   state_machine_regs.view = 0; | 
 |   state_machine_regs.op_index = 0; | 
 |   state_machine_regs.file = 1; | 
 |   state_machine_regs.line = 1; | 
 |   state_machine_regs.column = 0; | 
 |   state_machine_regs.is_stmt = is_stmt; | 
 |   state_machine_regs.basic_block = 0; | 
 |   state_machine_regs.end_sequence = 0; | 
 |   state_machine_regs.last_file_entry = 0; | 
 | } | 
 |  | 
 | /* Handled an extend line op. | 
 |    Returns the number of bytes read.  */ | 
 |  | 
 | static size_t | 
 | process_extended_line_op (unsigned char * data, | 
 | 			  int is_stmt, | 
 | 			  unsigned char * end) | 
 | { | 
 |   unsigned char op_code; | 
 |   size_t len, header_len; | 
 |   unsigned char *name; | 
 |   unsigned char *orig_data = data; | 
 |   uint64_t adr, val; | 
 |  | 
 |   READ_ULEB (len, data, end); | 
 |   header_len = data - orig_data; | 
 |  | 
 |   if (len == 0 || data >= end || len > (size_t) (end - data)) | 
 |     { | 
 |       warn (_("Badly formed extended line op encountered!\n")); | 
 |       return header_len; | 
 |     } | 
 |  | 
 |   op_code = *data++; | 
 |  | 
 |   printf (_("  Extended opcode %d: "), op_code); | 
 |  | 
 |   switch (op_code) | 
 |     { | 
 |     case DW_LNE_end_sequence: | 
 |       printf (_("End of Sequence\n\n")); | 
 |       reset_state_machine (is_stmt); | 
 |       break; | 
 |  | 
 |     case DW_LNE_set_address: | 
 |       /* PR 17512: file: 002-100480-0.004.  */ | 
 |       if (len - 1 > 8) | 
 | 	{ | 
 | 	  warn (_("Length (%zu) of DW_LNE_set_address op is too long\n"), | 
 | 		len - 1); | 
 | 	  adr = 0; | 
 | 	} | 
 |       else | 
 | 	SAFE_BYTE_GET (adr, data, len - 1, end); | 
 |       printf (_("set Address to %#" PRIx64 "\n"), adr); | 
 |       state_machine_regs.address = adr; | 
 |       state_machine_regs.view = 0; | 
 |       state_machine_regs.op_index = 0; | 
 |       break; | 
 |  | 
 |     case DW_LNE_define_file: | 
 |       printf (_("define new File Table entry\n")); | 
 |       printf (_("  Entry\tDir\tTime\tSize\tName\n")); | 
 |       printf ("   %d\t", ++state_machine_regs.last_file_entry); | 
 |  | 
 |       { | 
 | 	size_t l; | 
 |  | 
 | 	name = data; | 
 | 	l = strnlen ((char *) data, end - data); | 
 | 	data += l; | 
 | 	if (data < end) | 
 | 	  data++; | 
 | 	READ_ULEB (val, data, end); | 
 | 	printf ("%" PRIu64 "\t", val); | 
 | 	READ_ULEB (val, data, end); | 
 | 	printf ("%" PRIu64 "\t", val); | 
 | 	READ_ULEB (val, data, end); | 
 | 	printf ("%" PRIu64 "\t", val); | 
 | 	printf ("%.*s\n\n", (int) l, name); | 
 |       } | 
 |  | 
 |       if (((size_t) (data - orig_data) != len + header_len) || data >= end) | 
 | 	warn (_("DW_LNE_define_file: Bad opcode length\n")); | 
 |       break; | 
 |  | 
 |     case DW_LNE_set_discriminator: | 
 |       READ_ULEB (val, data, end); | 
 |       printf (_("set Discriminator to %" PRIu64 "\n"), val); | 
 |       break; | 
 |  | 
 |       /* HP extensions.  */ | 
 |     case DW_LNE_HP_negate_is_UV_update: | 
 |       printf ("DW_LNE_HP_negate_is_UV_update\n"); | 
 |       break; | 
 |     case DW_LNE_HP_push_context: | 
 |       printf ("DW_LNE_HP_push_context\n"); | 
 |       break; | 
 |     case DW_LNE_HP_pop_context: | 
 |       printf ("DW_LNE_HP_pop_context\n"); | 
 |       break; | 
 |     case DW_LNE_HP_set_file_line_column: | 
 |       printf ("DW_LNE_HP_set_file_line_column\n"); | 
 |       break; | 
 |     case DW_LNE_HP_set_routine_name: | 
 |       printf ("DW_LNE_HP_set_routine_name\n"); | 
 |       break; | 
 |     case DW_LNE_HP_set_sequence: | 
 |       printf ("DW_LNE_HP_set_sequence\n"); | 
 |       break; | 
 |     case DW_LNE_HP_negate_post_semantics: | 
 |       printf ("DW_LNE_HP_negate_post_semantics\n"); | 
 |       break; | 
 |     case DW_LNE_HP_negate_function_exit: | 
 |       printf ("DW_LNE_HP_negate_function_exit\n"); | 
 |       break; | 
 |     case DW_LNE_HP_negate_front_end_logical: | 
 |       printf ("DW_LNE_HP_negate_front_end_logical\n"); | 
 |       break; | 
 |     case DW_LNE_HP_define_proc: | 
 |       printf ("DW_LNE_HP_define_proc\n"); | 
 |       break; | 
 |     case DW_LNE_HP_source_file_correlation: | 
 |       { | 
 | 	unsigned char *edata = data + len - 1; | 
 |  | 
 | 	printf ("DW_LNE_HP_source_file_correlation\n"); | 
 |  | 
 | 	while (data < edata) | 
 | 	  { | 
 | 	    unsigned int opc; | 
 |  | 
 | 	    READ_ULEB (opc, data, edata); | 
 |  | 
 | 	    switch (opc) | 
 | 	      { | 
 | 	      case DW_LNE_HP_SFC_formfeed: | 
 | 		printf ("    DW_LNE_HP_SFC_formfeed\n"); | 
 | 		break; | 
 | 	      case DW_LNE_HP_SFC_set_listing_line: | 
 | 		READ_ULEB (val, data, edata); | 
 | 		printf ("    DW_LNE_HP_SFC_set_listing_line (%" PRIu64 ")\n", | 
 | 			val); | 
 | 		break; | 
 | 	      case DW_LNE_HP_SFC_associate: | 
 | 		printf ("    DW_LNE_HP_SFC_associate "); | 
 | 		READ_ULEB (val, data, edata); | 
 | 		printf ("(%" PRIu64 , val); | 
 | 		READ_ULEB (val, data, edata); | 
 | 		printf (",%" PRIu64, val); | 
 | 		READ_ULEB (val, data, edata); | 
 | 		printf (",%" PRIu64 ")\n", val); | 
 | 		break; | 
 | 	      default: | 
 | 		printf (_("    UNKNOWN DW_LNE_HP_SFC opcode (%u)\n"), opc); | 
 | 		data = edata; | 
 | 		break; | 
 | 	      } | 
 | 	  } | 
 |       } | 
 |       break; | 
 |  | 
 |     default: | 
 |       { | 
 | 	unsigned int rlen = len - 1; | 
 |  | 
 | 	if (op_code >= DW_LNE_lo_user | 
 | 	    /* The test against DW_LNW_hi_user is redundant due to | 
 | 	       the limited range of the unsigned char data type used | 
 | 	       for op_code.  */ | 
 | 	    /*&& op_code <= DW_LNE_hi_user*/) | 
 | 	  printf (_("user defined: ")); | 
 | 	else | 
 | 	  printf (_("UNKNOWN: ")); | 
 | 	printf (_("length %d ["), rlen); | 
 | 	for (; rlen; rlen--) | 
 | 	  printf (" %02x", *data++); | 
 | 	printf ("]\n"); | 
 |       } | 
 |       break; | 
 |     } | 
 |  | 
 |   return len + header_len; | 
 | } | 
 |  | 
 | static const char * | 
 | fetch_indirect_string (uint64_t offset) | 
 | { | 
 |   struct dwarf_section *section = &debug_displays [str].section; | 
 |   const char *ret; | 
 |  | 
 |   if (section->start == NULL) | 
 |     return _("<no .debug_str section>"); | 
 |  | 
 |   if (offset >= section->size) | 
 |     { | 
 |       warn (_("DW_FORM_strp offset too big: %#" PRIx64 "\n"), offset); | 
 |       return _("<offset is too big>"); | 
 |     } | 
 |  | 
 |   ret = (const char *) (section->start + offset); | 
 |   /* Unfortunately we cannot rely upon the .debug_str section ending with a | 
 |      NUL byte.  Since our caller is expecting to receive a well formed C | 
 |      string we test for the lack of a terminating byte here.  */ | 
 |   if (strnlen (ret, section->size - offset) | 
 |       == section->size - offset) | 
 |     ret = _("<no NUL byte at end of .debug_str section>"); | 
 |  | 
 |   return ret; | 
 | } | 
 |  | 
 | static const char * | 
 | fetch_indirect_line_string (uint64_t offset) | 
 | { | 
 |   struct dwarf_section *section = &debug_displays [line_str].section; | 
 |   const char *ret; | 
 |  | 
 |   if (section->start == NULL) | 
 |     return _("<no .debug_line_str section>"); | 
 |  | 
 |   if (offset >= section->size) | 
 |     { | 
 |       warn (_("DW_FORM_line_strp offset too big: %#" PRIx64 "\n"), offset); | 
 |       return _("<offset is too big>"); | 
 |     } | 
 |  | 
 |   ret = (const char *) (section->start + offset); | 
 |   /* Unfortunately we cannot rely upon the .debug_line_str section ending | 
 |      with a NUL byte.  Since our caller is expecting to receive a well formed | 
 |      C string we test for the lack of a terminating byte here.  */ | 
 |   if (strnlen (ret, section->size - offset) | 
 |       == section->size - offset) | 
 |     ret = _("<no NUL byte at end of .debug_line_str section>"); | 
 |  | 
 |   return ret; | 
 | } | 
 |  | 
 | static const char * | 
 | fetch_indexed_string (uint64_t idx, | 
 | 		      struct cu_tu_set *this_set, | 
 | 		      uint64_t offset_size, | 
 | 		      bool dwo, | 
 | 		      uint64_t str_offsets_base) | 
 | { | 
 |   enum dwarf_section_display_enum str_sec_idx = dwo ? str_dwo : str; | 
 |   enum dwarf_section_display_enum idx_sec_idx = dwo ? str_index_dwo : str_index; | 
 |   struct dwarf_section *index_section = &debug_displays [idx_sec_idx].section; | 
 |   struct dwarf_section *str_section = &debug_displays [str_sec_idx].section; | 
 |   uint64_t index_offset; | 
 |   uint64_t str_offset; | 
 |   const char * ret; | 
 |  | 
 |   if (index_section->start == NULL) | 
 |     return (dwo ? _("<no .debug_str_offsets.dwo section>") | 
 | 		: _("<no .debug_str_offsets section>")); | 
 |  | 
 |   if (str_section->start == NULL) | 
 |     return (dwo ? _("<no .debug_str.dwo section>") | 
 | 		: _("<no .debug_str section>")); | 
 |  | 
 |   if (_mul_overflow (idx, offset_size, &index_offset) | 
 |       || (this_set != NULL | 
 | 	  && ((index_offset += this_set->section_offsets [DW_SECT_STR_OFFSETS]) | 
 | 	      < this_set->section_offsets [DW_SECT_STR_OFFSETS])) | 
 |       || (index_offset += str_offsets_base) < str_offsets_base | 
 |       || index_offset + offset_size < offset_size | 
 |       || index_offset + offset_size > index_section->size) | 
 |     { | 
 |       warn (_("string index of %" PRIu64 " converts to an offset of %#" PRIx64 | 
 | 	      " which is too big for section %s\n"), | 
 | 	    idx, index_offset, str_section->name); | 
 |  | 
 |       return _("<string index too big>"); | 
 |     } | 
 |  | 
 |   str_offset = byte_get (index_section->start + index_offset, offset_size); | 
 |  | 
 |   if (str_offset >= str_section->size) | 
 |     { | 
 |       warn (_("indirect offset too big: %#" PRIx64 "\n"), str_offset); | 
 |       return _("<indirect index offset is too big>"); | 
 |     } | 
 |  | 
 |   ret = (const char *) str_section->start + str_offset; | 
 |  | 
 |   /* Unfortunately we cannot rely upon str_section ending with a NUL byte. | 
 |      Since our caller is expecting to receive a well formed C string we test | 
 |      for the lack of a terminating byte here.  */ | 
 |   if (strnlen (ret, str_section->size - str_offset) | 
 |       == str_section->size - str_offset) | 
 |     return _("<no NUL byte at end of section>"); | 
 |  | 
 |   return ret; | 
 | } | 
 |  | 
 | static uint64_t | 
 | fetch_indexed_addr (uint64_t offset, uint32_t num_bytes) | 
 | { | 
 |   struct dwarf_section *section = &debug_displays [debug_addr].section; | 
 |  | 
 |   if (section->start == NULL) | 
 |     { | 
 |       warn (_("Cannot fetch indexed address: the .debug_addr section is missing\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (offset + num_bytes > section->size) | 
 |     { | 
 |       warn (_("Offset into section %s too big: %#" PRIx64 "\n"), | 
 | 	    section->name, offset); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   return byte_get (section->start + offset, num_bytes); | 
 | } | 
 |  | 
 | /* This is for resolving DW_FORM_rnglistx and DW_FORM_loclistx. | 
 |  | 
 |    The memory layout is: base_address (taken from the CU's top DIE) points at a table of offsets, | 
 |    relative to the section start. | 
 |    The table of offsets contains the offsets of objects of interest relative to the table of offsets. | 
 |    IDX is the index of the desired object in said table of offsets. | 
 |  | 
 |    This returns the offset of the desired object relative to the section start or -1 upon failure.  */ | 
 |  | 
 | static uint64_t | 
 | fetch_indexed_offset (uint64_t                         idx, | 
 | 		      enum dwarf_section_display_enum  sec_enum, | 
 | 		      uint64_t                         base_address, | 
 | 		      uint64_t                         offset_size) | 
 | { | 
 |   uint64_t offset_of_offset = base_address + idx * offset_size; | 
 |   struct dwarf_section *section = &debug_displays [sec_enum].section; | 
 |  | 
 |   if (section->start == NULL) | 
 |     { | 
 |       warn (_("Unable to locate %s section\n"), section->uncompressed_name); | 
 |       return -1; | 
 |     } | 
 |  | 
 |   if (section->size < 4) | 
 |     { | 
 |       warn (_("Section %s is too small to contain an value indexed from another section!\n"), | 
 | 	    section->name); | 
 |       return -1; | 
 |     } | 
 |  | 
 |   if (offset_of_offset + offset_size >= section->size) | 
 |     { | 
 |       warn (_("Offset of %#" PRIx64 " is too big for section %s\n"), | 
 | 	    offset_of_offset, section->name); | 
 |       return -1; | 
 |     } | 
 |  | 
 |   return base_address + byte_get (section->start + offset_of_offset, offset_size); | 
 | } | 
 |  | 
 | /* FIXME:  There are better and more efficient ways to handle | 
 |    these structures.  For now though, I just want something that | 
 |    is simple to implement.  */ | 
 | /* Records a single attribute in an abbrev.  */ | 
 | typedef struct abbrev_attr | 
 | { | 
 |   unsigned long attribute; | 
 |   unsigned long form; | 
 |   int64_t implicit_const; | 
 |   struct abbrev_attr *next; | 
 | } | 
 | abbrev_attr; | 
 |  | 
 | /* Records a single abbrev.  */ | 
 | typedef struct abbrev_entry | 
 | { | 
 |   unsigned long          number; | 
 |   unsigned long          tag; | 
 |   int                    children; | 
 |   struct abbrev_attr *   first_attr; | 
 |   struct abbrev_attr *   last_attr; | 
 |   struct abbrev_entry *  next; | 
 | } | 
 | abbrev_entry; | 
 |  | 
 | /* Records a set of abbreviations.  */ | 
 | typedef struct abbrev_list | 
 | { | 
 |   abbrev_entry *        first_abbrev; | 
 |   abbrev_entry *        last_abbrev; | 
 |   unsigned char *       raw; | 
 |   struct abbrev_list *  next; | 
 |   unsigned char *       start_of_next_abbrevs; | 
 | } | 
 | abbrev_list; | 
 |  | 
 | /* Records all the abbrevs found so far.  */ | 
 | static struct abbrev_list * abbrev_lists = NULL; | 
 |  | 
 | typedef struct abbrev_map | 
 | { | 
 |   uint64_t start; | 
 |   uint64_t end; | 
 |   abbrev_list *list; | 
 | } abbrev_map; | 
 |  | 
 | /* Maps between CU offsets and abbrev sets.  */ | 
 | static abbrev_map *   cu_abbrev_map = NULL; | 
 | static unsigned long  num_abbrev_map_entries = 0; | 
 | static unsigned long  next_free_abbrev_map_entry = 0; | 
 |  | 
 | #define INITIAL_NUM_ABBREV_MAP_ENTRIES 8 | 
 | #define ABBREV_MAP_ENTRIES_INCREMENT   8 | 
 |  | 
 | static void | 
 | record_abbrev_list_for_cu (uint64_t start, uint64_t end, | 
 | 			   abbrev_list *list, abbrev_list *free_list) | 
 | { | 
 |   if (free_list != NULL) | 
 |     { | 
 |       list->next = abbrev_lists; | 
 |       abbrev_lists = list; | 
 |     } | 
 |  | 
 |   if (cu_abbrev_map == NULL) | 
 |     { | 
 |       num_abbrev_map_entries = INITIAL_NUM_ABBREV_MAP_ENTRIES; | 
 |       cu_abbrev_map = xmalloc (num_abbrev_map_entries * sizeof (* cu_abbrev_map)); | 
 |     } | 
 |   else if (next_free_abbrev_map_entry == num_abbrev_map_entries) | 
 |     { | 
 |       num_abbrev_map_entries += ABBREV_MAP_ENTRIES_INCREMENT; | 
 |       cu_abbrev_map = xrealloc (cu_abbrev_map, num_abbrev_map_entries * sizeof (* cu_abbrev_map)); | 
 |     } | 
 |  | 
 |   cu_abbrev_map[next_free_abbrev_map_entry].start = start; | 
 |   cu_abbrev_map[next_free_abbrev_map_entry].end = end; | 
 |   cu_abbrev_map[next_free_abbrev_map_entry].list = list; | 
 |   next_free_abbrev_map_entry ++; | 
 | } | 
 |  | 
 | static abbrev_list * | 
 | free_abbrev_list (abbrev_list *list) | 
 | { | 
 |   abbrev_entry *abbrv = list->first_abbrev; | 
 |  | 
 |   while (abbrv) | 
 |     { | 
 |       abbrev_attr *attr = abbrv->first_attr; | 
 |  | 
 |       while (attr) | 
 | 	{ | 
 | 	  abbrev_attr *next_attr = attr->next; | 
 | 	  free (attr); | 
 | 	  attr = next_attr; | 
 | 	} | 
 |  | 
 |       abbrev_entry *next_abbrev = abbrv->next; | 
 |       free (abbrv); | 
 |       abbrv = next_abbrev; | 
 |     } | 
 |  | 
 |   abbrev_list *next = list->next; | 
 |   free (list); | 
 |   return next; | 
 | } | 
 |  | 
 | static void | 
 | free_all_abbrevs (void) | 
 | { | 
 |   while (abbrev_lists) | 
 |     abbrev_lists = free_abbrev_list (abbrev_lists); | 
 |  | 
 |   free (cu_abbrev_map); | 
 |   cu_abbrev_map = NULL; | 
 |   next_free_abbrev_map_entry = 0; | 
 | } | 
 |  | 
 | static abbrev_list * | 
 | find_abbrev_list_by_raw_abbrev (unsigned char *raw) | 
 | { | 
 |   abbrev_list * list; | 
 |  | 
 |   for (list = abbrev_lists; list != NULL; list = list->next) | 
 |     if (list->raw == raw) | 
 |       return list; | 
 |  | 
 |   return NULL; | 
 | } | 
 |  | 
 | /* Find the abbreviation map for the CU that includes OFFSET. | 
 |    OFFSET is an absolute offset from the start of the .debug_info section.  */ | 
 | /* FIXME: This function is going to slow down readelf & objdump. | 
 |    Not caching abbrevs is likely the answer.  */ | 
 |  | 
 | static  abbrev_map * | 
 | find_abbrev_map_by_offset (uint64_t offset) | 
 | { | 
 |   unsigned long i; | 
 |  | 
 |   for (i = 0; i < next_free_abbrev_map_entry; i++) | 
 |     if (cu_abbrev_map[i].start <= offset | 
 | 	&& cu_abbrev_map[i].end > offset) | 
 |       return cu_abbrev_map + i; | 
 |  | 
 |   return NULL; | 
 | } | 
 |  | 
 | static void | 
 | add_abbrev (unsigned long  number, | 
 | 	    unsigned long  tag, | 
 | 	    int            children, | 
 | 	    abbrev_list *  list) | 
 | { | 
 |   abbrev_entry *  entry; | 
 |  | 
 |   entry = (abbrev_entry *) xmalloc (sizeof (*entry)); | 
 |  | 
 |   entry->number     = number; | 
 |   entry->tag        = tag; | 
 |   entry->children   = children; | 
 |   entry->first_attr = NULL; | 
 |   entry->last_attr  = NULL; | 
 |   entry->next       = NULL; | 
 |  | 
 |   assert (list != NULL); | 
 |  | 
 |   if (list->first_abbrev == NULL) | 
 |     list->first_abbrev = entry; | 
 |   else | 
 |     list->last_abbrev->next = entry; | 
 |  | 
 |   list->last_abbrev = entry; | 
 | } | 
 |  | 
 | static void | 
 | add_abbrev_attr (unsigned long attribute, | 
 | 		 unsigned long form, | 
 | 		 int64_t implicit_const, | 
 | 		 abbrev_list *list) | 
 | { | 
 |   abbrev_attr *attr; | 
 |  | 
 |   attr = (abbrev_attr *) xmalloc (sizeof (*attr)); | 
 |  | 
 |   attr->attribute      = attribute; | 
 |   attr->form           = form; | 
 |   attr->implicit_const = implicit_const; | 
 |   attr->next           = NULL; | 
 |  | 
 |   assert (list != NULL && list->last_abbrev != NULL); | 
 |  | 
 |   if (list->last_abbrev->first_attr == NULL) | 
 |     list->last_abbrev->first_attr = attr; | 
 |   else | 
 |     list->last_abbrev->last_attr->next = attr; | 
 |  | 
 |   list->last_abbrev->last_attr = attr; | 
 | } | 
 |  | 
 | /* Return processed (partial) contents of a .debug_abbrev section. | 
 |    Returns NULL on errors.  */ | 
 |  | 
 | static abbrev_list * | 
 | process_abbrev_set (struct dwarf_section *section, | 
 | 		    unsigned char *start, | 
 | 		    unsigned char *end) | 
 | { | 
 |   abbrev_list *list = xmalloc (sizeof (*list)); | 
 |   list->first_abbrev = NULL; | 
 |   list->last_abbrev = NULL; | 
 |   list->raw = start; | 
 |   list->next = NULL; | 
 |  | 
 |   while (start < end) | 
 |     { | 
 |       unsigned long entry; | 
 |       unsigned long tag; | 
 |       unsigned long attribute; | 
 |       int children; | 
 |  | 
 |       READ_ULEB (entry, start, end); | 
 |  | 
 |       /* A single zero is supposed to end the set according | 
 | 	 to the standard.  If there's more, then signal that to | 
 | 	 the caller.  */ | 
 |       if (start == end || entry == 0) | 
 | 	{ | 
 | 	  list->start_of_next_abbrevs = start != end ? start : NULL; | 
 | 	  return list; | 
 | 	} | 
 |  | 
 |       READ_ULEB (tag, start, end); | 
 |       if (start == end) | 
 | 	return free_abbrev_list (list); | 
 |  | 
 |       children = *start++; | 
 |  | 
 |       add_abbrev (entry, tag, children, list); | 
 |  | 
 |       do | 
 | 	{ | 
 | 	  unsigned long form; | 
 | 	  /* Initialize it due to a false compiler warning.  */ | 
 | 	  int64_t implicit_const = -1; | 
 |  | 
 | 	  READ_ULEB (attribute, start, end); | 
 | 	  if (start == end) | 
 | 	    break; | 
 |  | 
 | 	  READ_ULEB (form, start, end); | 
 | 	  if (start == end) | 
 | 	    break; | 
 |  | 
 | 	  if (form == DW_FORM_implicit_const) | 
 | 	    { | 
 | 	      READ_SLEB (implicit_const, start, end); | 
 | 	      if (start == end) | 
 | 		break; | 
 | 	    } | 
 |  | 
 | 	  add_abbrev_attr (attribute, form, implicit_const, list); | 
 | 	} | 
 |       while (attribute != 0); | 
 |     } | 
 |  | 
 |   /* Report the missing single zero which ends the section.  */ | 
 |   error (_("%s section not zero terminated\n"), section->name); | 
 |  | 
 |   return free_abbrev_list (list); | 
 | } | 
 |  | 
 | /* Return a sequence of abbrevs in SECTION starting at ABBREV_BASE | 
 |    plus ABBREV_OFFSET and finishing at ABBREV_BASE + ABBREV_SIZE. | 
 |    If FREE_LIST is non-NULL search the already decoded abbrevs on | 
 |    abbrev_lists first and if found set *FREE_LIST to NULL.  If | 
 |    searching doesn't find a matching abbrev, set *FREE_LIST to the | 
 |    newly allocated list.  If FREE_LIST is NULL, no search is done and | 
 |    the returned abbrev_list is always newly allocated.  */ | 
 |  | 
 | static abbrev_list * | 
 | find_and_process_abbrev_set (struct dwarf_section *section, | 
 | 			     uint64_t abbrev_base, | 
 | 			     uint64_t abbrev_size, | 
 | 			     uint64_t abbrev_offset, | 
 | 			     abbrev_list **free_list) | 
 | { | 
 |   if (free_list) | 
 |     *free_list = NULL; | 
 |  | 
 |   if (abbrev_base >= section->size | 
 |       || abbrev_size > section->size - abbrev_base) | 
 |     { | 
 |       /* PR 17531: file:4bcd9ce9.  */ | 
 |       warn (_("Debug info is corrupted, abbrev size (%#" PRIx64 ")" | 
 | 	      " is larger than abbrev section size (%#" PRIx64 ")\n"), | 
 | 	    abbrev_base + abbrev_size, section->size); | 
 |       return NULL; | 
 |     } | 
 |   if (abbrev_offset >= abbrev_size) | 
 |     { | 
 |       warn (_("Debug info is corrupted, abbrev offset (%#" PRIx64 ")" | 
 | 	      " is larger than abbrev section size (%#" PRIx64 ")\n"), | 
 | 	    abbrev_offset, abbrev_size); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   unsigned char *start = section->start + abbrev_base + abbrev_offset; | 
 |   unsigned char *end = section->start + abbrev_base + abbrev_size; | 
 |   abbrev_list *list = NULL; | 
 |   if (free_list) | 
 |     list = find_abbrev_list_by_raw_abbrev (start); | 
 |   if (list == NULL) | 
 |     { | 
 |       list = process_abbrev_set (section, start, end); | 
 |       if (free_list) | 
 | 	*free_list = list; | 
 |     } | 
 |   return list; | 
 | } | 
 |  | 
 | static const char * | 
 | get_TAG_name (uint64_t tag) | 
 | { | 
 |   const char *name = NULL; | 
 |  | 
 |   if ((unsigned int) tag == tag) | 
 |     name = get_DW_TAG_name ((unsigned int) tag); | 
 |   if (name == NULL) | 
 |     { | 
 |       static char buffer[100]; | 
 |  | 
 |       if (tag >= DW_TAG_lo_user && tag <= DW_TAG_hi_user) | 
 | 	snprintf (buffer, sizeof (buffer), | 
 | 		  _("User TAG value: %#" PRIx64), tag); | 
 |       else | 
 | 	snprintf (buffer, sizeof (buffer), | 
 | 		  _("Unknown TAG value: %#" PRIx64), tag); | 
 |       return buffer; | 
 |     } | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static const char * | 
 | get_FORM_name (unsigned long form) | 
 | { | 
 |   const char *name = NULL; | 
 |  | 
 |   if (form == 0) | 
 |     return "DW_FORM value: 0"; | 
 |  | 
 |   if ((unsigned int) form == form) | 
 |     name = get_DW_FORM_name ((unsigned int) form); | 
 |   if (name == NULL) | 
 |     { | 
 |       static char buffer[100]; | 
 |  | 
 |       snprintf (buffer, sizeof (buffer), _("Unknown FORM value: %lx"), form); | 
 |       return buffer; | 
 |     } | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static const char * | 
 | get_IDX_name (unsigned long idx) | 
 | { | 
 |   const char *name = NULL; | 
 |  | 
 |   if ((unsigned int) idx == idx) | 
 |     name = get_DW_IDX_name ((unsigned int) idx); | 
 |   if (name == NULL) | 
 |     { | 
 |       static char buffer[100]; | 
 |  | 
 |       snprintf (buffer, sizeof (buffer), _("Unknown IDX value: %lx"), idx); | 
 |       return buffer; | 
 |     } | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static unsigned char * | 
 | display_block (unsigned char *data, | 
 | 	       uint64_t length, | 
 | 	       const unsigned char * const end, char delimiter) | 
 | { | 
 |   size_t maxlen; | 
 |  | 
 |   printf (_("%c%" PRIu64 " byte block: "), delimiter, length); | 
 |   if (data > end) | 
 |     return (unsigned char *) end; | 
 |  | 
 |   maxlen = end - data; | 
 |   length = length > maxlen ? maxlen : length; | 
 |  | 
 |   while (length --) | 
 |     printf ("%" PRIx64 " ", byte_get (data++, 1)); | 
 |  | 
 |   return data; | 
 | } | 
 |  | 
 | static int | 
 | decode_location_expression (unsigned char * data, | 
 | 			    unsigned int pointer_size, | 
 | 			    unsigned int offset_size, | 
 | 			    int dwarf_version, | 
 | 			    uint64_t length, | 
 | 			    uint64_t cu_offset, | 
 | 			    struct dwarf_section * section) | 
 | { | 
 |   unsigned op; | 
 |   uint64_t uvalue; | 
 |   int64_t svalue; | 
 |   unsigned char *end = data + length; | 
 |   int need_frame_base = 0; | 
 |  | 
 |   while (data < end) | 
 |     { | 
 |       op = *data++; | 
 |  | 
 |       switch (op) | 
 | 	{ | 
 | 	case DW_OP_addr: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 | 	  printf ("DW_OP_addr: %" PRIx64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_deref: | 
 | 	  printf ("DW_OP_deref"); | 
 | 	  break; | 
 | 	case DW_OP_const1u: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  printf ("DW_OP_const1u: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_const1s: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 1, end); | 
 | 	  printf ("DW_OP_const1s: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_const2u: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 2, end); | 
 | 	  printf ("DW_OP_const2u: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_const2s: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 2, end); | 
 | 	  printf ("DW_OP_const2s: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_const4u: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 | 	  printf ("DW_OP_const4u: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_const4s: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 4, end); | 
 | 	  printf ("DW_OP_const4s: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_const8u: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 8, end); | 
 | 	  printf ("DW_OP_const8u: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_const8s: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 8, end); | 
 | 	  printf ("DW_OP_const8s: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_constu: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_constu: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_consts: | 
 | 	  READ_SLEB (svalue, data, end); | 
 | 	  printf ("DW_OP_consts: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_dup: | 
 | 	  printf ("DW_OP_dup"); | 
 | 	  break; | 
 | 	case DW_OP_drop: | 
 | 	  printf ("DW_OP_drop"); | 
 | 	  break; | 
 | 	case DW_OP_over: | 
 | 	  printf ("DW_OP_over"); | 
 | 	  break; | 
 | 	case DW_OP_pick: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  printf ("DW_OP_pick: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_swap: | 
 | 	  printf ("DW_OP_swap"); | 
 | 	  break; | 
 | 	case DW_OP_rot: | 
 | 	  printf ("DW_OP_rot"); | 
 | 	  break; | 
 | 	case DW_OP_xderef: | 
 | 	  printf ("DW_OP_xderef"); | 
 | 	  break; | 
 | 	case DW_OP_abs: | 
 | 	  printf ("DW_OP_abs"); | 
 | 	  break; | 
 | 	case DW_OP_and: | 
 | 	  printf ("DW_OP_and"); | 
 | 	  break; | 
 | 	case DW_OP_div: | 
 | 	  printf ("DW_OP_div"); | 
 | 	  break; | 
 | 	case DW_OP_minus: | 
 | 	  printf ("DW_OP_minus"); | 
 | 	  break; | 
 | 	case DW_OP_mod: | 
 | 	  printf ("DW_OP_mod"); | 
 | 	  break; | 
 | 	case DW_OP_mul: | 
 | 	  printf ("DW_OP_mul"); | 
 | 	  break; | 
 | 	case DW_OP_neg: | 
 | 	  printf ("DW_OP_neg"); | 
 | 	  break; | 
 | 	case DW_OP_not: | 
 | 	  printf ("DW_OP_not"); | 
 | 	  break; | 
 | 	case DW_OP_or: | 
 | 	  printf ("DW_OP_or"); | 
 | 	  break; | 
 | 	case DW_OP_plus: | 
 | 	  printf ("DW_OP_plus"); | 
 | 	  break; | 
 | 	case DW_OP_plus_uconst: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_plus_uconst: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_shl: | 
 | 	  printf ("DW_OP_shl"); | 
 | 	  break; | 
 | 	case DW_OP_shr: | 
 | 	  printf ("DW_OP_shr"); | 
 | 	  break; | 
 | 	case DW_OP_shra: | 
 | 	  printf ("DW_OP_shra"); | 
 | 	  break; | 
 | 	case DW_OP_xor: | 
 | 	  printf ("DW_OP_xor"); | 
 | 	  break; | 
 | 	case DW_OP_bra: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 2, end); | 
 | 	  printf ("DW_OP_bra: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_eq: | 
 | 	  printf ("DW_OP_eq"); | 
 | 	  break; | 
 | 	case DW_OP_ge: | 
 | 	  printf ("DW_OP_ge"); | 
 | 	  break; | 
 | 	case DW_OP_gt: | 
 | 	  printf ("DW_OP_gt"); | 
 | 	  break; | 
 | 	case DW_OP_le: | 
 | 	  printf ("DW_OP_le"); | 
 | 	  break; | 
 | 	case DW_OP_lt: | 
 | 	  printf ("DW_OP_lt"); | 
 | 	  break; | 
 | 	case DW_OP_ne: | 
 | 	  printf ("DW_OP_ne"); | 
 | 	  break; | 
 | 	case DW_OP_skip: | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 2, end); | 
 | 	  printf ("DW_OP_skip: %" PRId64, svalue); | 
 | 	  break; | 
 |  | 
 | 	case DW_OP_lit0: | 
 | 	case DW_OP_lit1: | 
 | 	case DW_OP_lit2: | 
 | 	case DW_OP_lit3: | 
 | 	case DW_OP_lit4: | 
 | 	case DW_OP_lit5: | 
 | 	case DW_OP_lit6: | 
 | 	case DW_OP_lit7: | 
 | 	case DW_OP_lit8: | 
 | 	case DW_OP_lit9: | 
 | 	case DW_OP_lit10: | 
 | 	case DW_OP_lit11: | 
 | 	case DW_OP_lit12: | 
 | 	case DW_OP_lit13: | 
 | 	case DW_OP_lit14: | 
 | 	case DW_OP_lit15: | 
 | 	case DW_OP_lit16: | 
 | 	case DW_OP_lit17: | 
 | 	case DW_OP_lit18: | 
 | 	case DW_OP_lit19: | 
 | 	case DW_OP_lit20: | 
 | 	case DW_OP_lit21: | 
 | 	case DW_OP_lit22: | 
 | 	case DW_OP_lit23: | 
 | 	case DW_OP_lit24: | 
 | 	case DW_OP_lit25: | 
 | 	case DW_OP_lit26: | 
 | 	case DW_OP_lit27: | 
 | 	case DW_OP_lit28: | 
 | 	case DW_OP_lit29: | 
 | 	case DW_OP_lit30: | 
 | 	case DW_OP_lit31: | 
 | 	  printf ("DW_OP_lit%d", op - DW_OP_lit0); | 
 | 	  break; | 
 |  | 
 | 	case DW_OP_reg0: | 
 | 	case DW_OP_reg1: | 
 | 	case DW_OP_reg2: | 
 | 	case DW_OP_reg3: | 
 | 	case DW_OP_reg4: | 
 | 	case DW_OP_reg5: | 
 | 	case DW_OP_reg6: | 
 | 	case DW_OP_reg7: | 
 | 	case DW_OP_reg8: | 
 | 	case DW_OP_reg9: | 
 | 	case DW_OP_reg10: | 
 | 	case DW_OP_reg11: | 
 | 	case DW_OP_reg12: | 
 | 	case DW_OP_reg13: | 
 | 	case DW_OP_reg14: | 
 | 	case DW_OP_reg15: | 
 | 	case DW_OP_reg16: | 
 | 	case DW_OP_reg17: | 
 | 	case DW_OP_reg18: | 
 | 	case DW_OP_reg19: | 
 | 	case DW_OP_reg20: | 
 | 	case DW_OP_reg21: | 
 | 	case DW_OP_reg22: | 
 | 	case DW_OP_reg23: | 
 | 	case DW_OP_reg24: | 
 | 	case DW_OP_reg25: | 
 | 	case DW_OP_reg26: | 
 | 	case DW_OP_reg27: | 
 | 	case DW_OP_reg28: | 
 | 	case DW_OP_reg29: | 
 | 	case DW_OP_reg30: | 
 | 	case DW_OP_reg31: | 
 | 	  printf ("DW_OP_reg%d (%s)", op - DW_OP_reg0, | 
 | 		  regname (op - DW_OP_reg0, 1)); | 
 | 	  break; | 
 |  | 
 | 	case DW_OP_breg0: | 
 | 	case DW_OP_breg1: | 
 | 	case DW_OP_breg2: | 
 | 	case DW_OP_breg3: | 
 | 	case DW_OP_breg4: | 
 | 	case DW_OP_breg5: | 
 | 	case DW_OP_breg6: | 
 | 	case DW_OP_breg7: | 
 | 	case DW_OP_breg8: | 
 | 	case DW_OP_breg9: | 
 | 	case DW_OP_breg10: | 
 | 	case DW_OP_breg11: | 
 | 	case DW_OP_breg12: | 
 | 	case DW_OP_breg13: | 
 | 	case DW_OP_breg14: | 
 | 	case DW_OP_breg15: | 
 | 	case DW_OP_breg16: | 
 | 	case DW_OP_breg17: | 
 | 	case DW_OP_breg18: | 
 | 	case DW_OP_breg19: | 
 | 	case DW_OP_breg20: | 
 | 	case DW_OP_breg21: | 
 | 	case DW_OP_breg22: | 
 | 	case DW_OP_breg23: | 
 | 	case DW_OP_breg24: | 
 | 	case DW_OP_breg25: | 
 | 	case DW_OP_breg26: | 
 | 	case DW_OP_breg27: | 
 | 	case DW_OP_breg28: | 
 | 	case DW_OP_breg29: | 
 | 	case DW_OP_breg30: | 
 | 	case DW_OP_breg31: | 
 | 	  READ_SLEB (svalue, data, end); | 
 | 	  printf ("DW_OP_breg%d (%s): %" PRId64, | 
 | 		  op - DW_OP_breg0, regname (op - DW_OP_breg0, 1), svalue); | 
 | 	  break; | 
 |  | 
 | 	case DW_OP_regx: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_regx: %" PRIu64 " (%s)", | 
 | 		  uvalue, regname (uvalue, 1)); | 
 | 	  break; | 
 | 	case DW_OP_fbreg: | 
 | 	  need_frame_base = 1; | 
 | 	  READ_SLEB (svalue, data, end); | 
 | 	  printf ("DW_OP_fbreg: %" PRId64, svalue); | 
 | 	  break; | 
 | 	case DW_OP_bregx: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  READ_SLEB (svalue, data, end); | 
 | 	  printf ("DW_OP_bregx: %" PRIu64 " (%s) %" PRId64, | 
 | 		  uvalue, regname (uvalue, 1), svalue); | 
 | 	  break; | 
 | 	case DW_OP_piece: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_piece: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_deref_size: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  printf ("DW_OP_deref_size: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_xderef_size: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  printf ("DW_OP_xderef_size: %" PRIu64, uvalue); | 
 | 	  break; | 
 | 	case DW_OP_nop: | 
 | 	  printf ("DW_OP_nop"); | 
 | 	  break; | 
 |  | 
 | 	  /* DWARF 3 extensions.  */ | 
 | 	case DW_OP_push_object_address: | 
 | 	  printf ("DW_OP_push_object_address"); | 
 | 	  break; | 
 | 	case DW_OP_call2: | 
 | 	  /* FIXME: Strictly speaking for 64-bit DWARF3 files | 
 | 	     this ought to be an 8-byte wide computation.  */ | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 2, end); | 
 | 	  printf ("DW_OP_call2: <%#" PRIx64 ">", svalue + cu_offset); | 
 | 	  break; | 
 | 	case DW_OP_call4: | 
 | 	  /* FIXME: Strictly speaking for 64-bit DWARF3 files | 
 | 	     this ought to be an 8-byte wide computation.  */ | 
 | 	  SAFE_SIGNED_BYTE_GET_AND_INC (svalue, data, 4, end); | 
 | 	  printf ("DW_OP_call4: <%#" PRIx64 ">", svalue + cu_offset); | 
 | 	  break; | 
 | 	case DW_OP_call_ref: | 
 | 	  /* FIXME: Strictly speaking for 64-bit DWARF3 files | 
 | 	     this ought to be an 8-byte wide computation.  */ | 
 | 	  if (dwarf_version == -1) | 
 | 	    { | 
 | 	      printf (_("(DW_OP_call_ref in frame info)")); | 
 | 	      /* No way to tell where the next op is, so just bail.  */ | 
 | 	      return need_frame_base; | 
 | 	    } | 
 | 	  if (dwarf_version == 2) | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 | 	    } | 
 | 	  printf ("DW_OP_call_ref: <%#" PRIx64 ">", uvalue); | 
 | 	  break; | 
 | 	case DW_OP_form_tls_address: | 
 | 	  printf ("DW_OP_form_tls_address"); | 
 | 	  break; | 
 | 	case DW_OP_call_frame_cfa: | 
 | 	  printf ("DW_OP_call_frame_cfa"); | 
 | 	  break; | 
 | 	case DW_OP_bit_piece: | 
 | 	  printf ("DW_OP_bit_piece: "); | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf (_("size: %" PRIu64 " "), uvalue); | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf (_("offset: %" PRIu64 " "), uvalue); | 
 | 	  break; | 
 |  | 
 | 	  /* DWARF 4 extensions.  */ | 
 | 	case DW_OP_stack_value: | 
 | 	  printf ("DW_OP_stack_value"); | 
 | 	  break; | 
 |  | 
 | 	case DW_OP_implicit_value: | 
 | 	  printf ("DW_OP_implicit_value"); | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  data = display_block (data, uvalue, end, ' '); | 
 | 	  break; | 
 |  | 
 | 	  /* GNU extensions.  */ | 
 | 	case DW_OP_GNU_push_tls_address: | 
 | 	  printf (_("DW_OP_GNU_push_tls_address or DW_OP_HP_unknown")); | 
 | 	  break; | 
 | 	case DW_OP_GNU_uninit: | 
 | 	  printf ("DW_OP_GNU_uninit"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_GNU_encoded_addr: | 
 | 	  { | 
 | 	    int encoding = 0; | 
 | 	    uint64_t addr; | 
 |  | 
 | 	    if (data < end) | 
 | 	      encoding = *data++; | 
 | 	    addr = get_encoded_value (&data, encoding, section, end); | 
 |  | 
 | 	    printf ("DW_OP_GNU_encoded_addr: fmt:%02x addr:", encoding); | 
 | 	    print_hex_ns (addr, pointer_size); | 
 | 	  } | 
 | 	  break; | 
 | 	case DW_OP_implicit_pointer: | 
 | 	case DW_OP_GNU_implicit_pointer: | 
 | 	  /* FIXME: Strictly speaking for 64-bit DWARF3 files | 
 | 	     this ought to be an 8-byte wide computation.  */ | 
 | 	  if (dwarf_version == -1) | 
 | 	    { | 
 | 	      printf (_("(%s in frame info)"), | 
 | 		      (op == DW_OP_implicit_pointer | 
 | 		       ? "DW_OP_implicit_pointer" | 
 | 		       : "DW_OP_GNU_implicit_pointer")); | 
 | 	      /* No way to tell where the next op is, so just bail.  */ | 
 | 	      return need_frame_base; | 
 | 	    } | 
 | 	  if (dwarf_version == 2) | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 | 	    } | 
 | 	  READ_SLEB (svalue, data, end); | 
 | 	  printf ("%s: <%#" PRIx64 "> %" PRId64, | 
 | 		  (op == DW_OP_implicit_pointer | 
 | 		   ? "DW_OP_implicit_pointer" : "DW_OP_GNU_implicit_pointer"), | 
 | 		  uvalue, svalue); | 
 | 	  break; | 
 | 	case DW_OP_entry_value: | 
 | 	case DW_OP_GNU_entry_value: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  /* PR 17531: file: 0cc9cd00.  */ | 
 | 	  if (uvalue > (size_t) (end - data)) | 
 | 	    uvalue = end - data; | 
 | 	  printf ("%s: (", (op == DW_OP_entry_value | 
 | 			    ? "DW_OP_entry_value" : "DW_OP_GNU_entry_value")); | 
 | 	  if (decode_location_expression (data, pointer_size, offset_size, | 
 | 					  dwarf_version, uvalue, | 
 | 					  cu_offset, section)) | 
 | 	    need_frame_base = 1; | 
 | 	  putchar (')'); | 
 | 	  data += uvalue; | 
 | 	  break; | 
 | 	case DW_OP_const_type: | 
 | 	case DW_OP_GNU_const_type: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("%s: <%#" PRIx64 "> ", | 
 | 		  (op == DW_OP_const_type | 
 | 		   ? "DW_OP_const_type" : "DW_OP_GNU_const_type"), | 
 | 		  cu_offset + uvalue); | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  data = display_block (data, uvalue, end, ' '); | 
 | 	  break; | 
 | 	case DW_OP_regval_type: | 
 | 	case DW_OP_GNU_regval_type: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("%s: %" PRIu64 " (%s)", | 
 | 		  (op == DW_OP_regval_type | 
 | 		   ? "DW_OP_regval_type" : "DW_OP_GNU_regval_type"), | 
 | 		  uvalue, regname (uvalue, 1)); | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf (" <%#" PRIx64 ">", cu_offset + uvalue); | 
 | 	  break; | 
 | 	case DW_OP_deref_type: | 
 | 	case DW_OP_GNU_deref_type: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 | 	  printf ("%s: %" PRId64, | 
 | 		  (op == DW_OP_deref_type | 
 | 		   ? "DW_OP_deref_type" : "DW_OP_GNU_deref_type"), | 
 | 		  uvalue); | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf (" <%#" PRIx64 ">", cu_offset + uvalue); | 
 | 	  break; | 
 | 	case DW_OP_convert: | 
 | 	case DW_OP_GNU_convert: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("%s <%#" PRIx64 ">", | 
 | 		  (op == DW_OP_convert ? "DW_OP_convert" : "DW_OP_GNU_convert"), | 
 | 		  uvalue ? cu_offset + uvalue : uvalue); | 
 | 	  break; | 
 | 	case DW_OP_reinterpret: | 
 | 	case DW_OP_GNU_reinterpret: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("%s <%#" PRIx64 ">", | 
 | 		  (op == DW_OP_reinterpret | 
 | 		   ? "DW_OP_reinterpret" : "DW_OP_GNU_reinterpret"), | 
 | 		  uvalue ? cu_offset + uvalue : uvalue); | 
 | 	  break; | 
 | 	case DW_OP_GNU_parameter_ref: | 
 | 	  SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 | 	  printf ("DW_OP_GNU_parameter_ref: <%#" PRIx64 ">", | 
 | 		  cu_offset + uvalue); | 
 | 	  break; | 
 | 	case DW_OP_addrx: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_addrx <%#" PRIx64 ">", uvalue); | 
 | 	  break; | 
 | 	case DW_OP_GNU_addr_index: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_GNU_addr_index <%#" PRIx64 ">", uvalue); | 
 | 	  break; | 
 | 	case DW_OP_GNU_const_index: | 
 | 	  READ_ULEB (uvalue, data, end); | 
 | 	  printf ("DW_OP_GNU_const_index <%#" PRIx64 ">", uvalue); | 
 | 	  break; | 
 | 	case DW_OP_GNU_variable_value: | 
 | 	  /* FIXME: Strictly speaking for 64-bit DWARF3 files | 
 | 	     this ought to be an 8-byte wide computation.  */ | 
 | 	  if (dwarf_version == -1) | 
 | 	    { | 
 | 	      printf (_("(DW_OP_GNU_variable_value in frame info)")); | 
 | 	      /* No way to tell where the next op is, so just bail.  */ | 
 | 	      return need_frame_base; | 
 | 	    } | 
 | 	  if (dwarf_version == 2) | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 | 	    } | 
 | 	  printf ("DW_OP_GNU_variable_value: <%#" PRIx64 ">", uvalue); | 
 | 	  break; | 
 |  | 
 | 	  /* HP extensions.  */ | 
 | 	case DW_OP_HP_is_value: | 
 | 	  printf ("DW_OP_HP_is_value"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_HP_fltconst4: | 
 | 	  printf ("DW_OP_HP_fltconst4"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_HP_fltconst8: | 
 | 	  printf ("DW_OP_HP_fltconst8"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_HP_mod_range: | 
 | 	  printf ("DW_OP_HP_mod_range"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_HP_unmod_range: | 
 | 	  printf ("DW_OP_HP_unmod_range"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 | 	case DW_OP_HP_tls: | 
 | 	  printf ("DW_OP_HP_tls"); | 
 | 	  /* FIXME: Is there data associated with this OP ?  */ | 
 | 	  break; | 
 |  | 
 | 	  /* PGI (STMicroelectronics) extensions.  */ | 
 | 	case DW_OP_PGI_omp_thread_num: | 
 | 	  /* Pushes the thread number for the current thread as it would be | 
 | 	     returned by the standard OpenMP library function: | 
 | 	     omp_get_thread_num().  The "current thread" is the thread for | 
 | 	     which the expression is being evaluated.  */ | 
 | 	  printf ("DW_OP_PGI_omp_thread_num"); | 
 | 	  break; | 
 |  | 
 | 	default: | 
 | 	  if (op >= DW_OP_lo_user | 
 | 	      && op <= DW_OP_hi_user) | 
 | 	    printf (_("(User defined location op %#x)"), op); | 
 | 	  else | 
 | 	    printf (_("(Unknown location op %#x)"), op); | 
 | 	  /* No way to tell where the next op is, so just bail.  */ | 
 | 	  return need_frame_base; | 
 | 	} | 
 |  | 
 |       /* Separate the ops.  */ | 
 |       if (data < end) | 
 | 	printf ("; "); | 
 |     } | 
 |  | 
 |   return need_frame_base; | 
 | } | 
 |  | 
 | /* Find the CU or TU set corresponding to the given CU_OFFSET. | 
 |    This is used for DWARF package files.  */ | 
 |  | 
 | static struct cu_tu_set * | 
 | find_cu_tu_set_v2 (uint64_t cu_offset, int do_types) | 
 | { | 
 |   struct cu_tu_set *p; | 
 |   unsigned int nsets; | 
 |   unsigned int dw_sect; | 
 |  | 
 |   if (do_types) | 
 |     { | 
 |       p = tu_sets; | 
 |       nsets = tu_count; | 
 |       dw_sect = DW_SECT_TYPES; | 
 |     } | 
 |   else | 
 |     { | 
 |       p = cu_sets; | 
 |       nsets = cu_count; | 
 |       dw_sect = DW_SECT_INFO; | 
 |     } | 
 |   while (nsets > 0) | 
 |     { | 
 |       if (p->section_offsets [dw_sect] == cu_offset) | 
 | 	return p; | 
 |       p++; | 
 |       nsets--; | 
 |     } | 
 |   return NULL; | 
 | } | 
 |  | 
 | static const char * | 
 | fetch_alt_indirect_string (uint64_t offset) | 
 | { | 
 |   separate_info * i; | 
 |  | 
 |   if (! do_follow_links) | 
 |     return ""; | 
 |  | 
 |   if (first_separate_info == NULL) | 
 |     return _("<no links available>"); | 
 |  | 
 |   for (i = first_separate_info; i != NULL; i = i->next) | 
 |     { | 
 |       struct dwarf_section * section; | 
 |       const char *           ret; | 
 |  | 
 |       if (! load_debug_section (separate_debug_str, i->handle)) | 
 | 	continue; | 
 |  | 
 |       section = &debug_displays [separate_debug_str].section; | 
 |  | 
 |       if (section->start == NULL) | 
 | 	continue; | 
 |  | 
 |       if (offset >= section->size) | 
 | 	continue; | 
 |  | 
 |       ret = (const char *) (section->start + offset); | 
 |       /* Unfortunately we cannot rely upon the .debug_str section ending with a | 
 | 	 NUL byte.  Since our caller is expecting to receive a well formed C | 
 | 	 string we test for the lack of a terminating byte here.  */ | 
 |       if (strnlen ((const char *) ret, section->size - offset) | 
 | 	  == section->size - offset) | 
 | 	return _("<no NUL byte at end of alt .debug_str section>"); | 
 |  | 
 |       return ret; | 
 |     } | 
 |  | 
 |   warn (_("DW_FORM_GNU_strp_alt offset (%#" PRIx64 ")" | 
 | 	  " too big or no string sections available\n"), offset); | 
 |   return _("<offset is too big>"); | 
 | } | 
 |  | 
 | static const char * | 
 | get_AT_name (unsigned long attribute) | 
 | { | 
 |   const char *name; | 
 |  | 
 |   if (attribute == 0) | 
 |     return "DW_AT value: 0"; | 
 |  | 
 |   /* One value is shared by the MIPS and HP extensions:  */ | 
 |   if (attribute == DW_AT_MIPS_fde) | 
 |     return "DW_AT_MIPS_fde or DW_AT_HP_unmodifiable"; | 
 |  | 
 |   name = get_DW_AT_name (attribute); | 
 |  | 
 |   if (name == NULL) | 
 |     { | 
 |       static char buffer[100]; | 
 |  | 
 |       snprintf (buffer, sizeof (buffer), _("Unknown AT value: %lx"), | 
 | 		attribute); | 
 |       return buffer; | 
 |     } | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static void | 
 | add_dwo_info (const char * value, uint64_t cu_offset, dwo_type type) | 
 | { | 
 |   dwo_info * dwinfo = xmalloc (sizeof * dwinfo); | 
 |  | 
 |   dwinfo->type   = type; | 
 |   dwinfo->value  = value; | 
 |   dwinfo->cu_offset = cu_offset; | 
 |   dwinfo->next   = first_dwo_info; | 
 |   first_dwo_info = dwinfo; | 
 | } | 
 |  | 
 | static void | 
 | add_dwo_name (const char * name, uint64_t cu_offset) | 
 | { | 
 |   add_dwo_info (name, cu_offset, DWO_NAME); | 
 | } | 
 |  | 
 | static void | 
 | add_dwo_dir (const char * dir, uint64_t cu_offset) | 
 | { | 
 |   add_dwo_info (dir, cu_offset, DWO_DIR); | 
 | } | 
 |  | 
 | static void | 
 | add_dwo_id (const char * id, uint64_t cu_offset) | 
 | { | 
 |   add_dwo_info (id, cu_offset, DWO_ID); | 
 | } | 
 |  | 
 | static void | 
 | free_dwo_info (void) | 
 | { | 
 |   dwo_info * dwinfo; | 
 |   dwo_info * next; | 
 |  | 
 |   for (dwinfo = first_dwo_info; dwinfo != NULL; dwinfo = next) | 
 |     { | 
 |       next = dwinfo->next; | 
 |       free (dwinfo); | 
 |     } | 
 |   first_dwo_info = NULL; | 
 | } | 
 |  | 
 | /* Ensure that START + UVALUE is less than END. | 
 |    Return an adjusted UVALUE if necessary to ensure this relationship.  */ | 
 |  | 
 | static inline uint64_t | 
 | check_uvalue (const unsigned char *start, | 
 | 	      uint64_t uvalue, | 
 | 	      const unsigned char *end) | 
 | { | 
 |   uint64_t max_uvalue = end - start; | 
 |  | 
 |   /* See PR 17512: file: 008-103549-0.001:0.1. | 
 |      and PR 24829 for examples of where these tests are triggered.  */ | 
 |   if (uvalue > max_uvalue) | 
 |     { | 
 |       warn (_("Corrupt attribute block length: %#" PRIx64 "\n"), uvalue); | 
 |       uvalue = max_uvalue; | 
 |     } | 
 |  | 
 |   return uvalue; | 
 | } | 
 |  | 
 | static unsigned char * | 
 | skip_attr_bytes (unsigned long form, | 
 | 		 unsigned char *data, | 
 | 		 unsigned char *end, | 
 | 		 uint64_t pointer_size, | 
 | 		 uint64_t offset_size, | 
 | 		 int dwarf_version, | 
 | 		 uint64_t *value_return) | 
 | { | 
 |   int64_t svalue; | 
 |   uint64_t uvalue = 0; | 
 |   uint64_t inc = 0; | 
 |  | 
 |   * value_return = 0; | 
 |  | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_ref_addr: | 
 |       if (dwarf_version == 2) | 
 | 	SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 |       else if (dwarf_version > 2) | 
 | 	SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 |       else | 
 | 	return NULL; | 
 |       break; | 
 |  | 
 |     case DW_FORM_addr: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_strp: | 
 |     case DW_FORM_line_strp: | 
 |     case DW_FORM_sec_offset: | 
 |     case DW_FORM_GNU_ref_alt: | 
 |     case DW_FORM_GNU_strp_alt: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_flag_present: | 
 |       uvalue = 1; | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref1: | 
 |     case DW_FORM_flag: | 
 |     case DW_FORM_data1: | 
 |     case DW_FORM_strx1: | 
 |     case DW_FORM_addrx1: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_strx3: | 
 |     case DW_FORM_addrx3: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 3, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref2: | 
 |     case DW_FORM_data2: | 
 |     case DW_FORM_strx2: | 
 |     case DW_FORM_addrx2: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 2, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref4: | 
 |     case DW_FORM_data4: | 
 |     case DW_FORM_strx4: | 
 |     case DW_FORM_addrx4: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_sdata: | 
 |       READ_SLEB (svalue, data, end); | 
 |       uvalue = svalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_udata: | 
 |     case DW_FORM_udata: | 
 |     case DW_FORM_GNU_str_index: | 
 |     case DW_FORM_strx: | 
 |     case DW_FORM_GNU_addr_index: | 
 |     case DW_FORM_addrx: | 
 |     case DW_FORM_loclistx: | 
 |     case DW_FORM_rnglistx: | 
 |       READ_ULEB (uvalue, data, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref8: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 8, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_data8: | 
 |     case DW_FORM_ref_sig8: | 
 |       inc = 8; | 
 |       break; | 
 |  | 
 |     case DW_FORM_data16: | 
 |       inc = 16; | 
 |       break; | 
 |  | 
 |     case DW_FORM_string: | 
 |       inc = strnlen ((char *) data, end - data) + 1; | 
 |       break; | 
 |  | 
 |     case DW_FORM_block: | 
 |     case DW_FORM_exprloc: | 
 |       READ_ULEB (uvalue, data, end); | 
 |       inc = uvalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_block1: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 |       inc = uvalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_block2: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 2, end); | 
 |       inc = uvalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_block4: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 |       inc = uvalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_indirect: | 
 |       READ_ULEB (form, data, end); | 
 |       if (form == DW_FORM_implicit_const) | 
 | 	SKIP_ULEB (data, end); | 
 |       return skip_attr_bytes (form, data, end, pointer_size, offset_size, | 
 | 			      dwarf_version, value_return); | 
 |  | 
 |     default: | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   * value_return = uvalue; | 
 |   if (inc <= (size_t) (end - data)) | 
 |     data += inc; | 
 |   else | 
 |     data = end; | 
 |   return data; | 
 | } | 
 |  | 
 | /* Given form FORM with value UVALUE, locate and return the abbreviation | 
 |    associated with it.  */ | 
 |  | 
 | static abbrev_entry * | 
 | get_type_abbrev_from_form (unsigned long form, | 
 | 			   uint64_t uvalue, | 
 | 			   uint64_t cu_offset, | 
 | 			   unsigned char *cu_end, | 
 | 			   const struct dwarf_section *section, | 
 | 			   unsigned long *abbrev_num_return, | 
 | 			   unsigned char **data_return, | 
 | 			   abbrev_map **map_return) | 
 | { | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_GNU_ref_alt: | 
 |     case DW_FORM_ref_sig8: | 
 |       /* FIXME: We are unable to handle this form at the moment.  */ | 
 |       return NULL; | 
 |  | 
 |     case DW_FORM_ref_addr: | 
 |       if (uvalue >= section->size) | 
 | 	{ | 
 | 	  warn (_("Unable to resolve ref_addr form: uvalue %" PRIx64 | 
 | 		  " >= section size %" PRIx64 " (%s)\n"), | 
 | 		uvalue, section->size, section->name); | 
 | 	  return NULL; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_sup4: | 
 |     case DW_FORM_ref_sup8: | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref1: | 
 |     case DW_FORM_ref2: | 
 |     case DW_FORM_ref4: | 
 |     case DW_FORM_ref8: | 
 |     case DW_FORM_ref_udata: | 
 |       if (uvalue + cu_offset < uvalue | 
 | 	  || uvalue + cu_offset > (size_t) (cu_end - section->start)) | 
 | 	{ | 
 | 	  warn (_("Unable to resolve ref form: uvalue %" PRIx64 | 
 | 		  " + cu_offset %" PRIx64 " > CU size %tx\n"), | 
 | 		uvalue, cu_offset, cu_end - section->start); | 
 | 	  return NULL; | 
 | 	} | 
 |       uvalue += cu_offset; | 
 |       break; | 
 |  | 
 |       /* FIXME: Are there other DW_FORMs that can be used by types ?  */ | 
 |  | 
 |     default: | 
 |       warn (_("Unexpected form %lx encountered whilst finding abbreviation for type\n"), form); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   abbrev_map *map = find_abbrev_map_by_offset (uvalue); | 
 |  | 
 |   if (map == NULL) | 
