Re: [rfc/patch] mm/slub: restore/expand unfreeze_partials() local exclusion scope

[Mike Galbraith]

On Tue, 2021-07-20 at 13:26 +0200, Mike Galbraith wrote:
> On Tue, 2021-07-20 at 10:56 +0200, Vlastimil Babka wrote:
> > > crash> bt -sx
> > > PID: 18761  TASK: ffff88812fff0000  CPU: 0   COMMAND: "hackbench"
> > >  #0 [ffff88818f8ff980] machine_kexec+0x14f at ffffffff81051c8f
> > >  #1 [ffff88818f8ff9c8] __crash_kexec+0xd2 at ffffffff8111ef72
> > >  #2 [ffff88818f8ffa88] crash_kexec+0x30 at ffffffff8111fd10
> > >  #3 [ffff88818f8ffa98] oops_end+0xd3 at ffffffff810267e3
> > >  #4 [ffff88818f8ffab8] exc_general_protection+0x195 at
> > > ffffffff8179fdb5
> > >  #5 [ffff88818f8ffb50] asm_exc_general_protection+0x1e at
> > > ffffffff81800a0e
> > >     [exception RIP: __unfreeze_partials+156]
> >
> > Hm going back to this report...
> > So could it be that it was stillput_cpu_partial() preempting
> > __slab_alloc() messing the partial list, but for some reason the
> > put_cpu_partial() side crashed this time?
>
> Thinking this bug is toast, I emptied the trash bin, so no can peek.

I made fireworks while waiting for bike riding time, boom #10 was
finally the right flavor, but...

crash> bt -sx
PID: 32     TASK: ffff888100a56000  CPU: 3   COMMAND: "rcuc/3"
 #0 [ffff888100aa7a90] machine_kexec+0x14f at ffffffff81051c8f
 #1 [ffff888100aa7ad8] __crash_kexec+0xd2 at ffffffff81120612
 #2 [ffff888100aa7b98] crash_kexec+0x30 at ffffffff811213b0
 #3 [ffff888100aa7ba8] oops_end+0xd3 at ffffffff810267e3
 #4 [ffff888100aa7bc8] exc_general_protection+0x195 at ffffffff817a2cc5
 #5 [ffff888100aa7c60] asm_exc_general_protection+0x1e at ffffffff81800a0e
    [exception RIP: __unfreeze_partials+149]
    RIP: ffffffff8124a295  RSP: ffff888100aa7d10  RFLAGS: 00010202
    RAX: 0000000000190016  RBX: 0000000000190016  RCX: 000000017fffffff
    RDX: 00000001ffffffff  RSI: 0000000000000023  RDI: ffffffff81e58b10
    RBP: ffff888100aa7da0   R8: 0000000000000000   R9: 0000000000190018
    R10: ffff888100aa7db8  R11: 000000000002d9e4  R12: ffff888100190500
    R13: ffff88810018c980  R14: 00000001ffffffff  R15: ffffea0004571588
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
 #6 [ffff888100aa7db0] put_cpu_partial+0x8e at ffffffff8124a56e
 #7 [ffff888100aa7dd0] kmem_cache_free+0x3a8 at ffffffff8124d238
 #8 [ffff888100aa7e08] rcu_do_batch+0x186 at ffffffff810eb246
 #9 [ffff888100aa7e70] rcu_core+0x25f at ffffffff810eeb2f
#10 [ffff888100aa7eb0] rcu_cpu_kthread+0x94 at ffffffff810eed24
#11 [ffff888100aa7ee0] smpboot_thread_fn+0x249 at ffffffff8109e559
#12 [ffff888100aa7f18] kthread+0x1ac at ffffffff810984dc
#13 [ffff888100aa7f50] ret_from_fork+0x1f at ffffffff81001b1f
crash> runq
...
CPU 3 RUNQUEUE: ffff88840ece9980
  CURRENT: PID: 32     TASK: ffff888100a56000  COMMAND: "rcuc/3"
  RT PRIO_ARRAY: ffff88840ece9bc0
     [ 94] PID: 32     TASK: ffff888100a56000  COMMAND: "rcuc/3"
  CFS RB_ROOT: ffff88840ece9a40
     [120] PID: 33     TASK: ffff888100a51000  COMMAND: "ksoftirqd/3"
...
crash> bt -sx 33
PID: 33     TASK: ffff888100a51000  CPU: 3   COMMAND: "ksoftirqd/3"
 #0 [ffff888100aabdf0] __schedule+0x2d7 at ffffffff817ad3a7
 #1 [ffff888100aabec8] schedule+0x3b at ffffffff817ae4eb
 #2 [ffff888100aabee0] smpboot_thread_fn+0x18c at ffffffff8109e49c
 #3 [ffff888100aabf18] kthread+0x1ac at ffffffff810984dc
 #4 [ffff888100aabf50] ret_from_fork+0x1f at ffffffff81001b1f
crash>