On 17 February 2015 at 11:56, Radu-Andrei Bulie
<[email protected]> wrote:
> Hi,
>
>
>
> There is a test in schedule validaton : test_schedule_pause_resume in which
> the following is performed:
>
> - a number of events (buf type ) are enqueued
>
> - a number of iteratons(NUM_BUFS_BEFORE_PAUSE) are performed and in
> each iteration a schedule of type NO_WAIT is executed
>
>
>
> Looking at the code one can see that there is a strict assertion that for
> each iteration, odp_schedule function must return a valid event.
>
> From an implementation perspective there is no guarantee that when a “NO
> WAIT schedule” is performed, there will always be a valid event returned.
Agree. The scheduler is executing asynchronously from the threads.
Only in linux-generic (with a SW scheduler)
>
> The code snippet is :
>
> for (i = 0; i < NUM_BUFS_BEFORE_PAUSE; i++) {
>
> ev = odp_schedule(&from, ODP_SCHED_NO_WAIT);
The code needs to check that 'ev' returns a valid event before it
continues with the code below.
>
> CU_ASSERT(from == queue);
>
> buf = odp_buffer_from_event(ev);
>
> odp_buffer_free(buf);
>
> }
>
>
>
> So I think that the assertion should be made only if the schedule returns a
> valid event.
Correct.
Probably the test needs to be slightly rewritten, we cannot have a
fixed number of loop iterations since not all iterations will return
an event. The test should loop until all expected events have been
received through odp_schedule() and have some timeout to handle the
situation where the ODP implementation doesn't actually return all
expected events (I don't think we want the validation to hang, better
to detect this situation and report a CU_FAIL).
>
>
>
>
>
> Regards,
>
>
>
> Radu
>
>
> _______________________________________________
> lng-odp mailing list
> [email protected]
> http://lists.linaro.org/mailman/listinfo/lng-odp
>
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp
