[PATCH i-g-t 2/2] tests/intel/perf_pmu: Migrate to unbind + unload

[Lucas De Marchi]

On Mon, Nov 04, 2024 at 11:25:06AM -0800, Umesh Nerlige Ramappa wrote:
>On Tue, Oct 22, 2024 at 10:05:02PM -0700, Lucas De Marchi wrote:
>>There isn't much point in testing we can't remove the module directly
>>after removing *some* known dependencies. The point in this test is to
>>make sure that if the HW vanishes, driver still behaves. That is
>>currently not true, as it will cause perf to explode.
>>
>>Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
>
>LGTM,
>
>Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>

I'm actually thinking about splitting this patch. See below.

>
>Thanks,
>Umesh
>>---
>>tests/intel/perf_pmu.c | 22 ++++++++++++----------
>>1 file changed, 12 insertions(+), 10 deletions(-)
>>
>>diff --git a/tests/intel/perf_pmu.c b/tests/intel/perf_pmu.c
>>index bfa2d501a..6eb4f09d1 100644
>>--- a/tests/intel/perf_pmu.c
>>+++ b/tests/intel/perf_pmu.c
>>@@ -2282,7 +2282,6 @@ static void test_unload(unsigned int num_engines)
>>		int fd[4 + num_engines * 3], i;
>>		uint64_t *buf;
>>		int count = 0, ret;
>>-		char *who = NULL;
>>		int i915;
>>
>>		i915 = __drm_open_driver(DRIVER_INTEL);
>>@@ -2337,23 +2336,26 @@ static void test_unload(unsigned int num_engines)
>>
>>		igt_debug("Read %d events from perf and trial unload\n", count);
>>		pmu_read_multi(fd[0], count, buf);
>>-		ret = __igt_i915_driver_unload(&who);

I think I will drop this part first, so I can remove the
__igt_i915_driver_unload() and finish migrating i915 to unbind + unload.

Then when the kernel is fixed (which will probably happen on 6.13 or
6.14) we can think about adding back the additions below for unbind
while perf is open. Thoughts?

Lucas De Marchi