[Intel-gfx] [PATCH v3] drm/i915/guc: Move GuC wq_check_space to alloc_request_extras

[yu.dai at intel.com]

From: Alex Dai <yu.dai at intel.com>

Split GuC work queue space checking from submission and move it to
ring_alloc_request_extras. The reason is that failure in later
i915_add_request() won't be handled. In the case timeout happens,
driver can return early in order to handle the error.

v1: Move wq_reserve_space to ring_reserve_space
v2: Move wq_reserve_space to alloc_request_extras (Chris Wilson)
v3: The work queue head pointer is cached by driver now. So we can
    quickly return if space is available.
    s/reserve/check/g (Dave Gordon)

Signed-off-by: Alex Dai <yu.dai at intel.com>
---
 drivers/gpu/drm/i915/i915_guc_submission.c | 34 ++++++++++++++++++------------
 drivers/gpu/drm/i915/intel_guc.h           |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c           |  6 ++++++
 3 files changed, 28 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 226e9c0..cb8e1f71 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -207,6 +207,9 @@ static int guc_ring_doorbell(struct i915_guc_client *gc)
 	/* Update the tail so it is visible to GuC */
 	desc->tail = gc->wq_tail;
 
+	/* Cache the head where GuC is processing */
+	gc->wq_head = desc->head;
+
 	/* current cookie */
 	db_cmp.db_status = GUC_DOORBELL_ENABLED;
 	db_cmp.cookie = gc->cookie;
@@ -472,28 +475,30 @@ static void guc_fini_ctx_desc(struct intel_guc *guc,
 			     sizeof(desc) * client->ctx_index);
 }
 
-/* Get valid workqueue item and return it back to offset */
-static int guc_get_workqueue_space(struct i915_guc_client *gc, u32 *offset)
+int i915_guc_wq_check_space(struct i915_guc_client *gc)
 {
 	struct guc_process_desc *desc;
 	void *base;
 	u32 size = sizeof(struct guc_wq_item);
 	int ret = -ETIMEDOUT, timeout_counter = 200;
 
+	if (!gc)
+		return 0;
+
+	/* Quickly return if wq space is available since last time we cache the
+	 * head position. */
+	if (CIRC_SPACE(gc->wq_tail, gc->wq_head, gc->wq_size) >= size)
+		return 0;
+
 	base = kmap_atomic(i915_gem_object_get_page(gc->client_obj, 0));
 	desc = base + gc->proc_desc_offset;
 
 	while (timeout_counter-- > 0) {
-		if (CIRC_SPACE(gc->wq_tail, desc->head, gc->wq_size) >= size) {
-			*offset = gc->wq_tail;
+		gc->wq_head = desc->head;
 
-			/* advance the tail for next workqueue item */
-			gc->wq_tail += size;
-			gc->wq_tail &= gc->wq_size - 1;
-
-			/* this will break the loop */
-			timeout_counter = 0;
+		if (CIRC_SPACE(gc->wq_tail, gc->wq_head, gc->wq_size) >= size) {
 			ret = 0;
+			break;
 		}
 
 		if (timeout_counter)
@@ -512,11 +517,8 @@ static int guc_add_workqueue_item(struct i915_guc_client *gc,
 	struct guc_wq_item *wqi;
 	void *base;
 	u32 tail, wq_len, wq_off = 0;
-	int ret;
 
-	ret = guc_get_workqueue_space(gc, &wq_off);
-	if (ret)
-		return ret;
+	wq_off = gc->wq_tail;
 
 	/* For now workqueue item is 4 DWs; workqueue buffer is 2 pages. So we
 	 * should not have the case where structure wqi is across page, neither
@@ -551,6 +553,10 @@ static int guc_add_workqueue_item(struct i915_guc_client *gc,
 
 	kunmap_atomic(base);
 
+	/* advance the tail for next workqueue item */
+	gc->wq_tail += sizeof(struct guc_wq_item);
+	gc->wq_tail &= gc->wq_size - 1;
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/i915/intel_guc.h b/drivers/gpu/drm/i915/intel_guc.h
index 0e048bf..5cf555d 100644
--- a/drivers/gpu/drm/i915/intel_guc.h
+++ b/drivers/gpu/drm/i915/intel_guc.h
@@ -43,6 +43,7 @@ struct i915_guc_client {
 	uint32_t wq_offset;
 	uint32_t wq_size;
 	uint32_t wq_tail;
+	uint32_t wq_head;
 
 	/* GuC submission statistics & status */
 	uint64_t submissions[I915_NUM_RINGS];
@@ -123,5 +124,6 @@ int i915_guc_submit(struct i915_guc_client *client,
 		    struct drm_i915_gem_request *rq);
 void i915_guc_submission_disable(struct drm_device *dev);
 void i915_guc_submission_fini(struct drm_device *dev);
+int i915_guc_wq_check_space(struct i915_guc_client *client);
 
 #endif
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index f96fb51..9605ce9 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -667,6 +667,12 @@ int intel_logical_ring_alloc_request_extras(struct drm_i915_gem_request *request
 			return ret;
 	}
 
+	/* Reserve GuC WQ space here (one request needs one WQ item) because
+	 * the later i915_add_request() call can't fail. */
+	ret = i915_guc_wq_check_space(request->i915->guc.execbuf_client);
+	if (ret)
+		return ret;
+
 	return 0;
 }
 
-- 
2.5.0