Re: [PATCH i-g-t 1/2] tests/intel/xe_sriov_flr: Add reset-only FLR subtest

["Bernatowicz, Marcin" <marcin.bernatowicz@linux.intel.com>]

On 2/27/2026 9:58 AM, Piotr Piórkowski wrote:
> Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com> wrote on czw [2026-lut-26 18:31:36 +0100]:
>> Add flr-reset-only subtest intended as a minimal smoke test
>> for the VF reset sysfs write path while still using the xe-vfio-pci
>> based synchronization: the test skips if xe-vfio-pci binding is disabled,
>> if the xe_vfio_pci module is not loaded, or if any VF under test is
>> not bound to xe-vfio-pci.
>>
>> Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com>
>> Cc: Lukasz Laguna <lukasz.laguna@intel.com>
>> Cc: Piotr Piórkowski <piotr.piorkowski@intel.com>
>> ---
>>   tests/intel/xe_sriov_flr.c | 65 ++++++++++++++++++++++++++++++++++++++
>>   1 file changed, 65 insertions(+)
>>
>> diff --git a/tests/intel/xe_sriov_flr.c b/tests/intel/xe_sriov_flr.c
>> index b73727787..846731697 100644
>> --- a/tests/intel/xe_sriov_flr.c
>> +++ b/tests/intel/xe_sriov_flr.c
>> @@ -29,6 +29,12 @@
>>    * Functionality: FLR
>>    * Description: Examine behavior of SR-IOV VF FLR
>>    *
>> + * SUBTEST: flr-reset-only
> Or maybe just flr-basic?
>
>> + * Run type: BAT
>> + * Description:
>> + *   Initiates FLR without any additional state checks.
>> + *   Useful as a basic smoke test of the reset sysfs write path.
>> + *
>>    * SUBTEST: flr-vf1-clear
>>    * Run type: BAT
>>    * Description:
>> @@ -1016,6 +1022,60 @@ static void regs_subcheck_cleanup(struct subcheck_data *data)
>>   {
>>   }
>>   
>> +static void reset_only_subcheck_init(struct subcheck_data *data)
>> +{
>> +	if (!g_use_xe_vfio_pci) {
>> +		set_skip_reason(data, "xe-vfio-pci binding is disabled\n");
>> +		return;
>> +	}
>> +
>> +	if (!igt_kmod_is_loaded("xe_vfio_pci"))
> Sometimes it is xe_vfio_pci, sometimes xe-vfio-pci

xe_vfio_pci is a module name

xe-vfio-pci is a driver name

> What is the reason for this difference? Shouldn't we keep it somewhere as a constant?
I'll add some constants with xe_sriov_vfio test
> Or have helpers to handle this module at all?
>
>> +		set_skip_reason(data, "xe_vfio_pci is not loaded\n");
>> +}
>> +
>> +static void reset_only_subcheck_prepare_vf(int vf_id, struct subcheck_data *data)
>> +{
>> +	char *slot = igt_sriov_get_vf_pci_slot_alloc(data->pf_fd, vf_id);
>> +	char bound[64];
>> +	int bound_ret;
>> +
>> +	igt_assert(slot);
>> +
>> +	bound_ret = igt_pci_get_bound_driver_name(slot, bound, sizeof(bound));
>> +	if (bound_ret <= 0 || strcmp(bound, "xe-vfio-pci") != 0)
>> +		set_skip_reason(data, "VF%u not bound to xe-vfio-pci\n", vf_id);
>> +
>> +	free(slot);
>> +}
>> +
>> +static void noop_subcheck_verify_vf(int vf_id, int flr_vf_id, struct subcheck_data *data)
>> +{
>> +}
>> +
>> +static void noop_subcheck_cleanup(struct subcheck_data *data)
>> +{
>> +}
>> +
>> +static void reset_only_test(int pf_fd, int num_vfs, flr_exec_strategy exec_strategy)
>> +{
>> +	struct subcheck_data base = {
>> +		.pf_fd = pf_fd,
>> +		.num_vfs = num_vfs,
>> +		.tile = 0,
>> +		.stop_reason = NULL,
>> +	};
>> +	struct subcheck check = {
>> +		.data = &base,
>> +		.name = "reset-only",
>> +		.init = reset_only_subcheck_init,
>> +		.prepare_vf = reset_only_subcheck_prepare_vf,
>> +		.verify_vf = noop_subcheck_verify_vf,
>> +		.cleanup = noop_subcheck_cleanup,
>> +	};
>> +
>> +	verify_flr(pf_fd, num_vfs, &check, 1, exec_strategy);
>> +}
>> +
>>   static void clear_tests(int pf_fd, int num_vfs, flr_exec_strategy exec_strategy)
>>   {
>>   	const uint8_t num_tiles = xe_tiles_count(pf_fd);
>> @@ -1141,6 +1201,11 @@ int igt_main_args("vw:", long_options, help_str, opt_handler, NULL)
>>   		autoprobe = igt_sriov_is_driver_autoprobe_enabled(pf_fd);
>>   	}
>>   
>> +	igt_describe("Initiate FLR without any additional state checks.");
>> +	igt_subtest("flr-reset-only") {
>> +		reset_only_test(pf_fd, 1, execute_sequential_flr);
> In this case, shouldn't we have dynamic subtests regarding the number of VFs?
> I know that this will be expanded in the next patch, but maybe it should be the default.
> Because if someone wants to run tests on only one Vf's, they can still use
> the dynamic subtets filter.
Currently VF1 is default to keep CI time bounded.
In --extended mode (next patch), if numvfs-N is specified, we enable N 
VFs and run FLR sequentially on each.
If no specific numvfs-* dynamic subtest is selected, we iterate over the 
full range (numvfs-1 … numvfs-totalVFs).

If you prefer dynamic by default, which set would you suggest running - 
VF1 only, random number of VFs, total VFs ?

Thanks,

Marcin

>
> Thanks,
> Piotr
>
>> +	}
>> +
>>   	igt_describe("Verify LMEM, GGTT, and SCRATCH_REGS are properly cleared after VF1 FLR");
>>   	igt_subtest("flr-vf1-clear") {
>>   		clear_tests(pf_fd, 1, execute_sequential_flr);
>> -- 
>> 2.43.0
>>