[Intel-gfx] [PATCH v2 3/3] drm/i915/guc: drop cached copy of 'wq_head'

[Dave Gordon]

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

Now that we keep the GuC client structure (and therefore the process
descriptor embedded therein) permanently mapped, we don't really need
to keep a local copy of the GuC's work-queue-head. So we can simplify
the code a little by not doing this. Also, optimise away a few calls
by caching results in local variables.

v2:
    Added local optimisations (Dave Gordon)

Signed-off-by: Alex Dai <yu.dai at intel.com>
Signed-off-by: Dave Gordon <david.s.gordon at intel.com>
---
 drivers/gpu/drm/i915/i915_guc_submission.c | 54 ++++++++++++------------------
 drivers/gpu/drm/i915/intel_guc.h           |  2 +-
 2 files changed, 23 insertions(+), 33 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 35231fd..5d35570 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -238,9 +238,6 @@ static int guc_ring_doorbell(struct i915_guc_client *gc)
 			db_exc.cookie = 1;
 	}
 
-	/* Finally, update the cached copy of the GuC's WQ head */
-	gc->wq_head = desc->head;
-
 	return ret;
 }
 
@@ -361,12 +358,14 @@ static void guc_init_proc_desc(struct intel_guc *guc,
 static void guc_init_ctx_desc(struct intel_guc *guc,
 			      struct i915_guc_client *client)
 {
+	struct drm_i915_gem_object *client_obj = client->client_obj;
 	struct drm_i915_private *dev_priv = guc_to_i915(guc);
 	struct intel_engine_cs *engine;
 	struct intel_context *ctx = client->owner;
 	struct guc_context_desc desc;
 	struct sg_table *sg;
 	enum intel_engine_id id;
+	u32 gfx_addr;
 
 	memset(&desc, 0, sizeof(desc));
 
@@ -395,16 +394,17 @@ static void guc_init_ctx_desc(struct intel_guc *guc,
 		lrc->context_desc = (u32)ctx_desc;
 
 		/* The state page is after PPHWSP */
-		lrc->ring_lcra = i915_gem_obj_ggtt_offset(obj) +
-				LRC_STATE_PN * PAGE_SIZE;
+		gfx_addr = i915_gem_obj_ggtt_offset(obj);
+		lrc->ring_lcra = gfx_addr + LRC_STATE_PN * PAGE_SIZE;
 		lrc->context_id = (client->ctx_index << GUC_ELC_CTXID_OFFSET) |
 				(engine->guc_id << GUC_ELC_ENGINE_OFFSET);
 
 		obj = ctx->engine[id].ringbuf->obj;
+		gfx_addr = i915_gem_obj_ggtt_offset(obj);
 
-		lrc->ring_begin = i915_gem_obj_ggtt_offset(obj);
-		lrc->ring_end = lrc->ring_begin + obj->base.size - 1;
-		lrc->ring_next_free_location = lrc->ring_begin;
+		lrc->ring_begin = gfx_addr;
+		lrc->ring_end = gfx_addr + obj->base.size - 1;
+		lrc->ring_next_free_location = gfx_addr;
 		lrc->ring_current_tail_pointer_value = 0;
 
 		desc.engines_used |= (1 << engine->guc_id);
@@ -413,22 +413,17 @@ static void guc_init_ctx_desc(struct intel_guc *guc,
 	WARN_ON(desc.engines_used == 0);
 
 	/*
-	 * The CPU address is only needed at certain points, so kmap_atomic on
-	 * demand instead of storing it in the ctx descriptor.
-	 * XXX: May make debug easier to have it mapped
+	 * The doorbell, process descriptor, and workqueue are all parts
+	 * of the client object, which the GuC will reference via the GGTT
 	 */
-	desc.db_trigger_cpu = 0;
-	desc.db_trigger_uk = client->doorbell_offset +
-		i915_gem_obj_ggtt_offset(client->client_obj);
-	desc.db_trigger_phy = client->doorbell_offset +
-		sg_dma_address(client->client_obj->pages->sgl);
-
-	desc.process_desc = client->proc_desc_offset +
-		i915_gem_obj_ggtt_offset(client->client_obj);
-
-	desc.wq_addr = client->wq_offset +
-		i915_gem_obj_ggtt_offset(client->client_obj);
-
+	gfx_addr = i915_gem_obj_ggtt_offset(client_obj);
+	desc.db_trigger_phy = sg_dma_address(client_obj->pages->sgl) +
+				client->doorbell_offset;
+	desc.db_trigger_cpu = (uintptr_t)client->client_base +
+				client->doorbell_offset;
+	desc.db_trigger_uk = gfx_addr + client->doorbell_offset;
+	desc.process_desc = gfx_addr + client->proc_desc_offset;
+	desc.wq_addr = gfx_addr + client->wq_offset;
 	desc.wq_size = client->wq_size;
 
 	/*
@@ -465,17 +460,10 @@ int i915_guc_wq_check_space(struct i915_guc_client *gc)
 	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;
-
 	desc = gc->client_base + gc->proc_desc_offset;
 
 	while (timeout_counter-- > 0) {
-		gc->wq_head = desc->head;
-
-		if (CIRC_SPACE(gc->wq_tail, gc->wq_head, gc->wq_size) >= size) {
+		if (CIRC_SPACE(gc->wq_tail, desc->head, gc->wq_size) >= size) {
 			ret = 0;
 			break;
 		}
@@ -490,10 +478,12 @@ int i915_guc_wq_check_space(struct i915_guc_client *gc)
 static int guc_add_workqueue_item(struct i915_guc_client *gc,
 				  struct drm_i915_gem_request *rq)
 {
+	struct guc_process_desc *desc;
 	struct guc_wq_item *wqi;
 	u32 tail, wq_len, wq_off, space;
 
-	space = CIRC_SPACE(gc->wq_tail, gc->wq_head, gc->wq_size);
+	desc = gc->client_base + gc->proc_desc_offset;
+	space = CIRC_SPACE(gc->wq_tail, desc->head, gc->wq_size);
 	if (WARN_ON(space < sizeof(struct guc_wq_item)))
 		return -ENOSPC; /* shouldn't happen */
 
diff --git a/drivers/gpu/drm/i915/intel_guc.h b/drivers/gpu/drm/i915/intel_guc.h
index f2c051e..cc32a18 100644
--- a/drivers/gpu/drm/i915/intel_guc.h
+++ b/drivers/gpu/drm/i915/intel_guc.h
@@ -68,7 +68,7 @@ struct i915_guc_client {
 	uint32_t wq_offset;
 	uint32_t wq_size;
 	uint32_t wq_tail;
-	uint32_t wq_head;
+	uint32_t unused;		/* Was 'wq_head'		*/
 
 	/* GuC submission statistics & status */
 	uint64_t submissions[GUC_MAX_ENGINES_NUM];
-- 
1.9.1