Re: [RT PATCH 1/2] kernel/cpu_chill: use schedule_hrtimeout()

[Thomas Gleixner]

On Fri, 16 Mar 2018, Sebastian Andrzej Siewior wrote:

> cpu_chill() uses __hrtimer_nanosleep() with a hack to get
> TASK_UNINTERRUPTIBLE state. It seems to work but it does not.
> The problem is that we may have a signal pending and receive a sporadic
> wake up _before_ the timer expires. At that point the code will look at

_We_ have nothing pending. Please describe it in a factual way and not like
you are running the code yourself.

> ->restart_block and use it based on its value from the previous syscall
> resulting in a possible copy-to-user.

What you want to describe is:

If a task calls cpu_chill() and gets woken up by a regular or spurious
wakeup and has a signal pending, then it exits the sleep loop in
do_nanosleep() and sets up the restart block. If restart->nanosleep.type is
not TI_NONE then this results in accessing a stale user pointer from a
previously interrupted syscall and a copy to user based on the stale
pointer or a BUG() when 'type' is not supported in nanosleep_copyout().

> instead all this trouble, we could use schedule_hrtimeout() which is not

Sentences start with an uppercase letter.

> instead all this trouble, we could use schedule_hrtimeout() which is not

s/we could// 

> an user-space interface and does not have this side effects.

hmm?

> Cc: stable...@vger.kernel.org
> Signed-off-by: Sebastian Andrzej Siewior 
> ---
>  kernel/time/hrtimer.c | 9 -
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 6c77643eaf02..c0efc22ba635 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1873,14 +1873,13 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct 
> compat_timespec __user *, rqtp,
>   */
>  void cpu_chill(void)
>  {
> - struct timespec64 tu = {
> - .tv_nsec = NSEC_PER_MSEC,
> - };
> + ktime_t chill_time;
>   unsigned int freeze_flag = current->flags & PF_NOFREEZE;
>  
> + chill_time = ktime_set(0, NSEC_PER_MSEC);
> + set_current_state(TASK_UNINTERRUPTIBLE);
>   current->flags |= PF_NOFREEZE;
> - __hrtimer_nanosleep(, HRTIMER_MODE_REL_HARD, CLOCK_MONOTONIC,
> - TASK_UNINTERRUPTIBLE);
> + schedule_hrtimeout(_time, HRTIMER_MODE_REL_HARD);
>   if (!freeze_flag)
>   current->flags &= ~PF_NOFREEZE;
>  }
> -- 
> 2.16.2
> 
> 

Re: [RT PATCH 1/2] kernel/cpu_chill: use schedule_hrtimeout()

[Thomas Gleixner]

On Fri, 16 Mar 2018, Sebastian Andrzej Siewior wrote:

> cpu_chill() uses __hrtimer_nanosleep() with a hack to get
> TASK_UNINTERRUPTIBLE state. It seems to work but it does not.
> The problem is that we may have a signal pending and receive a sporadic
> wake up _before_ the timer expires. At that point the code will look at

_We_ have nothing pending. Please describe it in a factual way and not like
you are running the code yourself.

> ->restart_block and use it based on its value from the previous syscall
> resulting in a possible copy-to-user.

What you want to describe is:

If a task calls cpu_chill() and gets woken up by a regular or spurious
wakeup and has a signal pending, then it exits the sleep loop in
do_nanosleep() and sets up the restart block. If restart->nanosleep.type is
not TI_NONE then this results in accessing a stale user pointer from a
previously interrupted syscall and a copy to user based on the stale
pointer or a BUG() when 'type' is not supported in nanosleep_copyout().

> instead all this trouble, we could use schedule_hrtimeout() which is not

Sentences start with an uppercase letter.

> instead all this trouble, we could use schedule_hrtimeout() which is not

s/we could// 

> an user-space interface and does not have this side effects.

hmm?

> Cc: stable...@vger.kernel.org
> Signed-off-by: Sebastian Andrzej Siewior 
> ---
>  kernel/time/hrtimer.c | 9 -
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 6c77643eaf02..c0efc22ba635 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1873,14 +1873,13 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct 
> compat_timespec __user *, rqtp,
>   */
>  void cpu_chill(void)
>  {
> - struct timespec64 tu = {
> - .tv_nsec = NSEC_PER_MSEC,
> - };
> + ktime_t chill_time;
>   unsigned int freeze_flag = current->flags & PF_NOFREEZE;
>  
> + chill_time = ktime_set(0, NSEC_PER_MSEC);
> + set_current_state(TASK_UNINTERRUPTIBLE);
>   current->flags |= PF_NOFREEZE;
> - __hrtimer_nanosleep(, HRTIMER_MODE_REL_HARD, CLOCK_MONOTONIC,
> - TASK_UNINTERRUPTIBLE);
> + schedule_hrtimeout(_time, HRTIMER_MODE_REL_HARD);
>   if (!freeze_flag)
>   current->flags &= ~PF_NOFREEZE;
>  }
> -- 
> 2.16.2
> 
>