2011-06-03 20:51:30

by Max Asbock

Subject: [PATCH] timerfd: really wake up processes when timer is cancelled on clock change

When the system time is set the clock_was_set() function calls
timerfd_clock_was_set() to cancel and wake up processes waiting on
potential cancelable timerfd timers. However the wake up currently has
no effect because in the case of timerfd_read it is dependent on
ctx->ticks not being 0. timerfd_poll also requires ctx->ticks being non
zero. As a consequence processes waiting on cancelable timers only get
woken up when the timers expire. This patch fixes this by incrementing
ctx->ticks before calling wake_up.

Signed-off-by: Max Asbock <[email protected]>
---

--- linux-3.0-rc1/fs/timerfd.c
+++ linux-3.0-rc1.timerfd/fs/timerfd.c
@@ -61,7 +61,9 @@ static enum hrtimer_restart timerfd_tmrp

/*
* Called when the clock was set to cancel the timers in the cancel
- * list.
+ * list. This will wake up processes waiting on these timers. The
+ * wake-up requires ctx->ticks to be non zero, therefore we increment
+ * it before calling wake_up_locked().
*/
void timerfd_clock_was_set(void)
{
@@ -76,6 +78,7 @@ void timerfd_clock_was_set(void)
spin_lock_irqsave(&ctx->wqh.lock, flags);
if (ctx->moffs.tv64 != moffs.tv64) {
ctx->moffs.tv64 = KTIME_MAX;
+ ctx->ticks++;
wake_up_locked(&ctx->wqh);
}
spin_unlock_irqrestore(&ctx->wqh.lock, flags);


--
Max Asb?ck

2011-06-14 22:22:01

by Andrew Morton

Subject: Re: [PATCH] timerfd: really wake up processes when timer is cancelled on clock change

On Fri, 03 Jun 2011 13:53:45 -0700 Max Asbock <[email protected]> wrote:

> When the system time is set the clock_was_set() function calls
> timerfd_clock_was_set() to cancel and wake up processes waiting on
> potential cancelable timerfd timers. However the wake up currently has
> no effect because in the case of timerfd_read it is dependent on
> ctx->ticks not being 0. timerfd_poll also requires ctx->ticks being non
> zero. As a consequence processes waiting on cancelable timers only get
> woken up when the timers expire. This patch fixes this by incrementing
> ctx->ticks before calling wake_up.
>
> Signed-off-by: Max Asbock <[email protected]>
> ---
>
> --- linux-3.0-rc1/fs/timerfd.c
> +++ linux-3.0-rc1.timerfd/fs/timerfd.c
> @@ -61,7 +61,9 @@ static enum hrtimer_restart timerfd_tmrp
>
> /*
> * Called when the clock was set to cancel the timers in the cancel
> - * list.
> + * list. This will wake up processes waiting on these timers. The
> + * wake-up requires ctx->ticks to be non zero, therefore we increment
> + * it before calling wake_up_locked().
> */
> void timerfd_clock_was_set(void)
> {
> @@ -76,6 +78,7 @@ void timerfd_clock_was_set(void)
> spin_lock_irqsave(&ctx->wqh.lock, flags);
> if (ctx->moffs.tv64 != moffs.tv64) {
> ctx->moffs.tv64 = KTIME_MAX;
> + ctx->ticks++;
> wake_up_locked(&ctx->wqh);
> }
> spin_unlock_irqrestore(&ctx->wqh.lock, flags);

Do you think this fix should be backported into -stable kernels? If so
(or if not), why?

It sounds like it _should_ be backported. I wonder if that will break
any apps which depend on (or work around) the current behaviour.

2011-06-15 17:41:12

by Max Asbock

Subject: Re: [PATCH] timerfd: really wake up processes when timer is cancelled on clock change

On Tue, 2011-06-14 at 15:24 -0700, Andrew Morton wrote:
> On Fri, 03 Jun 2011 13:53:45 -0700 Max Asbock <[email protected]> wrote:
>
> Do you think this fix should be backported into -stable kernels? If so
> (or if not), why?
>

The feature this patch fixes has been introduced in 3.0, therefore there
is no need for a packport to stable kernels.

- Max