[PATCH] tests/xe_spin_batch: Add spin-timestamp-check
Zbigniew Kempczyński
zbigniew.kempczynski at intel.com
Thu Nov 7 07:25:13 UTC 2024
On Tue, Nov 05, 2024 at 02:58:05PM +0530, Pravalika Gurram wrote:
> check the ctx_timestamp register post gt reset for each engine.
>
> Reference: VLK-64943
> Signed-off-by: Pravalika Gurram <pravalika.gurram at intel.com>
> Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski at intel.com>
> ---
> tests/intel/xe_spin_batch.c | 158 ++++++++++++++++++++++++++++++++++++
> 1 file changed, 158 insertions(+)
>
> diff --git a/tests/intel/xe_spin_batch.c b/tests/intel/xe_spin_batch.c
> index 9314e229e..547cb1c5b 100644
> --- a/tests/intel/xe_spin_batch.c
> +++ b/tests/intel/xe_spin_batch.c
> @@ -309,6 +309,158 @@ static void xe_spin_fixed_duration(int fd, int gt, int class, int flags)
> put_ahnd(ahnd);
> }
>
> +static void exec_store(int fd, struct drm_xe_engine_class_instance *eci,
> + bool hang)
> +{
> + uint64_t ahnd, bb_size, bb_addr;
> + uint32_t vm, exec_queue, bb;
> +#define USER_FENCE_VALUE 0xdeadbeefdeadbeefull
> + struct drm_xe_sync syncobj = {
> + .type = DRM_XE_SYNC_TYPE_USER_FENCE,
> + .flags = DRM_XE_SYNC_FLAG_SIGNAL,
> + .timeline_value = USER_FENCE_VALUE,
> + };
> +
> + struct drm_xe_exec exec = {
> + .num_batch_buffer = 1,
> + .num_syncs = 1,
> + .syncs = to_user_pointer(&syncobj),
> + };
> + struct {
> + uint32_t batch[16];
> + uint64_t pad;
> + uint32_t data;
> + uint64_t vm_sync;
> + uint64_t exec_sync;
> + } *data;
> + uint64_t batch_offset, batch_addr, sdi_offset, sdi_addr;
> + int64_t timeout = NSEC_PER_SEC;
> + int i, ret;
> +
> + ahnd = intel_allocator_open(fd, 0, INTEL_ALLOCATOR_RELOC);
> +
> + vm = xe_vm_create(fd, 0, 0);
> + exec_queue = xe_exec_queue_create(fd, vm, eci, 0);
> + bb_size = xe_bb_size(fd, sizeof(*data));
> + bb = xe_bo_create(fd, vm, bb_size, vram_if_possible(fd, eci->gt_id), 0);
> + bb_addr = intel_allocator_alloc_with_strategy(ahnd, bb, bb_size, 0,
> + ALLOC_STRATEGY_LOW_TO_HIGH);
> + data = xe_bo_map(fd, bb, bb_size);
> + syncobj.addr = to_user_pointer(&data->vm_sync);
> + xe_vm_bind_async(fd, vm, 0, bb, 0, bb_addr, bb_size, &syncobj, 1);
> + xe_wait_ufence(fd, &data->vm_sync, USER_FENCE_VALUE, 0, NSEC_PER_SEC);
> +
> + batch_offset = (char *)&data->batch - (char *)data;
> + batch_addr = bb_addr + batch_offset;
> + sdi_offset = (char *)&data->data - (char *)data;
> + sdi_addr = bb_addr + sdi_offset;
> +
> + i = 0;
> +
> + data->batch[i++] = MI_STORE_DWORD_IMM_GEN4;
> + data->batch[i++] = sdi_addr;
> + data->batch[i++] = sdi_addr >> 32;
> + data->batch[i++] = 0;
> + if (!hang)
> + data->batch[i++] = MI_BATCH_BUFFER_END;
> + igt_assert(i <= ARRAY_SIZE(data->batch));
> +
> + syncobj.addr = bb_addr + (char *)&data->exec_sync - (char *)data;
> + exec.exec_queue_id = exec_queue;
> + exec.address = batch_addr;
> + ret = __xe_exec(fd, &exec);
Just call xe_exec() here, it must succeed.