 |     { | 
 |       warn (_("Unable to find abbreviations for CU offset %" PRIx64 "\n"), | 
 | 	    uvalue); | 
 |       return NULL; | 
 |     } | 
 |   if (map->list == NULL) | 
 |     { | 
 |       warn (_("Empty abbreviation list encountered for CU offset %" PRIx64 "\n"), | 
 | 	    uvalue); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   if (map_return != NULL) | 
 |     { | 
 |       if (form == DW_FORM_ref_addr) | 
 | 	*map_return = map; | 
 |       else | 
 | 	*map_return = NULL; | 
 |     } | 
 |  | 
 |   unsigned char *data = section->start + uvalue; | 
 |   if (form == DW_FORM_ref_addr) | 
 |     cu_end = section->start + map->end; | 
 |  | 
 |   unsigned long abbrev_number; | 
 |   READ_ULEB (abbrev_number, data, cu_end); | 
 |  | 
 |   if (abbrev_num_return != NULL) | 
 |     *abbrev_num_return = abbrev_number; | 
 |  | 
 |   if (data_return != NULL) | 
 |     *data_return = data; | 
 |  | 
 |   abbrev_entry *entry; | 
 |   for (entry = map->list->first_abbrev; entry != NULL; entry = entry->next) | 
 |     if (entry->number == abbrev_number) | 
 |       break; | 
 |  | 
 |   if (entry == NULL) | 
 |     warn (_("Unable to find entry for abbreviation %lu\n"), abbrev_number); | 
 |  | 
 |   return entry; | 
 | } | 
 |  | 
 | /* Return IS_SIGNED set to TRUE if the type using abbreviation ENTRY | 
 |    can be determined to be a signed type.  The data for ENTRY can be | 
 |    found starting at DATA.  */ | 
 |  | 
 | static void | 
 | get_type_signedness (abbrev_entry *entry, | 
 | 		     const struct dwarf_section *section, | 
 | 		     unsigned char *data, | 
 | 		     unsigned char *end, | 
 | 		     uint64_t cu_offset, | 
 | 		     uint64_t pointer_size, | 
 | 		     uint64_t offset_size, | 
 | 		     int dwarf_version, | 
 | 		     bool *is_signed, | 
 | 		     unsigned int nesting) | 
 | { | 
 |   abbrev_attr *   attr; | 
 |  | 
 |   * is_signed = false; | 
 |  | 
 | #define MAX_NESTING 20 | 
 |   if (nesting > MAX_NESTING) | 
 |     { | 
 |       /* FIXME: Warn - or is this expected ? | 
 | 	 NB/ We need to avoid infinite recursion.  */ | 
 |       return; | 
 |     } | 
 |  | 
 |   for (attr = entry->first_attr; | 
 |        attr != NULL && attr->attribute; | 
 |        attr = attr->next) | 
 |     { | 
 |       unsigned char * orig_data = data; | 
 |       uint64_t uvalue = 0; | 
 |  | 
 |       data = skip_attr_bytes (attr->form, data, end, pointer_size, | 
 | 			      offset_size, dwarf_version, & uvalue); | 
 |       if (data == NULL) | 
 | 	return; | 
 |  | 
 |       switch (attr->attribute) | 
 | 	{ | 
 | 	case DW_AT_linkage_name: | 
 | 	case DW_AT_name: | 
 | 	  if (do_wide) | 
 | 	    { | 
 | 	      if (attr->form == DW_FORM_strp) | 
 | 		printf (", %s", fetch_indirect_string (uvalue)); | 
 | 	      else if (attr->form == DW_FORM_string) | 
 | 		printf (", %.*s", (int) (end - orig_data), orig_data); | 
 | 	    } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_type: | 
 | 	  /* Recurse.  */ | 
 | 	  { | 
 | 	    abbrev_entry *type_abbrev; | 
 | 	    unsigned char *type_data; | 
 | 	    abbrev_map *map; | 
 |  | 
 | 	    type_abbrev = get_type_abbrev_from_form (attr->form, | 
 | 						     uvalue, | 
 | 						     cu_offset, | 
 | 						     end, | 
 | 						     section, | 
 | 						     NULL /* abbrev num return */, | 
 | 						     &type_data, | 
 | 						     &map); | 
 | 	    if (type_abbrev == NULL) | 
 | 	      break; | 
 |  | 
 | 	    get_type_signedness (type_abbrev, section, type_data, | 
 | 				 map ? section->start + map->end : end, | 
 | 				 map ? map->start : cu_offset, | 
 | 				 pointer_size, offset_size, dwarf_version, | 
 | 				 is_signed, nesting + 1); | 
 | 	  } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_encoding: | 
 | 	  /* Determine signness.  */ | 
 | 	  switch (uvalue) | 
 | 	    { | 
 | 	    case DW_ATE_address: | 
 | 	      /* FIXME - some architectures have signed addresses.  */ | 
 | 	    case DW_ATE_boolean: | 
 | 	    case DW_ATE_unsigned: | 
 | 	    case DW_ATE_unsigned_char: | 
 | 	    case DW_ATE_unsigned_fixed: | 
 | 	      * is_signed = false; | 
 | 	      break; | 
 |  | 
 | 	    default: | 
 | 	    case DW_ATE_complex_float: | 
 | 	    case DW_ATE_float: | 
 | 	    case DW_ATE_signed: | 
 | 	    case DW_ATE_signed_char: | 
 | 	    case DW_ATE_imaginary_float: | 
 | 	    case DW_ATE_decimal_float: | 
 | 	    case DW_ATE_signed_fixed: | 
 | 	      * is_signed = true; | 
 | 	      break; | 
 | 	    } | 
 | 	  break; | 
 | 	} | 
 |     } | 
 | } | 
 |  | 
 | static void | 
 | read_and_print_leb128 (unsigned char *data, | 
 | 		       unsigned int *bytes_read, | 
 | 		       unsigned const char *end, | 
 | 		       bool is_signed) | 
 | { | 
 |   int status; | 
 |   uint64_t val = read_leb128 (data, end, is_signed, bytes_read, &status); | 
 |   if (status != 0) | 
 |     report_leb_status (status); | 
 |   else if (is_signed) | 
 |     printf ("%" PRId64, val); | 
 |   else | 
 |     printf ("%" PRIu64, val); | 
 | } | 
 |  | 
 | static void | 
 | display_discr_list (unsigned long form, | 
 | 		    uint64_t uvalue, | 
 | 		    unsigned char *data, | 
 | 		    int level) | 
 | { | 
 |   unsigned char *end = data; | 
 |  | 
 |   if (uvalue == 0) | 
 |     { | 
 |       printf ("[default]"); | 
 |       return; | 
 |     } | 
 |  | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_block: | 
 |     case DW_FORM_block1: | 
 |     case DW_FORM_block2: | 
 |     case DW_FORM_block4: | 
 |       /* Move data pointer back to the start of the byte array.  */ | 
 |       data -= uvalue; | 
 |       break; | 
 |     default: | 
 |       printf ("<corrupt>\n"); | 
 |       warn (_("corrupt discr_list - not using a block form\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   if (uvalue < 2) | 
 |     { | 
 |       printf ("<corrupt>\n"); | 
 |       warn (_("corrupt discr_list - block not long enough\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   bool is_signed = (level > 0 && level <= MAX_CU_NESTING | 
 | 		    ? level_type_signed [level - 1] : false); | 
 |  | 
 |   printf ("("); | 
 |   while (data < end) | 
 |     { | 
 |       unsigned char     discriminant; | 
 |       unsigned int      bytes_read; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (discriminant, data, 1, end); | 
 |  | 
 |       switch (discriminant) | 
 | 	{ | 
 | 	case DW_DSC_label: | 
 | 	  printf ("label "); | 
 | 	  read_and_print_leb128 (data, & bytes_read, end, is_signed); | 
 | 	  data += bytes_read; | 
 | 	  break; | 
 |  | 
 | 	case DW_DSC_range: | 
 | 	  printf ("range "); | 
 | 	  read_and_print_leb128 (data, & bytes_read, end, is_signed); | 
 | 	  data += bytes_read; | 
 |  | 
 | 	  printf (".."); | 
 | 	  read_and_print_leb128 (data, & bytes_read, end, is_signed); | 
 | 	  data += bytes_read; | 
 | 	  break; | 
 |  | 
 | 	default: | 
 | 	  printf ("<corrupt>\n"); | 
 | 	  warn (_("corrupt discr_list - unrecognized discriminant byte %#x\n"), | 
 | 		discriminant); | 
 | 	  return; | 
 | 	} | 
 |  | 
 |       if (data < end) | 
 | 	printf (", "); | 
 |     } | 
 |  | 
 |   if (is_signed) | 
 |     printf (")(signed)"); | 
 |   else | 
 |     printf (")(unsigned)"); | 
 | } | 
 |  | 
 | static void | 
 | display_lang (uint64_t uvalue) | 
 | { | 
 |   switch (uvalue) | 
 |     { | 
 |       /* Ordered by the numeric value of these constants.  */ | 
 |     case DW_LANG_C89:			printf ("ANSI C"); break; | 
 |     case DW_LANG_C:			printf ("non-ANSI C"); break; | 
 |     case DW_LANG_Ada83:			printf ("Ada"); break; | 
 |     case DW_LANG_C_plus_plus:		printf ("C++"); break; | 
 |     case DW_LANG_Cobol74:		printf ("Cobol 74"); break; | 
 |     case DW_LANG_Cobol85:		printf ("Cobol 85"); break; | 
 |     case DW_LANG_Fortran77:		printf ("FORTRAN 77"); break; | 
 |     case DW_LANG_Fortran90:		printf ("Fortran 90"); break; | 
 |     case DW_LANG_Pascal83:		printf ("ANSI Pascal"); break; | 
 |     case DW_LANG_Modula2:		printf ("Modula 2"); break; | 
 |  | 
 |       /* DWARF 2.1 values.	*/ | 
 |     case DW_LANG_Java:			printf ("Java"); break; | 
 |     case DW_LANG_C99:			printf ("ANSI C99"); break; | 
 |     case DW_LANG_Ada95:			printf ("ADA 95"); break; | 
 |     case DW_LANG_Fortran95:		printf ("Fortran 95"); break; | 
 |  | 
 |       /* DWARF 3 values.  */ | 
 |     case DW_LANG_PLI:			printf ("PLI"); break; | 
 |     case DW_LANG_ObjC:			printf ("Objective C"); break; | 
 |     case DW_LANG_ObjC_plus_plus:	printf ("Objective C++"); break; | 
 |     case DW_LANG_UPC:			printf ("Unified Parallel C"); break; | 
 |     case DW_LANG_D:			printf ("D"); break; | 
 |  | 
 |       /* DWARF 4 values.  */ | 
 |     case DW_LANG_Python:		printf ("Python"); break; | 
 |  | 
 |       /* DWARF 5 values.  */ | 
 |     case DW_LANG_OpenCL:		printf ("OpenCL"); break; | 
 |     case DW_LANG_Go:			printf ("Go"); break; | 
 |     case DW_LANG_Modula3:		printf ("Modula 3"); break; | 
 |     case DW_LANG_Haskell:		printf ("Haskell"); break; | 
 |     case DW_LANG_C_plus_plus_03:	printf ("C++03"); break; | 
 |     case DW_LANG_C_plus_plus_11:	printf ("C++11"); break; | 
 |     case DW_LANG_OCaml:			printf ("OCaml"); break; | 
 |     case DW_LANG_Rust:			printf ("Rust"); break; | 
 |     case DW_LANG_C11:			printf ("C11"); break; | 
 |     case DW_LANG_Swift:			printf ("Swift"); break; | 
 |     case DW_LANG_Julia:			printf ("Julia"); break; | 
 |     case DW_LANG_Dylan:			printf ("Dylan"); break; | 
 |     case DW_LANG_C_plus_plus_14:	printf ("C++14"); break; | 
 |     case DW_LANG_Fortran03:		printf ("Fortran 03"); break; | 
 |     case DW_LANG_Fortran08:		printf ("Fortran 08"); break; | 
 |     case DW_LANG_RenderScript:		printf ("RenderScript"); break; | 
 |     case DW_LANG_C17:                   printf ("C17"); break; | 
 |     case DW_LANG_Fortran18:             printf ("Fortran 18"); break; | 
 |     case DW_LANG_Ada2005:               printf ("Ada 2005"); break; | 
 |     case DW_LANG_Ada2012:               printf ("Ada 2012"); break; | 
 |     case DW_LANG_HIP:                   printf ("Hip"); break; | 
 |     case DW_LANG_Assembly:              printf ("Assembler"); break; | 
 |     case DW_LANG_C_sharp:               printf ("C Sharp"); break; | 
 |     case DW_LANG_Mojo:                  printf ("Mojo"); break; | 
 |     case DW_LANG_GLSL:                  printf ("GLSL"); break; | 
 |     case DW_LANG_GLSL_ES:               printf ("GLSL_ES"); break; | 
 |     case DW_LANG_HLSL:                  printf ("HLSL"); break; | 
 |     case DW_LANG_OpenCL_CPP:            printf ("OpenCL C++"); break; | 
 |     case DW_LANG_CPP_for_OpenCL:        printf ("C++ for OpenCL"); break; | 
 |     case DW_LANG_SYCL:                  printf ("SYCL"); break; | 
 |     case DW_LANG_C_plus_plus_17:        printf ("C++17"); break; | 
 |     case DW_LANG_C_plus_plus_20:        printf ("C++20"); break; | 
 |     case DW_LANG_C_plus_plus_23:	printf ("C++23"); break; | 
 |     case DW_LANG_Odin:                  printf ("Odin"); break; | 
 |     case DW_LANG_P4:                    printf ("P4"); break; | 
 |     case DW_LANG_Metal:                 printf ("C23"); break; | 
 |     case DW_LANG_C23:                   printf ("C23"); break; | 
 |     case DW_LANG_Fortran23:             printf ("Fortran 23"); break; | 
 |     case DW_LANG_Ruby:                  printf ("Ruby"); break; | 
 |     case DW_LANG_Move:                  printf ("Move"); break; | 
 |     case DW_LANG_Hylo:                  printf ("Hylo"); break; | 
 |  | 
 |       /* MIPS extension.  */ | 
 |     case DW_LANG_Mips_Assembler:	printf ("MIPS assembler"); break; | 
 |  | 
 |       /* UPC extension.  */ | 
 |     case DW_LANG_Upc:			printf ("Unified Parallel C"); break; | 
 |  | 
 |     default: | 
 |       if (uvalue >= DW_LANG_lo_user && uvalue <= DW_LANG_hi_user) | 
 | 	printf (_("implementation defined: %#" PRIx64 ""), uvalue); | 
 |       else | 
 | 	printf (_("unknown: %#" PRIx64 ""), uvalue); | 
 |       break; | 
 |     } | 
 | } | 
 |  | 
 | static unsigned char * | 
 | read_and_display_attr_value (unsigned long attribute, | 
 | 			     unsigned long form, | 
 | 			     int64_t implicit_const, | 
 | 			     unsigned char *start, | 
 | 			     unsigned char *data, | 
 | 			     unsigned char *end, | 
 | 			     uint64_t cu_offset, | 
 | 			     uint64_t pointer_size, | 
 | 			     uint64_t offset_size, | 
 | 			     int dwarf_version, | 
 | 			     debug_info *debug_info_p, | 
 | 			     int do_loc, | 
 | 			     struct dwarf_section *section, | 
 | 			     struct cu_tu_set *this_set, | 
 | 			     char delimiter, | 
 | 			     int level) | 
 | { | 
 |   int64_t svalue; | 
 |   uint64_t uvalue = 0; | 
 |   uint64_t uvalue_hi = 0; | 
 |   unsigned char *block_start = NULL; | 
 |   unsigned char *orig_data = data; | 
 |  | 
 |   if (data > end || (data == end && form != DW_FORM_flag_present)) | 
 |     { | 
 |       warn (_("Corrupt attribute\n")); | 
 |       return data; | 
 |     } | 
 |  | 
 |   if (do_wide && ! do_loc) | 
 |     { | 
 |       /* PR 26847: Display the name of the form.  */ | 
 |       const char * name = get_FORM_name (form); | 
 |  | 
 |       /* For convenience we skip the DW_FORM_ prefix to the name.  */ | 
 |       if (name[0] == 'D') | 
 | 	name += 8; /* strlen ("DW_FORM_")  */ | 
 |       printf ("%c(%s)", delimiter, name); | 
 |     } | 
 |  | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_ref_addr: | 
 |       if (dwarf_version == 2) | 
 | 	SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 |       else if (dwarf_version > 2) | 
 | 	SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 |       else | 
 | 	error (_("Internal error: DW_FORM_ref_addr is not supported in DWARF version 1.\n")); | 
 |       break; | 
 |  | 
 |     case DW_FORM_addr: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, pointer_size, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_strp_sup: | 
 |     case DW_FORM_strp: | 
 |     case DW_FORM_line_strp: | 
 |     case DW_FORM_sec_offset: | 
 |     case DW_FORM_GNU_ref_alt: | 
 |     case DW_FORM_GNU_strp_alt: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_flag_present: | 
 |       uvalue = 1; | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref1: | 
 |     case DW_FORM_flag: | 
 |     case DW_FORM_data1: | 
 |     case DW_FORM_strx1: | 
 |     case DW_FORM_addrx1: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref2: | 
 |     case DW_FORM_data2: | 
 |     case DW_FORM_strx2: | 
 |     case DW_FORM_addrx2: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 2, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_strx3: | 
 |     case DW_FORM_addrx3: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 3, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_sup4: | 
 |     case DW_FORM_ref4: | 
 |     case DW_FORM_data4: | 
 |     case DW_FORM_strx4: | 
 |     case DW_FORM_addrx4: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_sup8: | 
 |     case DW_FORM_ref8: | 
 |     case DW_FORM_data8: | 
 |     case DW_FORM_ref_sig8: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 8, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_data16: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 8, end); | 
 |       SAFE_BYTE_GET_AND_INC (uvalue_hi, data, 8, end); | 
 |       if (byte_get != byte_get_little_endian) | 
 | 	{ | 
 | 	  uint64_t utmp = uvalue; | 
 | 	  uvalue = uvalue_hi; | 
 | 	  uvalue_hi = utmp; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_sdata: | 
 |       READ_SLEB (svalue, data, end); | 
 |       uvalue = svalue; | 
 |       break; | 
 |  | 
 |     case DW_FORM_GNU_str_index: | 
 |     case DW_FORM_strx: | 
 |     case DW_FORM_ref_udata: | 
 |     case DW_FORM_udata: | 
 |     case DW_FORM_GNU_addr_index: | 
 |     case DW_FORM_addrx: | 
 |     case DW_FORM_loclistx: | 
 |     case DW_FORM_rnglistx: | 
 |       READ_ULEB (uvalue, data, end); | 
 |       break; | 
 |  | 
 |     case DW_FORM_indirect: | 
 |       READ_ULEB (form, data, end); | 
 |       if (!do_loc) | 
 | 	printf ("%c%s", delimiter, get_FORM_name (form)); | 
 |       if (form == DW_FORM_implicit_const) | 
 | 	READ_SLEB (implicit_const, data, end); | 
 |       return read_and_display_attr_value (attribute, form, implicit_const, | 
 | 					  start, data, end, | 
 | 					  cu_offset, pointer_size, | 
 | 					  offset_size, dwarf_version, | 
 | 					  debug_info_p, do_loc, | 
 | 					  section, this_set, delimiter, level); | 
 |  | 
 |     case DW_FORM_implicit_const: | 
 |       uvalue = implicit_const; | 
 |       break; | 
 |  | 
 |     default: | 
 |       break; | 
 |     } | 
 |  | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_ref_addr: | 
 |       if (!do_loc) | 
 | 	printf ("%c<%#" PRIx64 ">", delimiter, uvalue); | 
 |       break; | 
 |  | 
 |     case DW_FORM_GNU_ref_alt: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  if (do_wide) | 
 | 	    /* We have already printed the form name.  */ | 
 | 	    printf ("%c<%#" PRIx64 ">", delimiter, uvalue); | 
 | 	  else | 
 | 	    printf ("%c<alt %#" PRIx64 ">", delimiter, uvalue); | 
 | 	} | 
 |       /* FIXME: Follow the reference...  */ | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref1: | 
 |     case DW_FORM_ref2: | 
 |     case DW_FORM_ref4: | 
 |     case DW_FORM_ref_sup4: | 
 |     case DW_FORM_ref_udata: | 
 |       if (!do_loc) | 
 | 	printf ("%c<%#" PRIx64 ">", delimiter, uvalue + cu_offset); | 
 |       break; | 
 |  | 
 |     case DW_FORM_data4: | 
 |     case DW_FORM_addr: | 
 |     case DW_FORM_sec_offset: | 
 |       if (!do_loc) | 
 | 	printf ("%c%#" PRIx64, delimiter, uvalue); | 
 |       break; | 
 |  | 
 |     case DW_FORM_flag_present: | 
 |     case DW_FORM_flag: | 
 |     case DW_FORM_data1: | 
 |     case DW_FORM_data2: | 
 |     case DW_FORM_sdata: | 
 |       if (!do_loc) | 
 | 	printf ("%c%" PRId64, delimiter, uvalue); | 
 |       break; | 
 |  | 
 |     case DW_FORM_udata: | 
 |       if (!do_loc) | 
 | 	printf ("%c%" PRIu64, delimiter, uvalue); | 
 |       break; | 
 |  | 
 |     case DW_FORM_implicit_const: | 
 |       if (!do_loc) | 
 | 	printf ("%c%" PRId64, delimiter, implicit_const); | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_sup8: | 
 |     case DW_FORM_ref8: | 
 |     case DW_FORM_data8: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  uint64_t utmp = uvalue; | 
 | 	  if (form == DW_FORM_ref8) | 
 | 	    utmp += cu_offset; | 
 | 	  printf ("%c%#" PRIx64, delimiter, utmp); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_data16: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  if (uvalue_hi == 0) | 
 | 	    printf (" %#" PRIx64, uvalue); | 
 | 	  else | 
 | 	    printf (" %#" PRIx64 "%016" PRIx64, uvalue_hi, uvalue); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_string: | 
 |       if (!do_loc) | 
 | 	printf ("%c%.*s", delimiter, (int) (end - data), data); | 
 |       data += strnlen ((char *) data, end - data); | 
 |       if (data < end) | 
 | 	data++; | 
 |       break; | 
 |  | 
 |     case DW_FORM_block: | 
 |     case DW_FORM_exprloc: | 
 |       READ_ULEB (uvalue, data, end); | 
 |     do_block: | 
 |       block_start = data; | 
 |       if (block_start >= end) | 
 | 	{ | 
 | 	  warn (_("Block ends prematurely\n")); | 
 | 	  uvalue = 0; | 
 | 	  block_start = end; | 
 | 	} | 
 |  | 
 |       uvalue = check_uvalue (block_start, uvalue, end); | 
 |  | 
 |       data = block_start + uvalue; | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  unsigned char op; | 
 |  | 
 | 	  SAFE_BYTE_GET (op, block_start, sizeof (op), end); | 
 | 	  if (op != DW_OP_addrx) | 
 | 	    data = display_block (block_start, uvalue, end, delimiter); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_block1: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 1, end); | 
 |       goto do_block; | 
 |  | 
 |     case DW_FORM_block2: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 2, end); | 
 |       goto do_block; | 
 |  | 
 |     case DW_FORM_block4: | 
 |       SAFE_BYTE_GET_AND_INC (uvalue, data, 4, end); | 
 |       goto do_block; | 
 |  | 
 |     case DW_FORM_strp: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  if (do_wide) | 
 | 	    /* We have already displayed the form name.  */ | 
 | 	    printf (_("%c(offset: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, fetch_indirect_string (uvalue)); | 
 | 	  else | 
 | 	    printf (_("%c(indirect string, offset: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, fetch_indirect_string (uvalue)); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_line_strp: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  if (do_wide) | 
 | 	    /* We have already displayed the form name.  */ | 
 | 	    printf (_("%c(offset: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, fetch_indirect_line_string (uvalue)); | 
 | 	  else | 
 | 	    printf (_("%c(indirect line string, offset: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, fetch_indirect_line_string (uvalue)); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_GNU_str_index: | 
 |     case DW_FORM_strx: | 
 |     case DW_FORM_strx1: | 
 |     case DW_FORM_strx2: | 
 |     case DW_FORM_strx3: | 
 |     case DW_FORM_strx4: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  const char *suffix = section ? strrchr (section->name, '.') : NULL; | 
 | 	  bool dwo = suffix && strcmp (suffix, ".dwo") == 0; | 
 | 	  const char *strng; | 
 |  | 
 | 	  strng = fetch_indexed_string (uvalue, this_set, offset_size, dwo, | 
 | 					debug_info_p ? debug_info_p->str_offsets_base : 0); | 
 | 	  if (do_wide) | 
 | 	    /* We have already displayed the form name.  */ | 
 | 	    printf (_("%c(offset: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, strng); | 
 | 	  else | 
 | 	    printf (_("%c(indexed string: %#" PRIx64 "): %s"), | 
 | 		    delimiter, uvalue, strng); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_GNU_strp_alt: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  if (do_wide) | 
 | 	    /* We have already displayed the form name.  */ | 
 | 	    printf (_("%c(offset: %#" PRIx64 ") %s"), | 
 | 		    delimiter, uvalue, fetch_alt_indirect_string (uvalue)); | 
 | 	  else | 
 | 	    printf (_("%c(alt indirect string, offset: %#" PRIx64 ") %s"), | 
 | 		    delimiter, uvalue, fetch_alt_indirect_string (uvalue)); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_indirect: | 
 |       /* Handled above.  */ | 
 |       break; | 
 |  | 
 |     case DW_FORM_ref_sig8: | 
 |       if (!do_loc) | 
 | 	printf ("%c%s: %#" PRIx64, delimiter, do_wide ? "" : "signature", | 
 | 		uvalue); | 
 |       break; | 
 |  | 
 |     case DW_FORM_GNU_addr_index: | 
 |     case DW_FORM_addrx: | 
 |     case DW_FORM_addrx1: | 
 |     case DW_FORM_addrx2: | 
 |     case DW_FORM_addrx3: | 
 |     case DW_FORM_addrx4: | 
 |     case DW_FORM_loclistx: | 
 |     case DW_FORM_rnglistx: | 
 |       if (!do_loc) | 
 | 	{ | 
 | 	  uint64_t base, idx; | 
 | 	  const char *suffix = strrchr (section->name, '.'); | 
 | 	  bool dwo = suffix && strcmp (suffix, ".dwo") == 0; | 
 |  | 
 | 	  if (form == DW_FORM_loclistx) | 
 | 	    { | 
 | 	      if (debug_info_p == NULL) | 
 | 		idx = -1; | 
 | 	      else if (dwo) | 
 | 		{ | 
 | 		  idx = fetch_indexed_offset (uvalue, loclists_dwo, | 
 | 					      debug_info_p->loclists_base, | 
 | 					      debug_info_p->offset_size); | 
 | 		  if (idx != (uint64_t) -1) | 
 | 		    idx += (offset_size == 8) ? 20 : 12; | 
 | 		} | 
 | 	      else if (dwarf_version > 4) | 
 | 		{ | 
 | 		  idx = fetch_indexed_offset (uvalue, loclists, | 
 | 					      debug_info_p->loclists_base, | 
 | 					      debug_info_p->offset_size); | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  /* We want to compute: | 
 | 		       idx = fetch_indexed_value (uvalue, loclists, | 
 | 		                             debug_info_p->loclists_base); | 
 | 		       idx += debug_info_p->loclists_base; | 
 | 		      Fortunately we already have that sum cached in the | 
 | 		      loc_offsets array.  */ | 
 | 		  if (uvalue < debug_info_p->num_loc_offsets) | 
 | 		    idx = debug_info_p->loc_offsets [uvalue]; | 
 | 		  else | 
 | 		    { | 
 | 		      warn (_("loc_offset %" PRIu64 " too big\n"), uvalue); | 
 | 		      idx = -1; | 
 | 		    } | 
 | 		} | 
 | 	    } | 
 | 	  else if (form == DW_FORM_rnglistx) | 
 | 	    { | 
 | 	      if (debug_info_p == NULL) | 
 | 		idx = -1; | 
 | 	      else | 
 | 		idx = fetch_indexed_offset (uvalue, | 
 | 					    dwo ? rnglists_dwo : rnglists, | 
 | 					    debug_info_p->rnglists_base, | 
 | 					    debug_info_p->offset_size); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      if (debug_info_p == NULL) | 
 | 		base = 0; | 
 | 	      else if (debug_info_p->addr_base == DEBUG_INFO_UNAVAILABLE) | 
 | 		base = 0; | 
 | 	      else | 
 | 		base = debug_info_p->addr_base; | 
 |  | 
 | 	      base += uvalue * pointer_size; | 
 | 	      idx = fetch_indexed_addr (base, pointer_size); | 
 | 	    } | 
 |  | 
 | 	  /* We have already displayed the form name.  */ | 
 | 	  if (idx != (uint64_t) -1) | 
 | 	    printf (_("%c(index: %#" PRIx64 "): %#" PRIx64), | 
 | 		    delimiter, uvalue, idx); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_FORM_strp_sup: | 
 |       if (!do_loc) | 
 | 	printf ("%c<%#" PRIx64 ">", delimiter, uvalue + cu_offset); | 
 |       break; | 
 |  | 
 |     default: | 
 |       warn (_("Unrecognized form: %#lx\n"), form); | 
 |       /* What to do?  Consume a byte maybe?  */ | 
 |       ++data; | 
 |       break; | 
 |     } | 
 |  | 
 |   if ((do_loc || do_debug_loc || do_debug_ranges || do_debug_info) | 
 |       && num_debug_info_entries == 0 | 
 |       && debug_info_p != NULL) | 
 |     { | 
 |       switch (attribute) | 
 | 	{ | 
 | 	case DW_AT_loclists_base: | 
 | 	  if (debug_info_p->loclists_base) | 
 | 	    warn (_("CU @ %#" PRIx64 " has multiple loclists_base values " | 
 | 		    "(%#" PRIx64 " and %#" PRIx64 ")\n"), | 
 | 		  debug_info_p->cu_offset, | 
 | 		  debug_info_p->loclists_base, uvalue); | 
 | 	  svalue = uvalue; | 
 | 	  if (svalue < 0) | 
 | 	    { | 
 | 	      warn (_("CU @ %#" PRIx64 " has has a negative loclists_base " | 
 | 		      "value of %#" PRIx64 " - treating as zero\n"), | 
 | 		    debug_info_p->cu_offset, svalue); | 
 | 	      uvalue = 0; | 
 | 	    } | 
 | 	  debug_info_p->loclists_base = uvalue; | 
 | 	  break; | 
 |  | 
 |  	case DW_AT_rnglists_base: | 
 | 	  /* Assignment to debug_info_p->rnglists_base is now elsewhere.  */ | 
 |  	  break; | 
 |  | 
 | 	case DW_AT_str_offsets_base: | 
 | 	  if (debug_info_p->str_offsets_base) | 
 | 	    warn (_("CU @ %#" PRIx64 " has multiple str_offsets_base values " | 
 | 		    "%#" PRIx64 " and %#" PRIx64 ")\n"), | 
 | 		  debug_info_p->cu_offset, | 
 | 		  debug_info_p->str_offsets_base, uvalue); | 
 | 	  svalue = uvalue; | 
 | 	  if (svalue < 0) | 
 | 	    { | 
 | 	      warn (_("CU @ %#" PRIx64 " has has a negative stroffsets_base " | 
 | 		      "value of %#" PRIx64 " - treating as zero\n"), | 
 | 		    debug_info_p->cu_offset, svalue); | 
 | 	      uvalue = 0; | 
 | 	    } | 
 | 	  debug_info_p->str_offsets_base = uvalue; | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_frame_base: | 
 | 	  /* This is crude; the have_frame_base is reset on the next | 
 | 	     subprogram, not at the end of the current topmost one.  */ | 
 | 	  have_frame_base = 1; | 
 | 	  frame_base_level = level; | 
 | 	  /* Fall through.  */ | 
 | 	case DW_AT_location: | 
 | 	case DW_AT_GNU_locviews: | 
 | 	case DW_AT_string_length: | 
 | 	case DW_AT_return_addr: | 
 | 	case DW_AT_data_member_location: | 
 | 	case DW_AT_vtable_elem_location: | 
 | 	case DW_AT_segment: | 
 | 	case DW_AT_static_link: | 
 | 	case DW_AT_use_location: | 
 | 	case DW_AT_call_value: | 
 | 	case DW_AT_GNU_call_site_value: | 
 | 	case DW_AT_call_data_value: | 
 | 	case DW_AT_GNU_call_site_data_value: | 
 | 	case DW_AT_call_target: | 
 | 	case DW_AT_GNU_call_site_target: | 
 | 	case DW_AT_call_target_clobbered: | 
 | 	case DW_AT_GNU_call_site_target_clobbered: | 
 | 	  if ((dwarf_version < 4 | 
 | 	       && (form == DW_FORM_data4 || form == DW_FORM_data8)) | 
 | 	      || form == DW_FORM_sec_offset | 
 | 	      || form == DW_FORM_loclistx) | 
 | 	    { | 
 | 	      /* Process location list.  */ | 
 | 	      unsigned int lmax = debug_info_p->max_loc_offsets; | 
 | 	      unsigned int num = debug_info_p->num_loc_offsets; | 
 |  | 
 | 	      if (lmax == 0 || num >= lmax) | 
 | 		{ | 
 | 		  lmax += 1024; | 
 | 		  debug_info_p->loc_offsets = (uint64_t *) | 
 | 		    xcrealloc (debug_info_p->loc_offsets, | 
 | 			       lmax, sizeof (*debug_info_p->loc_offsets)); | 
 | 		  debug_info_p->loc_views = (uint64_t *) | 
 | 		    xcrealloc (debug_info_p->loc_views, | 
 | 			       lmax, sizeof (*debug_info_p->loc_views)); | 
 | 		  debug_info_p->have_frame_base = (int *) | 
 | 		    xcrealloc (debug_info_p->have_frame_base, | 
 | 			       lmax, sizeof (*debug_info_p->have_frame_base)); | 
 | 		  debug_info_p->max_loc_offsets = lmax; | 
 | 		} | 
 | 	      if (form == DW_FORM_loclistx) | 
 | 		uvalue = fetch_indexed_offset (num, loclists, | 
 | 					       debug_info_p->loclists_base, | 
 | 					       debug_info_p->offset_size); | 
 | 	      else if (this_set != NULL) | 
 | 		uvalue += this_set->section_offsets [DW_SECT_LOC]; | 
 |  | 
 | 	      debug_info_p->have_frame_base [num] = have_frame_base; | 
 | 	      if (attribute != DW_AT_GNU_locviews) | 
 | 		{ | 
 | 		  /* Corrupt DWARF info can produce more offsets than views. | 
 | 		     See PR 23062 for an example.  */ | 
 | 		  if (debug_info_p->num_loc_offsets | 
 | 		      > debug_info_p->num_loc_views) | 
 | 		    warn (_("More location offset attributes than DW_AT_GNU_locview attributes\n")); | 
 | 		  else | 
 | 		    { | 
 | 		      debug_info_p->loc_offsets [num] = uvalue; | 
 | 		      debug_info_p->num_loc_offsets++; | 
 | 		    } | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  if (debug_info_p->num_loc_views > num) | 
 | 		    { | 
 | 		      warn (_("The number of views (%u) is greater than the number of locations (%u)\n"), | 
 | 			    debug_info_p->num_loc_views, num); | 
 | 		      debug_info_p->num_loc_views = num; | 
 | 		    } | 
 | 		  else | 
 | 		    num = debug_info_p->num_loc_views; | 
 | 		  if (num > debug_info_p->num_loc_offsets) | 
 | 		    warn (_("More DW_AT_GNU_locview attributes than location offset attributes\n")); | 
 | 		  else | 
 | 		    { | 
 | 		      debug_info_p->loc_views [num] = uvalue; | 
 | 		      debug_info_p->num_loc_views++; | 
 | 		    } | 
 | 		} | 
 | 	    } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_low_pc: | 
 | 	  if (need_base_address) | 
 | 	    { | 
 | 	      if (form == DW_FORM_addrx) | 
 | 		uvalue = fetch_indexed_addr (debug_info_p->addr_base | 
 | 					     + uvalue * pointer_size, | 
 | 					     pointer_size); | 
 |  | 
 | 	      debug_info_p->base_address = uvalue; | 
 | 	    } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_GNU_addr_base: | 
 | 	case DW_AT_addr_base: | 
 | 	  debug_info_p->addr_base = uvalue; | 
 | 	  /* Retrieved elsewhere so that it is in | 
 | 	     place by the time we read low_pc.  */ | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_GNU_ranges_base: | 
 | 	  debug_info_p->ranges_base = uvalue; | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_ranges: | 
 | 	  if ((dwarf_version < 4 | 
 | 	       && (form == DW_FORM_data4 || form == DW_FORM_data8)) | 
 | 	      || form == DW_FORM_sec_offset | 
 | 	      || form == DW_FORM_rnglistx) | 
 | 	    { | 
 | 	      /* Process range list.  */ | 
 | 	      unsigned int lmax = debug_info_p->max_range_lists; | 
 | 	      unsigned int num = debug_info_p->num_range_lists; | 
 |  | 
 | 	      if (lmax == 0 || num >= lmax) | 
 | 		{ | 
 | 		  lmax += 1024; | 
 | 		  debug_info_p->range_lists = (uint64_t *) | 
 | 		    xcrealloc (debug_info_p->range_lists, | 
 | 			       lmax, sizeof (*debug_info_p->range_lists)); | 
 | 		  debug_info_p->max_range_lists = lmax; | 
 | 		} | 
 |  | 
 | 	      if (form == DW_FORM_rnglistx) | 
 | 		uvalue = fetch_indexed_offset (uvalue, rnglists, | 
 | 					       debug_info_p->rnglists_base, | 
 | 					       debug_info_p->offset_size); | 
 |  | 
 | 	      debug_info_p->range_lists [num] = uvalue; | 
 | 	      debug_info_p->num_range_lists++; | 
 | 	    } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_GNU_dwo_name: | 
 | 	case DW_AT_dwo_name: | 
 | 	  if (need_dwo_info) | 
 | 	    switch (form) | 
 | 	      { | 
 | 	      case DW_FORM_strp: | 
 | 		add_dwo_name (fetch_indirect_string (uvalue), | 
 | 			      cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_GNU_strp_alt: | 
 | 		add_dwo_name ((const char *) fetch_alt_indirect_string (uvalue), cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_GNU_str_index: | 
 | 	      case DW_FORM_strx: | 
 | 	      case DW_FORM_strx1: | 
 | 	      case DW_FORM_strx2: | 
 | 	      case DW_FORM_strx3: | 
 | 	      case DW_FORM_strx4: | 
 | 		add_dwo_name (fetch_indexed_string (uvalue, this_set, | 
 | 						    offset_size, false, | 
 | 						    debug_info_p->str_offsets_base), | 
 | 			      cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_string: | 
 | 		add_dwo_name ((const char *) orig_data, cu_offset); | 
 | 		break; | 
 | 	      default: | 
 | 		warn (_("Unsupported form (%s) for attribute %s\n"), | 
 | 		      get_FORM_name (form), get_AT_name (attribute)); | 
 | 		break; | 
 | 	      } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_comp_dir: | 
 | 	  /* FIXME: Also extract a build-id in a CU/TU.  */ | 
 | 	  if (need_dwo_info) | 
 | 	    switch (form) | 
 | 	      { | 
 | 	      case DW_FORM_strp: | 
 | 		add_dwo_dir (fetch_indirect_string (uvalue), cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_GNU_strp_alt: | 
 | 		add_dwo_dir (fetch_alt_indirect_string (uvalue), cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_line_strp: | 
 | 		add_dwo_dir (fetch_indirect_line_string (uvalue), cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_GNU_str_index: | 
 | 	      case DW_FORM_strx: | 
 | 	      case DW_FORM_strx1: | 
 | 	      case DW_FORM_strx2: | 
 | 	      case DW_FORM_strx3: | 
 | 	      case DW_FORM_strx4: | 
 | 		add_dwo_dir (fetch_indexed_string (uvalue, this_set, offset_size, false, | 
 | 						   debug_info_p->str_offsets_base), | 
 | 			     cu_offset); | 
 | 		break; | 
 | 	      case DW_FORM_string: | 
 | 		add_dwo_dir ((const char *) orig_data, cu_offset); | 
 | 		break; | 
 | 	      default: | 
 | 		warn (_("Unsupported form (%s) for attribute %s\n"), | 
 | 		      get_FORM_name (form), get_AT_name (attribute)); | 
 | 		break; | 
 | 	      } | 
 | 	  break; | 
 |  | 
 | 	case DW_AT_GNU_dwo_id: | 
 | 	  if (need_dwo_info) | 
 | 	    switch (form) | 
 | 	      { | 
 | 	      case DW_FORM_data8: | 
 | 		/* FIXME: Record the length of the ID as well ?  */ | 
 | 		add_dwo_id ((const char *) (data - 8), cu_offset); | 
 | 		break; | 
 | 	      default: | 
 | 		warn (_("Unsupported form (%s) for attribute %s\n"), | 
 | 		      get_FORM_name (form), get_AT_name (attribute)); | 
 | 		break; | 
 | 	      } | 
 | 	  break; | 
 |  | 
 | 	default: | 
 | 	  break; | 
 | 	} | 
 |     } | 
 |  | 
 |   if (do_loc || attribute == 0) | 
 |     return data; | 
 |  | 
 |   /* For some attributes we can display further information.  */ | 
 |   switch (attribute) | 
 |     { | 
 |     case DW_AT_type: | 
 |       if (level >= 0 && level < MAX_CU_NESTING | 
 | 	  && uvalue < (size_t) (end - start)) | 
 | 	{ | 
 | 	  bool is_signed = false; | 
 | 	  abbrev_entry *type_abbrev; | 
 | 	  unsigned char *type_data; | 
 | 	  abbrev_map *map; | 
 |  | 
 | 	  type_abbrev = get_type_abbrev_from_form (form, uvalue, | 
 | 						   cu_offset, end, | 
 | 						   section, NULL, | 
 | 						   &type_data, &map); | 
 | 	  if (type_abbrev != NULL) | 
 | 	    { | 
 | 	      get_type_signedness (type_abbrev, section, type_data, | 
 | 				   map ? section->start + map->end : end, | 
 | 				   map ? map->start : cu_offset, | 
 | 				   pointer_size, offset_size, dwarf_version, | 
 | 				   & is_signed, 0); | 
 | 	    } | 
 | 	  level_type_signed[level] = is_signed; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_inline: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_INL_not_inlined: | 
 | 	  printf (_("(not inlined)")); | 
 | 	  break; | 
 | 	case DW_INL_inlined: | 
 | 	  printf (_("(inlined)")); | 
 | 	  break; | 
 | 	case DW_INL_declared_not_inlined: | 
 | 	  printf (_("(declared as inline but ignored)")); | 
 | 	  break; | 
 | 	case DW_INL_declared_inlined: | 
 | 	  printf (_("(declared as inline and inlined)")); | 
 | 	  break; | 
 | 	default: | 
 | 	  printf (_("  (Unknown inline attribute value: %#" PRIx64 ")"), | 
 | 		  uvalue); | 
 | 	  break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_language: | 
 |       printf ("\t("); | 
 |       display_lang (uvalue); | 
 |       printf (")"); | 
 |       break; | 
 |  | 
 |     case DW_AT_encoding: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_ATE_void:		printf ("(void)"); break; | 
 | 	case DW_ATE_address:		printf ("(machine address)"); break; | 
 | 	case DW_ATE_boolean:		printf ("(boolean)"); break; | 
 | 	case DW_ATE_complex_float:	printf ("(complex float)"); break; | 
 | 	case DW_ATE_float:		printf ("(float)"); break; | 
 | 	case DW_ATE_signed:		printf ("(signed)"); break; | 
 | 	case DW_ATE_signed_char:	printf ("(signed char)"); break; | 
 | 	case DW_ATE_unsigned:		printf ("(unsigned)"); break; | 
 | 	case DW_ATE_unsigned_char:	printf ("(unsigned char)"); break; | 
 | 	  /* DWARF 2.1 values:  */ | 
 | 	case DW_ATE_imaginary_float:	printf ("(imaginary float)"); break; | 
 | 	case DW_ATE_decimal_float:	printf ("(decimal float)"); break; | 
 | 	  /* DWARF 3 values:  */ | 
 | 	case DW_ATE_packed_decimal:	printf ("(packed_decimal)"); break; | 
 | 	case DW_ATE_numeric_string:	printf ("(numeric_string)"); break; | 
 | 	case DW_ATE_edited:		printf ("(edited)"); break; | 
 | 	case DW_ATE_signed_fixed:	printf ("(signed_fixed)"); break; | 
 | 	case DW_ATE_unsigned_fixed:	printf ("(unsigned_fixed)"); break; | 
 | 	  /* DWARF 4 values:  */ | 
 | 	case DW_ATE_UTF:		printf ("(unicode string)"); break; | 
 | 	  /* DWARF 5 values:  */ | 
 | 	case DW_ATE_UCS:		printf ("(UCS)"); break; | 
 | 	case DW_ATE_ASCII:		printf ("(ASCII)"); break; | 
 |  | 
 | 	  /* HP extensions:  */ | 
 | 	case DW_ATE_HP_float80:		printf ("(HP_float80)"); break; | 
 | 	case DW_ATE_HP_complex_float80:	printf ("(HP_complex_float80)"); break; | 
 | 	case DW_ATE_HP_float128:	printf ("(HP_float128)"); break; | 
 | 	case DW_ATE_HP_complex_float128:printf ("(HP_complex_float128)"); break; | 
 | 	case DW_ATE_HP_floathpintel:	printf ("(HP_floathpintel)"); break; | 
 | 	case DW_ATE_HP_imaginary_float80:	printf ("(HP_imaginary_float80)"); break; | 
 | 	case DW_ATE_HP_imaginary_float128:	printf ("(HP_imaginary_float128)"); break; | 
 |  | 
 | 	default: | 
 | 	  if (uvalue >= DW_ATE_lo_user | 
 | 	      && uvalue <= DW_ATE_hi_user) | 
 | 	    printf (_("(user defined type)")); | 
 | 	  else | 
 | 	    printf (_("(unknown type)")); | 
 | 	  break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_accessibility: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_ACCESS_public:		printf ("(public)"); break; | 
 | 	case DW_ACCESS_protected:	printf ("(protected)"); break; | 
 | 	case DW_ACCESS_private:		printf ("(private)"); break; | 
 | 	default: | 
 | 	  printf (_("(unknown accessibility)")); | 
 | 	  break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_visibility: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_VIS_local:		printf ("(local)"); break; | 
 | 	case DW_VIS_exported:		printf ("(exported)"); break; | 
 | 	case DW_VIS_qualified:		printf ("(qualified)"); break; | 
 | 	default:			printf (_("(unknown visibility)")); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_endianity: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_END_default:		printf ("(default)"); break; | 
 | 	case DW_END_big:		printf ("(big)"); break; | 
 | 	case DW_END_little:		printf ("(little)"); break; | 
 | 	default: | 
 | 	  if (uvalue >= DW_END_lo_user && uvalue <= DW_END_hi_user) | 
 | 	    printf (_("(user specified)")); | 
 | 	  else | 
 | 	    printf (_("(unknown endianity)")); | 
 | 	  break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_virtuality: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_VIRTUALITY_none:	printf ("(none)"); break; | 
 | 	case DW_VIRTUALITY_virtual:	printf ("(virtual)"); break; | 
 | 	case DW_VIRTUALITY_pure_virtual:printf ("(pure_virtual)"); break; | 
 | 	default:			printf (_("(unknown virtuality)")); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_identifier_case: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_ID_case_sensitive:	printf ("(case_sensitive)"); break; | 
 | 	case DW_ID_up_case:		printf ("(up_case)"); break; | 
 | 	case DW_ID_down_case:		printf ("(down_case)"); break; | 
 | 	case DW_ID_case_insensitive:	printf ("(case_insensitive)"); break; | 
 | 	default:			printf (_("(unknown case)")); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_calling_convention: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_CC_normal:	printf ("(normal)"); break; | 
 | 	case DW_CC_program:	printf ("(program)"); break; | 
 | 	case DW_CC_nocall:	printf ("(nocall)"); break; | 
 | 	case DW_CC_pass_by_reference: printf ("(pass by ref)"); break; | 
 | 	case DW_CC_pass_by_value: printf ("(pass by value)"); break; | 
 | 	case DW_CC_GNU_renesas_sh: printf ("(Rensas SH)"); break; | 
 | 	case DW_CC_GNU_borland_fastcall_i386: printf ("(Borland fastcall i386)"); break; | 
 | 	default: | 
 | 	  if (uvalue >= DW_CC_lo_user | 
 | 	      && uvalue <= DW_CC_hi_user) | 
 | 	    printf (_("(user defined)")); | 
 | 	  else | 
 | 	    printf (_("(unknown convention)")); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_ordering: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case 255: | 
 | 	case -1: printf (_("(undefined)")); break; | 
 | 	case 0:  printf ("(row major)"); break; | 
 | 	case 1:  printf ("(column major)"); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_decimal_sign: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_DS_unsigned:            printf (_("(unsigned)")); break; | 
 | 	case DW_DS_leading_overpunch:   printf (_("(leading overpunch)")); break; | 
 | 	case DW_DS_trailing_overpunch:  printf (_("(trailing overpunch)")); break; | 
 | 	case DW_DS_leading_separate:    printf (_("(leading separate)")); break; | 
 | 	case DW_DS_trailing_separate:   printf (_("(trailing separate)")); break; | 
 | 	default:                        printf (_("(unrecognised)")); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_defaulted: | 
 |       printf ("\t"); | 
 |       switch (uvalue) | 
 | 	{ | 
 | 	case DW_DEFAULTED_no:           printf (_("(no)")); break; | 
 | 	case DW_DEFAULTED_in_class:     printf (_("(in class)")); break; | 
 | 	case DW_DEFAULTED_out_of_class: printf (_("(out of class)")); break; | 
 | 	default:                        printf (_("(unrecognised)")); break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_discr_list: | 
 |       printf ("\t"); | 
 |       display_discr_list (form, uvalue, data, level); | 
 |       break; | 
 |  | 
 |     case DW_AT_frame_base: | 
 |       have_frame_base = 1; | 
 |       /* Fall through.  */ | 
 |     case DW_AT_location: | 
 |     case DW_AT_loclists_base: | 
 |     case DW_AT_rnglists_base: | 
 |     case DW_AT_str_offsets_base: | 
 |     case DW_AT_string_length: | 
 |     case DW_AT_return_addr: | 
 |     case DW_AT_data_member_location: | 
 |     case DW_AT_vtable_elem_location: | 
 |     case DW_AT_segment: | 
 |     case DW_AT_static_link: | 
 |     case DW_AT_use_location: | 
 |     case DW_AT_call_value: | 
 |     case DW_AT_GNU_call_site_value: | 
 |     case DW_AT_call_data_value: | 
 |     case DW_AT_GNU_call_site_data_value: | 
 |     case DW_AT_call_target: | 
 |     case DW_AT_GNU_call_site_target: | 
 |     case DW_AT_call_target_clobbered: | 
 |     case DW_AT_GNU_call_site_target_clobbered: | 
 |       if ((dwarf_version < 4 | 
 | 	   && (form == DW_FORM_data4 || form == DW_FORM_data8)) | 
 | 	  || form == DW_FORM_sec_offset | 
 | 	  || form == DW_FORM_loclistx) | 
 | 	{ | 
 | 	  if (attribute != DW_AT_rnglists_base | 
 | 	      && attribute != DW_AT_str_offsets_base) | 
 | 	    printf (_(" (location list)")); | 
 | 	} | 
 |       /* Fall through.  */ | 
 |     case DW_AT_allocated: | 
 |     case DW_AT_associated: | 
 |     case DW_AT_data_location: | 
 |     case DW_AT_stride: | 
 |     case DW_AT_upper_bound: | 
 |     case DW_AT_lower_bound: | 
 |     case DW_AT_rank: | 
 |       if (block_start) | 
 | 	{ | 
 | 	  int need_frame_base; | 
 |  | 
 | 	  printf ("\t("); | 
 | 	  need_frame_base = decode_location_expression (block_start, | 
 | 							pointer_size, | 
 | 							offset_size, | 
 | 							dwarf_version, | 
 | 							uvalue, | 
 | 							cu_offset, section); | 
 | 	  printf (")"); | 
 | 	  if (need_frame_base && !have_frame_base) | 
 | 	    printf (_(" [without DW_AT_frame_base]")); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_data_bit_offset: | 
 |     case DW_AT_byte_size: | 
 |     case DW_AT_bit_size: | 
 |     case DW_AT_string_length_byte_size: | 
 |     case DW_AT_string_length_bit_size: | 
 |     case DW_AT_bit_stride: | 
 |       if (form == DW_FORM_exprloc) | 
 | 	{ | 
 | 	  printf ("\t("); | 
 | 	  (void) decode_location_expression (block_start, pointer_size, | 
 | 					     offset_size, dwarf_version, | 
 | 					     uvalue, cu_offset, section); | 
 | 	  printf (")"); | 
 | 	} | 
 |       break; | 
 |  | 
 |     case DW_AT_import: | 
 |       { | 
 | 	unsigned long abbrev_number; | 
 | 	abbrev_entry *entry; | 
 |  | 
 | 	entry = get_type_abbrev_from_form (form, uvalue, cu_offset, end, | 
 | 					   section, & abbrev_number, NULL, NULL); | 
 | 	if (entry == NULL) | 
 | 	  { | 
 | 	    if (form != DW_FORM_GNU_ref_alt) | 
 | 	      warn (_("Offset %#" PRIx64 " used as value for DW_AT_import attribute of DIE at offset %#tx is too big.\n"), | 
 | 		    uvalue, | 
 | 		    orig_data - section->start); | 
 | 	  } | 
 | 	else | 
 | 	  { | 
 | 	    printf (_("\t[Abbrev Number: %ld"), abbrev_number); | 
 | 	    printf (" (%s)", get_TAG_name (entry->tag)); | 
 | 	    printf ("]"); | 
 | 	  } | 
 |       } | 
 |       break; | 
 |  | 
 |     default: | 
 |       break; | 
 |     } | 
 |  | 
 |   return data; | 
 | } | 
 |  | 
 | static unsigned char * | 
 | read_and_display_attr (unsigned long attribute, | 
 | 		       unsigned long form, | 
 | 		       int64_t implicit_const, | 
 | 		       unsigned char *start, | 
 | 		       unsigned char *data, | 
 | 		       unsigned char *end, | 
 | 		       uint64_t cu_offset, | 
 | 		       uint64_t pointer_size, | 
 | 		       uint64_t offset_size, | 
 | 		       int dwarf_version, | 
 | 		       debug_info *debug_info_p, | 
 | 		       int do_loc, | 
 | 		       struct dwarf_section *section, | 
 | 		       struct cu_tu_set *this_set, | 
 | 		       int level) | 
 | { | 
 |   if (!do_loc) | 
 |     printf ("   %-18s:", get_AT_name (attribute)); | 
 |   data = read_and_display_attr_value (attribute, form, implicit_const, | 
 | 				      start, data, end, | 
 | 				      cu_offset, pointer_size, offset_size, | 
 | 				      dwarf_version, debug_info_p, | 
 | 				      do_loc, section, this_set, ' ', level); | 
 |   if (!do_loc) | 
 |     printf ("\n"); | 
 |   return data; | 
 | } | 
 |  | 
 | /* Like load_debug_section, but if the ordinary call fails, and we are | 
 |    following debug links, then attempt to load the requested section | 
 |    from one of the separate debug info files.  */ | 
 |  | 
 | static bool | 
 | load_debug_section_with_follow (enum dwarf_section_display_enum sec_enum, | 
 | 				void * handle) | 
 | { | 
 |   if (load_debug_section (sec_enum, handle)) | 
 |     { | 
 |       if (debug_displays[sec_enum].section.filename == NULL) | 
 | 	{ | 
 | 	  /* See if we can associate a filename with this section.  */ | 
 | 	  separate_info * i; | 
 |  | 
 | 	  for (i = first_separate_info; i != NULL; i = i->next) | 
 | 	    if (i->handle == handle) | 
 | 	      { | 
 | 		debug_displays[sec_enum].section.filename = i->filename; | 
 | 		break; | 
 | 	      } | 
 | 	} | 
 |  | 
 |       return true; | 
 |     } | 
 |  | 
 |   if (do_follow_links) | 
 |     { | 
 |       separate_info * i; | 
 |  | 
 |       for (i = first_separate_info; i != NULL; i = i->next) | 
 | 	{ | 
 | 	  if (load_debug_section (sec_enum, i->handle)) | 
 | 	    { | 
 | 	      debug_displays[sec_enum].section.filename = i->filename; | 
 |  | 
 | 	      /* FIXME: We should check to see if any of the remaining debug info | 
 | 		 files also contain this section, and, umm, do something about it.  */ | 
 | 	      return true; | 
 | 	    } | 
 | 	} | 
 |     } | 
 |  | 
 |   return false; | 
 | } | 
 |  | 
 | static void | 
 | introduce (struct dwarf_section * section, bool raw) | 
 | { | 
 |   if (raw) | 
 |     { | 
 |       if (do_follow_links && section->filename) | 
 | 	printf (_("Raw dump of debug contents of section %s (loaded from %s):\n\n"), | 
 | 		section->name, section->filename); | 
 |       else | 
 | 	printf (_("Raw dump of debug contents of section %s:\n\n"), section->name); | 
 |     } | 
 |   else | 
 |     { | 
 |       if (do_follow_links && section->filename) | 
 | 	printf (_("Contents of the %s section (loaded from %s):\n\n"), | 
 | 		section->name, section->filename); | 
 |       else | 
 | 	printf (_("Contents of the %s section:\n\n"), section->name); | 
 |     } | 
 | } | 
 |  | 
 | /* Free memory allocated for one unit in debug_information.  */ | 
 |  | 
 | static void | 
 | free_debug_information (debug_info *ent) | 
 | { | 
 |   if (ent->max_loc_offsets) | 
 |     { | 
 |       free (ent->loc_offsets); | 
 |       free (ent->loc_views); | 
 |       free (ent->have_frame_base); | 
 |     } | 
 |   if (ent->max_range_lists) | 
 |     { | 
 |       free (ent->range_lists); | 
 |     } | 
 | } | 
 |  | 
 | /* For look-ahead in attributes.  When you want to scan a DIE for one specific | 
 |    attribute and ignore the rest.  */ | 
 |  | 
 | static unsigned char * | 
 | skip_attribute (unsigned long    form, | 
 | 		unsigned char *  data, | 
 | 		unsigned char *  end, | 
 | 		uint64_t         pointer_size, | 
 | 		uint64_t         offset_size, | 
 | 		int              dwarf_version) | 
 | { | 
 |   uint64_t temp; | 
 |   size_t inc; | 
 |  | 
 |   switch (form) | 
 |     { | 
 |     case DW_FORM_ref_addr: | 
 |       inc = dwarf_version == 2 ? pointer_size : offset_size; | 
 |       break; | 
 |     case DW_FORM_addr: | 
 |       inc = pointer_size; | 
 |       break; | 
 |     case DW_FORM_strp_sup: | 
 |     case DW_FORM_strp: | 
 |     case DW_FORM_line_strp: | 
 |     case DW_FORM_sec_offset: | 
 |     case DW_FORM_GNU_ref_alt: | 
 |     case DW_FORM_GNU_strp_alt: | 
 |       inc = offset_size; | 
 |       break; | 
 |     case DW_FORM_ref1: | 
 |     case DW_FORM_flag: | 
 |     case DW_FORM_data1: | 
 |     case DW_FORM_strx1: | 
 |     case DW_FORM_addrx1: | 
 |       inc = 1; | 
 |       break; | 
 |     case DW_FORM_ref2: | 
 |     case DW_FORM_data2: | 
 |     case DW_FORM_strx2: | 
 |     case DW_FORM_addrx2: | 
 |       inc = 2; | 
 |       break; | 
 |     case DW_FORM_strx3: | 
 |     case DW_FORM_addrx3: | 
 |       inc = 3; | 
 |       break; | 
 |     case DW_FORM_ref_sup4: | 
 |     case DW_FORM_ref4: | 
 |     case DW_FORM_data4: | 
 |     case DW_FORM_strx4: | 
 |     case DW_FORM_addrx4: | 
 |       inc = 4; | 
 |       break; | 
 |     case DW_FORM_ref_sup8: | 
 |     case DW_FORM_ref8: | 
 |     case DW_FORM_data8: | 
 |     case DW_FORM_ref_sig8: | 
 |       inc = 8; | 
 |       break; | 
 |     case DW_FORM_data16: | 
 |       inc = 16; | 
 |       break; | 
 |     case DW_FORM_sdata: | 
 |       SKIP_SLEB (data, end); | 
 |       return data; | 
 |     case DW_FORM_GNU_str_index: | 
 |     case DW_FORM_strx: | 
 |     case DW_FORM_ref_udata: | 
 |     case DW_FORM_udata: | 
 |     case DW_FORM_GNU_addr_index: | 
 |     case DW_FORM_addrx: | 
 |     case DW_FORM_loclistx: | 
 |     case DW_FORM_rnglistx: | 
 |       SKIP_ULEB (data, end); | 
 |       return data; | 
 |     case DW_FORM_indirect: | 
 |       while (form == DW_FORM_indirect) | 
 |         READ_ULEB (form, data, end); | 
 |       return skip_attribute (form, data, end, pointer_size, offset_size, | 
 | 			     dwarf_version); | 
 |  | 
 |     case DW_FORM_string: | 
 |       inc = strnlen ((char *) data, end - data); | 
 |       break; | 
 |     case DW_FORM_block: | 
 |     case DW_FORM_exprloc: | 
 |       READ_ULEB (temp, data, end); | 
 |       inc = temp; | 
 |       break; | 
 |     case DW_FORM_block1: | 
 |       SAFE_BYTE_GET_AND_INC (temp, data, 1, end); | 
 |       inc = temp; | 
 |       break; | 
 |     case DW_FORM_block2: | 
 |       SAFE_BYTE_GET_AND_INC (temp, data, 2, end); | 
 |       inc = temp; | 
 |       break; | 
 |     case DW_FORM_block4: | 
 |       SAFE_BYTE_GET_AND_INC (temp, data, 4, end); | 
 |       inc = temp; | 
 |       break; | 
 |     case DW_FORM_implicit_const: | 
 |     case DW_FORM_flag_present: | 
 |       return data; | 
 |     default: | 
 |       warn (_("Unexpected form in top DIE\n")); | 
 |       return data; | 
 |     } | 
 |   if (inc <= (size_t) (end - data)) | 
 |     data += inc; | 
 |   else | 
 |     data = end; | 
 |   return data; | 
 | } | 
 |  | 
 | static void | 
 | read_bases (abbrev_entry *   entry, | 
 | 	    unsigned char *  data, | 
 | 	    unsigned char *  end, | 
 | 	    int64_t          pointer_size, | 
 | 	    uint64_t         offset_size, | 
 | 	    int              dwarf_version, | 
 | 	    debug_info *     debug_info_p) | 
 | { | 
 |   abbrev_attr *attr; | 
 |  | 
 |   for (attr = entry->first_attr; | 
 |        attr && attr->attribute; | 
 |        attr = attr->next) | 
 |     { | 
 |       uint64_t uvalue; | 
 |  | 
 |       if (attr->attribute == DW_AT_rnglists_base) | 
 | 	{ | 
 | 	  if (attr->form == DW_FORM_sec_offset) | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 | 	      debug_info_p->rnglists_base = uvalue; | 
 | 	    } | 
 | 	  else | 
 | 	    warn (_("Unexpected form of DW_AT_rnglists_base in the top DIE\n")); | 
 | 	} | 
 |       else if (attr->attribute == DW_AT_addr_base | 
 | 	       || attr->attribute == DW_AT_GNU_addr_base) | 
 | 	{ | 
 | 	  if (attr->form == DW_FORM_sec_offset) | 
 | 	    { | 
 | 	      SAFE_BYTE_GET_AND_INC (uvalue, data, offset_size, end); | 
 | 	      debug_info_p->addr_base = uvalue; | 
 | 	    } | 
 | 	  else | 
 | 	    warn (_("Unexpected form of DW_AT_addr_base in the top DIE\n")); | 
 | 	} | 
 |       else | 
 | 	data = skip_attribute (attr->form, data, end, pointer_size, | 
 | 			       offset_size, dwarf_version); | 
 |     } | 
 | } | 
 |  | 
 | /* Process the contents of a .debug_info section. | 
 |    If do_loc is TRUE then we are scanning for location lists and dwo tags | 
 |    and we do not want to display anything to the user. | 
 |    If do_types is TRUE, we are processing a .debug_types section instead of | 
 |    a .debug_info section. | 
 |    The information displayed is restricted by the values in DWARF_START_DIE | 
 |    and DWARF_CUTOFF_LEVEL. | 
 |    Returns TRUE upon success.  Otherwise an error or warning message is | 
 |    printed and FALSE is returned.  */ | 
 |  | 
 | static bool | 
 | process_debug_info (struct dwarf_section * section, | 
 | 		    void *file, | 
 | 		    enum dwarf_section_display_enum abbrev_sec, | 
 | 		    bool do_loc, | 
 | 		    bool do_types) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |   unsigned char *section_begin; | 
 |   unsigned int unit; | 
 |   unsigned int num_units = 0; | 
 |  | 
 |   /* First scan the section to get the number of comp units. | 
 |      Length sanity checks are done here.  */ | 
 |   for (section_begin = start, num_units = 0; section_begin < end; | 
 |        num_units ++) | 
 |     { | 
 |       uint64_t length; | 
 |  | 
 |       /* Read the first 4 bytes.  For a 32-bit DWARF section, this | 
 | 	 will be the length.  For a 64-bit DWARF section, it'll be | 
 | 	 the escape code 0xffffffff followed by an 8 byte length.  */ | 
 |       SAFE_BYTE_GET_AND_INC (length, section_begin, 4, end); | 
 |  | 
 |       if (length == 0xffffffff) | 
 | 	SAFE_BYTE_GET_AND_INC (length, section_begin, 8, end); | 
 |       else if (length >= 0xfffffff0 && length < 0xffffffff) | 
 | 	{ | 
 | 	  warn (_("Reserved length value (%#" PRIx64 ") found in section %s\n"), | 
 | 		length, section->name); | 
 | 	  return false; | 
 | 	} | 
 |  | 
 |       /* Negative values are illegal, they may even cause infinite | 
 | 	 looping.  This can happen if we can't accurately apply | 
 | 	 relocations to an object file, or if the file is corrupt.  */ | 
 |       if (length > (size_t) (end - section_begin)) | 
 | 	{ | 
 | 	  warn (_("Corrupt unit length (got %#" PRIx64 | 
 | 		  " expected at most %#tx) in section %s\n"), | 
 | 		length, end - section_begin, section->name); | 
 | 	  return false; | 
 | 	} | 
 |       section_begin += length; | 
 |     } | 
 |  | 
 |   if (num_units == 0) | 
 |     { | 
 |       error (_("No comp units in %s section ?\n"), section->name); | 
 |       return false; | 
 |     } | 
 |  | 
 |   if ((do_loc || do_debug_loc || do_debug_ranges || do_debug_info) | 
 |       && alloc_num_debug_info_entries == 0 | 
 |       && !do_types) | 
 |     { | 
 |       /* Then allocate an array to hold the information.  */ | 
 |       debug_information = cmalloc (num_units, sizeof (*debug_information)); | 
 |       if (debug_information == NULL) | 
 | 	{ | 
 | 	  error (_("Not enough memory for a debug info array of %u entries\n"), | 
 | 		 num_units); | 
 | 	  alloc_num_debug_info_entries = num_debug_info_entries = 0; | 
 | 	  return false; | 
 | 	} | 
 |  | 
 |       /* PR 17531: file: 92ca3797. | 
 | 	 We cannot rely upon the debug_information array being initialised | 
 | 	 before it is used.  A corrupt file could easily contain references | 
 | 	 to a unit for which information has not been made available.  So | 
 | 	 we ensure that the array is zeroed here.  */ | 
 |       memset (debug_information, 0, num_units * sizeof (*debug_information)); | 
 |  | 
 |       alloc_num_debug_info_entries = num_units; | 
 |     } | 
 |  | 
 |   if (!do_loc) | 
 |     { | 
 |       load_debug_section_with_follow (str, file); | 
 |       load_debug_section_with_follow (line_str, file); | 
 |       load_debug_section_with_follow (str_dwo, file); | 
 |       load_debug_section_with_follow (str_index, file); | 
 |       load_debug_section_with_follow (str_index_dwo, file); | 
 |       load_debug_section_with_follow (debug_addr, file); | 
 |     } | 
 |  | 
 |   load_debug_section_with_follow (abbrev_sec, file); | 
 |   load_debug_section_with_follow (loclists, file); | 
 |   load_debug_section_with_follow (rnglists, file); | 
 |   load_debug_section_with_follow (loclists_dwo, file); | 
 |   load_debug_section_with_follow (rnglists_dwo, file); | 
 |  | 
 |   if (debug_displays [abbrev_sec].section.start == NULL) | 
 |     { | 
 |       warn (_("Unable to locate %s section!\n"), | 
 | 	    debug_displays [abbrev_sec].section.uncompressed_name); | 
 |       return false; | 
 |     } | 
 |  | 
 |   if (!do_loc && dwarf_start_die == 0) | 
 |     introduce (section, false); | 
 |  | 
 |   free_all_abbrevs (); | 
 |  | 
 |   /* In order to be able to resolve DW_FORM_ref_addr forms we need | 
 |      to load *all* of the abbrevs for all CUs in this .debug_info | 
 |      section.  This does effectively mean that we (partially) read | 
 |      every CU header twice.  */ | 
 |   for (section_begin = start; start < end;) | 
 |     { | 
 |       DWARF2_Internal_CompUnit compunit; | 
 |       unsigned char *hdrptr; | 
 |       uint64_t abbrev_base; | 
 |       size_t abbrev_size; | 
 |       uint64_t cu_offset; | 
 |       unsigned int offset_size; | 
 |       struct cu_tu_set *this_set; | 
 |       unsigned char *end_cu; | 
 |  | 
 |       hdrptr = start; | 
 |       cu_offset = start - section_begin; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_length, hdrptr, 4, end); | 
 |  | 
 |       if (compunit.cu_length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_length, hdrptr, 8, end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |       end_cu = hdrptr + compunit.cu_length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_version, hdrptr, 2, end_cu); | 
 |  | 
 |       this_set = find_cu_tu_set_v2 (cu_offset, do_types); | 
 |  | 
 |       if (compunit.cu_version < 5) | 
 | 	{ | 
 | 	  compunit.cu_unit_type = DW_UT_compile; | 
 | 	  /* Initialize it due to a false compiler warning.  */ | 
 | 	  compunit.cu_pointer_size = -1; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_unit_type, hdrptr, 1, end_cu); | 
 | 	  do_types = (compunit.cu_unit_type == DW_UT_type); | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_pointer_size, hdrptr, 1, end_cu); | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_abbrev_offset, hdrptr, offset_size, | 
 | 			     end_cu); | 
 |  | 
 |       if (compunit.cu_unit_type == DW_UT_split_compile | 
 | 	  || compunit.cu_unit_type == DW_UT_skeleton) | 
 | 	{ | 
 | 	  uint64_t dwo_id; | 
 | 	  SAFE_BYTE_GET_AND_INC (dwo_id, hdrptr, 8, end_cu); | 
 | 	} | 
 |  | 
 |       if (this_set == NULL) | 
 | 	{ | 
 | 	  abbrev_base = 0; | 
 | 	  abbrev_size = debug_displays [abbrev_sec].section.size; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  abbrev_base = this_set->section_offsets [DW_SECT_ABBREV]; | 
 | 	  abbrev_size = this_set->section_sizes [DW_SECT_ABBREV]; | 
 | 	} | 
 |  | 
 |       abbrev_list *list; | 
 |       abbrev_list *free_list; | 
 |       list = find_and_process_abbrev_set (&debug_displays[abbrev_sec].section, | 
 | 					  abbrev_base, abbrev_size, | 
 | 					  compunit.cu_abbrev_offset, | 
 | 					  &free_list); | 
 |       start = end_cu; | 
 |       if (list != NULL && list->first_abbrev != NULL) | 
 | 	record_abbrev_list_for_cu (cu_offset, start - section_begin, | 
 | 				   list, free_list); | 
 |       else if (free_list != NULL) | 
 | 	free_abbrev_list (free_list); | 
 |     } | 
 |  | 
 |   for (start = section_begin, unit = 0; start < end; unit++) | 
 |     { | 
 |       DWARF2_Internal_CompUnit compunit; | 
 |       unsigned char *hdrptr; | 
 |       unsigned char *tags; | 
 |       int level, last_level, saved_level; | 
 |       uint64_t cu_offset; | 
 |       unsigned int offset_size; | 
 |       uint64_t signature = 0; | 
 |       uint64_t type_offset = 0; | 
 |       struct cu_tu_set *this_set; | 
 |       uint64_t abbrev_base; | 
 |       size_t abbrev_size; | 
 |       unsigned char *end_cu; | 
 |  | 
 |       hdrptr = start; | 
 |       cu_offset = start - section_begin; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_length, hdrptr, 4, end); | 
 |  | 
 |       if (compunit.cu_length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_length, hdrptr, 8, end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |       end_cu = hdrptr + compunit.cu_length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_version, hdrptr, 2, end_cu); | 
 |  | 
 |       this_set = find_cu_tu_set_v2 (cu_offset, do_types); | 
 |  | 
 |       if (compunit.cu_version < 5) | 
 | 	{ | 
 | 	  compunit.cu_unit_type = DW_UT_compile; | 
 | 	  /* Initialize it due to a false compiler warning.  */ | 
 | 	  compunit.cu_pointer_size = -1; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_unit_type, hdrptr, 1, end_cu); | 
 | 	  do_types = (compunit.cu_unit_type == DW_UT_type); | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (compunit.cu_pointer_size, hdrptr, 1, end_cu); | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (compunit.cu_abbrev_offset, hdrptr, offset_size, end_cu); | 
 |  | 
 |       if (this_set == NULL) | 
 | 	{ | 
 | 	  abbrev_base = 0; | 
 | 	  abbrev_size = debug_displays [abbrev_sec].section.size; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  abbrev_base = this_set->section_offsets [DW_SECT_ABBREV]; | 
 | 	  abbrev_size = this_set->section_sizes [DW_SECT_ABBREV]; | 
 | 	} | 
 |  | 
 |       if (compunit.cu_version < 5) | 
 | 	SAFE_BYTE_GET_AND_INC (compunit.cu_pointer_size, hdrptr, 1, end_cu); | 
 |  | 
 |       bool do_dwo_id = false; | 
 |       uint64_t dwo_id = 0; | 
 |       if (compunit.cu_unit_type == DW_UT_split_compile | 
 | 	  || compunit.cu_unit_type == DW_UT_skeleton) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (dwo_id, hdrptr, 8, end_cu); | 
 | 	  do_dwo_id = true; | 
 | 	} | 
 |  | 
 |       /* PR 17512: file: 001-108546-0.001:0.1.  */ | 
 |       if (compunit.cu_pointer_size < 2 || compunit.cu_pointer_size > 8) | 
 | 	{ | 
 | 	  warn (_("Invalid pointer size (%d) in compunit header, using %d instead\n"), | 
 | 		compunit.cu_pointer_size, offset_size); | 
 | 	  compunit.cu_pointer_size = offset_size; | 
 | 	} | 
 |  | 
 |       if (do_types) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (signature, hdrptr, 8, end_cu); | 
 | 	  SAFE_BYTE_GET_AND_INC (type_offset, hdrptr, offset_size, end_cu); | 
 | 	} | 
 |  | 
 |       if (dwarf_start_die >= (size_t) (end_cu - section_begin)) | 
 | 	{ | 
 | 	  start = end_cu; | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if ((do_loc || do_debug_loc || do_debug_ranges || do_debug_info) | 
 | 	  && num_debug_info_entries == 0 | 
 | 	  && alloc_num_debug_info_entries > unit | 
 | 	  && ! do_types) | 
 | 	{ | 
 | 	  free_debug_information (&debug_information[unit]); | 
 | 	  memset (&debug_information[unit], 0, sizeof (*debug_information)); | 
 | 	  debug_information[unit].pointer_size = compunit.cu_pointer_size; | 
 | 	  debug_information[unit].offset_size = offset_size; | 
 | 	  debug_information[unit].dwarf_version = compunit.cu_version; | 
 | 	  debug_information[unit].cu_offset = cu_offset; | 
 | 	  debug_information[unit].addr_base = DEBUG_INFO_UNAVAILABLE; | 
 | 	  debug_information[unit].ranges_base = DEBUG_INFO_UNAVAILABLE; | 
 | 	} | 
 |  | 
 |       if (!do_loc && dwarf_start_die == 0) | 
 | 	{ | 
 | 	  printf (_("  Compilation Unit @ offset %#" PRIx64 ":\n"), | 
 | 		  cu_offset); | 
 | 	  printf (_("   Length:        %#" PRIx64 " (%s)\n"), | 
 | 		  compunit.cu_length, | 
 | 		  offset_size == 8 ? "64-bit" : "32-bit"); | 
 | 	  printf (_("   Version:       %d\n"), compunit.cu_version); | 
 | 	  if (compunit.cu_version >= 5) | 
 | 	    { | 
 | 	      const char *name = get_DW_UT_name (compunit.cu_unit_type); | 
 |  | 
 | 	      printf (_("   Unit Type:     %s (%x)\n"), | 
 | 		      null_name (name), | 
 | 		      compunit.cu_unit_type); | 
 | 	    } | 
 | 	  printf (_("   Abbrev Offset: %#" PRIx64 "\n"), | 
 | 		  compunit.cu_abbrev_offset); | 
 | 	  printf (_("   Pointer Size:  %d\n"), compunit.cu_pointer_size); | 
 | 	  if (do_types) | 
 | 	    { | 
 | 	      printf (_("   Signature:     %#" PRIx64 "\n"), signature); | 
 | 	      printf (_("   Type Offset:   %#" PRIx64 "\n"), type_offset); | 
 | 	    } | 
 | 	  if (do_dwo_id) | 
 | 	    printf (_("   DWO ID:        %#" PRIx64 "\n"), dwo_id); | 
 | 	  if (this_set != NULL) | 
 | 	    { | 
 | 	      uint64_t *offsets = this_set->section_offsets; | 
 | 	      size_t *sizes = this_set->section_sizes; | 
 |  | 
 | 	      printf (_("   Section contributions:\n")); | 
 | 	      printf (_("    .debug_abbrev.dwo:       %#" PRIx64 "  %#zx\n"), | 
 | 		      offsets[DW_SECT_ABBREV], sizes[DW_SECT_ABBREV]); | 
 | 	      printf (_("    .debug_line.dwo:         %#" PRIx64 "  %#zx\n"), | 
 | 		      offsets[DW_SECT_LINE], sizes[DW_SECT_LINE]); | 
 | 	      printf (_("    .debug_loc.dwo:          %#" PRIx64 "  %#zx\n"), | 
 | 		      offsets[DW_SECT_LOC], sizes[DW_SECT_LOC]); | 
 | 	      printf (_("    .debug_str_offsets.dwo:  %#" PRIx64 "  %#zx\n"), | 
 | 		      offsets[DW_SECT_STR_OFFSETS], sizes[DW_SECT_STR_OFFSETS]); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       tags = hdrptr; | 
 |       start = end_cu; | 
 |  | 
 |       if (compunit.cu_version < 2 || compunit.cu_version > 5) | 
 | 	{ | 
 | 	  warn (_("CU at offset %#" PRIx64 " contains corrupt or " | 
 | 		  "unsupported version number: %d.\n"), | 
 | 		cu_offset, compunit.cu_version); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if (compunit.cu_unit_type != DW_UT_compile | 
 | 	  && compunit.cu_unit_type != DW_UT_partial | 
 | 	  && compunit.cu_unit_type != DW_UT_type | 
 | 	  && compunit.cu_unit_type != DW_UT_split_compile | 
 | 	  && compunit.cu_unit_type != DW_UT_skeleton) | 
 | 	{ | 
 | 	  warn (_("CU at offset %#" PRIx64 " contains corrupt or " | 
 | 		  "unsupported unit type: %d.\n"), | 
 | 		cu_offset, compunit.cu_unit_type); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       /* Process the abbrevs used by this compilation unit.  */ | 
 |       abbrev_list *list; | 
 |       list = find_and_process_abbrev_set (&debug_displays[abbrev_sec].section, | 
 | 					  abbrev_base, abbrev_size, | 
 | 					  compunit.cu_abbrev_offset, NULL); | 
 |       level = 0; | 
 |       last_level = level; | 
 |       saved_level = -1; | 
 |       while (tags < start) | 
 | 	{ | 
 | 	  unsigned long abbrev_number; | 
 | 	  unsigned long die_offset; | 
 | 	  abbrev_entry *entry; | 
 | 	  abbrev_attr *attr; | 
 | 	  int do_printing = 1; | 
 |  | 
 | 	  die_offset = tags - section_begin; | 
 |  | 
 | 	  READ_ULEB (abbrev_number, tags, start); | 
 |  | 
 | 	  /* A null DIE marks the end of a list of siblings or it may also be | 
 | 	     a section padding.  */ | 
 | 	  if (abbrev_number == 0) | 
 | 	    { | 
 | 	      /* Check if it can be a section padding for the last CU.  */ | 
 | 	      if (level == 0 && start == end) | 
 | 		{ | 
 | 		  unsigned char *chk; | 
 |  | 
 | 		  for (chk = tags; chk < start; chk++) | 
 | 		    if (*chk != 0) | 
 | 		      break; | 
 | 		  if (chk == start) | 
 | 		    break; | 
 | 		} | 
 |  | 
 | 	      if (!do_loc && die_offset >= dwarf_start_die | 
 | 		  && (dwarf_cutoff_level == -1 | 
 | 		      || level < dwarf_cutoff_level)) | 
 | 		printf (_(" <%d><%lx>: Abbrev Number: 0\n"), | 
 | 			level, die_offset); | 
 |  | 
 | 	      --level; | 
 | 	      if (level < 0) | 
 | 		{ | 
 | 		  static unsigned num_bogus_warns = 0; | 
 |  | 
 | 		  if (num_bogus_warns < 3) | 
 | 		    { | 
 | 		      warn (_("Bogus end-of-siblings marker detected at offset %lx in %s section\n"), | 
 | 			    die_offset, section->name); | 
 | 		      num_bogus_warns ++; | 
 | 		      if (num_bogus_warns == 3) | 
 | 			warn (_("Further warnings about bogus end-of-sibling markers suppressed\n")); | 
 | 		    } | 
 | 		} | 
 | 	      if (dwarf_start_die != 0 && level < saved_level) | 
 | 		{ | 
 | 		  if (list != NULL) | 
 | 		    free_abbrev_list (list); | 
 | 		  return true; | 
 | 		} | 
 | 	      continue; | 
 | 	    } | 
 |  | 
 | 	  if (!do_loc) | 
 | 	    { | 
 | 	      if (dwarf_start_die != 0 && die_offset < dwarf_start_die) | 
 | 		do_printing = 0; | 
 | 	      else | 
 | 		{ | 
 | 		  if (dwarf_start_die != 0 && die_offset == dwarf_start_die) | 
 | 		    saved_level = level; | 
 | 		  do_printing = (dwarf_cutoff_level == -1 | 
 | 				 || level < dwarf_cutoff_level); | 
 | 		  if (do_printing) | 
 | 		    printf (_(" <%d><%lx>: Abbrev Number: %lu"), | 
 | 			    level, die_offset, abbrev_number); | 
 | 		  else if (dwarf_cutoff_level == -1 | 
 | 			   || last_level < dwarf_cutoff_level) | 
 | 		    printf (_(" <%d><%lx>: ...\n"), level, die_offset); | 
 | 		  last_level = level; | 
 | 		} | 
 | 	    } | 
 |  | 
 | 	  /* Scan through the abbreviation list until we reach the | 
 | 	     correct entry.  */ | 
 | 	  entry = NULL; | 
 | 	  if (list != NULL) | 
 | 	    for (entry = list->first_abbrev; entry != NULL; entry = entry->next) | 
 | 	      if (entry->number == abbrev_number) | 
 | 		break; | 
 |  | 
 | 	  if (entry == NULL) | 
 | 	    { | 
 | 	      if (!do_loc && do_printing) | 
 | 		{ | 
 | 		  printf ("\n"); | 
 | 		  fflush (stdout); | 
 | 		} | 
 | 	      warn (_("DIE at offset %#lx refers to abbreviation number %lu which does not exist\n"), | 
 | 		    die_offset, abbrev_number); | 
 | 	      if (list != NULL) | 
 | 		free_abbrev_list (list); | 
 | 	      return false; | 
 | 	    } | 
 |  | 
 | 	  if (!do_loc && do_printing) | 
 | 	    printf (" (%s)\n", get_TAG_name (entry->tag)); | 
 |  | 
 | 	  switch (entry->tag) | 
 | 	    { | 
 | 	    default: | 
 | 	      need_base_address = 0; | 
 | 	      break; | 
 | 	    case DW_TAG_compile_unit: | 
 | 	    case DW_TAG_skeleton_unit: | 
 | 	      need_base_address = 1; | 
 | 	      need_dwo_info = do_loc; | 
 | 	      break; | 
 | 	    case DW_TAG_entry_point: | 
 | 	      need_base_address = 0; | 
 | 	      /* Assuming that there is no DW_AT_frame_base.  */ | 
 | 	      have_frame_base = 0; | 
 | 	      break; | 
 | 	    case DW_TAG_subprogram: | 
 | 	      need_base_address = 0; | 
 | 	      if (level <= frame_base_level) | 
 | 		/* Don't reset that for nested subprogram.  */ | 
 | 		have_frame_base = 0; | 
 | 	      break; | 
 | 	    } | 
 |  | 
 | 	  debug_info *debug_info_p = ((debug_information | 
 | 				       && unit < alloc_num_debug_info_entries) | 
 | 				      ? debug_information + unit : NULL); | 
 |  | 
 | 	  assert (!debug_info_p | 
 | 		  || (debug_info_p->num_loc_offsets | 
 | 		      == debug_info_p->num_loc_views)); | 
 |  | 
 | 	  /* Look ahead so that the values of DW_AT_rnglists_base, | 
 | 	     DW_AT_[GNU_]addr_base are available before attributes that | 
 | 	     reference them are parsed in the same DIE. | 
 | 	     Only needed for the top DIE on DWARFv5+. | 
 | 	     No simiar treatment for loclists_base because there should | 
 | 	     be no loclist attributes in top DIE.  */ | 
 | 	  if (debug_info_p && compunit.cu_version >= 5 && level == 0) | 
 | 	    { | 
 | 	      int64_t stemp; | 
 |  | 
 | 	      read_bases (entry, | 
 | 			  tags, | 
 | 			  start, | 
 | 			  compunit.cu_pointer_size, | 
 | 			  offset_size, | 
 | 			  compunit.cu_version, | 
 | 			  debug_info_p); | 
 |  | 
 | 	      /* This check was in place before, keep it.  */ | 
 | 	      stemp = debug_info_p->rnglists_base; | 
 | 	      if (stemp < 0) | 
 | 		{ | 
 | 		  warn (_("CU @ %#" PRIx64 " has has a negative rnglists_base " | 
 | 			  "value of %#" PRIx64 " - treating as zero\n"), | 
 | 			debug_info_p->cu_offset, stemp); | 
 | 		  debug_info_p->rnglists_base = 0; | 
 | 		} | 
 | 	    } | 
 |  | 
 | 	  for (attr = entry->first_attr; | 
 | 	       attr && attr->attribute; | 
 | 	       attr = attr->next) | 
 | 	    { | 
 | 	      if (! do_loc && do_printing) | 
 | 		/* Show the offset from where the tag was extracted.  */ | 
 | 		printf ("    <%tx>", tags - section_begin); | 
 | 	      tags = read_and_display_attr (attr->attribute, | 
 | 					    attr->form, | 
 | 					    attr->implicit_const, | 
 | 					    section_begin, | 
 | 					    tags, | 
 | 					    start, | 
 | 					    cu_offset, | 
 | 					    compunit.cu_pointer_size, | 
 | 					    offset_size, | 
 | 					    compunit.cu_version, | 
 | 					    debug_info_p, | 
 | 					    do_loc || ! do_printing, | 
 | 					    section, | 
 | 					    this_set, | 
 | 					    level); | 
 | 	    } | 
 |  | 
 | 	  /* If a locview attribute appears before a location one, | 
 | 	     make sure we don't associate it with an earlier | 
 | 	     loclist. */ | 
 | 	  if (debug_info_p) | 
 | 	    switch (debug_info_p->num_loc_offsets - debug_info_p->num_loc_views) | 
 | 	      { | 
 | 	      case 1: | 
 | 		debug_info_p->loc_views [debug_info_p->num_loc_views] = -1; | 
 | 		debug_info_p->num_loc_views++; | 
 | 		assert (debug_info_p->num_loc_views | 
 | 			== debug_info_p->num_loc_offsets); | 
 | 		break; | 
 |  | 
 | 	      case 0: | 
 | 		break; | 
 |  | 
 | 	      case -1: | 
 | 		warn (_("DIE has locviews without loclist\n")); | 
 | 		debug_info_p->num_loc_views--; | 
 | 		break; | 
 |  | 
 | 	      default: | 
 | 		assert (0); | 
 | 	      } | 
 |  | 
 | 	  if (entry->children) | 
 | 	    ++level; | 
 | 	} | 
 |       if (list != NULL) | 
 | 	free_abbrev_list (list); | 
 |     } | 
 |  | 
 |   /* Set num_debug_info_entries here so that it can be used to check if | 
 |      we need to process .debug_loc and .debug_ranges sections.  */ | 
 |   if ((do_loc || do_debug_loc || do_debug_ranges || do_debug_info) | 
 |       && num_debug_info_entries == 0 | 
 |       && ! do_types) | 
 |     { | 
 |       if (num_units > alloc_num_debug_info_entries) | 
 | 	num_debug_info_entries = alloc_num_debug_info_entries; | 
 |       else | 
 | 	num_debug_info_entries = num_units; | 
 |     } | 
 |  | 
 |   if (!do_loc) | 
 |     printf ("\n"); | 
 |  | 
 |   return true; | 
 | } | 
 |  | 
 | /* Locate and scan the .debug_info section in the file and record the pointer | 
 |    sizes and offsets for the compilation units in it.  Usually an executable | 
 |    will have just one pointer size, but this is not guaranteed, and so we try | 
 |    not to make any assumptions.  Returns zero upon failure, or the number of | 
 |    compilation units upon success.  */ | 
 |  | 
 | static unsigned int | 
 | load_debug_info (void * file) | 
 | { | 
 |   /* If we have already tried and failed to load the .debug_info | 
 |      section then do not bother to repeat the task.  */ | 
 |   if (num_debug_info_entries == DEBUG_INFO_UNAVAILABLE) | 
 |     return 0; | 
 |  | 
 |   /* If we already have the information there is nothing else to do.  */ | 
 |   if (num_debug_info_entries > 0) | 
 |     return num_debug_info_entries; | 
 |  | 
 |   /* If this is a DWARF package file, load the CU and TU indexes.  */ | 
 |   (void) load_cu_tu_indexes (file); | 
 |  | 
 |   if (load_debug_section_with_follow (info, file) | 
 |       && process_debug_info (&debug_displays [info].section, file, abbrev, true, false)) | 
 |     return num_debug_info_entries; | 
 |  | 
 |   if (load_debug_section_with_follow (info_dwo, file) | 
 |       && process_debug_info (&debug_displays [info_dwo].section, file, | 
 | 			     abbrev_dwo, true, false)) | 
 |     return num_debug_info_entries; | 
 |  | 
 |   num_debug_info_entries = DEBUG_INFO_UNAVAILABLE; | 
 |   return 0; | 
 | } | 
 |  | 
 | /* Read a DWARF .debug_line section header starting at DATA. | 
 |    Upon success returns an updated DATA pointer and the LINFO | 
 |    structure and the END_OF_SEQUENCE pointer will be filled in. | 
 |    Otherwise returns NULL.  */ | 
 |  | 
 | static unsigned char * | 
 | read_debug_line_header (struct dwarf_section * section, | 
 | 			unsigned char * data, | 
 | 			unsigned char * end, | 
 | 			DWARF2_Internal_LineInfo * linfo, | 
 | 			unsigned char ** end_of_sequence) | 
 | { | 
 |   unsigned char *hdrptr; | 
 |  | 
 |   /* Extract information from the Line Number Program Header. | 
 |      (section 6.2.4 in the Dwarf3 doc).  */ | 
 |   hdrptr = data; | 
 |  | 
 |   /* Get and check the length of the block.  */ | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_length, hdrptr, 4, end); | 
 |  | 
 |   if (linfo->li_length == 0xffffffff) | 
 |     { | 
 |       /* This section is 64-bit DWARF 3.  */ | 
 |       SAFE_BYTE_GET_AND_INC (linfo->li_length, hdrptr, 8, end); | 
 |       linfo->li_offset_size = 8; | 
 |     } | 
 |   else | 
 |     linfo->li_offset_size = 4; | 
 |  | 
 |   if (linfo->li_length > (size_t) (end - hdrptr)) | 
 |     { | 
 |       /* If the length field has a relocation against it, then we should | 
 | 	 not complain if it is inaccurate (and probably negative).  This | 
 | 	 happens in object files when the .debug_line section is actually | 
 | 	 comprised of several different .debug_line.* sections, (some of | 
 | 	 which may be removed by linker garbage collection), and a relocation | 
 | 	 is used to compute the correct length once that is done.  */ | 
 |       if (reloc_at (section, (hdrptr - section->start) - linfo->li_offset_size)) | 
 | 	{ | 
 | 	  linfo->li_length = end - hdrptr; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  warn (_("The length field (%#" PRIx64 ")" | 
 | 		  " in the debug_line header is wrong" | 
 | 		  " - the section is too small\n"), | 
 | 		linfo->li_length); | 
 | 	  return NULL; | 
 | 	} | 
 |     } | 
 |   end = hdrptr + linfo->li_length; | 
 |  | 
 |   /* Get and check the version number.  */ | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_version, hdrptr, 2, end); | 
 |  | 
 |   if (linfo->li_version != 2 | 
 |       && linfo->li_version != 3 | 
 |       && linfo->li_version != 4 | 
 |       && linfo->li_version != 5) | 
 |     { | 
 |       warn (_("Only DWARF version 2, 3, 4 and 5 line info " | 
 | 	      "is currently supported.\n")); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   if (linfo->li_version >= 5) | 
 |     { | 
 |       SAFE_BYTE_GET_AND_INC (linfo->li_address_size, hdrptr, 1, end); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (linfo->li_segment_size, hdrptr, 1, end); | 
 |       if (linfo->li_segment_size != 0) | 
 | 	{ | 
 | 	  warn (_("The %s section contains " | 
 | 		  "unsupported segment selector size: %d.\n"), | 
 | 		section->name, linfo->li_segment_size); | 
 | 	  return NULL; | 
 | 	} | 
 |     } | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_prologue_length, hdrptr, | 
 | 			 linfo->li_offset_size, end); | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_min_insn_length, hdrptr, 1, end); | 
 |  | 
 |   if (linfo->li_version >= 4) | 
 |     { | 
 |       SAFE_BYTE_GET_AND_INC (linfo->li_max_ops_per_insn, hdrptr, 1, end); | 
 |  | 
 |       if (linfo->li_max_ops_per_insn == 0) | 
 | 	{ | 
 | 	  warn (_("Invalid maximum operations per insn.\n")); | 
 | 	  return NULL; | 
 | 	} | 
 |     } | 
 |   else | 
 |     linfo->li_max_ops_per_insn = 1; | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_default_is_stmt, hdrptr, 1, end); | 
 |   SAFE_SIGNED_BYTE_GET_AND_INC (linfo->li_line_base, hdrptr, 1, end); | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_line_range, hdrptr, 1, end); | 
 |   SAFE_BYTE_GET_AND_INC (linfo->li_opcode_base, hdrptr, 1, end); | 
 |  | 
 |   *end_of_sequence = end; | 
 |   return hdrptr; | 
 | } | 
 |  | 
 | static unsigned char * | 
 | display_formatted_table (unsigned char *data, | 
 | 			 unsigned char *start, | 
 | 			 unsigned char *end, | 
 | 			 const DWARF2_Internal_LineInfo *linfo, | 
 | 			 struct dwarf_section *section, | 
 | 			 bool is_dir) | 
 | { | 
 |   unsigned char *format_start, format_count, *format, formati; | 
 |   uint64_t data_count, datai; | 
 |   unsigned int namepass, last_entry = 0; | 
 |   const char * table_name = is_dir ? N_("Directory Table") : N_("File Name Table"); | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (format_count, data, 1, end); | 
 |   if (do_checks && format_count > 5) | 
 |     warn (_("Unexpectedly large number of columns in the %s (%u)\n"), | 
 | 	  table_name, format_count); | 
 |  | 
 |   format_start = data; | 
 |   for (formati = 0; formati < format_count; formati++) | 
 |     { | 
 |       SKIP_ULEB (data, end); | 
 |       SKIP_ULEB (data, end); | 
 |       if (data >= end) | 
 | 	{ | 
 | 	  warn (_("%s: Corrupt format description entry\n"), table_name); | 
 | 	  return data; | 
 | 	} | 
 |     } | 
 |  | 
 |   READ_ULEB (data_count, data, end); | 
 |   if (data_count == 0) | 
 |     { | 
 |       printf (_("\n The %s is empty.\n"), table_name); | 
 |       return data; | 
 |     } | 
 |   else if (data >= end | 
 | 	   || data_count > (size_t) (end - data)) | 
 |     { | 
 |       warn (_("%s: Corrupt entry count %#" PRIx64 "\n"), table_name, data_count); | 
 |       return data; | 
 |     } | 
 |  | 
 |   else if (format_count == 0) | 
 |     { | 
 |       warn (_("%s: format count is zero, but the table is not empty\n"), | 
 | 	    table_name); | 
 |       return end; | 
 |     } | 
 |  | 
 |   printf (_("\n The %s (offset %#tx, lines %" PRIu64 ", columns %u):\n"), | 
 | 	  table_name, data - start, data_count, format_count); | 
 |  | 
 |   printf (_("  Entry")); | 
 |   /* Delay displaying name as the last entry for better screen layout.  */ | 
 |   for (namepass = 0; namepass < 2; namepass++) | 
 |     { | 
 |       format = format_start; | 
 |       for (formati = 0; formati < format_count; formati++) | 
 | 	{ | 
 | 	  uint64_t content_type; | 
 |  | 
 | 	  READ_ULEB (content_type, format, end); | 
 | 	  if ((content_type == DW_LNCT_path) == (namepass == 1)) | 
 | 	    switch (content_type) | 
 | 	      { | 
 | 	      case DW_LNCT_path: | 
 | 		printf (_("\tName")); | 
 | 		break; | 
 | 	      case DW_LNCT_directory_index: | 
 | 		printf (_("\tDir")); | 
 | 		break; | 
 | 	      case DW_LNCT_timestamp: | 
 | 		printf (_("\tTime")); | 
 | 		break; | 
 | 	      case DW_LNCT_size: | 
 | 		printf (_("\tSize")); | 
 | 		break; | 
 | 	      case DW_LNCT_MD5: | 
 | 		printf (_("\tMD5\t\t\t")); | 
 | 		break; | 
 | 	      default: | 
 | 		printf (_("\t(Unknown format content type %" PRIu64 ")"), | 
 | 			content_type); | 
 | 	      } | 
 | 	  SKIP_ULEB (format, end); | 
 | 	} | 
 |     } | 
 |   putchar ('\n'); | 
 |  | 
 |   for (datai = 0; datai < data_count; datai++) | 
 |     { | 
 |       unsigned char *datapass = data; | 
 |  | 
 |       printf ("  %d", last_entry++); | 
 |       /* Delay displaying name as the last entry for better screen layout.  */ | 
 |       for (namepass = 0; namepass < 2; namepass++) | 
 | 	{ | 
 | 	  format = format_start; | 
 | 	  data = datapass; | 
 | 	  for (formati = 0; formati < format_count; formati++) | 
 | 	    { | 
 | 	      uint64_t content_type, form; | 
 |  | 
 | 	      READ_ULEB (content_type, format, end); | 
 | 	      READ_ULEB (form, format, end); | 
 | 	      data = read_and_display_attr_value (0, form, 0, start, data, end, | 
 | 						  0, 0, linfo->li_offset_size, | 
 | 						  linfo->li_version, NULL, | 
 | 			    ((content_type == DW_LNCT_path) != (namepass == 1)), | 
 | 						  section, NULL, '\t', -1); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       if (data >= end && (datai < data_count - 1)) | 
 | 	{ | 
 | 	  warn (_("\n%s: Corrupt entries list\n"), table_name); | 
 | 	  return data; | 
 | 	} | 
 |       putchar ('\n'); | 
 |     } | 
 |   return data; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_sup (struct dwarf_section *  section, | 
 | 		   void *                  file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char * start = section->start; | 
 |   unsigned char * end = section->start + section->size; | 
 |   unsigned int version; | 
 |   char is_supplementary; | 
 |   const unsigned char * sup_filename; | 
 |   size_t sup_filename_len; | 
 |   unsigned int num_read; | 
 |   int status; | 
 |   uint64_t checksum_len; | 
 |  | 
 |  | 
 |   introduce (section, true); | 
 |   if (section->size < 4) | 
 |     { | 
 |       error (_("corrupt .debug_sup section: size is too small\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   /* Read the data.  */ | 
 |   SAFE_BYTE_GET_AND_INC (version, start, 2, end); | 
 |   if (version < 5) | 
 |     warn (_("corrupt .debug_sup section: version < 5\n")); | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (is_supplementary, start, 1, end); | 
 |   if (is_supplementary != 0 && is_supplementary != 1) | 
 |     warn (_("corrupt .debug_sup section: is_supplementary not 0 or 1\n")); | 
 |  | 
 |   sup_filename = start; | 
 |   if (is_supplementary && sup_filename[0] != 0) | 
 |     warn (_("corrupt .debug_sup section: filename not empty in supplementary section\n")); | 
 |  | 
 |   sup_filename_len = strnlen ((const char *) start, end - start); | 
 |   if (sup_filename_len == (size_t) (end - start)) | 
 |     { | 
 |       error (_("corrupt .debug_sup section: filename is not NUL terminated\n")); | 
 |       return 0; | 
 |     } | 
 |   start += sup_filename_len + 1; | 
 |  | 
 |   checksum_len = read_leb128 (start, end, false /* unsigned */, & num_read, & status); | 
 |   if (status) | 
 |     { | 
 |       error (_("corrupt .debug_sup section: bad LEB128 field for checksum length\n")); | 
 |       checksum_len = 0; | 
 |     } | 
 |   start += num_read; | 
 |   if (checksum_len > (size_t) (end - start)) | 
 |     { | 
 |       error (_("corrupt .debug_sup section: checksum length is longer than the remaining section length\n")); | 
 |       checksum_len = end - start; | 
 |     } | 
 |   else if (checksum_len < (size_t) (end - start)) | 
 |     { | 
 |       warn (_("corrupt .debug_sup section: there are %#" PRIx64 | 
 | 	      " extra, unused bytes at the end of the section\n"), | 
 | 	    (end - start) - checksum_len); | 
 |     } | 
 |  | 
 |   printf (_("  Version:      %u\n"), version); | 
 |   printf (_("  Is Supp:      %u\n"), is_supplementary); | 
 |   printf (_("  Filename:     %s\n"), sup_filename); | 
 |   printf (_("  Checksum Len: %" PRIu64 "\n"), checksum_len); | 
 |   if (checksum_len > 0) | 
 |     { | 
 |       printf (_("  Checksum:     ")); | 
 |       while (checksum_len--) | 
 | 	printf ("0x%x ", * start++ ); | 
 |       printf ("\n"); | 
 |     } | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_lines_raw (struct dwarf_section *  section, | 
 | 			 unsigned char *         data, | 
 | 			 unsigned char *         end, | 
 | 			 void *                  file) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   int verbose_view = 0; | 
 |  | 
 |   introduce (section, true); | 
 |  | 
 |   while (data < end) | 
 |     { | 
 |       static DWARF2_Internal_LineInfo saved_linfo; | 
 |       DWARF2_Internal_LineInfo linfo; | 
 |       unsigned char *standard_opcodes; | 
 |       unsigned char *end_of_sequence; | 
 |       int i; | 
 |  | 
 |       if (startswith (section->name, ".debug_line.") | 
 | 	  /* Note: the following does not apply to .debug_line.dwo sections. | 
 | 	     These are full debug_line sections.  */ | 
 | 	  && strcmp (section->name, ".debug_line.dwo") != 0) | 
 | 	{ | 
 | 	  /* Sections named .debug_line.<foo> are fragments of a .debug_line | 
 | 	     section containing just the Line Number Statements.  They are | 
 | 	     created by the assembler and intended to be used alongside gcc's | 
 | 	     -ffunction-sections command line option.  When the linker's | 
 | 	     garbage collection decides to discard a .text.<foo> section it | 
 | 	     can then also discard the line number information in .debug_line.<foo>. | 
 |  | 
 | 	     Since the section is a fragment it does not have the details | 
 | 	     needed to fill out a LineInfo structure, so instead we use the | 
 | 	     details from the last full debug_line section that we processed.  */ | 
 | 	  end_of_sequence = end; | 
 | 	  standard_opcodes = NULL; | 
 | 	  linfo = saved_linfo; | 
 | 	  /* PR 17531: file: 0522b371.  */ | 
 | 	  if (linfo.li_line_range == 0) | 
 | 	    { | 
 | 	      warn (_("Partial .debug_line. section encountered without a prior full .debug_line section\n")); | 
 | 	      return 0; | 
 | 	    } | 
 | 	  reset_state_machine (linfo.li_default_is_stmt); | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  unsigned char * hdrptr; | 
 |  | 
 | 	  if ((hdrptr = read_debug_line_header (section, data, end, & linfo, | 
 | 						& end_of_sequence)) == NULL) | 
 | 	    return 0; | 
 |  | 
 | 	  printf (_("  Offset:                      %#tx\n"), data - start); | 
 | 	  printf (_("  Length:                      %" PRId64 "\n"), linfo.li_length); | 
 | 	  printf (_("  DWARF Version:               %d\n"), linfo.li_version); | 
 | 	  if (linfo.li_version >= 5) | 
 | 	    { | 
 | 	      printf (_("  Address size (bytes):        %d\n"), linfo.li_address_size); | 
 | 	      printf (_("  Segment selector (bytes):    %d\n"), linfo.li_segment_size); | 
 | 	    } | 
 | 	  printf (_("  Prologue Length:             %d\n"), (int) linfo.li_prologue_length); | 
 | 	  printf (_("  Minimum Instruction Length:  %d\n"), linfo.li_min_insn_length); | 
 | 	  if (linfo.li_version >= 4) | 
 | 	    printf (_("  Maximum Ops per Instruction: %d\n"), linfo.li_max_ops_per_insn); | 
 | 	  printf (_("  Initial value of 'is_stmt':  %d\n"), linfo.li_default_is_stmt); | 
 | 	  printf (_("  Line Base:                   %d\n"), linfo.li_line_base); | 
 | 	  printf (_("  Line Range:                  %d\n"), linfo.li_line_range); | 
 | 	  printf (_("  Opcode Base:                 %d\n"), linfo.li_opcode_base); | 
 |  | 
 | 	  /* PR 17512: file: 1665-6428-0.004.  */ | 
 | 	  if (linfo.li_line_range == 0) | 
 | 	    { | 
 | 	      warn (_("Line range of 0 is invalid, using 1 instead\n")); | 
 | 	      linfo.li_line_range = 1; | 
 | 	    } | 
 |  | 
 | 	  reset_state_machine (linfo.li_default_is_stmt); | 
 |  | 
 | 	  /* Display the contents of the Opcodes table.  */ | 
 | 	  standard_opcodes = hdrptr; | 
 |  | 
 | 	  /* PR 17512: file: 002-417945-0.004.  */ | 
 | 	  if (standard_opcodes + linfo.li_opcode_base >= end) | 
 | 	    { | 
 | 	      warn (_("Line Base extends beyond end of section\n")); | 
 | 	      return 0; | 
 | 	    } | 
 |  | 
 | 	  printf (_("\n Opcodes:\n")); | 
 |  | 
 | 	  for (i = 1; i < linfo.li_opcode_base; i++) | 
 | 	    printf (ngettext ("  Opcode %d has %d arg\n", | 
 | 			      "  Opcode %d has %d args\n", | 
 | 			      standard_opcodes[i - 1]), | 
 | 		    i, standard_opcodes[i - 1]); | 
 |  | 
 | 	  /* Display the contents of the Directory table.  */ | 
 | 	  data = standard_opcodes + linfo.li_opcode_base - 1; | 
 |  | 
 | 	  if (linfo.li_version >= 5) | 
 | 	    { | 
 | 	      load_debug_section_with_follow (line_str, file); | 
 |  | 
 | 	      data = display_formatted_table (data, start, end, &linfo, section, | 
 | 					      true); | 
 | 	      data = display_formatted_table (data, start, end, &linfo, section, | 
 | 					      false); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      if (*data == 0) | 
 | 		printf (_("\n The Directory Table is empty.\n")); | 
 | 	      else | 
 | 		{ | 
 | 		  unsigned int last_dir_entry = 0; | 
 |  | 
 | 		  printf (_("\n The Directory Table (offset %#tx):\n"), | 
 | 			  data - start); | 
 |  | 
 | 		  while (data < end && *data != 0) | 
 | 		    { | 
 | 		      printf ("  %d\t%.*s\n", ++last_dir_entry, (int) (end - data), data); | 
 |  | 
 | 		      data += strnlen ((char *) data, end - data); | 
 | 		      if (data < end) | 
 | 			data++; | 
 | 		    } | 
 |  | 
 | 		  /* PR 17512: file: 002-132094-0.004.  */ | 
 | 		  if (data >= end - 1) | 
 | 		    break; | 
 | 		} | 
 |  | 
 | 	      /* Skip the NUL at the end of the table.  */ | 
 | 	      if (data < end) | 
 | 		data++; | 
 |  | 
 | 	      /* Display the contents of the File Name table.  */ | 
 | 	      if (data >= end || *data == 0) | 
 | 		printf (_("\n The File Name Table is empty.\n")); | 
 | 	      else | 
 | 		{ | 
 | 		  printf (_("\n The File Name Table (offset %#tx):\n"), | 
 | 			  data - start); | 
 | 		  printf (_("  Entry\tDir\tTime\tSize\tName\n")); | 
 |  | 
 | 		  while (data < end && *data != 0) | 
 | 		    { | 
 | 		      unsigned char *name; | 
 | 		      uint64_t val; | 
 |  | 
 | 		      printf ("  %d\t", ++state_machine_regs.last_file_entry); | 
 | 		      name = data; | 
 | 		      data += strnlen ((char *) data, end - data); | 
 | 		      if (data < end) | 
 | 			data++; | 
 |  | 
 | 		      READ_ULEB (val, data, end); | 
 | 		      printf ("%" PRIu64 "\t", val); | 
 | 		      READ_ULEB (val, data, end); | 
 | 		      printf ("%" PRIu64 "\t", val); | 
 | 		      READ_ULEB (val, data, end); | 
 | 		      printf ("%" PRIu64 "\t", val); | 
 | 		      printf ("%.*s\n", (int)(end - name), name); | 
 |  | 
 | 		      if (data >= end) | 
 | 			{ | 
 | 			  warn (_("Corrupt file name table entry\n")); | 
 | 			  break; | 
 | 			} | 
 | 		    } | 
 | 		} | 
 |  | 
 | 	      /* Skip the NUL at the end of the table.  */ | 
 | 	      if (data < end) | 
 | 		data++; | 
 | 	    } | 
 |  | 
 | 	  putchar ('\n'); | 
 | 	  saved_linfo = linfo; | 
 | 	} | 
 |  | 
 |       /* Now display the statements.  */ | 
 |       if (data >= end_of_sequence) | 
 | 	printf (_(" No Line Number Statements.\n")); | 
 |       else | 
 | 	{ | 
 | 	  printf (_(" Line Number Statements:\n")); | 
 |  | 
 | 	  while (data < end_of_sequence) | 
 | 	    { | 
 | 	      unsigned char op_code; | 
 | 	      int adv; | 
 | 	      uint64_t uladv; | 
 |  | 
 | 	      printf ("  [0x%08tx]", data - start); | 
 |  | 
 | 	      op_code = *data++; | 
 |  | 
 | 	      if (op_code >= linfo.li_opcode_base) | 
 | 		{ | 
 | 		  op_code -= linfo.li_opcode_base; | 
 | 		  uladv = (op_code / linfo.li_line_range); | 
 | 		  if (linfo.li_max_ops_per_insn == 1) | 
 | 		    { | 
 | 		      uladv *= linfo.li_min_insn_length; | 
 | 		      state_machine_regs.address += uladv; | 
 | 		      if (uladv) | 
 | 			state_machine_regs.view = 0; | 
 | 		      printf (_("  Special opcode %d: " | 
 | 				"advance Address by %" PRIu64 | 
 | 				" to %#" PRIx64 "%s"), | 
 | 			      op_code, uladv, state_machine_regs.address, | 
 | 			      verbose_view && uladv | 
 | 			      ? _(" (reset view)") : ""); | 
 | 		    } | 
 | 		  else | 
 | 		    { | 
 | 		      unsigned addrdelta | 
 | 			= ((state_machine_regs.op_index + uladv) | 
 | 			    / linfo.li_max_ops_per_insn) | 
 | 			* linfo.li_min_insn_length; | 
 |  | 
 | 		      state_machine_regs.address += addrdelta; | 
 | 		      state_machine_regs.op_index | 
 | 			= (state_machine_regs.op_index + uladv) | 
 | 			% linfo.li_max_ops_per_insn; | 
 | 		      if (addrdelta) | 
 | 			state_machine_regs.view = 0; | 
 | 		      printf (_("  Special opcode %d: " | 
 | 				"advance Address by %" PRIu64 | 
 | 				" to %#" PRIx64 "[%d]%s"), | 
 | 			      op_code, uladv, state_machine_regs.address, | 
 | 			      state_machine_regs.op_index, | 
 | 			      verbose_view && addrdelta | 
 | 			      ? _(" (reset view)") : ""); | 
 | 		    } | 
 | 		  adv = (op_code % linfo.li_line_range) + linfo.li_line_base; | 
 | 		  state_machine_regs.line += adv; | 
 | 		  printf (_(" and Line by %d to %d"), | 
 | 			  adv, state_machine_regs.line); | 
 | 		  if (verbose_view || state_machine_regs.view) | 
 | 		    printf (_(" (view %u)\n"), state_machine_regs.view); | 
 | 		  else | 
 | 		    putchar ('\n'); | 
 | 		  state_machine_regs.view++; | 
 | 		} | 
 | 	      else | 
 | 		switch (op_code) | 
 | 		  { | 
 | 		  case DW_LNS_extended_op: | 
 | 		    data += process_extended_line_op (data, | 
 | 						      linfo.li_default_is_stmt, | 
 | 						      end); | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_copy: | 
 | 		    printf (_("  Copy")); | 
 | 		    if (verbose_view || state_machine_regs.view) | 
 | 		      printf (_(" (view %u)\n"), state_machine_regs.view); | 
 | 		    else | 
 | 		      putchar ('\n'); | 
 | 		    state_machine_regs.view++; | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_advance_pc: | 
 | 		    READ_ULEB (uladv, data, end); | 
 | 		    if (linfo.li_max_ops_per_insn == 1) | 
 | 		      { | 
 | 			uladv *= linfo.li_min_insn_length; | 
 | 			state_machine_regs.address += uladv; | 
 | 			if (uladv) | 
 | 			  state_machine_regs.view = 0; | 
 | 			printf (_("  Advance PC by %" PRIu64 | 
 | 				  " to %#" PRIx64 "%s\n"), | 
 | 				uladv, state_machine_regs.address, | 
 | 				verbose_view && uladv | 
 | 				? _(" (reset view)") : ""); | 
 | 		      } | 
 | 		    else | 
 | 		      { | 
 | 			unsigned addrdelta | 
 | 			  = ((state_machine_regs.op_index + uladv) | 
 | 			     / linfo.li_max_ops_per_insn) | 
 | 			  * linfo.li_min_insn_length; | 
 | 			state_machine_regs.address | 
 | 			  += addrdelta; | 
 | 			state_machine_regs.op_index | 
 | 			  = (state_machine_regs.op_index + uladv) | 
 | 			  % linfo.li_max_ops_per_insn; | 
 | 			if (addrdelta) | 
 | 			  state_machine_regs.view = 0; | 
 | 			printf (_("  Advance PC by %" PRIu64 | 
 | 				  " to %#" PRIx64 "[%d]%s\n"), | 
 | 				uladv, state_machine_regs.address, | 
 | 				state_machine_regs.op_index, | 
 | 				verbose_view && addrdelta | 
 | 				? _(" (reset view)") : ""); | 
 | 		      } | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_advance_line: | 
 | 		    READ_SLEB (adv, data, end); | 
 | 		    state_machine_regs.line += adv; | 
 | 		    printf (_("  Advance Line by %d to %d\n"), | 
 | 			    adv, state_machine_regs.line); | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_file: | 
 | 		    READ_ULEB (uladv, data, end); | 
 | 		    printf (_("  Set File Name to entry %" PRIu64 | 
 | 			      " in the File Name Table\n"), uladv); | 
 | 		    state_machine_regs.file = uladv; | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_column: | 
 | 		    READ_ULEB (uladv, data, end); | 
 | 		    printf (_("  Set column to %" PRIu64 "\n"), uladv); | 
 | 		    state_machine_regs.column = uladv; | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_negate_stmt: | 
 | 		    adv = state_machine_regs.is_stmt; | 
 | 		    adv = ! adv; | 
 | 		    printf (_("  Set is_stmt to %d\n"), adv); | 
 | 		    state_machine_regs.is_stmt = adv; | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_basic_block: | 
 | 		    printf (_("  Set basic block\n")); | 
 | 		    state_machine_regs.basic_block = 1; | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_const_add_pc: | 
 | 		    uladv = ((255 - linfo.li_opcode_base) / linfo.li_line_range); | 
 | 		    if (linfo.li_max_ops_per_insn) | 
 | 		      { | 
 | 			uladv *= linfo.li_min_insn_length; | 
 | 			state_machine_regs.address += uladv; | 
 | 			if (uladv) | 
 | 			  state_machine_regs.view = 0; | 
 | 			printf (_("  Advance PC by constant %" PRIu64 | 
 | 				  " to %#" PRIx64 "%s\n"), | 
 | 				uladv, state_machine_regs.address, | 
 | 				verbose_view && uladv | 
 | 				? _(" (reset view)") : ""); | 
 | 		      } | 
 | 		    else | 
 | 		      { | 
 | 			unsigned addrdelta | 
 | 			  = ((state_machine_regs.op_index + uladv) | 
 | 			     / linfo.li_max_ops_per_insn) | 
 | 			  * linfo.li_min_insn_length; | 
 | 			state_machine_regs.address | 
 | 			  += addrdelta; | 
 | 			state_machine_regs.op_index | 
 | 			  = (state_machine_regs.op_index + uladv) | 
 | 			  % linfo.li_max_ops_per_insn; | 
 | 			if (addrdelta) | 
 | 			  state_machine_regs.view = 0; | 
 | 			printf (_("  Advance PC by constant %" PRIu64 | 
 | 				  " to %#" PRIx64 "[%d]%s\n"), | 
 | 				uladv, state_machine_regs.address, | 
 | 				state_machine_regs.op_index, | 
 | 				verbose_view && addrdelta | 
 | 				? _(" (reset view)") : ""); | 
 | 		      } | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_fixed_advance_pc: | 
 | 		    SAFE_BYTE_GET_AND_INC (uladv, data, 2, end); | 
 | 		    state_machine_regs.address += uladv; | 
 | 		    state_machine_regs.op_index = 0; | 
 | 		    printf (_("  Advance PC by fixed size amount %" PRIu64 | 
 | 			      " to %#" PRIx64 "\n"), | 
 | 			    uladv, state_machine_regs.address); | 
 | 		    /* Do NOT reset view.  */ | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_prologue_end: | 
 | 		    printf (_("  Set prologue_end to true\n")); | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_epilogue_begin: | 
 | 		    printf (_("  Set epilogue_begin to true\n")); | 
 | 		    break; | 
 |  | 
 | 		  case DW_LNS_set_isa: | 
 | 		    READ_ULEB (uladv, data, end); | 
 | 		    printf (_("  Set ISA to %" PRIu64 "\n"), uladv); | 
 | 		    break; | 
 |  | 
 | 		  default: | 
 | 		    printf (_("  Unknown opcode %d with operands: "), op_code); | 
 |  | 
 | 		    if (standard_opcodes != NULL) | 
 | 		      for (i = standard_opcodes[op_code - 1]; i > 0 ; --i) | 
 | 			{ | 
 | 			  READ_ULEB (uladv, data, end); | 
 | 			  printf ("%#" PRIx64 "%s", uladv, i == 1 ? "" : ", "); | 
 | 			} | 
 | 		    putchar ('\n'); | 
 | 		    break; | 
 | 		  } | 
 | 	    } | 
 | 	  putchar ('\n'); | 
 | 	} | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | typedef struct | 
 | { | 
 |   const char *name; | 
 |   unsigned int directory_index; | 
 |   unsigned int modification_date; | 
 |   unsigned int length; | 
 | } File_Entry; | 
 |  | 
 | /* Output a decoded representation of the .debug_line section.  */ | 
 |  | 
 | static int | 
 | display_debug_lines_decoded (struct dwarf_section *  section, | 
 | 			     unsigned char *         start, | 
 | 			     unsigned char *         data, | 
 | 			     unsigned char *         end, | 
 | 			     void *                  fileptr) | 
 | { | 
 |   static DWARF2_Internal_LineInfo saved_linfo; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (data < end) | 
 |     { | 
 |       /* This loop amounts to one iteration per compilation unit.  */ | 
 |       DWARF2_Internal_LineInfo linfo; | 
 |       unsigned char *standard_opcodes; | 
