Re: [PATCH] padata: do not leak refcount in reorder_work

[Dominik Grzegorzek <dominik.grzegorzek@oracle.com>]

On Thu, 2025-05-22 at 14:10 +0100, Aishwarya wrote:
> Hi Dominik,
> 
> I wanted to report a regression observed while running the
> `kselftest-mm` suite, specifically the
> `mm_run_vmtests_sh_migration_migration_shared_anon` test, on an
> Arm64 Marvell Thunder X2 (TX2) system.
> 
> The kernel was built using defconfig with the additional config
> fragment from:
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/
> testing/selftests/mm/config
> 
> This works fine on v6.15-rc7.
> 
> A bisect identified this patch as introducing the failure. Bisected
> it on the tag "v6.15-rc7-7-g4a95bc121ccd" at repo:
> git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> 
> Failure log:
> 11193 03:29:14.806502  # # running ./migration
> 11194 03:29:14.806876  # # -------------------
> 11195 03:29:14.820939  # # TAP version 13
> 11196 03:29:14.821236  # # 1..6
> 11197 03:29:14.821519  # # # Starting 6 tests from 1 test cases.
> 11198 03:29:14.821773  # # #  RUN           migration.private_anon ...
> 11199 03:29:34.602964  # # #            OK  migration.private_anon
> 11200 03:29:34.603418  # # ok 1 migration.private_anon
> 11201 03:29:34.603687  # # #  RUN           migration.shared_anon ...
> 11202 03:29:34.973479  # # Didn't migrate 1 pages
> 11203 03:29:34.973855  # # # migration.c:175:shared_anon:Expected migrate(ptr,
> self->n1, self->n2) (-2) == 0 (0)
> 11204 03:29:34.984787  # # # shared_anon: Test terminated by assertion
> 11205 03:29:34.985105  # # #          FAIL  migration.shared_anon
> 11206 03:29:34.985365  # # not ok 2 migration.shared_anon
> 11207 03:29:34.988568  # # #  RUN           migration.private_anon_thp ...
> 11208 03:29:54.597572  # # #            OK  migration.private_anon_thp
> 11209 03:29:54.597951  # # ok 3 migration.private_anon_thp
> 11210 03:29:54.598487  # # #  RUN           migration.shared_anon_thp ...
> 11211 03:29:55.011183  # # Didn't migrate 1 pages
> 11212 03:29:55.011524  # # # migration.c:241:shared_anon_thp:Expected
> migrate(ptr,
> self->n1, self->n2) (-2) == 0 (0)
> 11213 03:29:55.022519  # # # shared_anon_thp: Test terminated by assertion
> 11214 03:29:55.022834  # # #          FAIL  migration.shared_anon_thp
> 11215 03:29:55.027864  # # not ok 4 migration.shared_anon_thp
> 11216 03:29:55.028156  # # #  RUN           migration.private_anon_htlb ...
> 11217 03:30:14.595327  # # #            OK  migration.private_anon_htlb
> 11218 03:30:14.595777  # # ok 5 migration.private_anon_htlb
> 11219 03:30:14.596398  # # #  RUN           migration.shared_anon_htlb ...
> 11220 03:30:34.595239  # # #            OK  migration.shared_anon_htlb
> 11221 03:30:34.595623  # # ok 6 migration.shared_anon_htlb
> 11222 03:30:34.595859  # # # FAILED: 4 / 6 tests passed.
> 11223 03:30:34.603816  # # # Totals: pass:4 fail:2 xfail:0 xpass:0 skip:0
> error:0
> 11224 03:30:34.604110  # # [FAIL]
> 11225 03:30:34.604342  # not ok 55 migration # exit=1
> 
> Thanks,
> Aishwarya

Hi, 

Looking at the test, I don't think this is related. The test allocates some
pages and attempts to migrate them between two NUMA nodes. It seems to be a
selftest for memory management code, and I don't see how this patch could affect
its outcome.

Do you see this failure consistently, or is it only happening occasionally?
I'm wondering if it might have passed by chance when testing the real culprit
during the bisect.

Regards, 
Dominik