--
Zbigniew
> + ret = __xe_wait_ufence(fd, &data->exec_sync, USER_FENCE_VALUE, 0, &timeout);
> + igt_assert(hang ? ret < 0 : ret == 0);
> +
> + munmap(data, bb_size);
> + gem_close(fd, bb);
> +
> + xe_exec_queue_destroy(fd, exec_queue);
> + xe_vm_destroy(fd, vm);
> +
> + put_ahnd(ahnd);
> +}
> +
> +static void run_spinner(int fd, struct drm_xe_engine_class_instance *eci)
> +{
> + struct drm_xe_sync sync = {
> + .type = DRM_XE_SYNC_TYPE_USER_FENCE,
> + .flags = DRM_XE_SYNC_FLAG_SIGNAL,
> + .timeline_value = USER_FENCE_VALUE,
> + };
> + struct drm_xe_exec exec = {
> + .num_batch_buffer = 1,
> + .num_syncs = 1,
> + .syncs = to_user_pointer(&sync),
> + };
> + struct xe_spin *spin;
> + uint64_t vm_sync;
> + size_t bo_size;
> + uint32_t vm;
> + uint32_t exec_queue;
> + uint64_t spin_addr;
> + uint64_t ahnd;
> + uint32_t bo;
> + uint32_t ts_1, ts_2;
> +
> + vm = xe_vm_create(fd, 0, 0);
> + ahnd = intel_allocator_open(fd, 0, INTEL_ALLOCATOR_RELOC);
> + bo_size = xe_bb_size(fd, sizeof(*spin));
> + bo = xe_bo_create(fd, vm, bo_size, vram_if_possible(fd, eci->gt_id), 0);
> + spin = xe_bo_map(fd, bo, bo_size);
> +
> + exec_queue = xe_exec_queue_create(fd, vm, eci, 0);
> + spin_addr = intel_allocator_alloc_with_strategy(ahnd, bo, bo_size, 0,
> + ALLOC_STRATEGY_LOW_TO_HIGH);
> +
> + sync.addr = to_user_pointer(&vm_sync);
> + xe_vm_bind_async(fd, vm, 0, bo, 0, spin_addr, bo_size, &sync, 1);
> + xe_wait_ufence(fd, &vm_sync, USER_FENCE_VALUE, 0, NSEC_PER_SEC);
> +
> +
> + xe_spin_init_opts(spin, .addr = spin_addr, .write_timestamp = true);
> + sync.addr = spin_addr + (char *)&spin->exec_sync - (char *)spin;
> + exec.exec_queue_id = exec_queue;
> + exec.address = spin_addr;
> + xe_exec(fd, &exec);
> + xe_spin_wait_started(spin);
> +
> + /* Collect and check timestamps before stopping the spinner */
> + usleep(50000);
> + ts_1 = spin->timestamp;
> + usleep(50000);
> + ts_2 = spin->timestamp;
> + igt_assert_neq_u32(ts_1, ts_2);
> +
> + xe_spin_end(spin);
> + xe_wait_ufence(fd, &spin->exec_sync, USER_FENCE_VALUE, 0, NSEC_PER_SEC);
> +
> + sync.addr = to_user_pointer(&vm_sync);
> + xe_vm_unbind_async(fd, vm, 0, 0, spin_addr, bo_size, &sync, 1);
> + xe_wait_ufence(fd, &vm_sync, USER_FENCE_VALUE, 0, NSEC_PER_SEC);
> + munmap(spin, bo_size);
> + gem_close(fd, bo);
> +
> + xe_exec_queue_destroy(fd, exec_queue);
> + xe_vm_destroy(fd, vm);
> + put_ahnd(ahnd);
> +}
> +
> +/**
> + * SUBTEST: spin-timestamp-check
> + * Description: Intiate gt reset then check the timestamp register for each engine.
> + * Test category: functionality test
> + */
> +static void xe_spin_timestamp_check(int fd, struct drm_xe_engine_class_instance *eci)
> +{
> +
> + exec_store(fd, eci, false);
> +
> + exec_store(fd, eci, true);
> +
> + run_spinner(fd, eci);
> +}
> +
> igt_main
> {
> struct drm_xe_engine_class_instance *hwe;
> @@ -343,6 +495,12 @@ igt_main
> xe_for_each_engine_class(class)
> xe_spin_fixed_duration(fd, gt, class, SPIN_FIX_DURATION_PREEMPT);
>
> + igt_subtest_with_dynamic("spin-timestamp-check")
> + xe_for_each_engine(fd, hwe) {
> + igt_dynamic_f("engine-%s", xe_engine_class_string(hwe->engine_class))
> + xe_spin_timestamp_check(fd, hwe);
> + }
> +
> igt_fixture
> drm_close_driver(fd);
> }
> --
> 2.34.1
>
More information about the igt-dev
mailing list