 |       unsigned char *end_of_sequence; | 
 |       int i; | 
 |       File_Entry *file_table = NULL; | 
 |       unsigned int n_files = 0; | 
 |       const char **directory_table = NULL; | 
 |       unsigned int n_directories = 0; | 
 |  | 
 |       if (startswith (section->name, ".debug_line.") | 
 | 	  /* Note: the following does not apply to .debug_line.dwo sections. | 
 | 	     These are full debug_line sections.  */ | 
 | 	  && strcmp (section->name, ".debug_line.dwo") != 0) | 
 | 	{ | 
 | 	  /* See comment in display_debug_lines_raw().  */ | 
 | 	  end_of_sequence = end; | 
 | 	  standard_opcodes = NULL; | 
 | 	  linfo = saved_linfo; | 
 | 	  /* PR 17531: file: 0522b371.  */ | 
 | 	  if (linfo.li_line_range == 0) | 
 | 	    { | 
 | 	      warn (_("Partial .debug_line. section encountered without a prior full .debug_line section\n")); | 
 | 	      return 0; | 
 | 	    } | 
 | 	  reset_state_machine (linfo.li_default_is_stmt); | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  unsigned char *hdrptr; | 
 |  | 
 | 	  if ((hdrptr = read_debug_line_header (section, data, end, & linfo, | 
 | 						& end_of_sequence)) == NULL) | 
 | 	      return 0; | 
 |  | 
 | 	  /* PR 17531: file: 0522b371.  */ | 
 | 	  if (linfo.li_line_range == 0) | 
 | 	    { | 
 | 	      warn (_("Line range of 0 is invalid, using 1 instead\n")); | 
 | 	      linfo.li_line_range = 1; | 
 | 	    } | 
 | 	  reset_state_machine (linfo.li_default_is_stmt); | 
 |  | 
 | 	  /* Save a pointer to the contents of the Opcodes table.  */ | 
 | 	  standard_opcodes = hdrptr; | 
 |  | 
 | 	  /* Traverse the Directory table just to count entries.  */ | 
 | 	  data = standard_opcodes + linfo.li_opcode_base - 1; | 
 | 	  /* PR 20440 */ | 
 | 	  if (data >= end) | 
 | 	    { | 
 | 	      warn (_("opcode base of %d extends beyond end of section\n"), | 
 | 		    linfo.li_opcode_base); | 
 | 	      return 0; | 
 | 	    } | 
 |  | 
 | 	  if (linfo.li_version >= 5) | 
 | 	    { | 
 | 	      unsigned char *format_start, *format; | 
 | 	      unsigned int format_count, formati, entryi; | 
 |  | 
 | 	      load_debug_section_with_follow (line_str, fileptr); | 
 |  | 
 | 	      /* Skip directories format.  */ | 
 | 	      SAFE_BYTE_GET_AND_INC (format_count, data, 1, end); | 
 | 	      if (do_checks && format_count > 1) | 
 | 		warn (_("Unexpectedly large number of columns in the directory name table (%u)\n"), | 
 | 		      format_count); | 
 | 	      format_start = data; | 
 | 	      for (formati = 0; formati < format_count; formati++) | 
 | 		{ | 
 | 		  SKIP_ULEB (data, end); | 
 | 		  SKIP_ULEB (data, end); | 
 | 		} | 
 |  | 
 | 	      READ_ULEB (n_directories, data, end); | 
 | 	      if (data >= end) | 
 | 		{ | 
 | 		  warn (_("Corrupt directories list\n")); | 
 | 		  break; | 
 | 		} | 
 |  | 
 | 	      if (n_directories == 0) | 
 | 		directory_table = NULL; | 
 | 	      else if (n_directories > section->size) | 
 | 		{ | 
 | 		  warn (_("number of directories (0x%x) exceeds size of section %s\n"), | 
 | 			n_directories, section->name); | 
 | 		  return 0; | 
 | 		} | 
 | 	      else | 
 | 		directory_table = (const char **) | 
 | 		  xcalloc (n_directories, sizeof (const char *)); | 
 |  | 
 | 	      for (entryi = 0; entryi < n_directories; entryi++) | 
 | 		{ | 
 | 		  const char **pathp = &directory_table[entryi]; | 
 |  | 
 | 		  format = format_start; | 
 | 		  for (formati = 0; formati < format_count; formati++) | 
 | 		    { | 
 | 		      uint64_t content_type, form; | 
 | 		      uint64_t uvalue; | 
 |  | 
 | 		      READ_ULEB (content_type, format, end); | 
 | 		      READ_ULEB (form, format, end); | 
 | 		      if (data >= end) | 
 | 			{ | 
 | 			  warn (_("Corrupt directories list\n")); | 
 | 			  break; | 
 | 			} | 
 | 		      switch (content_type) | 
 | 			{ | 
 | 			case DW_LNCT_path: | 
 | 			  switch (form) | 
 | 			    { | 
 | 			    case DW_FORM_string: | 
 | 			      *pathp = (char *) data; | 
 | 			      break; | 
 | 			    case DW_FORM_line_strp: | 
 | 			      SAFE_BYTE_GET (uvalue, data, linfo.li_offset_size, | 
 | 					     end); | 
 | 			      /* Remove const by the cast.  */ | 
 | 			      *pathp = fetch_indirect_line_string (uvalue); | 
 | 			      break; | 
 | 			    } | 
 | 			  break; | 
 | 			} | 
 | 		      data = read_and_display_attr_value (0, form, 0, start, | 
 | 							  data, end, 0, 0, | 
 | 							  linfo.li_offset_size, | 
 | 							  linfo.li_version, | 
 | 							  NULL, 1, section, | 
 | 							  NULL, '\t', -1); | 
 | 		    } | 
 | 		  if (data >= end) | 
 | 		    { | 
 | 		      warn (_("Corrupt directories list\n")); | 
 | 		      break; | 
 | 		    } | 
 | 		} | 
 |  | 
 | 	      /* Skip files format.  */ | 
 | 	      SAFE_BYTE_GET_AND_INC (format_count, data, 1, end); | 
 | 	      if (do_checks && format_count > 5) | 
 | 		warn (_("Unexpectedly large number of columns in the file name table (%u)\n"), | 
 | 		      format_count); | 
 |  | 
 | 	      format_start = data; | 
 | 	      for (formati = 0; formati < format_count; formati++) | 
 | 		{ | 
 | 		  SKIP_ULEB (data, end); | 
 | 		  SKIP_ULEB (data, end); | 
 | 		} | 
 |  | 
 | 	      READ_ULEB (n_files, data, end); | 
 | 	      if (data >= end && n_files > 0) | 
 | 		{ | 
 | 		  warn (_("Corrupt file name list\n")); | 
 | 		  break; | 
 | 		} | 
 |  | 
 | 	      if (n_files == 0) | 
 | 		file_table = NULL; | 
 | 	      else if (n_files > section->size) | 
 | 		{ | 
 | 		  warn (_("number of files (0x%x) exceeds size of section %s\n"), | 
 | 			n_files, section->name); | 
 | 		  return 0; | 
 | 		} | 
 | 	      else | 
 | 		file_table = (File_Entry *) xcalloc (n_files, | 
 | 						     sizeof (File_Entry)); | 
 |  | 
 | 	      for (entryi = 0; entryi < n_files; entryi++) | 
 | 		{ | 
 | 		  File_Entry *file = &file_table[entryi]; | 
 |  | 
 | 		  format = format_start; | 
 | 		  for (formati = 0; formati < format_count; formati++) | 
 | 		    { | 
 | 		      uint64_t content_type, form; | 
 | 		      uint64_t uvalue; | 
 | 		      unsigned char *tmp; | 
 |  | 
 | 		      READ_ULEB (content_type, format, end); | 
 | 		      READ_ULEB (form, format, end); | 
 | 		      if (data >= end) | 
 | 			{ | 
 | 			  warn (_("Corrupt file name list\n")); | 
 | 			  break; | 
 | 			} | 
 | 		      switch (content_type) | 
 | 			{ | 
 | 			case DW_LNCT_path: | 
 | 			  switch (form) | 
 | 			    { | 
 | 			    case DW_FORM_string: | 
 | 			      file->name = (char *) data; | 
 | 			      break; | 
 | 			    case DW_FORM_line_strp: | 
 | 			      SAFE_BYTE_GET (uvalue, data, linfo.li_offset_size, | 
 | 					     end); | 
 | 			      /* Remove const by the cast.  */ | 
 | 			      file->name = fetch_indirect_line_string (uvalue); | 
 | 			      break; | 
 | 			    } | 
 | 			  break; | 
 | 			case DW_LNCT_directory_index: | 
 | 			  switch (form) | 
 | 			    { | 
 | 			    case DW_FORM_data1: | 
 | 			      SAFE_BYTE_GET (file->directory_index, data, 1, | 
 | 					     end); | 
 | 			      break; | 
 | 			    case DW_FORM_data2: | 
 | 			      SAFE_BYTE_GET (file->directory_index, data, 2, | 
 | 					     end); | 
 | 			      break; | 
 | 			    case DW_FORM_udata: | 
 | 			      tmp = data; | 
 | 			      READ_ULEB (file->directory_index, tmp, end); | 
 | 			      break; | 
 | 			    } | 
 | 			  break; | 
 | 			} | 
 | 		      data = read_and_display_attr_value (0, form, 0, start, | 
 | 							  data, end, 0, 0, | 
 | 							  linfo.li_offset_size, | 
 | 							  linfo.li_version, | 
 | 							  NULL, 1, section, | 
 | 							  NULL, '\t', -1); | 
 | 		    } | 
 | 		  if (data >= end) | 
 | 		    { | 
 | 		      warn (_("Corrupt file name list\n")); | 
 | 		      break; | 
 | 		    } | 
 | 		} | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      if (*data != 0) | 
 | 		{ | 
 | 		  char *ptr_directory_table = (char *) data; | 
 |  | 
 | 		  while (data < end && *data != 0) | 
 | 		    { | 
 | 		      data += strnlen ((char *) data, end - data); | 
 | 		      if (data < end) | 
 | 			data++; | 
 | 		      n_directories++; | 
 | 		    } | 
 |  | 
 | 		  /* PR 20440 */ | 
 | 		  if (data >= end) | 
 | 		    { | 
 | 		      warn (_("directory table ends unexpectedly\n")); | 
 | 		      n_directories = 0; | 
 | 		      break; | 
 | 		    } | 
 |  | 
 | 		  /* Go through the directory table again to save the directories.  */ | 
 | 		  directory_table = (const char **) | 
 | 		    xmalloc (n_directories * sizeof (const char *)); | 
 |  | 
 | 		  i = 0; | 
 | 		  while (*ptr_directory_table != 0) | 
 | 		    { | 
 | 		      directory_table[i] = ptr_directory_table; | 
 | 		      ptr_directory_table += strlen (ptr_directory_table) + 1; | 
 | 		      i++; | 
 | 		    } | 
 | 		} | 
 | 	      /* Skip the NUL at the end of the table.  */ | 
 | 	      data++; | 
 |  | 
 | 	      /* Traverse the File Name table just to count the entries.  */ | 
 | 	      if (data < end && *data != 0) | 
 | 		{ | 
 | 		  unsigned char *ptr_file_name_table = data; | 
 |  | 
 | 		  while (data < end && *data != 0) | 
 | 		    { | 
 | 		      /* Skip Name, directory index, last modification | 
 | 			 time and length of file.  */ | 
 | 		      data += strnlen ((char *) data, end - data); | 
 | 		      if (data < end) | 
 | 			data++; | 
 | 		      SKIP_ULEB (data, end); | 
 | 		      SKIP_ULEB (data, end); | 
 | 		      SKIP_ULEB (data, end); | 
 | 		      n_files++; | 
 | 		    } | 
 |  | 
 | 		  if (data >= end) | 
 | 		    { | 
 | 		      warn (_("file table ends unexpectedly\n")); | 
 | 		      n_files = 0; | 
 | 		      break; | 
 | 		    } | 
 |  | 
 | 		  /* Go through the file table again to save the strings.  */ | 
 | 		  file_table = (File_Entry *) xmalloc (n_files * sizeof (File_Entry)); | 
 |  | 
 | 		  i = 0; | 
 | 		  while (*ptr_file_name_table != 0) | 
 | 		    { | 
 | 		      file_table[i].name = (char *) ptr_file_name_table; | 
 | 		      ptr_file_name_table | 
 | 			+= strlen ((char *) ptr_file_name_table) + 1; | 
 |  | 
 | 		      /* We are not interested in directory, time or size.  */ | 
 | 		      READ_ULEB (file_table[i].directory_index, | 
 | 				 ptr_file_name_table, end); | 
 | 		      READ_ULEB (file_table[i].modification_date, | 
 | 				 ptr_file_name_table, end); | 
 | 		      READ_ULEB (file_table[i].length, | 
 | 				 ptr_file_name_table, end); | 
 | 		      i++; | 
 | 		    } | 
 | 		  i = 0; | 
 | 		} | 
 |  | 
 | 	      /* Skip the NUL at the end of the table.  */ | 
 | 	      data++; | 
 | 	    } | 
 |  | 
 | 	  /* Print the Compilation Unit's name and a header.  */ | 
 | 	  if (file_table == NULL) | 
 | 	    printf (_("CU: No directory table\n")); | 
 | 	  else if (directory_table == NULL) | 
 | 	    printf (_("CU: %s:\n"), null_name (file_table[0].name)); | 
 | 	  else | 
 | 	    { | 
 | 	      unsigned int ix = file_table[0].directory_index; | 
 | 	      const char *directory; | 
 |  | 
 | 	      if (ix == 0 && linfo.li_version < 5) | 
 | 		directory = "."; | 
 | 	      /* PR 20439 */ | 
 | 	      else if (n_directories == 0) | 
 | 		directory = _("<unknown>"); | 
 | 	      else | 
 | 		{ | 
 | 		  if (linfo.li_version < 5) | 
 | 		    --ix; | 
 | 		  if (ix >= n_directories) | 
 | 		    { | 
 | 		      warn (_("directory index %u " | 
 | 			      ">= number of directories %u\n"), | 
 | 			    ix, n_directories); | 
 | 		      directory = _("<corrupt>"); | 
 | 		    } | 
 | 		  else | 
 | 		    directory = directory_table[ix]; | 
 | 		} | 
 | 	      if (do_wide) | 
 | 		printf (_("CU: %s/%s:\n"), | 
 | 			null_name (directory), | 
 | 			null_name (file_table[0].name)); | 
 | 	      else | 
 | 		printf ("%s:\n", null_name (file_table[0].name)); | 
 | 	    } | 
 |  | 
 | 	  if (n_files > 0) | 
 | 	    { | 
 | 	      if (do_wide) | 
 | 		printf (_("File name                            Line number    Starting address    View    Stmt\n")); | 
 | 	      else | 
 | 		printf (_("File name                        Line number    Starting address    View    Stmt\n")); | 
 | 	    } | 
 | 	  else | 
 | 	    printf (_("CU: Empty file name table\n")); | 
 | 	  saved_linfo = linfo; | 
 | 	} | 
 |  | 
 |       /* This loop iterates through the Dwarf Line Number Program.  */ | 
 |       while (data < end_of_sequence) | 
 | 	{ | 
 | 	  unsigned char op_code; | 
 | 	  int xop; | 
 | 	  int adv; | 
 | 	  unsigned long int uladv; | 
 | 	  int is_special_opcode = 0; | 
 |  | 
 | 	  op_code = *data++; | 
 | 	  xop = op_code; | 
 |  | 
 | 	  if (op_code >= linfo.li_opcode_base) | 
 | 	    { | 
 | 	      op_code -= linfo.li_opcode_base; | 
 | 	      uladv = (op_code / linfo.li_line_range); | 
 | 	      if (linfo.li_max_ops_per_insn == 1) | 
 | 		{ | 
 | 		  uladv *= linfo.li_min_insn_length; | 
 | 		  state_machine_regs.address += uladv; | 
 | 		  if (uladv) | 
 | 		    state_machine_regs.view = 0; | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  unsigned addrdelta | 
 | 		    = ((state_machine_regs.op_index + uladv) | 
 | 		       / linfo.li_max_ops_per_insn) | 
 | 		    * linfo.li_min_insn_length; | 
 | 		  state_machine_regs.address | 
 | 		    += addrdelta; | 
 | 		  state_machine_regs.op_index | 
 | 		    = (state_machine_regs.op_index + uladv) | 
 | 		    % linfo.li_max_ops_per_insn; | 
 | 		  if (addrdelta) | 
 | 		    state_machine_regs.view = 0; | 
 | 		} | 
 |  | 
 | 	      adv = (op_code % linfo.li_line_range) + linfo.li_line_base; | 
 | 	      state_machine_regs.line += adv; | 
 | 	      is_special_opcode = 1; | 
 | 	      /* Increment view after printing this row.  */ | 
 | 	    } | 
 | 	  else | 
 | 	    switch (op_code) | 
 | 	      { | 
 | 	      case DW_LNS_extended_op: | 
 | 		{ | 
 | 		  unsigned int ext_op_code_len; | 
 | 		  unsigned char ext_op_code; | 
 | 		  unsigned char *op_code_end; | 
 | 		  unsigned char *op_code_data = data; | 
 |  | 
 | 		  READ_ULEB (ext_op_code_len, op_code_data, end_of_sequence); | 
 | 		  op_code_end = op_code_data + ext_op_code_len; | 
 | 		  if (ext_op_code_len == 0 || op_code_end > end_of_sequence) | 
 | 		    { | 
 | 		      warn (_("Badly formed extended line op encountered!\n")); | 
 | 		      break; | 
 | 		    } | 
 | 		  ext_op_code = *op_code_data++; | 
 | 		  xop = ext_op_code; | 
 | 		  xop = -xop; | 
 |  | 
 | 		  switch (ext_op_code) | 
 | 		    { | 
 | 		    case DW_LNE_end_sequence: | 
 | 		      /* Reset stuff after printing this row.  */ | 
 | 		      break; | 
 | 		    case DW_LNE_set_address: | 
 | 		      SAFE_BYTE_GET_AND_INC (state_machine_regs.address, | 
 | 					     op_code_data, | 
 | 					     op_code_end - op_code_data, | 
 | 					     op_code_end); | 
 | 		      state_machine_regs.op_index = 0; | 
 | 		      state_machine_regs.view = 0; | 
 | 		      break; | 
 | 		    case DW_LNE_define_file: | 
 | 		      file_table = (File_Entry *) xrealloc | 
 | 			(file_table, (n_files + 1) * sizeof (File_Entry)); | 
 |  | 
 | 		      ++state_machine_regs.last_file_entry; | 
 | 		      /* Source file name.  */ | 
 | 		      file_table[n_files].name = (char *) op_code_data; | 
 | 		      op_code_data += strlen ((char *) op_code_data) + 1; | 
 | 		      /* Directory index.  */ | 
 | 		      READ_ULEB (file_table[n_files].directory_index, | 
 | 				 op_code_data, op_code_end); | 
 | 		      /* Last modification time.  */ | 
 | 		      READ_ULEB (file_table[n_files].modification_date, | 
 | 				 op_code_data, op_code_end); | 
 | 		      /* File length.  */ | 
 | 		      READ_ULEB (file_table[n_files].length, | 
 | 				 op_code_data, op_code_end); | 
 | 		      n_files++; | 
 | 		      break; | 
 |  | 
 | 		    case DW_LNE_set_discriminator: | 
 | 		    case DW_LNE_HP_set_sequence: | 
 | 		      /* Simply ignored.  */ | 
 | 		      break; | 
 |  | 
 | 		    default: | 
 | 		      printf (_("UNKNOWN (%u): length %ld\n"), | 
 | 			      ext_op_code, (long int) (op_code_data - data)); | 
 | 		      break; | 
 | 		    } | 
 | 		  data = op_code_end; | 
 | 		  break; | 
 | 		} | 
 | 	      case DW_LNS_copy: | 
 | 		/* Increment view after printing this row.  */ | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_advance_pc: | 
 | 		READ_ULEB (uladv, data, end); | 
 | 		if (linfo.li_max_ops_per_insn == 1) | 
 | 		  { | 
 | 		    uladv *= linfo.li_min_insn_length; | 
 | 		    state_machine_regs.address += uladv; | 
 | 		    if (uladv) | 
 | 		      state_machine_regs.view = 0; | 
 | 		  } | 
 | 		else | 
 | 		  { | 
 | 		    unsigned addrdelta | 
 | 		      = ((state_machine_regs.op_index + uladv) | 
 | 			 / linfo.li_max_ops_per_insn) | 
 | 		      * linfo.li_min_insn_length; | 
 | 		    state_machine_regs.address | 
 | 		      += addrdelta; | 
 | 		    state_machine_regs.op_index | 
 | 		      = (state_machine_regs.op_index + uladv) | 
 | 		      % linfo.li_max_ops_per_insn; | 
 | 		    if (addrdelta) | 
 | 		      state_machine_regs.view = 0; | 
 | 		  } | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_advance_line: | 
 | 		READ_SLEB (adv, data, end); | 
 | 		state_machine_regs.line += adv; | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_file: | 
 | 		READ_ULEB (uladv, data, end); | 
 | 		state_machine_regs.file = uladv; | 
 |  | 
 | 		unsigned file = state_machine_regs.file; | 
 | 		if (linfo.li_version < 5) | 
 | 		  --file; | 
 |  | 
 | 		if (file_table == NULL || n_files == 0) | 
 | 		  printf (_("\n [Use file table entry %d]\n"), file); | 
 | 		/* PR 20439 */ | 
 | 		else if (file >= n_files) | 
 | 		  { | 
 | 		    warn (_("file index %u >= number of files %u\n"), | 
 | 			  file, n_files); | 
 | 		    printf (_("\n <over large file table index %u>"), file); | 
 | 		  } | 
 | 		else | 
 | 		  { | 
 | 		    unsigned dir = file_table[file].directory_index; | 
 | 		    if (dir == 0 && linfo.li_version < 5) | 
 | 		      /* If directory index is 0, that means compilation | 
 | 			 current directory.  bfd/dwarf2.c shows | 
 | 			 DW_AT_comp_dir here but in keeping with the | 
 | 			 readelf practice of minimal interpretation of | 
 | 			 file data, we show "./".  */ | 
 | 		      printf ("\n./%s:[++]\n", | 
 | 			      null_name (file_table[file].name)); | 
 | 		    else if (directory_table == NULL || n_directories == 0) | 
 | 		      printf (_("\n [Use file %s " | 
 | 				"in directory table entry %d]\n"), | 
 | 			      null_name (file_table[file].name), dir); | 
 | 		    else | 
 | 		      { | 
 | 			if (linfo.li_version < 5) | 
 | 			  --dir; | 
 | 			/* PR 20439 */ | 
 | 			if (dir >= n_directories) | 
 | 			  { | 
 | 			    warn (_("directory index %u " | 
 | 				    ">= number of directories %u\n"), | 
 | 				  dir, n_directories); | 
 | 			    printf (_("\n <over large directory table entry " | 
 | 				      "%u>\n"), dir); | 
 | 			  } | 
 | 			else | 
 | 			  printf ("\n%s/%s:\n", | 
 | 				  null_name (directory_table[dir]), | 
 | 				  null_name (file_table[file].name)); | 
 | 		      } | 
 | 		  } | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_column: | 
 | 		READ_ULEB (uladv, data, end); | 
 | 		state_machine_regs.column = uladv; | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_negate_stmt: | 
 | 		adv = state_machine_regs.is_stmt; | 
 | 		adv = ! adv; | 
 | 		state_machine_regs.is_stmt = adv; | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_basic_block: | 
 | 		state_machine_regs.basic_block = 1; | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_const_add_pc: | 
 | 		uladv = ((255 - linfo.li_opcode_base) / linfo.li_line_range); | 
 | 		if (linfo.li_max_ops_per_insn == 1) | 
 | 		  { | 
 | 		    uladv *= linfo.li_min_insn_length; | 
 | 		    state_machine_regs.address += uladv; | 
 | 		    if (uladv) | 
 | 		      state_machine_regs.view = 0; | 
 | 		  } | 
 | 		else | 
 | 		  { | 
 | 		    unsigned addrdelta | 
 | 		      = ((state_machine_regs.op_index + uladv) | 
 | 			 / linfo.li_max_ops_per_insn) | 
 | 		      * linfo.li_min_insn_length; | 
 | 		    state_machine_regs.address | 
 | 		      += addrdelta; | 
 | 		    state_machine_regs.op_index | 
 | 		      = (state_machine_regs.op_index + uladv) | 
 | 		      % linfo.li_max_ops_per_insn; | 
 | 		    if (addrdelta) | 
 | 		      state_machine_regs.view = 0; | 
 | 		  } | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_fixed_advance_pc: | 
 | 		SAFE_BYTE_GET_AND_INC (uladv, data, 2, end); | 
 | 		state_machine_regs.address += uladv; | 
 | 		state_machine_regs.op_index = 0; | 
 | 		/* Do NOT reset view.  */ | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_prologue_end: | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_epilogue_begin: | 
 | 		break; | 
 |  | 
 | 	      case DW_LNS_set_isa: | 
 | 		READ_ULEB (uladv, data, end); | 
 | 		printf (_("  Set ISA to %lu\n"), uladv); | 
 | 		break; | 
 |  | 
 | 	      default: | 
 | 		printf (_("  Unknown opcode %d with operands: "), op_code); | 
 |  | 
 | 		if (standard_opcodes != NULL) | 
 | 		  for (i = standard_opcodes[op_code - 1]; i > 0 ; --i) | 
 | 		    { | 
 | 		      uint64_t val; | 
 |  | 
 | 		      READ_ULEB (val, data, end); | 
 | 		      printf ("%#" PRIx64 "%s", val, i == 1 ? "" : ", "); | 
 | 		    } | 
 | 		putchar ('\n'); | 
 | 		break; | 
 | 	      } | 
 |  | 
 | 	  /* Only Special opcodes, DW_LNS_copy and DW_LNE_end_sequence adds a row | 
 | 	     to the DWARF address/line matrix.  */ | 
 | 	  if ((is_special_opcode) || (xop == -DW_LNE_end_sequence) | 
 | 	      || (xop == DW_LNS_copy)) | 
 | 	    { | 
 | 	      const unsigned int MAX_FILENAME_LENGTH = 35; | 
 | 	      char *fileName = NULL; | 
 | 	      char *newFileName = NULL; | 
 | 	      size_t fileNameLength; | 
 |  | 
 | 	      if (file_table) | 
 | 		{ | 
 | 		  unsigned indx = state_machine_regs.file; | 
 |  | 
 | 		  if (linfo.li_version < 5) | 
 | 		    --indx; | 
 | 		  /* PR 20439  */ | 
 | 		  if (indx >= n_files) | 
 | 		    { | 
 | 		      warn (_("file index %u >= number of files %u\n"), | 
 | 			    indx, n_files); | 
 | 		      fileName = _("<corrupt>"); | 
 | 		    } | 
 | 		  else | 
 | 		    fileName = (char *) file_table[indx].name; | 
 | 		} | 
 | 	      if (!fileName) | 
 | 		fileName = _("<unknown>"); | 
 |  | 
 | 	      fileNameLength = strlen (fileName); | 
 | 	      newFileName = fileName; | 
 | 	      if (fileNameLength > MAX_FILENAME_LENGTH && !do_wide) | 
 | 		{ | 
 | 		  newFileName = (char *) xmalloc (MAX_FILENAME_LENGTH + 1); | 
 | 		  /* Truncate file name */ | 
 | 		  memcpy (newFileName, | 
 | 			  fileName + fileNameLength - MAX_FILENAME_LENGTH, | 
 | 			  MAX_FILENAME_LENGTH); | 
 | 		  newFileName[MAX_FILENAME_LENGTH] = 0; | 
 | 		} | 
 |  | 
 | 	      /* A row with end_seq set to true has a meaningful address, but | 
 | 		 the other information in the same row is not significant. | 
 | 		 In such a row, print line as "-", and don't print | 
 | 		 view/is_stmt.  */ | 
 | 	      if (!do_wide || fileNameLength <= MAX_FILENAME_LENGTH) | 
 | 		{ | 
 | 		  if (linfo.li_max_ops_per_insn == 1) | 
 | 		    { | 
 | 		      if (xop == -DW_LNE_end_sequence) | 
 | 			printf ("%-31s  %11s  %#18" PRIx64, | 
 | 				newFileName, "-", | 
 | 				state_machine_regs.address); | 
 | 		      else | 
 | 			printf ("%-31s  %11d  %#18" PRIx64, | 
 | 				newFileName, state_machine_regs.line, | 
 | 				state_machine_regs.address); | 
 | 		    } | 
 | 		  else | 
 | 		    { | 
 | 		      if (xop == -DW_LNE_end_sequence) | 
 | 			printf ("%-31s  %11s  %#18" PRIx64 "[%d]", | 
 | 				newFileName, "-", | 
 | 				state_machine_regs.address, | 
 | 				state_machine_regs.op_index); | 
 | 		      else | 
 | 			printf ("%-31s  %11d  %#18" PRIx64 "[%d]", | 
 | 				newFileName, state_machine_regs.line, | 
 | 				state_machine_regs.address, | 
 | 				state_machine_regs.op_index); | 
 | 		    } | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  if (linfo.li_max_ops_per_insn == 1) | 
 | 		    { | 
 | 		      if (xop == -DW_LNE_end_sequence) | 
 | 			printf ("%s  %11s  %#18" PRIx64, | 
 | 				newFileName, "-", | 
 | 				state_machine_regs.address); | 
 | 		      else | 
 | 			printf ("%s  %11d  %#18" PRIx64, | 
 | 				newFileName, state_machine_regs.line, | 
 | 				state_machine_regs.address); | 
 | 		    } | 
 | 		  else | 
 | 		    { | 
 | 		      if (xop == -DW_LNE_end_sequence) | 
 | 			printf ("%s  %11s  %#18" PRIx64 "[%d]", | 
 | 				newFileName, "-", | 
 | 				state_machine_regs.address, | 
 | 				state_machine_regs.op_index); | 
 | 		      else | 
 | 			printf ("%s  %11d  %#18" PRIx64 "[%d]", | 
 | 				newFileName, state_machine_regs.line, | 
 | 				state_machine_regs.address, | 
 | 				state_machine_regs.op_index); | 
 | 		    } | 
 | 		} | 
 |  | 
 | 	      if (xop != -DW_LNE_end_sequence) | 
 | 		{ | 
 | 		  if (state_machine_regs.view) | 
 | 		    printf ("  %6u", state_machine_regs.view); | 
 | 		  else | 
 | 		    printf ("        "); | 
 |  | 
 | 		  if (state_machine_regs.is_stmt) | 
 | 		    printf ("       x"); | 
 | 		} | 
 |  | 
 | 	      putchar ('\n'); | 
 | 	      state_machine_regs.view++; | 
 |  | 
 | 	      if (xop == -DW_LNE_end_sequence) | 
 | 		{ | 
 | 		  reset_state_machine (linfo.li_default_is_stmt); | 
 | 		  putchar ('\n'); | 
 | 		} | 
 |  | 
 | 	      if (newFileName != fileName) | 
 | 		free (newFileName); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       if (file_table) | 
 | 	{ | 
 | 	  free (file_table); | 
 | 	  file_table = NULL; | 
 | 	  n_files = 0; | 
 | 	} | 
 |  | 
 |       if (directory_table) | 
 | 	{ | 
 | 	  free (directory_table); | 
 | 	  directory_table = NULL; | 
 | 	  n_directories = 0; | 
 | 	} | 
 |  | 
 |       putchar ('\n'); | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_lines (struct dwarf_section *section, void *file) | 
 | { | 
 |   unsigned char *data = section->start; | 
 |   unsigned char *end = data + section->size; | 
 |   int retValRaw = 1; | 
 |   int retValDecoded = 1; | 
 |  | 
 |   if (do_debug_lines == 0) | 
 |     do_debug_lines |= FLAG_DEBUG_LINES_RAW; | 
 |  | 
 |   if (do_debug_lines & FLAG_DEBUG_LINES_RAW) | 
 |     retValRaw = display_debug_lines_raw (section, data, end, file); | 
 |  | 
 |   if (do_debug_lines & FLAG_DEBUG_LINES_DECODED) | 
 |     retValDecoded = display_debug_lines_decoded (section, data, data, end, file); | 
 |  | 
 |   if (!retValRaw || !retValDecoded) | 
 |     return 0; | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static debug_info * | 
 | find_debug_info_for_offset (uint64_t offset) | 
 | { | 
 |   unsigned int i; | 
 |  | 
 |   if (num_debug_info_entries == DEBUG_INFO_UNAVAILABLE) | 
 |     return NULL; | 
 |  | 
 |   for (i = 0; i < num_debug_info_entries; i++) | 
 |     if (debug_information[i].cu_offset == offset) | 
 |       return debug_information + i; | 
 |  | 
 |   return NULL; | 
 | } | 
 |  | 
 | static const char * | 
 | get_gdb_index_symbol_kind_name (gdb_index_symbol_kind kind) | 
 | { | 
 |   /* See gdb/gdb-index.h.  */ | 
 |   static const char * const kinds[] = | 
 |   { | 
 |     N_ ("no info"), | 
 |     N_ ("type"), | 
 |     N_ ("variable"), | 
 |     N_ ("function"), | 
 |     N_ ("other"), | 
 |     N_ ("unused5"), | 
 |     N_ ("unused6"), | 
 |     N_ ("unused7") | 
 |   }; | 
 |  | 
 |   return _ (kinds[kind]); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_pubnames_worker (struct dwarf_section *section, | 
 | 			       void *file ATTRIBUTE_UNUSED, | 
 | 			       int is_gnu) | 
 | { | 
 |   DWARF2_Internal_PubNames names; | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |  | 
 |   /* It does not matter if this load fails, | 
 |      we test for that later on.  */ | 
 |   load_debug_info (file); | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (start < end) | 
 |     { | 
 |       unsigned char *data; | 
 |       unsigned long sec_off = start - section->start; | 
 |       unsigned int offset_size; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (names.pn_length, start, 4, end); | 
 |       if (names.pn_length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (names.pn_length, start, 8, end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |  | 
 |       if (names.pn_length > (size_t) (end - start)) | 
 | 	{ | 
 | 	  warn (_("Debug info is corrupted, " | 
 | 		  "%s header at %#lx has length %#" PRIx64 "\n"), | 
 | 		section->name, sec_off, names.pn_length); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       data = start; | 
 |       start += names.pn_length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (names.pn_version, data, 2, start); | 
 |       SAFE_BYTE_GET_AND_INC (names.pn_offset, data, offset_size, start); | 
 |  | 
 |       if (num_debug_info_entries != DEBUG_INFO_UNAVAILABLE | 
 | 	  && num_debug_info_entries > 0 | 
 | 	  && find_debug_info_for_offset (names.pn_offset) == NULL) | 
 | 	warn (_(".debug_info offset of %#" PRIx64 | 
 | 		" in %s section does not point to a CU header.\n"), | 
 | 	      names.pn_offset, section->name); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (names.pn_size, data, offset_size, start); | 
 |  | 
 |       printf (_("  Length:                              %" PRId64 "\n"), | 
 | 	      names.pn_length); | 
 |       printf (_("  Version:                             %d\n"), | 
 | 	      names.pn_version); | 
 |       printf (_("  Offset into .debug_info section:     %#" PRIx64 "\n"), | 
 | 	      names.pn_offset); | 
 |       printf (_("  Size of area in .debug_info section: %" PRId64 "\n"), | 
 | 	      names.pn_size); | 
 |  | 
 |       if (names.pn_version != 2 && names.pn_version != 3) | 
 | 	{ | 
 | 	  static int warned = 0; | 
 |  | 
 | 	  if (! warned) | 
 | 	    { | 
 | 	      warn (_("Only DWARF 2 and 3 pubnames are currently supported\n")); | 
 | 	      warned = 1; | 
 | 	    } | 
 |  | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if (is_gnu) | 
 | 	printf (_("\n    Offset  Kind          Name\n")); | 
 |       else | 
 | 	printf (_("\n    Offset\tName\n")); | 
 |  | 
 |       while (1) | 
 | 	{ | 
 | 	  size_t maxprint; | 
 | 	  uint64_t offset; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (offset, data, offset_size, start); | 
 |  | 
 | 	  if (offset == 0) | 
 | 	    break; | 
 |  | 
 | 	  if (data >= start) | 
 | 	    break; | 
 | 	  maxprint = (start - data) - 1; | 
 |  | 
 | 	  if (is_gnu) | 
 | 	    { | 
 | 	      unsigned int kind_data; | 
 | 	      gdb_index_symbol_kind kind; | 
 | 	      const char *kind_name; | 
 | 	      int is_static; | 
 |  | 
 | 	      SAFE_BYTE_GET_AND_INC (kind_data, data, 1, start); | 
 | 	      maxprint --; | 
 | 	      /* GCC computes the kind as the upper byte in the CU index | 
 | 		 word, and then right shifts it by the CU index size. | 
 | 		 Left shift KIND to where the gdb-index.h accessor macros | 
 | 		 can use it.  */ | 
 | 	      kind_data <<= GDB_INDEX_CU_BITSIZE; | 
 | 	      kind = GDB_INDEX_SYMBOL_KIND_VALUE (kind_data); | 
 | 	      kind_name = get_gdb_index_symbol_kind_name (kind); | 
 | 	      is_static = GDB_INDEX_SYMBOL_STATIC_VALUE (kind_data); | 
 | 	      printf ("    %-6" PRIx64 "  %s,%-10s  %.*s\n", | 
 | 		      offset, is_static ? _("s") : _("g"), | 
 | 		      kind_name, (int) maxprint, data); | 
 | 	    } | 
 | 	  else | 
 | 	    printf ("    %-6" PRIx64 "\t%.*s\n", | 
 | 		    offset, (int) maxprint, data); | 
 |  | 
 | 	  data += strnlen ((char *) data, maxprint); | 
 | 	  if (data < start) | 
 | 	    data++; | 
 | 	  if (data >= start) | 
 | 	    break; | 
 | 	} | 
 |     } | 
 |  | 
 |   printf ("\n"); | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_pubnames (struct dwarf_section *section, void *file) | 
 | { | 
 |   return display_debug_pubnames_worker (section, file, 0); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_gnu_pubnames (struct dwarf_section *section, void *file) | 
 | { | 
 |   return display_debug_pubnames_worker (section, file, 1); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_macinfo (struct dwarf_section *section, | 
 | 		       void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |   unsigned char *curr = start; | 
 |   enum dwarf_macinfo_record_type op; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (curr < end) | 
 |     { | 
 |       unsigned int lineno; | 
 |       const unsigned char *string; | 
 |  | 
 |       op = (enum dwarf_macinfo_record_type) *curr; | 
 |       curr++; | 
 |  | 
 |       switch (op) | 
 | 	{ | 
 | 	case DW_MACINFO_start_file: | 
 | 	  { | 
 | 	    unsigned int filenum; | 
 |  | 
 | 	    READ_ULEB (lineno, curr, end); | 
 | 	    READ_ULEB (filenum, curr, end); | 
 | 	    printf (_(" DW_MACINFO_start_file - lineno: %d filenum: %d\n"), | 
 | 		    lineno, filenum); | 
 | 	  } | 
 | 	  break; | 
 |  | 
 | 	case DW_MACINFO_end_file: | 
 | 	  printf (_(" DW_MACINFO_end_file\n")); | 
 | 	  break; | 
 |  | 
 | 	case DW_MACINFO_define: | 
 | 	  READ_ULEB (lineno, curr, end); | 
 | 	  string = curr; | 
 | 	  curr += strnlen ((char *) string, end - string); | 
 | 	  printf (_(" DW_MACINFO_define - lineno : %d macro : %*s\n"), | 
 | 		  lineno, (int) (curr - string), string); | 
 | 	  if (curr < end) | 
 | 	    curr++; | 
 | 	  break; | 
 |  | 
 | 	case DW_MACINFO_undef: | 
 | 	  READ_ULEB (lineno, curr, end); | 
 | 	  string = curr; | 
 | 	  curr += strnlen ((char *) string, end - string); | 
 | 	  printf (_(" DW_MACINFO_undef - lineno : %d macro : %*s\n"), | 
 | 		  lineno, (int) (curr - string), string); | 
 | 	  if (curr < end) | 
 | 	    curr++; | 
 | 	  break; | 
 |  | 
 | 	case DW_MACINFO_vendor_ext: | 
 | 	  { | 
 | 	    unsigned int constant; | 
 |  | 
 | 	    READ_ULEB (constant, curr, end); | 
 | 	    string = curr; | 
 | 	    curr += strnlen ((char *) string, end - string); | 
 | 	    printf (_(" DW_MACINFO_vendor_ext - constant : %d string : %*s\n"), | 
 | 		    constant, (int) (curr - string), string); | 
 | 	    if (curr < end) | 
 | 	      curr++; | 
 | 	  } | 
 | 	  break; | 
 | 	} | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Given LINE_OFFSET into the .debug_line section, attempt to return | 
 |    filename and dirname corresponding to file name table entry with index | 
 |    FILEIDX.  Return NULL on failure.  */ | 
 |  | 
 | static unsigned char * | 
 | get_line_filename_and_dirname (uint64_t line_offset, | 
 | 			       uint64_t fileidx, | 
 | 			       unsigned char **dir_name) | 
 | { | 
 |   struct dwarf_section *section = &debug_displays [line].section; | 
 |   unsigned char *hdrptr, *dirtable, *file_name; | 
 |   unsigned int offset_size; | 
 |   unsigned int version, opcode_base; | 
 |   uint64_t length, diridx; | 
 |   const unsigned char * end; | 
 |  | 
 |   *dir_name = NULL; | 
 |   if (section->start == NULL | 
 |       || line_offset >= section->size | 
 |       || fileidx == 0) | 
 |     return NULL; | 
 |  | 
 |   hdrptr = section->start + line_offset; | 
 |   end = section->start + section->size; | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (length, hdrptr, 4, end); | 
 |   if (length == 0xffffffff) | 
 |     { | 
 |       /* This section is 64-bit DWARF 3.  */ | 
 |       SAFE_BYTE_GET_AND_INC (length, hdrptr, 8, end); | 
 |       offset_size = 8; | 
 |     } | 
 |   else | 
 |     offset_size = 4; | 
 |  | 
 |   if (length > (size_t) (end - hdrptr) | 
 |       || length < 2 + offset_size + 1 + 3 + 1) | 
 |     return NULL; | 
 |   end = hdrptr + length; | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (version, hdrptr, 2, end); | 
 |   if (version != 2 && version != 3 && version != 4) | 
 |     return NULL; | 
 |   hdrptr += offset_size + 1;/* Skip prologue_length and min_insn_length.  */ | 
 |   if (version >= 4) | 
 |     hdrptr++;		    /* Skip max_ops_per_insn.  */ | 
 |   hdrptr += 3;		    /* Skip default_is_stmt, line_base, line_range.  */ | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (opcode_base, hdrptr, 1, end); | 
 |   if (opcode_base == 0 | 
 |       || opcode_base - 1 >= (size_t) (end - hdrptr)) | 
 |     return NULL; | 
 |  | 
 |   hdrptr += opcode_base - 1; | 
 |  | 
 |   dirtable = hdrptr; | 
 |   /* Skip over dirname table.  */ | 
 |   while (*hdrptr != '\0') | 
 |     { | 
 |       hdrptr += strnlen ((char *) hdrptr, end - hdrptr); | 
 |       if (hdrptr < end) | 
 | 	hdrptr++; | 
 |       if (hdrptr >= end) | 
 | 	return NULL; | 
 |     } | 
 |   hdrptr++;		    /* Skip the NUL at the end of the table.  */ | 
 |  | 
 |   /* Now skip over preceding filename table entries.  */ | 
 |   for (; hdrptr < end && *hdrptr != '\0' && fileidx > 1; fileidx--) | 
 |     { | 
 |       hdrptr += strnlen ((char *) hdrptr, end - hdrptr); | 
 |       if (hdrptr < end) | 
 | 	hdrptr++; | 
 |       SKIP_ULEB (hdrptr, end); | 
 |       SKIP_ULEB (hdrptr, end); | 
 |       SKIP_ULEB (hdrptr, end); | 
 |     } | 
 |   if (hdrptr >= end || *hdrptr == '\0') | 
 |     return NULL; | 
 |  | 
 |   file_name = hdrptr; | 
 |   hdrptr += strnlen ((char *) hdrptr, end - hdrptr); | 
 |   if (hdrptr < end) | 
 |     hdrptr++; | 
 |   if (hdrptr >= end) | 
 |     return NULL; | 
 |   READ_ULEB (diridx, hdrptr, end); | 
 |   if (diridx == 0) | 
 |     return file_name; | 
 |   for (; dirtable < end && *dirtable != '\0' && diridx > 1; diridx--) | 
 |     { | 
 |       dirtable += strnlen ((char *) dirtable, end - dirtable); | 
 |       if (dirtable < end) | 
 | 	dirtable++; | 
 |     } | 
 |   if (dirtable >= end || *dirtable == '\0') | 
 |     return NULL; | 
 |   *dir_name = dirtable; | 
 |   return file_name; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_macro (struct dwarf_section *section, | 
 | 		     void *file) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |   unsigned char *curr = start; | 
 |   unsigned char *extended_op_buf[256]; | 
 |   bool is_dwo = false; | 
 |   const char *suffix = strrchr (section->name, '.'); | 
 |  | 
 |   if (suffix && strcmp (suffix, ".dwo") == 0) | 
 |     is_dwo = true; | 
 |  | 
 |   if (is_dwo) | 
 |     { | 
 |       load_debug_section_with_follow (str_dwo, file); | 
 |       load_debug_section_with_follow (str_index_dwo, file); | 
 |     } | 
 |   else | 
 |     { | 
 |       load_debug_section_with_follow (str, file); | 
 |       load_debug_section_with_follow (str_index, file); | 
 |     } | 
 |   load_debug_section_with_follow (line, file); | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (curr < end) | 
 |     { | 
 |       unsigned int lineno, version, flags; | 
 |       unsigned int offset_size; | 
 |       uint64_t line_offset = 0, sec_offset = curr - start, offset; | 
 |       unsigned char **extended_ops = NULL; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (version, curr, 2, end); | 
 |       if (version != 4 && version != 5) | 
 | 	{ | 
 | 	  error (_("Expected to find a version number of 4 or 5 in section %s but found %d instead\n"), | 
 | 		 section->name, version); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (flags, curr, 1, end); | 
 |       offset_size = (flags & 1) ? 8 : 4; | 
 |       printf (_("  Offset:                      %#" PRIx64 "\n"), sec_offset); | 
 |       printf (_("  Version:                     %d\n"), version); | 
 |       printf (_("  Offset size:                 %d\n"), offset_size); | 
 |       if (flags & 2) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (line_offset, curr, offset_size, end); | 
 | 	  printf (_("  Offset into .debug_line:     %#" PRIx64 "\n"), | 
 | 		  line_offset); | 
 | 	} | 
 |       if (flags & 4) | 
 | 	{ | 
 | 	  unsigned int i, count, op; | 
 | 	  uint64_t nargs, n; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (count, curr, 1, end); | 
 |  | 
 | 	  memset (extended_op_buf, 0, sizeof (extended_op_buf)); | 
 | 	  extended_ops = extended_op_buf; | 
 | 	  if (count) | 
 | 	    { | 
 | 	      printf (_("  Extension opcode arguments:\n")); | 
 | 	      for (i = 0; i < count; i++) | 
 | 		{ | 
 | 		  SAFE_BYTE_GET_AND_INC (op, curr, 1, end); | 
 | 		  extended_ops[op] = curr; | 
 | 		  READ_ULEB (nargs, curr, end); | 
 | 		  if (nargs == 0) | 
 | 		    printf (_("    DW_MACRO_%02x has no arguments\n"), op); | 
 | 		  else | 
 | 		    { | 
 | 		      printf (_("    DW_MACRO_%02x arguments: "), op); | 
 | 		      for (n = 0; n < nargs; n++) | 
 | 			{ | 
 | 			  unsigned int form; | 
 |  | 
 | 			  SAFE_BYTE_GET_AND_INC (form, curr, 1, end); | 
 | 			  printf ("%s%s", get_FORM_name (form), | 
 | 				  n == nargs - 1 ? "\n" : ", "); | 
 | 			  switch (form) | 
 | 			    { | 
 | 			    case DW_FORM_data1: | 
 | 			    case DW_FORM_data2: | 
 | 			    case DW_FORM_data4: | 
 | 			    case DW_FORM_data8: | 
 | 			    case DW_FORM_sdata: | 
 | 			    case DW_FORM_udata: | 
 | 			    case DW_FORM_block: | 
 | 			    case DW_FORM_block1: | 
 | 			    case DW_FORM_block2: | 
 | 			    case DW_FORM_block4: | 
 | 			    case DW_FORM_flag: | 
 | 			    case DW_FORM_string: | 
 | 			    case DW_FORM_strp: | 
 | 			    case DW_FORM_sec_offset: | 
 | 			      break; | 
 | 			    default: | 
 | 			      error (_("Invalid extension opcode form %s\n"), | 
 | 				     get_FORM_name (form)); | 
 | 			      return 0; | 
 | 			    } | 
 | 			} | 
 | 		    } | 
 | 		} | 
 | 	    } | 
 | 	} | 
 |       printf ("\n"); | 
 |  | 
 |       while (1) | 
 | 	{ | 
 | 	  unsigned int op; | 
 | 	  const char *string; | 
 |  | 
 | 	  if (curr >= end) | 
 | 	    { | 
 | 	      error (_(".debug_macro section not zero terminated\n")); | 
 | 	      return 0; | 
 | 	    } | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (op, curr, 1, end); | 
 | 	  if (op == 0) | 
 | 	    break; | 
 |  | 
 | 	  switch (op) | 
 | 	    { | 
 | 	    case DW_MACRO_define: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      string = (const char *) curr; | 
 | 	      op = strnlen (string, end - curr); | 
 | 	      curr += op; | 
 | 	      printf (_(" DW_MACRO_define - lineno : %d macro : %*s\n"), | 
 | 		      lineno, (int) op, string); | 
 | 	      if (curr < end) | 
 | 		curr++; | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_undef: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      string = (const char *) curr; | 
 | 	      op = strnlen (string, end - curr); | 
 | 	      curr += op; | 
 | 	      printf (_(" DW_MACRO_undef - lineno : %d macro : %*s\n"), | 
 | 		      lineno, (int) op, string); | 
 | 	      if (curr < end) | 
 | 		curr++; | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_start_file: | 
 | 	      { | 
 | 		unsigned int filenum; | 
 | 		unsigned char *file_name = NULL, *dir_name = NULL; | 
 |  | 
 | 		READ_ULEB (lineno, curr, end); | 
 | 		READ_ULEB (filenum, curr, end); | 
 |  | 
 | 		if ((flags & 2) == 0) | 
 | 		  error (_("DW_MACRO_start_file used, but no .debug_line offset provided.\n")); | 
 | 		else | 
 | 		  file_name | 
 | 		    = get_line_filename_and_dirname (line_offset, filenum, | 
 | 						     &dir_name); | 
 | 		if (file_name == NULL) | 
 | 		  printf (_(" DW_MACRO_start_file - lineno: %d filenum: %d\n"), | 
 | 			  lineno, filenum); | 
 | 		else | 
 | 		  printf (_(" DW_MACRO_start_file - lineno: %d filenum: %d filename: %s%s%s\n"), | 
 | 			  lineno, filenum, | 
 | 			  dir_name != NULL ? (const char *) dir_name : "", | 
 | 			  dir_name != NULL ? "/" : "", file_name); | 
 | 	      } | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_end_file: | 
 | 	      printf (_(" DW_MACRO_end_file\n")); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_define_strp: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      if (version == 4 && is_dwo) | 
 | 		READ_ULEB (offset, curr, end); | 
 | 	      else | 
 | 		SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      string = fetch_indirect_string (offset); | 
 | 	      printf (_(" DW_MACRO_define_strp - lineno : %d macro : %s\n"), | 
 | 		      lineno, string); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_undef_strp: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      if (version == 4 && is_dwo) | 
 | 		READ_ULEB (offset, curr, end); | 
 | 	      else | 
 | 		SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      string = fetch_indirect_string (offset); | 
 | 	      printf (_(" DW_MACRO_undef_strp - lineno : %d macro : %s\n"), | 
 | 		      lineno, string); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_import: | 
 | 	      SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      printf (_(" DW_MACRO_import - offset : %#" PRIx64 "\n"), | 
 | 		      offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_define_sup: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      printf (_(" DW_MACRO_define_sup - lineno : %d" | 
 | 			" macro offset : %#" PRIx64 "\n"), | 
 | 		      lineno, offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_undef_sup: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      printf (_(" DW_MACRO_undef_sup - lineno : %d" | 
 | 			" macro offset : %#" PRIx64 "\n"), | 
 | 		      lineno, offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_import_sup: | 
 | 	      SAFE_BYTE_GET_AND_INC (offset, curr, offset_size, end); | 
 | 	      printf (_(" DW_MACRO_import_sup - offset : %#" PRIx64 "\n"), | 
 | 		      offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_MACRO_define_strx: | 
 | 	    case DW_MACRO_undef_strx: | 
 | 	      READ_ULEB (lineno, curr, end); | 
 | 	      READ_ULEB (offset, curr, end); | 
 | 	      string = fetch_indexed_string (offset, NULL, offset_size, | 
 | 					     is_dwo, 0); | 
 | 	      if (op == DW_MACRO_define_strx) | 
 | 		printf (" DW_MACRO_define_strx "); | 
 | 	      else | 
 | 		printf (" DW_MACRO_undef_strx "); | 
 | 	      if (do_wide) | 
 | 		printf (_("(with offset %#" PRIx64 ") "), offset); | 
 | 	      printf (_("lineno : %d macro : %s\n"), | 
 | 		      lineno, string); | 
 | 	      break; | 
 |  | 
 | 	    default: | 
 | 	      if (op >= DW_MACRO_lo_user && op <= DW_MACRO_hi_user) | 
 | 		{ | 
 | 		  printf (_(" <Target Specific macro op: %#x - UNHANDLED"), op); | 
 | 		  break; | 
 | 		} | 
 |  | 
 | 	      if (extended_ops == NULL || extended_ops[op] == NULL) | 
 | 		{ | 
 | 		  error (_(" Unknown macro opcode %02x seen\n"), op); | 
 | 		  return 0; | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  /* Skip over unhandled opcodes.  */ | 
 | 		  uint64_t nargs, n; | 
 | 		  unsigned char *desc = extended_ops[op]; | 
 | 		  READ_ULEB (nargs, desc, end); | 
 | 		  if (nargs == 0) | 
 | 		    { | 
 | 		      printf (_(" DW_MACRO_%02x\n"), op); | 
 | 		      break; | 
 | 		    } | 
 | 		  printf (_(" DW_MACRO_%02x -"), op); | 
 | 		  for (n = 0; n < nargs; n++) | 
 | 		    { | 
 | 		      int val; | 
 |  | 
 | 		      /* DW_FORM_implicit_const is not expected here.  */ | 
 | 		      SAFE_BYTE_GET_AND_INC (val, desc, 1, end); | 
 | 		      curr | 
 | 			= read_and_display_attr_value (0, val, 0, | 
 | 						       start, curr, end, 0, 0, | 
 | 						       offset_size, version, | 
 | 						       NULL, 0, section, | 
 | 						       NULL, ' ', -1); | 
 | 		      if (n != nargs - 1) | 
 | 			printf (","); | 
 | 		    } | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      break; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       printf ("\n"); | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_abbrev (struct dwarf_section *section, | 
 | 		      void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   abbrev_entry *entry; | 
 |   unsigned char *start = section->start; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   do | 
 |     { | 
 |       uint64_t offset = start - section->start; | 
 |       abbrev_list *list = find_and_process_abbrev_set (section, 0, | 
 | 						       section->size, offset, | 
 | 						       NULL); | 
 |       if (list == NULL) | 
 | 	break; | 
 |  | 
 |       if (list->first_abbrev) | 
 | 	printf (_("  Number TAG (%#" PRIx64 ")\n"), offset); | 
 |  | 
 |       for (entry = list->first_abbrev; entry; entry = entry->next) | 
 | 	{ | 
 | 	  abbrev_attr *attr; | 
 |  | 
 | 	  printf ("   %ld      %s    [%s]\n", | 
 | 		  entry->number, | 
 | 		  get_TAG_name (entry->tag), | 
 | 		  entry->children ? _("has children") : _("no children")); | 
 |  | 
 | 	  for (attr = entry->first_attr; attr; attr = attr->next) | 
 | 	    { | 
 | 	      printf ("    %-18s %s", | 
 | 		      get_AT_name (attr->attribute), | 
 | 		      get_FORM_name (attr->form)); | 
 | 	      if (attr->form == DW_FORM_implicit_const) | 
 | 		printf (": %" PRId64, attr->implicit_const); | 
 | 	      putchar ('\n'); | 
 | 	    } | 
 | 	} | 
 |       start = list->start_of_next_abbrevs; | 
 |       free_abbrev_list (list); | 
 |     } | 
 |   while (start); | 
 |  | 
 |   printf ("\n"); | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Return true when ADDR is the maximum address, when addresses are | 
 |    POINTER_SIZE bytes long.  */ | 
 |  | 
 | static bool | 
 | is_max_address (uint64_t addr, unsigned int pointer_size) | 
 | { | 
 |   uint64_t mask = ~(~(uint64_t) 0 << 1 << (pointer_size * 8 - 1)); | 
 |   return ((addr & mask) == mask); | 
 | } | 
 |  | 
 | /* Display a view pair list starting at *VSTART_PTR and ending at | 
 |    VLISTEND within SECTION.  */ | 
 |  | 
 | static void | 
 | display_view_pair_list (struct dwarf_section *section, | 
 | 			unsigned char **vstart_ptr, | 
 | 			unsigned int debug_info_entry, | 
 | 			unsigned char *vlistend) | 
 | { | 
 |   unsigned char *vstart = *vstart_ptr; | 
 |   unsigned char *section_end = section->start + section->size; | 
 |   unsigned int pointer_size = debug_information [debug_info_entry].pointer_size; | 
 |  | 
 |   if (vlistend < section_end) | 
 |     section_end = vlistend; | 
 |  | 
 |   putchar ('\n'); | 
 |  | 
 |   while (vstart < section_end) | 
 |     { | 
 |       uint64_t off = vstart - section->start; | 
 |       uint64_t vbegin, vend; | 
 |  | 
 |       READ_ULEB (vbegin, vstart, section_end); | 
 |       if (vstart == section_end) | 
 | 	break; | 
 |  | 
 |       READ_ULEB (vend, vstart, section_end); | 
 |       printf ("    %8.8" PRIx64 " ", off); | 
 |  | 
 |       print_view (vbegin, pointer_size); | 
 |       print_view (vend, pointer_size); | 
 |       printf (_("location view pair\n")); | 
 |     } | 
 |  | 
 |   putchar ('\n'); | 
 |   *vstart_ptr = vstart; | 
 | } | 
 |  | 
 | /* Display a location list from a normal (ie, non-dwo) .debug_loc section.  */ | 
 |  | 
 | static void | 
 | display_loc_list (struct dwarf_section *section, | 
 | 		  unsigned char **start_ptr, | 
 | 		  unsigned int debug_info_entry, | 
 | 		  uint64_t offset, | 
 | 		  uint64_t base_address, | 
 | 		  unsigned char **vstart_ptr, | 
 | 		  int has_frame_base) | 
 | { | 
 |   unsigned char *start = *start_ptr, *vstart = *vstart_ptr; | 
 |   unsigned char *section_end = section->start + section->size; | 
 |   uint64_t cu_offset; | 
 |   unsigned int pointer_size; | 
 |   unsigned int offset_size; | 
 |   int dwarf_version; | 
 |   uint64_t begin; | 
 |   uint64_t end; | 
 |   unsigned short length; | 
 |   int need_frame_base; | 
 |  | 
 |   if (debug_info_entry >= num_debug_info_entries) | 
 |     { | 
 |       warn (_("No debug information available for loc lists of entry: %u\n"), | 
 | 	    debug_info_entry); | 
 |       return; | 
 |     } | 
 |  | 
 |   cu_offset = debug_information [debug_info_entry].cu_offset; | 
 |   pointer_size = debug_information [debug_info_entry].pointer_size; | 
 |   offset_size = debug_information [debug_info_entry].offset_size; | 
 |   dwarf_version = debug_information [debug_info_entry].dwarf_version; | 
 |  | 
 |   if (pointer_size < 2 || pointer_size > 8) | 
 |     { | 
 |       warn (_("Invalid pointer size (%d) in debug info for entry %d\n"), | 
 | 	    pointer_size, debug_info_entry); | 
 |       return; | 
 |     } | 
 |  | 
 |   while (1) | 
 |     { | 
 |       uint64_t off = offset + (start - *start_ptr); | 
 |       uint64_t vbegin = -1, vend = -1; | 
 |  | 
 |       if (2 * pointer_size > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       printf ("    "); | 
 |       print_hex (off, 4); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, section_end); | 
 |       SAFE_BYTE_GET_AND_INC (end, start, pointer_size, section_end); | 
 |  | 
 |       if (begin == 0 && end == 0) | 
 | 	{ | 
 | 	  /* PR 18374: In a object file we can have a location list that | 
 | 	     starts with a begin and end of 0 because there are relocations | 
 | 	     that need to be applied to the addresses.  Actually applying | 
 | 	     the relocations now does not help as they will probably resolve | 
 | 	     to 0, since the object file has not been fully linked.  Real | 
 | 	     end of list markers will not have any relocations against them.  */ | 
