[Intel-gfx] [RESEND 2/6] drm/i915/selftests: Exercise reset to break stuck GTT eviction

[Chris Wilson]

We must be able to reset the GPU while we are waiting on it to perform
an eviction (unbinding an active vma). So attach a spinning request to a
target vma and try and it evict it from a thread to see if that blocks
indefinitely.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 .../gpu/drm/i915/selftests/intel_hangcheck.c  | 153 +++++++++++++++++-
 1 file changed, 151 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 73462a65a330..344973664206 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -27,6 +27,7 @@
 #include "../i915_selftest.h"
 #include "i915_random.h"
 #include "igt_flush_test.h"
+#include "igt_wedge_me.h"
 
 #include "mock_context.h"
 #include "mock_drm.h"
@@ -921,7 +922,7 @@ static u32 fake_hangcheck(struct i915_request *rq, u32 mask)
 	return reset_count;
 }
 
-static int igt_wait_reset(void *arg)
+static int igt_reset_wait(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
 	struct i915_request *rq;
@@ -995,6 +996,152 @@ static int igt_wait_reset(void *arg)
 	return err;
 }
 
+static int evict_vma(void *data)
+{
+	struct i915_vma *vma = data;
+	struct drm_i915_private *i915 = vma->vm->i915;
+	struct drm_mm_node evict = vma->node;
+	int err;
+
+	mutex_lock(&i915->drm.struct_mutex);
+	err = i915_gem_evict_for_node(vma->vm, &evict, 0);
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	return err;
+}
+
+static int __igt_reset_evict_vma(struct drm_i915_private *i915,
+				 struct i915_address_space *vm)
+{
+	struct drm_i915_gem_object *obj;
+	struct task_struct *tsk = NULL;
+	struct i915_request *rq;
+	struct i915_vma *vma;
+	struct hang h;
+	int err;
+
+	if (!intel_engine_can_store_dword(i915->engine[RCS]))
+		return 0;
+
+	/* Check that we can recover an unbind stuck on a hanging request */
+
+	global_reset_lock(i915);
+
+	mutex_lock(&i915->drm.struct_mutex);
+	err = hang_init(&h, i915);
+	if (err)
+		goto unlock;
+
+	obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto fini;
+	}
+
+	vma = i915_vma_instance(obj, vm, NULL);
+	if (IS_ERR(vma)) {
+		err = PTR_ERR(vma);
+		goto out_obj;
+	}
+
+	rq = hang_create_request(&h, i915->engine[RCS]);
+	if (IS_ERR(rq)) {
+		err = PTR_ERR(rq);
+		goto out_obj;
+	}
+
+	err = i915_vma_pin(vma, 0, 0,
+			   i915_vma_is_ggtt(vma) ? PIN_GLOBAL : PIN_USER);
+	if (err)
+		goto out_obj;
+
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	i915_vma_unpin(vma);
+
+	i915_request_get(rq);
+	i915_request_add(rq);
+	if (err)
+		goto out_rq;
+
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	if (!wait_until_running(&h, rq)) {
+		struct drm_printer p = drm_info_printer(i915->drm.dev);
+
+		pr_err("%s: Failed to start request %x, at %x\n",
+		       __func__, rq->fence.seqno, hws_seqno(&h, rq));
+		intel_engine_dump(rq->engine, &p, "%s\n", rq->engine->name);
+
+		i915_gem_set_wedged(i915);
+		goto out_reset;
+	}
+
+	tsk = kthread_run(evict_vma, vma, "igt/evict_vma");
+
+	if (wait_for(waitqueue_active(&rq->execute), 10)) {
+		struct drm_printer p = drm_info_printer(i915->drm.dev);
+
+		pr_err("igt/evict_vma kthread did not wait\n");
+		intel_engine_dump(rq->engine, &p, "%s\n", rq->engine->name);
+
+		i915_gem_set_wedged(i915);
+		goto out_reset;
+	}
+
+out_reset:
+	fake_hangcheck(rq, intel_engine_flag(rq->engine));
+
+	if (tsk) {
+		struct igt_wedge_me w;
+
+		igt_wedge_on_timeout(&w, i915, HZ / 5)
+			err = kthread_stop(tsk);
+	}
+
+	mutex_lock(&i915->drm.struct_mutex);
+out_rq:
+	i915_request_put(rq);
+out_obj:
+	i915_gem_object_put(obj);
+fini:
+	hang_fini(&h);
+unlock:
+	mutex_unlock(&i915->drm.struct_mutex);
+	global_reset_unlock(i915);
+
+	if (i915_terminally_wedged(&i915->gpu_error))
+		return -EIO;
+
+	return err;
+}
+
+static int igt_reset_evict_ggtt(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+
+	return __igt_reset_evict_vma(i915, &i915->ggtt.vm);
+}
+
+static int igt_reset_evict_ppgtt(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct i915_gem_context *ctx;
+	int err;
+
+	mutex_lock(&i915->drm.struct_mutex);
+	ctx = kernel_context(i915);
+	mutex_unlock(&i915->drm.struct_mutex);
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
+
+	err = 0;
+	if (ctx->ppgtt)
+		err = __igt_reset_evict_vma(i915, &ctx->ppgtt->vm);
+
+	kernel_context_close(ctx);
+	return err;
+}
+
 static int wait_for_others(struct drm_i915_private *i915,
 			   struct intel_engine_cs *exclude)
 {
@@ -1240,8 +1387,10 @@ int intel_hangcheck_live_selftests(struct drm_i915_private *i915)
 		SUBTEST(igt_reset_idle_engine),
 		SUBTEST(igt_reset_active_engine),
 		SUBTEST(igt_reset_engines),
-		SUBTEST(igt_wait_reset),
 		SUBTEST(igt_reset_queue),
+		SUBTEST(igt_reset_wait),
+		SUBTEST(igt_reset_evict_ggtt),
+		SUBTEST(igt_reset_evict_ppgtt),
 		SUBTEST(igt_handle_error),
 	};
 	bool saved_hangcheck;
-- 
2.18.0