1 files changed, 161 insertions, 76 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 7b5a9d7c9593..4c9d2a6f7d28 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -35,6 +35,7 @@
 #include "intel_drv.h"
 #include "intel_frontbuffer.h"
 #include "intel_mocs.h"
+#include "intel_workarounds.h"
 #include "i915_gemfs.h"
 #include <linux/dma-fence-array.h>
 #include <linux/kthread.h>
@@ -136,6 +137,100 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)
 	return 0;
 }
 
+static u32 __i915_gem_park(struct drm_i915_private *i915)
+{
+	lockdep_assert_held(&i915->drm.struct_mutex);
+	GEM_BUG_ON(i915->gt.active_requests);
+
+	if (!i915->gt.awake)
+		return I915_EPOCH_INVALID;
+
+	GEM_BUG_ON(i915->gt.epoch == I915_EPOCH_INVALID);
+
+	/*
+	 * Be paranoid and flush a concurrent interrupt to make sure
+	 * we don't reactivate any irq tasklets after parking.
+	 *
+	 * FIXME: Note that even though we have waited for execlists to be idle,
+	 * there may still be an in-flight interrupt even though the CSB
+	 * is now empty. synchronize_irq() makes sure that a residual interrupt
+	 * is completed before we continue, but it doesn't prevent the HW from
+	 * raising a spurious interrupt later. To complete the shield we should
+	 * coordinate disabling the CS irq with flushing the interrupts.
+	 */
+	synchronize_irq(i915->drm.irq);
+
+	intel_engines_park(i915);
+	i915_gem_timelines_park(i915);
+
+	i915_pmu_gt_parked(i915);
+
+	i915->gt.awake = false;
+
+	if (INTEL_GEN(i915) >= 6)
+		gen6_rps_idle(i915);
+
+	intel_display_power_put(i915, POWER_DOMAIN_GT_IRQ);
+
+	intel_runtime_pm_put(i915);
+
+	return i915->gt.epoch;
+}
+
+void i915_gem_park(struct drm_i915_private *i915)
+{
+	lockdep_assert_held(&i915->drm.struct_mutex);
+	GEM_BUG_ON(i915->gt.active_requests);
+
+	if (!i915->gt.awake)
+		return;
+
+	/* Defer the actual call to __i915_gem_park() to prevent ping-pongs */
+	mod_delayed_work(i915->wq, &i915->gt.idle_work, msecs_to_jiffies(100));
+}
+
+void i915_gem_unpark(struct drm_i915_private *i915)
+{
+	lockdep_assert_held(&i915->drm.struct_mutex);
+	GEM_BUG_ON(!i915->gt.active_requests);
+
+	if (i915->gt.awake)
+		return;
+
+	intel_runtime_pm_get_noresume(i915);
+
+	/*
+	 * It seems that the DMC likes to transition between the DC states a lot
+	 * when there are no connected displays (no active power domains) during
+	 * command submission.
+	 *
+	 * This activity has negative impact on the performance of the chip with
+	 * huge latencies observed in the interrupt handler and elsewhere.
+	 *
+	 * Work around it by grabbing a GT IRQ power domain whilst there is any
+	 * GT activity, preventing any DC state transitions.
+	 */
+	intel_display_power_get(i915, POWER_DOMAIN_GT_IRQ);
+
+	i915->gt.awake = true;
+	if (unlikely(++i915->gt.epoch == 0)) /* keep 0 as invalid */
+		i915->gt.epoch = 1;
+
+	intel_enable_gt_powersave(i915);
+	i915_update_gfx_val(i915);
+	if (INTEL_GEN(i915) >= 6)
+		gen6_rps_busy(i915);
+	i915_pmu_gt_unparked(i915);
+
+	intel_engines_unpark(i915);
+
+	i915_queue_hangcheck(i915);
+
+	queue_delayed_work(i915->wq,
+			   &i915->gt.retire_work,
+			   round_jiffies_up_relative(HZ));
+}
+
 int
 i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 			    struct drm_file *file)
@@ -2896,20 +2991,6 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
 	return active;
 }
 
