[PATCH 2/2] tests/intel/xe_eudebug_online: Add pagefault stress tests

[Gwan-gyeong Mun]

On 5/26/25 3:13 AM, Jan Maslak wrote:
> Add a stress variant of the already existing
> xe_eudebug_online at pagefault-[read|write] tests that increases
> number of threads in the shader and number of threads in
> threadgroup to their maximum available amounts.
> 
> Signed-off-by: Jan Maslak <jan.maslak at intel.com>
> ---
>   tests/intel/xe_eudebug_online.c | 27 ++++++++++++++++++++++++++-
>   1 file changed, 26 insertions(+), 1 deletion(-)
> 
> diff --git a/tests/intel/xe_eudebug_online.c b/tests/intel/xe_eudebug_online.c
> index 3d138bdc3..ad62dd929 100644
> --- a/tests/intel/xe_eudebug_online.c
> +++ b/tests/intel/xe_eudebug_online.c
> @@ -39,6 +39,7 @@
>   #define SHADER_PAGEFAULT_READ		(1 << 14)
>   #define SHADER_PAGEFAULT_WRITE		(1 << 15)
>   #define FAULTABLE_VM			(1 << 16)
> +#define PAGEFAULT_STRESS_TEST		(1 << 17)
>   #define TRIGGER_UFENCE_SET_BREAKPOINT	(1 << 24)
>   #define TRIGGER_RESUME_SINGLE_WALK	(1 << 25)
>   #define TRIGGER_RESUME_PARALLEL_WALK	(1 << 26)
> @@ -126,6 +127,9 @@ static struct intel_buf *create_uc_buf(int fd, int width, int height, uint64_t r
>   
>   static int get_number_of_threads(uint64_t flags)
>   {
> +	if (flags & (PAGEFAULT_STRESS_TEST))
> +		return 65535; /* 64K - 1 */
This value causes the Local Y Maximum and Local X Maximum values to be 
set higher than the values that can be set in the COMPUTE_WALKER_BODY 
structure used by the COMPUTE_WALKER command.
Please check that these values are set correctly in the 
__xehp_gpgpu_execfunc() function

G.G.
> +
>   	if (flags & (SHADER_MIN_THREADS | SHADER_PAGEFAULT))
>   		return 16;
>   
> @@ -1068,6 +1072,10 @@ static void run_online_client(struct xe_eudebug_client *c)
>   	struct intel_buf *buf;
>   	uint32_t *ptr;
>   	int fd, vm_flags;
> +	uint32_t num_threads_in_tg = 1;
> +
> +	if (c->flags & PAGEFAULT_STRESS_TEST)
> +		num_threads_in_tg = 64;
>   
>   	metadata[0] = calloc(2, sizeof(**metadata));
>   	metadata[1] = calloc(2, sizeof(**metadata));
> @@ -1115,7 +1123,7 @@ static void run_online_client(struct xe_eudebug_client *c)
>   	sip = get_sip(fd, c->flags);
>   
>   	igt_nsec_elapsed(&ts);
> -	gpgpu_shader_exec(ibb, buf, w_dim.x, w_dim.y, 1, shader, sip, 0, 0);
> +	gpgpu_shader_exec(ibb, buf, w_dim.x, w_dim.y, num_threads_in_tg, shader, sip, 0, 0);
>   
>   	gpgpu_shader_destroy(sip);
>   	gpgpu_shader_destroy(shader);
> @@ -1629,6 +1637,18 @@ static void test_set_breakpoint_online_sigint_debugger(int fd,
>    * Description:
>    *     Check whether KMD sends pagefault event for workload in debug mode that
>    *     triggers a write pagefault.
> + *
> + * SUBTEST: pagefault-read-stress
> + * Functionality: page faults
> + * Description:
> + *     Check whether KMD sends read pagefault event for workload in debug mode
> + *     with many threads.
> + *
> + * SUBTEST: pagefault-write-stress
> + * Functionality: page faults
> + * Description:
> + *     Check whether KMD sends write pagefault event for workload in debug mode
> + *     with many threads.
>    */
>   static void test_pagefault_online(int fd, struct drm_xe_engine_class_instance *hwe,
>   				  int flags)
> @@ -2664,6 +2684,11 @@ igt_main
>   	test_gt_render_or_compute("pagefault-write", fd, hwe)
>   		test_pagefault_online(fd, hwe, SHADER_PAGEFAULT_WRITE);
>   
> +	test_gt_render_or_compute("pagefault-read-stress", fd, hwe)
> +		test_pagefault_online(fd, hwe, SHADER_PAGEFAULT_READ | PAGEFAULT_STRESS_TEST);
> +	test_gt_render_or_compute("pagefault-write-stress", fd, hwe)
> +		test_pagefault_online(fd, hwe, SHADER_PAGEFAULT_WRITE | PAGEFAULT_STRESS_TEST);
> +
>   	igt_fixture {
>   		xe_eudebug_enable(fd, was_enabled);
>