Hello,
The semaphore API provides several flavors of the down primitive:
down, down_interruptible, down_killable, down_trylock, down_timeout
As far as I can tell, they all call __down_common (except down_trylock,
which returns 1 where the others would sleep).
I was looking for a version
1) with a timeout
2) that could be interrupted
e.g. down_interruptible_timeout, but it doesn't exist.
It seems
__down_common(sem, TASK_INTERRUPTIBLE, timeout);
would work as expected, no?
Do you know why it is not offered?
(Maybe there is a better way to achieve the same thing?)
[POST SCRIPTUM EDIT]
I found this 2007 discussion:
http://thread.gmane.org/gmane.linux.kernel/498034
At the time, Andrew said
"Nobody else has needed to invent new locking infrastructure
to do such things and I'd prefer not to do so now."
I suppose this is still true :-)
[/EDIT]
My use-case is pretty simple:
A) process-context kernel thread fills a FIFO and calls down(&fifo_empty);
B) ISR handles the FIFO-empty interrupt with up(&fifo_empty);
However, in case something goes wrong and the interrupt never fires,
I don't want the process to be stuck in an uninterruptible sleep.
Perhaps I can set a tiny timeout (e.g. 10 ?s) and not worry about
the interruptible part for such a small duration? (Hmm, __down_common
calls schedule_timeout, which is jiffies-based. I don't think there
is a hrtimers flavor. So ?s timeouts would be off the table?)
Or I could use the interruptible version, and let the user kill the
operation if necessary.
I'd like to hear your comments and suggestions.
Regards.
On Mon, 15 Jun 2015, Mason wrote:
> A) process-context kernel thread fills a FIFO and calls down(&fifo_empty);
> B) ISR handles the FIFO-empty interrupt with up(&fifo_empty);
>
> However, in case something goes wrong and the interrupt never fires,
> I don't want the process to be stuck in an uninterruptible sleep.
>
> Perhaps I can set a tiny timeout (e.g. 10 ?s) and not worry about
> the interruptible part for such a small duration? (Hmm, __down_common
> calls schedule_timeout, which is jiffies-based. I don't think there
> is a hrtimers flavor. So ?s timeouts would be off the table?)
>
> Or I could use the interruptible version, and let the user kill the
> operation if necessary.
Use a completion.
Thanks,
tglx
On 15/06/2015 18:56, Thomas Gleixner wrote:
> On Mon, 15 Jun 2015, Mason wrote:
>
>> A) process-context kernel thread fills a FIFO and calls down(&fifo_empty);
>> B) ISR handles the FIFO-empty interrupt with up(&fifo_empty);
>>
>> However, in case something goes wrong and the interrupt never fires,
>> I don't want the process to be stuck in an uninterruptible sleep.
>>
>> Perhaps I can set a tiny timeout (e.g. 10 ?s) and not worry about
>> the interruptible part for such a small duration? (Hmm, __down_common
>> calls schedule_timeout, which is jiffies-based. I don't think there
>> is a hrtimers flavor. So ?s timeouts would be off the table?)
>>
>> Or I could use the interruptible version, and let the user kill the
>> operation if necessary.
>
> Use a completion.
Thanks for the pointer. I will also read the following LKML thread.
http://thread.gmane.org/gmane.linux.kernel/664514
Regards.