[PATCH 01/37] drm/i915: Support asynchronous waits on struct fence from i915_gem_request

[Chris Wilson]

We will need to wait on DMA completion (as signaled via struct fence)
before executing our i915_gem_request. Therefore we want to expose a
method for adding the await on the fence itself to the request.

v2: Add a comment detailing a failure to handle a signal-on-any
fence-array.
v3: Pretend that magic numbers don't exist.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h            |  3 ++
 drivers/gpu/drm/i915/i915_gem.c            |  2 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |  3 +-
 drivers/gpu/drm/i915/i915_gem_request.c    | 48 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_request.h    |  6 ++--
 5 files changed, 58 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 55afb664b2f8..6989df2bf6d1 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1417,6 +1417,9 @@ struct i915_error_state_file_priv {
 	struct drm_i915_error_state *error;
 };
 
+#define I915_RESET_TIMEOUT (10*HZ)
+#define I915_FENCE_TIMEOUT (10*HZ)
+
 struct i915_gpu_error {
 	/* For hangcheck timer */
 #define DRM_I915_HANGCHECK_PERIOD 1500 /* in ms */
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 87018df94d68..0c186b75ec86 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -114,7 +114,7 @@ i915_gem_wait_for_error(struct i915_gpu_error *error)
 	 */
 	ret = wait_event_interruptible_timeout(error->reset_queue,
 					       !i915_reset_in_progress(error),
-					       10*HZ);
+					       I915_RESET_TIMEOUT);
 	if (ret == 0) {
 		DRM_ERROR("Timed out waiting for the gpu reset to complete\n");
 		return -EIO;
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index e52affdcc125..cdd7b3d49a41 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1142,7 +1142,8 @@ i915_gem_execbuffer_move_to_gpu(struct drm_i915_gem_request *req,
 		if (resv) {
 			ret = i915_sw_fence_await_reservation
 				(&req->submit, resv, &i915_fence_ops,
-				 obj->base.pending_write_domain, 10*HZ,
+				 obj->base.pending_write_domain,
+				 I915_FENCE_TIMEOUT,
 				 GFP_KERNEL | __GFP_NOWARN);
 			if (ret < 0)
 				return ret;
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index f9af2a00625e..5e38bc04a4f0 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -23,6 +23,7 @@
  */
 
 #include <linux/prefetch.h>
+#include <linux/dma-fence-array.h>
 
 #include "i915_drv.h"
 
@@ -496,6 +497,53 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to,
 	return 0;
 }
 
+int
+i915_gem_request_await_dma_fence(struct drm_i915_gem_request *req,
+				 struct dma_fence *fence)
+{
+	struct dma_fence_array *array;
+	int ret;
+	int i;
+
+	if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
+		return 0;
+
+	if (dma_fence_is_i915(fence))
+		return i915_gem_request_await_request(req, to_request(fence));
+
+	if (!dma_fence_is_array(fence)) {
+		ret = i915_sw_fence_await_dma_fence(&req->submit,
+						    fence, I915_FENCE_TIMEOUT,
+						    GFP_KERNEL);
+		return ret < 0 ? ret : 0;
+	}
+
+	/* Note that if the fence-array was created in signal-on-any mode,
+	 * we should *not* decompose it into its individual fences. However,
+	 * we don't currently store which mode the fence-array is operating
+	 * in. Fortunately, the only user of signal-on-any is private to
+	 * amdgpu and we should not see any incoming fence-array from
+	 * sync-file being in signal-on-any mode.
+	 */
+
+	array = to_dma_fence_array(fence);
+	for (i = 0; i < array->num_fences; i++) {
+		struct dma_fence *child = array->fences[i];
+
+		if (dma_fence_is_i915(child))
+			ret = i915_gem_request_await_request(req,
+							     to_request(child));
+		else
+			ret = i915_sw_fence_await_dma_fence(&req->submit,
+							    child, I915_FENCE_TIMEOUT,
+							    GFP_KERNEL);
+		if (ret < 0)
+			return ret;
+	}
+
+	return 0;
+}
+
 /**
  * i915_gem_request_await_object - set this request to (async) wait upon a bo
  *
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index bceeaa3a5193..4e6d038cc9de 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -147,7 +147,7 @@ struct drm_i915_gem_request {
 
 extern const struct dma_fence_ops i915_fence_ops;
 
-static inline bool fence_is_i915(struct dma_fence *fence)
+static inline bool dma_fence_is_i915(const struct dma_fence *fence)
 {
 	return fence->ops == &i915_fence_ops;
 }
@@ -176,7 +176,7 @@ to_request(struct dma_fence *fence)
 {
 	/* We assume that NULL fence/request are interoperable */
 	BUILD_BUG_ON(offsetof(struct drm_i915_gem_request, fence) != 0);
-	GEM_BUG_ON(fence && !fence_is_i915(fence));
+	GEM_BUG_ON(fence && !dma_fence_is_i915(fence));
 	return container_of(fence, struct drm_i915_gem_request, fence);
 }
 
@@ -214,6 +214,8 @@ int
 i915_gem_request_await_object(struct drm_i915_gem_request *to,
 			      struct drm_i915_gem_object *obj,
 			      bool write);
+int i915_gem_request_await_dma_fence(struct drm_i915_gem_request *req,
+				     struct dma_fence *fence);
 
 void __i915_add_request(struct drm_i915_gem_request *req, bool flush_caches);
 #define i915_add_request(req) \
-- 
2.10.1