-static bool engine_stalled(struct intel_engine_cs *engine)
-{
-	if (!engine->hangcheck.stalled)
-		return false;
-
-	/* Check for possible seqno movement after hang declaration */
-	if (engine->hangcheck.seqno != intel_engine_get_seqno(engine)) {
-		DRM_DEBUG_DRIVER("%s pardoned\n", engine->name);
-		return false;
-	}
-
-	return true;
-}
-
 /*
  * Ensure irq handler finishes, and not run again.
  * Also return the active request so that we only search for it once.
@@ -2998,6 +3079,7 @@ int i915_gem_reset_prepare(struct drm_i915_private *dev_priv)
 	}
 
 	i915_gem_revoke_fences(dev_priv);
+	intel_uc_sanitize(dev_priv);
 
 	return err;
 }
@@ -3047,7 +3129,8 @@ static void engine_skip_context(struct i915_request *request)
 /* Returns the request if it was guilty of the hang */
 static struct i915_request *
 i915_gem_reset_request(struct intel_engine_cs *engine,
-		       struct i915_request *request)
+		       struct i915_request *request,
+		       bool stalled)
 {
 	/* The guilty request will get skipped on a hung engine.
 	 *
@@ -3070,7 +3153,15 @@ i915_gem_reset_request(struct intel_engine_cs *engine,
 	 * subsequent hangs.
 	 */
 
-	if (engine_stalled(engine)) {
+	if (i915_request_completed(request)) {
+		GEM_TRACE("%s pardoned global=%d (fence %llx:%d), current %d\n",
+			  engine->name, request->global_seqno,
+			  request->fence.context, request->fence.seqno,
+			  intel_engine_get_seqno(engine));
+		stalled = false;
+	}
+
+	if (stalled) {
 		i915_gem_context_mark_guilty(request->ctx);
 		skip_request(request);
 
@@ -3101,7 +3192,8 @@ i915_gem_reset_request(struct intel_engine_cs *engine,
 }
 
 void i915_gem_reset_engine(struct intel_engine_cs *engine,
-			   struct i915_request *request)
+			   struct i915_request *request,
+			   bool stalled)
 {
 	/*
 	 * Make sure this write is visible before we re-enable the interrupt
@@ -3111,7 +3203,7 @@ void i915_gem_reset_engine(struct intel_engine_cs *engine,
 	smp_store_mb(engine->irq_posted, 0);
 
 	if (request)
-		request = i915_gem_reset_request(engine, request);
+		request = i915_gem_reset_request(engine, request, stalled);
 
 	if (request) {
 		DRM_DEBUG_DRIVER("resetting %s to restart from tail of request 0x%x\n",
@@ -3122,7 +3214,8 @@ void i915_gem_reset_engine(struct intel_engine_cs *engine,
 	engine->reset_hw(engine, request);
 }
 
-void i915_gem_reset(struct drm_i915_private *dev_priv)
+void i915_gem_reset(struct drm_i915_private *dev_priv,
+		    unsigned int stalled_mask)
 {
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
@@ -3134,7 +3227,9 @@ void i915_gem_reset(struct drm_i915_private *dev_priv)
 	for_each_engine(engine, dev_priv, id) {
 		struct i915_gem_context *ctx;
 
-		i915_gem_reset_engine(engine, engine->hangcheck.active_request);
+		i915_gem_reset_engine(engine,
+				      engine->hangcheck.active_request,
+				      stalled_mask & ENGINE_MASK(id));
 		ctx = fetch_and_zero(&engine->last_retired_context);
 		if (ctx)
 			engine->context_unpin(engine, ctx);
@@ -3160,13 +3255,6 @@ void i915_gem_reset(struct drm_i915_private *dev_priv)
 	}
 
 	i915_gem_restore_fences(dev_priv);
-
-	if (dev_priv->gt.awake) {
-		intel_sanitize_gt_powersave(dev_priv);
-		intel_enable_gt_powersave(dev_priv);
-		if (INTEL_GEN(dev_priv) >= 6)
-			gen6_rps_busy(dev_priv);
-	}
 }
 
 void i915_gem_reset_finish_engine(struct intel_engine_cs *engine)
@@ -3192,6 +3280,9 @@ void i915_gem_reset_finish(struct drm_i915_private *dev_priv)
 
 static void nop_submit_request(struct i915_request *request)
 {
+	GEM_TRACE("%s fence %llx:%d -> -EIO\n",
+		  request->engine->name,
+		  request->fence.context, request->fence.seqno);
 	dma_fence_set_error(&request->fence, -EIO);
 
 	i915_request_submit(request);
@@ -3201,6 +3292,9 @@ static void nop_complete_submit_request(struct i915_request *request)
 {
 	unsigned long flags;
 
