Re: Synchronization mechanism between wait_for_completion_interruptible_timeout() & complete()

From: Naveen Kumar Parna
Date: Mon Mar 09 2015 - 02:59:34 EST


I am having a PCIe device , it asserts two types if Tx interrupts i.e.
TX Queue count interrupt & TX Queue timer interrupt.

TX Queue count interrupt related to âTX Queue count registerâ : No. of
TXQ transactions to move from host before asserting TX Queue count
interrupt

TX Queue timer interrupt related to âTX Queue timer registerâ: Time
out value in uSec. The timeout covers the case where a burst of
transactions are sent but not enough to trigger the count interrupt.

I am getting lot many TX Queue timer interrupts for larger write
requests and the current driver implementation is calling complete()
API in both TX Queue count & timer interrupts. Here âx->done > 0â
because of multiple TX Queue timer interrupts.

Yes I agree with you. I need to correct driver write() function.

Thanks,
Naveen

On Fri, Mar 6, 2015 at 2:18 PM, Michal Hocko <mhocko@xxxxxxx> wrote:
> On Fri 06-03-15 14:04:53, Naveen Kumar Parna wrote:
>> Thanks Michal.
>>
>> Now I am clearing x->done to zero before
>> wait_for_completion_interruptible_timeout.
>
> You shouldn't touch this internal thing. Why would you want that
> anyway? If x->done > 0 then there has been something completed.
>
>> On Thu, Mar 5, 2015 at 6:53 PM, Michal Hocko <mhocko@xxxxxxx> wrote:
>> > On Wed 04-03-15 18:46:43, Naveen Kumar Parna wrote:
>> >> Hello,
>> >>
>> >> I have a question regarding the synchronization mechanism between
>> >> wait_for_completion_interruptible_timeout() & complete(). I used
>> >> complete() API in the ISR Tx interrupt path and
>> >> wait_for_completion_interruptible_timeout() in struct file_operations
>> >> .write method.
>> >>
>> >> Let assume a scenario in which Tx interrupt occurred and ISR called
>> >> complete() API before any process actually blocked at write method on
>> >> wait_for_completion_interruptible_timeout().
>> >>
>> >> After a few milli seconds, a process called write() API. Does it gets
>> >> blocked on wait_for_completion_interruptible_timeout() for completion
>> >> of a specific task to be signaled from new Tx interrupt?
>> >
>> > No, see do_wait_for_common resp. complete and x->done handling.
>> >
>> > --
>> > Michal Hocko
>> > SUSE Labs
>
> --
> Michal Hocko
> SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/