Re: [PATCH RT 4/2] hrtimer: Don't lose state in cpu_chill()

[Mike Galbraith]

On Mon, 2019-02-25 at 17:34 +0100, Sebastian Andrzej Siewior wrote:
> On 2019-02-25 15:43:35 [+0100], Mike Galbraith wrote:
> > Hi Sebastian,
> Hi Mike,
> 
> > My box claims that this patch is busted.  It argues its case by IO
> > deadlocking any kernel this patch is applied to when spinning rust is
> > flogged, including virgin 4.19-rt14, said kernel becoming stable again
> > when I whack the accused.
> 
> does the following hunk make any difference?

Ah.  Yup, box says you're right.

> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 6ecdb9469ca9..e154632b90b4 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1884,20 +1884,29 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct old_timespec32 __user *, rqtp,
>   */
>  void cpu_chill(void)
>  {
> +	struct task_struct *self = current;
>  	ktime_t chill_time;
>  	unsigned int freeze_flag = current->flags & PF_NOFREEZE;
> -	long saved_state;
>  
> -	saved_state = current->state;
> -	chill_time = ktime_set(0, NSEC_PER_MSEC);
> +	raw_spin_lock_irq(&self->pi_lock);
> +	WARN_ON(self->saved_state != TASK_RUNNING);
> +	self->saved_state = self->state;
>  	__set_current_state_no_track(TASK_UNINTERRUPTIBLE);
> +	raw_spin_unlock_irq(&self->pi_lock);
> +
> +	chill_time = ktime_set(0, NSEC_PER_MSEC);
> +
>  	current->flags |= PF_NOFREEZE;
>  	sleeping_lock_inc();
>  	schedule_hrtimeout(&chill_time, HRTIMER_MODE_REL_HARD);
>  	sleeping_lock_dec();
>  	if (!freeze_flag)
>  		current->flags &= ~PF_NOFREEZE;
> -	__set_current_state_no_track(saved_state);
> +
> +	raw_spin_lock_irq(&self->pi_lock);
> +	__set_current_state_no_track(self->saved_state);
> +	self->saved_state = TASK_RUNNING;
> +	raw_spin_unlock_irq(&self->pi_lock);
>  }
>  EXPORT_SYMBOL(cpu_chill);
>  #endif
> 
> Sebastian