+	GEM_TRACE("%s fence %llx:%d -> -EIO\n",
+		  request->engine->name,
+		  request->fence.context, request->fence.seqno);
 	dma_fence_set_error(&request->fence, -EIO);
 
 	spin_lock_irqsave(&request->engine->timeline->lock, flags);
@@ -3214,6 +3308,8 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 
+	GEM_TRACE("start\n");
+
 	if (drm_debug & DRM_UT_DRIVER) {
 		struct drm_printer p = drm_debug_printer(__func__);
 
@@ -3237,6 +3333,9 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 	}
 	i915->caps.scheduler = 0;
 
+	/* Even if the GPU reset fails, it should still stop the engines */
+	intel_gpu_reset(i915, ALL_ENGINES);
+
 	/*
 	 * Make sure no one is running the old callback before we proceed with
 	 * cancelling requests and resetting the completion tracking. Otherwise
@@ -3278,6 +3377,8 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 		i915_gem_reset_finish_engine(engine);
 	}
 
+	GEM_TRACE("end\n");
+
 	wake_up_all(&i915->gpu_error.reset_queue);
 }
 
@@ -3290,7 +3391,10 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	if (!test_bit(I915_WEDGED, &i915->gpu_error.flags))
 		return true;
 
-	/* Before unwedging, make sure that all pending operations
+	GEM_TRACE("start\n");
+
+	/*
+	 * Before unwedging, make sure that all pending operations
 	 * are flushed and errored out - we may have requests waiting upon
 	 * third party fences. We marked all inflight requests as EIO, and
 	 * every execbuf since returned EIO, for consistency we want all
@@ -3308,7 +3412,8 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 			if (!rq)
 				continue;
 
-			/* We can't use our normal waiter as we want to
+			/*
+			 * We can't use our normal waiter as we want to
 			 * avoid recursively trying to handle the current
 			 * reset. The basic dma_fence_default_wait() installs
 			 * a callback for dma_fence_signal(), which is
@@ -3323,8 +3428,11 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 				return false;
 		}
 	}
+	i915_retire_requests(i915);
+	GEM_BUG_ON(i915->gt.active_requests);
 
-	/* Undo nop_submit_request. We prevent all new i915 requests from
+	/*
+	 * Undo nop_submit_request. We prevent all new i915 requests from
 	 * being queued (by disallowing execbuf whilst wedged) so having
 	 * waited for all active requests above, we know the system is idle
 	 * and do not have to worry about a thread being inside
@@ -3335,6 +3443,8 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	intel_engines_reset_default_submission(i915);
 	i915_gem_contexts_lost(i915);
 
+	GEM_TRACE("end\n");
+
 	smp_mb__before_atomic(); /* complete takeover before enabling execbuf */
 	clear_bit(I915_WEDGED, &i915->gpu_error.flags);
 
@@ -3473,36 +3583,9 @@ i915_gem_idle_work_handler(struct work_struct *work)
 	if (new_requests_since_last_retire(dev_priv))
 		goto out_unlock;
 
-	/*
-	 * Be paranoid and flush a concurrent interrupt to make sure
-	 * we don't reactivate any irq tasklets after parking.
-	 *
-	 * FIXME: Note that even though we have waited for execlists to be idle,
-	 * there may still be an in-flight interrupt even though the CSB
-	 * is now empty. synchronize_irq() makes sure that a residual interrupt
-	 * is completed before we continue, but it doesn't prevent the HW from
-	 * raising a spurious interrupt later. To complete the shield we should
-	 * coordinate disabling the CS irq with flushing the interrupts.
-	 */
-	synchronize_irq(dev_priv->drm.irq);
-
-	intel_engines_park(dev_priv);
-	i915_gem_timelines_park(dev_priv);
-
-	i915_pmu_gt_parked(dev_priv);
+	epoch = __i915_gem_park(dev_priv);
 
-	GEM_BUG_ON(!dev_priv->gt.awake);
-	dev_priv->gt.awake = false;
-	epoch = dev_priv->gt.epoch;
-	GEM_BUG_ON(epoch == I915_EPOCH_INVALID);
 	rearm_hangcheck = false;
-
-	if (INTEL_GEN(dev_priv) >= 6)
-		gen6_rps_idle(dev_priv);
-
-	intel_display_power_put(dev_priv, POWER_DOMAIN_GT_IRQ);
-
-	intel_runtime_pm_put(dev_priv);
 out_unlock:
 	mutex_unlock(&dev_priv->drm.struct_mutex);
 
