[igt-dev] [PATCH] i915/gem_ctx_persistence: adjust hostile test timeout

[Andrzej Hajda]

On 14.07.2022 17:38, Kamil Konieczny wrote:
> On 2022-07-14 at 12:01:25 +0200, Andrzej Hajda wrote:
>>
>> On 13.07.2022 16:57, Kamil Konieczny wrote:
>>> Hi Andrzej,
>>>
>>> On 2022-07-12 at 09:17:22 +0200, Andrzej Hajda wrote:
>>>> GPU occasionally can hang during hostile test. Detection of such case and
>>>> then reset can take up to 5 seconds.
>>>>
>>>> Closes: https://gitlab.freedesktop.org/drm/intel/issues/2410
>>>> Suggested-by: Chris Wilson <chris.p.wilson at intel.com>
>>>> Signed-off-by: Andrzej Hajda <andrzej.hajda at intel.com>
>>>> ---
>>>>    tests/i915/gem_ctx_persistence.c | 2 +-
>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/tests/i915/gem_ctx_persistence.c b/tests/i915/gem_ctx_persistence.c
>>>> index 00dda3a8b52..50196edb19f 100644
>>>> --- a/tests/i915/gem_ctx_persistence.c
>>>> +++ b/tests/i915/gem_ctx_persistence.c
>>>> @@ -370,7 +370,7 @@ static void test_nohangcheck_hostile(int i915, const intel_ctx_cfg_t *cfg)
>>>>    	igt_require(__enable_hangcheck(dir, false));
>>>>    	for_each_ctx_cfg_engine(i915, cfg, e) {
>>>> -		int64_t timeout = reset_timeout_ms * NSEC_PER_MSEC;
>>>> +		int64_t timeout = 10000 * NSEC_PER_MSEC;
>>> May we extend this to other hostile reset timeouts ?
>> Thanks for review.
>>
>> You are right, that similar problem exists also for other persistence tests,
>> but I guess we need to check them carefully as they differ from this one,
>> for example by preemption.
>>
>>> Btw I think about limiting this to only new gens (like >= 11)
>>> where GuC dumps can take some time, but maybe I am overthinking.
>>> Reviewed-by: Kamil Konieczny <kamil.konieczny at linux.intel.com>
>>>
>>> One more question: should we restore preemption timeout at exit
>>> after test fails ?
>> No, it is assigned to local var, so no global change.
> I should stated this differently, when test dies inside do_test
> we do not restore preemtion timeout (which was set with sysfs).

This test (test_nohangcheck_hostile) is run directly, not via do_test.

>
> Maybe this should go in seperate patch.

Yes, fixing do_test is a different story.

Regards
Andrzej

>
> Regards,
> Kamil
>
>>>>    		const intel_ctx_t *ctx = intel_ctx_create(i915, cfg);
>>>>    		uint64_t ahnd = get_reloc_ahnd(i915, ctx->id);
>>>>    		igt_spin_t *spin;
>>>> -- 
>>>> 2.25.1
>>>>