[Bug gcov-profile/97461] [11 Regression] allocate_gcov_kvp() deadlocks in firefox LTO+PGO build (overridden malloc() recursion)

["marxin at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97461

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |---

--- Comment #14 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Sergei Trofimovich from comment #13)
> Tried firefox-82 with LTO+PGO today on gcc-11-4428-g4a369d199bf. It gets a
> lot more forward, but still gets stuck. This time on
> free()->allocate_gcov_kvp ()->alloc() deadloc.

Hmm, that's quite unpleasant. Anyway, thanks for testing that.

> 
> The backtrace:
> 
> (gdb) bt
> #0  __lll_lock_wait (futex=0x7ffff7800018, private=0) at lowlevellock.c:52
> #1  0x00007ffff7f69305 in __GI___pthread_mutex_lock (mutex=0x7ffff7800018)
> at ../nptl/pthread_mutex_lock.c:135
> #2  0x00005555555b2151 in arena_t::MallocSmall(unsigned long, bool) ()
> #3  0x00005555555b307c in arena_t::Malloc(unsigned long, bool) ()
> #4  0x00005555555b41f2 in calloc ()
> #5  0x00005555556a49ab in allocate_gcov_kvp () at
> /var/tmp/portage/sys-devel/gcc-11.0.0_pre9999/work/gcc-11.0.0_pre9999/libgcc/
> libgcov.h:431
> #6  gcov_topn_add_value (count=1, use_atomic=1, increment_total=1,
> value=496, counters=0x5555557158c0
> <__gcov3._ZL20arena_run_reg_dallocP11arena_run_tP11arena_bin_tPvm>)
>     at
> /var/tmp/portage/sys-devel/gcc-11.0.0_pre9999/work/gcc-11.0.0_pre9999/libgcc/
> libgcov.h:477
> #7  __gcov_topn_values_profiler_body (use_atomic=1, value=496,
> counters=0x5555557158c0
> <__gcov3._ZL20arena_run_reg_dallocP11arena_run_tP11arena_bin_tPvm>)
>     at
> /var/tmp/portage/sys-devel/gcc-11.0.0_pre9999/work/gcc-11.0.0_pre9999/libgcc/
> libgcov-profiler.c:103
> #8  __gcov_topn_values_profiler_atomic (counters=0x5555557158c0
> <__gcov3._ZL20arena_run_reg_dallocP11arena_run_tP11arena_bin_tPvm>,
> value=496)
>     at
> /var/tmp/portage/sys-devel/gcc-11.0.0_pre9999/work/gcc-11.0.0_pre9999/libgcc/
> libgcov-profiler.c:128
> #9  0x00005555555b1065 in arena_t::DallocSmall(arena_chunk_t*, void*,
> arena_chunk_map_t*) ()
> #10 0x00005555555b166c in free ()
> #11 0x00007fffe1b0ef1a in CollectProcessInfo(ProcessInfo&) () at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #12 0x00007fffea0afa72 in PreRecordMetaInformation() () at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #13 0x00007fffea0b28c8 in profiler_shutdown(IsFastShutdown) () at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #14 0x00007fffea7c6aa0 in XREMain::XRE_main(int, char**,
> mozilla::BootstrapConfig const&) ()
>     at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #15 0x00007fffea7c7543 in XRE_main(int, char**, mozilla::BootstrapConfig
> const&) () at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #16 0x00007fffea7cfbd0 in mozilla::BootstrapImpl::XRE_main(int, char**,
> mozilla::BootstrapConfig const&) ()
>     at
> /home/slyfox/tmp/portage/www-client/firefox-82.0/work/firefox_build/
> instrumented/dist/firefox/libxul.so
> #17 0x000055555559a370 in do_main(int, char**, char**) ()
> #18 0x00005555555980a1 in main ()

Can you please run in gdb how many times is allocate_gcov_kvp called before
we reach the deadlock?

> 
> Worth filing a new bug, or this would be good enough?

We can stay with this PR as it contains useful discussion.

Anyway, one another fix can be usage of the suggested
__gcov_supress_malloc/__gcov_allow_malloc
that will be used in an alternative memory allocator..