 | 	  if (! reloc_at (section, off) | 
 | 	      && ! reloc_at (section, off + pointer_size)) | 
 | 	    { | 
 | 	      printf (_("<End of list>\n")); | 
 | 	      break; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       /* Check base address specifiers.  */ | 
 |       if (is_max_address (begin, pointer_size) | 
 | 	  && !is_max_address (end, pointer_size)) | 
 | 	{ | 
 | 	  base_address = end; | 
 | 	  print_hex (begin, pointer_size); | 
 | 	  print_hex (end, pointer_size); | 
 | 	  printf (_("(base address)\n")); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if (vstart) | 
 | 	{ | 
 | 	  off = offset + (vstart - *start_ptr); | 
 |  | 
 | 	  READ_ULEB (vbegin, vstart, section_end); | 
 | 	  print_view (vbegin, pointer_size); | 
 |  | 
 | 	  READ_ULEB (vend, vstart, section_end); | 
 | 	  print_view (vend, pointer_size); | 
 |  | 
 | 	  printf (_("views at %8.8" PRIx64 " for:\n    %*s "), off, 8, ""); | 
 | 	} | 
 |  | 
 |       if (2 > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (length, start, 2, section_end); | 
 |  | 
 |       if (length > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       print_hex (begin + base_address, pointer_size); | 
 |       print_hex (end + base_address, pointer_size); | 
 |  | 
 |       putchar ('('); | 
 |       need_frame_base = decode_location_expression (start, | 
 | 						    pointer_size, | 
 | 						    offset_size, | 
 | 						    dwarf_version, | 
 | 						    length, | 
 | 						    cu_offset, section); | 
 |       putchar (')'); | 
 |  | 
 |       if (need_frame_base && !has_frame_base) | 
 | 	printf (_(" [without DW_AT_frame_base]")); | 
 |  | 
 |       if (begin == end && vbegin == vend) | 
 | 	fputs (_(" (start == end)"), stdout); | 
 |       else if (begin > end || (begin == end && vbegin > vend)) | 
 | 	fputs (_(" (start > end)"), stdout); | 
 |  | 
 |       putchar ('\n'); | 
 |  | 
 |       start += length; | 
 |     } | 
 |  | 
 |   *start_ptr = start; | 
 |   *vstart_ptr = vstart; | 
 | } | 
 |  | 
 | /* Display a location list from a normal (ie, non-dwo) .debug_loclists section.  */ | 
 |  | 
 | static void | 
 | display_loclists_list (struct dwarf_section *  section, | 
 | 		       unsigned char **        start_ptr, | 
 | 		       debug_info *            debug_info_p, | 
 | 		       uint64_t                offset, | 
 | 		       uint64_t                base_address, | 
 | 		       unsigned char **        vstart_ptr, | 
 | 		       int                     has_frame_base) | 
 | { | 
 |   unsigned char *start = *start_ptr; | 
 |   unsigned char *vstart = *vstart_ptr; | 
 |   unsigned char *section_end = section->start + section->size; | 
 |   uint64_t cu_offset; | 
 |   unsigned int pointer_size; | 
 |   unsigned int offset_size; | 
 |   unsigned int dwarf_version; | 
 |   uint64_t idx; | 
 |  | 
 |   /* Initialize it due to a false compiler warning.  */ | 
 |   uint64_t begin = -1, vbegin = -1; | 
 |   uint64_t end = -1, vend = -1; | 
 |   uint64_t length; | 
 |   int need_frame_base; | 
 |  | 
 |   cu_offset = debug_info_p->cu_offset; | 
 |   pointer_size = debug_info_p->pointer_size; | 
 |   offset_size = debug_info_p->offset_size; | 
 |   dwarf_version = debug_info_p->dwarf_version; | 
 |  | 
 |   if (pointer_size < 2 || pointer_size > 8) | 
 |     { | 
 |       warn (_("Invalid pointer size (%d) in debug info for entry %d\n"), | 
 | 	    pointer_size, (int)(debug_info_p - debug_information)); | 
 |       return; | 
 |     } | 
 |  | 
 |   while (1) | 
 |     { | 
 |       uint64_t off = offset + (start - *start_ptr); | 
 |       enum dwarf_location_list_entry_type llet; | 
 |  | 
 |       if (start + 1 > section_end) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       printf ("    "); | 
 |       print_hex (off, 4); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (llet, start, 1, section_end); | 
 |  | 
 |       if (vstart && (llet == DW_LLE_offset_pair | 
 | 		     || llet == DW_LLE_start_end | 
 | 		     || llet == DW_LLE_start_length)) | 
 | 	{ | 
 | 	  off = offset + (vstart - *start_ptr); | 
 |  | 
 | 	  READ_ULEB (vbegin, vstart, section_end); | 
 | 	  print_view (vbegin, pointer_size); | 
 |  | 
 | 	  READ_ULEB (vend, vstart, section_end); | 
 | 	  print_view (vend, pointer_size); | 
 |  | 
 | 	  printf (_("views at %8.8" PRIx64 " for:\n    %*s "), off, 8, ""); | 
 | 	} | 
 |  | 
 |       switch (llet) | 
 | 	{ | 
 | 	case DW_LLE_end_of_list: | 
 | 	  printf (_("<End of list>\n")); | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_base_addressx: | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  print_hex (idx, pointer_size); | 
 | 	  printf (_("(index into .debug_addr) ")); | 
 | 	  base_address = fetch_indexed_addr | 
 | 	    (debug_info_p->addr_base + idx * pointer_size, pointer_size); | 
 | 	  print_hex (base_address, pointer_size); | 
 | 	  printf (_("(base address)\n")); | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_startx_endx: | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  begin = fetch_indexed_addr | 
 | 	    (debug_info_p->addr_base + idx * pointer_size, pointer_size); | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  end = fetch_indexed_addr | 
 | 	    (debug_info_p->addr_base + idx * pointer_size, pointer_size); | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_startx_length: | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  begin = fetch_indexed_addr | 
 | 	    (debug_info_p->addr_base + idx * pointer_size, pointer_size); | 
 | 	  READ_ULEB (end, start, section_end); | 
 | 	  end += begin; | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_default_location: | 
 | 	  begin = end = 0; | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_offset_pair: | 
 | 	  READ_ULEB (begin, start, section_end); | 
 | 	  begin += base_address; | 
 | 	  READ_ULEB (end, start, section_end); | 
 | 	  end += base_address; | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_base_address: | 
 | 	  SAFE_BYTE_GET_AND_INC (base_address, start, pointer_size, | 
 | 				 section_end); | 
 | 	  print_hex (base_address, pointer_size); | 
 | 	  printf (_("(base address)\n")); | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_start_end: | 
 | 	  SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, section_end); | 
 | 	  SAFE_BYTE_GET_AND_INC (end, start, pointer_size, section_end); | 
 | 	  break; | 
 |  | 
 | 	case DW_LLE_start_length: | 
 | 	  SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, section_end); | 
 | 	  READ_ULEB (end, start, section_end); | 
 | 	  end += begin; | 
 | 	  break; | 
 |  | 
 | #ifdef DW_LLE_view_pair | 
 | 	case DW_LLE_view_pair: | 
 | 	  if (vstart) | 
 | 	    printf (_("View pair entry in loclist with locviews attribute\n")); | 
 | 	  READ_ULEB (vbegin, start, section_end); | 
 | 	  print_view (vbegin, pointer_size); | 
 |  | 
 | 	  READ_ULEB (vend, start, section_end); | 
 | 	  print_view (vend, pointer_size); | 
 |  | 
 | 	  printf (_("views for:\n")); | 
 | 	  continue; | 
 | #endif | 
 |  | 
 | 	default: | 
 | 	  error (_("Invalid location list entry type %d\n"), llet); | 
 | 	  return; | 
 | 	} | 
 |  | 
 |       if (llet == DW_LLE_end_of_list) | 
 | 	break; | 
 |  | 
 |       if (llet == DW_LLE_base_address | 
 | 	  || llet == DW_LLE_base_addressx) | 
 | 	continue; | 
 |  | 
 |       if (start == section_end) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |       READ_ULEB (length, start, section_end); | 
 |  | 
 |       if (length > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       print_hex (begin, pointer_size); | 
 |       print_hex (end, pointer_size); | 
 |  | 
 |       putchar ('('); | 
 |       need_frame_base = decode_location_expression (start, | 
 | 						    pointer_size, | 
 | 						    offset_size, | 
 | 						    dwarf_version, | 
 | 						    length, | 
 | 						    cu_offset, section); | 
 |       putchar (')'); | 
 |  | 
 |       if (need_frame_base && !has_frame_base) | 
 | 	printf (_(" [without DW_AT_frame_base]")); | 
 |  | 
 |       if (begin == end && vbegin == vend) | 
 | 	fputs (_(" (start == end)"), stdout); | 
 |       else if (begin > end || (begin == end && vbegin > vend)) | 
 | 	fputs (_(" (start > end)"), stdout); | 
 |  | 
 |       putchar ('\n'); | 
 |  | 
 |       start += length; | 
 |       vbegin = vend = -1; | 
 |     } | 
 |  | 
 |   if (vbegin != (uint64_t) -1 || vend != (uint64_t) -1) | 
 |     printf (_("Trailing view pair not used in a range")); | 
 |  | 
 |   *start_ptr = start; | 
 |   *vstart_ptr = vstart; | 
 | } | 
 |  | 
 | /* Print a .debug_addr table index in decimal, surrounded by square brackets, | 
 |    right-adjusted in a field of length LEN, and followed by a space.  */ | 
 |  | 
 | static void | 
 | print_addr_index (unsigned int idx, unsigned int len) | 
 | { | 
 |   static char buf[15]; | 
 |   snprintf (buf, sizeof (buf), "[%d]", idx); | 
 |   printf ("%*s ", len, buf); | 
 | } | 
 |  | 
 | /* Display a location list from a .dwo section. It uses address indexes rather | 
 |    than embedded addresses.  This code closely follows display_loc_list, but the | 
 |    two are sufficiently different that combining things is very ugly.  */ | 
 |  | 
 | static void | 
 | display_loc_list_dwo (struct dwarf_section *section, | 
 | 		      unsigned char **start_ptr, | 
 | 		      unsigned int debug_info_entry, | 
 | 		      uint64_t offset, | 
 | 		      unsigned char **vstart_ptr, | 
 | 		      int has_frame_base) | 
 | { | 
 |   unsigned char *start = *start_ptr, *vstart = *vstart_ptr; | 
 |   unsigned char *section_end = section->start + section->size; | 
 |   uint64_t cu_offset; | 
 |   unsigned int pointer_size; | 
 |   unsigned int offset_size; | 
 |   int dwarf_version; | 
 |   int entry_type; | 
 |   unsigned short length; | 
 |   int need_frame_base; | 
 |   unsigned int idx; | 
 |  | 
 |   if (debug_info_entry >= num_debug_info_entries) | 
 |     { | 
 |       warn (_("No debug information for loc lists of entry: %u\n"), | 
 | 	    debug_info_entry); | 
 |       return; | 
 |     } | 
 |  | 
 |   cu_offset = debug_information [debug_info_entry].cu_offset; | 
 |   pointer_size = debug_information [debug_info_entry].pointer_size; | 
 |   offset_size = debug_information [debug_info_entry].offset_size; | 
 |   dwarf_version = debug_information [debug_info_entry].dwarf_version; | 
 |  | 
 |   if (pointer_size < 2 || pointer_size > 8) | 
 |     { | 
 |       warn (_("Invalid pointer size (%d) in debug info for entry %d\n"), | 
 | 	    pointer_size, debug_info_entry); | 
 |       return; | 
 |     } | 
 |  | 
 |   while (1) | 
 |     { | 
 |       printf ("    "); | 
 |       print_hex (offset + (start - *start_ptr), 4); | 
 |  | 
 |       if (start >= section_end) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (entry_type, start, 1, section_end); | 
 |  | 
 |       if (vstart) | 
 | 	switch (entry_type) | 
 | 	  { | 
 | 	  default: | 
 | 	    break; | 
 |  | 
 | 	  case 2: | 
 | 	  case 3: | 
 | 	  case 4: | 
 | 	    { | 
 | 	      uint64_t view; | 
 | 	      uint64_t off = offset + (vstart - *start_ptr); | 
 |  | 
 | 	      READ_ULEB (view, vstart, section_end); | 
 | 	      print_view (view, 8); | 
 |  | 
 | 	      READ_ULEB (view, vstart, section_end); | 
 | 	      print_view (view, 8); | 
 |  | 
 | 	      printf (_("views at %8.8" PRIx64 " for:\n    %*s "), off, 8, ""); | 
 |  | 
 | 	    } | 
 | 	    break; | 
 | 	  } | 
 |  | 
 |       switch (entry_type) | 
 | 	{ | 
 | 	case 0: /* A terminating entry.  */ | 
 | 	  *start_ptr = start; | 
 | 	  *vstart_ptr = vstart; | 
 | 	  printf (_("<End of list>\n")); | 
 | 	  return; | 
 | 	case 1: /* A base-address entry.  */ | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  print_addr_index (idx, 8); | 
 | 	  printf ("%*s", 9 + (vstart ? 2 * 6 : 0), ""); | 
 | 	  printf (_("(base address selection entry)\n")); | 
 | 	  continue; | 
 | 	case 2: /* A start/end entry.  */ | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  print_addr_index (idx, 8); | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  print_addr_index (idx, 8); | 
 | 	  break; | 
 | 	case 3: /* A start/length entry.  */ | 
 | 	  READ_ULEB (idx, start, section_end); | 
 | 	  print_addr_index (idx, 8); | 
 | 	  SAFE_BYTE_GET_AND_INC (idx, start, 4, section_end); | 
 | 	  printf ("%08x ", idx); | 
 | 	  break; | 
 | 	case 4: /* An offset pair entry.  */ | 
 | 	  SAFE_BYTE_GET_AND_INC (idx, start, 4, section_end); | 
 | 	  printf ("%08x ", idx); | 
 | 	  SAFE_BYTE_GET_AND_INC (idx, start, 4, section_end); | 
 | 	  printf ("%08x ", idx); | 
 | 	  break; | 
 | 	default: | 
 | 	  warn (_("Unknown location list entry type 0x%x.\n"), entry_type); | 
 | 	  *start_ptr = start; | 
 | 	  *vstart_ptr = vstart; | 
 | 	  return; | 
 | 	} | 
 |  | 
 |       if (2 > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (length, start, 2, section_end); | 
 |       if (length > (size_t) (section_end - start)) | 
 | 	{ | 
 | 	  warn (_("Location list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       putchar ('('); | 
 |       need_frame_base = decode_location_expression (start, | 
 | 						    pointer_size, | 
 | 						    offset_size, | 
 | 						    dwarf_version, | 
 | 						    length, | 
 | 						    cu_offset, section); | 
 |       putchar (')'); | 
 |  | 
 |       if (need_frame_base && !has_frame_base) | 
 | 	printf (_(" [without DW_AT_frame_base]")); | 
 |  | 
 |       putchar ('\n'); | 
 |  | 
 |       start += length; | 
 |     } | 
 |  | 
 |   *start_ptr = start; | 
 |   *vstart_ptr = vstart; | 
 | } | 
 |  | 
 | /* Sort array of indexes in ascending order of loc_offsets[idx] and | 
 |    loc_views.  */ | 
 |  | 
 | static uint64_t *loc_offsets, *loc_views; | 
 |  | 
 | static int | 
 | loc_offsets_compar (const void *ap, const void *bp) | 
 | { | 
 |   uint64_t a = loc_offsets[*(const unsigned int *) ap]; | 
 |   uint64_t b = loc_offsets[*(const unsigned int *) bp]; | 
 |  | 
 |   int ret = (a > b) - (b > a); | 
 |   if (ret) | 
 |     return ret; | 
 |  | 
 |   a = loc_views[*(const unsigned int *) ap]; | 
 |   b = loc_views[*(const unsigned int *) bp]; | 
 |  | 
 |   ret = (a > b) - (b > a); | 
 |  | 
 |   return ret; | 
 | } | 
 |  | 
 | /* Reads and dumps the DWARFv5 loclists compiler unit header, | 
 |    including the offset table. | 
 |    Returns the offset of the next compile unit header.  */ | 
 |  | 
 | static uint64_t | 
 | display_loclists_unit_header (struct dwarf_section *  section, | 
 | 			      uint64_t                header_offset, | 
 | 			      uint32_t *              offset_count, | 
 | 			      unsigned char **        loclists_start) | 
 | { | 
 |   uint64_t length; | 
 |   unsigned char *start = section->start + header_offset; | 
 |   unsigned char *end = section->start + section->size; | 
 |   unsigned short version; | 
 |   unsigned char address_size; | 
 |   unsigned char segment_selector_size; | 
 |   bool is_64bit; | 
 |   uint32_t i; | 
 |  | 
 |   printf (_("Table at Offset %#" PRIx64 "\n"), header_offset); | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (length, start, 4, end); | 
 |   if (length == 0xffffffff) | 
 |     { | 
 |       is_64bit = true; | 
 |       SAFE_BYTE_GET_AND_INC (length, start, 8, end); | 
 |     } | 
 |   else | 
 |     is_64bit = false; | 
 |  | 
 |   SAFE_BYTE_GET_AND_INC (version, start, 2, end); | 
 |   SAFE_BYTE_GET_AND_INC (address_size, start, 1, end); | 
 |   SAFE_BYTE_GET_AND_INC (segment_selector_size, start, 1, end); | 
 |   SAFE_BYTE_GET_AND_INC (*offset_count, start, 4, end); | 
 |  | 
 |   printf (_("  Length:          %#" PRIx64 "\n"), length); | 
 |   printf (_("  DWARF version:   %u\n"), version); | 
 |   printf (_("  Address size:    %u\n"), address_size); | 
 |   printf (_("  Segment size:    %u\n"), segment_selector_size); | 
 |   printf (_("  Offset entries:  %u\n"), *offset_count); | 
 |  | 
 |   if (segment_selector_size != 0) | 
 |     { | 
 |       warn (_("The %s section contains an " | 
 | 	      "unsupported segment selector size: %d.\n"), | 
 | 	    section->name, segment_selector_size); | 
 |       return (uint64_t)-1; | 
 |     } | 
 |  | 
 |   if ( *offset_count) | 
 |     { | 
 |       printf (_("\n   Offset Entries starting at %#tx:\n"), | 
 | 	      start - section->start); | 
 |  | 
 |       for (i = 0; i < *offset_count; i++) | 
 | 	{ | 
 | 	  uint64_t entry; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (entry, start, is_64bit ? 8 : 4, end); | 
 | 	  printf (_("    [%6u] %#" PRIx64 "\n"), i, entry); | 
 | 	} | 
 |     } | 
 |  | 
 |   putchar ('\n'); | 
 |   *loclists_start = start; | 
 |  | 
 |   /* The length field doesn't include the length field itself.  */ | 
 |   return header_offset + length + (is_64bit ? 12 : 4); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_loc (struct dwarf_section *section, void *file) | 
 | { | 
 |   unsigned char *start = section->start, *vstart = NULL; | 
 |   uint64_t bytes; | 
 |   unsigned char *section_begin = start; | 
 |   unsigned int num_loc_list = 0; | 
 |   uint64_t last_offset = 0; | 
 |   uint64_t last_view = 0; | 
 |   unsigned int first = 0; | 
 |   unsigned int i; | 
 |   unsigned int j; | 
 |   int seen_first_offset = 0; | 
 |   int locs_sorted = 1; | 
 |   unsigned char *next = start, *vnext = vstart; | 
 |   unsigned int *array = NULL; | 
 |   const char *suffix = strrchr (section->name, '.'); | 
 |   bool is_dwo = false; | 
 |   bool is_loclists = strstr (section->name, "debug_loclists") != NULL; | 
 |   uint64_t next_header_offset = 0; | 
 |  | 
 |   if (suffix && strcmp (suffix, ".dwo") == 0) | 
 |     is_dwo = true; | 
 |  | 
 |   bytes = section->size; | 
 |  | 
 |   if (bytes == 0) | 
 |     { | 
 |       printf (_("\nThe %s section is empty.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (is_loclists) | 
 |     { | 
 |       unsigned char *hdrptr = section_begin; | 
 |       uint64_t ll_length; | 
 |       unsigned short ll_version; | 
 |       unsigned char *end = section_begin + section->size; | 
 |       unsigned char address_size, segment_selector_size; | 
 |       uint32_t offset_entry_count; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (ll_length, hdrptr, 4, end); | 
 |       if (ll_length == 0xffffffff) | 
 | 	SAFE_BYTE_GET_AND_INC (ll_length, hdrptr, 8, end); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (ll_version, hdrptr, 2, end); | 
 |       if (ll_version != 5) | 
 | 	{ | 
 | 	  warn (_("The %s section contains corrupt or " | 
 | 		  "unsupported version number: %d.\n"), | 
 | 		section->name, ll_version); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (address_size, hdrptr, 1, end); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (segment_selector_size, hdrptr, 1, end); | 
 |       if (segment_selector_size != 0) | 
 | 	{ | 
 | 	  warn (_("The %s section contains " | 
 | 		  "unsupported segment selector size: %d.\n"), | 
 | 		section->name, segment_selector_size); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (offset_entry_count, hdrptr, 4, end); | 
 |  | 
 |       /*if (offset_entry_count != 0) | 
 | 	return display_offset_entry_loclists (section);*/ | 
 |  | 
 |       //header_size = hdrptr - section_begin; | 
 |     } | 
 |  | 
 |   if (load_debug_info (file) == 0) | 
 |     { | 
 |       warn (_("Unable to load/parse the .debug_info section, so cannot interpret the %s section.\n"), | 
 | 	    section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   /* Check the order of location list in .debug_info section. If | 
 |      offsets of location lists are in the ascending order, we can | 
 |      use `debug_information' directly.  */ | 
 |   for (i = 0; i < num_debug_info_entries; i++) | 
 |     { | 
 |       unsigned int num; | 
 |  | 
 |       num = debug_information [i].num_loc_offsets; | 
 |       if (num > num_loc_list) | 
 | 	num_loc_list = num; | 
 |  | 
 |       /* Check if we can use `debug_information' directly.  */ | 
 |       if (locs_sorted && num != 0) | 
 | 	{ | 
 | 	  if (!seen_first_offset) | 
 | 	    { | 
 | 	      /* This is the first location list.  */ | 
 | 	      last_offset = debug_information [i].loc_offsets [0]; | 
 | 	      last_view = debug_information [i].loc_views [0]; | 
 | 	      first = i; | 
 | 	      seen_first_offset = 1; | 
 | 	      j = 1; | 
 | 	    } | 
 | 	  else | 
 | 	    j = 0; | 
 |  | 
 | 	  for (; j < num; j++) | 
 | 	    { | 
 | 	      if (last_offset > debug_information [i].loc_offsets [j] | 
 | 		  || (last_offset == debug_information [i].loc_offsets [j] | 
 | 		      && last_view > debug_information [i].loc_views [j])) | 
 | 		{ | 
 | 		  locs_sorted = 0; | 
 | 		  break; | 
 | 		} | 
 | 	      last_offset = debug_information [i].loc_offsets [j]; | 
 | 	      last_view = debug_information [i].loc_views [j]; | 
 | 	    } | 
 | 	} | 
 |     } | 
 |  | 
 |   if (!seen_first_offset) | 
 |     error (_("No location lists in .debug_info section!\n")); | 
 |  | 
 |   if (!locs_sorted) | 
 |     array = (unsigned int *) xcmalloc (num_loc_list, sizeof (unsigned int)); | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   if (reloc_at (section, 0)) | 
 |     printf (_(" Warning: This section has relocations - addresses seen here may not be accurate.\n\n")); | 
 |  | 
 |   if (!is_loclists) | 
 |     printf (_("    Offset   Begin            End              Expression\n")); | 
 |  | 
 |   for (i = first; i < num_debug_info_entries; i++) | 
 |     { | 
 |       uint64_t offset = 0, voffset = 0; | 
 |       uint64_t base_address; | 
 |       unsigned int k; | 
 |       int has_frame_base; | 
 |       debug_info *debug_info_p = debug_information + i; | 
 |       uint32_t offset_count; | 
 |  | 
 |       /* .debug_loclists section is loaded into debug_information as | 
 | 	 DWARF-5 debug info and .debug_loc section is loaded into | 
 | 	 debug_information as pre-DWARF-5 debug info.  When dumping | 
 | 	 .debug_loc section, we should only process pre-DWARF-5 debug | 
 | 	 info in debug_information.  When dumping .debug_loclists | 
 | 	 section, we should only process DWARF-5 info in | 
 | 	 debug_information.  */ | 
 |       if ((debug_info_p->dwarf_version >= 5) != is_loclists) | 
 | 	continue; | 
 |  | 
 |       if (!locs_sorted) | 
 | 	{ | 
 | 	  for (k = 0; k < debug_info_p->num_loc_offsets; k++) | 
 | 	    array[k] = k; | 
 | 	  loc_offsets = debug_info_p->loc_offsets; | 
 | 	  loc_views = debug_info_p->loc_views; | 
 | 	  qsort (array, debug_info_p->num_loc_offsets, | 
 | 		 sizeof (*array), loc_offsets_compar); | 
 | 	} | 
 |  | 
 |       /* .debug_loclists has a per-unit header. | 
 | 	 Update start if we are detecting it.  */ | 
 |       if (debug_info_p->dwarf_version >= 5) | 
 | 	{ | 
 | 	  j = locs_sorted ? 0 : array [0]; | 
 |  | 
 | 	  if (debug_info_p->num_loc_offsets) | 
 | 	    offset = debug_info_p->loc_offsets [j]; | 
 |  | 
 | 	  if (debug_info_p->num_loc_views) | 
 | 	    voffset = debug_info_p->loc_views [j]; | 
 |  | 
 | 	  /* Parse and dump unit headers in loclists. | 
 | 	     This will misbehave if the order of CUs in debug_info | 
 | 	     doesn't match the one in loclists.  */ | 
 | 	  if (next_header_offset < offset) | 
 | 	    { | 
 | 	      while (next_header_offset < offset) | 
 | 		{ | 
 | 		  next_header_offset = display_loclists_unit_header | 
 | 		    (section, next_header_offset, &offset_count, &start); | 
 |  | 
 | 		  if (next_header_offset == (uint64_t)-1) | 
 | 		    /* Header parsing error.  */ | 
 | 		    return 0; | 
 | 		} | 
 |  | 
 | 	      printf (_("\ | 
 |     Offset   Begin            End              Expression\n")); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       int adjacent_view_loclists = 1; | 
 |  | 
 |       for (k = 0; k < debug_info_p->num_loc_offsets; k++) | 
 | 	{ | 
 | 	  j = locs_sorted ? k : array[k]; | 
 | 	  if (k | 
 | 	      && (debug_info_p->loc_offsets [locs_sorted | 
 | 					     ? k - 1 : array [k - 1]] | 
 | 		  == debug_info_p->loc_offsets [j]) | 
 | 	      && (debug_info_p->loc_views [locs_sorted | 
 | 					   ? k - 1 : array [k - 1]] | 
 | 		  == debug_info_p->loc_views [j])) | 
 | 	    continue; | 
 | 	  has_frame_base = debug_info_p->have_frame_base [j]; | 
 | 	  offset = debug_info_p->loc_offsets [j]; | 
 | 	  next = section_begin + offset; | 
 | 	  voffset = debug_info_p->loc_views [j]; | 
 | 	  if (voffset != (uint64_t) -1) | 
 | 	    vnext = section_begin + voffset; | 
 | 	  else | 
 | 	    vnext = NULL; | 
 | 	  base_address = debug_info_p->base_address; | 
 |  | 
 | 	  if (vnext && vnext < next) | 
 | 	    { | 
 | 	      vstart = vnext; | 
 | 	      display_view_pair_list (section, &vstart, i, next); | 
 | 	      if (start == vnext) | 
 | 		start = vstart; | 
 | 	    } | 
 |  | 
 | 	  if (start < next) | 
 | 	    { | 
 | 	      if (vnext && vnext < next) | 
 | 		warn (_("There is a hole [%#tx - %#" PRIx64 "]" | 
 | 			" in %s section.\n"), | 
 | 		      start - section_begin, voffset, section->name); | 
 | 	      else | 
 | 		warn (_("There is a hole [%#tx - %#" PRIx64 "]" | 
 | 			" in %s section.\n"), | 
 | 		      start - section_begin, offset, section->name); | 
 | 	    } | 
 | 	  else if (start > next) | 
 | 	    warn (_("There is an overlap [%#tx - %#" PRIx64 "]" | 
 | 		    " in %s section.\n"), | 
 | 		  start - section_begin, offset, section->name); | 
 | 	  start = next; | 
 | 	  vstart = vnext; | 
 |  | 
 | 	  if (offset >= bytes) | 
 | 	    { | 
 | 	      warn (_("Offset %#" PRIx64 " is bigger than %s section size.\n"), | 
 | 		    offset, section->name); | 
 | 	      continue; | 
 | 	    } | 
 |  | 
 | 	  if (vnext && voffset >= bytes) | 
 | 	    { | 
 | 	      warn (_("View Offset %#" PRIx64 " is bigger than %s section size.\n"), | 
 | 		    voffset, section->name); | 
 | 	      continue; | 
 | 	    } | 
 |  | 
 | 	  if (!is_loclists) | 
 | 	    { | 
 | 	      if (is_dwo) | 
 | 		display_loc_list_dwo (section, &start, i, offset, | 
 | 				      &vstart, has_frame_base); | 
 | 	      else | 
 | 		display_loc_list (section, &start, i, offset, base_address, | 
 | 				  &vstart, has_frame_base); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      if (is_dwo) | 
 | 		warn (_("DWO is not yet supported.\n")); | 
 | 	      else | 
 | 		display_loclists_list (section, &start, debug_info_p, offset, | 
 | 				       base_address, &vstart, has_frame_base); | 
 | 	    } | 
 |  | 
 | 	  /* FIXME: this arrangement is quite simplistic.  Nothing | 
 | 	     requires locview lists to be adjacent to corresponding | 
 | 	     loclists, and a single loclist could be augmented by | 
 | 	     different locview lists, and vice-versa, unlikely as it | 
 | 	     is that it would make sense to do so.  Hopefully we'll | 
 | 	     have view pair support built into loclists before we ever | 
 | 	     need to address all these possibilities.  */ | 
 | 	  if (adjacent_view_loclists && vnext | 
 | 	      && vnext != start && vstart != next) | 
 | 	    { | 
 | 	      adjacent_view_loclists = 0; | 
 | 	      warn (_("Hole and overlap detection requires adjacent view lists and loclists.\n")); | 
 | 	    } | 
 |  | 
 | 	  if (vnext && vnext == start) | 
 | 	    display_view_pair_list (section, &start, i, vstart); | 
 | 	} | 
 |     } | 
 |  | 
 |   if (start < section->start + section->size) | 
 |     warn (ngettext ("There is %ld unused byte at the end of section %s\n", | 
 | 		    "There are %ld unused bytes at the end of section %s\n", | 
 | 		    (long) (section->start + section->size - start)), | 
 | 	  (long) (section->start + section->size - start), section->name); | 
 |   putchar ('\n'); | 
 |   free (array); | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_str (struct dwarf_section *section, | 
 | 		   void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   uint64_t bytes = section->size; | 
 |   uint64_t addr = section->address; | 
 |  | 
 |   if (bytes == 0) | 
 |     { | 
 |       printf (_("\nThe %s section is empty.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (bytes) | 
 |     { | 
 |       int j; | 
 |       int k; | 
 |       int lbytes; | 
 |  | 
 |       lbytes = (bytes > 16 ? 16 : bytes); | 
 |  | 
 |       printf ("  0x%8.8" PRIx64 " ", addr); | 
 |  | 
 |       for (j = 0; j < 16; j++) | 
 | 	{ | 
 | 	  if (j < lbytes) | 
 | 	    printf ("%2.2x", start[j]); | 
 | 	  else | 
 | 	    printf ("  "); | 
 |  | 
 | 	  if ((j & 3) == 3) | 
 | 	    printf (" "); | 
 | 	} | 
 |  | 
 |       for (j = 0; j < lbytes; j++) | 
 | 	{ | 
 | 	  k = start[j]; | 
 | 	  if (k >= ' ' && k < 0x80) | 
 | 	    printf ("%c", k); | 
 | 	  else | 
 | 	    printf ("."); | 
 | 	} | 
 |  | 
 |       putchar ('\n'); | 
 |  | 
 |       start += lbytes; | 
 |       addr  += lbytes; | 
 |       bytes -= lbytes; | 
 |     } | 
 |  | 
 |   putchar ('\n'); | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_info (struct dwarf_section *section, void *file) | 
 | { | 
 |   return process_debug_info (section, file, section->abbrev_sec, false, false); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_types (struct dwarf_section *section, void *file) | 
 | { | 
 |   return process_debug_info (section, file, section->abbrev_sec, false, true); | 
 | } | 
 |  | 
 | static int | 
 | display_trace_info (struct dwarf_section *section, void *file) | 
 | { | 
 |   return process_debug_info (section, file, section->abbrev_sec, false, true); | 
 | } | 
 |  | 
 | static int | 
 | display_sframe (struct dwarf_section *section, void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   sframe_decoder_ctx *sfd_ctx = NULL; | 
 |   unsigned char *data = section->start; | 
 |   size_t sf_size = section->size; | 
 |   int err = 0; | 
 |  | 
 |   if (strcmp (section->name, "") == 0) | 
 |     { | 
 |       error (_("Section name must be provided \n")); | 
 |       return false; | 
 |     } | 
 |  | 
 |   /* Decode the contents of the section.  */ | 
 |   sfd_ctx = sframe_decode ((const char*)data, sf_size, &err); | 
 |   if (!sfd_ctx || err) | 
 |     { | 
 |       error (_("SFrame decode failure: %s\n"), sframe_errmsg (err)); | 
 |       return false; | 
 |     } | 
 |  | 
 |   printf (_("Contents of the SFrame section %s:"), section->name); | 
 |   /* Dump the contents as text.  */ | 
 |   dump_sframe (sfd_ctx, section->address); | 
 |  | 
 |   sframe_decoder_free (&sfd_ctx); | 
 |  | 
 |   return true; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_aranges (struct dwarf_section *section, | 
 | 		       void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   /* It does not matter if this load fails, | 
 |      we test for that later on.  */ | 
 |   load_debug_info (file); | 
 |  | 
 |   while (start < end) | 
 |     { | 
 |       unsigned char *hdrptr; | 
 |       DWARF2_Internal_ARange arange; | 
 |       unsigned char *addr_ranges; | 
 |       uint64_t length; | 
 |       uint64_t address; | 
 |       uint64_t sec_off; | 
 |       unsigned char address_size; | 
 |       unsigned int offset_size; | 
 |       unsigned char *end_ranges; | 
 |  | 
 |       hdrptr = start; | 
 |       sec_off = hdrptr - section->start; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (arange.ar_length, hdrptr, 4, end); | 
 |       if (arange.ar_length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (arange.ar_length, hdrptr, 8, end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |  | 
 |       if (arange.ar_length > (size_t) (end - hdrptr)) | 
 | 	{ | 
 | 	  warn (_("Debug info is corrupted, %s header at %#" PRIx64 | 
 | 		  " has length %#" PRIx64 "\n"), | 
 | 		section->name, sec_off, arange.ar_length); | 
 | 	  break; | 
 | 	} | 
 |       end_ranges = hdrptr + arange.ar_length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (arange.ar_version, hdrptr, 2, end_ranges); | 
 |       SAFE_BYTE_GET_AND_INC (arange.ar_info_offset, hdrptr, offset_size, | 
 | 			     end_ranges); | 
 |  | 
 |       if (num_debug_info_entries != DEBUG_INFO_UNAVAILABLE | 
 | 	  && num_debug_info_entries > 0 | 
 | 	  && find_debug_info_for_offset (arange.ar_info_offset) == NULL) | 
 | 	warn (_(".debug_info offset of %#" PRIx64 | 
 | 		" in %s section does not point to a CU header.\n"), | 
 | 	      arange.ar_info_offset, section->name); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (arange.ar_pointer_size, hdrptr, 1, end_ranges); | 
 |       SAFE_BYTE_GET_AND_INC (arange.ar_segment_size, hdrptr, 1, end_ranges); | 
 |  | 
 |       if (arange.ar_version != 2 && arange.ar_version != 3) | 
 | 	{ | 
 | 	  /* PR 19872: A version number of 0 probably means that there is | 
 | 	     padding at the end of the .debug_aranges section.  Gold puts | 
 | 	     it there when performing an incremental link, for example. | 
 | 	     So do not generate a warning in this case.  */ | 
 | 	  if (arange.ar_version) | 
 | 	    warn (_("Only DWARF 2 and 3 aranges are currently supported.\n")); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       printf (_("  Length:                   %" PRId64 "\n"), arange.ar_length); | 
 |       printf (_("  Version:                  %d\n"), arange.ar_version); | 
 |       printf (_("  Offset into .debug_info:  %#" PRIx64 "\n"), | 
 | 	      arange.ar_info_offset); | 
 |       printf (_("  Pointer Size:             %d\n"), arange.ar_pointer_size); | 
 |       printf (_("  Segment Size:             %d\n"), arange.ar_segment_size); | 
 |  | 
 |       address_size = arange.ar_pointer_size + arange.ar_segment_size; | 
 |  | 
 |       /* PR 17512: file: 001-108546-0.001:0.1.  */ | 
 |       if (address_size == 0 || address_size > 8) | 
 | 	{ | 
 | 	  error (_("Invalid address size in %s section!\n"), | 
 | 		 section->name); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       /* The DWARF spec does not require that the address size be a power | 
 | 	 of two, but we do.  This will have to change if we ever encounter | 
 | 	 an uneven architecture.  */ | 
 |       if ((address_size & (address_size - 1)) != 0) | 
 | 	{ | 
 | 	  warn (_("Pointer size + Segment size is not a power of two.\n")); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       if (address_size > 4) | 
 | 	printf (_("\n    Address            Length\n")); | 
 |       else | 
 | 	printf (_("\n    Address    Length\n")); | 
 |  | 
 |       addr_ranges = hdrptr; | 
 |  | 
 |       /* Must pad to an alignment boundary that is twice the address size.  */ | 
 |       addr_ranges += (2 * address_size - 1 | 
 | 		      - (hdrptr - start - 1) % (2 * address_size)); | 
 |  | 
 |       while (2 * address_size <= end_ranges - addr_ranges) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (address, addr_ranges, address_size, | 
 | 				 end_ranges); | 
 | 	  SAFE_BYTE_GET_AND_INC (length, addr_ranges, address_size, | 
 | 				 end_ranges); | 
 | 	  printf ("    "); | 
 | 	  print_hex (address, address_size); | 
 | 	  print_hex_ns (length, address_size); | 
 | 	  putchar ('\n'); | 
 | 	} | 
 |  | 
 |       start = end_ranges; | 
 |     } | 
 |  | 
 |   printf ("\n"); | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Comparison function for qsort.  */ | 
 | static int | 
 | comp_addr_base (const void * v0, const void * v1) | 
 | { | 
 |   debug_info *info0 = *(debug_info **) v0; | 
 |   debug_info *info1 = *(debug_info **) v1; | 
 |   return info0->addr_base - info1->addr_base; | 
 | } | 
 |  | 
 | /* Display the debug_addr section.  */ | 
 | static int | 
 | display_debug_addr (struct dwarf_section *section, | 
 | 		    void *file) | 
 | { | 
 |   debug_info **debug_addr_info; | 
 |   unsigned char *entry; | 
 |   unsigned char *end; | 
 |   unsigned int i; | 
 |   unsigned int count; | 
 |   unsigned char * header; | 
 |  | 
 |   if (section->size == 0) | 
 |     { | 
 |       printf (_("\nThe %s section is empty.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (load_debug_info (file) == 0) | 
 |     { | 
 |       warn (_("Unable to load/parse the .debug_info section, so cannot interpret the %s section.\n"), | 
 | 	    section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   /* PR  17531: file: cf38d01b. | 
 |      We use xcalloc because a corrupt file may not have initialised all of the | 
 |      fields in the debug_info structure, which means that the sort below might | 
 |      try to move uninitialised data.  */ | 
 |   debug_addr_info = (debug_info **) xcalloc ((num_debug_info_entries + 1), | 
 | 					     sizeof (debug_info *)); | 
 |  | 
 |   count = 0; | 
 |   for (i = 0; i < num_debug_info_entries; i++) | 
 |     if (debug_information [i].addr_base != DEBUG_INFO_UNAVAILABLE) | 
 |       { | 
 | 	/* PR 17531: file: cf38d01b.  */ | 
 | 	if (debug_information[i].addr_base >= section->size) | 
 | 	  warn (_("Corrupt address base (%#" PRIx64 ")" | 
 | 		  " found in debug section %u\n"), | 
 | 		debug_information[i].addr_base, i); | 
 | 	else | 
 | 	  debug_addr_info [count++] = debug_information + i; | 
 |       } | 
 |  | 
 |   /* Add a sentinel to make iteration convenient.  */ | 
 |   debug_addr_info [count] = (debug_info *) xmalloc (sizeof (debug_info)); | 
 |   debug_addr_info [count]->addr_base = section->size; | 
 |   qsort (debug_addr_info, count, sizeof (debug_info *), comp_addr_base); | 
 |  | 
 |   header = section->start; | 
 |   for (i = 0; i < count; i++) | 
 |     { | 
 |       unsigned int idx; | 
 |       unsigned int address_size = debug_addr_info [i]->pointer_size; | 
 |  | 
 |       printf (_("  For compilation unit at offset %#" PRIx64 ":\n"), | 
 | 	      debug_addr_info [i]->cu_offset); | 
 |  | 
 |       printf (_("\tIndex\tAddress\n")); | 
 |       entry = section->start + debug_addr_info [i]->addr_base; | 
 |       if (debug_addr_info [i]->dwarf_version >= 5) | 
 | 	{ | 
 | 	  size_t header_size = entry - header; | 
 | 	  unsigned char *curr_header = header; | 
 | 	  uint64_t length; | 
 | 	  int version; | 
 | 	  int segment_selector_size; | 
 |  | 
 | 	  if (header_size != 8 && header_size != 16) | 
 | 	    { | 
 | 	      warn (_("Corrupt %s section: expecting header size of 8 or 16, but found %zd instead\n"), | 
 | 		    section->name, header_size); | 
 | 	      break; | 
 | 	    } | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (length, curr_header, 4, entry); | 
 | 	  if (length == 0xffffffff) | 
 | 	    SAFE_BYTE_GET_AND_INC (length, curr_header, 8, entry); | 
 | 	  if (length > (size_t) (section->start + section->size - curr_header) | 
 | 	      || length < (size_t) (entry - curr_header)) | 
 | 	    { | 
 | 	      warn (_("Corrupt %s section: unit_length field of %#" PRIx64 | 
 | 		      " is invalid\n"), section->name, length); | 
 | 	      break; | 
 | 	    } | 
 | 	  end = curr_header + length; | 
 | 	  SAFE_BYTE_GET_AND_INC (version, curr_header, 2, entry); | 
 | 	  if (version != 5) | 
 | 	    warn (_("Corrupt %s section: expecting version number 5 in header but found %d instead\n"), | 
 | 		  section->name, version); | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (address_size, curr_header, 1, entry); | 
 | 	  SAFE_BYTE_GET_AND_INC (segment_selector_size, curr_header, 1, entry); | 
 | 	  address_size += segment_selector_size; | 
 | 	} | 
 |       else | 
 | 	end = section->start + debug_addr_info [i + 1]->addr_base; | 
 |  | 
 |       header = end; | 
 |       idx = 0; | 
 |  | 
 |       if (address_size < 1 || address_size > sizeof (uint64_t)) | 
 | 	{ | 
 | 	  warn (_("Corrupt %s section: address size (%x) is wrong\n"), | 
 | 		section->name, address_size); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       while ((size_t) (end - entry) >= address_size) | 
 | 	{ | 
 | 	  uint64_t base = byte_get (entry, address_size); | 
 | 	  printf (_("\t%d:\t"), idx); | 
 | 	  print_hex_ns (base, address_size); | 
 | 	  printf ("\n"); | 
 | 	  entry += address_size; | 
 | 	  idx++; | 
 | 	} | 
 |     } | 
 |   printf ("\n"); | 
 |  | 
 |   free (debug_addr_info[count]); | 
 |   free (debug_addr_info); | 
 |   return i == count; | 
 | } | 
 |  | 
 | /* Display the .debug_str_offsets and .debug_str_offsets.dwo sections.  */ | 
 |  | 
 | static int | 
 | display_debug_str_offsets (struct dwarf_section *section, | 
 | 			   void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned long idx; | 
 |  | 
 |   if (section->size == 0) | 
 |     { | 
 |       printf (_("\nThe %s section is empty.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |   unsigned char *curr = start; | 
 |   uint64_t debug_str_offsets_hdr_len; | 
 |  | 
 |   const char *suffix = strrchr (section->name, '.'); | 
 |   bool dwo = suffix && strcmp (suffix, ".dwo") == 0; | 
 |  | 
 |   if (dwo) | 
 |     load_debug_section_with_follow (str_dwo, file); | 
 |   else | 
 |     load_debug_section_with_follow (str, file); | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (curr < end) | 
 |     { | 
 |       uint64_t length; | 
 |       uint64_t entry_length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (length, curr, 4, end); | 
 |       /* FIXME: We assume that this means 64-bit DWARF is being used.  */ | 
 |       if (length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (length, curr, 8, end); | 
 | 	  entry_length = 8; | 
 | 	  debug_str_offsets_hdr_len = 16; | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  entry_length = 4; | 
 | 	  debug_str_offsets_hdr_len = 8; | 
 | 	} | 
 |  | 
 |       unsigned char *entries_end; | 
 |       if (length == 0) | 
 | 	{ | 
 | 	  /* This is probably an old style .debug_str_offset section which | 
 | 	     just contains offsets and no header (and the first offset is 0).  */ | 
 | 	  length = section->size; | 
 | 	  curr   = section->start; | 
 | 	  entries_end = end; | 
 | 	  debug_str_offsets_hdr_len = 0; | 
 |  | 
 | 	  printf (_("    Length: %#" PRIx64 "\n"), length); | 
 | 	  printf (_("       Index   Offset [String]\n")); | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  if (length <= (size_t) (end - curr)) | 
 | 	    entries_end = curr + length; | 
 | 	  else | 
 | 	    { | 
 | 	      warn (_("Section %s is too small %#" PRIx64 "\n"), | 
 | 		    section->name, section->size); | 
 | 	      entries_end = end; | 
 | 	    } | 
 |  | 
 | 	  int version; | 
 | 	  SAFE_BYTE_GET_AND_INC (version, curr, 2, entries_end); | 
 | 	  if (version != 5) | 
 | 	    warn (_("Unexpected version number in str_offset header: %#x\n"), version); | 
 |  | 
 | 	  int padding; | 
 | 	  SAFE_BYTE_GET_AND_INC (padding, curr, 2, entries_end); | 
 | 	  if (padding != 0) | 
 | 	    warn (_("Unexpected value in str_offset header's padding field: %#x\n"), padding); | 
 |  | 
 | 	  printf (_("    Length: %#" PRIx64 "\n"), length); | 
 | 	  printf (_("    Version: %#x\n"), version); | 
 | 	  printf (_("       Index   Offset [String]\n")); | 
 | 	} | 
 |  | 
 |       for (idx = 0; curr < entries_end; idx++) | 
 | 	{ | 
 | 	  uint64_t offset; | 
 | 	  const char *string; | 
 |  | 
 | 	  if ((size_t) (entries_end - curr) < entry_length) | 
 | 	    /* Not enough space to read one entry_length, give up.  */ | 
 | 	    return 0; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (offset, curr, entry_length, entries_end); | 
 | 	  if (dwo) | 
 | 	    string = fetch_indexed_string (idx, NULL, entry_length, dwo, | 
 | 					   debug_str_offsets_hdr_len); | 
 | 	  else | 
 | 	    string = fetch_indirect_string (offset); | 
 |  | 
 | 	  printf ("    %8lu ", idx); | 
 | 	  print_hex (offset, entry_length); | 
 | 	  printf (" %s\n", string); | 
 | 	} | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Each debug_information[x].range_lists[y] gets this representation for | 
 |    sorting purposes.  */ | 
 |  | 
 | struct range_entry | 
 | { | 
 |   /* The debug_information[x].range_lists[y] value.  */ | 
 |   uint64_t ranges_offset; | 
 |  | 
 |   /* Original debug_information to find parameters of the data.  */ | 
 |   debug_info *debug_info_p; | 
 | }; | 
 |  | 
 | /* Sort struct range_entry in ascending order of its RANGES_OFFSET.  */ | 
 |  | 
 | static int | 
 | range_entry_compar (const void *ap, const void *bp) | 
 | { | 
 |   const struct range_entry *a_re = (const struct range_entry *) ap; | 
 |   const struct range_entry *b_re = (const struct range_entry *) bp; | 
 |   const uint64_t a = a_re->ranges_offset; | 
 |   const uint64_t b = b_re->ranges_offset; | 
 |  | 
 |   return (a > b) - (b > a); | 
 | } | 
 |  | 
 | static unsigned char * | 
 | display_debug_ranges_list (unsigned char *  start, | 
 | 			   unsigned char *  finish, | 
 | 			   unsigned int     pointer_size, | 
 | 			   uint64_t         offset, | 
 | 			   uint64_t         base_address) | 
 | { | 
 |   while (start < finish) | 
 |     { | 
 |       uint64_t begin; | 
 |       uint64_t end; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, finish); | 
 |       if (start >= finish) | 
 | 	break; | 
 |       SAFE_SIGNED_BYTE_GET_AND_INC (end, start, pointer_size, finish); | 
 |  | 
 |       printf ("    "); | 
 |       print_hex (offset, 4); | 
 |  | 
 |       if (begin == 0 && end == 0) | 
 | 	{ | 
 | 	  printf (_("<End of list>\n")); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       /* Check base address specifiers.  */ | 
 |       if (is_max_address (begin, pointer_size) | 
 | 	  && !is_max_address (end, pointer_size)) | 
 | 	{ | 
 | 	  base_address = end; | 
 | 	  print_hex (begin, pointer_size); | 
 | 	  print_hex (end, pointer_size); | 
 | 	  printf ("(base address)\n"); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       print_hex (begin + base_address, pointer_size); | 
 |       print_hex_ns (end + base_address, pointer_size); | 
 |  | 
 |       if (begin == end) | 
 | 	fputs (_(" (start == end)"), stdout); | 
 |       else if (begin > end) | 
 | 	fputs (_(" (start > end)"), stdout); | 
 |  | 
 |       putchar ('\n'); | 
 |     } | 
 |  | 
 |   return start; | 
 | } | 
 |  | 
 | static unsigned char * | 
 | display_debug_rnglists_list (unsigned char * start, | 
 | 			     unsigned char * finish, | 
 | 			     unsigned int    pointer_size, | 
 | 			     uint64_t        offset, | 
 | 			     uint64_t        base_address, | 
 | 			     uint64_t        addr_base) | 
 | { | 
 |   unsigned char *next = start; | 
 |  | 
 |   while (1) | 
 |     { | 
 |       uint64_t off = offset + (start - next); | 
 |       enum dwarf_range_list_entry rlet; | 
 |       /* Initialize it due to a false compiler warning.  */ | 
 |       uint64_t begin = -1, length, end = -1; | 
 |  | 
 |       if (start >= finish) | 
 | 	{ | 
 | 	  warn (_("Range list starting at offset %#" PRIx64 | 
 | 		  " is not terminated.\n"), offset); | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       printf ("    "); | 
 |       print_hex (off, 4); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (rlet, start, 1, finish); | 
 |  | 
 |       switch (rlet) | 
 | 	{ | 
 | 	case DW_RLE_end_of_list: | 
 | 	  printf (_("<End of list>\n")); | 
 | 	  break; | 
 | 	case DW_RLE_base_addressx: | 
 | 	  READ_ULEB (base_address, start, finish); | 
 | 	  print_hex (base_address, pointer_size); | 
 | 	  printf (_("(base address index) ")); | 
 | 	  base_address = fetch_indexed_addr (base_address * pointer_size + addr_base, | 
 | 					     pointer_size); | 
 | 	  print_hex (base_address, pointer_size); | 
 | 	  printf (_("(base address)\n")); | 
 | 	  break; | 
 | 	case DW_RLE_startx_endx: | 
 | 	  READ_ULEB (begin, start, finish); | 
 | 	  READ_ULEB (end, start, finish); | 
 | 	  begin = fetch_indexed_addr (begin * pointer_size + addr_base, | 
 | 				      pointer_size); | 
 | 	  end   = fetch_indexed_addr (end * pointer_size + addr_base, | 
 | 				      pointer_size); | 
 | 	  break; | 
 | 	case DW_RLE_startx_length: | 
 | 	  READ_ULEB (begin, start, finish); | 
 | 	  READ_ULEB (length, start, finish); | 
 | 	  begin = fetch_indexed_addr (begin * pointer_size + addr_base, | 
 | 				      pointer_size); | 
 | 	  end = begin + length; | 
 | 	  break; | 
 | 	case DW_RLE_offset_pair: | 
 | 	  READ_ULEB (begin, start, finish); | 
 | 	  READ_ULEB (end, start, finish); | 
 | 	  break; | 
 | 	case DW_RLE_base_address: | 
 | 	  SAFE_BYTE_GET_AND_INC (base_address, start, pointer_size, finish); | 
 | 	  print_hex (base_address, pointer_size); | 
 | 	  printf (_("(base address)\n")); | 
 | 	  break; | 
 | 	case DW_RLE_start_end: | 
 | 	  SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, finish); | 
 | 	  SAFE_BYTE_GET_AND_INC (end, start, pointer_size, finish); | 
 | 	  break; | 
 | 	case DW_RLE_start_length: | 
 | 	  SAFE_BYTE_GET_AND_INC (begin, start, pointer_size, finish); | 
 | 	  READ_ULEB (length, start, finish); | 
 | 	  end = begin + length; | 
 | 	  break; | 
 | 	default: | 
 | 	  error (_("Invalid range list entry type %d\n"), rlet); | 
 | 	  rlet = DW_RLE_end_of_list; | 
 | 	  break; | 
 | 	} | 
 |  | 
 |       if (rlet == DW_RLE_end_of_list) | 
 | 	break; | 
 |       if (rlet == DW_RLE_base_address || rlet == DW_RLE_base_addressx) | 
 | 	continue; | 
 |  | 
 |       /* Only a DW_RLE_offset_pair needs the base address added.  */ | 
 |       if (rlet == DW_RLE_offset_pair) | 
 | 	{ | 
 | 	  begin += base_address; | 
 | 	  end += base_address; | 
 | 	} | 
 |  | 
 |       print_hex (begin, pointer_size); | 
 |       print_hex (end, pointer_size); | 
 |  | 
 |       if (begin == end) | 
 | 	fputs (_(" (start == end)"), stdout); | 
 |       else if (begin > end) | 
 | 	fputs (_(" (start > end)"), stdout); | 
 |  | 
 |       putchar ('\n'); | 
 |     } | 
 |  | 
 |   return start; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_rnglists_unit_header (struct dwarf_section *  section, | 
 | 				    uint64_t *              unit_offset, | 
 | 				    unsigned char *         poffset_size) | 
 | { | 
 |   uint64_t        start_offset = *unit_offset; | 
 |   unsigned char * p = section->start + start_offset; | 
 |   unsigned char * finish = section->start + section->size; | 
 |   uint64_t        initial_length; | 
 |   unsigned char   segment_selector_size; | 
 |   unsigned int    offset_entry_count; | 
 |   unsigned int    i; | 
 |   unsigned short  version; | 
 |   unsigned char   address_size = 0; | 
 |   unsigned char   offset_size; | 
 |  | 
 |   /* Get and check the length of the block.  */ | 
 |   SAFE_BYTE_GET_AND_INC (initial_length, p, 4, finish); | 
 |  | 
 |   if (initial_length == 0xffffffff) | 
 |     { | 
 |       /* This section is 64-bit DWARF 3.  */ | 
 |       SAFE_BYTE_GET_AND_INC (initial_length, p, 8, finish); | 
 |       *poffset_size = offset_size = 8; | 
 |     } | 
 |   else | 
 |     *poffset_size = offset_size = 4; | 
 |  | 
 |   if (initial_length > (size_t) (finish - p)) | 
 |     { | 
 |       /* If the length field has a relocation against it, then we should | 
 | 	 not complain if it is inaccurate (and probably negative). | 
 | 	 It is copied from .debug_line handling code.  */ | 
 |       if (reloc_at (section, (p - section->start) - offset_size)) | 
 | 	initial_length = finish - p; | 
 |       else | 
 | 	{ | 
 | 	  warn (_("The length field (%#" PRIx64 | 
 | 		  ") in the debug_rnglists header is wrong" | 
 | 		  " - the section is too small\n"), | 
 | 		initial_length); | 
 | 	  return 0; | 
 | 	} | 
 |     } | 
 |  | 
 |   /* Report the next unit offset to the caller.  */ | 
 |   *unit_offset = (p - section->start) + initial_length; | 
 |  | 
 |   /* Get the other fields in the header.  */ | 
 |   SAFE_BYTE_GET_AND_INC (version, p, 2, finish); | 
 |   SAFE_BYTE_GET_AND_INC (address_size, p, 1, finish); | 
 |   SAFE_BYTE_GET_AND_INC (segment_selector_size, p, 1, finish); | 
 |   SAFE_BYTE_GET_AND_INC (offset_entry_count, p, 4, finish); | 
 |  | 
 |   printf (_(" Table at Offset: %#" PRIx64 ":\n"), start_offset); | 
 |   printf (_("  Length:          %#" PRIx64 "\n"), initial_length); | 
 |   printf (_("  DWARF version:   %u\n"), version); | 
 |   printf (_("  Address size:    %u\n"), address_size); | 
 |   printf (_("  Segment size:    %u\n"), segment_selector_size); | 
 |   printf (_("  Offset entries:  %u\n"), offset_entry_count); | 
 |  | 
 |   /* Check the fields.  */ | 
 |   if (segment_selector_size != 0) | 
 |     { | 
 |       warn (_("The %s section contains " | 
 | 	      "unsupported segment selector size: %d.\n"), | 
 | 	    section->name, segment_selector_size); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (version < 5) | 
 |     { | 
 |       warn (_("Only DWARF version 5+ debug_rnglists info " | 
 | 	      "is currently supported.\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (offset_entry_count != 0) | 
