[Intel-gfx] [PATCH i-g-t 2/2] tests/i915/perf: Exercise barrier race

[Kamil Konieczny]

Hi Janusz,

please send patches to igt ML and add other addresses to cc:
I have one nit, see below.

On 2023-01-31 at 10:17:31 +0100, Janusz Krzysztofik wrote:
> Add a new subtest focused on exercising interaction between perf open /
> close operations, which replace active barriers with perf requests, and
> concurrent barrier preallocate / acquire operations performed during
> context first pin / last unpin.
> 
> References: https://gitlab.freedesktop.org/drm/intel/-/issues/6333
> Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik at linux.intel.com>
> Cc: Chris Wilson <chris.p.wilson at linux.intel.com>
> ---
>  tests/i915/perf.c | 41 +++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 39 insertions(+), 2 deletions(-)
> 
> diff --git a/tests/i915/perf.c b/tests/i915/perf.c
> index e33cacc443..11a3ec21ab 100644
> --- a/tests/i915/perf.c
> +++ b/tests/i915/perf.c
> @@ -39,6 +39,7 @@
>  #include <math.h>
>  
>  #include "i915/gem.h"
> +#include "i915/gem_create.h"
>  #include "i915/perf.h"
>  #include "igt.h"
>  #include "igt_perf.h"
> @@ -4885,7 +4886,27 @@ test_whitelisted_registers_userspace_config(void)
>  	i915_perf_remove_config(drm_fd, config_id);
>  }
>  
> -static void test_open_race(const struct intel_execution_engine2 *e, int timeout)
> +static void gem_exec_nop(int i915, const struct intel_execution_engine2 *e)
> +{
> +	const uint32_t bbe = MI_BATCH_BUFFER_END;
> +	struct drm_i915_gem_exec_object2 obj = { };
> +	struct drm_i915_gem_execbuffer2 execbuf = {
> +		.buffers_ptr = to_user_pointer(&obj),
> +		.buffer_count = 1,
> +	};
> +
> +	obj.handle = gem_create(i915, 4096);
> +	gem_write(i915, obj.handle, 0, &bbe, sizeof(bbe));
> +
> +	execbuf.flags = e->flags;
> +	gem_execbuf(i915, &execbuf);
> +
> +	gem_sync(i915, obj.handle);
> +	gem_close(i915, obj.handle);
> +}
> +
> +static void test_open_race(const struct intel_execution_engine2 *e, int timeout,
> +			   bool use_spin)
-------------------------- ^
This is not the best way to develop new code, it may be good
for fast development but imho it is better to refactor existing
code and avoiding to add bool logic into function.
It can be done later as this is revealing the bug.

>  {
>  	int *done;
>  
> @@ -4926,6 +4947,12 @@ static void test_open_race(const struct intel_execution_engine2 *e, int timeout)
>  				ctx = gem_context_create_for_engine(i915, e->class, e->instance);
>  				gem_context_set_persistence(i915, ctx, persistence);
>  
> +				if (!use_spin) {
> +					gem_exec_nop(i915, e);
> +					gem_context_destroy(i915, ctx);
> +					continue;
> +				}
> +
>  				spin = __igt_spin_new(i915, ctx, .ahnd = ahnd);
>  				for (int i = random() % 7; i--; ) {
>  					if (random() & 1) {
> @@ -5330,7 +5357,17 @@ igt_main
>  		for_each_physical_engine(drm_fd, e)
>  			if (e->class == I915_ENGINE_CLASS_RENDER)
>  				igt_dynamic_f("%s", e->name)
> -					test_open_race(e, 5);
> +					test_open_race(e, 5, true);
> +	}
> +

Please add here some TODO comments from discussions and a note
which will help bug filling team to classify that.

Regards,
Kamil

> +	igt_describe("Exercise perf open/close against intensive barrier preallocate/acquire");
> +	igt_subtest_with_dynamic("barrier-race") {
> +		const struct intel_execution_engine2 *e;
> +
> +		for_each_physical_engine(drm_fd, e)
> +			if (e->class == I915_ENGINE_CLASS_RENDER)
> +				igt_dynamic_f("%s", e->name)
> +					test_open_race(e, 5, false);
>  	}
>  
>  	igt_fixture {
> -- 
> 2.25.1
>