[PATCH 15/15] drm/i915: Allow user control over preempt timeout on the important context

[Chris Wilson <chris@chris-wilson.co.uk>]

EGL_NV_realtime_priority?

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_context.c    | 22 ++++++++
 drivers/gpu/drm/i915/i915_gem_context.h    | 13 +++++
 drivers/gpu/drm/i915/i915_request.c        |  8 ++-
 drivers/gpu/drm/i915/intel_lrc.c           |  2 +-
 drivers/gpu/drm/i915/selftests/intel_lrc.c | 85 ++++++++++++++++++++++++++++++
 include/uapi/drm/i915_drm.h                | 12 +++++
 6 files changed, 139 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 5cfac0255758..dfb0a2b698c3 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -749,6 +749,15 @@ int i915_gem_context_getparam_ioctl(struct drm_device *dev, void *data,
 	case I915_CONTEXT_PARAM_PRIORITY:
 		args->value = ctx->priority;
 		break;
+	case I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
+		if (!(to_i915(dev)->caps.scheduler & I915_SCHEDULER_CAP_PREEMPTION))
+			ret = -ENODEV;
+		else if (args->size)
+			ret = -EINVAL;
+		else
+			args->value = ctx->preempt_timeout;
+		break;
+
 	default:
 		ret = -EINVAL;
 		break;
@@ -824,6 +833,19 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
 		}
 		break;
 
+	case I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
+		if (args->size)
+			ret = -EINVAL;
+		else if (args->value >= NSEC_PER_SEC)
+			ret = -EINVAL;
+		else if (!(to_i915(dev)->caps.scheduler & I915_SCHEDULER_CAP_PREEMPTION))
+			ret = -ENODEV;
+		else if (args->value && !capable(CAP_SYS_ADMIN))
+			ret = -EPERM;
+		else
+			ctx->preempt_timeout = args->value;
+		break;
+
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
index 7854262ddfd9..74d4cadd729e 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.h
+++ b/drivers/gpu/drm/i915/i915_gem_context.h
@@ -150,6 +150,19 @@ struct i915_gem_context {
 	 */
 	int priority;
 
+	/**
+	 * @preempt_timeout: QoS guarantee for the high priority context
+	 *
+	 * Some clients need a guarantee that they will start executing
+	 * within a certain window, even at the expense of others. This entails
+	 * that if a preemption request is not honoured by the active context
+	 * within the timeout, we will reset the GPU to evict the hog and
+	 * run the high priority context instead.
+	 *
+	 * Timeout is stored in nanoseconds; exclusive max of 1s.
+	 */
+	u32 preempt_timeout;
+
 	/** ggtt_offset_bias: placement restriction for context objects */
 	u32 ggtt_offset_bias;
 
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 9d8dcebd9649..2cd4ea75d127 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1101,8 +1101,12 @@ void __i915_request_add(struct i915_request *request, bool flush_caches)
 	 * run at the earliest possible convenience.
 	 */
 	rcu_read_lock();
-	if (engine->schedule)
-		engine->schedule(request, request->ctx->priority, 0);
+	if (engine->schedule) {
+		unsigned int timeout = request->ctx->preempt_timeout;
+		int priority = request->ctx->priority;
+
+		engine->schedule(request, priority, timeout);
+	}
 	rcu_read_unlock();
 
 	local_bh_disable();
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 37150ad421ad..d1569928e300 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1192,7 +1192,7 @@ static void execlists_submit_request(struct i915_request *request)
 	spin_lock_irqsave(&engine->timeline->lock, flags);
 
 	queue_request(engine, &request->priotree, rq_prio(request));
-	submit_queue(engine, rq_prio(request), 0);
+	submit_queue(engine, rq_prio(request), request->ctx->preempt_timeout);
 
 	GEM_BUG_ON(!engine->execlists.first);
 	GEM_BUG_ON(list_empty(&request->priotree.link));
diff --git a/drivers/gpu/drm/i915/selftests/intel_lrc.c b/drivers/gpu/drm/i915/selftests/intel_lrc.c
index d942afb40bc6..5cd355c6a2da 100644
--- a/drivers/gpu/drm/i915/selftests/intel_lrc.c
+++ b/drivers/gpu/drm/i915/selftests/intel_lrc.c
@@ -727,6 +727,90 @@ static int live_late_preempt_timeout(void *arg)
 	goto err_ctx_lo;
 }
 
+static int live_context_preempt_timeout(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct i915_gem_context *ctx_hi, *ctx_lo;
+	struct spinner spin_hi, spin_lo;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	int err = -ENOMEM;
+
+	if (!HAS_LOGICAL_RING_PREEMPTION(i915))
+		return 0;
+
+	mutex_lock(&i915->drm.struct_mutex);
+
+	if (spinner_init(&spin_hi, i915))
+		goto err_unlock;
+
+	if (spinner_init(&spin_lo, i915))
+		goto err_spin_hi;
+
+	ctx_hi = kernel_context(i915);
+	if (!ctx_hi)
+		goto err_spin_lo;
+	ctx_hi->priority = I915_CONTEXT_MAX_USER_PRIORITY;
+	ctx_hi->preempt_timeout = 50 * 1000; /* 50us */
+
+	ctx_lo = kernel_context(i915);
+	if (!ctx_lo)
+		goto err_ctx_hi;
+	ctx_lo->priority = I915_CONTEXT_MIN_USER_PRIORITY;
+
+	for_each_engine(engine, i915, id) {
+		struct i915_request *rq;
+
+		rq = spinner_create_request(&spin_lo, ctx_lo, engine, MI_NOOP);
+		if (IS_ERR(rq)) {
+			err = PTR_ERR(rq);
+			goto err_ctx_lo;
+		}
+
+		i915_request_add(rq);
+		if (!wait_for_spinner(&spin_lo, rq)) {
+			i915_gem_set_wedged(i915);
+			err = -EIO;
+			goto err_ctx_lo;
+		}
+
+		rq = spinner_create_request(&spin_hi, ctx_hi, engine, MI_NOOP);
+		if (IS_ERR(rq)) {
+			spinner_end(&spin_lo);
+			err = PTR_ERR(rq);
+			goto err_ctx_lo;
+		}
+
+		i915_request_add(rq);
+		if (!wait_for_spinner(&spin_hi, rq)) {
+			i915_gem_set_wedged(i915);
+			err = -EIO;
+			goto err_ctx_lo;
+		}
+
+		spinner_end(&spin_hi);
+		spinner_end(&spin_lo);
+		if (flush_test(i915, I915_WAIT_LOCKED)) {
+			err = -EIO;
+			goto err_ctx_lo;
+		}
+	}
+
+	err = 0;
+err_ctx_lo:
+	kernel_context_close(ctx_lo);
+err_ctx_hi:
+	kernel_context_close(ctx_hi);
+err_spin_lo:
+	spinner_fini(&spin_lo);
+err_spin_hi:
+	spinner_fini(&spin_hi);
+err_unlock:
+	flush_test(i915, I915_WAIT_LOCKED);
+	mutex_unlock(&i915->drm.struct_mutex);
+	return err;
+}
+
 int intel_execlists_live_selftests(struct drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
@@ -736,6 +820,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915)
 		SUBTEST(live_preempt_timeout),
 		SUBTEST(live_preempt_reset),
 		SUBTEST(live_late_preempt_timeout),
+		SUBTEST(live_context_preempt_timeout),
 	};
 	return i915_subtests(tests, i915);
 }
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 7f5634ce8e88..6f10bbe90304 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -1456,6 +1456,18 @@ struct drm_i915_gem_context_param {
 #define   I915_CONTEXT_MAX_USER_PRIORITY	1023 /* inclusive */
 #define   I915_CONTEXT_DEFAULT_PRIORITY		0
 #define   I915_CONTEXT_MIN_USER_PRIORITY	-1023 /* inclusive */
+
+/*
+ * I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
+ *
+ * Preemption timeout give in nanoseconds.
+ *
+ * Only allowed for privileged clients (CAP_SYS_ADMIN), this property allows
+ * the preempting context to kick out a GPU hog using a GPU reset if they do
+ * not honour the preemption request in time.
+ */
+#define I915_CONTEXT_PARAM_PREEMPT_TIMEOUT	0x7
+
 	__u64 value;
 };
 
-- 
2.16.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx