]> git.ipfire.org Git - thirdparty/postgresql.git/commit
Silence valgrind about pg_numa_touch_mem_if_required
authorTomas Vondra <tomas.vondra@postgresql.org>
Tue, 1 Jul 2025 10:32:23 +0000 (12:32 +0200)
committerTomas Vondra <tomas.vondra@postgresql.org>
Tue, 1 Jul 2025 10:32:23 +0000 (12:32 +0200)
commit81f287dc923f565722f46b18d71969926bc3c64f
tree8e8e88a70156a04a144865a7fb5d4a9c49c57a39
parent953050236ab2640055d79532981f958271a33292
Silence valgrind about pg_numa_touch_mem_if_required

When querying NUMA status of pages in shared memory, we need to touch
the memory first to get valid results. This may trigger valgrind
reports, because some of the memory (e.g. unpinned buffers) may be
marked as noaccess.

Solved by adding a valgrind suppresion. An alternative would be to
adjust the access/noaccess status before touching the memory, but that
seems far too invasive. It would require all those places to have
detailed knowledge of what the shared memory stores.

The pg_numa_touch_mem_if_required() macro is replaced with a function.
Macros are invisible to suppressions, so it'd have to suppress reports
for the caller - e.g. pg_get_shmem_allocations_numa(). So we'd suppress
reports for the whole function, and that seems to heavy-handed. It might
easily hide other valid issues.

Reviewed-by: Christoph Berg <myon@debian.org>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/aEtDozLmtZddARdB@msg.df7cb.de
Backpatch-through: 18
contrib/pg_buffercache/pg_buffercache_pages.c
src/backend/storage/ipc/shmem.c
src/include/port/pg_numa.h
src/tools/valgrind.supp