Re: mm: ksm: WARNING: CPU: 3 PID: 22593 at mm/ksm.c:715 remove_stable_node+0xc7/0xf0()

[Hugh Dickins]

On Thu, 17 Sep 2015, Sasha Levin wrote:

> Hi all,
> 
> I've observed the following warning while fuzzing with trinity inside a KVM tools
> guest running -next:
> 
> [1385507.811807] Out of memory (oom_kill_allocating_task): Kill process 22593 (ksm04) score 0 or sacrifice child
> [1385507.815277] Killed process 22612 (ksm04) total-vm:139476kB, anon-rss:131204kB, file-rss:896kB
> [1385507.821799] Out of memory (oom_kill_allocating_task): Kill process 22593 (ksm04) score 0 or sacrifice child
> [1385507.823082] Killed process 22613 (ksm04) total-vm:139476kB, anon-rss:131204kB, file-rss:896kB
> [1385508.569555] Out of memory (oom_kill_allocating_task): Kill process 22593 (ksm04) score 0 or sacrifice child
> [1385508.574114] Killed process 22614 (ksm04) total-vm:139476kB, anon-rss:131204kB, file-rss:896kB
> [1385508.589529] Out of memory (oom_kill_allocating_task): Kill process 22593 (ksm04) score 0 or sacrifice child
> [1385508.591203] Killed process 22593 (ksm04) total-vm:8408kB, anon-rss:148kB, file-rss:1508kB
> [1385509.046298] ------------[ cut here ]------------
> [1385509.047136] WARNING: CPU: 3 PID: 22593 at mm/ksm.c:715 remove_stable_node+0xc7/0xf0()
> [1385509.048308] Modules linked in:
> [1385509.069698] CPU: 3 PID: 22593 Comm: ksm04 Not tainted 4.3.0-rc1-next-20150914-sasha-00043-geddd763-dirty #2557
> [1385509.072158]  ffffffffa4750740 ffff8803f60bfab0 ffffffff9bf8bc6a 0000000000000000
> [1385509.073347]  ffff8803f60bfaf0 ffffffff9a369096 ffffffff9a7bd007 ffffea000042a5c0
> [1385509.074533]  00000000fffffff0 0000000000000000 0000000000000000 ffffffffaaa16540
> [1385509.075700] Call Trace:
> [1385509.076152] dump_stack (lib/dump_stack.c:52)
> [1385509.076971] warn_slowpath_common (kernel/panic.c:448)
> [1385509.078823] warn_slowpath_null (kernel/panic.c:482)
> [1385509.079700] remove_stable_node (mm/ksm.c:715 (discriminator 3))
> [1385509.080565] remove_all_stable_nodes (mm/ksm.c:751)
> [1385509.081515] run_store (include/linux/oom.h:65 mm/ksm.c:2162)
> [1385509.089983] kobj_attr_store (lib/kobject.c:780)
> [1385509.092142] sysfs_kf_write (fs/sysfs/file.c:131)
> [1385509.093899] kernfs_fop_write (fs/kernfs/file.c:312)
> [1385509.094793] __vfs_write (fs/read_write.c:487)
> [1385509.101787] vfs_write (fs/read_write.c:539)
> [1385509.102575] SyS_write (fs/read_write.c:586 fs/read_write.c:577)
> [1385509.106273] tracesys_phase2 (arch/x86/entry/entry_64.S:273)
> [1385509.188672] ---[ end trace 66cda70045475cf9 ]---

Thanks for the report, Sasha.

My first inclination is to be lazy, and take this as confirmation
that the WARN_ON_ONCE was indeed just the right thing to do.  There
are some holes in undoing KSM, but nobody's data is in danger from
them, there's no unbounded leak, and no user has yet complained
(useful though it is, trinity doesn't quite count as a user).

But...

I've been assuming you're just doing something new with trinity,
turning it towards KSM and now hitting this.  But perhaps that's
not so, and you're doing the same as before, but -next contains
a change somewhere (not in ksm.c) which is in conflict with KSM's
assumptions.  Please reassure me that this is not the case, that
you can reproduce this warning on older kernels.

Thanks,
Hugh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/