[PATCH 1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[PATCH 1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[Chris Wilson]

Currently all callers are responsible for adding the vma to the active
timeline and then exporting its fence. Combine the two operations into
i915_vma_move_to_active() to move all the extra handling from the
callers to the single site.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_execbuffer.c    | 47 +++++++++----------
 drivers/gpu/drm/i915/selftests/huge_pages.c   |  4 --
 .../drm/i915/selftests/i915_gem_coherency.c   |  4 --
 .../gpu/drm/i915/selftests/i915_gem_context.c |  4 --
 .../gpu/drm/i915/selftests/i915_gem_object.c  |  4 --
 .../drm/i915/selftests/intel_workarounds.c    |  3 --
 6 files changed, 21 insertions(+), 45 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index c2dd9b4cdace..91f20445147f 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1166,15 +1166,9 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 
 	GEM_BUG_ON(!reservation_object_test_signaled_rcu(batch->resv, true));
 	i915_vma_move_to_active(batch, rq, 0);
-	reservation_object_lock(batch->resv, NULL);
-	reservation_object_add_excl_fence(batch->resv, &rq->fence);
-	reservation_object_unlock(batch->resv);
 	i915_vma_unpin(batch);
 
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
-	reservation_object_lock(vma->resv, NULL);
-	reservation_object_add_excl_fence(vma->resv, &rq->fence);
-	reservation_object_unlock(vma->resv);
 
 	rq->batch = batch;
 
@@ -1771,25 +1765,6 @@ static int eb_relocate(struct i915_execbuffer *eb)
 	return eb_relocate_slow(eb);
 }
 
-static void eb_export_fence(struct i915_vma *vma,
-			    struct i915_request *rq,
-			    unsigned int flags)
-{
-	struct reservation_object *resv = vma->resv;
-
-	/*
-	 * Ignore errors from failing to allocate the new fence, we can't
-	 * handle an error right now. Worst case should be missed
-	 * synchronisation leading to rendering corruption.
-	 */
-	reservation_object_lock(resv, NULL);
-	if (flags & EXEC_OBJECT_WRITE)
-		reservation_object_add_excl_fence(resv, &rq->fence);
-	else if (reservation_object_reserve_shared(resv) == 0)
-		reservation_object_add_shared_fence(resv, &rq->fence);
-	reservation_object_unlock(resv);
-}
-
 static int eb_move_to_gpu(struct i915_execbuffer *eb)
 {
 	const unsigned int count = eb->buffer_count;
@@ -1844,7 +1819,6 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		struct i915_vma *vma = eb->vma[i];
 
 		i915_vma_move_to_active(vma, eb->request, flags);
-		eb_export_fence(vma, eb->request, flags);
 
 		__eb_unreserve_vma(vma, flags);
 		vma->exec_flags = NULL;
@@ -1884,6 +1858,25 @@ static bool i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec)
 	return true;
 }
 
+static void export_fence(struct i915_vma *vma,
+			 struct i915_request *rq,
+			 unsigned int flags)
+{
+	struct reservation_object *resv = vma->resv;
+
+	/*
+	 * Ignore errors from failing to allocate the new fence, we can't
+	 * handle an error right now. Worst case should be missed
+	 * synchronisation leading to rendering corruption.
+	 */
+	reservation_object_lock(resv, NULL);
+	if (flags & EXEC_OBJECT_WRITE)
+		reservation_object_add_excl_fence(resv, &rq->fence);
+	else if (reservation_object_reserve_shared(resv) == 0)
+		reservation_object_add_shared_fence(resv, &rq->fence);
+	reservation_object_unlock(resv);
+}
+
 void i915_vma_move_to_active(struct i915_vma *vma,
 			     struct i915_request *rq,
 			     unsigned int flags)
@@ -1921,6 +1914,8 @@ void i915_vma_move_to_active(struct i915_vma *vma,
 
 	if (flags & EXEC_OBJECT_NEEDS_FENCE)
 		i915_gem_active_set(&vma->last_fence, rq);
+
+	export_fence(vma, rq, flags);
 }
 
 static int i915_reset_gen7_sol_offsets(struct i915_request *rq)
diff --git a/drivers/gpu/drm/i915/selftests/huge_pages.c b/drivers/gpu/drm/i915/selftests/huge_pages.c
index 1193dd36913a..97b44bf72ea5 100644
--- a/drivers/gpu/drm/i915/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/selftests/huge_pages.c
@@ -998,10 +998,6 @@ static int gpu_write(struct i915_vma *vma,
 
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 
-	reservation_object_lock(vma->resv, NULL);
-	reservation_object_add_excl_fence(vma->resv, &rq->fence);
-	reservation_object_unlock(vma->resv);
-
 err_request:
 	i915_request_add(rq);
 
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
index cb9eef1635e1..caec9787e1d9 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
@@ -225,10 +225,6 @@ static int gpu_set(struct drm_i915_gem_object *obj,
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 	i915_vma_unpin(vma);
 
-	reservation_object_lock(obj->resv, NULL);
-	reservation_object_add_excl_fence(obj->resv, &rq->fence);
-	reservation_object_unlock(obj->resv);
-
 	i915_request_add(rq);
 
 	return 0;
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_context.c b/drivers/gpu/drm/i915/selftests/i915_gem_context.c
index 0b36265a0f96..443df707273d 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_context.c
@@ -178,10 +178,6 @@ static int gpu_fill(struct drm_i915_gem_object *obj,
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 	i915_vma_unpin(vma);
 
-	reservation_object_lock(obj->resv, NULL);
-	reservation_object_add_excl_fence(obj->resv, &rq->fence);
-	reservation_object_unlock(obj->resv);
-
 	i915_request_add(rq);
 
 	return 0;
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_object.c b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
index 6fe71865b710..b2ccbc5e2bbe 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
@@ -466,10 +466,6 @@ static int make_obj_busy(struct drm_i915_gem_object *obj)
 
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 
-	reservation_object_lock(vma->resv, NULL);
-	reservation_object_add_excl_fence(vma->resv, &rq->fence);
-	reservation_object_unlock(vma->resv);
-
 	i915_request_add(rq);
 
 	i915_gem_object_set_active_reference(obj);
diff --git a/drivers/gpu/drm/i915/selftests/intel_workarounds.c b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
index 3d86b90ab722..4a9dc01a364a 100644
--- a/drivers/gpu/drm/i915/selftests/intel_workarounds.c
+++ b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
@@ -68,9 +68,6 @@ read_nonprivs(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
 	intel_ring_advance(rq, cs);
 
 	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
-	reservation_object_lock(vma->resv, NULL);
-	reservation_object_add_excl_fence(vma->resv, &rq->fence);
-	reservation_object_unlock(vma->resv);
 
 	i915_gem_object_get(result);
 	i915_gem_object_set_active_reference(result);
-- 
2.18.0

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

[PATCH 2/6] drm/i915: Export i915_request_skip()

[Chris Wilson]

In the next patch, we will want to start skipping requests on failing to
complete their payloads. So export the utility function current used to
make requests inoperable following a failed gpu reset.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c     | 25 +++----------------------
 drivers/gpu/drm/i915/i915_request.c | 21 +++++++++++++++++++++
 drivers/gpu/drm/i915/i915_request.h |  2 ++
 3 files changed, 26 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 0c0a1a959d0b..f516c289647c 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3081,25 +3081,6 @@ int i915_gem_reset_prepare(struct drm_i915_private *dev_priv)
 	return err;
 }
 
-static void skip_request(struct i915_request *request)
-{
-	void *vaddr = request->ring->vaddr;
-	u32 head;
-
-	/* As this request likely depends on state from the lost
-	 * context, clear out all the user operations leaving the
-	 * breadcrumb at the end (so we get the fence notifications).
-	 */
-	head = request->head;
-	if (request->postfix < head) {
-		memset(vaddr + head, 0, request->ring->size - head);
-		head = 0;
-	}
-	memset(vaddr + head, 0, request->postfix - head);
-
-	dma_fence_set_error(&request->fence, -EIO);
-}
-
 static void engine_skip_context(struct i915_request *request)
 {
 	struct intel_engine_cs *engine = request->engine;
@@ -3114,10 +3095,10 @@ static void engine_skip_context(struct i915_request *request)
 
 	list_for_each_entry_continue(request, &engine->timeline.requests, link)
 		if (request->gem_context == hung_ctx)
-			skip_request(request);
+			i915_request_skip(request, -EIO);
 
 	list_for_each_entry(request, &timeline->requests, link)
-		skip_request(request);
+		i915_request_skip(request, -EIO);
 
 	spin_unlock(&timeline->lock);
 	spin_unlock_irqrestore(&engine->timeline.lock, flags);
@@ -3160,7 +3141,7 @@ i915_gem_reset_request(struct intel_engine_cs *engine,
 
 	if (stalled) {
 		i915_gem_context_mark_guilty(request->gem_context);
-		skip_request(request);
+		i915_request_skip(request, -EIO);
 
 		/* If this context is now banned, skip all pending requests. */
 		if (i915_gem_context_is_banned(request->gem_context))
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index a2f7e9358450..7ae08b68121e 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1013,6 +1013,27 @@ i915_request_await_object(struct i915_request *to,
 	return ret;
 }
 
+void i915_request_skip(struct i915_request *rq, int error)
+{
+	void *vaddr = rq->ring->vaddr;
+	u32 head;
+
+	GEM_BUG_ON(!IS_ERR_VALUE((long)error));
+	dma_fence_set_error(&rq->fence, error);
+
+	/*
+	 * As this request likely depends on state from the lost
+	 * context, clear out all the user operations leaving the
+	 * breadcrumb at the end (so we get the fence notifications).
+	 */
+	head = rq->infix;
+	if (rq->postfix < head) {
+		memset(vaddr + head, 0, rq->ring->size - head);
+		head = 0;
+	}
+	memset(vaddr + head, 0, rq->postfix - head);
+}
+
 /*
  * NB: This function is not allowed to fail. Doing so would mean the the
  * request is not being tracked for completion but the work itself is
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 7ee220ded9c9..a355a081485f 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -258,6 +258,8 @@ void i915_request_add(struct i915_request *rq);
 void __i915_request_submit(struct i915_request *request);
 void i915_request_submit(struct i915_request *request);
 
+void i915_request_skip(struct i915_request *request, int error);
+
 void __i915_request_unsubmit(struct i915_request *request);
 void i915_request_unsubmit(struct i915_request *request);
 
-- 
2.18.0

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

[PATCH 3/6] drm/i915: Start returning an error from i915_vma_move_to_active()

[Chris Wilson]

Handling such a late error in request construction is tricky, but to
accommodate future patches which may allocate here, we potentially could
err. To handle the error after already adjusting global state to track
the new request, we must finish and submit the request. But we don't
want to use the request as not everything is being tracked by it, so we
opt to cancel the commands inside the request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/gvt/scheduler.c          |  6 ++++-
 drivers/gpu/drm/i915/i915_drv.h               |  6 ++---
 drivers/gpu/drm/i915/i915_gem_execbuffer.c    | 25 +++++++++++++------
 drivers/gpu/drm/i915/i915_gem_render_state.c  |  2 +-
 drivers/gpu/drm/i915/selftests/huge_pages.c   |  9 +++++--
 .../drm/i915/selftests/i915_gem_coherency.c   |  4 +--
 .../gpu/drm/i915/selftests/i915_gem_context.c | 12 +++++++--
 .../gpu/drm/i915/selftests/i915_gem_object.c  |  7 +++---
 drivers/gpu/drm/i915/selftests/i915_request.c |  8 ++++--
 .../gpu/drm/i915/selftests/intel_hangcheck.c  | 11 ++++++--
 drivers/gpu/drm/i915/selftests/intel_lrc.c    | 11 ++++++--
 .../drm/i915/selftests/intel_workarounds.c    |  6 +++--
 12 files changed, 78 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 928818f218f7..b0e566956b8d 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -476,7 +476,11 @@ static int prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 			i915_gem_obj_finish_shmem_access(bb->obj);
 			bb->accessing = false;
 
-			i915_vma_move_to_active(bb->vma, workload->req, 0);
+			ret = i915_vma_move_to_active(bb->vma,
+						      workload->req,
+						      0);
+			if (ret)
+				goto err;
 		}
 	}
 	return 0;
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 550e86dfbfe8..e5464f5a2a78 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3095,9 +3095,9 @@ i915_gem_obj_finish_shmem_access(struct drm_i915_gem_object *obj)
 }
 
 int __must_check i915_mutex_lock_interruptible(struct drm_device *dev);
-void i915_vma_move_to_active(struct i915_vma *vma,
-			     struct i915_request *rq,
-			     unsigned int flags);
+int __must_check i915_vma_move_to_active(struct i915_vma *vma,
+					 struct i915_request *rq,
+					 unsigned int flags);
 int i915_gem_dumb_create(struct drm_file *file_priv,
 			 struct drm_device *dev,
 			 struct drm_mode_create_dumb *args);
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 91f20445147f..97136e4ce91d 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1165,12 +1165,16 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 		goto err_request;
 
 	GEM_BUG_ON(!reservation_object_test_signaled_rcu(batch->resv, true));
-	i915_vma_move_to_active(batch, rq, 0);
-	i915_vma_unpin(batch);
+	err = i915_vma_move_to_active(batch, rq, 0);
+	if (err)
+		goto skip_request;
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		goto skip_request;
 
 	rq->batch = batch;
+	i915_vma_unpin(batch);
 
 	cache->rq = rq;
 	cache->rq_cmd = cmd;
@@ -1179,6 +1183,8 @@ static int __reloc_gpu_alloc(struct i915_execbuffer *eb,
 	/* Return with batch mapping (cmd) still pinned */
 	return 0;
 
+skip_request:
+	i915_request_skip(rq, err);
 err_request:
 	i915_request_add(rq);
 err_unpin:
@@ -1818,7 +1824,11 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		unsigned int flags = eb->flags[i];
 		struct i915_vma *vma = eb->vma[i];
 
-		i915_vma_move_to_active(vma, eb->request, flags);
+		err = i915_vma_move_to_active(vma, eb->request, flags);
+		if (unlikely(err)) {
+			i915_request_skip(eb->request, err);
+			return err;
+		}
 
 		__eb_unreserve_vma(vma, flags);
 		vma->exec_flags = NULL;
@@ -1877,9 +1887,9 @@ static void export_fence(struct i915_vma *vma,
 	reservation_object_unlock(resv);
 }
 
-void i915_vma_move_to_active(struct i915_vma *vma,
-			     struct i915_request *rq,
-			     unsigned int flags)
+int i915_vma_move_to_active(struct i915_vma *vma,
+			    struct i915_request *rq,
+			    unsigned int flags)
 {
 	struct drm_i915_gem_object *obj = vma->obj;
 	const unsigned int idx = rq->engine->id;
@@ -1916,6 +1926,7 @@ void i915_vma_move_to_active(struct i915_vma *vma,
 		i915_gem_active_set(&vma->last_fence, rq);
 
 	export_fence(vma, rq, flags);
+	return 0;
 }
 
 static int i915_reset_gen7_sol_offsets(struct i915_request *rq)
diff --git a/drivers/gpu/drm/i915/i915_gem_render_state.c b/drivers/gpu/drm/i915/i915_gem_render_state.c
index 3210cedfa46c..90baf9086d0a 100644
--- a/drivers/gpu/drm/i915/i915_gem_render_state.c
+++ b/drivers/gpu/drm/i915/i915_gem_render_state.c
@@ -222,7 +222,7 @@ int i915_gem_render_state_emit(struct i915_request *rq)
 			goto err_unpin;
 	}
 
-	i915_vma_move_to_active(so.vma, rq, 0);
+	err = i915_vma_move_to_active(so.vma, rq, 0);
 err_unpin:
 	i915_vma_unpin(so.vma);
 err_vma:
diff --git a/drivers/gpu/drm/i915/selftests/huge_pages.c b/drivers/gpu/drm/i915/selftests/huge_pages.c
index 97b44bf72ea5..c399e7d69365 100644
--- a/drivers/gpu/drm/i915/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/selftests/huge_pages.c
@@ -985,7 +985,10 @@ static int gpu_write(struct i915_vma *vma,
 		goto err_request;
 	}
 
-	i915_vma_move_to_active(batch, rq, 0);
+	err = i915_vma_move_to_active(batch, rq, 0);
+	if (err)
+		goto err_request;
+
 	i915_gem_object_set_active_reference(batch->obj);
 	i915_vma_unpin(batch);
 	i915_vma_close(batch);
@@ -996,7 +999,9 @@ static int gpu_write(struct i915_vma *vma,
 	if (err)
 		goto err_request;
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		i915_request_skip(rq, err);
 
 err_request:
 	i915_request_add(rq);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
index caec9787e1d9..d03b226579bb 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
@@ -222,12 +222,12 @@ static int gpu_set(struct drm_i915_gem_object *obj,
 	}
 	intel_ring_advance(rq, cs);
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 	i915_vma_unpin(vma);
 
 	i915_request_add(rq);
 
-	return 0;
+	return err;
 }
 
 static bool always_valid(struct drm_i915_private *i915)
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_context.c b/drivers/gpu/drm/i915/selftests/i915_gem_context.c
index 443df707273d..eeb91009557c 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_context.c
@@ -170,18 +170,26 @@ static int gpu_fill(struct drm_i915_gem_object *obj,
 	if (err)
 		goto err_request;
 
-	i915_vma_move_to_active(batch, rq, 0);
+	err = i915_vma_move_to_active(batch, rq, 0);
+	if (err)
+		goto skip_request;
+
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		goto skip_request;
+
 	i915_gem_object_set_active_reference(batch->obj);
 	i915_vma_unpin(batch);
 	i915_vma_close(batch);
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 	i915_vma_unpin(vma);
 
 	i915_request_add(rq);
 
 	return 0;
 
+skip_request:
+	i915_request_skip(rq, err);
 err_request:
 	i915_request_add(rq);
 err_batch:
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_object.c b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
index b2ccbc5e2bbe..25c2b2d433bd 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
@@ -464,13 +464,14 @@ static int make_obj_busy(struct drm_i915_gem_object *obj)
 		return PTR_ERR(rq);
 	}
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
 
 	i915_request_add(rq);
 
-	i915_gem_object_set_active_reference(obj);
+	__i915_gem_object_release_unless_active(obj);
 	i915_vma_unpin(vma);
-	return 0;
+
+	return err;
 }
 
 static bool assert_mmap_offset(struct drm_i915_private *i915,
diff --git a/drivers/gpu/drm/i915/selftests/i915_request.c b/drivers/gpu/drm/i915/selftests/i915_request.c
index cc27edc40356..43995fc3534d 100644
--- a/drivers/gpu/drm/i915/selftests/i915_request.c
+++ b/drivers/gpu/drm/i915/selftests/i915_request.c
@@ -675,7 +675,9 @@ static int live_all_engines(void *arg)
 			i915_gem_object_set_active_reference(batch->obj);
 		}
 
-		i915_vma_move_to_active(batch, request[id], 0);
+		err = i915_vma_move_to_active(batch, request[id], 0);
+		GEM_BUG_ON(err);
+
 		i915_request_get(request[id]);
 		i915_request_add(request[id]);
 	}
@@ -785,7 +787,9 @@ static int live_sequential_engines(void *arg)
 		GEM_BUG_ON(err);
 		request[id]->batch = batch;
 
-		i915_vma_move_to_active(batch, request[id], 0);
+		err = i915_vma_move_to_active(batch, request[id], 0);
+		GEM_BUG_ON(err);
+
 		i915_gem_object_set_active_reference(batch->obj);
 		i915_vma_get(batch);
 
diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 5cb808dc5b50..f17c22000e6b 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -130,13 +130,19 @@ static int emit_recurse_batch(struct hang *h,
 	if (err)
 		goto unpin_vma;
 
-	i915_vma_move_to_active(vma, rq, 0);
+	err = i915_vma_move_to_active(vma, rq, 0);
+	if (err)
+		goto unpin_hws;
+
 	if (!i915_gem_object_has_active_reference(vma->obj)) {
 		i915_gem_object_get(vma->obj);
 		i915_gem_object_set_active_reference(vma->obj);
 	}
 
-	i915_vma_move_to_active(hws, rq, 0);
+	err = i915_vma_move_to_active(hws, rq, 0);
+	if (err)
+		goto unpin_hws;
+
 	if (!i915_gem_object_has_active_reference(hws->obj)) {
 		i915_gem_object_get(hws->obj);
 		i915_gem_object_set_active_reference(hws->obj);
@@ -205,6 +211,7 @@ static int emit_recurse_batch(struct hang *h,
 
 	err = rq->engine->emit_bb_start(rq, vma->node.start, PAGE_SIZE, flags);
 
+unpin_hws:
 	i915_vma_unpin(hws);
 unpin_vma:
 	i915_vma_unpin(vma);
diff --git a/drivers/gpu/drm/i915/selftests/intel_lrc.c b/drivers/gpu/drm/i915/selftests/intel_lrc.c
index ea27c7cfbf96..b34f7ac6631e 100644
--- a/drivers/gpu/drm/i915/selftests/intel_lrc.c
+++ b/drivers/gpu/drm/i915/selftests/intel_lrc.c
@@ -104,13 +104,19 @@ static int emit_recurse_batch(struct spinner *spin,
 	if (err)
 		goto unpin_vma;
 
-	i915_vma_move_to_active(vma, rq, 0);
+	err = i915_vma_move_to_active(vma, rq, 0);
+	if (err)
+		goto unpin_hws;
+
 	if (!i915_gem_object_has_active_reference(vma->obj)) {
 		i915_gem_object_get(vma->obj);
 		i915_gem_object_set_active_reference(vma->obj);
 	}
 
-	i915_vma_move_to_active(hws, rq, 0);
+	err = i915_vma_move_to_active(hws, rq, 0);
+	if (err)
+		goto unpin_hws;
+
 	if (!i915_gem_object_has_active_reference(hws->obj)) {
 		i915_gem_object_get(hws->obj);
 		i915_gem_object_set_active_reference(hws->obj);
@@ -134,6 +140,7 @@ static int emit_recurse_batch(struct spinner *spin,
 
 	err = rq->engine->emit_bb_start(rq, vma->node.start, PAGE_SIZE, 0);
 
+unpin_hws:
 	i915_vma_unpin(hws);
 unpin_vma:
 	i915_vma_unpin(vma);
diff --git a/drivers/gpu/drm/i915/selftests/intel_workarounds.c b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
index 4a9dc01a364a..fafdec3fe83e 100644
--- a/drivers/gpu/drm/i915/selftests/intel_workarounds.c
+++ b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
@@ -49,6 +49,10 @@ read_nonprivs(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
 		goto err_pin;
 	}
 
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		goto err_req;
+
 	srm = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
 	if (INTEL_GEN(ctx->i915) >= 8)
 		srm++;
@@ -67,8 +71,6 @@ read_nonprivs(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
 	}
 	intel_ring_advance(rq, cs);
 
-	i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
-
 	i915_gem_object_get(result);
 	i915_gem_object_set_active_reference(result);
 
-- 
2.18.0

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

[PATCH 4/6] drm/i915: Move i915_vma_move_to_active() to i915_vma.c

[Chris Wilson]

i915_vma_move_to_active() has grown beyond its execbuf origins, and
should take its rightful place in i915_vma.c as a method for i915_vma!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h            |  3 --
 drivers/gpu/drm/i915/i915_gem_execbuffer.c | 61 ----------------------
 drivers/gpu/drm/i915/i915_vma.c            | 61 ++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_vma.h            |  4 ++
 4 files changed, 65 insertions(+), 64 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index e5464f5a2a78..70ae2b37571a 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3095,9 +3095,6 @@ i915_gem_obj_finish_shmem_access(struct drm_i915_gem_object *obj)
 }
 
 int __must_check i915_mutex_lock_interruptible(struct drm_device *dev);
-int __must_check i915_vma_move_to_active(struct i915_vma *vma,
-					 struct i915_request *rq,
-					 unsigned int flags);
 int i915_gem_dumb_create(struct drm_file *file_priv,
 			 struct drm_device *dev,
 			 struct drm_mode_create_dumb *args);
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 97136e4ce91d..3f0c612d42e7 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1868,67 +1868,6 @@ static bool i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec)
 	return true;
 }
 
-static void export_fence(struct i915_vma *vma,
-			 struct i915_request *rq,
-			 unsigned int flags)
-{
-	struct reservation_object *resv = vma->resv;
-
-	/*
-	 * Ignore errors from failing to allocate the new fence, we can't
-	 * handle an error right now. Worst case should be missed
-	 * synchronisation leading to rendering corruption.
-	 */
-	reservation_object_lock(resv, NULL);
-	if (flags & EXEC_OBJECT_WRITE)
-		reservation_object_add_excl_fence(resv, &rq->fence);
-	else if (reservation_object_reserve_shared(resv) == 0)
-		reservation_object_add_shared_fence(resv, &rq->fence);
-	reservation_object_unlock(resv);
-}
-
-int i915_vma_move_to_active(struct i915_vma *vma,
-			    struct i915_request *rq,
-			    unsigned int flags)
-{
-	struct drm_i915_gem_object *obj = vma->obj;
-	const unsigned int idx = rq->engine->id;
-
-	lockdep_assert_held(&rq->i915->drm.struct_mutex);
-	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
-
-	/*
-	 * Add a reference if we're newly entering the active list.
-	 * The order in which we add operations to the retirement queue is
-	 * vital here: mark_active adds to the start of the callback list,
-	 * such that subsequent callbacks are called first. Therefore we
-	 * add the active reference first and queue for it to be dropped
-	 * *last*.
-	 */
-	if (!i915_vma_is_active(vma))
-		obj->active_count++;
-	i915_vma_set_active(vma, idx);
-	i915_gem_active_set(&vma->last_read[idx], rq);
-	list_move_tail(&vma->vm_link, &vma->vm->active_list);
-
-	obj->write_domain = 0;
-	if (flags & EXEC_OBJECT_WRITE) {
-		obj->write_domain = I915_GEM_DOMAIN_RENDER;
-
-		if (intel_fb_obj_invalidate(obj, ORIGIN_CS))
-			i915_gem_active_set(&obj->frontbuffer_write, rq);
-
-		obj->read_domains = 0;
-	}
-	obj->read_domains |= I915_GEM_GPU_DOMAINS;
-
-	if (flags & EXEC_OBJECT_NEEDS_FENCE)
-		i915_gem_active_set(&vma->last_fence, rq);
-
-	export_fence(vma, rq, flags);
-	return 0;
-}
-
 static int i915_reset_gen7_sol_offsets(struct i915_request *rq)
 {
 	u32 *cs;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 518de47111ff..6f3a0f2296c2 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -859,6 +859,67 @@ void i915_vma_revoke_mmap(struct i915_vma *vma)
 		list_del(&vma->obj->userfault_link);
 }
 
+static void export_fence(struct i915_vma *vma,
+			 struct i915_request *rq,
+			 unsigned int flags)
+{
+	struct reservation_object *resv = vma->resv;
+
+	/*
+	 * Ignore errors from failing to allocate the new fence, we can't
+	 * handle an error right now. Worst case should be missed
+	 * synchronisation leading to rendering corruption.
+	 */
+	reservation_object_lock(resv, NULL);
+	if (flags & EXEC_OBJECT_WRITE)
+		reservation_object_add_excl_fence(resv, &rq->fence);
+	else if (reservation_object_reserve_shared(resv) == 0)
+		reservation_object_add_shared_fence(resv, &rq->fence);
+	reservation_object_unlock(resv);
+}
+
+int i915_vma_move_to_active(struct i915_vma *vma,
+			    struct i915_request *rq,
+			    unsigned int flags)
+{
+	struct drm_i915_gem_object *obj = vma->obj;
+	const unsigned int idx = rq->engine->id;
+
+	lockdep_assert_held(&rq->i915->drm.struct_mutex);
+	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
+
+	/*
+	 * Add a reference if we're newly entering the active list.
+	 * The order in which we add operations to the retirement queue is
+	 * vital here: mark_active adds to the start of the callback list,
+	 * such that subsequent callbacks are called first. Therefore we
+	 * add the active reference first and queue for it to be dropped
+	 * *last*.
+	 */
+	if (!i915_vma_is_active(vma))
+		obj->active_count++;
+	i915_vma_set_active(vma, idx);
+	i915_gem_active_set(&vma->last_read[idx], rq);
+	list_move_tail(&vma->vm_link, &vma->vm->active_list);
+
+	obj->write_domain = 0;
+	if (flags & EXEC_OBJECT_WRITE) {
+		obj->write_domain = I915_GEM_DOMAIN_RENDER;
+
+		if (intel_fb_obj_invalidate(obj, ORIGIN_CS))
+			i915_gem_active_set(&obj->frontbuffer_write, rq);
+
+		obj->read_domains = 0;
+	}
+	obj->read_domains |= I915_GEM_GPU_DOMAINS;
+
+	if (flags & EXEC_OBJECT_NEEDS_FENCE)
+		i915_gem_active_set(&vma->last_fence, rq);
+
+	export_fence(vma, rq, flags);
+	return 0;
+}
+
 int i915_vma_unbind(struct i915_vma *vma)
 {
 	unsigned long active;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 66a228931517..a218b689e418 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -215,6 +215,10 @@ static inline bool i915_vma_has_active_engine(const struct i915_vma *vma,
 	return vma->active & BIT(engine);
 }
 
+int __must_check i915_vma_move_to_active(struct i915_vma *vma,
+					 struct i915_request *rq,
+					 unsigned int flags);
+
 static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
 {
 	GEM_BUG_ON(!i915_vma_is_ggtt(vma));
-- 
2.18.0

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

[PATCH 5/6] drm/i915: Track vma activity per fence.context, not per engine

[Chris Wilson]

In the next patch, we will want to be able to use more flexible request
timelines that can hop between engines. From the vma pov, we can then
not rely on the binding of this request to an engine and so can not
ensure that different requests are ordered through a per-engine
timeline, and so we must track activity of all timelines. (We track
activity on the vma itself to prevent unbinding from HW before the HW
has finished accessing it.)

v2: Switch to a rbtree for 32b safety (since using u64 as a radixtree
index is fraught with aliasing of unsigned longs).
v3: s/lookup_active/active_instance/ because we can never agree on names

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c   |   5 +-
 drivers/gpu/drm/i915/i915_gpu_error.c |  14 +---
 drivers/gpu/drm/i915/i915_gpu_error.h |   2 +-
 drivers/gpu/drm/i915/i915_request.h   |   1 +
 drivers/gpu/drm/i915/i915_vma.c       | 112 +++++++++++++++++++-------
 drivers/gpu/drm/i915/i915_vma.h       |  46 +++--------
 6 files changed, 100 insertions(+), 80 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 2f10f5580230..801d17ab6f72 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2068,7 +2068,6 @@ static struct i915_vma *pd_vma_create(struct gen6_hw_ppgtt *ppgtt, int size)
 	struct drm_i915_private *i915 = ppgtt->base.vm.i915;
 	struct i915_ggtt *ggtt = &i915->ggtt;
 	struct i915_vma *vma;
-	int i;
 
 	GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE));
 	GEM_BUG_ON(size > ggtt->vm.total);
@@ -2077,14 +2076,14 @@ static struct i915_vma *pd_vma_create(struct gen6_hw_ppgtt *ppgtt, int size)
 	if (!vma)
 		return ERR_PTR(-ENOMEM);
 
-	for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
-		init_request_active(&vma->last_read[i], NULL);
 	init_request_active(&vma->last_fence, NULL);
 
 	vma->vm = &ggtt->vm;
 	vma->ops = &pd_vma_ops;
 	vma->private = ppgtt;
 
+	vma->active = RB_ROOT;
+
 	vma->size = size;
 	vma->fence_size = size;
 	vma->flags = I915_VMA_GGTT;
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index df524c9cad40..8c81cf3aa182 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -335,21 +335,16 @@ static void print_error_buffers(struct drm_i915_error_state_buf *m,
 				struct drm_i915_error_buffer *err,
 				int count)
 {
-	int i;
-
 	err_printf(m, "%s [%d]:\n", name, count);
 
 	while (count--) {
-		err_printf(m, "    %08x_%08x %8u %02x %02x [ ",
+		err_printf(m, "    %08x_%08x %8u %02x %02x %02x",
 			   upper_32_bits(err->gtt_offset),
 			   lower_32_bits(err->gtt_offset),
 			   err->size,
 			   err->read_domains,
-			   err->write_domain);
-		for (i = 0; i < I915_NUM_ENGINES; i++)
-			err_printf(m, "%02x ", err->rseqno[i]);
-
-		err_printf(m, "] %02x", err->wseqno);
+			   err->write_domain,
+			   err->wseqno);
 		err_puts(m, tiling_flag(err->tiling));
 		err_puts(m, dirty_flag(err->dirty));
 		err_puts(m, purgeable_flag(err->purgeable));
@@ -1021,13 +1016,10 @@ static void capture_bo(struct drm_i915_error_buffer *err,
 		       struct i915_vma *vma)
 {
 	struct drm_i915_gem_object *obj = vma->obj;
-	int i;
 
 	err->size = obj->base.size;
 	err->name = obj->base.name;
 
-	for (i = 0; i < I915_NUM_ENGINES; i++)
-		err->rseqno[i] = __active_get_seqno(&vma->last_read[i]);
 	err->wseqno = __active_get_seqno(&obj->frontbuffer_write);
 	err->engine = __active_get_engine_id(&obj->frontbuffer_write);
 
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
index 58910f1dc67c..f893a4e8b783 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.h
+++ b/drivers/gpu/drm/i915/i915_gpu_error.h
@@ -177,7 +177,7 @@ struct i915_gpu_state {
 	struct drm_i915_error_buffer {
 		u32 size;
 		u32 name;
-		u32 rseqno[I915_NUM_ENGINES], wseqno;
+		u32 wseqno;
 		u64 gtt_offset;
 		u32 read_domains;
 		u32 write_domain;
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index a355a081485f..e1c9365dfefb 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -380,6 +380,7 @@ static inline void
 init_request_active(struct i915_gem_active *active,
 		    i915_gem_retire_fn retire)
 {
+	RCU_INIT_POINTER(active->request, NULL);
 	INIT_LIST_HEAD(&active->link);
 	active->retire = retire ?: i915_gem_retire_noop;
 }
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 6f3a0f2296c2..b4cc98330225 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -63,18 +63,20 @@ static void vma_print_allocator(struct i915_vma *vma, const char *reason)
 
 #endif
 
+struct i915_vma_active {
+	struct i915_gem_active base;
+	struct i915_vma *vma;
+	struct rb_node node;
+	u64 timeline;
+};
+
 static void
-i915_vma_retire(struct i915_gem_active *active, struct i915_request *rq)
+__i915_vma_retire(struct i915_vma *vma, struct i915_request *rq)
 {
-	const unsigned int idx = rq->engine->id;
-	struct i915_vma *vma =
-		container_of(active, struct i915_vma, last_read[idx]);
 	struct drm_i915_gem_object *obj = vma->obj;
 
-	GEM_BUG_ON(!i915_vma_has_active_engine(vma, idx));
-
-	i915_vma_clear_active(vma, idx);
-	if (i915_vma_is_active(vma))
+	GEM_BUG_ON(!i915_vma_is_active(vma));
+	if (--vma->active_count)
 		return;
 
 	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
@@ -108,6 +110,15 @@ i915_vma_retire(struct i915_gem_active *active, struct i915_request *rq)
 	}
 }
 
+static void
+i915_vma_retire(struct i915_gem_active *base, struct i915_request *rq)
+{
+	struct i915_vma_active *active =
+		container_of(base, typeof(*active), base);
+
+	__i915_vma_retire(active->vma, rq);
+}
+
 static struct i915_vma *
 vma_create(struct drm_i915_gem_object *obj,
 	   struct i915_address_space *vm,
@@ -115,7 +126,6 @@ vma_create(struct drm_i915_gem_object *obj,
 {
 	struct i915_vma *vma;
 	struct rb_node *rb, **p;
-	int i;
 
 	/* The aliasing_ppgtt should never be used directly! */
 	GEM_BUG_ON(vm == &vm->i915->mm.aliasing_ppgtt->vm);
@@ -124,8 +134,8 @@ vma_create(struct drm_i915_gem_object *obj,
 	if (vma == NULL)
 		return ERR_PTR(-ENOMEM);
 
-	for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
-		init_request_active(&vma->last_read[i], i915_vma_retire);
+	vma->active = RB_ROOT;
+
 	init_request_active(&vma->last_fence, NULL);
 	vma->vm = vm;
 	vma->ops = &vm->vma_ops;
@@ -778,13 +788,11 @@ void i915_vma_reopen(struct i915_vma *vma)
 static void __i915_vma_destroy(struct i915_vma *vma)
 {
 	struct drm_i915_private *i915 = vma->vm->i915;
-	int i;
+	struct i915_vma_active *iter, *n;
 
 	GEM_BUG_ON(vma->node.allocated);
 	GEM_BUG_ON(vma->fence);
 
-	for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
-		GEM_BUG_ON(i915_gem_active_isset(&vma->last_read[i]));
 	GEM_BUG_ON(i915_gem_active_isset(&vma->last_fence));
 
 	list_del(&vma->obj_link);
@@ -795,6 +803,11 @@ static void __i915_vma_destroy(struct i915_vma *vma)
 	if (!i915_vma_is_ggtt(vma))
 		i915_ppgtt_put(i915_vm_to_ppgtt(vma->vm));
 
+	rbtree_postorder_for_each_entry_safe(iter, n, &vma->active, node) {
+		GEM_BUG_ON(i915_gem_active_isset(&iter->base));
+		kfree(iter);
+	}
+
 	kmem_cache_free(i915->vmas, vma);
 }
 
@@ -878,16 +891,54 @@ static void export_fence(struct i915_vma *vma,
 	reservation_object_unlock(resv);
 }
 
+static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
+{
+	struct i915_vma_active *active;
+	struct rb_node **p, *parent;
+
+	parent = NULL;
+	p = &vma->active.rb_node;
+	while (*p) {
+		parent = *p;
+
+		active = rb_entry(parent, struct i915_vma_active, node);
+		if (active->timeline == idx)
+			return &active->base;
+
+		if (active->timeline < idx)
+			p = &parent->rb_right;
+		else
+			p = &parent->rb_left;
+	}
+
+	active = kmalloc(sizeof(*active), GFP_KERNEL);
+	if (unlikely(!active))
+		return ERR_PTR(-ENOMEM);
+
+	init_request_active(&active->base, i915_vma_retire);
+	active->vma = vma;
+	active->timeline = idx;
+
+	rb_link_node(&active->node, parent, p);
+	rb_insert_color(&active->node, &vma->active);
+
+	return &active->base;
+}
+
 int i915_vma_move_to_active(struct i915_vma *vma,
 			    struct i915_request *rq,
 			    unsigned int flags)
 {
 	struct drm_i915_gem_object *obj = vma->obj;
-	const unsigned int idx = rq->engine->id;
+	struct i915_gem_active *active;
 
 	lockdep_assert_held(&rq->i915->drm.struct_mutex);
 	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 
+	active = active_instance(vma, rq->fence.context);
+	if (IS_ERR(active))
+		return PTR_ERR(active);
+
 	/*
 	 * Add a reference if we're newly entering the active list.
 	 * The order in which we add operations to the retirement queue is
@@ -896,11 +947,13 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 	 * add the active reference first and queue for it to be dropped
 	 * *last*.
 	 */
-	if (!i915_vma_is_active(vma))
+	if (!i915_gem_active_isset(active) && !vma->active_count++) {
+		list_move_tail(&vma->vm_link, &vma->vm->active_list);
 		obj->active_count++;
-	i915_vma_set_active(vma, idx);
-	i915_gem_active_set(&vma->last_read[idx], rq);
-	list_move_tail(&vma->vm_link, &vma->vm->active_list);
+	}
+	i915_gem_active_set(active, rq);
+	GEM_BUG_ON(!i915_vma_is_active(vma));
+	GEM_BUG_ON(!obj->active_count);
 
 	obj->write_domain = 0;
 	if (flags & EXEC_OBJECT_WRITE) {
@@ -922,7 +975,6 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 
 int i915_vma_unbind(struct i915_vma *vma)
 {
-	unsigned long active;
 	int ret;
 
 	lockdep_assert_held(&vma->vm->i915->drm.struct_mutex);
@@ -932,9 +984,8 @@ int i915_vma_unbind(struct i915_vma *vma)
 	 * have side-effects such as unpinning or even unbinding this vma.
 	 */
 	might_sleep();
-	active = i915_vma_get_active(vma);
-	if (active) {
-		int idx;
+	if (i915_vma_is_active(vma)) {
+		struct i915_vma_active *active, *n;
 
 		/*
 		 * When a closed VMA is retired, it is unbound - eek.
@@ -951,18 +1002,17 @@ int i915_vma_unbind(struct i915_vma *vma)
 		 */
 		__i915_vma_pin(vma);
 
-		for_each_active(active, idx) {
-			ret = i915_gem_active_retire(&vma->last_read[idx],
+		rbtree_postorder_for_each_entry_safe(active, n,
+						     &vma->active, node) {
+			ret = i915_gem_active_retire(&active->base,
 						     &vma->vm->i915->drm.struct_mutex);
 			if (ret)
-				break;
-		}
-
-		if (!ret) {
-			ret = i915_gem_active_retire(&vma->last_fence,
-						     &vma->vm->i915->drm.struct_mutex);
+				goto unpin;
 		}
 
+		ret = i915_gem_active_retire(&vma->last_fence,
+					     &vma->vm->i915->drm.struct_mutex);
+unpin:
 		__i915_vma_unpin(vma);
 		if (ret)
 			return ret;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index a218b689e418..c297b0a0dc47 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -26,6 +26,7 @@
 #define __I915_VMA_H__
 
 #include <linux/io-mapping.h>
+#include <linux/rbtree.h>
 
 #include <drm/drm_mm.h>
 
@@ -94,8 +95,8 @@ struct i915_vma {
 #define I915_VMA_USERFAULT	BIT(I915_VMA_USERFAULT_BIT)
 #define I915_VMA_GGTT_WRITE	BIT(12)
 
-	unsigned int active;
-	struct i915_gem_active last_read[I915_NUM_ENGINES];
+	unsigned int active_count;
+	struct rb_root active;
 	struct i915_gem_active last_fence;
 
 	/**
@@ -138,6 +139,15 @@ i915_vma_instance(struct drm_i915_gem_object *obj,
 
 void i915_vma_unpin_and_release(struct i915_vma **p_vma);
 
+static inline bool i915_vma_is_active(struct i915_vma *vma)
+{
+	return vma->active_count;
+}
+
+int __must_check i915_vma_move_to_active(struct i915_vma *vma,
+					 struct i915_request *rq,
+					 unsigned int flags);
+
 static inline bool i915_vma_is_ggtt(const struct i915_vma *vma)
 {
 	return vma->flags & I915_VMA_GGTT;
@@ -187,38 +197,6 @@ static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
 	return test_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 }
 
-static inline unsigned int i915_vma_get_active(const struct i915_vma *vma)
-{
-	return vma->active;
-}
-
-static inline bool i915_vma_is_active(const struct i915_vma *vma)
-{
-	return i915_vma_get_active(vma);
-}
-
-static inline void i915_vma_set_active(struct i915_vma *vma,
-				       unsigned int engine)
-{
-	vma->active |= BIT(engine);
-}
-
-static inline void i915_vma_clear_active(struct i915_vma *vma,
-					 unsigned int engine)
-{
-	vma->active &= ~BIT(engine);
-}
-
-static inline bool i915_vma_has_active_engine(const struct i915_vma *vma,
-					      unsigned int engine)
-{
-	return vma->active & BIT(engine);
-}
-
-int __must_check i915_vma_move_to_active(struct i915_vma *vma,
-					 struct i915_request *rq,
-					 unsigned int flags);
-
 static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
 {
 	GEM_BUG_ON(!i915_vma_is_ggtt(vma));
-- 
2.18.0

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

[PATCH 6/6] drm/i915: Track the last-active inside the i915_vma

[Chris Wilson]

Using a VMA on more than one timeline concurrently is the exception
rather than the rule (using it concurrently on multiple engines). As we
expect to only use one active tracker, store the most recently used
tracker inside the i915_vma itself and only fallback to the rbtree if
we need a second or more concurrent active trackers.

v2: Comments on how we overwrite any existing last_active cache.
v3: __list_del_entry() before list_replace_init() is confusing and, much
more important, entirely redundant.
v4: Note that both last_active and the rbtree may be simultaneously
tracking this timeline, albeit with different requests, and so the vma
may be retired twice for the same timeline.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_vma.c | 56 +++++++++++++++++++++++++++++++--
 drivers/gpu/drm/i915/i915_vma.h |  1 +
 2 files changed, 55 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index b4cc98330225..6fbd09d6af28 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -119,6 +119,12 @@ i915_vma_retire(struct i915_gem_active *base, struct i915_request *rq)
 	__i915_vma_retire(active->vma, rq);
 }
 
+static void
+i915_vma_last_retire(struct i915_gem_active *base, struct i915_request *rq)
+{
+	__i915_vma_retire(container_of(base, struct i915_vma, last_active), rq);
+}
+
 static struct i915_vma *
 vma_create(struct drm_i915_gem_object *obj,
 	   struct i915_address_space *vm,
@@ -136,6 +142,7 @@ vma_create(struct drm_i915_gem_object *obj,
 
 	vma->active = RB_ROOT;
 
+	init_request_active(&vma->last_active, i915_vma_last_retire);
 	init_request_active(&vma->last_fence, NULL);
 	vma->vm = vm;
 	vma->ops = &vm->vma_ops;
@@ -895,6 +902,29 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 {
 	struct i915_vma_active *active;
 	struct rb_node **p, *parent;
+	struct i915_request *old;
+
+	/*
+	 * We track the most recently used timeline to skip a rbtree search
+	 * for the common case, under typical loads we never need the rbtree
+	 * at all. We can reuse the last_active slot if it is empty, that is
+	 * after the previous activity has been retired, or if the active
+	 * matches the current timeline.
+	 *
+	 * Note that we allow the timeline to be active simultaneously in
+	 * the rbtree and the last_active cache. We do this to avoid having
+	 * to search and replace the rbtree element for a new timeline, with
+	 * the cost being that we must be aware that the vma may be retired
+	 * twice for the same timeline (as the older rbtree element will be
+	 * retired before the new request added to last_active).
+	 */
+	old = i915_gem_active_raw(&vma->last_active,
+				  &vma->vm->i915->drm.struct_mutex);
+	if (!old || old->fence.context == idx)
+		goto out;
+
+	/* Move the currently active fence into the rbtree */
+	idx = old->fence.context;
 
 	parent = NULL;
 	p = &vma->active.rb_node;
@@ -903,7 +933,7 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 
 		active = rb_entry(parent, struct i915_vma_active, node);
 		if (active->timeline == idx)
-			return &active->base;
+			goto replace;
 
 		if (active->timeline < idx)
 			p = &parent->rb_right;
@@ -922,7 +952,24 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 	rb_link_node(&active->node, parent, p);
 	rb_insert_color(&active->node, &vma->active);
 
-	return &active->base;
+replace:
+	/*
+	 * Overwrite the previous active slot in the rbtree with last_active,
+	 * leaving last_active zeroed. If the previous slot is still active,
+	 * we must be careful as we now only expect to recieve one retire
+	 * callback not two, and so much undo the active counting for the
+	 * overwritten slot.
+	 */
+	if (i915_gem_active_isset(&active->base)) {
+		vma->active_count--;
+		GEM_BUG_ON(!vma->active_count);
+	}
+	GEM_BUG_ON(list_empty(&vma->last_active.link));
+	list_replace_init(&vma->last_active.link, &active->base.link);
+	active->base.request = fetch_and_zero(&vma->last_active.request);
+
+out:
+	return &vma->last_active;
 }
 
 int i915_vma_move_to_active(struct i915_vma *vma,
@@ -1002,6 +1049,11 @@ int i915_vma_unbind(struct i915_vma *vma)
 		 */
 		__i915_vma_pin(vma);
 
+		ret = i915_gem_active_retire(&vma->last_active,
+					     &vma->vm->i915->drm.struct_mutex);
+		if (ret)
+			goto unpin;
+
 		rbtree_postorder_for_each_entry_safe(active, n,
 						     &vma->active, node) {
 			ret = i915_gem_active_retire(&active->base,
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index c297b0a0dc47..f06d66377107 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -97,6 +97,7 @@ struct i915_vma {
 
 	unsigned int active_count;
 	struct rb_root active;
+	struct i915_gem_active last_active;
 	struct i915_gem_active last_fence;
 
 	/**
-- 
2.18.0

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

Re: [PATCH 6/6] drm/i915: Track the last-active inside the i915_vma

[Tvrtko Ursulin]

On 06/07/2018 11:39, Chris Wilson wrote:
> Using a VMA on more than one timeline concurrently is the exception
> rather than the rule (using it concurrently on multiple engines). As we
> expect to only use one active tracker, store the most recently used
> tracker inside the i915_vma itself and only fallback to the rbtree if
> we need a second or more concurrent active trackers.
> 
> v2: Comments on how we overwrite any existing last_active cache.
> v3: __list_del_entry() before list_replace_init() is confusing and, much
> more important, entirely redundant.
> v4: Note that both last_active and the rbtree may be simultaneously
> tracking this timeline, albeit with different requests, and so the vma
> may be retired twice for the same timeline.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> ---
>   drivers/gpu/drm/i915/i915_vma.c | 56 +++++++++++++++++++++++++++++++--
>   drivers/gpu/drm/i915/i915_vma.h |  1 +
>   2 files changed, 55 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index b4cc98330225..6fbd09d6af28 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -119,6 +119,12 @@ i915_vma_retire(struct i915_gem_active *base, struct i915_request *rq)
>   	__i915_vma_retire(active->vma, rq);
>   }
>   
> +static void
> +i915_vma_last_retire(struct i915_gem_active *base, struct i915_request *rq)
> +{
> +	__i915_vma_retire(container_of(base, struct i915_vma, last_active), rq);
> +}
> +
>   static struct i915_vma *
>   vma_create(struct drm_i915_gem_object *obj,
>   	   struct i915_address_space *vm,
> @@ -136,6 +142,7 @@ vma_create(struct drm_i915_gem_object *obj,
>   
>   	vma->active = RB_ROOT;
>   
> +	init_request_active(&vma->last_active, i915_vma_last_retire);
>   	init_request_active(&vma->last_fence, NULL);
>   	vma->vm = vm;
>   	vma->ops = &vm->vma_ops;
> @@ -895,6 +902,29 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
>   {
>   	struct i915_vma_active *active;
>   	struct rb_node **p, *parent;
> +	struct i915_request *old;
> +
> +	/*
> +	 * We track the most recently used timeline to skip a rbtree search
> +	 * for the common case, under typical loads we never need the rbtree
> +	 * at all. We can reuse the last_active slot if it is empty, that is
> +	 * after the previous activity has been retired, or if the active
> +	 * matches the current timeline.
> +	 *
> +	 * Note that we allow the timeline to be active simultaneously in
> +	 * the rbtree and the last_active cache. We do this to avoid having
> +	 * to search and replace the rbtree element for a new timeline, with
> +	 * the cost being that we must be aware that the vma may be retired
> +	 * twice for the same timeline (as the older rbtree element will be
> +	 * retired before the new request added to last_active).
> +	 */
> +	old = i915_gem_active_raw(&vma->last_active,
> +				  &vma->vm->i915->drm.struct_mutex);
> +	if (!old || old->fence.context == idx)
> +		goto out;
> +
> +	/* Move the currently active fence into the rbtree */
> +	idx = old->fence.context;
>   
>   	parent = NULL;
>   	p = &vma->active.rb_node;
> @@ -903,7 +933,7 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
>   
>   		active = rb_entry(parent, struct i915_vma_active, node);
>   		if (active->timeline == idx)
> -			return &active->base;
> +			goto replace;
>   
>   		if (active->timeline < idx)
>   			p = &parent->rb_right;
> @@ -922,7 +952,24 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
>   	rb_link_node(&active->node, parent, p);
>   	rb_insert_color(&active->node, &vma->active);
>   
> -	return &active->base;
> +replace:
> +	/*
> +	 * Overwrite the previous active slot in the rbtree with last_active,
> +	 * leaving last_active zeroed. If the previous slot is still active,
> +	 * we must be careful as we now only expect to recieve one retire
> +	 * callback not two, and so much undo the active counting for the
> +	 * overwritten slot.
> +	 */
> +	if (i915_gem_active_isset(&active->base)) {
> +		vma->active_count--;
> +		GEM_BUG_ON(!vma->active_count);
> +	}
> +	GEM_BUG_ON(list_empty(&vma->last_active.link));
> +	list_replace_init(&vma->last_active.link, &active->base.link);
> +	active->base.request = fetch_and_zero(&vma->last_active.request);
> +
> +out:
> +	return &vma->last_active;
>   }
>   
>   int i915_vma_move_to_active(struct i915_vma *vma,
> @@ -1002,6 +1049,11 @@ int i915_vma_unbind(struct i915_vma *vma)
>   		 */
>   		__i915_vma_pin(vma);
>   
> +		ret = i915_gem_active_retire(&vma->last_active,
> +					     &vma->vm->i915->drm.struct_mutex);
> +		if (ret)
> +			goto unpin;
> +
>   		rbtree_postorder_for_each_entry_safe(active, n,
>   						     &vma->active, node) {
>   			ret = i915_gem_active_retire(&active->base,
> diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
> index c297b0a0dc47..f06d66377107 100644
> --- a/drivers/gpu/drm/i915/i915_vma.h
> +++ b/drivers/gpu/drm/i915/i915_vma.h
> @@ -97,6 +97,7 @@ struct i915_vma {
>   
>   	unsigned int active_count;
>   	struct rb_root active;
> +	struct i915_gem_active last_active;
>   	struct i915_gem_active last_fence;
>   
>   	/**
> 

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Regards,

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

✗ Fi.CI.CHECKPATCH: warning for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()
URL   : https://patchwork.freedesktop.org/series/46067/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
478556e2da64 drm/i915: Refactor export_fence() after i915_vma_move_to_active()
fd54dbed2799 drm/i915: Export i915_request_skip()
809b8f016f80 drm/i915: Start returning an error from i915_vma_move_to_active()
ec40b966f936 drm/i915: Move i915_vma_move_to_active() to i915_vma.c
a3a860c56328 drm/i915: Track vma activity per fence.context, not per engine
bacf87522060 drm/i915: Track the last-active inside the i915_vma
-:96: WARNING:TYPO_SPELLING: 'recieve' may be misspelled - perhaps 'receive'?
#96: FILE: drivers/gpu/drm/i915/i915_vma.c:959:
+	 * we must be careful as we now only expect to recieve one retire

total: 0 errors, 1 warnings, 0 checks, 99 lines checked

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

✗ Fi.CI.SPARSE: warning for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()
URL   : https://patchwork.freedesktop.org/series/46067/
State : warning

== Summary ==

$ dim sparse origin/drm-tip
Commit: drm/i915: Refactor export_fence() after i915_vma_move_to_active()
Okay!

Commit: drm/i915: Export i915_request_skip()
Okay!

Commit: drm/i915: Start returning an error from i915_vma_move_to_active()
Okay!

Commit: drm/i915: Move i915_vma_move_to_active() to i915_vma.c
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3663:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3660:16: warning: expression using sizeof(void)

Commit: drm/i915: Track vma activity per fence.context, not per engine
Okay!

Commit: drm/i915: Track the last-active inside the i915_vma
Okay!

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

✗ Fi.CI.BAT: failure for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()
URL   : https://patchwork.freedesktop.org/series/46067/
State : failure

== Summary ==

= CI Bug Log - changes from CI_DRM_4444 -> Patchwork_9564 =

== Summary - FAILURE ==

  Serious unknown changes coming with Patchwork_9564 absolutely need to be
  verified manually.
  
  If you think the reported changes have nothing to do with the changes
  introduced in Patchwork_9564, please notify your bug team to allow them
  to document this new failure mode, which will reduce false positives in CI.

  External URL: https://patchwork.freedesktop.org/api/1.0/series/46067/revisions/1/mbox/

== Possible new issues ==

  Here are the unknown changes that may have been introduced in Patchwork_9564:

  === IGT changes ===

    ==== Possible regressions ====

    igt@gem_exec_nop@basic-parallel:
      {fi-kbl-x1275}:     PASS -> INCOMPLETE
      fi-kbl-7567u:       PASS -> INCOMPLETE
      fi-skl-6600u:       PASS -> INCOMPLETE
      fi-glk-dsi:         PASS -> INCOMPLETE
      {fi-kbl-8809g}:     PASS -> INCOMPLETE
      fi-kbl-r:           PASS -> INCOMPLETE
      fi-cfl-8700k:       PASS -> INCOMPLETE
      {fi-cfl-8109u}:     PASS -> INCOMPLETE
      fi-cfl-s3:          PASS -> INCOMPLETE
      fi-kbl-7500u:       PASS -> INCOMPLETE
      fi-skl-6770hq:      PASS -> INCOMPLETE
      fi-whl-u:           PASS -> INCOMPLETE
      fi-kbl-7560u:       PASS -> INCOMPLETE
      fi-skl-6700hq:      PASS -> INCOMPLETE
      fi-skl-6260u:       PASS -> INCOMPLETE
      fi-skl-6700k2:      PASS -> INCOMPLETE
      fi-bsw-n3050:       PASS -> INCOMPLETE

    igt@gem_exec_store@basic-all:
      fi-snb-2600:        PASS -> INCOMPLETE

    igt@gem_wait@basic-wait-all:
      fi-bdw-5557u:       PASS -> INCOMPLETE

    
== Known issues ==

  Here are the changes found in Patchwork_9564 that come from known issues:

  === IGT changes ===

    ==== Issues hit ====

    igt@gem_exec_nop@basic-parallel:
      fi-cfl-guc:         PASS -> INCOMPLETE (fdo#106693)
      fi-skl-guc:         PASS -> INCOMPLETE (fdo#106693)
      fi-kbl-guc:         PASS -> INCOMPLETE (fdo#106693)
      fi-bxt-dsi:         PASS -> INCOMPLETE (fdo#103927)
      fi-bxt-j4205:       PASS -> INCOMPLETE (fdo#103927)
      fi-glk-j4005:       PASS -> INCOMPLETE (fdo#103359, k.org#198133)

    igt@gem_exec_store@basic-all:
      fi-snb-2520m:       PASS -> INCOMPLETE (fdo#103713)

    igt@gem_exec_suspend@basic:
      fi-byt-j1900:       PASS -> INCOMPLETE (fdo#102657)

    igt@gem_exec_suspend@basic-s4-devices:
      fi-byt-n2820:       PASS -> INCOMPLETE (fdo#102657)

    igt@gem_wait@basic-wait-all:
      fi-skl-gvtdvm:      PASS -> INCOMPLETE (fdo#105600)
      fi-bdw-gvtdvm:      PASS -> INCOMPLETE (fdo#105600)

    
  {name}: This element is suppressed. This means it is ignored when computing
          the status of the difference (SUCCESS, WARNING, or FAILURE).

  fdo#102657 https://bugs.freedesktop.org/show_bug.cgi?id=102657
  fdo#103359 https://bugs.freedesktop.org/show_bug.cgi?id=103359
  fdo#103713 https://bugs.freedesktop.org/show_bug.cgi?id=103713
  fdo#103927 https://bugs.freedesktop.org/show_bug.cgi?id=103927
  fdo#105600 https://bugs.freedesktop.org/show_bug.cgi?id=105600
  fdo#106693 https://bugs.freedesktop.org/show_bug.cgi?id=106693
  k.org#198133 https://bugzilla.kernel.org/show_bug.cgi?id=198133


== Participating hosts (47 -> 42) ==

  Missing    (5): fi-ctg-p8600 fi-ilk-m540 fi-byt-squawks fi-bsw-cyan fi-hsw-4200u 


== Build changes ==

    * Linux: CI_DRM_4444 -> Patchwork_9564

  CI_DRM_4444: b153ce72348cbf6639dc5bd8e5e1bb0c180e9b86 @ git://anongit.freedesktop.org/gfx-ci/linux
  IGT_4540: 78071c2fa53db2f04b8eddc6e6118be4fbc5c2fe @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
  Patchwork_9564: bacf87522060199f7a786f2cc819a8a3cc29cfc4 @ git://anongit.freedesktop.org/gfx-ci/linux


== Linux commits ==

bacf87522060 drm/i915: Track the last-active inside the i915_vma
a3a860c56328 drm/i915: Track vma activity per fence.context, not per engine
ec40b966f936 drm/i915: Move i915_vma_move_to_active() to i915_vma.c
809b8f016f80 drm/i915: Start returning an error from i915_vma_move_to_active()
fd54dbed2799 drm/i915: Export i915_request_skip()
478556e2da64 drm/i915: Refactor export_fence() after i915_vma_move_to_active()

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_9564/issues.html
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: ✗ Fi.CI.BAT: failure for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()

[Chris Wilson]

Quoting Patchwork (2018-07-06 13:20:19)
> == Series Details ==
> 
> Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active()
> URL   : https://patchwork.freedesktop.org/series/46067/
> State : failure
> 
> == Summary ==
> 
> = CI Bug Log - changes from CI_DRM_4444 -> Patchwork_9564 =
> 
> == Summary - FAILURE ==
> 
>   Serious unknown changes coming with Patchwork_9564 absolutely need to be
>   verified manually.
>   
>   If you think the reported changes have nothing to do with the changes
>   introduced in Patchwork_9564, please notify your bug team to allow them
>   to document this new failure mode, which will reduce false positives in CI.
> 
>   External URL: https://patchwork.freedesktop.org/api/1.0/series/46067/revisions/1/mbox/
> 
> == Possible new issues ==
> 
>   Here are the unknown changes that may have been introduced in Patchwork_9564:
> 
>   === IGT changes ===
> 
>     ==== Possible regressions ====
> 
>     igt@gem_exec_nop@basic-parallel:
>       {fi-kbl-x1275}:     PASS -> INCOMPLETE
>       fi-kbl-7567u:       PASS -> INCOMPLETE
>       fi-skl-6600u:       PASS -> INCOMPLETE
>       fi-glk-dsi:         PASS -> INCOMPLETE
>       {fi-kbl-8809g}:     PASS -> INCOMPLETE
>       fi-kbl-r:           PASS -> INCOMPLETE
>       fi-cfl-8700k:       PASS -> INCOMPLETE
>       {fi-cfl-8109u}:     PASS -> INCOMPLETE
>       fi-cfl-s3:          PASS -> INCOMPLETE
>       fi-kbl-7500u:       PASS -> INCOMPLETE
>       fi-skl-6770hq:      PASS -> INCOMPLETE
>       fi-whl-u:           PASS -> INCOMPLETE
>       fi-kbl-7560u:       PASS -> INCOMPLETE
>       fi-skl-6700hq:      PASS -> INCOMPLETE
>       fi-skl-6260u:       PASS -> INCOMPLETE
>       fi-skl-6700k2:      PASS -> INCOMPLETE
>       fi-bsw-n3050:       PASS -> INCOMPLETE

Not so redundant after all! Hah. Oh, it is just doing a list_del and is
definitely required.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v5] drm/i915: Track the last-active inside the i915_vma

[Chris Wilson]

Using a VMA on more than one timeline concurrently is the exception
rather than the rule (using it concurrently on multiple engines). As we
expect to only use one active tracker, store the most recently used
tracker inside the i915_vma itself and only fallback to the rbtree if
we need a second or more concurrent active trackers.

v2: Comments on how we overwrite any existing last_active cache.
v3: __list_del_entry() before list_replace_init() is confusing and, much
more important, entirely redundant.
v4: Note that both last_active and the rbtree may be simultaneously
tracking this timeline, albeit with different requests, and so the vma
may be retired twice for the same timeline.
v5: No, that list_del is required!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_vma.c | 58 +++++++++++++++++++++++++++++++--
 drivers/gpu/drm/i915/i915_vma.h |  1 +
 2 files changed, 57 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index b4cc98330225..8943daa5c6e3 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -119,6 +119,12 @@ i915_vma_retire(struct i915_gem_active *base, struct i915_request *rq)
 	__i915_vma_retire(active->vma, rq);
 }
 
+static void
+i915_vma_last_retire(struct i915_gem_active *base, struct i915_request *rq)
+{
+	__i915_vma_retire(container_of(base, struct i915_vma, last_active), rq);
+}
+
 static struct i915_vma *
 vma_create(struct drm_i915_gem_object *obj,
 	   struct i915_address_space *vm,
@@ -136,6 +142,7 @@ vma_create(struct drm_i915_gem_object *obj,
 
 	vma->active = RB_ROOT;
 
+	init_request_active(&vma->last_active, i915_vma_last_retire);
 	init_request_active(&vma->last_fence, NULL);
 	vma->vm = vm;
 	vma->ops = &vm->vma_ops;
@@ -895,6 +902,29 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 {
 	struct i915_vma_active *active;
 	struct rb_node **p, *parent;
+	struct i915_request *old;
+
+	/*
+	 * We track the most recently used timeline to skip a rbtree search
+	 * for the common case, under typical loads we never need the rbtree
+	 * at all. We can reuse the last_active slot if it is empty, that is
+	 * after the previous activity has been retired, or if the active
+	 * matches the current timeline.
+	 *
+	 * Note that we allow the timeline to be active simultaneously in
+	 * the rbtree and the last_active cache. We do this to avoid having
+	 * to search and replace the rbtree element for a new timeline, with
+	 * the cost being that we must be aware that the vma may be retired
+	 * twice for the same timeline (as the older rbtree element will be
+	 * retired before the new request added to last_active).
+	 */
+	old = i915_gem_active_raw(&vma->last_active,
+				  &vma->vm->i915->drm.struct_mutex);
+	if (!old || old->fence.context == idx)
+		goto out;
+
+	/* Move the currently active fence into the rbtree */
+	idx = old->fence.context;
 
 	parent = NULL;
 	p = &vma->active.rb_node;
@@ -903,7 +933,7 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 
 		active = rb_entry(parent, struct i915_vma_active, node);
 		if (active->timeline == idx)
-			return &active->base;
+			goto replace;
 
 		if (active->timeline < idx)
 			p = &parent->rb_right;
@@ -922,7 +952,26 @@ static struct i915_gem_active *active_instance(struct i915_vma *vma, u64 idx)
 	rb_link_node(&active->node, parent, p);
 	rb_insert_color(&active->node, &vma->active);
 
-	return &active->base;
+replace:
+	/*
+	 * Overwrite the previous active slot in the rbtree with last_active,
+	 * leaving last_active zeroed. If the previous slot is still active,
+	 * we must be careful as we now only expect to recieve one retire
+	 * callback not two, and so much undo the active counting for the
+	 * overwritten slot.
+	 */
+	if (i915_gem_active_isset(&active->base)) {
+		/* Retire ourselves from the old rq->active_list */
+		__list_del_entry(&active->base.link);
+		vma->active_count--;
+		GEM_BUG_ON(!vma->active_count);
+	}
+	GEM_BUG_ON(list_empty(&vma->last_active.link));
+	list_replace_init(&vma->last_active.link, &active->base.link);
+	active->base.request = fetch_and_zero(&vma->last_active.request);
+
+out:
+	return &vma->last_active;
 }
 
 int i915_vma_move_to_active(struct i915_vma *vma,
@@ -1002,6 +1051,11 @@ int i915_vma_unbind(struct i915_vma *vma)
 		 */
 		__i915_vma_pin(vma);
 
+		ret = i915_gem_active_retire(&vma->last_active,
+					     &vma->vm->i915->drm.struct_mutex);
+		if (ret)
+			goto unpin;
+
 		rbtree_postorder_for_each_entry_safe(active, n,
 						     &vma->active, node) {
 			ret = i915_gem_active_retire(&active->base,
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index c297b0a0dc47..f06d66377107 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -97,6 +97,7 @@ struct i915_vma {
 
 	unsigned int active_count;
 	struct rb_root active;
+	struct i915_gem_active last_active;
 	struct i915_gem_active last_fence;
 
 	/**
-- 
2.18.0

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

✗ Fi.CI.CHECKPATCH: warning for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)
URL   : https://patchwork.freedesktop.org/series/46067/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
c404016bcc1d drm/i915: Refactor export_fence() after i915_vma_move_to_active()
951d50a80f49 drm/i915: Export i915_request_skip()
39fbb254f607 drm/i915: Start returning an error from i915_vma_move_to_active()
78318c338989 drm/i915: Move i915_vma_move_to_active() to i915_vma.c
42d0e21316dd drm/i915: Track vma activity per fence.context, not per engine
a65a91a8871c drm/i915: Track the last-active inside the i915_vma
-:97: WARNING:TYPO_SPELLING: 'recieve' may be misspelled - perhaps 'receive'?
#97: FILE: drivers/gpu/drm/i915/i915_vma.c:959:
+	 * we must be careful as we now only expect to recieve one retire

total: 0 errors, 1 warnings, 0 checks, 101 lines checked

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

✗ Fi.CI.SPARSE: warning for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)
URL   : https://patchwork.freedesktop.org/series/46067/
State : warning

== Summary ==

$ dim sparse origin/drm-tip
Commit: drm/i915: Refactor export_fence() after i915_vma_move_to_active()
Okay!

Commit: drm/i915: Export i915_request_skip()
Okay!

Commit: drm/i915: Start returning an error from i915_vma_move_to_active()
Okay!

Commit: drm/i915: Move i915_vma_move_to_active() to i915_vma.c
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3655:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3652:16: warning: expression using sizeof(void)

Commit: drm/i915: Track vma activity per fence.context, not per engine
Okay!

Commit: drm/i915: Track the last-active inside the i915_vma
Okay!

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

✓ Fi.CI.BAT: success for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)
URL   : https://patchwork.freedesktop.org/series/46067/
State : success

== Summary ==

= CI Bug Log - changes from CI_DRM_4446 -> Patchwork_9568 =

== Summary - SUCCESS ==

  No regressions found.

  External URL: https://patchwork.freedesktop.org/api/1.0/series/46067/revisions/2/mbox/

== Known issues ==

  Here are the changes found in Patchwork_9568 that come from known issues:

  === IGT changes ===

    ==== Issues hit ====

    igt@prime_vgem@basic-fence-flip:
      fi-ilk-650:         PASS -> FAIL (fdo#104008)

    
  fdo#104008 https://bugs.freedesktop.org/show_bug.cgi?id=104008


== Participating hosts (47 -> 42) ==

  Missing    (5): fi-ctg-p8600 fi-ilk-m540 fi-byt-squawks fi-bsw-cyan fi-hsw-4200u 


== Build changes ==

    * Linux: CI_DRM_4446 -> Patchwork_9568

  CI_DRM_4446: 95944426a9ffda186843c78f2f925494e1bc53c5 @ git://anongit.freedesktop.org/gfx-ci/linux
  IGT_4543: 366eed37c7c71217e1cb1f3be5e26358a41f0001 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
  Patchwork_9568: a65a91a8871c69c8a80db2a1da700aaa4eb113b4 @ git://anongit.freedesktop.org/gfx-ci/linux


== Linux commits ==

a65a91a8871c drm/i915: Track the last-active inside the i915_vma
42d0e21316dd drm/i915: Track vma activity per fence.context, not per engine
78318c338989 drm/i915: Move i915_vma_move_to_active() to i915_vma.c
39fbb254f607 drm/i915: Start returning an error from i915_vma_move_to_active()
951d50a80f49 drm/i915: Export i915_request_skip()
c404016bcc1d drm/i915: Refactor export_fence() after i915_vma_move_to_active()

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_9568/issues.html
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✓ Fi.CI.IGT: success for series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)

[Patchwork]

== Series Details ==

Series: series starting with [1/6] drm/i915: Refactor export_fence() after i915_vma_move_to_active() (rev2)
URL   : https://patchwork.freedesktop.org/series/46067/
State : success

== Summary ==

= CI Bug Log - changes from CI_DRM_4446_full -> Patchwork_9568_full =

== Summary - WARNING ==

  Minor unknown changes coming with Patchwork_9568_full need to be verified
  manually.
  
  If you think the reported changes have nothing to do with the changes
  introduced in Patchwork_9568_full, please notify your bug team to allow them
  to document this new failure mode, which will reduce false positives in CI.

  

== Possible new issues ==

  Here are the unknown changes that may have been introduced in Patchwork_9568_full:

  === IGT changes ===

    ==== Warnings ====

    igt@gem_exec_schedule@deep-bsd1:
      shard-kbl:          SKIP -> PASS

    igt@kms_vblank@pipe-b-query-busy-hang:
      shard-snb:          PASS -> SKIP

    
== Known issues ==

  Here are the changes found in Patchwork_9568_full that come from known issues:

  === IGT changes ===

    ==== Issues hit ====

    igt@drv_selftest@live_gtt:
      shard-glk:          PASS -> INCOMPLETE (fdo#103359, k.org#198133, fdo#107127)

    igt@kms_cursor_legacy@2x-nonblocking-modeset-vs-cursor-atomic:
      shard-glk:          PASS -> FAIL (fdo#106509, fdo#105454)

    igt@kms_rotation_crc@primary-rotation-180:
      shard-hsw:          PASS -> FAIL (fdo#103925)

    igt@kms_setmode@basic:
      shard-apl:          PASS -> FAIL (fdo#99912)

    
    ==== Possible fixes ====

    igt@gem_workarounds@suspend-resume-context:
      shard-kbl:          INCOMPLETE (fdo#103665) -> PASS

    igt@kms_busy@extended-pageflip-hang-oldfb-render-a:
      shard-snb:          INCOMPLETE (fdo#105411) -> PASS

    igt@kms_cursor_legacy@2x-long-nonblocking-modeset-vs-cursor-atomic:
      shard-glk:          FAIL (fdo#106509) -> PASS

    igt@kms_flip_tiling@flip-to-y-tiled:
      shard-glk:          FAIL (fdo#103822) -> PASS

    igt@kms_flip_tiling@flip-y-tiled:
      shard-glk:          FAIL -> PASS

    igt@kms_universal_plane@cursor-fb-leak-pipe-c:
      shard-apl:          FAIL -> PASS

    
  fdo#103359 https://bugs.freedesktop.org/show_bug.cgi?id=103359
  fdo#103665 https://bugs.freedesktop.org/show_bug.cgi?id=103665
  fdo#103822 https://bugs.freedesktop.org/show_bug.cgi?id=103822
  fdo#103925 https://bugs.freedesktop.org/show_bug.cgi?id=103925
  fdo#105411 https://bugs.freedesktop.org/show_bug.cgi?id=105411
  fdo#105454 https://bugs.freedesktop.org/show_bug.cgi?id=105454
  fdo#106509 https://bugs.freedesktop.org/show_bug.cgi?id=106509
  fdo#107127 https://bugs.freedesktop.org/show_bug.cgi?id=107127
  fdo#99912 https://bugs.freedesktop.org/show_bug.cgi?id=99912
  k.org#198133 https://bugzilla.kernel.org/show_bug.cgi?id=198133


== Participating hosts (5 -> 5) ==

  No changes in participating hosts


== Build changes ==

    * Linux: CI_DRM_4446 -> Patchwork_9568

  CI_DRM_4446: 95944426a9ffda186843c78f2f925494e1bc53c5 @ git://anongit.freedesktop.org/gfx-ci/linux
  IGT_4543: 366eed37c7c71217e1cb1f3be5e26358a41f0001 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
  Patchwork_9568: a65a91a8871c69c8a80db2a1da700aaa4eb113b4 @ git://anongit.freedesktop.org/gfx-ci/linux
  piglit_4509: fdc5a4ca11124ab8413c7988896eec4c97336694 @ git://anongit.freedesktop.org/piglit

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_9568/shards.html
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx