On Wed, Jul 17, 2019 at 01:26:29PM +0200, Alexander Bluhm wrote: > I got a strange panic on my daily amd64 regress machine > reordering libraries:panic: pool_do_get: scxspl free list modified: page
I see more strange effects on my regress machines for a while now. The SSH connection that controls my tests fails with broken pipe. packet_write_wait: Connection to 10.0.1.26 port 22: Broken pipe This is my excerpt of the previous two weeks. I never see it on i386, that architecture is stable for my use case. arch GENERIC.MP aborted amd64 Fri Jun 28 10:59:55 MDT 2019 ok arm64 Fri Jun 28 12:09:23 MDT 2019 ok amd64 Thu Jul 4 21:30:05 MDT 2019 ok arm64 Thu Jul 4 23:42:09 MDT 2019 ok amd64 Sat Jul 6 00:15:43 MDT 2019 broken pipe arm64 Sat Jul 6 02:28:11 MDT 2019 broken pipe amd64 Sat Jul 6 19:30:27 MDT 2019 ok arm64 Sat Jul 6 21:42:52 MDT 2019 broken pipe amd64 Sun Jul 7 22:21:36 MDT 2019 broken pipe arm64 Mon Jul 8 00:34:07 MDT 2019 broken pipe amd64 Mon Jul 8 11:54:20 MDT 2019 ok arm64 Tue Jul 9 03:12:08 MDT 2019 broken pipe amd64 Wed Jul 10 00:20:15 MDT 2019 ok arm64 Wed Jul 10 02:33:40 MDT 2019 ok amd64 Wed Jul 10 23:35:33 MDT 2019 broken pipe arm64 Thu Jul 11 04:32:25 MDT 2019 ok amd64 Thu Jul 11 15:44:36 MDT 2019 broken pipe arm64 Thu Jul 11 17:56:31 MDT 2019 ok amd64 Fri Jul 12 09:59:59 MDT 2019 broken pipe arm64 Fri Jul 12 12:12:19 MDT 2019 ok amd64 Fri Jul 12 09:59:59 MDT 2019 ok arm64 Fri Jul 12 12:12:19 MDT 2019 ok amd64 Fri Jul 12 09:59:59 MDT 2019 broken pipe arm64 Fri Jul 12 12:12:19 MDT 2019 broken pipe amd64 Mon Jul 15 18:54:34 MDT 2019 broken pipe arm64 Mon Jul 15 21:07:06 MDT 2019 ok amd64 Mon Jul 15 18:54:34 MDT 2019 scxspl free list modified Sometimes the test before the broken pipe has an unusual result. Sun Jul 7 22:21:36 MDT 2019 malloc_duel(65986) in free(): chunk canary corrupted 0x873508a1770 0x4@0x4 Tue Jul 9 03:12:08 MDT 2019 ln: fstest_fc3618a74baf021f5226c3cb938e91f0/32767: Too many links Fri Jul 12 09:59:59 MDT 2019 malloc_duel(22164) in free(): chunk canary corrupted 0x4fd491216a0 0x4@0x4 (double free?) Fri Jul 12 12:12:19 MDT 2019 ln: fstest_c7e1a3c857443619516751530ca0a85e/32767: Too many links It is not reproduceable, the broken pipe happens at different regress tests. But there are some regress tests where it is more likely, during the first one it happens quite often, the last two hit it only once. regress/sbin/iked/parser regress/lib/libpthread/malloc_duel regress/sys/ffs/ffs regress/sys/ffs/nfs regress/lib/libc/stdio_threading/fread bluhm
