Prev: [PATCH] hvc_console: Fix a race between hvc_close and hvc_remove
Next: [PATCH 12/43] workqueue: merge feature parameters into flags
From: Thomas Gleixner on 26 Feb 2010 07:00
On Fri, 26 Feb 2010, Rafał Miłecki wrote:

> Forwarding to ppl I could often notice in git log time.h

And how is this related to time.h ?

>
> ---------- Wiadomość przekazana dalej ----------
> From: Rafał Miłecki <zajec5(a)gmail.com>
> Date: 21 lutego 2010 15:10
> Subject: [PATCH][RFC] time: add wait_interruptible_timeout macro to
> sleep (w. timeout) until wake_up
> To: Linux Kernel Mailing List <linux-kernel(a)vger.kernel.org>,
> dri-devel(a)lists.sourceforge.net
> CC: Rafał Miłecki <zajec5(a)gmail.com>
>
>
> Signed-off-by: Rafał Miłecki <zajec5(a)gmail.com>
> ---
> We try to implement some PM in radeon KMS and we need to sync with VLBANK for
> reclocking engine/memory. The easiest and cleanest way seems to be sleeping in
> timer handler just before reclocking. Then our IRQ handler calls wake_up and we

Sleeping in the timer handler ? In which context runs this timer handler ?

> continue reclocking.
>
> As you see our sleeping is condition-less, we just wait for waking up queue.

Your sleep is not condition-less at all. See below.

> We hope this waking will happen from IRQ handler, but for less-happy case we
> also use some timeout (this will probably cause some single corruption, but
> we can live with it).
>
> Following macro is soemthing that seems to work fine for us, but instead
> introducing this to radeon KMS only, I'd like to propose adding this to whole
> wait.h. Do you this it's something we should place there? Can someone take this
> patch for me? Or maybe you find this rather useless and we should keep this
> marco locally?

You better delete it right away. It's broken and racy.

If the interrupt happens right before you enqueue to the wake queue,
then you miss it. The old interruptible_sleep_on_timeout() was
replaced by the event wait functions which have a condition exaclty to
avoid that race.

And you have a condition: Did an interrupt happen?

Thanks,

tglx
 | 
Pages: 1
Prev: [PATCH] hvc_console: Fix a race between hvc_close and hvc_remove
Next: [PATCH 12/43] workqueue: merge feature parameters into flags