Re: [PATCH] sched/wait: Fix a kthread_park race with wait_woken()

[Peter Zijlstra <peterz@infradead.org>]

On Thu, May 11, 2023 at 05:52:24PM -0700, John Stultz wrote:
> On Thu, May 11, 2023 at 3:31 PM Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > On Thu, May 11, 2023 at 09:41:30PM +0000, John Stultz wrote:
> > > From: Arve Hjønnevåg <arve@android.com>
> > >
> > > kthread_park and wait_woken have a similar race that kthread_stop and
> > > wait_woken used to have before it was fixed in
> > > cb6538e740d7543cd989128625cf8cac4b471e0a. Extend that fix to also cover
> >
> >   cb6538e740d7 ("sched/wait: Fix a kthread race with wait_woken()")
> >
> > > kthread_park.
> > >
> > > Cc: Ingo Molnar <mingo@redhat.com>
> > > Cc: Peter Zijlstra <peterz@infradead.org>
> > > Cc: Juri Lelli <juri.lelli@redhat.com>
> > > Cc: Vincent Guittot <vincent.guittot@linaro.org>
> > > Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> > > Cc: Steven Rostedt <rostedt@goodmis.org>
> > > Cc: Ben Segall <bsegall@google.com>
> > > Cc: Mel Gorman <mgorman@suse.de>
> > > Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
> > > Cc: Valentin Schneider <vschneid@redhat.com>
> > > Signed-off-by: Arve Hjønnevåg <arve@android.com>
> > > Signed-off-by: John Stultz <jstultz@google.com>
> > > ---
> > > This seemingly slipped by, so I wanted to resend it
> > > for review.
> > > ---
> > >  kernel/sched/wait.c | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/kernel/sched/wait.c b/kernel/sched/wait.c
> > > index 133b74730738..a9cf49da884b 100644
> > > --- a/kernel/sched/wait.c
> > > +++ b/kernel/sched/wait.c
> > > @@ -425,9 +425,9 @@ int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, i
> > >  }
> > >  EXPORT_SYMBOL(autoremove_wake_function);
> > >
> > > -static inline bool is_kthread_should_stop(void)
> > > +static inline bool is_kthread_should_stop_or_park(void)
> > >  {
> > > -     return (current->flags & PF_KTHREAD) && kthread_should_stop();
> > > +     return (current->flags & PF_KTHREAD) && (kthread_should_stop() || kthread_should_park());
> > >  }
> > >
> > >  /*
> >
> > That's a bit sad; that two function calls for checking two consecutive
> > bits in the same word :-(
> >
> > If we move this to kthread.c and write it like:
> >
> >         kthread = __to_kthread(current);
> >         if (!kthread)
> >                 return false;
> >
> >         return test_bit(KTHREAD_SHOULD_STOP, &kthread->flags) ||
> >                test_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
> >
> > Then the compiler should be able to merge the two bits in a single load
> > and test due to constant_test_bit() -- do check though.
> 
> Hrm. Apologies, as it's been awhile since I've looked at disassembled
> asm, so I may be wrong, but I don't think that's happening here.
> 
> With the logic above I'm seeing it build as:
> 0000000000000a50 <kthread_should_stop_or_park>:
>      a50:       65 48 8b 14 25 00 00    mov    %gs:0x0,%rdx
>      a57:       00 00
>      a59:       48 8b 8a 78 0a 00 00    mov    0xa78(%rdx),%rcx
>      a60:       31 c0                   xor    %eax,%eax
>      a62:       48 85 c9                test   %rcx,%rcx
>      a65:       74 11                   je     a78  <kthread_should_stop_or_park+0x28>
>      a67:       f6 42 2e 20             testb  $0x20,0x2e(%rdx)
>      a6b:       74 0b                   je     a78  <kthread_should_stop_or_park+0x28>
>      a6d:       48 8b 01                mov    (%rcx),%rax
>      a70:       48 d1 e8                shr    %rax
>      a73:       83 e0 01                and    $0x1,%eax
>      a76:       74 05                   je     a7d  <kthread_should_stop_or_park+0x2d>
>      a78:       e9 00 00 00 00          jmp    a7d  <kthread_should_stop_or_park+0x2d>
>      a7d:       48 8b 01                mov    (%rcx),%rax
>      a80:       48 c1 e8 02             shr    $0x2,%rax
>      a84:       83 e0 01                and    $0x1,%eax
>      a87:       e9 00 00 00 00          jmp    a8c  <kthread_should_stop_or_park+0x3c>
>      a8c:       0f 1f 40 00             nopl   0x0(%rax)

Moo, where is that optimization pass when you need it ;-)

If we forgo test_bit() and write it plainly it seems to work:

diff --git a/kernel/kthread.c b/kernel/kthread.c
index 7e6751b29101..36f94616cae5 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -164,6 +164,15 @@ bool __kthread_should_park(struct task_struct *k)
 }
 EXPORT_SYMBOL_GPL(__kthread_should_park);
 
+bool kthread_should_stop_or_park(void)
+{
+	struct kthread *kthread = __to_kthread(current);
+	if (!kthread)
+		return false;
+
+	return kthread->flags & (BIT(KTHREAD_SHOULD_STOP) | BIT(KTHREAD_SHOULD_PARK));
+}
+
 /**
  * kthread_should_park - should this kthread park now?
  *


0000000000001850 <kthread_should_stop_or_park>:
    1850:       f3 0f 1e fa             endbr64
    1854:       65 48 8b 04 25 00 00 00 00      mov    %gs:0x0,%rax     1859: R_X86_64_32S      pcpu_hot
    185d:       48 8b 88 08 06 00 00    mov    0x608(%rax),%rcx
    1864:       31 d2                   xor    %edx,%edx
    1866:       48 85 c9                test   %rcx,%rcx
    1869:       74 0c                   je     1877 <kthread_should_stop_or_park+0x27>
    186b:       f6 40 2e 20             testb  $0x20,0x2e(%rax)
    186f:       74 06                   je     1877 <kthread_should_stop_or_park+0x27>
    1871:       f6 01 06                testb  $0x6,(%rcx)
    1874:       0f 95 c2                setne  %dl
    1877:       89 d0                   mov    %edx,%eax
    1879:       e9 00 00 00 00          jmp    187e <kthread_should_stop_or_park+0x2e>  187a: R_X86_64_PLT32    __x86_return_thunk-0x4