[igt-dev] [PATCH] tests/intel/xe_exec_thread: Increase ufence timeout

[Mishra, Pallavi]

> -----Original Message-----
> From: Brost, Matthew <matthew.brost at intel.com>
> Sent: Thursday, January 11, 2024 5:00 PM
> To: Mishra, Pallavi <pallavi.mishra at intel.com>
> Cc: igt-dev at lists.freedesktop.org
> Subject: Re: [igt-dev] [PATCH] tests/intel/xe_exec_thread: Increase ufence
> timeout
> 
> On Fri, Jan 12, 2024 at 06:04:50AM +0530, Pallavi Mishra wrote:
> > For invalidate scenario, the ufence occasionally times out. Increase
> > timeout for invalidate case.
> >
> > Fixes: xe_exec_threads at threads-mixed-userptr-invalidate
> > and xe_exec_threads at threads-bal-mixed-userptr-invalidate
> >
> > Signed-off-by: Pallavi Mishra <pallavi.mishra at intel.com>
> > ---
> >  tests/intel/xe_exec_threads.c | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/tests/intel/xe_exec_threads.c
> > b/tests/intel/xe_exec_threads.c index 17ee57a49..04c0e2196 100644
> > --- a/tests/intel/xe_exec_threads.c
> > +++ b/tests/intel/xe_exec_threads.c
> > @@ -327,6 +327,7 @@ test_compute_mode(int fd, uint32_t vm, uint64_t
> addr, uint64_t userptr,
> >  					 bo_size, sync, 1);
> >  #define THREE_SEC	MS_TO_NS(3000)
> >  #define THIRTY_SEC	MS_TO_NS(30000)
> > +#define FIVE_SEC	MS_TO_NS(5000)
> >
> >  	fence_timeout = igt_run_in_simulation() ? THIRTY_SEC : THREE_SEC;
> 
> The correct fix would be:
> fence_timeout = igt_run_in_simulation() ? THIRTY_SEC : FIVE_SEC;

I did not want to change the timeout duration for all the scenarios hence..
> 
> I also suspect of this change as 3 seconds is already a very long time.

Agree. It did resolve the failures for LNL though. Might be some underlying problem causing this delay.

But now on the latest code, I do not see the failures. For now I will hold off.

Thanks!
> 
> Matt
> 
> >
> > @@ -389,7 +390,7 @@ test_compute_mode(int fd, uint32_t vm, uint64_t
> addr, uint64_t userptr,
> >  					xe_wait_ufence(fd,
> &data[j].exec_sync,
> >  						       USER_FENCE_VALUE,
> >  						       exec_queues[e],
> > -						       fence_timeout);
> > +						       FIVE_SEC);
> >  				igt_assert_eq(data[i].data, 0xc0ffee);
> >  			} else if (i * 2 != n_execs) {
> >  				/*
> > @@ -422,7 +423,7 @@ test_compute_mode(int fd, uint32_t vm, uint64_t
> addr, uint64_t userptr,
> >  		(flags & RACE ? n_execs / 2 + 1 : n_execs - 1) : 0;
> >  	for (i = j; i < n_execs; i++)
> >  		xe_wait_ufence(fd, &data[i].exec_sync, USER_FENCE_VALUE,
> > -			       exec_queues[i % n_exec_queues],
> fence_timeout);
> > +			       exec_queues[i % n_exec_queues], FIVE_SEC);
> >
> >  	/* Wait for all execs to complete */
> >  	if (flags & INVALIDATE)
> > --
> > 2.25.1
> >