drm/i915: Keep rings pinned while the context is active

Remember to keep the rings pinned as well as the context image until the
GPU is no longer active.

v2: Introduce a ring->pin_count primarily to hide the
mock_ring that doesn't fit into the normal GGTT vma picture.

v3: Order is important in teardown, ringbuffer submission needs to drop
the pin count on the engine->kernel_context before it can gleefully free
its ring.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110946
Fixes: ce476c80b8bf ("drm/i915: Keep contexts pinned until after the next kernel context switch")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190619170135.15281-1-chris@chris-wilson.co.uk

5 files changed, 53 insertions, 28 deletions

diff --git a/drivers/gpu/drm/i915/gt/intel_context.c b/drivers/gpu/drm/i915/gt/intel_context.c
index 2c454f227c2e..23120901c55f 100644
--- a/drivers/gpu/drm/i915/gt/intel_context.c
+++ b/drivers/gpu/drm/i915/gt/intel_context.c
@@ -126,6 +126,7 @@ static void intel_context_retire(struct i915_active *active)
 	if (ce->state)
 		__context_unpin_state(ce->state);
 
+	intel_ring_unpin(ce->ring);
 	intel_context_put(ce);
 }
 
@@ -160,27 +161,35 @@ int intel_context_active_acquire(struct intel_context *ce, unsigned long flags)
 
 	intel_context_get(ce);
 
+	err = intel_ring_pin(ce->ring);
+	if (err)
+		goto err_put;
+
 	if (!ce->state)
 		return 0;
 
 	err = __context_pin_state(ce->state, flags);
-	if (err) {
-		i915_active_cancel(&ce->active);
-		intel_context_put(ce);
-		return err;
-	}
+	if (err)
+		goto err_ring;
 
 	/* Preallocate tracking nodes */
 	if (!i915_gem_context_is_kernel(ce->gem_context)) {
 		err = i915_active_acquire_preallocate_barrier(&ce->active,
 							      ce->engine);
-		if (err) {
-			i915_active_release(&ce->active);
-			return err;
-		}
+		if (err)
+			goto err_state;
 	}
 
 	return 0;
+
+err_state:
+	__context_unpin_state(ce->state);
+err_ring:
+	intel_ring_unpin(ce->ring);
+err_put:
+	intel_context_put(ce);
+	i915_active_cancel(&ce->active);
+	return err;
 }
 
 void intel_context_active_release(struct intel_context *ce)
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 868b220214f8..43e975a26016 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -70,6 +70,18 @@ struct intel_ring {
 	struct list_head request_list;
 	struct list_head active_link;
 
+	/*
+	 * As we have two types of rings, one global to the engine used
+	 * by ringbuffer submission and those that are exclusive to a
+	 * context used by execlists, we have to play safe and allow
+	 * atomic updates to the pin_count. However, the actual pinning
+	 * of the context is either done during initialisation for
+	 * ringbuffer submission or serialised as part of the context
+	 * pinning for execlists, and so we do not need a mutex ourselves
+	 * to serialise intel_ring_pin/intel_ring_unpin.
+	 */
+	atomic_t pin_count;
+
 	u32 head;
 	u32 tail;
 	u32 emit;
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
index b42b5f158295..82b7ace62d97 100644
--- a/drivers/gpu/drm/i915/gt/intel_lrc.c
+++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
@@ -1414,6 +1414,7 @@ static void execlists_context_destroy(struct kref *kref)
 {
 	struct intel_context *ce = container_of(kref, typeof(*ce), ref);
 
+	GEM_BUG_ON(!i915_active_is_idle(&ce->active));
 	GEM_BUG_ON(intel_context_is_pinned(ce));
 
 	if (ce->state)
@@ -1426,7 +1427,6 @@ static void execlists_context_unpin(struct intel_context *ce)
 {
 	i915_gem_context_unpin_hw_id(ce->gem_context);
 	i915_gem_object_unpin_map(ce->state->obj);
-	intel_ring_unpin(ce->ring);
 }
 
 static void
@@ -1478,13 +1478,9 @@ __execlists_context_pin(struct intel_context *ce,
 		goto unpin_active;
 	}
 
-	ret = intel_ring_pin(ce->ring);
-	if (ret)
-		goto unpin_map;
-
 	ret = i915_gem_context_pin_hw_id(ce->gem_context);
 	if (ret)
-		goto unpin_ring;
+		goto unpin_map;
 
 	ce->lrc_desc = lrc_descriptor(ce, engine);
 	ce->lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE;
@@ -1492,8 +1488,6 @@ __execlists_context_pin(struct intel_context *ce,
 
 	return 0;
 
-unpin_ring:
-	intel_ring_unpin(ce->ring);
 unpin_map:
 	i915_gem_object_unpin_map(ce->state->obj);
 unpin_active:
diff --git a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
index c6023bc9452d..12010e798868 100644
--- a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
@@ -1149,16 +1149,16 @@ i915_emit_bb_start(struct i915_request *rq,
 int intel_ring_pin(struct intel_ring *ring)
 {
 	struct i915_vma *vma = ring->vma;
-	enum i915_map_type map = i915_coherent_map_type(vma->vm->i915);
 	unsigned int flags;
 	void *addr;
 	int ret;
 
-	GEM_BUG_ON(ring->vaddr);
+	if (atomic_fetch_inc(&ring->pin_count))
+		return 0;
 
 	ret = i915_timeline_pin(ring->timeline);
 	if (ret)
-		return ret;
+		goto err_unpin;
 
 	flags = PIN_GLOBAL;
 
@@ -1172,26 +1172,31 @@ int intel_ring_pin(struct intel_ring *ring)
 
 	ret = i915_vma_pin(vma, 0, 0, flags);
 	if (unlikely(ret))
-		goto unpin_timeline;
+		goto err_timeline;
 
 	if (i915_vma_is_map_and_fenceable(vma))
 		addr = (void __force *)i915_vma_pin_iomap(vma);
 	else
-		addr = i915_gem_object_pin_map(vma->obj, map);
+		addr = i915_gem_object_pin_map(vma->obj,
+					       i915_coherent_map_type(vma->vm->i915));
 	if (IS_ERR(addr)) {
 		ret = PTR_ERR(addr);
-		goto unpin_ring;
+		goto err_ring;
 	}
 
 	vma->obj->pin_global++;
 
+	GEM_BUG_ON(ring->vaddr);
 	ring->vaddr = addr;
+
 	return 0;
 
-unpin_ring:
+err_ring:
 	i915_vma_unpin(vma);
-unpin_timeline:
+err_timeline:
 	i915_timeline_unpin(ring->timeline);
+err_unpin:
+	atomic_dec(&ring->pin_count);
 	return ret;
 }
 
@@ -1207,16 +1212,19 @@ void intel_ring_reset(struct intel_ring *ring, u32 tail)
 
 void intel_ring_unpin(struct intel_ring *ring)
 {
-	GEM_BUG_ON(!ring->vma);
-	GEM_BUG_ON(!ring->vaddr);
+	if (!atomic_dec_and_test(&ring->pin_count))
+		return;
 
 	/* Discard any unused bytes beyond that submitted to hw. */
 	intel_ring_reset(ring, ring->tail);
 
+	GEM_BUG_ON(!ring->vma);
 	if (i915_vma_is_map_and_fenceable(ring->vma))
 		i915_vma_unpin_iomap(ring->vma);
 	else
 		i915_gem_object_unpin_map(ring->vma->obj);
+
+	GEM_BUG_ON(!ring->vaddr);
 	ring->vaddr = NULL;
 
 	ring->vma->obj->pin_global--;
@@ -2081,10 +2089,11 @@ static void ring_destroy(struct intel_engine_cs *engine)
 	WARN_ON(INTEL_GEN(dev_priv) > 2 &&
 		(ENGINE_READ(engine, RING_MI_MODE) & MODE_IDLE) == 0);
 
+	intel_engine_cleanup_common(engine);
+
 	intel_ring_unpin(engine->buffer);
 	intel_ring_put(engine->buffer);
 
-	intel_engine_cleanup_common(engine);
 	kfree(engine);
 }
 
diff --git a/drivers/gpu/drm/i915/gt/mock_engine.c b/drivers/gpu/drm/i915/gt/mock_engine.c
index 086801b51441..486c6953dcb1 100644
--- a/drivers/gpu/drm/i915/gt/mock_engine.c
+++ b/drivers/gpu/drm/i915/gt/mock_engine.c
@@ -66,6 +66,7 @@ static struct intel_ring *mock_ring(struct intel_engine_cs *engine)
 	ring->base.effective_size = sz;
 	ring->base.vaddr = (void *)(ring + 1);
 	ring->base.timeline = &ring->timeline;
+	atomic_set(&ring->base.pin_count, 1);
 
 	INIT_LIST_HEAD(&ring->base.request_list);
 	intel_ring_update_space(&ring->base);