drm/i915/guc: Add selftest for a hung GuC
authorRahul Kumar Singh <rahul.kumar.singh@intel.com>
Thu, 28 Jul 2022 18:26:16 +0000 (11:26 -0700)
committerJohn Harrison <John.C.Harrison@Intel.com>
Fri, 29 Jul 2022 17:35:55 +0000 (10:35 -0700)
Add a test to check that the hangcheck will recover from a submission
hang in the GuC.

Signed-off-by: Rahul Kumar Singh <rahul.kumar.singh@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728182616.2417491-1-John.C.Harrison@Intel.com
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
drivers/gpu/drm/i915/gt/uc/selftest_guc_hangcheck.c [new file with mode: 0644]
drivers/gpu/drm/i915/selftests/i915_live_selftests.h

index 0b8c645..ff205c4 100644 (file)
@@ -5177,4 +5177,5 @@ bool intel_guc_virtual_engine_has_heartbeat(const struct intel_engine_cs *ve)
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftest_guc.c"
 #include "selftest_guc_multi_lrc.c"
+#include "selftest_guc_hangcheck.c"
 #endif
diff --git a/drivers/gpu/drm/i915/gt/uc/selftest_guc_hangcheck.c b/drivers/gpu/drm/i915/gt/uc/selftest_guc_hangcheck.c
new file mode 100644 (file)
index 0000000..01f8cd3
--- /dev/null
@@ -0,0 +1,159 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "selftests/igt_spinner.h"
+#include "selftests/igt_reset.h"
+#include "selftests/intel_scheduler_helpers.h"
+#include "gt/intel_engine_heartbeat.h"
+#include "gem/selftests/mock_context.h"
+
+#define BEAT_INTERVAL  100
+
+static struct i915_request *nop_request(struct intel_engine_cs *engine)
+{
+       struct i915_request *rq;
+
+       rq = intel_engine_create_kernel_request(engine);
+       if (IS_ERR(rq))
+               return rq;
+
+       i915_request_get(rq);
+       i915_request_add(rq);
+
+       return rq;
+}
+
+static int intel_hang_guc(void *arg)
+{
+       struct intel_gt *gt = arg;
+       int ret = 0;
+       struct i915_gem_context *ctx;
+       struct intel_context *ce;
+       struct igt_spinner spin;
+       struct i915_request *rq;
+       intel_wakeref_t wakeref;
+       struct i915_gpu_error *global = &gt->i915->gpu_error;
+       struct intel_engine_cs *engine;
+       unsigned int reset_count;
+       u32 guc_status;
+       u32 old_beat;
+
+       ctx = kernel_context(gt->i915, NULL);
+       if (IS_ERR(ctx)) {
+               drm_err(&gt->i915->drm, "Failed get kernel context: %ld\n", PTR_ERR(ctx));
+               return PTR_ERR(ctx);
+       }
+
+       wakeref = intel_runtime_pm_get(gt->uncore->rpm);
+
+       ce = intel_context_create(gt->engine[BCS0]);
+       if (IS_ERR(ce)) {
+               ret = PTR_ERR(ce);
+               drm_err(&gt->i915->drm, "Failed to create spinner request: %d\n", ret);
+               goto err;
+       }
+
+       engine = ce->engine;
+       reset_count = i915_reset_count(global);
+
+       old_beat = engine->props.heartbeat_interval_ms;
+       ret = intel_engine_set_heartbeat(engine, BEAT_INTERVAL);
+       if (ret) {
+               drm_err(&gt->i915->drm, "Failed to boost heatbeat interval: %d\n", ret);
+               goto err;
+       }
+
+       ret = igt_spinner_init(&spin, engine->gt);
+       if (ret) {
+               drm_err(&gt->i915->drm, "Failed to create spinner: %d\n", ret);
+               goto err;
+       }
+
+       rq = igt_spinner_create_request(&spin, ce, MI_NOOP);
+       intel_context_put(ce);
+       if (IS_ERR(rq)) {
+               ret = PTR_ERR(rq);
+               drm_err(&gt->i915->drm, "Failed to create spinner request: %d\n", ret);
+               goto err_spin;
+       }
+
+       ret = request_add_spin(rq, &spin);
+       if (ret) {
+               i915_request_put(rq);
+               drm_err(&gt->i915->drm, "Failed to add Spinner request: %d\n", ret);
+               goto err_spin;
+       }
+
+       ret = intel_reset_guc(gt);
+       if (ret) {
+               i915_request_put(rq);
+               drm_err(&gt->i915->drm, "Failed to reset GuC, ret = %d\n", ret);
+               goto err_spin;
+       }
+
+       guc_status = intel_uncore_read(gt->uncore, GUC_STATUS);
+       if (!(guc_status & GS_MIA_IN_RESET)) {
+               i915_request_put(rq);
+               drm_err(&gt->i915->drm, "GuC failed to reset: status = 0x%08X\n", guc_status);
+               ret = -EIO;
+               goto err_spin;
+       }
+
+       /* Wait for the heartbeat to cause a reset */
+       ret = intel_selftest_wait_for_rq(rq);
+       i915_request_put(rq);
+       if (ret) {
+               drm_err(&gt->i915->drm, "Request failed to complete: %d\n", ret);
+               goto err_spin;
+       }
+
+       if (i915_reset_count(global) == reset_count) {
+               drm_err(&gt->i915->drm, "Failed to record a GPU reset\n");
+               ret = -EINVAL;
+               goto err_spin;
+       }
+
+err_spin:
+       igt_spinner_end(&spin);
+       igt_spinner_fini(&spin);
+       intel_engine_set_heartbeat(engine, old_beat);
+
+       if (ret == 0) {
+               rq = nop_request(engine);
+               if (IS_ERR(rq)) {
+                       ret = PTR_ERR(rq);
+                       goto err;
+               }
+
+               ret = intel_selftest_wait_for_rq(rq);
+               i915_request_put(rq);
+               if (ret) {
+                       drm_err(&gt->i915->drm, "No-op failed to complete: %d\n", ret);
+                       goto err;
+               }
+       }
+
+err:
+       intel_runtime_pm_put(gt->uncore->rpm, wakeref);
+       kernel_context_close(ctx);
+
+       return ret;
+}
+
+int intel_guc_hang_check(struct drm_i915_private *i915)
+{
+       static const struct i915_subtest tests[] = {
+               SUBTEST(intel_hang_guc),
+       };
+       struct intel_gt *gt = to_gt(i915);
+
+       if (intel_gt_is_wedged(gt))
+               return 0;
+
+       if (!intel_uc_uses_guc_submission(&gt->uc))
+               return 0;
+
+       return intel_gt_live_subtests(tests, gt);
+}
index bdd290f..aaf8a38 100644 (file)
@@ -49,5 +49,6 @@ selftest(perf, i915_perf_live_selftests)
 selftest(slpc, intel_slpc_live_selftests)
 selftest(guc, intel_guc_live_selftests)
 selftest(guc_multi_lrc, intel_guc_multi_lrc_live_selftests)
+selftest(guc_hang, intel_guc_hang_check)
 /* Here be dragons: keep last to run last! */
 selftest(late_gt_pm, intel_gt_pm_late_selftests)