Re: [PATCH i-g-t 2/2] tests/intel/xe_oa: Tests for OA syncs

["Dixit, Ashutosh" <ashutosh.dixit@intel.com>]

On Fri, 25 Oct 2024 12:35:35 -0700, Umesh Nerlige Ramappa wrote:
>

Hi Umesh,

> On Fri, Oct 25, 2024 at 11:52:05AM -0700, Ashutosh Dixit wrote:
> > Verify OA syncs signal correctly in open and change config code
> > paths. Verify with different types of sync objects as well as by both
> > waiting and skipping the wait for syncs to signal.
> >
> > v2: Significantly expand oa syncs testing as described above
>
> Are we just enabling output sync for the OA use case, as in, user can only
> wait for the OA configuration to complete, but cannot control when the OA
> config starts?

No, both input and output syncs are implemented in the kernel. But this
patch only adds tests for the output syncs. I am tracking writing tests for
input syncs elsewhere.

Input syncs are also straightforward, xe_sync_entry_add_deps() basically
does that for you, you just need to call that function from the right
place. The output sync's are quite tricky and I have fixed many uaf bugs
with these tests.

> Which is what existed earlier, so it should be fine either ways.

You need both input and output syncs for it to be useful (that allows
pipelining the workload while keeping it sync'd with a particular OA
config). Mesa is using and have tested both.

>
> LGTM,
>
> Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Thanks.
--
Ashutosh