1// SPDX-License-Identifier: MIT
  2/*
  3 * Copyright © 2022 Intel Corporation
  4 */
  5
  6#include "selftests/igt_spinner.h"
  7#include "selftests/igt_reset.h"
  8#include "selftests/intel_scheduler_helpers.h"
  9#include "gt/intel_engine_heartbeat.h"
 10#include "gem/selftests/mock_context.h"
 11
 12#define BEAT_INTERVAL	100
 13
 14static struct i915_request *nop_request(struct intel_engine_cs *engine)
 15{
 16	struct i915_request *rq;
 17
 18	rq = intel_engine_create_kernel_request(engine);
 19	if (IS_ERR(rq))
 20		return rq;
 21
 22	i915_request_get(rq);
 23	i915_request_add(rq);
 24
 25	return rq;
 26}
 27
 28static int intel_hang_guc(void *arg)
 29{
 30	struct intel_gt *gt = arg;
 31	int ret = 0;
 32	struct i915_gem_context *ctx;
 33	struct intel_context *ce;
 34	struct igt_spinner spin;
 35	struct i915_request *rq;
 36	intel_wakeref_t wakeref;
 37	struct i915_gpu_error *global = &gt->i915->gpu_error;
 38	struct intel_engine_cs *engine = intel_selftest_find_any_engine(gt);
 39	unsigned int reset_count;
 40	u32 guc_status;
 41	u32 old_beat;
 42
 43	if (!engine)
 44		return 0;
 45
 46	ctx = kernel_context(gt->i915, NULL);
 47	if (IS_ERR(ctx)) {
 48		drm_err(&gt->i915->drm, "Failed get kernel context: %ld\n", PTR_ERR(ctx));
 49		return PTR_ERR(ctx);
 50	}
 51
 52	wakeref = intel_runtime_pm_get(gt->uncore->rpm);
 53
 54	ce = intel_context_create(engine);
 55	if (IS_ERR(ce)) {
 56		ret = PTR_ERR(ce);
 57		drm_err(&gt->i915->drm, "Failed to create spinner request: %d\n", ret);
 58		goto err;
 59	}
 60
 61	reset_count = i915_reset_count(global);
 62
 63	old_beat = engine->props.heartbeat_interval_ms;
 64	ret = intel_engine_set_heartbeat(engine, BEAT_INTERVAL);
 65	if (ret) {
 66		drm_err(&gt->i915->drm, "Failed to boost heatbeat interval: %d\n", ret);
 67		goto err;
 68	}
 69
 70	ret = igt_spinner_init(&spin, engine->gt);
 71	if (ret) {
 72		drm_err(&gt->i915->drm, "Failed to create spinner: %d\n", ret);
 73		goto err;
 74	}
 75
 76	rq = igt_spinner_create_request(&spin, ce, MI_NOOP);
 77	intel_context_put(ce);
 78	if (IS_ERR(rq)) {
 79		ret = PTR_ERR(rq);
 80		drm_err(&gt->i915->drm, "Failed to create spinner request: %d\n", ret);
 81		goto err_spin;
 82	}
 83
 84	ret = request_add_spin(rq, &spin);
 85	if (ret) {
 86		i915_request_put(rq);
 87		drm_err(&gt->i915->drm, "Failed to add Spinner request: %d\n", ret);
 88		goto err_spin;
 89	}
 90
 91	ret = intel_reset_guc(gt);
 92	if (ret) {
 93		i915_request_put(rq);
 94		drm_err(&gt->i915->drm, "Failed to reset GuC, ret = %d\n", ret);
 95		goto err_spin;
 96	}
 97
 98	guc_status = intel_uncore_read(gt->uncore, GUC_STATUS);
 99	if (!(guc_status & GS_MIA_IN_RESET)) {
100		i915_request_put(rq);
101		drm_err(&gt->i915->drm, "GuC failed to reset: status = 0x%08X\n", guc_status);
102		ret = -EIO;
103		goto err_spin;
104	}
105
106	/* Wait for the heartbeat to cause a reset */
107	ret = intel_selftest_wait_for_rq(rq);
108	i915_request_put(rq);
109	if (ret) {
110		drm_err(&gt->i915->drm, "Request failed to complete: %d\n", ret);
111		goto err_spin;
112	}
113
114	if (i915_reset_count(global) == reset_count) {
115		drm_err(&gt->i915->drm, "Failed to record a GPU reset\n");
116		ret = -EINVAL;
117		goto err_spin;
118	}
119
120err_spin:
121	igt_spinner_end(&spin);
122	igt_spinner_fini(&spin);
123	intel_engine_set_heartbeat(engine, old_beat);
124
125	if (ret == 0) {
126		rq = nop_request(engine);
127		if (IS_ERR(rq)) {
128			ret = PTR_ERR(rq);
129			goto err;
130		}
131
132		ret = intel_selftest_wait_for_rq(rq);
133		i915_request_put(rq);
134		if (ret) {
135			drm_err(&gt->i915->drm, "No-op failed to complete: %d\n", ret);
136			goto err;
137		}
138	}
139
140err:
141	intel_runtime_pm_put(gt->uncore->rpm, wakeref);
142	kernel_context_close(ctx);
143
144	return ret;
145}
146
147int intel_guc_hang_check(struct drm_i915_private *i915)
148{
149	static const struct i915_subtest tests[] = {
150		SUBTEST(intel_hang_guc),
151	};
152	struct intel_gt *gt = to_gt(i915);
153
154	if (intel_gt_is_wedged(gt))
155		return 0;
156
157	if (!intel_uc_uses_guc_submission(&gt->uc))
158		return 0;
159
160	return intel_gt_live_subtests(tests, gt);
161}