[Intel-gfx] [PATCH 35/53] drm/i915/bdw: Workload submission mechanism for Execlists

oscar.mateo at intel.com oscar.mateo at intel.com
Fri Jun 13 17:37:53 CEST 2014


From: Oscar Mateo <oscar.mateo at intel.com>

This is what i915_gem_do_execbuffer calls when it wants to execute some
worload in an Execlists world. It's a candidate for abstracting the
submission mechanism away.

Signed-off-by: Oscar Mateo <oscar.mateo at intel.com>
---
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |   8 +-
 drivers/gpu/drm/i915/intel_lrc.c           | 138 +++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/intel_lrc.h           |   8 ++
 3 files changed, 152 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 366469d..36c7f0c 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1379,8 +1379,12 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 	else
 		exec_start += i915_gem_obj_offset(batch_obj, vm);
 
-	ret = legacy_ringbuffer_submission(dev, file, ring, ctx,
-			args, &eb->vmas, batch_obj, exec_start, flags);
+	if (intel_enable_execlists(dev))
+		ret = intel_execlists_submission(dev, file, ring, ctx,
+				args, &eb->vmas, batch_obj, exec_start, flags);
+	else
+		ret = legacy_ringbuffer_submission(dev, file, ring, ctx,
+				args, &eb->vmas, batch_obj, exec_start, flags);
 	if (ret)
 		goto err;
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 27fde8d..c9a5e00 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -130,6 +130,144 @@ static int logical_ring_flush_all_caches(struct intel_engine_cs *ring,
 	return 0;
 }
 
+static int logical_ring_invalidate_all_caches(struct intel_engine_cs *ring,
+					      struct intel_context *ctx)
+{
+	uint32_t flush_domains;
+	int ret;
+
+	flush_domains = 0;
+	if (ring->gpu_caches_dirty)
+		flush_domains = I915_GEM_GPU_DOMAINS;
+
+	ret = ring->emit_flush(ring, ctx, I915_GEM_GPU_DOMAINS, flush_domains);
+	if (ret)
+		return ret;
+
+	ring->gpu_caches_dirty = false;
+	return 0;
+}
+
+static int execlists_move_to_gpu(struct intel_engine_cs *ring,
+				 struct intel_context *ctx,
+				 struct list_head *vmas)
+{
+	struct i915_vma *vma;
+	uint32_t flush_domains = 0;
+	bool flush_chipset = false;
+	int ret;
+
+	list_for_each_entry(vma, vmas, exec_list) {
+		struct drm_i915_gem_object *obj = vma->obj;
+		ret = i915_gem_object_sync(obj, ring);
+		if (ret)
+			return ret;
+
+		if (obj->base.write_domain & I915_GEM_DOMAIN_CPU)
+			flush_chipset |= i915_gem_clflush_object(obj, false);
+
+		flush_domains |= obj->base.write_domain;
+	}
+
+	if (flush_chipset)
+		i915_gem_chipset_flush(ring->dev);
+
+	if (flush_domains & I915_GEM_DOMAIN_GTT)
+		wmb();
+
+	/* Unconditionally invalidate gpu caches and ensure that we do flush
+	 * any residual writes from the previous batch.
+	 */
+	return logical_ring_invalidate_all_caches(ring, ctx);
+}
+
+int intel_execlists_submission(struct drm_device *dev, struct drm_file *file,
+			       struct intel_engine_cs *ring,
+			       struct intel_context *ctx,
+			       struct drm_i915_gem_execbuffer2 *args,
+			       struct list_head *vmas,
+			       struct drm_i915_gem_object *batch_obj,
+			       u64 exec_start, u32 flags)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct intel_ringbuffer *ringbuf = logical_ringbuf_get(ring, ctx);
+	int instp_mode;
+	u32 instp_mask;
+	int ret;
+
+	if (args->num_cliprects != 0) {
+		DRM_DEBUG("clip rectangles are only valid on pre-gen5\n");
+		return -EINVAL;
+	} else {
+		if (args->DR4 == 0xffffffff) {
+			DRM_DEBUG("UXA submitting garbage DR4, fixing up\n");
+			args->DR4 = 0;
+		}
+
+		if (args->DR1 || args->DR4 || args->cliprects_ptr) {
+			DRM_DEBUG("0 cliprects but dirt in cliprects fields\n");
+			return -EINVAL;
+		}
+	}
+
+	ret = execlists_move_to_gpu(ring, ctx, vmas);
+	if (ret)
+		return ret;
+
+	instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK;
+	instp_mask = I915_EXEC_CONSTANTS_MASK;
+	/* The HW changed the meaning on this bit on gen6 */
+	instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE;
+	switch (instp_mode) {
+	case I915_EXEC_CONSTANTS_REL_GENERAL:
+		break;
+
+	case I915_EXEC_CONSTANTS_ABSOLUTE:
+		if (ring != &dev_priv->ring[RCS]) {
+			DRM_DEBUG("non-0 rel constants mode on non-RCS\n");
+			return -EINVAL;
+		}
+		break;
+
+	case I915_EXEC_CONSTANTS_REL_SURFACE:
+		DRM_DEBUG("rel surface constants mode invalid on gen5+\n");
+		return -EINVAL;
+
+	default:
+		DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode);
+		return -EINVAL;
+	}
+
+	if (ring == &dev_priv->ring[RCS] &&
+			instp_mode != dev_priv->relative_constants_mode) {
+		ret = intel_logical_ring_begin(ring, ctx, 4);
+		if (ret)
+			return ret;
+
+		intel_logical_ring_emit(ringbuf, MI_NOOP);
+		intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1));
+		intel_logical_ring_emit(ringbuf, INSTPM);
+		intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode);
+		intel_logical_ring_advance(ringbuf);
+
+		dev_priv->relative_constants_mode = instp_mode;
+	}
+
+	if (args->flags & I915_EXEC_GEN7_SOL_RESET) {
+		DRM_DEBUG("sol reset is gen7 only\n");
+		return -EINVAL;
+	}
+
+	ret = ring->emit_bb_start(ring, ctx, exec_start, flags);
+	if (ret)
+		return ret;
+
+	i915_gem_execbuffer_move_to_active(vmas, ring);
+	i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj);
+
+	return 0;
+}
+
 int intel_logical_ring_add_request(struct intel_engine_cs *ring,
 				   struct drm_file *file,
 				   struct drm_i915_gem_object *obj,
diff --git a/drivers/gpu/drm/i915/intel_lrc.h b/drivers/gpu/drm/i915/intel_lrc.h
index 4495359..0cb7cb5 100644
--- a/drivers/gpu/drm/i915/intel_lrc.h
+++ b/drivers/gpu/drm/i915/intel_lrc.h
@@ -5,6 +5,14 @@
 void intel_logical_ring_cleanup(struct intel_engine_cs *ring);
 int intel_logical_rings_init(struct drm_device *dev);
 
+int intel_execlists_submission(struct drm_device *dev, struct drm_file *file,
+			       struct intel_engine_cs *ring,
+			       struct intel_context *ctx,
+			       struct drm_i915_gem_execbuffer2 *args,
+			       struct list_head *vmas,
+			       struct drm_i915_gem_object *batch_obj,
+			       u64 exec_start, u32 flags);
+
 int intel_logical_ring_add_request(struct intel_engine_cs *ring,
 				   struct drm_file *file,
 				   struct drm_i915_gem_object *obj,
-- 
1.9.0




More information about the Intel-gfx mailing list