[Bug gdb/31254] [gdb, arm] FAIL: gdb.threads/staticthreads.exp: up 10

["thiago.bauermann at linaro dot org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=31254

--- Comment #17 from Thiago Jung Bauermann <thiago.bauermann at linaro dot org> ---
Created attachment 15347
  --> https://sourceware.org/bugzilla/attachment.cgi?id=15347&action=edit
memcheck output of corruption problem

Valgrind pointed out what the problem is. I'm attaching the full output (the
line numbers may be a bit off due to some debug statements I added), but here
are the highlights:

As I mentioned in the previous comment, arm_find_exidx_entry () returns an
entry containing garbage. Then arm_exidx_fill_cache () reads it:

==3763513== Invalid read of size 1
==3763513==    at 0x3DE8CC: arm_exidx_fill_cache(frame_info_ptr, unsigned
char*) (arm-tdep.c:2865)
==3763513==    by 0x3DFFE1: arm_exidx_unwind_sniffer(frame_unwind const*,
frame_info_ptr, void**) (arm-tdep.c:3277)
==3763513==    by 0x6317ED: frame_unwind_try_unwinder(frame_info_ptr, void**,
frame_unwind const*) (frame-unwind.c:138)

The problem is that arm_exidx_new_objfile () stores the unwind instructions in
the objfile's obstack but it is freed before arm_exidx_fill_cache () has a
chance to use it:

==3763513==  Address 0xa128e10 is 608 bytes inside a block of size 4,072 free'd
==3763513==    at 0x4867A68: free (vg_replace_malloc.c:872)
==3763513==    by 0x374A3F: void xfree<void>(void*) (gdb-xfree.h:37)
==3763513==    by 0x518A351: obstack_free (obstack.c:357)
==3763513==    by 0x9A3B45: reread_symbols(int) (symfile.c:2587)
==3763513==    by 0x6C18CF: run_command_1(char const*, int, run_how)
(infcmd.c:398)
==3763513==    by 0x6C1CBB: run_command(char const*, int) (infcmd.c:512)

This happens because reread_symbols () detects that the staticthreads
executable has changed since it was loaded (I don't know why, though) and thus
reinitializes the objfile obstacks.

I'll see if I can make arm-tdep.c add a hook in objfiles_changed () so that it
can detect that the objfiles where reloaded and re-read the exception tables.
Or perhaps it's cleaner to convert objfiles_changed () into an observer.

-- 
You are receiving this mail because:
You are on the CC list for the bug.