 |     { | 
 |       printf (_("\n   Offsets starting at %#tx:\n"), p - section->start); | 
 |  | 
 |       for (i = 0; i < offset_entry_count; i++) | 
 | 	{ | 
 | 	  uint64_t entry; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (entry, p, offset_size, finish); | 
 | 	  printf (_("    [%6u] %#" PRIx64 "\n"), i, entry); | 
 | 	} | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static bool | 
 | is_range_list_for_this_section (bool is_rnglists, unsigned int version) | 
 | { | 
 |   if (is_rnglists && version > 4) | 
 |     return true; | 
 |  | 
 |   if (! is_rnglists && version < 5) | 
 |     return true; | 
 |  | 
 |   return false; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_ranges (struct dwarf_section *section, | 
 | 		      void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *last_start = start; | 
 |   unsigned char *last_end; | 
 |   uint64_t bytes = section->size; | 
 |   unsigned char *section_begin = start; | 
 |   unsigned char *finish = start + bytes; | 
 |   unsigned int num_range_list, i; | 
 |   struct range_entry *range_entries; | 
 |   struct range_entry *range_entry_fill; | 
 |   bool is_rnglists = strstr (section->name, "debug_rnglists") != NULL; | 
 |   uint64_t last_offset = 0; | 
 |   uint64_t next_rnglists_cu_offset = 0; | 
 |   unsigned char offset_size; | 
 |  | 
 |   if (bytes == 0) | 
 |     { | 
 |       printf (_("\nThe %s section is empty.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   if (load_debug_info (file) == 0) | 
 |     { | 
 |       warn (_("Unable to load/parse the .debug_info section, so cannot interpret the %s section.\n"), | 
 | 	    section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   num_range_list = 0; | 
 |   for (i = 0; i < num_debug_info_entries; i++) | 
 |     if (is_range_list_for_this_section (is_rnglists, debug_information [i].dwarf_version)) | 
 |       num_range_list += debug_information [i].num_range_lists; | 
 |  | 
 |   if (num_range_list == 0) | 
 |     { | 
 |       /* This can happen when the file was compiled with -gsplit-debug | 
 | 	 which removes references to range lists from the primary .o file.  */ | 
 |       printf (_("No range lists referenced by .debug_info section.\n")); | 
 |       return 1; | 
 |     } | 
 |  | 
 |   range_entry_fill = range_entries = XNEWVEC (struct range_entry, num_range_list); | 
 |  | 
 |   for (i = 0; i < num_debug_info_entries; i++) | 
 |     { | 
 |       debug_info *debug_info_p = &debug_information[i]; | 
 |       unsigned int j; | 
 |  | 
 |       for (j = 0; j < debug_info_p->num_range_lists; j++) | 
 | 	{ | 
 | 	  if (is_range_list_for_this_section (is_rnglists, debug_info_p->dwarf_version)) | 
 | 	    { | 
 | 	      range_entry_fill->ranges_offset = debug_info_p->range_lists[j]; | 
 | 	      range_entry_fill->debug_info_p = debug_info_p; | 
 | 	      range_entry_fill++; | 
 | 	    } | 
 | 	} | 
 |     } | 
 |  | 
 |   assert (range_entry_fill >= range_entries); | 
 |   assert (num_range_list >= (unsigned int)(range_entry_fill - range_entries)); | 
 |   num_range_list = range_entry_fill - range_entries; | 
 |   qsort (range_entries, num_range_list, sizeof (*range_entries), | 
 | 	 range_entry_compar); | 
 |  | 
 |   putchar ('\n'); | 
 |   if (!is_rnglists) | 
 |     printf (_("    Offset   Begin    End\n")); | 
 |  | 
 |   last_end = NULL; | 
 |   for (i = 0; i < num_range_list; i++) | 
 |     { | 
 |       struct range_entry *range_entry = &range_entries[i]; | 
 |       debug_info *debug_info_p = range_entry->debug_info_p; | 
 |       unsigned int pointer_size; | 
 |       uint64_t offset; | 
 |       unsigned char *next; | 
 |       uint64_t base_address; | 
 |  | 
 |       pointer_size = debug_info_p->pointer_size; | 
 |       offset = range_entry->ranges_offset; | 
 |       base_address = debug_info_p->base_address; | 
 |  | 
 |       /* PR 17512: file: 001-101485-0.001:0.1.  */ | 
 |       if (pointer_size < 2 || pointer_size > 8) | 
 | 	{ | 
 | 	  warn (_("Corrupt pointer size (%d) in debug entry at offset %#" PRIx64 "\n"), | 
 | 		pointer_size, offset); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if (offset > (size_t) (finish - section_begin)) | 
 | 	{ | 
 | 	  warn (_("Corrupt offset (%#" PRIx64 ") in range entry %u\n"), | 
 | 		offset, i); | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       /* If we've moved on to the next compile unit in the rnglists section - dump the unit header(s).  */ | 
 |       if (is_rnglists && next_rnglists_cu_offset < offset) | 
 | 	{ | 
 | 	  while (next_rnglists_cu_offset < offset) | 
 | 	    display_debug_rnglists_unit_header (section, &next_rnglists_cu_offset, &offset_size); | 
 | 	  printf (_("    Offset   Begin    End\n")); | 
 | 	} | 
 |  | 
 |       next = section_begin + offset; /* Offset is from the section start, the base has already been added.  */ | 
 |  | 
 |       if (i == 0) | 
 | 	{ | 
 | 	  last_end = section_begin; | 
 | 	  if (is_rnglists) | 
 | 	    last_end += 2 * offset_size - 4 + 2 + 1 + 1 + 4; | 
 | 	} | 
 |       /* If multiple DWARF entities reference the same range then we will | 
 | 	 have multiple entries in the `range_entries' list for the same | 
 | 	 offset.  Thanks to the sort above these will all be consecutive in | 
 | 	 the `range_entries' list, so we can easily ignore duplicates | 
 | 	 here.  */ | 
 |       if (i > 0 && last_offset == offset) | 
 | 	continue; | 
 |       last_offset = offset; | 
 |  | 
 |       if (dwarf_check != 0) | 
 | 	{ | 
 | 	  if (start < next) | 
 | 	    { | 
 | 	      if (last_end != next) | 
 | 		warn (_("There is a hole [%#tx - %#tx] in %s section.\n"), | 
 | 		      last_end - section_begin, next - section_begin, | 
 | 		      section->name); | 
 | 	    } | 
 | 	  else if (start > next) | 
 | 	    { | 
 | 	      if (next == last_start) | 
 | 		continue; | 
 | 	      warn (_("There is an overlap [%#tx - %#tx] in %s section.\n"), | 
 | 		    start - section_begin, next - section_begin, section->name); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       start = next; | 
 |       last_start = next; | 
 |  | 
 |       if (is_rnglists) | 
 | 	last_end | 
 | 	  = display_debug_rnglists_list | 
 | 	      (start, finish, pointer_size, offset, base_address, | 
 | 	       debug_info_p->addr_base); | 
 |       else | 
 | 	last_end | 
 | 	  = display_debug_ranges_list | 
 | 	      (start, finish, pointer_size, offset, base_address); | 
 |     } | 
 |  | 
 |   /* Display trailing empty (or unreferenced) compile units, if any.  */ | 
 |   if (is_rnglists) | 
 |     while (next_rnglists_cu_offset < section->size) | 
 |       display_debug_rnglists_unit_header (section, &next_rnglists_cu_offset, &offset_size); | 
 |  | 
 |   putchar ('\n'); | 
 |  | 
 |   free (range_entries); | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | typedef struct Frame_Chunk | 
 | { | 
 |   struct Frame_Chunk *next; | 
 |   unsigned char *chunk_start; | 
 |   unsigned int ncols; | 
 |   /* DW_CFA_{undefined,same_value,offset,register,unreferenced}  */ | 
 |   short int *col_type; | 
 |   int64_t *col_offset; | 
 |   char *augmentation; | 
 |   unsigned int code_factor; | 
 |   int data_factor; | 
 |   uint64_t pc_begin; | 
 |   uint64_t pc_range; | 
 |   unsigned int cfa_reg; | 
 |   uint64_t cfa_offset; | 
 |   bool cfa_ofs_signed_p; | 
 |   unsigned int ra; | 
 |   unsigned char fde_encoding; | 
 |   unsigned char cfa_exp; | 
 |   unsigned char ptr_size; | 
 |   unsigned char segment_size; | 
 | } | 
 | Frame_Chunk; | 
 |  | 
 | typedef bool (*is_mach_augmentation_ftype) (char c); | 
 | static is_mach_augmentation_ftype is_mach_augmentation; | 
 | typedef const char *(*dwarf_regname_lookup_ftype) (unsigned int); | 
 | static dwarf_regname_lookup_ftype dwarf_regnames_lookup_func; | 
 | static const char *const *dwarf_regnames; | 
 | static unsigned int dwarf_regnames_count; | 
 | static bool is_aarch64; | 
 |  | 
 | /* A marker for a col_type that means this column was never referenced | 
 |    in the frame info.  */ | 
 | #define DW_CFA_unreferenced (-1) | 
 |  | 
 | /* Return 0 if no more space is needed, 1 if more space is needed, | 
 |    -1 for invalid reg.  */ | 
 |  | 
 | static int | 
 | frame_need_space (Frame_Chunk *fc, unsigned int reg) | 
 | { | 
 |   unsigned int prev = fc->ncols; | 
 |  | 
 |   if (reg < (unsigned int) fc->ncols) | 
 |     return 0; | 
 |  | 
 |   if (dwarf_regnames_count > 0 | 
 |       && reg > dwarf_regnames_count) | 
 |     return -1; | 
 |  | 
 |   fc->ncols = reg + 1; | 
 |   /* PR 17512: file: 10450-2643-0.004. | 
 |      If reg == -1 then this can happen...  */ | 
 |   if (fc->ncols == 0) | 
 |     return -1; | 
 |  | 
 |   /* PR 17512: file: 2844a11d.  */ | 
 |   if (fc->ncols > 1024 && dwarf_regnames_count == 0) | 
 |     { | 
 |       error (_("Unfeasibly large register number: %u\n"), reg); | 
 |       fc->ncols = 0; | 
 |       /* FIXME: 1024 is an arbitrary limit.  Increase it if | 
 | 	 we ever encounter a valid binary that exceeds it.  */ | 
 |       return -1; | 
 |     } | 
 |  | 
 |   fc->col_type = xcrealloc (fc->col_type, fc->ncols, | 
 | 			    sizeof (*fc->col_type)); | 
 |   fc->col_offset = xcrealloc (fc->col_offset, fc->ncols, | 
 | 			      sizeof (*fc->col_offset)); | 
 |   /* PR 17512: file:002-10025-0.005.  */ | 
 |   if (fc->col_type == NULL || fc->col_offset == NULL) | 
 |     { | 
 |       error (_("Out of memory allocating %u columns in dwarf frame arrays\n"), | 
 | 	     fc->ncols); | 
 |       fc->ncols = 0; | 
 |       return -1; | 
 |     } | 
 |  | 
 |   while (prev < fc->ncols) | 
 |     { | 
 |       fc->col_type[prev] = DW_CFA_unreferenced; | 
 |       fc->col_offset[prev] = 0; | 
 |       prev++; | 
 |     } | 
 |   return 1; | 
 | } | 
 |  | 
 | static const char *const dwarf_regnames_i386[] = | 
 | { | 
 |   "eax", "ecx", "edx", "ebx",			  /* 0 - 3  */ | 
 |   "esp", "ebp", "esi", "edi",			  /* 4 - 7  */ | 
 |   "eip", "eflags", NULL,			  /* 8 - 10  */ | 
 |   "st0", "st1", "st2", "st3",			  /* 11 - 14  */ | 
 |   "st4", "st5", "st6", "st7",			  /* 15 - 18  */ | 
 |   NULL, NULL,					  /* 19 - 20  */ | 
 |   "xmm0", "xmm1", "xmm2", "xmm3",		  /* 21 - 24  */ | 
 |   "xmm4", "xmm5", "xmm6", "xmm7",		  /* 25 - 28  */ | 
 |   "mm0", "mm1", "mm2", "mm3",			  /* 29 - 32  */ | 
 |   "mm4", "mm5", "mm6", "mm7",			  /* 33 - 36  */ | 
 |   "fcw", "fsw", "mxcsr",			  /* 37 - 39  */ | 
 |   "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL, /* 40 - 47  */ | 
 |   "tr", "ldtr",					  /* 48 - 49  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 50 - 57  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 58 - 65  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 66 - 73  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 74 - 81  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 82 - 89  */ | 
 |   NULL, NULL, NULL,				  /* 90 - 92  */ | 
 |   "k0", "k1", "k2", "k3", "k4", "k5", "k6", "k7"  /* 93 - 100  */ | 
 | }; | 
 |  | 
 | static const char *const dwarf_regnames_iamcu[] = | 
 | { | 
 |   "eax", "ecx", "edx", "ebx",			  /* 0 - 3  */ | 
 |   "esp", "ebp", "esi", "edi",			  /* 4 - 7  */ | 
 |   "eip", "eflags", NULL,			  /* 8 - 10  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 11 - 18  */ | 
 |   NULL, NULL,					  /* 19 - 20  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 21 - 28  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 29 - 36  */ | 
 |   NULL, NULL, NULL,				  /* 37 - 39  */ | 
 |   "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL, /* 40 - 47  */ | 
 |   "tr", "ldtr",					  /* 48 - 49  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 50 - 57  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 58 - 65  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 66 - 73  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 74 - 81  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 82 - 89  */ | 
 |   NULL, NULL, NULL,				  /* 90 - 92  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL  /* 93 - 100  */ | 
 | }; | 
 |  | 
 | static void | 
 | init_dwarf_regnames_i386 (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_i386; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_i386); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 | } | 
 |  | 
 | static void | 
 | init_dwarf_regnames_iamcu (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_iamcu; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_iamcu); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 | } | 
 |  | 
 | static const char *const DW_CFA_GNU_window_save_name[] = | 
 | { | 
 |   "DW_CFA_GNU_window_save", | 
 |   "DW_CFA_AARCH64_negate_ra_state" | 
 | }; | 
 |  | 
 | static const char *const dwarf_regnames_x86_64[] = | 
 | { | 
 |   "rax", "rdx", "rcx", "rbx", | 
 |   "rsi", "rdi", "rbp", "rsp", | 
 |   "r8",  "r9",  "r10", "r11", | 
 |   "r12", "r13", "r14", "r15", | 
 |   "rip", | 
 |   "xmm0",  "xmm1",  "xmm2",  "xmm3", | 
 |   "xmm4",  "xmm5",  "xmm6",  "xmm7", | 
 |   "xmm8",  "xmm9",  "xmm10", "xmm11", | 
 |   "xmm12", "xmm13", "xmm14", "xmm15", | 
 |   "st0", "st1", "st2", "st3", | 
 |   "st4", "st5", "st6", "st7", | 
 |   "mm0", "mm1", "mm2", "mm3", | 
 |   "mm4", "mm5", "mm6", "mm7", | 
 |   "rflags", | 
 |   "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL, | 
 |   "fs.base", "gs.base", NULL, NULL, | 
 |   "tr", "ldtr", | 
 |   "mxcsr", "fcw", "fsw", | 
 |   "xmm16",  "xmm17",  "xmm18",  "xmm19", | 
 |   "xmm20",  "xmm21",  "xmm22",  "xmm23", | 
 |   "xmm24",  "xmm25",  "xmm26",  "xmm27", | 
 |   "xmm28",  "xmm29",  "xmm30",  "xmm31", | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 83 - 90  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 91 - 98  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 99 - 106  */ | 
 |   NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, /* 107 - 114  */ | 
 |   NULL, NULL, NULL,				  /* 115 - 117  */ | 
 |   "k0", "k1", "k2", "k3", "k4", "k5", "k6", "k7", | 
 |   "bnd0", "bnd1", "bnd2", "bnd3", | 
 | }; | 
 |  | 
 | static void | 
 | init_dwarf_regnames_x86_64 (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_x86_64; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_x86_64); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 | } | 
 |  | 
 | static const char *const dwarf_regnames_aarch64[] = | 
 | { | 
 |    "x0",  "x1",  "x2",  "x3",  "x4",  "x5",  "x6",  "x7", | 
 |    "x8",  "x9", "x10", "x11", "x12", "x13", "x14", "x15", | 
 |   "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", | 
 |   "x24", "x25", "x26", "x27", "x28", "x29", "x30", "sp", | 
 |    NULL, "elr",  NULL,  NULL,  NULL,  NULL,  NULL,  NULL, | 
 |    NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  "vg", "ffr", | 
 |    "p0",  "p1",  "p2",  "p3",  "p4",  "p5",  "p6",  "p7", | 
 |    "p8",  "p9", "p10", "p11", "p12", "p13", "p14", "p15", | 
 |    "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7", | 
 |    "v8",  "v9", "v10", "v11", "v12", "v13", "v14", "v15", | 
 |   "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", | 
 |   "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", | 
 |    "z0",  "z1",  "z2",  "z3",  "z4",  "z5",  "z6",  "z7", | 
 |    "z8",  "z9", "z10", "z11", "z12", "z13", "z14", "z15", | 
 |   "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", | 
 |   "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31", | 
 | }; | 
 |  | 
 | static void | 
 | init_dwarf_regnames_aarch64 (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_aarch64; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_aarch64); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 |   is_aarch64 = true; | 
 | } | 
 |  | 
 | static const char *const dwarf_regnames_s390[] = | 
 | { | 
 |   /* Avoid saying "r5 (r5)", so omit the names of r0-r15.  */ | 
 |   NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL, | 
 |   NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL, | 
 |   "f0",  "f2",  "f4",  "f6",  "f1",  "f3",  "f5",  "f7", | 
 |   "f8",  "f10", "f12", "f14", "f9",  "f11", "f13", "f15", | 
 |   "cr0", "cr1", "cr2", "cr3", "cr4", "cr5", "cr6", "cr7", | 
 |   "cr8", "cr9", "cr10", "cr11", "cr12", "cr13", "cr14", "cr15", | 
 |   "a0",  "a1",  "a2",  "a3",  "a4",  "a5",  "a6",  "a7", | 
 |   "a8",  "a9",  "a10", "a11", "a12", "a13", "a14", "a15", | 
 |   "pswm", "pswa", | 
 |   NULL, NULL, | 
 |   "v16", "v18", "v20", "v22", "v17", "v19", "v21", "v23", | 
 |   "v24", "v26", "v28", "v30", "v25", "v27", "v29", "v31", | 
 | }; | 
 |  | 
 | static void | 
 | init_dwarf_regnames_s390 (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_s390; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_s390); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 | } | 
 |  | 
 | static const char *const dwarf_regnames_riscv[] = | 
 | { | 
 |  "zero", "ra",   "sp",   "gp",  "tp",  "t0",  "t1",  "t2",  /*   0 -   7 */ | 
 |  "s0",   "s1",   "a0",   "a1",  "a2",  "a3",  "a4",  "a5",  /*   8 -  15 */ | 
 |  "a6",   "a7",   "s2",   "s3",  "s4",  "s5",  "s6",  "s7",  /*  16 -  23 */ | 
 |  "s8",   "s9",   "s10",  "s11", "t3",  "t4",  "t5",  "t6",  /*  24 -  31 */ | 
 |  "ft0",  "ft1",  "ft2",  "ft3", "ft4", "ft5", "ft6", "ft7", /*  32 -  39 */ | 
 |  "fs0",  "fs1",                                             /*  40 -  41 */ | 
 |  "fa0",  "fa1",  "fa2",  "fa3", "fa4", "fa5", "fa6", "fa7", /*  42 -  49 */ | 
 |  "fs2",  "fs3",  "fs4",  "fs5", "fs6", "fs7", "fs8", "fs9", /*  50 -  57 */ | 
 |  "fs10", "fs11",                                            /*  58 -  59 */ | 
 |  "ft8",  "ft9",  "ft10", "ft11",                            /*  60 -  63 */ | 
 |  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,     /*  64 -  71 */ | 
 |  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,     /*  72 -  79 */ | 
 |  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,     /*  80 -  87 */ | 
 |  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,  NULL,     /*  88 -  95 */ | 
 |  "v0",  "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",     /*  96 - 103 */ | 
 |  "v8",  "v9",  "v10", "v11", "v12", "v13", "v14", "v15",    /* 104 - 111 */ | 
 |  "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",    /* 112 - 119 */ | 
 |  "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",    /* 120 - 127 */ | 
 | }; | 
 |  | 
 | /* A RISC-V replacement for REGNAME_INTERNAL_BY_TABLE_ONLY which handles | 
 |    the large number of CSRs.  */ | 
 |  | 
 | static const char * | 
 | regname_internal_riscv (unsigned int regno) | 
 | { | 
 |   const char *name = NULL; | 
 |  | 
 |   /* Lookup in the table first, this covers GPR and FPR.  */ | 
 |   if (regno < ARRAY_SIZE (dwarf_regnames_riscv)) | 
 |     name = dwarf_regnames_riscv [regno]; | 
 |   else if (regno >= 4096 && regno <= 8191) | 
 |     { | 
 |       /* This might be a CSR, these live in a sparse number space from 4096 | 
 | 	 to 8191  These numbers are defined in the RISC-V ELF ABI | 
 | 	 document.  */ | 
 |       switch (regno) | 
 | 	{ | 
 | #define DECLARE_CSR(NAME,VALUE,CLASS,DEFINE_VER,ABORT_VER) \ | 
 |   case VALUE + 4096: name = #NAME; break; | 
 | #include "opcode/riscv-opc.h" | 
 | #undef DECLARE_CSR | 
 |  | 
 | 	default: | 
 | 	  { | 
 | 	    static char csr_name[10]; | 
 | 	    snprintf (csr_name, sizeof (csr_name), "csr%d", (regno - 4096)); | 
 | 	    name = csr_name; | 
 | 	  } | 
 | 	  break; | 
 | 	} | 
 |     } | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static void | 
 | init_dwarf_regnames_riscv (void) | 
 | { | 
 |   dwarf_regnames = NULL; | 
 |   dwarf_regnames_count = 8192; | 
 |   dwarf_regnames_lookup_func = regname_internal_riscv; | 
 | } | 
 |  | 
 | static const char *const dwarf_regnames_loongarch[] = | 
 | { | 
 |   "$zero", "$ra", "$tp", "$sp", "$a0", "$a1", "$a2", "$a3",  /* 0-7   */ | 
 |   "$a4",   "$a5", "$a6", "$a7", "$t0", "$t1", "$t2", "$t3",  /* 8-15  */ | 
 |   "$t4",   "$t5", "$t6", "$t7", "$t8", "$r21","$fp", "$s0",  /* 16-23 */ | 
 |   "$s1",   "$s2", "$s3", "$s4", "$s5", "$s6", "$s7", "$s8",  /* 24-31 */ | 
 |   "$fa0", "$fa1", "$fa2", "$fa3", "$fa4",  "$fa5",  "$fa6",  /* 32-38 */ | 
 |   "$fa7", "$ft0", "$ft1", "$ft2", "$ft3",  "$ft4",  "$ft5",  /* 39-45 */ | 
 |   "$ft6", "$ft7", "$ft8", "$ft9", "$ft10", "$ft11", "$ft12", /* 46-52 */ | 
 |   "$ft13", "$ft14", "$ft15", "$fs0", "$fs1", "$fs2", "$fs3", /* 53-59 */ | 
 |   "$fs4",  "$fs5",  "$fs6",  "$fs7",			     /* 60-63 */ | 
 | }; | 
 |  | 
 | static void | 
 | init_dwarf_regnames_loongarch (void) | 
 | { | 
 |   dwarf_regnames = dwarf_regnames_loongarch; | 
 |   dwarf_regnames_count = ARRAY_SIZE (dwarf_regnames_loongarch); | 
 |   dwarf_regnames_lookup_func = regname_internal_by_table_only; | 
 | } | 
 |  | 
 | static bool | 
 | is_nomach_augmentation (char c ATTRIBUTE_UNUSED) | 
 | { | 
 |   return false; | 
 | } | 
 |  | 
 | static bool | 
 | is_aarch64_augmentation (char c) | 
 | { | 
 |   return (c == 'B' || c == 'G'); | 
 | } | 
 |  | 
 | void | 
 | init_dwarf_by_elf_machine_code (unsigned int e_machine) | 
 | { | 
 |   is_mach_augmentation = is_nomach_augmentation; | 
 |   dwarf_regnames_lookup_func = NULL; | 
 |   is_aarch64 = false; | 
 |  | 
 |   switch (e_machine) | 
 |     { | 
 |     case EM_386: | 
 |       init_dwarf_regnames_i386 (); | 
 |       break; | 
 |  | 
 |     case EM_IAMCU: | 
 |       init_dwarf_regnames_iamcu (); | 
 |       break; | 
 |  | 
 |     case EM_X86_64: | 
 |     case EM_L1OM: | 
 |     case EM_K1OM: | 
 |       init_dwarf_regnames_x86_64 (); | 
 |       break; | 
 |  | 
 |     case EM_AARCH64: | 
 |       init_dwarf_regnames_aarch64 (); | 
 |       is_mach_augmentation = is_aarch64_augmentation; | 
 |       break; | 
 |  | 
 |     case EM_S390: | 
 |       init_dwarf_regnames_s390 (); | 
 |       break; | 
 |  | 
 |     case EM_RISCV: | 
 |       init_dwarf_regnames_riscv (); | 
 |       break; | 
 |  | 
 |     case EM_LOONGARCH: | 
 |       init_dwarf_regnames_loongarch (); | 
 |       break; | 
 |  | 
 |     default: | 
 |       break; | 
 |     } | 
 | } | 
 |  | 
 | /* Initialize the DWARF register name lookup state based on the | 
 |    architecture and specific machine type of a BFD.  */ | 
 |  | 
 | void | 
 | init_dwarf_by_bfd_arch_and_mach (enum bfd_architecture arch, | 
 | 				 unsigned long mach) | 
 | { | 
 |   is_mach_augmentation = is_nomach_augmentation; | 
 |   dwarf_regnames_lookup_func = NULL; | 
 |   is_aarch64 = false; | 
 |  | 
 |   switch (arch) | 
 |     { | 
 |     case bfd_arch_i386: | 
 |       switch (mach) | 
 | 	{ | 
 | 	case bfd_mach_x86_64: | 
 | 	case bfd_mach_x86_64_intel_syntax: | 
 | 	case bfd_mach_x64_32: | 
 | 	case bfd_mach_x64_32_intel_syntax: | 
 | 	  init_dwarf_regnames_x86_64 (); | 
 | 	  break; | 
 |  | 
 | 	default: | 
 | 	  init_dwarf_regnames_i386 (); | 
 | 	  break; | 
 | 	} | 
 |       break; | 
 |  | 
 |     case bfd_arch_iamcu: | 
 |       init_dwarf_regnames_iamcu (); | 
 |       break; | 
 |  | 
 |     case bfd_arch_aarch64: | 
 |       init_dwarf_regnames_aarch64(); | 
 |       is_mach_augmentation = is_aarch64_augmentation; | 
 |       break; | 
 |  | 
 |     case bfd_arch_s390: | 
 |       init_dwarf_regnames_s390 (); | 
 |       break; | 
 |  | 
 |     case bfd_arch_riscv: | 
 |       init_dwarf_regnames_riscv (); | 
 |       break; | 
 |  | 
 |     case bfd_arch_loongarch: | 
 |       init_dwarf_regnames_loongarch (); | 
 |       break; | 
 |  | 
 |     default: | 
 |       break; | 
 |     } | 
 | } | 
 |  | 
 | static const char * | 
 | regname_internal_by_table_only (unsigned int regno) | 
 | { | 
 |   if (dwarf_regnames != NULL | 
 |       && regno < dwarf_regnames_count | 
 |       && dwarf_regnames [regno] != NULL) | 
 |     return dwarf_regnames [regno]; | 
 |  | 
 |   return NULL; | 
 | } | 
 |  | 
 | static const char * | 
 | regname (unsigned int regno, int name_only_p) | 
 | { | 
 |   static char reg[64]; | 
 |  | 
 |   const char *name = NULL; | 
 |  | 
 |   if (dwarf_regnames_lookup_func != NULL) | 
 |     name = dwarf_regnames_lookup_func (regno); | 
 |  | 
 |   if (name != NULL) | 
 |     { | 
 |       if (name_only_p) | 
 | 	return name; | 
 |       snprintf (reg, sizeof (reg), "r%d (%s)", regno, name); | 
 |     } | 
 |   else | 
 |     snprintf (reg, sizeof (reg), "r%d", regno); | 
 |   return reg; | 
 | } | 
 |  | 
 | static void | 
 | frame_display_row (Frame_Chunk *fc, int *need_col_headers, unsigned int *max_regs) | 
 | { | 
 |   unsigned int r; | 
 |   char tmp[100]; | 
 |  | 
 |   if (*max_regs != fc->ncols) | 
 |     *max_regs = fc->ncols; | 
 |  | 
 |   if (*need_col_headers) | 
 |     { | 
 |       *need_col_headers = 0; | 
 |  | 
 |       printf ("%-*s CFA      ", eh_addr_size * 2, "   LOC"); | 
 |  | 
 |       for (r = 0; r < *max_regs; r++) | 
 | 	if (fc->col_type[r] != DW_CFA_unreferenced) | 
 | 	  { | 
 | 	    if (r == fc->ra) | 
 | 	      printf ("ra    "); | 
 | 	    else | 
 | 	      printf ("%-5s ", regname (r, 1)); | 
 | 	  } | 
 |  | 
 |       printf ("\n"); | 
 |     } | 
 |  | 
 |   print_hex (fc->pc_begin, eh_addr_size); | 
 |   if (fc->cfa_exp) | 
 |     strcpy (tmp, "exp"); | 
 |   else | 
 |     sprintf (tmp, (fc->cfa_ofs_signed_p ? "%s%+" PRId64 : "%s+%" PRIu64), | 
 | 	     regname (fc->cfa_reg, 1), fc->cfa_offset); | 
 |   printf ("%-8s ", tmp); | 
 |  | 
 |   for (r = 0; r < fc->ncols; r++) | 
 |     { | 
 |       if (fc->col_type[r] != DW_CFA_unreferenced) | 
 | 	{ | 
 | 	  switch (fc->col_type[r]) | 
 | 	    { | 
 | 	    case DW_CFA_undefined: | 
 | 	      strcpy (tmp, "u"); | 
 | 	      break; | 
 | 	    case DW_CFA_same_value: | 
 | 	      strcpy (tmp, "s"); | 
 | 	      break; | 
 | 	    case DW_CFA_offset: | 
 | 	      sprintf (tmp, "c%+" PRId64, fc->col_offset[r]); | 
 | 	      break; | 
 | 	    case DW_CFA_val_offset: | 
 | 	      sprintf (tmp, "v%+" PRId64, fc->col_offset[r]); | 
 | 	      break; | 
 | 	    case DW_CFA_register: | 
 | 	      sprintf (tmp, "%s", regname (fc->col_offset[r], 0)); | 
 | 	      break; | 
 | 	    case DW_CFA_expression: | 
 | 	      strcpy (tmp, "exp"); | 
 | 	      break; | 
 | 	    case DW_CFA_val_expression: | 
 | 	      strcpy (tmp, "vexp"); | 
 | 	      break; | 
 | 	    default: | 
 | 	      strcpy (tmp, "n/a"); | 
 | 	      break; | 
 | 	    } | 
 | 	  printf ("%-5s ", tmp); | 
 | 	} | 
 |     } | 
 |   printf ("\n"); | 
 | } | 
 |  | 
 | #define GET(VAR, N)	SAFE_BYTE_GET_AND_INC (VAR, start, N, end) | 
 |  | 
 | static unsigned char * | 
 | read_cie (unsigned char *start, unsigned char *end, | 
 | 	  Frame_Chunk **p_cie, int *p_version, | 
 | 	  uint64_t *p_aug_len, unsigned char **p_aug) | 
 | { | 
 |   int version; | 
 |   Frame_Chunk *fc; | 
 |   unsigned char *augmentation_data = NULL; | 
 |   uint64_t augmentation_data_len = 0; | 
 |  | 
 |   * p_cie = NULL; | 
 |   /* PR 17512: file: 001-228113-0.004.  */ | 
 |   if (start >= end) | 
 |     return end; | 
 |  | 
 |   fc = (Frame_Chunk *) xmalloc (sizeof (Frame_Chunk)); | 
 |   memset (fc, 0, sizeof (Frame_Chunk)); | 
 |  | 
 |   fc->col_type = xmalloc (sizeof (*fc->col_type)); | 
 |   fc->col_offset = xmalloc (sizeof (*fc->col_offset)); | 
 |  | 
 |   version = *start++; | 
 |  | 
 |   fc->augmentation = (char *) start; | 
 |   /* PR 17512: file: 001-228113-0.004. | 
 |      Skip past augmentation name, but avoid running off the end of the data.  */ | 
 |   while (start < end) | 
 |     if (* start ++ == '\0') | 
 |       break; | 
 |   if (start == end) | 
 |     { | 
 |       warn (_("No terminator for augmentation name\n")); | 
 |       goto fail; | 
 |     } | 
 |  | 
 |   if (strcmp (fc->augmentation, "eh") == 0) | 
 |     { | 
 |       if (eh_addr_size > (size_t) (end - start)) | 
 | 	goto fail; | 
 |       start += eh_addr_size; | 
 |     } | 
 |  | 
 |   if (version >= 4) | 
 |     { | 
 |       if (2 > (size_t) (end - start)) | 
 | 	goto fail; | 
 |       GET (fc->ptr_size, 1); | 
 |       if (fc->ptr_size < 1 || fc->ptr_size > 8) | 
 | 	{ | 
 | 	  warn (_("Invalid pointer size (%d) in CIE data\n"), fc->ptr_size); | 
 | 	  goto fail; | 
 | 	} | 
 |  | 
 |       GET (fc->segment_size, 1); | 
 |       /* PR 17512: file: e99d2804.  */ | 
 |       if (fc->segment_size > 8 || fc->segment_size + fc->ptr_size > 8) | 
 | 	{ | 
 | 	  warn (_("Invalid segment size (%d) in CIE data\n"), fc->segment_size); | 
 | 	  goto fail; | 
 | 	} | 
 |  | 
 |       eh_addr_size = fc->ptr_size; | 
 |     } | 
 |   else | 
 |     { | 
 |       fc->ptr_size = eh_addr_size; | 
 |       fc->segment_size = 0; | 
 |     } | 
 |  | 
 |   READ_ULEB (fc->code_factor, start, end); | 
 |   READ_SLEB (fc->data_factor, start, end); | 
 |  | 
 |   if (start >= end) | 
 |     goto fail; | 
 |  | 
 |   if (version == 1) | 
 |     { | 
 |       GET (fc->ra, 1); | 
 |     } | 
 |   else | 
 |     { | 
 |       READ_ULEB (fc->ra, start, end); | 
 |     } | 
 |  | 
 |   if (fc->augmentation[0] == 'z') | 
 |     { | 
 |       if (start >= end) | 
 | 	goto fail; | 
 |       READ_ULEB (augmentation_data_len, start, end); | 
 |       augmentation_data = start; | 
 |       /* PR 17512: file: 11042-2589-0.004.  */ | 
 |       if (augmentation_data_len > (size_t) (end - start)) | 
 | 	{ | 
 | 	  warn (_("Augmentation data too long: %#" PRIx64 | 
 | 		  ", expected at most %#tx\n"), | 
 | 		augmentation_data_len, end - start); | 
 | 	  goto fail; | 
 | 	} | 
 |       start += augmentation_data_len; | 
 |     } | 
 |  | 
 |   if (augmentation_data_len) | 
 |     { | 
 |       unsigned char *p; | 
 |       unsigned char *q; | 
 |       unsigned char *qend; | 
 |  | 
 |       p = (unsigned char *) fc->augmentation + 1; | 
 |       q = augmentation_data; | 
 |       qend = q + augmentation_data_len; | 
 |  | 
 |       while (p < end && q < qend) | 
 | 	{ | 
 | 	  if (*p == 'L') | 
 | 	    q++; | 
 | 	  else if (*p == 'P') | 
 | 	    q += 1 + size_of_encoded_value (*q); | 
 | 	  else if (*p == 'R') | 
 | 	    fc->fde_encoding = *q++; | 
 | 	  else if (*p == 'S') | 
 | 	    ; | 
 | 	  else if (is_mach_augmentation (*p)) | 
 | 	    ; | 
 | 	  else | 
 | 	    break; | 
 | 	  p++; | 
 | 	} | 
 |       /* Note - it is OK if this loop terminates with q < qend. | 
 | 	 Padding may have been inserted to align the end of the CIE.  */ | 
 |     } | 
 |  | 
 |   *p_cie = fc; | 
 |   if (p_version) | 
 |     *p_version = version; | 
 |   if (p_aug_len) | 
 |     { | 
 |       *p_aug_len = augmentation_data_len; | 
 |       *p_aug = augmentation_data; | 
 |     } | 
 |   return start; | 
 |  | 
 |  fail: | 
 |   free (fc->col_offset); | 
 |   free (fc->col_type); | 
 |   free (fc); | 
 |   return end; | 
 | } | 
 |  | 
 | /* Prints out the contents on the DATA array formatted as unsigned bytes. | 
 |    If do_wide is not enabled, then formats the output to fit into 80 columns. | 
 |    PRINTED contains the number of characters already written to the current | 
 |    output line.  */ | 
 |  | 
 | static void | 
 | display_data (size_t printed, const unsigned char *data, size_t len) | 
 | { | 
 |   if (do_wide || len < ((80 - printed) / 3)) | 
 |     for (printed = 0; printed < len; ++printed) | 
 |       printf (" %02x", data[printed]); | 
 |   else | 
 |     { | 
 |       for (printed = 0; printed < len; ++printed) | 
 | 	{ | 
 | 	  if (printed % (80 / 3) == 0) | 
 | 	    putchar ('\n'); | 
 | 	  printf (" %02x", data[printed]); | 
 | 	} | 
 |     } | 
 | } | 
 |  | 
 | /* Prints out the contents on the augmentation data array. | 
 |    If do_wide is not enabled, then formats the output to fit into 80 columns.  */ | 
 |  | 
 | static void | 
 | display_augmentation_data (const unsigned char * data, uint64_t len) | 
 | { | 
 |   size_t i; | 
 |  | 
 |   i = printf (_("  Augmentation data:    ")); | 
 |   display_data (i, data, len); | 
 | } | 
 |  | 
 | static const char * | 
 | decode_eh_encoding (unsigned int value) | 
 | { | 
 |   if (value == DW_EH_PE_omit) | 
 |     return "omit"; | 
 |  | 
 |   char * format; | 
 |   switch (value & 0x0f) | 
 |     { | 
 |     case DW_EH_PE_uleb128: format = "uleb128"; break; | 
 |     case DW_EH_PE_udata2:  format = "udata2"; break; | 
 |     case DW_EH_PE_udata4:  format = "udata4"; break; | 
 |     case DW_EH_PE_udata8:  format = "udata8"; break; | 
 |     case DW_EH_PE_sleb128: format = "sleb128"; break; | 
 |     case DW_EH_PE_sdata2:  format = "sdata2"; break; | 
 |     case DW_EH_PE_sdata4:  format = "sdata4"; break; | 
 |     case DW_EH_PE_sdata8:  format = "sdata8"; break; | 
 |  | 
 |     default: format = "<unknown format>"; break; /* FIXME: Generate a warning ?  */ | 
 |     } | 
 |  | 
 |   char * application; | 
 |   switch (value & 0xf0) | 
 |     { | 
 |     case DW_EH_PE_absptr:   application = "absolute"; break; | 
 |     case DW_EH_PE_pcrel:    application = "pcrel"; break; | 
 |     case DW_EH_PE_textrel:  application = "textrel"; break; /* FIXME: Is this allowed ?  */ | 
 |     case DW_EH_PE_datarel:  application = "datarel"; break; | 
 |     case DW_EH_PE_funcrel:  application = "funcrel"; break; /* FIXME: Is this allowed ?  */ | 
 |     case DW_EH_PE_aligned:  application = "aligned"; break; /* FIXME: Is this allowed ?  */ | 
 |     case DW_EH_PE_indirect: application = "indirect"; break; /* FIXME: Is this allowed ?  */ | 
 |  | 
 |     default: application = "<unknown application method>"; break;  /* FIXME: Generate a warning ?  */ | 
 |     } | 
 |  | 
 |   static char buffer[128]; | 
 |   sprintf (buffer, "%s, %s", format, application); | 
 |   return buffer; | 
 | } | 
 |  | 
 | /* Reads a value stored at START encoded according to ENCODING. | 
 |    Does not read from, or past, END. | 
 |    Upon success, returns the read value and sets * RETURN_LEN to | 
 |    the number of bytes read. | 
 |    Upon failure returns zero and sets * RETURN_LEN to 0. | 
 |  | 
 |    Note: does not perform any application transformations to the value.  */ | 
 |  | 
 | static uint64_t | 
 | get_encoded_eh_value (unsigned int     encoding, | 
 | 		      unsigned char *  start, | 
 | 		      unsigned char *  end, | 
 | 		      unsigned int *   return_len) | 
 | { | 
 |   uint64_t val; | 
 |   unsigned int len; | 
 |   int status; | 
 |   unsigned char * old_start; | 
 |  | 
 |   switch (encoding & 0x0f) | 
 |     { | 
 |     case DW_EH_PE_uleb128: | 
 |       val = read_leb128 (start, end, false, & len, & status); | 
 |       if (status != 0) | 
 | 	len = 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_sleb128: | 
 |       val = read_leb128 (start, end, true, & len, & status); | 
 |       if (status != 0) | 
 | 	len = 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_udata2: | 
 |       old_start = start; | 
 |       SAFE_BYTE_GET_AND_INC (val, start, 2, end); | 
 |       len = start - old_start == 2 ? 2 : 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_udata4: | 
 |       old_start = start; | 
 |       SAFE_BYTE_GET_AND_INC (val, start, 4, end); | 
 |       len = start - old_start == 4 ? 4 : 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_udata8: | 
 |       old_start = start; | 
 |       SAFE_BYTE_GET_AND_INC (val, start, 8, end); | 
 |       len = start - old_start == 8 ? 8 : 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_sdata2: | 
 |       old_start = start; | 
 |       SAFE_SIGNED_BYTE_GET_AND_INC (val, start, 2, end); | 
 |       len = start - old_start == 2 ? 2 : 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_sdata4: | 
 |       old_start = start; | 
 |       SAFE_SIGNED_BYTE_GET_AND_INC (val, start, 4, end); | 
 |       len = start - old_start == 4 ? 4 : 0; | 
 |       break; | 
 |  | 
 |     case DW_EH_PE_sdata8: | 
 |       old_start = start; | 
 |       SAFE_SIGNED_BYTE_GET_AND_INC (val, start, 8, end); | 
 |       len = start - old_start == 8 ? 8 : 0; | 
 |       break; | 
 |  | 
 |     default: | 
 |       goto fail; | 
 |     } | 
 |  | 
 |   * return_len = len; | 
 |   return val; | 
 |  | 
 |  fail: | 
 |   * return_len = 0; | 
 |   return 0; | 
 | } | 
 |  | 
 | static uint64_t | 
 | encoded_eh_offset (unsigned int            encoding, | 
 | 		   struct dwarf_section *  section, | 
 | 		   uint64_t                section_offset, | 
 | 		   uint64_t                value) | 
 | { | 
 |   switch (encoding & 0xf0) | 
 |     { | 
 |     default: | 
 |       /* This should not happen.  FIXME: warn ?  */ | 
 |     case DW_EH_PE_absptr: | 
 |       return value; | 
 |  | 
 |     case DW_EH_PE_pcrel: | 
 |       return value + (uint64_t)(section->address + section_offset); | 
 |  | 
 |     case DW_EH_PE_datarel: | 
 |       return value + (uint64_t)section->address; | 
 |     } | 
 | } | 
 |  | 
 | static int | 
 | display_eh_frame_hdr (struct dwarf_section *section, | 
 | 		      void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   if (section->size < 6) | 
 |     { | 
 |       warn (_(".eh_frame_hdr section is too small\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   unsigned int version = start[0]; | 
 |   if (version != 1) | 
 |     { | 
 |       warn (_("Unsupported .eh_frame_hdr version %u\n"), version); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   printf (_("  Version:                 %u\n"), version); | 
 |  | 
 |   unsigned int ptr_enc = start[1]; | 
 |   /* Strictly speaking this is the encoding format of the eh_frame_ptr field below.  */ | 
 |   printf (_("  Pointer Encoding Format: %#x (%s)\n"), ptr_enc, decode_eh_encoding (ptr_enc)); | 
 |  | 
 |   unsigned int count_enc = start[2]; | 
 |   printf (_("  Count Encoding Format:   %#x (%s)\n"), count_enc, decode_eh_encoding (count_enc)); | 
 |  | 
 |   unsigned int table_enc = start[3]; | 
 |   printf (_("  Table Encoding Format:   %#x (%s)\n"), table_enc, decode_eh_encoding (table_enc)); | 
 |  | 
 |   start += 4; | 
 |  | 
 |   unsigned int len; | 
 |  | 
 |   uint64_t eh_frame_ptr = get_encoded_eh_value (ptr_enc, start, end, & len); | 
 |   if (len == 0) | 
 |     { | 
 |       warn (_("unable to read eh_frame_ptr field in .eh_frame_hdr section\n")); | 
 |       return 0; | 
 |     } | 
 |   printf (_("  Start of frame section:  %#" PRIx64), eh_frame_ptr); | 
 |  | 
 |   uint64_t offset_eh_frame_ptr = encoded_eh_offset (ptr_enc, section, 4, eh_frame_ptr); | 
 |   if (offset_eh_frame_ptr != eh_frame_ptr) | 
 |     printf (_(" (offset: %#" PRIx64 ")"), offset_eh_frame_ptr); | 
 |  | 
 |   printf ("\n"); | 
 |   start += len; | 
 |  | 
 |   if (count_enc == DW_EH_PE_omit) | 
 |     { | 
 |       warn (_("It is suspicious to have a .eh_frame_hdr section with an empty search table\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   if (count_enc & 0xf0) | 
 |     { | 
 |       warn (_("The count field format should be absolute, not relative to an address\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   uint64_t fde_count = get_encoded_eh_value (count_enc, start, end, & len); | 
 |   if (len == 0) | 
 |     { | 
 |       warn (_("unable to read fde_count field in .eh_frame_hdr section\n")); | 
 |       return 0; | 
 |     } | 
 |   printf (_("  Entries in search table: %#" PRIx64), fde_count); | 
 |   printf ("\n"); | 
 |   start += len; | 
 |  | 
 |   if (fde_count != 0 && table_enc == DW_EH_PE_omit) | 
 |     { | 
 |       warn (_("It is suspicious to have a .eh_frame_hdr section an empty table but a non empty count field\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   uint64_t i; | 
 |   /* Read and display the search table.  */ | 
 |   for (i = 0; i < fde_count; i++) | 
 |     { | 
 |       uint64_t location, address; | 
 |       unsigned char * row_start = start; | 
 |  | 
 |       location = get_encoded_eh_value (table_enc, start, end, & len); | 
 |       if (len == 0) | 
 | 	{ | 
 | 	  warn (_("Failed to read location field for entry %#" PRIx64 " in the .eh_frame_hdr's search table\n"), i); | 
 | 	  return 0; | 
 | 	} | 
 |       start += len; | 
 |  | 
 |       address = get_encoded_eh_value (table_enc, start, end, & len); | 
 |       if (len == 0) | 
 | 	{ | 
 | 	  warn (_("Failed to read address field for entry %#" PRIx64 " in the .eh_frame_hdr's search table\n"), i); | 
 | 	  return 0; | 
 | 	} | 
 |       start += len; | 
 |  | 
 |       /* This format is intended to be compatible with the output of eu-readelf's -e option.  */ | 
 |       printf ("  %#" PRIx64 " (offset: %#" PRIx64 ") -> %#" PRIx64 " fde=[ %5" PRIx64 "]\n", | 
 | 	      location, | 
 | 	      encoded_eh_offset (table_enc, section, row_start - section->start, location), | 
 | 	      address, | 
 | 	      encoded_eh_offset (table_enc, section, row_start - section->start, address) - offset_eh_frame_ptr); | 
 |     } | 
 |  | 
 |   printf ("\n"); | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_frames (struct dwarf_section *section, | 
 | 		      void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   unsigned char *end = start + section->size; | 
 |   unsigned char *section_start = start; | 
 |   Frame_Chunk *chunks = NULL, *forward_refs = NULL; | 
 |   Frame_Chunk *remembered_state = NULL; | 
 |   Frame_Chunk *rs; | 
 |   bool is_eh = strcmp (section->name, ".eh_frame") == 0; | 
 |   unsigned int max_regs = 0; | 
 |   const char *bad_reg = _("bad register: "); | 
 |   unsigned int saved_eh_addr_size = eh_addr_size; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   while (start < end) | 
 |     { | 
 |       unsigned char *saved_start; | 
 |       unsigned char *block_end; | 
 |       uint64_t length; | 
 |       uint64_t cie_id; | 
 |       Frame_Chunk *fc; | 
 |       Frame_Chunk *cie; | 
 |       int need_col_headers = 1; | 
 |       unsigned char *augmentation_data = NULL; | 
 |       uint64_t augmentation_data_len = 0; | 
 |       unsigned int encoded_ptr_size = saved_eh_addr_size; | 
 |       unsigned int offset_size; | 
 |       bool all_nops; | 
 |       static Frame_Chunk fde_fc; | 
 |  | 
 |       saved_start = start; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (length, start, 4, end); | 
 |  | 
 |       if (length == 0) | 
 | 	{ | 
 | 	  printf ("\n%08tx ZERO terminator\n\n", | 
 | 		  saved_start - section_start); | 
 | 	  /* Skip any zero terminators that directly follow. | 
 | 	     A corrupt section size could have loaded a whole | 
 | 	     slew of zero filled memory bytes.  eg | 
 | 	     PR 17512: file: 070-19381-0.004.  */ | 
 | 	  while (start < end && * start == 0) | 
 | 	    ++ start; | 
 | 	  continue; | 
 | 	} | 
 |  | 
 |       if (length == 0xffffffff) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET_AND_INC (length, start, 8, end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |  | 
 |       if (length > (size_t) (end - start)) | 
 | 	{ | 
 | 	  warn ("Invalid length %#" PRIx64 " in FDE at %#tx\n", | 
 | 		length, saved_start - section_start); | 
 | 	  block_end = end; | 
 | 	} | 
 |       else | 
 | 	block_end = start + length; | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (cie_id, start, offset_size, block_end); | 
 |  | 
 |       if (is_eh ? (cie_id == 0) : ((offset_size == 4 && cie_id == DW_CIE_ID) | 
 | 				   || (offset_size == 8 && cie_id == DW64_CIE_ID))) | 
 | 	{ | 
 | 	  int version; | 
 | 	  unsigned int mreg; | 
 |  | 
 | 	  start = read_cie (start, block_end, &cie, &version, | 
 | 			    &augmentation_data_len, &augmentation_data); | 
 | 	  /* PR 17512: file: 027-135133-0.005.  */ | 
 | 	  if (cie == NULL) | 
 | 	    break; | 
 |  | 
 | 	  fc = cie; | 
 | 	  fc->next = chunks; | 
 | 	  chunks = fc; | 
 | 	  fc->chunk_start = saved_start; | 
 | 	  mreg = max_regs > 0 ? max_regs - 1 : 0; | 
 | 	  if (mreg < fc->ra) | 
 | 	    mreg = fc->ra; | 
 | 	  if (frame_need_space (fc, mreg) < 0) | 
 | 	    break; | 
 | 	  if (fc->fde_encoding) | 
 | 	    encoded_ptr_size = size_of_encoded_value (fc->fde_encoding); | 
 |  | 
 | 	  printf ("\n%08tx ", saved_start - section_start); | 
 | 	  print_hex (length, fc->ptr_size); | 
 | 	  print_hex (cie_id, offset_size); | 
 |  | 
 | 	  if (do_debug_frames_interp) | 
 | 	    { | 
 | 	      printf ("CIE \"%s\" cf=%d df=%d ra=%d\n", fc->augmentation, | 
 | 		      fc->code_factor, fc->data_factor, fc->ra); | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      printf ("CIE\n"); | 
 | 	      printf ("  Version:               %d\n", version); | 
 | 	      printf ("  Augmentation:          \"%s\"\n", fc->augmentation); | 
 | 	      if (version >= 4) | 
 | 		{ | 
 | 		  printf ("  Pointer Size:          %u\n", fc->ptr_size); | 
 | 		  printf ("  Segment Size:          %u\n", fc->segment_size); | 
 | 		} | 
 | 	      printf ("  Code alignment factor: %u\n", fc->code_factor); | 
 | 	      printf ("  Data alignment factor: %d\n", fc->data_factor); | 
 | 	      printf ("  Return address column: %d\n", fc->ra); | 
 |  | 
 | 	      if (augmentation_data_len) | 
 | 		display_augmentation_data (augmentation_data, augmentation_data_len); | 
 |  | 
 | 	      putchar ('\n'); | 
 | 	    } | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  unsigned char *look_for; | 
 | 	  unsigned long segment_selector; | 
 | 	  uint64_t cie_off; | 
 |  | 
 | 	  cie_off = cie_id; | 
 | 	  if (is_eh) | 
 | 	    { | 
 | 	      uint64_t sign = (uint64_t) 1 << (offset_size * 8 - 1); | 
 | 	      cie_off = (cie_off ^ sign) - sign; | 
 | 	      cie_off = start - 4 - section_start - cie_off; | 
 | 	    } | 
 |  | 
 | 	  look_for = section_start + cie_off; | 
 | 	  if (cie_off <= (size_t) (saved_start - section_start)) | 
 | 	    { | 
 | 	      for (cie = chunks; cie ; cie = cie->next) | 
 | 		if (cie->chunk_start == look_for) | 
 | 		  break; | 
 | 	    } | 
 | 	  else if (cie_off >= section->size) | 
 | 	    cie = NULL; | 
 | 	  else | 
 | 	    { | 
 | 	      for (cie = forward_refs; cie ; cie = cie->next) | 
 | 		if (cie->chunk_start == look_for) | 
 | 		  break; | 
 | 	      if (!cie) | 
 | 		{ | 
 | 		  unsigned int off_size; | 
 | 		  unsigned char *cie_scan; | 
 |  | 
 | 		  cie_scan = look_for; | 
 | 		  off_size = 4; | 
 | 		  SAFE_BYTE_GET_AND_INC (length, cie_scan, 4, end); | 
 | 		  if (length == 0xffffffff) | 
 | 		    { | 
 | 		      SAFE_BYTE_GET_AND_INC (length, cie_scan, 8, end); | 
 | 		      off_size = 8; | 
 | 		    } | 
 | 		  if (length != 0 && length <= (size_t) (end - cie_scan)) | 
 | 		    { | 
 | 		      uint64_t c_id; | 
 | 		      unsigned char *cie_end = cie_scan + length; | 
 |  | 
 | 		      SAFE_BYTE_GET_AND_INC (c_id, cie_scan, off_size, | 
 | 					     cie_end); | 
 | 		      if (is_eh | 
 | 			  ? c_id == 0 | 
 | 			  : ((off_size == 4 && c_id == DW_CIE_ID) | 
 | 			     || (off_size == 8 && c_id == DW64_CIE_ID))) | 
 | 			{ | 
 | 			  int version; | 
 | 			  unsigned int mreg; | 
 |  | 
 | 			  read_cie (cie_scan, cie_end, &cie, &version, | 
 | 				    &augmentation_data_len, &augmentation_data); | 
 | 			  /* PR 17512: file: 3450-2098-0.004.  */ | 
 | 			  if (cie == NULL) | 
 | 			    { | 
 | 			      warn (_("Failed to read CIE information\n")); | 
 | 			      break; | 
 | 			    } | 
 | 			  cie->next = forward_refs; | 
 | 			  forward_refs = cie; | 
 | 			  cie->chunk_start = look_for; | 
 | 			  mreg = max_regs > 0 ? max_regs - 1 : 0; | 
 | 			  if (mreg < cie->ra) | 
 | 			    mreg = cie->ra; | 
 | 			  if (frame_need_space (cie, mreg) < 0) | 
 | 			    { | 
 | 			      warn (_("Invalid max register\n")); | 
 | 			      break; | 
 | 			    } | 
 | 			  if (cie->fde_encoding) | 
 | 			    encoded_ptr_size | 
 | 			      = size_of_encoded_value (cie->fde_encoding); | 
 | 			} | 
 | 		    } | 
 | 		} | 
 | 	    } | 
 |  | 
 | 	  fc = &fde_fc; | 
 | 	  memset (fc, 0, sizeof (Frame_Chunk)); | 
 |  | 
 | 	  if (!cie) | 
 | 	    { | 
 | 	      fc->ncols = 0; | 
 | 	      fc->col_type = xmalloc (sizeof (*fc->col_type)); | 
 | 	      fc->col_offset = xmalloc (sizeof (*fc->col_offset)); | 
 | 	      if (frame_need_space (fc, max_regs > 0 ? max_regs - 1 : 0) < 0) | 
 | 		{ | 
 | 		  warn (_("Invalid max register\n")); | 
 | 		  break; | 
 | 		} | 
 | 	      cie = fc; | 
 | 	      fc->augmentation = ""; | 
 | 	      fc->fde_encoding = 0; | 
 | 	      fc->ptr_size = eh_addr_size; | 
 | 	      fc->segment_size = 0; | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      fc->ncols = cie->ncols; | 
 | 	      fc->col_type = xcmalloc (fc->ncols, sizeof (*fc->col_type)); | 
 | 	      fc->col_offset =  xcmalloc (fc->ncols, sizeof (*fc->col_offset)); | 
 | 	      memcpy (fc->col_type, cie->col_type, | 
 | 		      fc->ncols * sizeof (*fc->col_type)); | 
 | 	      memcpy (fc->col_offset, cie->col_offset, | 
 | 		      fc->ncols * sizeof (*fc->col_offset)); | 
 | 	      fc->augmentation = cie->augmentation; | 
 | 	      fc->ptr_size = cie->ptr_size; | 
 | 	      eh_addr_size = cie->ptr_size; | 
 | 	      fc->segment_size = cie->segment_size; | 
 | 	      fc->code_factor = cie->code_factor; | 
 | 	      fc->data_factor = cie->data_factor; | 
 | 	      fc->cfa_reg = cie->cfa_reg; | 
 | 	      fc->cfa_offset = cie->cfa_offset; | 
 | 	      fc->cfa_ofs_signed_p = cie->cfa_ofs_signed_p; | 
 | 	      fc->ra = cie->ra; | 
 | 	      if (frame_need_space (fc, max_regs > 0 ? max_regs - 1: 0) < 0) | 
 | 		{ | 
 | 		  warn (_("Invalid max register\n")); | 
 | 		  break; | 
 | 		} | 
 | 	      fc->fde_encoding = cie->fde_encoding; | 
 | 	    } | 
 |  | 
 | 	  if (fc->fde_encoding) | 
 | 	    encoded_ptr_size = size_of_encoded_value (fc->fde_encoding); | 
 |  | 
 | 	  segment_selector = 0; | 
 | 	  if (fc->segment_size) | 
 | 	    { | 
 | 	      if (fc->segment_size > sizeof (segment_selector)) | 
 | 		{ | 
 | 		  /* PR 17512: file: 9e196b3e.  */ | 
 | 		  warn (_("Probably corrupt segment size: %d - using 4 instead\n"), fc->segment_size); | 
 | 		  fc->segment_size = 4; | 
 | 		} | 
 | 	      SAFE_BYTE_GET_AND_INC (segment_selector, start, | 
 | 				     fc->segment_size, block_end); | 
 | 	    } | 
 |  | 
 | 	  fc->pc_begin = get_encoded_value (&start, fc->fde_encoding, section, | 
 | 					    block_end); | 
 |  | 
 | 	  /* FIXME: It appears that sometimes the final pc_range value is | 
 | 	     encoded in less than encoded_ptr_size bytes.  See the x86_64 | 
 | 	     run of the "objcopy on compressed debug sections" test for an | 
 | 	     example of this.  */ | 
 | 	  SAFE_BYTE_GET_AND_INC (fc->pc_range, start, encoded_ptr_size, | 
 | 				 block_end); | 
 |  | 
 | 	  if (cie->augmentation[0] == 'z') | 
 | 	    { | 
 | 	      READ_ULEB (augmentation_data_len, start, block_end); | 
 | 	      augmentation_data = start; | 
 | 	      /* PR 17512 file: 722-8446-0.004 and PR 22386.  */ | 
 | 	      if (augmentation_data_len > (size_t) (block_end - start)) | 
 | 		{ | 
 | 		  warn (_("Augmentation data too long: %#" PRIx64 ", " | 
 | 			  "expected at most %#tx\n"), | 
 | 			augmentation_data_len, block_end - start); | 
 | 		  start = block_end; | 
 | 		  augmentation_data = NULL; | 
 | 		  augmentation_data_len = 0; | 
 | 		} | 
 | 	      start += augmentation_data_len; | 
 | 	    } | 
 |  | 
 | 	  printf ("\n%08tx ", saved_start - section_start); | 
 | 	  print_hex (length, fc->ptr_size); | 
 | 	  print_hex (cie_id, offset_size); | 
 | 	  printf ("FDE "); | 
 |  | 
 | 	  if (cie->chunk_start) | 
 | 	    printf ("cie=%08tx", cie->chunk_start - section_start); | 
 | 	  else | 
 | 	    /* Ideally translate "invalid " to 8 chars, trailing space | 
 | 	       is optional.  */ | 
 | 	    printf (_("cie=invalid ")); | 
 |  | 
 | 	  printf (" pc="); | 
 | 	  if (fc->segment_size) | 
 | 	    printf ("%04lx:", segment_selector); | 
 |  | 
 | 	  print_hex_ns (fc->pc_begin, fc->ptr_size); | 
 | 	  printf (".."); | 
 | 	  print_hex_ns (fc->pc_begin + fc->pc_range, fc->ptr_size); | 
 | 	  printf ("\n"); | 
 |  | 
 | 	  if (! do_debug_frames_interp && augmentation_data_len) | 
 | 	    { | 
 | 	      display_augmentation_data (augmentation_data, augmentation_data_len); | 
 | 	      putchar ('\n'); | 
 | 	    } | 
 | 	} | 
 |  | 
 |       /* At this point, fc is the current chunk, cie (if any) is set, and | 
 | 	 we're about to interpret instructions for the chunk.  */ | 
 |       /* ??? At present we need to do this always, since this sizes the | 
 | 	 fc->col_type and fc->col_offset arrays, which we write into always. | 
 | 	 We should probably split the interpreted and non-interpreted bits | 
 | 	 into two different routines, since there's so much that doesn't | 
 | 	 really overlap between them.  */ | 
 |       if (1 || do_debug_frames_interp) | 
 | 	{ | 
