Hi, Shortly speaking, we've found out that actual version of libunwind returns broken stack for PostgreSQL processes. But fix is already in upstream (not yet released): https://github.com/libunwind/libunwind/commit/400b3f819ad44ff4e15487b163cc3613389cb4c8 Details. We're trying to use libunwind to get stacktrace in case of PostgreSQL crash. Sometimes stack looks hang on wrong position like: 1 libthr.so.3 + 0x1a4b0 0x0008009ba58e 0x00007fffffffc130 _pthread_sigmask + 0x54e 2 libthr.so.3 + 0x19a00 0x0008009b9b3f 0x00007fffffffc4f0 _pthread_setschedparam + 0x82f 3 [vdso] + 0x2d0 0x7ffffffff8a3 0x00007fffffffc560 + 0xffffffff 4 libc.so.7 + 0x138570 0x00080143457a 0x00007fffffffc560 __sys_kevent + 0xa 5 libthr.so.3 + 0x19870 0x0008009b9a00 0x00007fffffffc568 _pthread_setschedparam + 0x6f0 6 <unknown> 0x000000000000 0x00007fffffffc570 + 0xffffffff 7 libthr.so.3 + 0x1ccc0 0x0008009bcd13 0x00007fffffffc578 _pthread_getspecific + 0x1903 8 libthr.so.3 + 0x1ccc0 0x0008009bcd13 0x56415741e5894865 _pthread_getspecific + 0x1903 9 libthr.so.3 + 0x1ccc0 0x0008009bcd13 0x56415741e5894865 _pthread_getspecific + 0x1903 10 libthr.so.3 + 0x1ccc0 0x0008009bcd13 0x56415741e5894865 _pthread_getspecific + 0x1903 The problem isn't reproduced on latest commit of libunwind. Git-bisect pointed to commit: https://github.com/libunwind/libunwind/commit/400b3f819ad44ff4e15487b163cc3613389cb4c8 On this commit stack looks good in all test invocations: 1 libthr.so.3 + 0x1a4b0 0x0008009ba58e 0x00007fffffffc130 _pthread_sigmask + 0x54e 2 libthr.so.3 + 0x19a00 0x0008009b9b3f 0x00007fffffffc4f0 _pthread_setschedparam + 0x82f 3 [vdso] + 0x2d0 0x7ffffffff8a3 0x00007fffffffc560 + 0xffffffff 4 libc.so.7 + 0x138570 0x00080143457a 0x00007fffffffc9b8 __sys_kevent + 0xa 5 libthr.so.3 + 0x1ccc0 0x0008009bcd13 0x00007fffffffc9c0 _pthread_getspecific + 0x1903 6 postgres + 0x7acc00 0x0000007acd44 0x00007fffffffca10 WaitEventSetWait + 0x144 7 postgres + 0x7ac790 0x0000007ac8d6 0x00007fffffffcac0 WaitLatchOrSocket + 0x146 8 postgres + 0x746170 0x000000746284 0x00007fffffffcb20 pgstat_start + 0x1b4 9 postgres + 0x7460d0 0x000000746129 0x00007fffffffd350 pgstat_start + 0x59 10 postgres + 0x74f2e0 0x00000074f81b 0x00007fffffffd370 InitProcessGlobals + 0x129b 11 libthr.so.3 + 0x1a4b0 0x0008009ba580 0x00007fffffffd7e0 _pthread_sigmask + 0x540 12 libthr.so.3 + 0x19a00 0x0008009b9b3f 0x00007fffffffdba0 _pthread_setschedparam + 0x82f 13 [vdso] + 0x2d0 0x7ffffffff8a3 0x00007fffffffdc10 + 0xffffffff 14 libc.so.7 + 0x138730 0x00080143473a 0x00007fffffffe068 __sys_select + 0xa 15 libthr.so.3 + 0x1c9c0 0x0008009bca02 0x00007fffffffe070 _pthread_getspecific + 0x15f2 16 postgres + 0x7504b0 0x000000750ac4 0x00007fffffffe0b0 InitProcessGlobals + 0x2544 17 postgres + 0x74d6a0 0x00000074e371 0x00007fffffffe610 PostmasterMain + 0xcd1 18 postgres + 0x6ad8d0 0x0000006adbac 0x00007fffffffea70 main + 0x2dc 19 postgres + 0x4b4100 0x0000004b4100 0x00007fffffffeaa0 _start + 0x100 20 <unknown> 0x000800981008 0x00007fffffffeae0 + 0xffffffff It's hard to provide simple reproducible case without compiling PostgreSQL and extensions. Also worth mentioning that issue is reproduced only on 64-bit FreeBSD, neither on Linux distributions (Debian/Ubuntu) nor on 32-bit FreeBSD. What is best way to solve it? To cherry-pick patch to port? I'll appreciate your help. Thanks, Michael.
Sorry for late reply. I think it's OK to cherry-pick this commit to fix the issue before 1.7.0 release.
Committed. Thanks!
A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=42ca033358a793bf186420844535e97df3000df6 commit 42ca033358a793bf186420844535e97df3000df6 Author: Po-Chuan Hsieh <sunpoet@FreeBSD.org> AuthorDate: 2023-04-09 23:38:47 +0000 Commit: Po-Chuan Hsieh <sunpoet@FreeBSD.org> CommitDate: 2023-04-09 23:38:47 +0000 devel/libunwind: Stop aliasing RSP and CFA - Bump PORTREVISION for package change PR: 269875 Obtained from: https://github.com/libunwind/libunwind/commit/400b3f819ad44ff4e15487b163cc3613389cb4c8 Reported by: mizhka devel/libunwind/Makefile | 2 +- devel/libunwind/files/patch-bug-269875 (new) | 310 +++++++++++++++++++++++++++ 2 files changed, 311 insertions(+), 1 deletion(-)
Many thanks! Now it works fine (at least for me)