diff mbox series

[v2,PR,symtab/32658] Fix parsing .debug_aranges section for MIPS signed addresses

Message ID jea59631ge.fsf@localhost
State New
Headers
Series [v2,PR,symtab/32658] Fix parsing .debug_aranges section for MIPS signed addresses |

Checks

Context Check Description
linaro-tcwg-bot/tcwg_gdb_build--master-aarch64 success Build passed
linaro-tcwg-bot/tcwg_gdb_build--master-arm success Build passed
linaro-tcwg-bot/tcwg_gdb_check--master-aarch64 fail Test failed
linaro-tcwg-bot/tcwg_gdb_check--master-arm fail Test failed

Commit Message

Martin Simmons March 27, 2025, 4:03 p.m. UTC 
  >>>>> On Wed, 19 Mar 2025 10:48:09 -0600, Tom Tromey said:
> 
>>>>> "Martin" == Martin Simmons <qqxnjvamvxwx@dyxyl.com> writes:
> 
> Martin> This patch fixes https://sourceware.org/bugzilla/show_bug.cgi?id=32658 by sign
> 
> There should be a 'Bug: ...' trailer in the comment with this link.
> 
> FWIW I see now that other spots in the DWARF reader do use the BFD
> approach:
> 
>   int signed_addr_p = bfd_get_sign_extend_vma (objfile->obfd.get ());
> 
> Martin> I made it explicitly sign extend the addresses after they are read
> Martin> (instead of using extract_signed_integer) because I don't know if the
> Martin> address_size in the section will always be equal to the gdb arch_size.
> 
> I don't think this should be a concern, unless you have an example where
> it actually happens.
> 
> The .debug_aranges header spec explicitly says "size of an address in
> bytes on the target architecture".

I've not found an example (and both gcc and clang explicitly write them
using the target's pointer size).

Here is v2 of the patch addressing the above.


Fix parsing .debug_aranges section for signed addresses.

Some architectures, such as MIPS, have signed addresses and this changes
read_addrmap_from_aranges to record them as signed when required.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=32658

Comments

Tom Tromey April 3, 2025, 3:03 p.m. UTC | #1 
>>>>> "Martin" == Martin Simmons <qqxnjvamvxwx@dyxyl.com> writes:

Martin> I've not found an example (and both gcc and clang explicitly write them
Martin> using the target's pointer size).

Martin> Here is v2 of the patch addressing the above.

Thanks.  I'll check it in shortly.

Tom
diff mbox series

Patch

diff --git a/gdb/dwarf2/aranges.c b/gdb/dwarf2/aranges.c
index 7016eee557c..ac530abd6b3 100644
--- a/gdb/dwarf2/aranges.c
+++ b/gdb/dwarf2/aranges.c
@@ -59,6 +59,7 @@  read_addrmap_from_aranges (dwarf2_per_objfile *per_objfile,
 
   gdb::unordered_set<sect_offset> debug_info_offset_seen;
   const bfd_endian dwarf5_byte_order = gdbarch_byte_order (gdbarch);
+  const int signed_addr_p = bfd_get_sign_extend_vma (abfd);
   const gdb_byte *addr = section->buffer;
   while (addr < section->buffer + section->size)
     {
@@ -167,8 +168,13 @@  read_addrmap_from_aranges (dwarf2_per_objfile *per_objfile,
 			  plongest (entry_addr - section->buffer));
 	      return false;
 	    }
-	  ULONGEST start = extract_unsigned_integer (addr, address_size,
-						     dwarf5_byte_order);
+	  ULONGEST start;
+	  if (signed_addr_p)
+	    start = extract_signed_integer (addr, address_size,
+					    dwarf5_byte_order);
+	  else
+	    start = extract_unsigned_integer (addr, address_size,
+					      dwarf5_byte_order);
 	  addr += address_size;
 	  ULONGEST length = extract_unsigned_integer (addr, address_size,
 						      dwarf5_byte_order);