Re: [peterz-queue:sched/core] [sched/fair] 420356c350: WARNING:at_kernel/sched/core.c:#__might_sleep

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Aug 16, 2024 at 05:15:12PM +0800, kernel test robot wrote:
> kernel test robot noticed "WARNING:at_kernel/sched/core.c:#__might_sleep" on:
> 
> commit: 420356c3504091f0f6021974389df7c58f365dad ("sched/fair: Implement delayed dequeue")
> https://git.kernel.org/cgit/linux/kernel/git/peterz/queue.git sched/core

> [   86.252370][  T674] ------------[ cut here ]------------
> [ 86.252945][ T674] do not call blocking ops when !TASK_RUNNING; state=1 set at kthread_worker_fn (kernel/kthread.c:?) 
> [ 86.254001][ T674] WARNING: CPU: 1 PID: 674 at kernel/sched/core.c:8469 __might_sleep (kernel/sched/core.c:8465) 

> [ 86.283398][ T674] ? handle_bug (arch/x86/kernel/traps.c:239) 
> [ 86.283995][ T674] ? exc_invalid_op (arch/x86/kernel/traps.c:260) 
> [ 86.284787][ T674] ? asm_exc_invalid_op (arch/x86/include/asm/idtentry.h:621) 
> [ 86.285682][ T674] ? __might_sleep (kernel/sched/core.c:8465) 
> [ 86.286380][ T674] ? __might_sleep (kernel/sched/core.c:8465) 
> [ 86.287116][ T674] kthread_worker_fn (include/linux/kernel.h:73 include/linux/freezer.h:53 kernel/kthread.c:851) 
> [ 86.287701][ T674] ? kthread_worker_fn (kernel/kthread.c:?) 
> [ 86.288138][ T674] kthread (kernel/kthread.c:391) 
> [ 86.288482][ T674] ? __cfi_kthread_worker_fn (kernel/kthread.c:803) 
> [ 86.288951][ T674] ? __cfi_kthread (kernel/kthread.c:342) 
> [ 86.289560][ T674] ret_from_fork (arch/x86/kernel/process.c:153) 
> [ 86.290162][ T674] ? __cfi_kthread (kernel/kthread.c:342) 
> [ 86.291465][ T674] ret_from_fork_asm (arch/x86/entry/entry_64.S:254) 

AFAICT this is a pre-existing issue. Notably that all transcribes to:

kthread_worker_fn()
  ...
repeat:
  set_current_state(TASK_INTERRUPTIBLE);
  ...
  if (work) { // false
    __set_current_state(TASK_RUNNING);
    ...
  } else if (!freezing(current)) // false -- we are freezing
    schedule();

  // so state really is still TASK_INTERRUPTIBLE here
  try_to_freeze()
    might_sleep() <--- boom, per the above.