2008-02-01 13:35:48

by Oleg Nesterov

Subject: [PATCH 1/5] hrtimer_nanosleep: use -EINTR, not -ERESTARTNOHAND

hrtimer_nanosleep:

/* Absolute timers do not update the rmtp value and restart: */
return -ERESTARTNOHAND;

This is not right, -ERESTARTNOHAND means we will do restart if there is no
in fact pending signal, and worse, this restart will use the same unchanged
"__user *rmtp" parameter.

Use -EINTR instead.

Signed-off-by: Oleg Nesterov <[email protected]>

--- MM/kernel/hrtimer.c~HRT_RMTP~ 2008-01-27 17:07:39.000000000 +0300
+++ MM/kernel/hrtimer.c~HRT_RMTP 2008-02-01 13:43:52.000000000 +0300
@@ -1359,7 +1359,7 @@ long hrtimer_nanosleep(struct timespec *

/* Absolute timers do not update the rmtp value and restart: */
if (mode == HRTIMER_MODE_ABS)
- return -ERESTARTNOHAND;
+ return -EINTR;

if (rmtp) {
rem = ktime_sub(t.timer.expires, t.timer.base->get_time());

2008-02-01 13:51:50

by Thomas Gleixner

Subject: Re: [PATCH 1/5] hrtimer_nanosleep: use -EINTR, not -ERESTARTNOHAND

On Fri, 1 Feb 2008, Oleg Nesterov wrote:

> hrtimer_nanosleep:
>
> /* Absolute timers do not update the rmtp value and restart: */
> return -ERESTARTNOHAND;
>
> This is not right, -ERESTARTNOHAND means we will do restart if there is no
> in fact pending signal, and worse, this restart will use the same unchanged
> "__user *rmtp" parameter.

-ERESTARTNOHAND is safe here. We want to restart the timer.

The rmtp problem is separate, but that's addressed by your follow up changes.

Thanks,
tglx

> Use -EINTR instead.
>
> Signed-off-by: Oleg Nesterov <[email protected]>
>
> --- MM/kernel/hrtimer.c~HRT_RMTP~ 2008-01-27 17:07:39.000000000 +0300
> +++ MM/kernel/hrtimer.c~HRT_RMTP 2008-02-01 13:43:52.000000000 +0300
> @@ -1359,7 +1359,7 @@ long hrtimer_nanosleep(struct timespec *
>
> /* Absolute timers do not update the rmtp value and restart: */
> if (mode == HRTIMER_MODE_ABS)
> - return -ERESTARTNOHAND;
> + return -EINTR;
>
> if (rmtp) {
> rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
>

2008-02-01 14:17:34

by Oleg Nesterov

Subject: Re: [PATCH 1/5] hrtimer_nanosleep: use -EINTR, not -ERESTARTNOHAND

On 02/01, Thomas Gleixner wrote:
>
> On Fri, 1 Feb 2008, Oleg Nesterov wrote:
>
> > hrtimer_nanosleep:
> >
> > /* Absolute timers do not update the rmtp value and restart: */
> > return -ERESTARTNOHAND;
> >
> > This is not right, -ERESTARTNOHAND means we will do restart if there is no
> > in fact pending signal, and worse, this restart will use the same unchanged
> > "__user *rmtp" parameter.
>
> -ERESTARTNOHAND is safe here. We want to restart the timer.

Ah. I was greatly confused by the comment, it says "do not ... restart",
now I understand what this _actually_ means.

And yes, I was wrong. If restart is wanted, it is safe to use the original
*rqtp, the timer is HRTIMER_MODE_ABS.

Thanks Thomas!

Andrew, please ignore this patch.

Oleg.