@@ -3666,16 +3749,7 @@ static int wait_for_engines(struct drm_i915_private *i915)
 	if (wait_for(intel_engines_are_idle(i915), I915_IDLE_ENGINES_TIMEOUT)) {
 		dev_err(i915->drm.dev,
 			"Failed to idle engines, declaring wedged!\n");
-		if (drm_debug & DRM_UT_DRIVER) {
-			struct drm_printer p = drm_debug_printer(__func__);
-			struct intel_engine_cs *engine;
-			enum intel_engine_id id;
-
-			for_each_engine(engine, i915, id)
-				intel_engine_dump(engine, &p,
-						  "%s\n", engine->name);
-		}
-
+		GEM_TRACE_DUMP();
 		i915_gem_set_wedged(i915);
 		return -EIO;
 	}
@@ -4088,9 +4162,10 @@ out:
 }
 
 /*
- * Prepare buffer for display plane (scanout, cursors, etc).
- * Can be called from an uninterruptible phase (modesetting) and allows
- * any flushes to be pipelined (for pageflips).
+ * Prepare buffer for display plane (scanout, cursors, etc). Can be called from
+ * an uninterruptible phase (modesetting) and allows any flushes to be pipelined
+ * (for pageflips). We only flush the caches while preparing the buffer for
+ * display, the callers are responsible for frontbuffer flush.
  */
 struct i915_vma *
 i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
@@ -4146,9 +4221,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
 
 	vma->display_alignment = max_t(u64, vma->display_alignment, alignment);
 
-	/* Treat this as an end-of-frame, like intel_user_framebuffer_dirty() */
 	__i915_gem_object_flush_for_display(obj);
-	intel_fb_obj_flush(obj, ORIGIN_DIRTYFB);
 
 	/* It should now be out of any other write domains, and we can update
 	 * the domain values for our changes.
@@ -4973,6 +5046,7 @@ int i915_gem_suspend(struct drm_i915_private *dev_priv)
 	 * machines is a good idea, we don't - just in case it leaves the
 	 * machine in an unusable condition.
 	 */
+	intel_uc_sanitize(dev_priv);
 	i915_gem_sanitize(dev_priv);
 
 	intel_runtime_pm_put(dev_priv);
@@ -5118,6 +5192,8 @@ int i915_gem_init_hw(struct drm_i915_private *dev_priv)
 		}
 	}
 
+	intel_gt_workarounds_apply(dev_priv);
+
 	i915_gem_init_swizzling(dev_priv);
 
 	/*
@@ -5140,6 +5216,12 @@ int i915_gem_init_hw(struct drm_i915_private *dev_priv)
 		goto out;
 	}
 
+	ret = intel_wopcm_init_hw(&dev_priv->wopcm);
+	if (ret) {
+		DRM_ERROR("Enabling WOPCM failed (%d)\n", ret);
+		goto out;
+	}
+
 	/* We can't enable contexts until all firmware is loaded */
 	ret = intel_uc_init_hw(dev_priv);
 	if (ret) {
@@ -5297,6 +5379,10 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 	if (ret)
 		return ret;
 
+	ret = intel_wopcm_init(&dev_priv->wopcm);
+	if (ret)
+		return ret;
+
 	ret = intel_uc_init_misc(dev_priv);
 	if (ret)
 		return ret;
@@ -5478,8 +5564,7 @@ static void i915_gem_init__mm(struct drm_i915_private *i915)
 	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
 }
 
-int
-i915_gem_load_init(struct drm_i915_private *dev_priv)
+int i915_gem_init_early(struct drm_i915_private *dev_priv)
 {
 	int err = -ENOMEM;
 
@@ -5554,7 +5639,7 @@ err_out:
 	return err;
 }
 
-void i915_gem_load_cleanup(struct drm_i915_private *dev_priv)
+void i915_gem_cleanup_early(struct drm_i915_private *dev_priv)
 {
 	i915_gem_drain_freed_objects(dev_priv);
 	GEM_BUG_ON(!llist_empty(&dev_priv->mm.free_list));