 | 	  /* Start by making a pass over the chunk, allocating storage | 
 | 	     and taking note of what registers are used.  */ | 
 | 	  unsigned char *tmp = start; | 
 |  | 
 | 	  while (start < block_end) | 
 | 	    { | 
 | 	      unsigned int reg, op, opa; | 
 | 	      unsigned long temp; | 
 |  | 
 | 	      op = *start++; | 
 | 	      opa = op & 0x3f; | 
 | 	      if (op & 0xc0) | 
 | 		op &= 0xc0; | 
 |  | 
 | 	      /* Warning: if you add any more cases to this switch, be | 
 | 		 sure to add them to the corresponding switch below.  */ | 
 | 	      reg = -1u; | 
 | 	      switch (op) | 
 | 		{ | 
 | 		case DW_CFA_advance_loc: | 
 | 		  break; | 
 | 		case DW_CFA_offset: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  reg = opa; | 
 | 		  break; | 
 | 		case DW_CFA_restore: | 
 | 		  reg = opa; | 
 | 		  break; | 
 | 		case DW_CFA_set_loc: | 
 | 		  if ((size_t) (block_end - start) < encoded_ptr_size) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += encoded_ptr_size; | 
 | 		  break; | 
 | 		case DW_CFA_advance_loc1: | 
 | 		  if ((size_t) (block_end - start) < 1) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += 1; | 
 | 		  break; | 
 | 		case DW_CFA_advance_loc2: | 
 | 		  if ((size_t) (block_end - start) < 2) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += 2; | 
 | 		  break; | 
 | 		case DW_CFA_advance_loc4: | 
 | 		  if ((size_t) (block_end - start) < 4) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += 4; | 
 | 		  break; | 
 | 		case DW_CFA_offset_extended: | 
 | 		case DW_CFA_val_offset: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_restore_extended: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_undefined: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_same_value: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_register: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa_register: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa_offset: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa_expression: | 
 | 		  READ_ULEB (temp, start, block_end); | 
 | 		  if ((size_t) (block_end - start) < temp) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += temp; | 
 | 		  break; | 
 | 		case DW_CFA_expression: | 
 | 		case DW_CFA_val_expression: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  READ_ULEB (temp, start, block_end); | 
 | 		  if ((size_t) (block_end - start) < temp) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += temp; | 
 | 		  break; | 
 | 		case DW_CFA_offset_extended_sf: | 
 | 		case DW_CFA_val_offset_sf: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  SKIP_SLEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa_sf: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  SKIP_SLEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_def_cfa_offset_sf: | 
 | 		  SKIP_SLEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_MIPS_advance_loc8: | 
 | 		  if ((size_t) (block_end - start) < 8) | 
 | 		    start = block_end; | 
 | 		  else | 
 | 		    start += 8; | 
 | 		  break; | 
 | 		case DW_CFA_GNU_args_size: | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		case DW_CFA_GNU_negative_offset_extended: | 
 | 		  READ_ULEB (reg, start, block_end); | 
 | 		  SKIP_ULEB (start, block_end); | 
 | 		  break; | 
 | 		default: | 
 | 		  break; | 
 | 		} | 
 | 	      if (reg != -1u && frame_need_space (fc, reg) >= 0) | 
 | 		{ | 
 | 		  /* Don't leave any reg as DW_CFA_unreferenced so | 
 | 		     that frame_display_row prints name of regs in | 
 | 		     header, and all referenced regs in each line.  */ | 
 | 		  if (reg >= cie->ncols | 
 | 		      || cie->col_type[reg] == DW_CFA_unreferenced) | 
 | 		    fc->col_type[reg] = DW_CFA_undefined; | 
 | 		  else | 
 | 		    fc->col_type[reg] = cie->col_type[reg]; | 
 | 		} | 
 | 	    } | 
 | 	  start = tmp; | 
 | 	} | 
 |  | 
 |       all_nops = true; | 
 |  | 
 |       /* Now we know what registers are used, make a second pass over | 
 | 	 the chunk, this time actually printing out the info.  */ | 
 |  | 
 |       while (start < block_end) | 
 | 	{ | 
 | 	  unsigned op, opa; | 
 | 	  /* Note: It is tempting to use an unsigned long for 'reg' but there | 
 | 	     are various functions, notably frame_space_needed() that assume that | 
 | 	     reg is an unsigned int.  */ | 
 | 	  unsigned int reg; | 
 | 	  int64_t sofs; | 
 | 	  uint64_t ofs; | 
 | 	  const char *reg_prefix = ""; | 
 |  | 
 | 	  op = *start++; | 
 | 	  opa = op & 0x3f; | 
 | 	  if (op & 0xc0) | 
 | 	    op &= 0xc0; | 
 |  | 
 | 	  /* Make a note if something other than DW_CFA_nop happens.  */ | 
 | 	  if (op != DW_CFA_nop) | 
 | 	    all_nops = false; | 
 |  | 
 | 	  /* Warning: if you add any more cases to this switch, be | 
 | 	     sure to add them to the corresponding switch above.  */ | 
 | 	  switch (op) | 
 | 	    { | 
 | 	    case DW_CFA_advance_loc: | 
 | 	      opa *= fc->code_factor; | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_advance_loc: %d to ", opa); | 
 | 		  print_hex_ns (fc->pc_begin + opa, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin += opa; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_offset: | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      ofs *= fc->data_factor; | 
 | 	      if (opa >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_offset: %s%s at cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (opa, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[opa] = DW_CFA_offset; | 
 | 		  fc->col_offset[opa] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_restore: | 
 | 	      if (opa >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_restore: %s%s\n", | 
 | 			reg_prefix, regname (opa, 0)); | 
 | 	      if (*reg_prefix != '\0') | 
 | 		break; | 
 |  | 
 | 	      if (opa >= cie->ncols | 
 | 		  || cie->col_type[opa] == DW_CFA_unreferenced) | 
 | 		{ | 
 | 		  fc->col_type[opa] = DW_CFA_undefined; | 
 | 		  fc->col_offset[opa] = 0; | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  fc->col_type[opa] = cie->col_type[opa]; | 
 | 		  fc->col_offset[opa] = cie->col_offset[opa]; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_set_loc: | 
 | 	      ofs = get_encoded_value (&start, fc->fde_encoding, section, | 
 | 				       block_end); | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_set_loc: "); | 
 | 		  print_hex_ns (ofs, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin = ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_advance_loc1: | 
 | 	      SAFE_BYTE_GET_AND_INC (ofs, start, 1, block_end); | 
 | 	      ofs *= fc->code_factor; | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_advance_loc1: %" PRId64 " to ", ofs); | 
 | 		  print_hex_ns (fc->pc_begin + ofs, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_advance_loc2: | 
 | 	      SAFE_BYTE_GET_AND_INC (ofs, start, 2, block_end); | 
 | 	      ofs *= fc->code_factor; | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_advance_loc2: %" PRId64 " to ", ofs); | 
 | 		  print_hex_ns (fc->pc_begin + ofs, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_advance_loc4: | 
 | 	      SAFE_BYTE_GET_AND_INC (ofs, start, 4, block_end); | 
 | 	      ofs *= fc->code_factor; | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_advance_loc4: %" PRId64 " to ", ofs); | 
 | 		  print_hex_ns (fc->pc_begin + ofs, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_offset_extended: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      ofs *= fc->data_factor; | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_offset_extended: %s%s at cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (reg, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_offset; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_val_offset: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      ofs *= fc->data_factor; | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_val_offset: %s%s is cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (reg, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_val_offset; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_restore_extended: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_restore_extended: %s%s\n", | 
 | 			reg_prefix, regname (reg, 0)); | 
 | 	      if (*reg_prefix != '\0') | 
 | 		break; | 
 |  | 
 | 	      if (reg >= cie->ncols | 
 | 		  || cie->col_type[reg] == DW_CFA_unreferenced) | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_undefined; | 
 | 		  fc->col_offset[reg] = 0; | 
 | 		} | 
 | 	      else | 
 | 		{ | 
 | 		  fc->col_type[reg] = cie->col_type[reg]; | 
 | 		  fc->col_offset[reg] = cie->col_offset[reg]; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_undefined: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_undefined: %s%s\n", | 
 | 			reg_prefix, regname (reg, 0)); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_undefined; | 
 | 		  fc->col_offset[reg] = 0; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_same_value: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_same_value: %s%s\n", | 
 | 			reg_prefix, regname (reg, 0)); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_same_value; | 
 | 		  fc->col_offset[reg] = 0; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_register: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		{ | 
 | 		  printf ("  DW_CFA_register: %s%s in ", | 
 | 			  reg_prefix, regname (reg, 0)); | 
 | 		  puts (regname (ofs, 0)); | 
 | 		} | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_register; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_remember_state: | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_remember_state\n"); | 
 | 	      rs = (Frame_Chunk *) xmalloc (sizeof (Frame_Chunk)); | 
 | 	      rs->cfa_offset = fc->cfa_offset; | 
 | 	      rs->cfa_ofs_signed_p = fc->cfa_ofs_signed_p; | 
 | 	      rs->cfa_reg = fc->cfa_reg; | 
 | 	      rs->ra = fc->ra; | 
 | 	      rs->cfa_exp = fc->cfa_exp; | 
 | 	      rs->ncols = fc->ncols; | 
 | 	      rs->col_type = xcmalloc (rs->ncols, sizeof (*rs->col_type)); | 
 | 	      rs->col_offset = xcmalloc (rs->ncols, sizeof (*rs->col_offset)); | 
 | 	      memcpy (rs->col_type, fc->col_type, | 
 | 		      rs->ncols * sizeof (*fc->col_type)); | 
 | 	      memcpy (rs->col_offset, fc->col_offset, | 
 | 		      rs->ncols * sizeof (*fc->col_offset)); | 
 | 	      rs->next = remembered_state; | 
 | 	      remembered_state = rs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_restore_state: | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_restore_state\n"); | 
 | 	      rs = remembered_state; | 
 | 	      if (rs) | 
 | 		{ | 
 | 		  remembered_state = rs->next; | 
 | 		  fc->cfa_offset = rs->cfa_offset; | 
 | 		  fc->cfa_ofs_signed_p = rs->cfa_ofs_signed_p; | 
 | 		  fc->cfa_reg = rs->cfa_reg; | 
 | 		  fc->ra = rs->ra; | 
 | 		  fc->cfa_exp = rs->cfa_exp; | 
 | 		  if (frame_need_space (fc, rs->ncols - 1) < 0) | 
 | 		    { | 
 | 		      warn (_("Invalid column number in saved frame state\n")); | 
 | 		      fc->ncols = 0; | 
 | 		    } | 
 | 		  else | 
 | 		    { | 
 | 		      memcpy (fc->col_type, rs->col_type, | 
 | 			      rs->ncols * sizeof (*rs->col_type)); | 
 | 		      memcpy (fc->col_offset, rs->col_offset, | 
 | 			      rs->ncols * sizeof (*rs->col_offset)); | 
 | 		    } | 
 | 		  free (rs->col_type); | 
 | 		  free (rs->col_offset); | 
 | 		  free (rs); | 
 | 		} | 
 | 	      else if (do_debug_frames_interp) | 
 | 		printf ("Mismatched DW_CFA_restore_state\n"); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa: | 
 | 	      READ_ULEB (fc->cfa_reg, start, block_end); | 
 | 	      READ_ULEB (fc->cfa_offset, start, block_end); | 
 | 	      fc->cfa_ofs_signed_p = false; | 
 | 	      fc->cfa_exp = 0; | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_def_cfa: %s ofs %" PRIu64 "\n", | 
 | 			regname (fc->cfa_reg, 0), fc->cfa_offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa_register: | 
 | 	      READ_ULEB (fc->cfa_reg, start, block_end); | 
 | 	      fc->cfa_exp = 0; | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_def_cfa_register: %s\n", | 
 | 			regname (fc->cfa_reg, 0)); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa_offset: | 
 | 	      READ_ULEB (fc->cfa_offset, start, block_end); | 
 | 	      fc->cfa_ofs_signed_p = false; | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_def_cfa_offset: %" PRIu64 "\n", fc->cfa_offset); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_nop: | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_nop\n"); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa_expression: | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      if (ofs > (size_t) (block_end - start)) | 
 | 		{ | 
 | 		  printf (_("  %s: <corrupt len %" PRIu64 ">\n"), | 
 | 			  "DW_CFA_def_cfa_expression", ofs); | 
 | 		  break; | 
 | 		} | 
 | 	      if (! do_debug_frames_interp) | 
 | 		{ | 
 | 		  printf ("  DW_CFA_def_cfa_expression ("); | 
 | 		  decode_location_expression (start, eh_addr_size, 0, -1, | 
 | 					      ofs, 0, section); | 
 | 		  printf (")\n"); | 
 | 		} | 
 | 	      fc->cfa_exp = 1; | 
 | 	      start += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_expression: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      /* PR 17512: file: 069-133014-0.006.  */ | 
 | 	      /* PR 17512: file: 98c02eb4.  */ | 
 | 	      if (ofs > (size_t) (block_end - start)) | 
 | 		{ | 
 | 		  printf (_("  %s: <corrupt len %" PRIu64 ">\n"), | 
 | 			  "DW_CFA_expression", ofs); | 
 | 		  break; | 
 | 		} | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		{ | 
 | 		  printf ("  DW_CFA_expression: %s%s (", | 
 | 			  reg_prefix, regname (reg, 0)); | 
 | 		  decode_location_expression (start, eh_addr_size, 0, -1, | 
 | 					      ofs, 0, section); | 
 | 		  printf (")\n"); | 
 | 		} | 
 | 	      if (*reg_prefix == '\0') | 
 | 		fc->col_type[reg] = DW_CFA_expression; | 
 | 	      start += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_val_expression: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (ofs > (size_t) (block_end - start)) | 
 | 		{ | 
 | 		  printf ("  %s: <corrupt len %" PRIu64 ">\n", | 
 | 			  "DW_CFA_val_expression", ofs); | 
 | 		  break; | 
 | 		} | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		{ | 
 | 		  printf ("  DW_CFA_val_expression: %s%s (", | 
 | 			  reg_prefix, regname (reg, 0)); | 
 | 		  decode_location_expression (start, eh_addr_size, 0, -1, | 
 | 					      ofs, 0, section); | 
 | 		  printf (")\n"); | 
 | 		} | 
 | 	      if (*reg_prefix == '\0') | 
 | 		fc->col_type[reg] = DW_CFA_val_expression; | 
 | 	      start += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_offset_extended_sf: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_SLEB (sofs, start, block_end); | 
 | 	      /* data_factor multiplicaton done here as unsigned to | 
 | 		 avoid integer overflow warnings from asan on fuzzed | 
 | 		 objects.  */ | 
 | 	      ofs = sofs; | 
 | 	      ofs *= fc->data_factor; | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_offset_extended_sf: %s%s at cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (reg, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_offset; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_val_offset_sf: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_SLEB (sofs, start, block_end); | 
 | 	      ofs = sofs; | 
 | 	      ofs *= fc->data_factor; | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_val_offset_sf: %s%s is cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (reg, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_val_offset; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa_sf: | 
 | 	      READ_ULEB (fc->cfa_reg, start, block_end); | 
 | 	      READ_SLEB (sofs, start, block_end); | 
 | 	      ofs = sofs; | 
 | 	      ofs *= fc->data_factor; | 
 | 	      fc->cfa_offset = ofs; | 
 | 	      fc->cfa_ofs_signed_p = true; | 
 | 	      fc->cfa_exp = 0; | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_def_cfa_sf: %s ofs %" PRId64 "\n", | 
 | 			regname (fc->cfa_reg, 0), ofs); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_def_cfa_offset_sf: | 
 | 	      READ_SLEB (sofs, start, block_end); | 
 | 	      ofs = sofs; | 
 | 	      ofs *= fc->data_factor; | 
 | 	      fc->cfa_offset = ofs; | 
 | 	      fc->cfa_ofs_signed_p = true; | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_def_cfa_offset_sf: %" PRId64 "\n", ofs); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_MIPS_advance_loc8: | 
 | 	      SAFE_BYTE_GET_AND_INC (ofs, start, 8, block_end); | 
 | 	      ofs *= fc->code_factor; | 
 | 	      if (do_debug_frames_interp) | 
 | 		frame_display_row (fc, &need_col_headers, &max_regs); | 
 | 	      else | 
 | 		{ | 
 | 		  printf ("  DW_CFA_MIPS_advance_loc8: %" PRId64 " to ", ofs); | 
 | 		  print_hex_ns (fc->pc_begin + ofs, fc->ptr_size); | 
 | 		  printf ("\n"); | 
 | 		} | 
 | 	      fc->pc_begin += ofs; | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_AARCH64_negate_ra_state_with_pc: | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_AARCH64_negate_ra_state_with_pc\n"); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_GNU_window_save: | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  %s\n", DW_CFA_GNU_window_save_name[is_aarch64]); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_GNU_args_size: | 
 | 	      READ_ULEB (ofs, start, block_end); | 
 | 	      if (! do_debug_frames_interp) | 
 | 		printf ("  DW_CFA_GNU_args_size: %" PRIu64 "\n", ofs); | 
 | 	      break; | 
 |  | 
 | 	    case DW_CFA_GNU_negative_offset_extended: | 
 | 	      READ_ULEB (reg, start, block_end); | 
 | 	      READ_SLEB (sofs, start, block_end); | 
 | 	      ofs = sofs; | 
 | 	      ofs = -ofs * fc->data_factor; | 
 | 	      if (reg >= fc->ncols) | 
 | 		reg_prefix = bad_reg; | 
 | 	      if (! do_debug_frames_interp || *reg_prefix != '\0') | 
 | 		printf ("  DW_CFA_GNU_negative_offset_extended: %s%s " | 
 | 			"at cfa%+" PRId64 "\n", | 
 | 			reg_prefix, regname (reg, 0), ofs); | 
 | 	      if (*reg_prefix == '\0') | 
 | 		{ | 
 | 		  fc->col_type[reg] = DW_CFA_offset; | 
 | 		  fc->col_offset[reg] = ofs; | 
 | 		} | 
 | 	      break; | 
 |  | 
 | 	    default: | 
 | 	      if (op >= DW_CFA_lo_user && op <= DW_CFA_hi_user) | 
 | 		printf (_("  DW_CFA_??? (User defined call frame op: %#x)\n"), op); | 
 | 	      else | 
 | 		warn (_("Unsupported or unknown Dwarf Call Frame Instruction number: %#x\n"), op); | 
 | 	      start = block_end; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       /* Interpret the CFA - as long as it is not completely full of NOPs.  */ | 
 |       if (do_debug_frames_interp && ! all_nops) | 
 | 	frame_display_row (fc, &need_col_headers, &max_regs); | 
 |  | 
 |       if (fde_fc.col_type != NULL) | 
 | 	{ | 
 | 	  free (fde_fc.col_type); | 
 | 	  fde_fc.col_type = NULL; | 
 | 	} | 
 |       if (fde_fc.col_offset != NULL) | 
 | 	{ | 
 | 	  free (fde_fc.col_offset); | 
 | 	  fde_fc.col_offset = NULL; | 
 | 	} | 
 |  | 
 |       start = block_end; | 
 |       eh_addr_size = saved_eh_addr_size; | 
 |     } | 
 |  | 
 |   printf ("\n"); | 
 |  | 
 |   while (remembered_state != NULL) | 
 |     { | 
 |       rs = remembered_state; | 
 |       remembered_state = rs->next; | 
 |       free (rs->col_type); | 
 |       free (rs->col_offset); | 
 |       rs->next = NULL; /* Paranoia.  */ | 
 |       free (rs); | 
 |     } | 
 |  | 
 |   while (chunks != NULL) | 
 |     { | 
 |       rs = chunks; | 
 |       chunks = rs->next; | 
 |       free (rs->col_type); | 
 |       free (rs->col_offset); | 
 |       rs->next = NULL; /* Paranoia.  */ | 
 |       free (rs); | 
 |     } | 
 |  | 
 |   while (forward_refs != NULL) | 
 |     { | 
 |       rs = forward_refs; | 
 |       forward_refs = rs->next; | 
 |       free (rs->col_type); | 
 |       free (rs->col_offset); | 
 |       rs->next = NULL; /* Paranoia.  */ | 
 |       free (rs); | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | #undef GET | 
 |  | 
 | static int | 
 | display_debug_names (struct dwarf_section *section, void *file) | 
 | { | 
 |   unsigned char *hdrptr = section->start; | 
 |   uint64_t unit_length; | 
 |   unsigned char *unit_start; | 
 |   const unsigned char *const section_end = section->start + section->size; | 
 |   unsigned char *unit_end; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   load_debug_section_with_follow (str, file); | 
 |  | 
 |   for (; hdrptr < section_end; hdrptr = unit_end) | 
 |     { | 
 |       unsigned int offset_size; | 
 |       uint16_t dwarf_version, padding; | 
 |       uint32_t comp_unit_count, local_type_unit_count, foreign_type_unit_count; | 
 |       uint64_t bucket_count, name_count, abbrev_table_size; | 
 |       uint32_t augmentation_string_size; | 
 |       unsigned int i; | 
 |       bool augmentation_printable; | 
 |       const char *augmentation_string; | 
 |       size_t total; | 
 |  | 
 |       unit_start = hdrptr; | 
 |  | 
 |       /* Get and check the length of the block.  */ | 
 |       SAFE_BYTE_GET_AND_INC (unit_length, hdrptr, 4, section_end); | 
 |  | 
 |       if (unit_length == 0xffffffff) | 
 | 	{ | 
 | 	  /* This section is 64-bit DWARF.  */ | 
 | 	  SAFE_BYTE_GET_AND_INC (unit_length, hdrptr, 8, section_end); | 
 | 	  offset_size = 8; | 
 | 	} | 
 |       else | 
 | 	offset_size = 4; | 
 |  | 
 |       if (unit_length > (size_t) (section_end - hdrptr) | 
 | 	  || unit_length < 2 + 2 + 4 * 7) | 
 | 	{ | 
 | 	too_short: | 
 | 	  warn (_("Debug info is corrupted, %s header at %#tx" | 
 | 		  " has length %#" PRIx64 "\n"), | 
 | 		section->name, unit_start - section->start, unit_length); | 
 | 	  return 0; | 
 | 	} | 
 |       unit_end = hdrptr + unit_length; | 
 |  | 
 |       /* Get and check the version number.  */ | 
 |       SAFE_BYTE_GET_AND_INC (dwarf_version, hdrptr, 2, unit_end); | 
 |       printf (_("Version %d\n"), (int) dwarf_version); | 
 |  | 
 |       /* Prior versions did not exist, and future versions may not be | 
 | 	 backwards compatible.  */ | 
 |       if (dwarf_version != 5) | 
 | 	{ | 
 | 	  warn (_("Only DWARF version 5 .debug_names " | 
 | 		  "is currently supported.\n")); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (padding, hdrptr, 2, unit_end); | 
 |       if (padding != 0) | 
 | 	warn (_("Padding field of .debug_names must be 0 (found 0x%x)\n"), | 
 | 	      padding); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (comp_unit_count, hdrptr, 4, unit_end); | 
 |       if (comp_unit_count == 0) | 
 | 	warn (_("Compilation unit count must be >= 1 in .debug_names\n")); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (local_type_unit_count, hdrptr, 4, unit_end); | 
 |       SAFE_BYTE_GET_AND_INC (foreign_type_unit_count, hdrptr, 4, unit_end); | 
 |       SAFE_BYTE_GET_AND_INC (bucket_count, hdrptr, 4, unit_end); | 
 |       SAFE_BYTE_GET_AND_INC (name_count, hdrptr, 4, unit_end); | 
 |       SAFE_BYTE_GET_AND_INC (abbrev_table_size, hdrptr, 4, unit_end); | 
 |  | 
 |       SAFE_BYTE_GET_AND_INC (augmentation_string_size, hdrptr, 4, unit_end); | 
 |       if (augmentation_string_size % 4 != 0) | 
 | 	{ | 
 | 	  warn (_("Augmentation string length %u must be rounded up " | 
 | 		  "to a multiple of 4 in .debug_names.\n"), | 
 | 		augmentation_string_size); | 
 | 	  augmentation_string_size += (-augmentation_string_size) & 3; | 
 | 	} | 
 |       if (augmentation_string_size > (size_t) (unit_end - hdrptr)) | 
 | 	goto too_short; | 
 |  | 
 |       printf (_("Augmentation string:")); | 
 |  | 
 |       augmentation_printable = true; | 
 |       augmentation_string = (const char *) hdrptr; | 
 |  | 
 |       for (i = 0; i < augmentation_string_size; i++) | 
 | 	{ | 
 | 	  unsigned char uc; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (uc, hdrptr, 1, unit_end); | 
 | 	  printf (" %02x", uc); | 
 |  | 
 | 	  if (uc != 0 && !ISPRINT (uc)) | 
 | 	    augmentation_printable = false; | 
 | 	} | 
 |  | 
 |       if (augmentation_printable) | 
 | 	{ | 
 | 	  printf ("  (\""); | 
 | 	  for (i = 0; | 
 | 	       i < augmentation_string_size && augmentation_string[i]; | 
 | 	       ++i) | 
 | 	    putchar (augmentation_string[i]); | 
 | 	  printf ("\")"); | 
 | 	} | 
 |       putchar ('\n'); | 
 |  | 
 |       printf (_("CU table:\n")); | 
 |       if (_mul_overflow (comp_unit_count, offset_size, &total) | 
 | 	  || total > (size_t) (unit_end - hdrptr)) | 
 | 	goto too_short; | 
 |       for (i = 0; i < comp_unit_count; i++) | 
 | 	{ | 
 | 	  uint64_t cu_offset; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (cu_offset, hdrptr, offset_size, unit_end); | 
 | 	  printf ("[%3u] %#" PRIx64 "\n", i, cu_offset); | 
 | 	} | 
 |       putchar ('\n'); | 
 |  | 
 |       printf (_("TU table:\n")); | 
 |       if (_mul_overflow (local_type_unit_count, offset_size, &total) | 
 | 	  || total > (size_t) (unit_end - hdrptr)) | 
 | 	goto too_short; | 
 |       for (i = 0; i < local_type_unit_count; i++) | 
 | 	{ | 
 | 	  uint64_t tu_offset; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (tu_offset, hdrptr, offset_size, unit_end); | 
 | 	  printf ("[%3u] %#" PRIx64 "\n", i, tu_offset); | 
 | 	} | 
 |       putchar ('\n'); | 
 |  | 
 |       printf (_("Foreign TU table:\n")); | 
 |       if (_mul_overflow (foreign_type_unit_count, 8, &total) | 
 | 	  || total > (size_t) (unit_end - hdrptr)) | 
 | 	goto too_short; | 
 |       for (i = 0; i < foreign_type_unit_count; i++) | 
 | 	{ | 
 | 	  uint64_t signature; | 
 |  | 
 | 	  SAFE_BYTE_GET_AND_INC (signature, hdrptr, 8, unit_end); | 
 | 	  printf (_("[%3u] "), i); | 
 | 	  print_hex_ns (signature, 8); | 
 | 	  putchar ('\n'); | 
 | 	} | 
 |       putchar ('\n'); | 
 |  | 
 |       uint64_t xtra = (bucket_count * sizeof (uint32_t) | 
 | 		       + name_count * (sizeof (uint32_t) + 2 * offset_size) | 
 | 		       + abbrev_table_size); | 
 |       if (xtra > (size_t) (unit_end - hdrptr)) | 
 | 	{ | 
 | 	  warn (_("Entry pool offset (%#" PRIx64 ") exceeds unit size %#tx " | 
 | 		  "for unit %#tx in the debug_names\n"), | 
 | 		xtra, unit_end - unit_start, unit_start - section->start); | 
 | 	  return 0; | 
 | 	} | 
 |       const uint32_t *const hash_table_buckets = (uint32_t *) hdrptr; | 
 |       hdrptr += bucket_count * sizeof (uint32_t); | 
 |       const uint32_t *const hash_table_hashes = (uint32_t *) hdrptr; | 
 |       if (bucket_count != 0) | 
 | 	hdrptr += name_count * sizeof (uint32_t); | 
 |       unsigned char *const name_table_string_offsets = hdrptr; | 
 |       hdrptr += name_count * offset_size; | 
 |       unsigned char *const name_table_entry_offsets = hdrptr; | 
 |       hdrptr += name_count * offset_size; | 
 |       unsigned char *const abbrev_table = hdrptr; | 
 |       hdrptr += abbrev_table_size; | 
 |       const unsigned char *const abbrev_table_end = hdrptr; | 
 |       unsigned char *const entry_pool = hdrptr; | 
 |  | 
 |       size_t buckets_filled = 0; | 
 |       size_t bucketi; | 
 |       for (bucketi = 0; bucketi < bucket_count; bucketi++) | 
 | 	{ | 
 | 	  const uint32_t bucket = hash_table_buckets[bucketi]; | 
 |  | 
 | 	  if (bucket != 0) | 
 | 	    ++buckets_filled; | 
 | 	} | 
 |       printf (ngettext ("Used %zu of %lu bucket.\n", | 
 | 			"Used %zu of %lu buckets.\n", | 
 | 			(unsigned long) bucket_count), | 
 | 	      buckets_filled, (unsigned long) bucket_count); | 
 |  | 
 |       if (bucket_count != 0) | 
 | 	{ | 
 | 	  uint32_t hash_prev = 0; | 
 | 	  size_t hash_clash_count = 0; | 
 | 	  size_t longest_clash = 0; | 
 | 	  size_t this_length = 0; | 
 | 	  size_t hashi; | 
 | 	  for (hashi = 0; hashi < name_count; hashi++) | 
 | 	    { | 
 | 	      const uint32_t hash_this = hash_table_hashes[hashi]; | 
 |  | 
 | 	      if (hashi > 0) | 
 | 		{ | 
 | 		  if (hash_prev % bucket_count == hash_this % bucket_count) | 
 | 		    { | 
 | 		      ++hash_clash_count; | 
 | 		      ++this_length; | 
 | 		      longest_clash = MAX (longest_clash, this_length); | 
 | 		    } | 
 | 		  else | 
 | 		    this_length = 0; | 
 | 		} | 
 | 	      hash_prev = hash_this; | 
 | 	    } | 
 | 	  printf (_("Out of %" PRIu64 " items there are %zu bucket clashes" | 
 | 		    " (longest of %zu entries).\n"), | 
 | 		  name_count, hash_clash_count, longest_clash); | 
 |  | 
 | 	  if (name_count != buckets_filled + hash_clash_count) | 
 | 	    warn (_("The name_count (%" PRIu64 ")" | 
 | 		    " is not the same as the used bucket_count" | 
 | 		    " (%zu) + the hash clash count (%zu)\n"), | 
 | 		  name_count, buckets_filled, hash_clash_count); | 
 | 	} | 
 |  | 
 |       struct abbrev_lookup_entry | 
 |       { | 
 | 	uint64_t abbrev_tag; | 
 | 	unsigned char *abbrev_lookup_ptr; | 
 |       }; | 
 |       struct abbrev_lookup_entry *abbrev_lookup = NULL; | 
 |       size_t abbrev_lookup_used = 0; | 
 |       size_t abbrev_lookup_allocated = 0; | 
 |  | 
 |       unsigned char *abbrevptr = abbrev_table; | 
 |       for (;;) | 
 | 	{ | 
 | 	  uint64_t abbrev_tag; | 
 |  | 
 | 	  READ_ULEB (abbrev_tag, abbrevptr, abbrev_table_end); | 
 | 	  if (abbrev_tag == 0) | 
 | 	    break; | 
 | 	  if (abbrev_lookup_used == abbrev_lookup_allocated) | 
 | 	    { | 
 | 	      abbrev_lookup_allocated = MAX (0x100, | 
 | 					     abbrev_lookup_allocated * 2); | 
 | 	      abbrev_lookup = xrealloc (abbrev_lookup, | 
 | 					(abbrev_lookup_allocated | 
 | 					 * sizeof (*abbrev_lookup))); | 
 | 	    } | 
 | 	  assert (abbrev_lookup_used < abbrev_lookup_allocated); | 
 | 	  struct abbrev_lookup_entry *entry; | 
 | 	  for (entry = abbrev_lookup; | 
 | 	       entry < abbrev_lookup + abbrev_lookup_used; | 
 | 	       entry++) | 
 | 	    if (entry->abbrev_tag == abbrev_tag) | 
 | 	      { | 
 | 		warn (_("Duplicate abbreviation tag %" PRIu64 | 
 | 			" in unit %#tx in the debug_names section\n"), | 
 | 		      abbrev_tag, unit_start - section->start); | 
 | 		break; | 
 | 	      } | 
 | 	  entry = &abbrev_lookup[abbrev_lookup_used++]; | 
 | 	  entry->abbrev_tag = abbrev_tag; | 
 | 	  entry->abbrev_lookup_ptr = abbrevptr; | 
 |  | 
 | 	  /* Skip DWARF tag.  */ | 
 | 	  SKIP_ULEB (abbrevptr, abbrev_table_end); | 
 | 	  for (;;) | 
 | 	    { | 
 | 	      uint64_t xindex, form; | 
 |  | 
 | 	      READ_ULEB (xindex, abbrevptr, abbrev_table_end); | 
 | 	      READ_ULEB (form, abbrevptr, abbrev_table_end); | 
 | 	      if (xindex == 0 && form == 0) | 
 | 		break; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       printf (_("\nSymbol table:\n")); | 
 |       uint32_t namei; | 
 |       for (namei = 0; namei < name_count; ++namei) | 
 | 	{ | 
 | 	  uint64_t string_offset, entry_offset; | 
 | 	  unsigned char *p; | 
 | 	  /* We need to scan first whether there is a single or multiple | 
 | 	     entries.  TAGNO is -2 for the first entry, it is -1 for the | 
 | 	     initial tag read of the second entry, then it becomes 0 for the | 
 | 	     first entry for real printing etc.  */ | 
 | 	  int tagno = -2; | 
 | 	  /* Initialize it due to a false compiler warning.  */ | 
 | 	  uint64_t second_abbrev_tag = -1; | 
 | 	  unsigned char *entryptr; | 
 |  | 
 | 	  p = name_table_string_offsets + namei * offset_size; | 
 | 	  SAFE_BYTE_GET (string_offset, p, offset_size, unit_end); | 
 |  | 
 | 	  p = name_table_entry_offsets + namei * offset_size; | 
 | 	  SAFE_BYTE_GET (entry_offset, p, offset_size, unit_end); | 
 |  | 
 | 	  /* The name table is indexed starting at 1 according to | 
 | 	     DWARF, so be sure to use the DWARF numbering here.  */ | 
 | 	  printf ("[%3u] ", namei + 1); | 
 | 	  if (bucket_count != 0) | 
 | 	    printf ("#%08x ", hash_table_hashes[namei]); | 
 |  | 
 | 	  printf ("%s:", fetch_indirect_string (string_offset)); | 
 |  | 
 | 	  entryptr = entry_pool + entry_offset; | 
 | 	  /* PR 31456: Check for invalid entry offset.  */ | 
 | 	  if (entryptr < entry_pool || entryptr >= unit_end) | 
 | 	    { | 
 | 	      warn (_("Invalid entry offset value: %" PRIx64 "\n"), entry_offset); | 
 | 	      break; | 
 | 	    } | 
 |  | 
 | 	  for (;;) | 
 | 	    { | 
 | 	      uint64_t abbrev_tag; | 
 | 	      uint64_t dwarf_tag; | 
 | 	      const struct abbrev_lookup_entry *entry; | 
 | 	      uint64_t this_entry = entryptr - entry_pool; | 
 |  | 
 | 	      READ_ULEB (abbrev_tag, entryptr, unit_end); | 
 | 	      if (tagno == -1) | 
 | 		{ | 
 | 		  second_abbrev_tag = abbrev_tag; | 
 | 		  tagno = 0; | 
 | 		  entryptr = entry_pool + entry_offset; | 
 | 		  continue; | 
 | 		} | 
 | 	      if (abbrev_tag == 0) | 
 | 		break; | 
 | 	      if (tagno >= 0) | 
 | 		printf ("%s<%#" PRIx64 "><%" PRIu64 ">", | 
 | 			(tagno == 0 && second_abbrev_tag == 0 ? " " : "\n\t"), | 
 | 			this_entry, abbrev_tag); | 
 |  | 
 | 	      for (entry = abbrev_lookup; | 
 | 		   entry < abbrev_lookup + abbrev_lookup_used; | 
 | 		   entry++) | 
 | 		if (entry->abbrev_tag == abbrev_tag) | 
 | 		  break; | 
 | 	      if (entry >= abbrev_lookup + abbrev_lookup_used) | 
 | 		{ | 
 | 		  warn (_("Undefined abbreviation tag %" PRId64 | 
 | 			  " in unit %#tx in the debug_names section\n"), | 
 | 			abbrev_tag, | 
 | 			unit_start - section->start); | 
 | 		  break; | 
 | 		} | 
 | 	      abbrevptr = entry->abbrev_lookup_ptr; | 
 | 	      READ_ULEB (dwarf_tag, abbrevptr, abbrev_table_end); | 
 | 	      if (tagno >= 0) | 
 | 		printf (" %s", get_TAG_name (dwarf_tag)); | 
 | 	      for (;;) | 
 | 		{ | 
 | 		  uint64_t xindex, form; | 
 |  | 
 | 		  READ_ULEB (xindex, abbrevptr, abbrev_table_end); | 
 | 		  READ_ULEB (form, abbrevptr, abbrev_table_end); | 
 | 		  if (xindex == 0 && form == 0) | 
 | 		    break; | 
 |  | 
 | 		  if (tagno >= 0) | 
 | 		    printf (" %s", get_IDX_name (xindex)); | 
 | 		  entryptr = read_and_display_attr_value (0, form, 0, | 
 | 							  unit_start, entryptr, unit_end, | 
 | 							  0, 0, offset_size, | 
 | 							  dwarf_version, NULL, | 
 | 							  (tagno < 0), section, | 
 | 							  NULL, '=', -1); | 
 | 		} | 
 | 	      ++tagno; | 
 | 	    } | 
 | 	  if (tagno <= 0) | 
 | 	    printf (_(" <no entries>")); | 
 | 	  putchar ('\n'); | 
 | 	} | 
 |  | 
 |       free (abbrev_lookup); | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_debug_links (struct dwarf_section *  section, | 
 | 		     void *                  file ATTRIBUTE_UNUSED) | 
 | { | 
 |   const unsigned char * filename; | 
 |   unsigned int          filelen; | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   /* The .gnu_debuglink section is formatted as: | 
 |       (c-string)  Filename. | 
 |       (padding)   If needed to reach a 4 byte boundary. | 
 |       (uint32_t)  CRC32 value. | 
 |  | 
 |     The .gnu_debugaltlink section is formatted as: | 
 |       (c-string)  Filename. | 
 |       (binary)    Build-ID.  */ | 
 |  | 
 |   filename =  section->start; | 
 |   filelen = strnlen ((const char *) filename, section->size); | 
 |   if (filelen == section->size) | 
 |     { | 
 |       warn (_("The debuglink filename is corrupt/missing\n")); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   printf (_("  Separate debug info file: %s\n"), filename); | 
 |  | 
 |   if (startswith (section->name, ".gnu_debuglink")) | 
 |     { | 
 |       unsigned int          crc32; | 
 |       unsigned int          crc_offset; | 
 |  | 
 |       crc_offset = filelen + 1; | 
 |       crc_offset = (crc_offset + 3) & ~3; | 
 |       if (crc_offset + 4 > section->size) | 
 | 	{ | 
 | 	  warn (_("CRC offset missing/truncated\n")); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       crc32 = byte_get (filename + crc_offset, 4); | 
 |  | 
 |       printf (_("  CRC value: %#x\n"), crc32); | 
 |  | 
 |       if (crc_offset + 4 < section->size) | 
 | 	{ | 
 | 	  warn (_("There are %#" PRIx64 | 
 | 		  " extraneous bytes at the end of the section\n"), | 
 | 		section->size - (crc_offset + 4)); | 
 | 	  return 0; | 
 | 	} | 
 |     } | 
 |   else /* startswith (section->name, ".gnu_debugaltlink") */ | 
 |     { | 
 |       const unsigned char *build_id = section->start + filelen + 1; | 
 |       size_t build_id_len = section->size - (filelen + 1); | 
 |       size_t printed; | 
 |  | 
 |       /* FIXME: Should we support smaller build-id notes ?  */ | 
 |       if (build_id_len < 0x14) | 
 | 	{ | 
 | 	  warn (_("Build-ID is too short (%#zx bytes)\n"), build_id_len); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       printed = printf (_("  Build-ID (%#zx bytes):"), build_id_len); | 
 |       display_data (printed, build_id, build_id_len); | 
 |       putchar ('\n'); | 
 |     } | 
 |  | 
 |   putchar ('\n'); | 
 |   return 1; | 
 | } | 
 |  | 
 | static int | 
 | display_gdb_index (struct dwarf_section *section, | 
 | 		   void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   unsigned char *start = section->start; | 
 |   uint32_t version; | 
 |   uint32_t cu_list_offset, tu_list_offset; | 
 |   uint32_t address_table_offset, symbol_table_offset, constant_pool_offset, | 
 |     shortcut_table_offset; | 
 |   unsigned int cu_list_elements, tu_list_elements; | 
 |   unsigned int address_table_elements, symbol_table_slots; | 
 |   unsigned char *cu_list, *tu_list; | 
 |   unsigned char *address_table, *symbol_table, *shortcut_table, *constant_pool; | 
 |   unsigned int i; | 
 |  | 
 |   /* The documentation for the format of this file is in gdb/dwarf2read.c.  */ | 
 |  | 
 |   introduce (section, false); | 
 |  | 
 |   version = section->size < 4 ? 0 : byte_get_little_endian (start, 4); | 
 |   size_t header_size = (version < 9 ? 6 : 7) * sizeof (uint32_t); | 
 |   if (section->size < header_size) | 
 |     { | 
 |       warn (_("Truncated header in the %s section.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   printf (_("Version %lu\n"), (unsigned long) version); | 
 |  | 
 |   /* Prior versions are obsolete, and future versions may not be | 
 |      backwards compatible.  */ | 
 |   if (version < 3 || version > 9) | 
 |     { | 
 |       warn (_("Unsupported version %lu.\n"), (unsigned long) version); | 
 |       return 0; | 
 |     } | 
 |   if (version < 4) | 
 |     warn (_("The address table data in version 3 may be wrong.\n")); | 
 |   if (version < 5) | 
 |     warn (_("Version 4 does not support case insensitive lookups.\n")); | 
 |   if (version < 6) | 
 |     warn (_("Version 5 does not include inlined functions.\n")); | 
 |   if (version < 7) | 
 |     warn (_("Version 6 does not include symbol attributes.\n")); | 
 |   /* Version 7 indices generated by Gold have bad type unit references, | 
 |      PR binutils/15021.  But we don't know if the index was generated by | 
 |      Gold or not, so to avoid worrying users with gdb-generated indices | 
 |      we say nothing for version 7 here.  */ | 
 |  | 
 |   cu_list_offset = byte_get_little_endian (start + 4, 4); | 
 |   tu_list_offset = byte_get_little_endian (start + 8, 4); | 
 |   address_table_offset = byte_get_little_endian (start + 12, 4); | 
 |   symbol_table_offset = byte_get_little_endian (start + 16, 4); | 
 |   shortcut_table_offset = byte_get_little_endian (start + 20, 4); | 
 |   if (version < 9) | 
 |     constant_pool_offset = shortcut_table_offset; | 
 |   else | 
 |     constant_pool_offset = byte_get_little_endian (start + 24, 4); | 
 |  | 
 |   if (cu_list_offset > section->size | 
 |       || tu_list_offset > section->size | 
 |       || address_table_offset > section->size | 
 |       || symbol_table_offset > section->size | 
 |       || shortcut_table_offset > section->size | 
 |       || constant_pool_offset > section->size | 
 |       || tu_list_offset < cu_list_offset | 
 |       || address_table_offset < tu_list_offset | 
 |       || symbol_table_offset < address_table_offset | 
 |       || shortcut_table_offset < symbol_table_offset | 
 |       || constant_pool_offset < shortcut_table_offset) | 
 |     { | 
 |       warn (_("Corrupt header in the %s section.\n"), section->name); | 
 |       return 0; | 
 |     } | 
 |  | 
 |   cu_list_elements = (tu_list_offset - cu_list_offset) / 16; | 
 |   tu_list_elements = (address_table_offset - tu_list_offset) / 24; | 
 |   address_table_elements = (symbol_table_offset - address_table_offset) / 20; | 
 |   symbol_table_slots = (shortcut_table_offset - symbol_table_offset) / 8; | 
 |  | 
 |   cu_list = start + cu_list_offset; | 
 |   tu_list = start + tu_list_offset; | 
 |   address_table = start + address_table_offset; | 
 |   symbol_table = start + symbol_table_offset; | 
 |   shortcut_table = start + shortcut_table_offset; | 
 |   constant_pool = start + constant_pool_offset; | 
 |  | 
 |   printf (_("\nCU table:\n")); | 
 |   for (i = 0; i < cu_list_elements; i++) | 
 |     { | 
 |       uint64_t cu_offset = byte_get_little_endian (cu_list + i * 16, 8); | 
 |       uint64_t cu_length = byte_get_little_endian (cu_list + i * 16 + 8, 8); | 
 |  | 
 |       printf ("[%3u] %#" PRIx64 " - %#" PRIx64 "\n", | 
 | 	      i, cu_offset, cu_offset + cu_length - 1); | 
 |     } | 
 |  | 
 |   printf (_("\nTU table:\n")); | 
 |   for (i = 0; i < tu_list_elements; i++) | 
 |     { | 
 |       uint64_t tu_offset = byte_get_little_endian (tu_list + i * 24, 8); | 
 |       uint64_t type_offset = byte_get_little_endian (tu_list + i * 24 + 8, 8); | 
 |       uint64_t signature = byte_get_little_endian (tu_list + i * 24 + 16, 8); | 
 |  | 
 |       printf ("[%3u] %#" PRIx64 " %#" PRIx64 " ", | 
 | 	      i, tu_offset, type_offset); | 
 |       print_hex_ns (signature, 8); | 
 |       printf ("\n"); | 
 |     } | 
 |  | 
 |   printf (_("\nAddress table:\n")); | 
 |   for (i = 0; i < address_table_elements; i++) | 
 |     { | 
 |       uint64_t low = byte_get_little_endian (address_table + i * 20, 8); | 
 |       uint64_t high = byte_get_little_endian (address_table + i * 20 + 8, 8); | 
 |       uint32_t cu_index = byte_get_little_endian (address_table + i * 20 + 16, 4); | 
 |  | 
 |       print_hex (low, 8); | 
 |       print_hex (high, 8); | 
 |       printf ("%" PRIu32 "\n", cu_index); | 
 |     } | 
 |  | 
 |   printf (_("\nSymbol table:\n")); | 
 |   for (i = 0; i < symbol_table_slots; ++i) | 
 |     { | 
 |       uint32_t name_offset = byte_get_little_endian (symbol_table + i * 8, 4); | 
 |       uint32_t cu_vector_offset = byte_get_little_endian (symbol_table + i * 8 + 4, 4); | 
 |       uint32_t num_cus, cu; | 
 |  | 
 |       if (name_offset != 0 | 
 | 	  || cu_vector_offset != 0) | 
 | 	{ | 
 | 	  unsigned int j; | 
 |  | 
 | 	  /* PR 17531: file: 5b7b07ad.  */ | 
 | 	  if (name_offset >= section->size - constant_pool_offset) | 
 | 	    { | 
 | 	      printf (_("[%3u] <corrupt offset: %x>"), i, name_offset); | 
 | 	      warn (_("Corrupt name offset of 0x%x found for symbol table slot %d\n"), | 
 | 		    name_offset, i); | 
 | 	    } | 
 | 	  else | 
 | 	    printf ("[%3u] %.*s:", i, | 
 | 		    (int) (section->size - (constant_pool_offset + name_offset)), | 
 | 		    constant_pool + name_offset); | 
 |  | 
 | 	  if (section->size - constant_pool_offset < 4 | 
 | 	      || cu_vector_offset > section->size - constant_pool_offset - 4) | 
 | 	    { | 
 | 	      printf (_("<invalid CU vector offset: %x>\n"), cu_vector_offset); | 
 | 	      warn (_("Corrupt CU vector offset of 0x%x found for symbol table slot %d\n"), | 
 | 		    cu_vector_offset, i); | 
 | 	      continue; | 
 | 	    } | 
 |  | 
 | 	  num_cus = byte_get_little_endian (constant_pool + cu_vector_offset, 4); | 
 |  | 
 | 	  if ((uint64_t) num_cus * 4 > section->size - (constant_pool_offset | 
 | 							+ cu_vector_offset + 4)) | 
 | 	    { | 
 | 	      printf ("<invalid number of CUs: %d>\n", num_cus); | 
 | 	      warn (_("Invalid number of CUs (0x%x) for symbol table slot %d\n"), | 
 | 		    num_cus, i); | 
 | 	      continue; | 
 | 	    } | 
 |  | 
 | 	  if (num_cus > 1) | 
 | 	    printf ("\n"); | 
 |  | 
 | 	  for (j = 0; j < num_cus; ++j) | 
 | 	    { | 
 | 	      int is_static; | 
 | 	      gdb_index_symbol_kind kind; | 
 |  | 
 | 	      cu = byte_get_little_endian (constant_pool + cu_vector_offset + 4 + j * 4, 4); | 
 | 	      is_static = GDB_INDEX_SYMBOL_STATIC_VALUE (cu); | 
 | 	      kind = GDB_INDEX_SYMBOL_KIND_VALUE (cu); | 
 | 	      cu = GDB_INDEX_CU_VALUE (cu); | 
 | 	      /* Convert to TU number if it's for a type unit.  */ | 
 | 	      if (cu >= cu_list_elements) | 
 | 		printf ("%cT%lu", num_cus > 1 ? '\t' : ' ', | 
 | 			(unsigned long) cu - cu_list_elements); | 
 | 	      else | 
 | 		printf ("%c%lu", num_cus > 1 ? '\t' : ' ', (unsigned long) cu); | 
 |  | 
 | 	      printf (" [%s, %s]", | 
 | 		      is_static ? _("static") : _("global"), | 
 | 		      get_gdb_index_symbol_kind_name (kind)); | 
 | 	      if (num_cus > 1) | 
 | 		printf ("\n"); | 
 | 	    } | 
 | 	  if (num_cus <= 1) | 
 | 	    printf ("\n"); | 
 | 	} | 
 |     } | 
 |  | 
 |   if (version >= 9) | 
 |     { | 
 |       printf (_("\nShortcut table:\n")); | 
 |  | 
 |       if (shortcut_table_offset + 8 > constant_pool_offset) | 
 | 	{ | 
 | 	  warn (_("Corrupt shortcut table in the %s section.\n"), section->name); | 
 | 	  return 0; | 
 | 	} | 
 |  | 
 |       uint32_t lang = byte_get_little_endian (shortcut_table, 4); | 
 |       printf (_("Language of main: ")); | 
 |       display_lang (lang); | 
 |       printf ("\n"); | 
 |  | 
 |       printf (_("Name of main: ")); | 
 |       if (lang == 0) | 
 | 	printf (_("<unknown>\n")); | 
 |       else | 
 | 	{ | 
 | 	  uint32_t name_offset = byte_get_little_endian (shortcut_table + 4, 4); | 
 | 	  if (name_offset >= section->size - constant_pool_offset) | 
 | 	    { | 
 | 	      printf (_("<corrupt offset: %x>\n"), name_offset); | 
 | 	      warn (_("Corrupt name offset of 0x%x found for name of main\n"), | 
 | 		    name_offset); | 
 | 	    } | 
 | 	  else | 
 | 	    printf ("%s\n", constant_pool + name_offset); | 
 | 	} | 
 |     } | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Pre-allocate enough space for the CU/TU sets needed.  */ | 
 |  | 
 | static void | 
 | prealloc_cu_tu_list (unsigned int nshndx) | 
 | { | 
 |   if (nshndx == 0) | 
 |     /* Always allocate at least one entry for the end-marker.  */ | 
 |     nshndx = 1; | 
 |  | 
 |   if (shndx_pool == NULL) | 
 |     { | 
 |       shndx_pool_size = nshndx; | 
 |       shndx_pool_used = 0; | 
 |       shndx_pool = (unsigned int *) xcmalloc (shndx_pool_size, | 
 | 					      sizeof (unsigned int)); | 
 |     } | 
 |   else | 
 |     { | 
 |       shndx_pool_size = shndx_pool_used + nshndx; | 
 |       shndx_pool = (unsigned int *) xcrealloc (shndx_pool, shndx_pool_size, | 
 | 					       sizeof (unsigned int)); | 
 |     } | 
 | } | 
 |  | 
 | static void | 
 | add_shndx_to_cu_tu_entry (unsigned int shndx) | 
 | { | 
 |   shndx_pool [shndx_pool_used++] = shndx; | 
 | } | 
 |  | 
 | static void | 
 | end_cu_tu_entry (void) | 
 | { | 
 |   shndx_pool [shndx_pool_used++] = 0; | 
 | } | 
 |  | 
 | /* Return the short name of a DWARF section given by a DW_SECT enumerator.  */ | 
 |  | 
 | static const char * | 
 | get_DW_SECT_short_name (unsigned int dw_sect) | 
 | { | 
 |   static char buf[16]; | 
 |  | 
 |   switch (dw_sect) | 
 |     { | 
 |     case DW_SECT_INFO: | 
 |       return "info"; | 
 |     case DW_SECT_TYPES: | 
 |       return "types"; | 
 |     case DW_SECT_ABBREV: | 
 |       return "abbrev"; | 
 |     case DW_SECT_LINE: | 
 |       return "line"; | 
 |     case DW_SECT_LOC: | 
 |       return "loc"; | 
 |     case DW_SECT_STR_OFFSETS: | 
 |       return "str_off"; | 
 |     case DW_SECT_MACINFO: | 
 |       return "macinfo"; | 
 |     case DW_SECT_MACRO: | 
 |       return "macro"; | 
 |     default: | 
 |       break; | 
 |     } | 
 |  | 
 |   snprintf (buf, sizeof (buf), "%d", dw_sect); | 
 |   return buf; | 
 | } | 
 |  | 
 | /* Process a CU or TU index.  If DO_DISPLAY is true, print the contents. | 
 |    These sections are extensions for Fission. | 
 |    See http://gcc.gnu.org/wiki/DebugFissionDWP.  */ | 
 |  | 
 | static bool | 
 | process_cu_tu_index (struct dwarf_section *section, int do_display) | 
 | { | 
 |   unsigned char *phdr = section->start; | 
 |   unsigned char *limit = phdr + section->size; | 
 |   unsigned char *phash; | 
 |   unsigned char *pindex; | 
 |   unsigned char *ppool; | 
 |   unsigned int version; | 
 |   unsigned int ncols = 0; | 
 |   unsigned int nused; | 
 |   unsigned int nslots; | 
 |   unsigned int i; | 
 |   unsigned int j; | 
 |   uint64_t signature; | 
 |   size_t total; | 
 |  | 
 |   /* PR 17512: file: 002-168123-0.004.  */ | 
 |   if (phdr == NULL) | 
 |     { | 
 |       warn (_("Section %s is empty\n"), section->name); | 
 |       return false; | 
 |     } | 
 |   /* PR 17512: file: 002-376-0.004.  */ | 
 |   if (section->size < 24) | 
 |     { | 
 |       warn (_("Section %s is too small to contain a CU/TU header\n"), | 
 | 	    section->name); | 
 |       return false; | 
 |     } | 
 |  | 
 |   phash = phdr; | 
 |   SAFE_BYTE_GET_AND_INC (version, phash, 4, limit); | 
 |   if (version >= 2) | 
 |     SAFE_BYTE_GET_AND_INC (ncols, phash, 4, limit); | 
 |   SAFE_BYTE_GET_AND_INC (nused, phash, 4, limit); | 
 |   SAFE_BYTE_GET_AND_INC (nslots, phash, 4, limit); | 
 |  | 
 |   pindex = phash + (size_t) nslots * 8; | 
 |   ppool = pindex + (size_t) nslots * 4; | 
 |  | 
 |   if (do_display) | 
 |     { | 
 |       introduce (section, false); | 
 |  | 
 |       printf (_("  Version:                 %u\n"), version); | 
 |       if (version >= 2) | 
 | 	printf (_("  Number of columns:       %u\n"), ncols); | 
 |       printf (_("  Number of used entries:  %u\n"), nused); | 
 |       printf (_("  Number of slots:         %u\n\n"), nslots); | 
 |     } | 
 |  | 
 |   /* PR 17531: file: 45d69832.  */ | 
 |   if (_mul_overflow ((size_t) nslots, 12, &total) | 
 |       || total > (size_t) (limit - phash)) | 
 |     { | 
 |       warn (ngettext ("Section %s is too small for %u slot\n", | 
 | 		      "Section %s is too small for %u slots\n", | 
 | 		      nslots), | 
 | 	    section->name, nslots); | 
 |       return false; | 
 |     } | 
 |  | 
 |   if (version == 1) | 
 |     { | 
 |       unsigned char *shndx_list; | 
 |       unsigned int shndx; | 
 |  | 
 |       if (!do_display) | 
 | 	{ | 
 | 	  prealloc_cu_tu_list ((limit - ppool) / 4); | 
 | 	  for (shndx_list = ppool + 4; shndx_list <= limit - 4; shndx_list += 4) | 
 | 	    { | 
 | 	      shndx = byte_get (shndx_list, 4); | 
 | 	      add_shndx_to_cu_tu_entry (shndx); | 
 | 	    } | 
 | 	  end_cu_tu_entry (); | 
 | 	} | 
 |       else | 
 | 	for (i = 0; i < nslots; i++) | 
 | 	  { | 
 | 	    SAFE_BYTE_GET (signature, phash, 8, limit); | 
 | 	    if (signature != 0) | 
 | 	      { | 
 | 		SAFE_BYTE_GET (j, pindex, 4, limit); | 
 | 		shndx_list = ppool + j * 4; | 
 | 		/* PR 17531: file: 705e010d.  */ | 
 | 		if (shndx_list < ppool) | 
 | 		  { | 
 | 		    warn (_("Section index pool located before start of section\n")); | 
 | 		    return false; | 
 | 		  } | 
 |  | 
 | 		printf (_("  [%3d] Signature:  %#" PRIx64 "  Sections: "), | 
 | 			i, signature); | 
 | 		for (;;) | 
 | 		  { | 
 | 		    if (shndx_list >= limit) | 
 | 		      { | 
 | 			warn (_("Section %s too small for shndx pool\n"), | 
 | 			      section->name); | 
 | 			return false; | 
 | 		      } | 
 | 		    SAFE_BYTE_GET (shndx, shndx_list, 4, limit); | 
 | 		    if (shndx == 0) | 
 | 		      break; | 
 | 		    printf (" %d", shndx); | 
 | 		    shndx_list += 4; | 
 | 		  } | 
 | 		printf ("\n"); | 
 | 	      } | 
 | 	    phash += 8; | 
 | 	    pindex += 4; | 
 | 	  } | 
 |     } | 
 |   else if (version == 2) | 
 |     { | 
 |       unsigned int val; | 
 |       unsigned int dw_sect; | 
 |       unsigned char *ph = phash; | 
 |       unsigned char *pi = pindex; | 
 |       unsigned char *poffsets = ppool + (size_t) ncols * 4; | 
 |       unsigned char *psizes = poffsets + (size_t) nused * ncols * 4; | 
 |       bool is_tu_index; | 
 |       struct cu_tu_set *this_set = NULL; | 
 |       unsigned int row; | 
 |       unsigned char *prow; | 
 |       size_t temp; | 
 |  | 
 |       is_tu_index = strcmp (section->name, ".debug_tu_index") == 0; | 
 |  | 
 |       /* PR 17531: file: 0dd159bf. | 
 | 	 Check for integer overflow (can occur when size_t is 32-bit) | 
 | 	 with overlarge ncols or nused values.  */ | 
 |       if (nused == -1u | 
 | 	  || _mul_overflow ((size_t) ncols, 4, &temp) | 
 | 	  || _mul_overflow ((size_t) nused + 1, temp, &total) | 
 | 	  || total > (size_t) (limit - ppool) | 
 | 	  /* PR 30227: ncols could be 0.  */ | 
