Re: [PATCH i-g-t] tests/intel/xe_evict: Drop blanket skip when system memory < VRAM

["Gupta, Varun" <varun.gupta@intel.com>]

On 30-Sep-26 9:19 AM, Sharma, Nishit wrote:
> Hi Varun
>
> On 9/29/2026 11:09 PM, Varun Gupta wrote:
>> Since working_set() already clamps each subtest's allocation to what
>> fits in VRAM plus 4/5 of available system memory, and every subtest
>> loop has its own igt_skip_on_f(!ws) for cases that end up too small
>> to run. The fixture-level skip is therefore redundant and skips the
>> entire eviction suite.
>> Remove the skip and rely on the per-subtest working_set() sizing.
>>
>> Signed-off-by: Varun Gupta <varun.gupta@intel.com>
>> ---
>>   tests/intel/xe_evict.c | 9 ---------
>>   1 file changed, 9 deletions(-)
>>
>> diff --git a/tests/intel/xe_evict.c b/tests/intel/xe_evict.c
>> index 5a7298db0..230357419 100644
>> --- a/tests/intel/xe_evict.c
>> +++ b/tests/intel/xe_evict.c
>> @@ -911,15 +911,6 @@ int igt_main()
>>           igt_assert(vram_size);
>>           system_size = igt_get_avail_ram_mb() << 20;
>>   -        /* Test requires SRAM to about as big as VRAM. For 
>> example, small-cm creates
>> -         * (448 / 2) BOs with a size (1 / 128) of the total VRAM 
>> size. For
>> -         * simplicity ensure the SRAM size >= VRAM before running 
>> this test.
>> -         */
>> -        igt_skip_on_f(system_size < vram_size,
>> -                  "System memory %llu MiB is less than local memory 
>> %llu MiB\n",
>> -                  (unsigned long long)system_size >> 20,
>> -                  (unsigned long long)vram_size >> 20);
>> -
>
> Can we put some message here why skip was removed, Something like:
>
> No System vs VRAM size check needed as working_set() below limits each 
> subtest to what fits in VRAM/SRAM
>
> and a subtest skips on its own when nothing fits i.e. igt_skip(!ws, ...)
>
Thanks for the review Nishit, I believe the commit message indicates why 
this was removed and it would be
redundant to add same thing as a comment. If you still  believe its not 
clear in commit and we should add a
comment I would gladly add.

Thanks, Varun

>>           xe_for_each_engine(fd, hwe)
>>               if (hwe->engine_class != DRM_XE_ENGINE_CLASS_COPY)
>>                   break;
>
> Rest LGTM:
>
> Reviewed-by: Nishit Sharma <nishit.sharma@intel.com>