[Intel-gfx] [PATCH] drm/i915/selftest: If reconfigure_sseu is busy, try again

[Tvrtko Ursulin]

On 26/11/2019 17:08, Tvrtko Ursulin wrote:
> 
> On 26/11/2019 17:05, Chris Wilson wrote:
>> Quoting Tvrtko Ursulin (2019-11-26 17:00:53)
>>>
>>> On 26/11/2019 16:47, Chris Wilson wrote:
>>>> Following 58b4c1a07ada ("drm/i915: Reduce nested 
>>>> prepare_remote_context()
>>>> to a trylock"), we punt to the caller if the local intel_context
>>>> happens to be busy as we try to rewrite the sseu (due to retiring in
>>>> another thread). As the interlude should be short, spin until the lock
>>>> is available.
>>>>
>>>> The regret for using mutex_trylock() and not an atomic insertion of the
>>>> barrier is growing...
>>>>
>>>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>> ---
>>>>    drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c | 5 ++++-
>>>>    1 file changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c 
>>>> b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
>>>> index 2ea4790f3721..571cc996577c 100644
>>>> --- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
>>>> +++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
>>>> @@ -1197,7 +1197,10 @@ __sseu_test(const char *name,
>>>>        if (ret)
>>>>                goto out_pm;
>>>> -     ret = intel_context_reconfigure_sseu(ce, sseu);
>>>> +     do {
>>>> +             ret = intel_context_reconfigure_sseu(ce, sseu);
>>>> +             cond_resched();
>>>> +     } while (ret == -EAGAIN);
>>>>        if (ret)
>>>>                goto out_spin;
>>>>
>>>
>>> Hm I looked at the selftests, saw error is correctly propagated, and
>>> concluded it will be fine. I missed the problem selftests will not
>>> actually retry. But wait, can we even count that userspace will if all
>>> of a sudden ctx.set_param starts returning -EAGAIN sporadically? Feels
>>> like we may need to revert.
>>
>> We invoke the principle of drmIoctl() catches -EAGAIN.
> 
> I'm not comfortable with that. :( Not least how we are saying not to use 
> libdrm. man 2 ioctl does not mention -EAGAIN. :(

Or duct tape by looping in set_sseu as well?

Regards,

Tvrtko