 | 	  || _mul_overflow ((size_t) nused + 1, 4, &total) | 
 | 	  || total > (size_t) (limit - ppool)) | 
 | 	{ | 
 | 	  warn (_("Section %s too small for offset and size tables\n"), | 
 | 		section->name); | 
 | 	  return false; | 
 | 	} | 
 |  | 
 |       if (do_display) | 
 | 	{ | 
 | 	  printf (_("  Offset table\n")); | 
 | 	  printf ("  slot  %-16s  ", | 
 | 		  is_tu_index ? _("signature") : _("dwo_id")); | 
 | 	} | 
 |       else | 
 | 	{ | 
 | 	  if (is_tu_index) | 
 | 	    { | 
 | 	      tu_count = nused; | 
 | 	      tu_sets = xcalloc2 (nused, sizeof (struct cu_tu_set)); | 
 | 	      this_set = tu_sets; | 
 | 	    } | 
 | 	  else | 
 | 	    { | 
 | 	      cu_count = nused; | 
 | 	      cu_sets = xcalloc2 (nused, sizeof (struct cu_tu_set)); | 
 | 	      this_set = cu_sets; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       if (do_display) | 
 | 	{ | 
 | 	  for (j = 0; j < ncols; j++) | 
 | 	    { | 
 | 	      unsigned char *p = ppool + j * 4; | 
 | 	      SAFE_BYTE_GET (dw_sect, p, 4, limit); | 
 | 	      printf (" %8s", get_DW_SECT_short_name (dw_sect)); | 
 | 	    } | 
 | 	  printf ("\n"); | 
 | 	} | 
 |  | 
 |       for (i = 0; i < nslots; i++) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET (signature, ph, 8, limit); | 
 |  | 
 | 	  SAFE_BYTE_GET (row, pi, 4, limit); | 
 | 	  if (row != 0) | 
 | 	    { | 
 | 	      /* PR 17531: file: a05f6ab3.  */ | 
 | 	      if (row > nused) | 
 | 		{ | 
 | 		  warn (_("Row index (%u) is larger than number of used entries (%u)\n"), | 
 | 			row, nused); | 
 | 		  return false; | 
 | 		} | 
 |  | 
 | 	      if (!do_display) | 
 | 		{ | 
 | 		  size_t num_copy = sizeof (uint64_t); | 
 |  | 
 | 		  memcpy (&this_set[row - 1].signature, ph, num_copy); | 
 | 		} | 
 |  | 
 | 	      prow = poffsets + (row - 1) * ncols * 4; | 
 | 	      if (do_display) | 
 | 		printf ("  [%3d] %#" PRIx64, i, signature); | 
 | 	      for (j = 0; j < ncols; j++) | 
 | 		{ | 
 | 		  unsigned char *p = prow + j * 4; | 
 | 		  SAFE_BYTE_GET (val, p, 4, limit); | 
 | 		  if (do_display) | 
 | 		    printf (" %8d", val); | 
 | 		  else | 
 | 		    { | 
 | 		      p = ppool + j * 4; | 
 | 		      SAFE_BYTE_GET (dw_sect, p, 4, limit); | 
 |  | 
 | 		      /* PR 17531: file: 10796eb3.  */ | 
 | 		      if (dw_sect >= DW_SECT_MAX) | 
 | 			warn (_("Overlarge Dwarf section index detected: %u\n"), dw_sect); | 
 | 		      else | 
 | 			this_set [row - 1].section_offsets [dw_sect] = val; | 
 | 		    } | 
 | 		} | 
 |  | 
 | 	      if (do_display) | 
 | 		printf ("\n"); | 
 | 	    } | 
 | 	  ph += 8; | 
 | 	  pi += 4; | 
 | 	} | 
 |  | 
 |       ph = phash; | 
 |       pi = pindex; | 
 |       if (do_display) | 
 | 	{ | 
 | 	  printf ("\n"); | 
 | 	  printf (_("  Size table\n")); | 
 | 	  printf ("  slot  %-16s  ", | 
 | 		  is_tu_index ? _("signature") : _("dwo_id")); | 
 | 	} | 
 |  | 
 |       for (j = 0; j < ncols; j++) | 
 | 	{ | 
 | 	  unsigned char *p = ppool + j * 4; | 
 | 	  SAFE_BYTE_GET (val, p, 4, limit); | 
 | 	  if (do_display) | 
 | 	    printf (" %8s", get_DW_SECT_short_name (val)); | 
 | 	} | 
 |  | 
 |       if (do_display) | 
 | 	printf ("\n"); | 
 |  | 
 |       for (i = 0; i < nslots; i++) | 
 | 	{ | 
 | 	  SAFE_BYTE_GET (signature, ph, 8, limit); | 
 |  | 
 | 	  SAFE_BYTE_GET (row, pi, 4, limit); | 
 | 	  if (row != 0) | 
 | 	    { | 
 | 	      prow = psizes + (row - 1) * ncols * 4; | 
 |  | 
 | 	      if (do_display) | 
 | 		printf ("  [%3d] %#" PRIx64, i, signature); | 
 |  | 
 | 	      for (j = 0; j < ncols; j++) | 
 | 		{ | 
 | 		  unsigned char *p = prow + j * 4; | 
 |  | 
 | 		  /* PR 28645: Check for overflow.  Since we do not know how | 
 | 		     many populated rows there will be, we cannot just | 
 | 		     perform a single check at the start of this function.  */ | 
 | 		  if (p > (limit - 4)) | 
 | 		    { | 
 | 		      if (do_display) | 
 | 			printf ("\n"); | 
 | 		      warn (_("Too many rows/columns in DWARF index section %s\n"), | 
 | 			    section->name); | 
 | 		      return false; | 
 | 		    } | 
 |  | 
 | 		  SAFE_BYTE_GET (val, p, 4, limit); | 
 |  | 
 | 		  if (do_display) | 
 | 		    printf (" %8d", val); | 
 | 		  else | 
 | 		    { | 
 | 		      p = ppool + j * 4; | 
 | 		      SAFE_BYTE_GET (dw_sect, p, 4, limit); | 
 | 		      if (dw_sect >= DW_SECT_MAX) | 
 | 			warn (_("Overlarge Dwarf section index detected: %u\n"), dw_sect); | 
 | 		      else | 
 | 			this_set [row - 1].section_sizes [dw_sect] = val; | 
 | 		    } | 
 | 		} | 
 |  | 
 | 	      if (do_display) | 
 | 		printf ("\n"); | 
 | 	    } | 
 |  | 
 | 	  ph += 8; | 
 | 	  pi += 4; | 
 | 	} | 
 |     } | 
 |   else if (do_display) | 
 |     printf (_("  Unsupported version (%d)\n"), version); | 
 |  | 
 |   if (do_display) | 
 |     printf ("\n"); | 
 |  | 
 |   return true; | 
 | } | 
 |  | 
 | static int cu_tu_indexes_read = -1; /* Tri-state variable.  */ | 
 |  | 
 | /* Load the CU and TU indexes if present.  This will build a list of | 
 |    section sets that we can use to associate a .debug_info.dwo section | 
 |    with its associated .debug_abbrev.dwo section in a .dwp file.  */ | 
 |  | 
 | static bool | 
 | load_cu_tu_indexes (void *file) | 
 | { | 
 |   /* If we have already loaded (or tried to load) the CU and TU indexes | 
 |      then do not bother to repeat the task.  */ | 
 |   if (cu_tu_indexes_read == -1) | 
 |     { | 
 |       cu_tu_indexes_read = true; | 
 |  | 
 |       if (load_debug_section_with_follow (dwp_cu_index, file)) | 
 | 	if (! process_cu_tu_index (&debug_displays [dwp_cu_index].section, 0)) | 
 | 	  cu_tu_indexes_read = false; | 
 |  | 
 |       if (load_debug_section_with_follow (dwp_tu_index, file)) | 
 | 	if (! process_cu_tu_index (&debug_displays [dwp_tu_index].section, 0)) | 
 | 	  cu_tu_indexes_read = false; | 
 |     } | 
 |  | 
 |   return (bool) cu_tu_indexes_read; | 
 | } | 
 |  | 
 | /* Find the set of sections that includes section SHNDX.  */ | 
 |  | 
 | unsigned int * | 
 | find_cu_tu_set (void *file, unsigned int shndx) | 
 | { | 
 |   unsigned int i; | 
 |  | 
 |   if (! load_cu_tu_indexes (file)) | 
 |     return NULL; | 
 |  | 
 |   /* Find SHNDX in the shndx pool.  */ | 
 |   for (i = 0; i < shndx_pool_used; i++) | 
 |     if (shndx_pool [i] == shndx) | 
 |       break; | 
 |  | 
 |   if (i >= shndx_pool_used) | 
 |     return NULL; | 
 |  | 
 |   /* Now backup to find the first entry in the set.  */ | 
 |   while (i > 0 && shndx_pool [i - 1] != 0) | 
 |     i--; | 
 |  | 
 |   return shndx_pool + i; | 
 | } | 
 |  | 
 | /* Display a .debug_cu_index or .debug_tu_index section.  */ | 
 |  | 
 | static int | 
 | display_cu_index (struct dwarf_section *section, void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   return process_cu_tu_index (section, 1); | 
 | } | 
 |  | 
 | static int | 
 | display_debug_not_supported (struct dwarf_section *section, | 
 | 			     void *file ATTRIBUTE_UNUSED) | 
 | { | 
 |   printf (_("Displaying the debug contents of section %s is not yet supported.\n"), | 
 | 	  section->name); | 
 |  | 
 |   return 1; | 
 | } | 
 |  | 
 | /* Like malloc, but takes two parameters like calloc. | 
 |    Verifies that the first parameter is not too large. | 
 |    Note: does *not* initialise the allocated memory to zero.  */ | 
 |  | 
 | void * | 
 | cmalloc (uint64_t nmemb, size_t size) | 
 | { | 
 |   /* Check for overflow.  */ | 
 |   if (nmemb >= ~(size_t) 0 / size) | 
 |     return NULL; | 
 |  | 
 |   return xmalloc (nmemb * size); | 
 | } | 
 |  | 
 | /* Like xmalloc, but takes two parameters like calloc. | 
 |    Verifies that the first parameter is not too large. | 
 |    Note: does *not* initialise the allocated memory to zero.  */ | 
 |  | 
 | void * | 
 | xcmalloc (uint64_t nmemb, size_t size) | 
 | { | 
 |   /* Check for overflow.  */ | 
 |   if (nmemb >= ~(size_t) 0 / size) | 
 |     { | 
 |       fprintf (stderr, | 
 | 	       _("Attempt to allocate an array with an excessive number of elements: %#" PRIx64 "\n"), | 
 | 	       nmemb); | 
 |       xexit (1); | 
 |     } | 
 |  | 
 |   return xmalloc (nmemb * size); | 
 | } | 
 |  | 
 | /* Like xrealloc, but takes three parameters. | 
 |    Verifies that the second parameter is not too large. | 
 |    Note: does *not* initialise any new memory to zero.  */ | 
 |  | 
 | void * | 
 | xcrealloc (void *ptr, uint64_t nmemb, size_t size) | 
 | { | 
 |   /* Check for overflow.  */ | 
 |   if (nmemb >= ~(size_t) 0 / size) | 
 |     { | 
 |       error (_("Attempt to re-allocate an array with an excessive number of elements: %#" PRIx64 "\n"), | 
 | 	     nmemb); | 
 |       xexit (1); | 
 |     } | 
 |  | 
 |   return xrealloc (ptr, nmemb * size); | 
 | } | 
 |  | 
 | /* Like xcalloc, but verifies that the first parameter is not too large.  */ | 
 |  | 
 | void * | 
 | xcalloc2 (uint64_t nmemb, size_t size) | 
 | { | 
 |   /* Check for overflow.  */ | 
 |   if (nmemb >= ~(size_t) 0 / size) | 
 |     { | 
 |       error (_("Attempt to allocate a zero'ed array with an excessive number of elements: %#" PRIx64 "\n"), | 
 | 	     nmemb); | 
 |       xexit (1); | 
 |     } | 
 |  | 
 |   return xcalloc (nmemb, size); | 
 | } | 
 |  | 
 | static unsigned long | 
 | calc_gnu_debuglink_crc32 (unsigned long crc, | 
 | 			  const unsigned char *buf, | 
 | 			  size_t len) | 
 | { | 
 |   static const unsigned long crc32_table[256] = | 
 |     { | 
 |       0x00000000, 0x77073096, 0xee0e612c, 0x990951ba, 0x076dc419, | 
 |       0x706af48f, 0xe963a535, 0x9e6495a3, 0x0edb8832, 0x79dcb8a4, | 
 |       0xe0d5e91e, 0x97d2d988, 0x09b64c2b, 0x7eb17cbd, 0xe7b82d07, | 
 |       0x90bf1d91, 0x1db71064, 0x6ab020f2, 0xf3b97148, 0x84be41de, | 
 |       0x1adad47d, 0x6ddde4eb, 0xf4d4b551, 0x83d385c7, 0x136c9856, | 
 |       0x646ba8c0, 0xfd62f97a, 0x8a65c9ec, 0x14015c4f, 0x63066cd9, | 
 |       0xfa0f3d63, 0x8d080df5, 0x3b6e20c8, 0x4c69105e, 0xd56041e4, | 
 |       0xa2677172, 0x3c03e4d1, 0x4b04d447, 0xd20d85fd, 0xa50ab56b, | 
 |       0x35b5a8fa, 0x42b2986c, 0xdbbbc9d6, 0xacbcf940, 0x32d86ce3, | 
 |       0x45df5c75, 0xdcd60dcf, 0xabd13d59, 0x26d930ac, 0x51de003a, | 
 |       0xc8d75180, 0xbfd06116, 0x21b4f4b5, 0x56b3c423, 0xcfba9599, | 
 |       0xb8bda50f, 0x2802b89e, 0x5f058808, 0xc60cd9b2, 0xb10be924, | 
 |       0x2f6f7c87, 0x58684c11, 0xc1611dab, 0xb6662d3d, 0x76dc4190, | 
 |       0x01db7106, 0x98d220bc, 0xefd5102a, 0x71b18589, 0x06b6b51f, | 
 |       0x9fbfe4a5, 0xe8b8d433, 0x7807c9a2, 0x0f00f934, 0x9609a88e, | 
 |       0xe10e9818, 0x7f6a0dbb, 0x086d3d2d, 0x91646c97, 0xe6635c01, | 
 |       0x6b6b51f4, 0x1c6c6162, 0x856530d8, 0xf262004e, 0x6c0695ed, | 
 |       0x1b01a57b, 0x8208f4c1, 0xf50fc457, 0x65b0d9c6, 0x12b7e950, | 
 |       0x8bbeb8ea, 0xfcb9887c, 0x62dd1ddf, 0x15da2d49, 0x8cd37cf3, | 
 |       0xfbd44c65, 0x4db26158, 0x3ab551ce, 0xa3bc0074, 0xd4bb30e2, | 
 |       0x4adfa541, 0x3dd895d7, 0xa4d1c46d, 0xd3d6f4fb, 0x4369e96a, | 
 |       0x346ed9fc, 0xad678846, 0xda60b8d0, 0x44042d73, 0x33031de5, | 
 |       0xaa0a4c5f, 0xdd0d7cc9, 0x5005713c, 0x270241aa, 0xbe0b1010, | 
 |       0xc90c2086, 0x5768b525, 0x206f85b3, 0xb966d409, 0xce61e49f, | 
 |       0x5edef90e, 0x29d9c998, 0xb0d09822, 0xc7d7a8b4, 0x59b33d17, | 
 |       0x2eb40d81, 0xb7bd5c3b, 0xc0ba6cad, 0xedb88320, 0x9abfb3b6, | 
 |       0x03b6e20c, 0x74b1d29a, 0xead54739, 0x9dd277af, 0x04db2615, | 
 |       0x73dc1683, 0xe3630b12, 0x94643b84, 0x0d6d6a3e, 0x7a6a5aa8, | 
 |       0xe40ecf0b, 0x9309ff9d, 0x0a00ae27, 0x7d079eb1, 0xf00f9344, | 
 |       0x8708a3d2, 0x1e01f268, 0x6906c2fe, 0xf762575d, 0x806567cb, | 
 |       0x196c3671, 0x6e6b06e7, 0xfed41b76, 0x89d32be0, 0x10da7a5a, | 
 |       0x67dd4acc, 0xf9b9df6f, 0x8ebeeff9, 0x17b7be43, 0x60b08ed5, | 
 |       0xd6d6a3e8, 0xa1d1937e, 0x38d8c2c4, 0x4fdff252, 0xd1bb67f1, | 
 |       0xa6bc5767, 0x3fb506dd, 0x48b2364b, 0xd80d2bda, 0xaf0a1b4c, | 
 |       0x36034af6, 0x41047a60, 0xdf60efc3, 0xa867df55, 0x316e8eef, | 
 |       0x4669be79, 0xcb61b38c, 0xbc66831a, 0x256fd2a0, 0x5268e236, | 
 |       0xcc0c7795, 0xbb0b4703, 0x220216b9, 0x5505262f, 0xc5ba3bbe, | 
 |       0xb2bd0b28, 0x2bb45a92, 0x5cb36a04, 0xc2d7ffa7, 0xb5d0cf31, | 
 |       0x2cd99e8b, 0x5bdeae1d, 0x9b64c2b0, 0xec63f226, 0x756aa39c, | 
 |       0x026d930a, 0x9c0906a9, 0xeb0e363f, 0x72076785, 0x05005713, | 
 |       0x95bf4a82, 0xe2b87a14, 0x7bb12bae, 0x0cb61b38, 0x92d28e9b, | 
 |       0xe5d5be0d, 0x7cdcefb7, 0x0bdbdf21, 0x86d3d2d4, 0xf1d4e242, | 
 |       0x68ddb3f8, 0x1fda836e, 0x81be16cd, 0xf6b9265b, 0x6fb077e1, | 
 |       0x18b74777, 0x88085ae6, 0xff0f6a70, 0x66063bca, 0x11010b5c, | 
 |       0x8f659eff, 0xf862ae69, 0x616bffd3, 0x166ccf45, 0xa00ae278, | 
 |       0xd70dd2ee, 0x4e048354, 0x3903b3c2, 0xa7672661, 0xd06016f7, | 
 |       0x4969474d, 0x3e6e77db, 0xaed16a4a, 0xd9d65adc, 0x40df0b66, | 
 |       0x37d83bf0, 0xa9bcae53, 0xdebb9ec5, 0x47b2cf7f, 0x30b5ffe9, | 
 |       0xbdbdf21c, 0xcabac28a, 0x53b39330, 0x24b4a3a6, 0xbad03605, | 
 |       0xcdd70693, 0x54de5729, 0x23d967bf, 0xb3667a2e, 0xc4614ab8, | 
 |       0x5d681b02, 0x2a6f2b94, 0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, | 
 |       0x2d02ef8d | 
 |     }; | 
 |   const unsigned char *end; | 
 |  | 
 |   crc = ~crc & 0xffffffff; | 
 |   for (end = buf + len; buf < end; ++ buf) | 
 |     crc = crc32_table[(crc ^ *buf) & 0xff] ^ (crc >> 8); | 
 |   return ~crc & 0xffffffff; | 
 | } | 
 |  | 
 | typedef bool (*check_func_type) (const char *, void *); | 
 | typedef const char *(* parse_func_type) (struct dwarf_section *, void *); | 
 |  | 
 | static bool | 
 | check_gnu_debuglink (const char * pathname, void * crc_pointer) | 
 | { | 
 |   static unsigned char buffer[8 * 1024]; | 
 |   FILE *f; | 
 |   size_t count; | 
 |   unsigned long crc = 0; | 
 |   void *sep_data; | 
 |  | 
 |   sep_data = open_debug_file (pathname); | 
 |   if (sep_data == NULL) | 
 |     return false; | 
 |  | 
 |   /* Yes - we are opening the file twice...  */ | 
 |   f = fopen (pathname, "rb"); | 
 |   if (f == NULL) | 
 |     { | 
 |       /* Paranoia: This should never happen.  */ | 
 |       close_debug_file (sep_data); | 
 |       warn (_("Unable to reopen separate debug info file: %s\n"), pathname); | 
 |       return false; | 
 |     } | 
 |  | 
 |   while ((count = fread (buffer, 1, sizeof (buffer), f)) > 0) | 
 |     crc = calc_gnu_debuglink_crc32 (crc, buffer, count); | 
 |  | 
 |   fclose (f); | 
 |   close_debug_file (sep_data); | 
 |  | 
 |   if (crc != * (unsigned long *) crc_pointer) | 
 |     { | 
 |       warn (_("Separate debug info file %s found, but CRC does not match - ignoring\n"), | 
 | 	    pathname); | 
 |       return false; | 
 |     } | 
 |  | 
 |   return true; | 
 | } | 
 |  | 
 | static const char * | 
 | parse_gnu_debuglink (struct dwarf_section * section, void * data) | 
 | { | 
 |   const char *     name; | 
 |   unsigned int     crc_offset; | 
 |   unsigned long *  crc32 = (unsigned long *) data; | 
 |  | 
 |   /* The name is first. | 
 |      The CRC value is stored after the filename, aligned up to 4 bytes.  */ | 
 |   name = (const char *) section->start; | 
 |  | 
 |   crc_offset = strnlen (name, section->size) + 1; | 
 |   if (crc_offset == 1) | 
 |     return NULL; | 
 |   crc_offset = (crc_offset + 3) & ~3; | 
 |   if (crc_offset + 4 > section->size) | 
 |     return NULL; | 
 |  | 
 |   * crc32 = byte_get (section->start + crc_offset, 4); | 
 |   return name; | 
 | } | 
 |  | 
 | static bool | 
 | check_gnu_debugaltlink (const char * filename, void * data ATTRIBUTE_UNUSED) | 
 | { | 
 |   void * sep_data = open_debug_file (filename); | 
 |  | 
 |   if (sep_data == NULL) | 
 |     return false; | 
 |  | 
 |   /* FIXME: We should now extract the build-id in the separate file | 
 |      and check it...  */ | 
 |  | 
 |   close_debug_file (sep_data); | 
 |   return true; | 
 | } | 
 |  | 
 | typedef struct build_id_data | 
 | { | 
 |   size_t len; | 
 |   const unsigned char *data; | 
 | } Build_id_data; | 
 |  | 
 | static const char * | 
 | parse_gnu_debugaltlink (struct dwarf_section * section, void * data) | 
 | { | 
 |   const char *name; | 
 |   size_t namelen; | 
 |   size_t id_len; | 
 |   Build_id_data *build_id_data; | 
 |  | 
 |   /* The name is first. | 
 |      The build-id follows immediately, with no padding, up to the section's end.  */ | 
 |  | 
 |   name = (const char *) section->start; | 
 |   namelen = strnlen (name, section->size) + 1; | 
 |   if (namelen == 1) | 
 |     return NULL; | 
 |   if (namelen >= section->size) | 
 |     return NULL; | 
 |  | 
 |   id_len = section->size - namelen; | 
 |   if (id_len < 0x14) | 
 |     return NULL; | 
 |  | 
 |   build_id_data = (Build_id_data *) data; | 
 |   build_id_data->len = id_len; | 
 |   build_id_data->data = section->start + namelen; | 
 |  | 
 |   return name; | 
 | } | 
 |  | 
 | static void | 
 | add_separate_debug_file (const char * filename, void * handle) | 
 | { | 
 |   separate_info * i = xmalloc (sizeof * i); | 
 |  | 
 |   i->filename = filename; | 
 |   i->handle   = handle; | 
 |   i->next     = first_separate_info; | 
 |   first_separate_info = i; | 
 | } | 
 |  | 
 | #if HAVE_LIBDEBUGINFOD | 
 | /* Query debuginfod servers for the target debuglink or debugaltlink | 
 |    file. If successful, store the path of the file in filename and | 
 |    return TRUE, otherwise return FALSE.  */ | 
 |  | 
 | static bool | 
 | debuginfod_fetch_separate_debug_info (struct dwarf_section * section, | 
 | 				      char ** filename, | 
 | 				      void * file) | 
 | { | 
 |   size_t build_id_len; | 
 |   unsigned char * build_id; | 
 |  | 
 |   if (strcmp (section->uncompressed_name, ".gnu_debuglink") == 0) | 
 |     { | 
 |       /* Get the build-id of file.  */ | 
 |       build_id = get_build_id (file); | 
 |       build_id_len = 0; | 
 |     } | 
 |   else if (strcmp (section->uncompressed_name, ".gnu_debugaltlink") == 0) | 
 |     { | 
 |       /* Get the build-id of the debugaltlink file.  */ | 
 |       unsigned int filelen; | 
 |  | 
 |       filelen = strnlen ((const char *)section->start, section->size); | 
 |       if (filelen == section->size) | 
 | 	/* Corrupt debugaltlink.  */ | 
 | 	return false; | 
 |  | 
 |       build_id = section->start + filelen + 1; | 
 |       build_id_len = section->size - (filelen + 1); | 
 |  | 
 |       if (build_id_len == 0) | 
 | 	return false; | 
 |     } | 
 |   else | 
 |     return false; | 
 |  | 
 |   if (build_id) | 
 |     { | 
 |       int fd; | 
 |       debuginfod_client * client; | 
 |  | 
 |       client = debuginfod_begin (); | 
 |       if (client == NULL) | 
 | 	return false; | 
 |  | 
 |       /* Query debuginfod servers for the target file. If found its path | 
 | 	 will be stored in filename.  */ | 
 |       fd = debuginfod_find_debuginfo (client, build_id, build_id_len, filename); | 
 |       debuginfod_end (client); | 
 |  | 
 |       /* Only free build_id if we allocated space for a hex string | 
 | 	 in get_build_id ().  */ | 
 |       if (build_id_len == 0) | 
 | 	free (build_id); | 
 |  | 
 |       if (fd >= 0) | 
 | 	{ | 
 | 	  /* File successfully retrieved. Close fd since we want to | 
 | 	     use open_debug_file () on filename instead.  */ | 
 | 	  close (fd); | 
 | 	  return true; | 
 | 	} | 
 |     } | 
 |  | 
 |   return false; | 
 | } | 
 | #endif /* HAVE_LIBDEBUGINFOD  */ | 
 |  | 
 | static void * | 
 | load_separate_debug_info (const char *            main_filename, | 
 | 			  struct dwarf_section *  xlink, | 
 | 			  parse_func_type         parse_func, | 
 | 			  check_func_type         check_func, | 
 | 			  void *                  func_data, | 
 | 			  void *                  file ATTRIBUTE_UNUSED) | 
 | { | 
 |   const char *   separate_filename; | 
 |   char *         debug_filename; | 
 |   char *         canon_dir; | 
 |   size_t         canon_dirlen; | 
 |   size_t         dirlen; | 
 |   char *         canon_filename; | 
 |   char *         canon_debug_filename; | 
 |   bool		 self; | 
 |  | 
 |   if ((separate_filename = parse_func (xlink, func_data)) == NULL) | 
 |     { | 
 |       warn (_("Corrupt debuglink section: %s\n"), | 
 | 	    xlink->name ? xlink->name : xlink->uncompressed_name); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   /* Attempt to locate the separate file. | 
 |      This should duplicate the logic in bfd/opncls.c:find_separate_debug_file().  */ | 
 |  | 
 |   canon_filename = lrealpath (main_filename); | 
 |   canon_dir = xstrdup (canon_filename); | 
 |  | 
 |   for (canon_dirlen = strlen (canon_dir); canon_dirlen > 0; canon_dirlen--) | 
 |     if (IS_DIR_SEPARATOR (canon_dir[canon_dirlen - 1])) | 
 |       break; | 
 |   canon_dir[canon_dirlen] = '\0'; | 
 |  | 
 | #ifndef DEBUGDIR | 
 | #define DEBUGDIR "/lib/debug" | 
 | #endif | 
 | #ifndef EXTRA_DEBUG_ROOT1 | 
 | #define EXTRA_DEBUG_ROOT1 "/usr/lib/debug" | 
 | #endif | 
 | #ifndef EXTRA_DEBUG_ROOT2 | 
 | #define EXTRA_DEBUG_ROOT2 "/usr/lib/debug/usr" | 
 | #endif | 
 |  | 
 |   debug_filename = (char *) malloc (strlen (DEBUGDIR) + 1 | 
 | 				    + canon_dirlen | 
 | 				    + strlen (".debug/") | 
 | #ifdef EXTRA_DEBUG_ROOT1 | 
 | 				    + strlen (EXTRA_DEBUG_ROOT1) | 
 | #endif | 
 | #ifdef EXTRA_DEBUG_ROOT2 | 
 | 				    + strlen (EXTRA_DEBUG_ROOT2) | 
 | #endif | 
 | 				    + strlen (separate_filename) | 
 | 				    + 1); | 
 |   if (debug_filename == NULL) | 
 |     { | 
 |       warn (_("Out of memory\n")); | 
 |       free (canon_dir); | 
 |       free (canon_filename); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   /* First try in the current directory.  */ | 
 |   sprintf (debug_filename, "%s", separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 |   /* Then try in a subdirectory called .debug.  */ | 
 |   sprintf (debug_filename, ".debug/%s", separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 |   /* Then try in the same directory as the original file.  */ | 
 |   sprintf (debug_filename, "%s%s", canon_dir, separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 |   /* And the .debug subdirectory of that directory.  */ | 
 |   sprintf (debug_filename, "%s.debug/%s", canon_dir, separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 | #ifdef EXTRA_DEBUG_ROOT1 | 
 |   /* Try the first extra debug file root.  */ | 
 |   sprintf (debug_filename, "%s/%s", EXTRA_DEBUG_ROOT1, separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 |   /* Try the first extra debug file root.  */ | 
 |   sprintf (debug_filename, "%s/%s/%s", EXTRA_DEBUG_ROOT1, canon_dir, separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 | #endif | 
 |  | 
 | #ifdef EXTRA_DEBUG_ROOT2 | 
 |   /* Try the second extra debug file root.  */ | 
 |   sprintf (debug_filename, "%s/%s", EXTRA_DEBUG_ROOT2, separate_filename); | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 | #endif | 
 |  | 
 |   /* Then try in the global debug_filename directory.  */ | 
 |   strcpy (debug_filename, DEBUGDIR); | 
 |   dirlen = strlen (DEBUGDIR) - 1; | 
 |   if (dirlen > 0 && DEBUGDIR[dirlen] != '/') | 
 |     strcat (debug_filename, "/"); | 
 |   strcat (debug_filename, (const char *) separate_filename); | 
 |  | 
 |   if (check_func (debug_filename, func_data)) | 
 |     goto found; | 
 |  | 
 | #if HAVE_LIBDEBUGINFOD | 
 |   { | 
 |     char * tmp_filename; | 
 |  | 
 |     if (use_debuginfod | 
 | 	&& debuginfod_fetch_separate_debug_info (xlink, | 
 | 						 & tmp_filename, | 
 | 						 file)) | 
 |       { | 
 | 	/* File successfully downloaded from server, replace | 
 | 	   debug_filename with the file's path.  */ | 
 | 	free (debug_filename); | 
 | 	debug_filename = tmp_filename; | 
 | 	goto found; | 
 |       } | 
 |   } | 
 | #endif | 
 |  | 
 |   if (do_debug_links) | 
 |     { | 
 |       /* Failed to find the file.  */ | 
 |       warn (_("could not find separate debug file '%s'\n"), | 
 | 	    separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 | #ifdef EXTRA_DEBUG_ROOT2 | 
 |       sprintf (debug_filename, "%s/%s", EXTRA_DEBUG_ROOT2, | 
 | 	       separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 | #endif | 
 |  | 
 | #ifdef EXTRA_DEBUG_ROOT1 | 
 |       sprintf (debug_filename, "%s/%s/%s", EXTRA_DEBUG_ROOT1, | 
 | 	       canon_dir, separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 |       sprintf (debug_filename, "%s/%s", EXTRA_DEBUG_ROOT1, | 
 | 	       separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 | #endif | 
 |  | 
 |       sprintf (debug_filename, "%s.debug/%s", canon_dir, | 
 | 	       separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 |       sprintf (debug_filename, "%s%s", canon_dir, separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 |       sprintf (debug_filename, ".debug/%s", separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 |       sprintf (debug_filename, "%s", separate_filename); | 
 |       warn (_("tried: %s\n"), debug_filename); | 
 |  | 
 | #if HAVE_LIBDEBUGINFOD | 
 |       if (use_debuginfod) | 
 | 	{ | 
 | 	  char *urls = getenv (DEBUGINFOD_URLS_ENV_VAR); | 
 |  | 
 | 	  if (urls == NULL) | 
 | 	    urls = ""; | 
 |  | 
 | 	  warn (_("tried: DEBUGINFOD_URLS=%s\n"), urls); | 
 | 	} | 
 | #endif | 
 |     } | 
 |  | 
 |   free (canon_dir); | 
 |   free (debug_filename); | 
 |   free (canon_filename); | 
 |   return NULL; | 
 |  | 
 |  found: | 
 |   free (canon_dir); | 
 |  | 
 |   canon_debug_filename = lrealpath (debug_filename); | 
 |   self = strcmp (canon_debug_filename, canon_filename) == 0; | 
 |   free (canon_filename); | 
 |   free (canon_debug_filename); | 
 |   if (self) | 
 |     { | 
 |       free (debug_filename); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   void * debug_handle; | 
 |  | 
 |   /* Now open the file.... */ | 
 |   if ((debug_handle = open_debug_file (debug_filename)) == NULL) | 
 |     { | 
 |       warn (_("failed to open separate debug file: %s\n"), debug_filename); | 
 |       free (debug_filename); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   /* FIXME: We do not check to see if there are any other separate debug info | 
 |      files that would also match.  */ | 
 |  | 
 |   if (do_debug_links) | 
 |     printf (_("\n%s: Found separate debug info file: %s\n"), main_filename, debug_filename); | 
 |   add_separate_debug_file (debug_filename, debug_handle); | 
 |  | 
 |   /* Do not free debug_filename - it might be referenced inside | 
 |      the structure returned by open_debug_file().  */ | 
 |   return debug_handle; | 
 | } | 
 |  | 
 | /* Attempt to load a separate dwarf object file.  */ | 
 |  | 
 | static void * | 
 | load_dwo_file (const char * main_filename, const char * name, const char * dir, const char * id ATTRIBUTE_UNUSED) | 
 | { | 
 |   char * separate_filename; | 
 |   void * separate_handle; | 
 |  | 
 |   if (IS_ABSOLUTE_PATH (name)) | 
 |     separate_filename = strdup (name); | 
 |   else | 
 |     /* FIXME: Skip adding / if dwo_dir ends in /.  */ | 
 |     separate_filename = concat (dir, "/", name, NULL); | 
 |   if (separate_filename == NULL) | 
 |     { | 
 |       warn (_("Out of memory allocating dwo filename\n")); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   if ((separate_handle = open_debug_file (separate_filename)) == NULL) | 
 |     { | 
 |       warn (_("Unable to load dwo file: %s\n"), separate_filename); | 
 |       free (separate_filename); | 
 |       return NULL; | 
 |     } | 
 |  | 
 |   /* FIXME: We should check the dwo_id.  */ | 
 |  | 
 |   printf (_("%s: Found separate debug object file: %s\n\n"), main_filename, separate_filename); | 
 |  | 
 |   add_separate_debug_file (separate_filename, separate_handle); | 
 |   /* Note - separate_filename will be freed in free_debug_memory().  */ | 
 |   return separate_handle; | 
 | } | 
 |  | 
 | static void * | 
 | try_build_id_prefix (const char * prefix, char * filename, const unsigned char * data, unsigned long id_len) | 
 | { | 
 |   char * f = filename; | 
 |  | 
 |   f += sprintf (f, "%s.build-id/%02x/", prefix, (unsigned) *data++); | 
 |   id_len --; | 
 |   while (id_len --) | 
 |     f += sprintf (f, "%02x", (unsigned) *data++); | 
 |   strcpy (f, ".debug"); | 
 |  | 
 |   return open_debug_file (filename); | 
 | } | 
 |  | 
 | /* Try to load a debug file based upon the build-id held in the .note.gnu.build-id section.  */ | 
 |  | 
 | static void | 
 | load_build_id_debug_file (const char * main_filename ATTRIBUTE_UNUSED, void * main_file) | 
 | { | 
 |   if (! load_debug_section (note_gnu_build_id, main_file)) | 
 |     return; /* No .note.gnu.build-id section.  */ | 
 |  | 
 |   struct dwarf_section * section = & debug_displays [note_gnu_build_id].section; | 
 |   if (section == NULL) | 
 |     { | 
 |       warn (_("Unable to load the .note.gnu.build-id section\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   if (section->start == NULL || section->size < 0x18) | 
 |     { | 
 |       warn (_(".note.gnu.build-id section is corrupt/empty\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   /* In theory we should extract the contents of the section into | 
 |      a note structure and then check the fields.  For now though | 
 |      just use hard coded offsets instead: | 
 |  | 
 |        Field  Bytes    Contents | 
 | 	NSize  0...3   4 | 
 | 	DSize  4...7   8+ | 
 | 	Type   8..11   3  (NT_GNU_BUILD_ID) | 
 | 	Name   12.15   GNU\0 | 
 | 	Data   16....   */ | 
 |  | 
 |   /* FIXME: Check the name size, name and type fields.  */ | 
 |  | 
 |   unsigned long build_id_size; | 
 |   build_id_size = byte_get (section->start + 4, 4); | 
 |   if (build_id_size < 8) | 
 |     { | 
 |       warn (_(".note.gnu.build-id data size is too small\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   if (build_id_size > (section->size - 16)) | 
 |     { | 
 |       warn (_(".note.gnu.build-id data size is too big\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   char * filename; | 
 |   filename = xmalloc (strlen (".build-id/") | 
 | 		      + build_id_size * 2 + 2 | 
 | 		      + strlen (".debug") | 
 | 		      /* The next string should be the same as the longest | 
 | 			 name found in the prefixes[] array below.  */ | 
 | 		      + strlen ("/usr/lib64/debug/usr/") | 
 | 		      + 1); | 
 |   void * handle; | 
 |  | 
 |   static const char * prefixes[] = | 
 |     { | 
 |       "", | 
 |       ".debug/", | 
 |       "/usr/lib/debug/", | 
 |       "/usr/lib/debug/usr/", | 
 |       "/usr/lib64/debug/", | 
 |       "/usr/lib64/debug/usr/" | 
 |     }; | 
 |   long unsigned int i; | 
 |  | 
 |   for (i = 0; i < ARRAY_SIZE (prefixes); i++) | 
 |     { | 
 |       handle = try_build_id_prefix (prefixes[i], filename, | 
 | 				    section->start + 16, build_id_size); | 
 |       if (handle != NULL) | 
 | 	break; | 
 |     } | 
 |   /* FIXME: TYhe BFD library also tries a global debugfile directory prefix.  */ | 
 |   if (handle == NULL) | 
 |     { | 
 |       /* Failed to find a debug file associated with the build-id. | 
 | 	 This is not an error however, rather it just means that | 
 | 	 the debug info has probably not been loaded on the system, | 
 | 	 or that another method is being used to link to the debug | 
 | 	 info.  */ | 
 |       free (filename); | 
 |       return; | 
 |     } | 
 |  | 
 |   add_separate_debug_file (filename, handle); | 
 | } | 
 |  | 
 | /* Try to load a debug file pointed to by the .debug_sup section.  */ | 
 |  | 
 | static void | 
 | load_debug_sup_file (const char * main_filename, void * file) | 
 | { | 
 |   if (! load_debug_section (debug_sup, file)) | 
 |     return; /* No .debug_sup section.  */ | 
 |  | 
 |   struct dwarf_section * section; | 
 |   section = & debug_displays [debug_sup].section; | 
 |   assert (section != NULL); | 
 |  | 
 |   if (section->start == NULL || section->size < 5) | 
 |     { | 
 |       warn (_(".debug_sup section is corrupt/empty\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   if (section->start[2] != 0) | 
 |     return; /* This is a supplementary file.  */ | 
 |  | 
 |   const char * filename = (const char *) section->start + 3; | 
 |   if (strnlen (filename, section->size - 3) == section->size - 3) | 
 |     { | 
 |       warn (_("filename in .debug_sup section is corrupt\n")); | 
 |       return; | 
 |     } | 
 |  | 
 |   if (filename[0] != '/' && strchr (main_filename, '/')) | 
 |     filename = xasprintf ("%.*s/%s", | 
 | 			  (int) (strrchr (main_filename, '/') - main_filename), | 
 | 			  main_filename, | 
 | 			  filename); | 
 |   else | 
 |     /* PR 27796: Make sure that we pass a filename that can be free'd to | 
 |        add_separate_debug_file().  */ | 
 |     filename = xstrdup (filename); | 
 |  | 
 |   void * handle = open_debug_file (filename); | 
 |   if (handle == NULL) | 
 |     { | 
 |       warn (_("unable to open file '%s' referenced from .debug_sup section\n"), filename); | 
 |       free ((void *) filename); | 
 |       return; | 
 |     } | 
 |  | 
 |   printf (_("%s: Found supplementary debug file: %s\n\n"), main_filename, filename); | 
 |  | 
 |   /* FIXME: Compare the checksums, if present.  */ | 
 |   add_separate_debug_file (filename, handle); | 
 | } | 
 |  | 
 | /* Load a debuglink section and/or a debugaltlink section, if either are present. | 
 |    Recursively check the loaded files for more of these sections. | 
 |    Also follow any links in .debug_sup sections. | 
 |    FIXME: Should also check for DWO_* entries in the newly loaded files.  */ | 
 |  | 
 | static void | 
 | check_for_and_load_links (void * file, const char * filename) | 
 | { | 
 |   void * handle = NULL; | 
 |  | 
 |   if (load_debug_section (gnu_debugaltlink, file)) | 
 |     { | 
 |       Build_id_data build_id_data; | 
 |  | 
 |       handle = load_separate_debug_info (filename, | 
 | 					 & debug_displays[gnu_debugaltlink].section, | 
 | 					 parse_gnu_debugaltlink, | 
 | 					 check_gnu_debugaltlink, | 
 | 					 & build_id_data, | 
 | 					 file); | 
 |       if (handle) | 
 | 	{ | 
 | 	  assert (handle == first_separate_info->handle); | 
 | 	  check_for_and_load_links (first_separate_info->handle, | 
 | 				    first_separate_info->filename); | 
 | 	} | 
 |     } | 
 |  | 
 |   if (load_debug_section (gnu_debuglink, file)) | 
 |     { | 
 |       unsigned long crc32; | 
 |  | 
 |       handle = load_separate_debug_info (filename, | 
 | 					 & debug_displays[gnu_debuglink].section, | 
 | 					 parse_gnu_debuglink, | 
 | 					 check_gnu_debuglink, | 
 | 					 & crc32, | 
 | 					 file); | 
 |       if (handle) | 
 | 	{ | 
 | 	  assert (handle == first_separate_info->handle); | 
 | 	  check_for_and_load_links (first_separate_info->handle, | 
 | 				    first_separate_info->filename); | 
 | 	} | 
 |     } | 
 |  | 
 |   load_debug_sup_file (filename, file); | 
 |  | 
 |   load_build_id_debug_file (filename, file); | 
 | } | 
 |  | 
 | /* Load the separate debug info file(s) attached to FILE, if any exist. | 
 |    Returns TRUE if any were found, FALSE otherwise. | 
 |    If TRUE is returned then the linked list starting at first_separate_info | 
 |    will be populated with open file handles.  */ | 
 |  | 
 | bool | 
 | load_separate_debug_files (void * file, const char * filename) | 
 | { | 
 |   /* Skip this operation if we are not interested in debug links.  */ | 
 |   if (! do_follow_links && ! do_debug_links) | 
 |     return false; | 
 |  | 
 |   /* See if there are any dwo links.  */ | 
 |   if (load_debug_section (str, file) | 
 |       && load_debug_section (abbrev, file) | 
 |       && load_debug_section (info, file)) | 
 |     { | 
 |       /* Load the .debug_addr section, if it exists.  */ | 
 |       load_debug_section (debug_addr, file); | 
 |       /* Load the .debug_str_offsets section, if it exists.  */ | 
 |       load_debug_section (str_index, file); | 
 |       /* Load the .debug_loclists section, if it exists.  */ | 
 |       load_debug_section (loclists, file); | 
 |       /* Load the .debug_rnglists section, if it exists.  */ | 
 |       load_debug_section (rnglists, file); | 
 |  | 
 |       free_dwo_info (); | 
 |  | 
 |       if (process_debug_info (& debug_displays[info].section, file, abbrev, | 
 | 			      true, false)) | 
 | 	{ | 
 | 	  bool introduced = false; | 
 | 	  dwo_info *dwinfo; | 
 | 	  const char *dir = NULL; | 
 | 	  const char *id = NULL; | 
 | 	  const char *name = NULL; | 
 |  | 
 | 	  for (dwinfo = first_dwo_info; dwinfo != NULL; dwinfo = dwinfo->next) | 
 | 	    { | 
 | 	      /* Accumulate NAME, DIR and ID fields.  */ | 
 | 	      switch (dwinfo->type) | 
 | 		{ | 
 | 		case DWO_NAME: | 
 | 		  if (name != NULL) | 
 | 		    warn (_("Multiple DWO_NAMEs encountered for the same CU\n")); | 
 | 		  name = dwinfo->value; | 
 | 		  break; | 
 |  | 
 | 		case DWO_DIR: | 
 | 		  /* There can be multiple DW_AT_comp_dir entries in a CU, | 
 | 		     so do not complain.  */ | 
 | 		  dir = dwinfo->value; | 
 | 		  break; | 
 |  | 
 | 		case DWO_ID: | 
 | 		  if (id != NULL) | 
 | 		    warn (_("multiple DWO_IDs encountered for the same CU\n")); | 
 | 		  id = dwinfo->value; | 
 | 		  break; | 
 |  | 
 | 		default: | 
 | 		  error (_("Unexpected DWO INFO type")); | 
 | 		  break; | 
 | 		} | 
 |  | 
 | 	      /* If we have reached the end of our list, or we are changing | 
 | 		 CUs, then display the information that we have accumulated | 
 | 		 so far.  */ | 
 | 	      if (name != NULL | 
 | 		  && (dwinfo->next == NULL | 
 | 		      || dwinfo->next->cu_offset != dwinfo->cu_offset)) | 
 | 		{ | 
 | 		  if (do_debug_links) | 
 | 		    { | 
 | 		      if (! introduced) | 
 | 			{ | 
 | 			  printf (_("The %s section contains link(s) to dwo file(s):\n\n"), | 
 | 				  debug_displays [info].section.uncompressed_name); | 
 | 			  introduced = true; | 
 | 			} | 
 |  | 
 | 		      printf (_("  Name:      %s\n"), name); | 
 | 		      printf (_("  Directory: %s\n"), dir ? dir : _("<not-found>")); | 
 | 		      if (id != NULL) | 
 | 			display_data (printf (_("  ID:       ")), (unsigned char *) id, 8); | 
 | 		      else if (debug_information[0].dwarf_version != 5) | 
 | 			printf (_("  ID:        <not specified>\n")); | 
 | 		      printf ("\n\n"); | 
 | 		    } | 
 |  | 
 | 		  if (do_follow_links) | 
 | 		    load_dwo_file (filename, name, dir, id); | 
 |  | 
 | 		  name = dir = id = NULL; | 
 | 		} | 
 | 	    } | 
 | 	} | 
 |     } | 
 |  | 
 |   if (! do_follow_links) | 
 |     /* The other debug links will be displayed by display_debug_links() | 
 |        so we do not need to do any further processing here.  */ | 
 |     return false; | 
 |  | 
 |   /* FIXME: We do not check for the presence of both link sections in the same file.  */ | 
 |   /* FIXME: We do not check for the presence of multiple, same-name debuglink sections.  */ | 
 |   /* FIXME: We do not check for the presence of a dwo link as well as a debuglink.  */ | 
 |  | 
 |   check_for_and_load_links (file, filename); | 
 |   if (first_separate_info != NULL) | 
 |     return true; | 
 |  | 
 |   do_follow_links = 0; | 
 |   return false; | 
 | } | 
 |  | 
 | void | 
 | free_debug_memory (void) | 
 | { | 
 |   unsigned int i; | 
 |  | 
 |   free_all_abbrevs (); | 
 |  | 
 |   free (shndx_pool); | 
 |   shndx_pool = NULL; | 
 |   shndx_pool_size = 0; | 
 |   shndx_pool_used = 0; | 
 |   free (cu_sets); | 
 |   cu_sets = NULL; | 
 |   cu_count = 0; | 
 |   free (tu_sets); | 
 |   tu_sets = NULL; | 
 |   tu_count = 0; | 
 |  | 
 |   memset (level_type_signed, 0, sizeof level_type_signed); | 
 |   cu_tu_indexes_read = -1; | 
 |  | 
 |   for (i = 0; i < max; i++) | 
 |     free_debug_section ((enum dwarf_section_display_enum) i); | 
 |  | 
 |   if (debug_information != NULL) | 
 |     { | 
 |       for (i = 0; i < alloc_num_debug_info_entries; i++) | 
 | 	free_debug_information (&debug_information[i]); | 
 |       free (debug_information); | 
 |       debug_information = NULL; | 
 |       alloc_num_debug_info_entries = num_debug_info_entries = 0; | 
 |     } | 
 |  | 
 |   separate_info * d; | 
 |   separate_info * next; | 
 |  | 
 |   for (d = first_separate_info; d != NULL; d = next) | 
 |     { | 
 |       close_debug_file (d->handle); | 
 |       free ((void *) d->filename); | 
 |       next = d->next; | 
 |       free ((void *) d); | 
 |     } | 
 |   first_separate_info = NULL; | 
 |  | 
 |   free_dwo_info (); | 
 | } | 
 |  | 
 | typedef struct | 
 | { | 
 |   const char letter; | 
 |   const char *option; | 
 |   int *variable; | 
 |   int val; | 
 | } debug_dump_long_opts; | 
 |  | 
 | static const debug_dump_long_opts debug_option_table[] = | 
 | { | 
 |   { 'A', "addr", &do_debug_addr, 1 }, | 
 |   { 'a', "abbrev", &do_debug_abbrevs, 1 }, | 
 |   { 'c', "cu_index", &do_debug_cu_index, 1 }, | 
 | #ifdef HAVE_LIBDEBUGINFOD | 
 |   { 'D', "use-debuginfod", &use_debuginfod, 1 }, | 
 |   { 'E', "do-not-use-debuginfod", &use_debuginfod, 0 }, | 
 | #endif | 
 |   { 'F', "frames-interp", &do_debug_frames_interp, 1 }, | 
 |   { 'f', "frames", &do_debug_frames, 1 }, | 
 |   { 'g', "gdb_index", &do_gdb_index, 1 }, | 
 |   { 'i', "info", &do_debug_info, 1 }, | 
 |   { 'K', "follow-links", &do_follow_links, 1 }, | 
 |   { 'k', "links", &do_debug_links, 1 }, | 
 |   { 'L', "decodedline", &do_debug_lines, FLAG_DEBUG_LINES_DECODED }, | 
 |   { 'l', "rawline", &do_debug_lines, FLAG_DEBUG_LINES_RAW }, | 
 |   /* For compatibility with earlier versions of readelf.  */ | 
 |   { 'l', "line", &do_debug_lines, FLAG_DEBUG_LINES_RAW }, | 
 |   { 'm', "macro", &do_debug_macinfo, 1 }, | 
 |   { 'N', "no-follow-links", &do_follow_links, 0 }, | 
 |   { 'O', "str-offsets", &do_debug_str_offsets, 1 }, | 
 |   { 'o', "loc", &do_debug_loc, 1 }, | 
 |   { 'p', "pubnames", &do_debug_pubnames, 1 }, | 
 |   { 'R', "Ranges", &do_debug_ranges, 1 }, | 
 |   { 'r', "aranges", &do_debug_aranges, 1 }, | 
 |   /* For compatibility with earlier versions of readelf.  */ | 
 |   { 'r', "ranges", &do_debug_aranges, 1 }, | 
 |   { 's', "str", &do_debug_str, 1 }, | 
 |   { '\0', "sframe-internal-only", &do_sframe, 1 }, | 
 |   { 'T', "trace_aranges", &do_trace_aranges, 1 }, | 
 |   { 't', "pubtypes", &do_debug_pubtypes, 1 }, | 
 |   { 'U', "trace_info", &do_trace_info, 1 }, | 
 |   { 'u', "trace_abbrev", &do_trace_abbrevs, 1 }, | 
 |   { 0, NULL, NULL, 0 } | 
 | }; | 
 |  | 
 | /* Enable display of specific DWARF sections as determined by the comma | 
 |    separated strings in NAMES.  Returns non-zero if any displaying was | 
 |    enabled.  */ | 
 |  | 
 | int | 
 | dwarf_select_sections_by_names (const char *names) | 
 | { | 
 |   const char *p; | 
 |   int result = 0; | 
 |  | 
 |   p = names; | 
 |   while (*p) | 
 |     { | 
 |       const debug_dump_long_opts *entry; | 
 |  | 
 |       for (entry = debug_option_table; entry->option; entry++) | 
 | 	{ | 
 | 	  size_t len = strlen (entry->option); | 
 |  | 
 | 	  if (strncmp (p, entry->option, len) == 0 | 
 | 	      && (p[len] == ',' || p[len] == '\0')) | 
 | 	    { | 
 | 	      if (entry->val == 0) | 
 | 		* entry->variable = 0; | 
 | 	      else | 
 | 		* entry->variable = entry->val; | 
 | 	      result |= entry->val; | 
 |  | 
 | 	      p += len; | 
 | 	      break; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       if (entry->option == NULL) | 
 | 	{ | 
 | 	  warn (_("Unrecognized debug option '%s'\n"), p); | 
 | 	  p = strchr (p, ','); | 
 | 	  if (p == NULL) | 
 | 	    break; | 
 | 	} | 
 |  | 
 |       if (*p == ',') | 
 | 	p++; | 
 |     } | 
 |  | 
 |   /* The --debug-dump=frames-interp option also enables the | 
 |      --debug-dump=frames option.  */ | 
 |   if (do_debug_frames_interp) | 
 |     do_debug_frames = 1; | 
 |  | 
 |   return result; | 
 | } | 
 |  | 
 | /* Enable display of specific DWARF sections as determined by the characters | 
 |    in LETTERS.  Returns non-zero if any displaying was enabled.  */ | 
 |  | 
 | int | 
 | dwarf_select_sections_by_letters (const char *letters) | 
 | { | 
 |   int result = 0; | 
 |  | 
 |   while (* letters) | 
 |     { | 
 |       const debug_dump_long_opts *entry; | 
 |  | 
 |       for (entry = debug_option_table; entry->letter; entry++) | 
 | 	{ | 
 | 	  if (entry->letter == * letters) | 
 | 	    { | 
 | 	      if (entry->val == 0) | 
 | 		* entry->variable = 0; | 
 | 	      else | 
 | 		* entry->variable |= entry->val; | 
 | 	      result |= entry->val; | 
 | 	      break; | 
 | 	    } | 
 | 	} | 
 |  | 
 |       if (entry->letter == 0) | 
 | 	warn (_("Unrecognized debug letter option '%c'\n"), * letters); | 
 |  | 
 |       letters ++; | 
 |     } | 
 |  | 
 |   /* The --debug-dump=frames-interp option also enables the | 
 |      --debug-dump=frames option.  */ | 
 |   if (do_debug_frames_interp) | 
 |     do_debug_frames = 1; | 
 |  | 
 |   return result; | 
 | } | 
 |  | 
 | void | 
 | dwarf_select_sections_all (void) | 
 | { | 
 |   do_debug_info = 1; | 
 |   do_debug_abbrevs = 1; | 
 |   do_debug_lines = FLAG_DEBUG_LINES_RAW; | 
 |   do_debug_pubnames = 1; | 
 |   do_debug_pubtypes = 1; | 
 |   do_debug_aranges = 1; | 
 |   do_debug_ranges = 1; | 
 |   do_debug_frames = 1; | 
 |   do_debug_macinfo = 1; | 
 |   do_debug_str = 1; | 
 |   do_debug_loc = 1; | 
 |   do_gdb_index = 1; | 
 |   do_trace_info = 1; | 
 |   do_trace_abbrevs = 1; | 
 |   do_trace_aranges = 1; | 
 |   do_debug_addr = 1; | 
 |   do_debug_cu_index = 1; | 
 |   do_follow_links = 1; | 
 |   do_debug_links = 1; | 
 |   do_debug_str_offsets = 1; | 
 | } | 
 |  | 
 | #define NO_ABBREVS   NULL, NULL, NULL, 0, 0, 0, NULL, 0 | 
 | #define ABBREV(N)    NULL, NULL, NULL, 0, 0, N, NULL, 0 | 
 |  | 
 | /* N.B. The order here must match the order in section_display_enum.  */ | 
 |  | 
 | struct dwarf_section_display debug_displays[] = | 
 | { | 
 |   { { ".debug_abbrev",	    ".zdebug_abbrev",	     ".dwabrev", NO_ABBREVS },	    display_debug_abbrev,   &do_debug_abbrevs,	false }, | 
 |   { { ".debug_aranges",	    ".zdebug_aranges",	     ".dwarnge", NO_ABBREVS },	    display_debug_aranges,  &do_debug_aranges,	true }, | 
 |   { { ".debug_frame",	    ".zdebug_frame",	     ".dwframe", NO_ABBREVS },	    display_debug_frames,   &do_debug_frames,	true }, | 
 |   { { ".debug_info",	    ".zdebug_info",	     ".dwinfo",	 ABBREV (abbrev)},  display_debug_info,	    &do_debug_info,	true }, | 
 |   { { ".debug_line",	    ".zdebug_line",	     ".dwline",	 NO_ABBREVS },	    display_debug_lines,    &do_debug_lines,	true }, | 
 |   { { ".debug_pubnames",    ".zdebug_pubnames",	     ".dwpbnms", NO_ABBREVS },	    display_debug_pubnames, &do_debug_pubnames, false }, | 
 |   { { ".debug_gnu_pubnames", ".zdebug_gnu_pubnames", "",	 NO_ABBREVS },	    display_debug_gnu_pubnames, &do_debug_pubnames, false }, | 
 |   { { ".eh_frame",	    "",			     "",	 NO_ABBREVS },	    display_debug_frames,   &do_debug_frames,	true }, | 
 |   { { ".eh_frame_hdr",	    "",			     "",	 NO_ABBREVS },	    display_eh_frame_hdr,   &do_debug_frames,	true }, | 
 |   { { ".debug_macinfo",	    ".zdebug_macinfo",	     "",	 NO_ABBREVS },	    display_debug_macinfo,  &do_debug_macinfo,	false }, | 
 |   { { ".debug_macro",	    ".zdebug_macro",	     ".dwmac",	 NO_ABBREVS },	    display_debug_macro,    &do_debug_macinfo,	true }, | 
 |   { { ".debug_str",	    ".zdebug_str",	     ".dwstr",	 NO_ABBREVS },	    display_debug_str,	    &do_debug_str,	false }, | 
 |   { { ".debug_line_str",    ".zdebug_line_str",	     "",	 NO_ABBREVS },	    display_debug_str,	    &do_debug_str,	false }, | 
 |   { { ".debug_loc",	    ".zdebug_loc",	     ".dwloc",	 NO_ABBREVS },	    display_debug_loc,	    &do_debug_loc,	true }, | 
 |   { { ".debug_loclists",    ".zdebug_loclists",	     "",	 NO_ABBREVS },	    display_debug_loc,	    &do_debug_loc,	true }, | 
 |   { { ".debug_loclists.dwo", ".zdebug_loclists.dwo", "",         NO_ABBREVS },      display_debug_loc,      &do_debug_loc,      true }, | 
 |   { { ".debug_pubtypes",    ".zdebug_pubtypes",	     ".dwpbtyp", NO_ABBREVS },	    display_debug_pubnames, &do_debug_pubtypes, false }, | 
 |   { { ".debug_gnu_pubtypes", ".zdebug_gnu_pubtypes", "",	 NO_ABBREVS },	    display_debug_gnu_pubnames, &do_debug_pubtypes, false }, | 
 |   { { ".debug_ranges",	    ".zdebug_ranges",	     ".dwrnges", NO_ABBREVS },	    display_debug_ranges,   &do_debug_ranges,	true }, | 
 |   { { ".debug_rnglists",    ".zdebug_rnglists",	     "",	 NO_ABBREVS },	    display_debug_ranges,   &do_debug_ranges,	true }, | 
 |   { { ".debug_rnglists.dwo", ".zdebug_rnglists.dwo", "",         NO_ABBREVS },      display_debug_ranges,   &do_debug_ranges,   true }, | 
 |   { { ".debug_static_func", ".zdebug_static_func",   "",	 NO_ABBREVS },	    display_debug_not_supported, NULL,		false }, | 
 |   { { ".debug_static_vars", ".zdebug_static_vars",   "",	 NO_ABBREVS },	    display_debug_not_supported, NULL,		false }, | 
 |   { { ".debug_types",	    ".zdebug_types",	     "",	 ABBREV (abbrev) }, display_debug_types,    &do_debug_info,	true }, | 
 |   { { ".debug_weaknames",   ".zdebug_weaknames",     "",	 NO_ABBREVS },	    display_debug_not_supported, NULL,		false }, | 
 |   { { ".gdb_index",	    "",			     "",	 NO_ABBREVS },	    display_gdb_index,	    &do_gdb_index,	false }, | 
 |   { { ".debug_names",	    "",			     "",	 NO_ABBREVS },	    display_debug_names,    &do_gdb_index,	false }, | 
 |   { { ".sframe",	    "",			     "",	 NO_ABBREVS },	    display_sframe,	    &do_sframe,		true }, | 
 |   { { ".trace_info",	    "",			     "",	 ABBREV (trace_abbrev) }, display_trace_info, &do_trace_info,	true }, | 
 |   { { ".trace_abbrev",	    "",			     "",	 NO_ABBREVS },	    display_debug_abbrev,   &do_trace_abbrevs,	false }, | 
 |   { { ".trace_aranges",	    "",			     "",	 NO_ABBREVS },	    display_debug_aranges,  &do_trace_aranges,	false }, | 
 |   { { ".debug_info.dwo",    ".zdebug_info.dwo",	     "",	 ABBREV (abbrev_dwo) }, display_debug_info, &do_debug_info,	true }, | 
 |   { { ".debug_abbrev.dwo",  ".zdebug_abbrev.dwo",    "",	 NO_ABBREVS },	  display_debug_abbrev,	    &do_debug_abbrevs,	false }, | 
 |   { { ".debug_types.dwo",   ".zdebug_types.dwo",     "",	 ABBREV (abbrev_dwo) }, display_debug_types, &do_debug_info,	true }, | 
 |   { { ".debug_line.dwo",    ".zdebug_line.dwo",	     "",	 NO_ABBREVS },	    display_debug_lines,    &do_debug_lines,	true }, | 
 |   { { ".debug_loc.dwo",	    ".zdebug_loc.dwo",	     "",	 NO_ABBREVS },	    display_debug_loc,	    &do_debug_loc,	true }, | 
 |   { { ".debug_macro.dwo",   ".zdebug_macro.dwo",     "",	 NO_ABBREVS },	    display_debug_macro,    &do_debug_macinfo,	true }, | 
 |   { { ".debug_macinfo.dwo", ".zdebug_macinfo.dwo",   "",	 NO_ABBREVS },	    display_debug_macinfo,  &do_debug_macinfo,	false }, | 
 |   { { ".debug_str.dwo",	    ".zdebug_str.dwo",	     "",	 NO_ABBREVS },	    display_debug_str,	    &do_debug_str,	true }, | 
 |   { { ".debug_str_offsets", ".zdebug_str_offsets",   "",	 NO_ABBREVS },	    display_debug_str_offsets, &do_debug_str_offsets, true }, | 
 |   { { ".debug_str_offsets.dwo", ".zdebug_str_offsets.dwo", "",	 NO_ABBREVS },	    display_debug_str_offsets, &do_debug_str_offsets, true }, | 
 |   { { ".debug_addr",	    ".zdebug_addr",	     "",	 NO_ABBREVS },	    display_debug_addr,	    &do_debug_addr,	true }, | 
 |   { { ".debug_cu_index",    "",			     "",	 NO_ABBREVS },	    display_cu_index,	    &do_debug_cu_index, false }, | 
 |   { { ".debug_tu_index",    "",			     "",	 NO_ABBREVS },	    display_cu_index,	    &do_debug_cu_index, false }, | 
 |   { { ".gnu_debuglink",	    "",			     "",	 NO_ABBREVS },	    display_debug_links,    &do_debug_links,	false }, | 
 |   { { ".gnu_debugaltlink",  "",			     "",	 NO_ABBREVS },	    display_debug_links,    &do_debug_links,	false }, | 
 |   { { ".debug_sup",	    "",			     "",	 NO_ABBREVS },	    display_debug_sup,	    &do_debug_links,	false }, | 
 |   /* Separate debug info files can containt their own .debug_str section, | 
 |      and this might be in *addition* to a .debug_str section already present | 
 |      in the main file.	Hence we need to have two entries for .debug_str.  */ | 
 |   { { ".debug_str",	    ".zdebug_str",	     "",	 NO_ABBREVS },	    display_debug_str,	    &do_debug_str,	false }, | 
 |   { { ".note.gnu.build-id", "",                      "",	 NO_ABBREVS },	    display_debug_not_supported, NULL,		false }, | 
 | }; | 
 |  | 
 | /* A static assertion.  */ | 
 | extern int debug_displays_assert[ARRAY_SIZE (debug_displays) == max ? 1 : -1]; |