[PATCH 1/5] drm/i915/guc: Remove obsolete comments and remove unused variable

[PATCH 1/5] drm/i915/guc: Remove obsolete comments and remove unused variable

[Michał Winiarski]

Originally removed in:
c1adab970348 ("drm/i915/guc: Remove failed doorbell stat from debugfs")
f1448a62a103 ("drm/i915/guc: Remove last submission result from debugfs")

Were accidentally restored in:
925344ccc91d ("BackMerge tag 'v4.12-rc5' into drm-next")

We can also remove unused variable and replace it with a WARN.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_guc_submission.c | 3 +--
 drivers/gpu/drm/i915/intel_uc.h            | 4 ----
 2 files changed, 1 insertion(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 48a1e9349a2c..8a550785b257 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -602,7 +602,6 @@ static void __i915_guc_submit(struct drm_i915_gem_request *rq)
 	struct intel_guc *guc = &rq->i915->guc;
 	struct i915_guc_client *client = guc->execbuf_client;
 	unsigned long flags;
-	int b_ret;
 
 	/* WA to flush out the pending GMADR writes to ring buffer. */
 	if (i915_vma_is_map_and_fenceable(rq->ring->vma))
@@ -611,7 +610,7 @@ static void __i915_guc_submit(struct drm_i915_gem_request *rq)
 	spin_lock_irqsave(&client->wq_lock, flags);
 
 	guc_wq_item_append(client, rq);
-	b_ret = guc_ring_doorbell(client);
+	WARN_ON(guc_ring_doorbell(client));
 
 	client->submissions[engine_id] += 1;
 
diff --git a/drivers/gpu/drm/i915/intel_uc.h b/drivers/gpu/drm/i915/intel_uc.h
index 22ae52b17b0f..69daf4c01cd0 100644
--- a/drivers/gpu/drm/i915/intel_uc.h
+++ b/drivers/gpu/drm/i915/intel_uc.h
@@ -59,10 +59,6 @@ struct drm_i915_gem_request;
  *                available in the work queue (note, the queue is shared,
  *                not per-engine). It is OK for this to be nonzero, but
  *                it should not be huge!
- *   b_fail: failed to ring the doorbell. This should never happen, unless
- *           somehow the hardware misbehaves, or maybe if the GuC firmware
- *           crashes? We probably need to reset the GPU to recover.
- *   retcode: errno from last guc_submit()
  */
 struct i915_guc_client {
 	struct i915_vma *vma;
-- 
2.13.5

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

[PATCH 2/5] drm/i915/guc: Submit GuC workitems containing coalesced requests

[Michał Winiarski]

To create an upper bound on number of GuC workitems, we need to change
the way that requests are being submitted. Rather than submitting each
request as an individual workitem, we can do coalescing in a similar way
we're handlig execlist submission ports. We also need to stop pretending
that we're doing "lite-restore" in GuC submission (we would create a
workitem each time we hit this condition). This allows us to completely
remove the reservation, replacing it with a compile time check.

v2: Also coalesce when replaying on reset (Daniele)
v3: Consistent wq_resv - per-request (Daniele)
v4: Squash removing wq_resv

References: https://bugs.freedesktop.org/show_bug.cgi?id=101873
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Jeff McGee <jeff.mcgee@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Oscar Mateo <oscar.mateo@intel.com>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |   2 -
 drivers/gpu/drm/i915/i915_guc_submission.c | 179 ++++++++++-------------------
 drivers/gpu/drm/i915/intel_lrc.c           |  25 +---
 drivers/gpu/drm/i915/intel_uc.h            |  11 --
 4 files changed, 62 insertions(+), 155 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 6338018f655d..f5fd00cfb3b0 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -2450,8 +2450,6 @@ static void i915_guc_client_info(struct seq_file *m,
 	seq_printf(m, "\tWQ size %d, offset: 0x%x, tail %d\n",
 		client->wq_size, client->wq_offset, client->wq_tail);
 
-	seq_printf(m, "\tWork queue full: %u\n", client->no_wq_space);
-
 	for_each_engine(engine, dev_priv, id) {
 		u64 submissions = client->submissions[id];
 		tot += submissions;
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 8a550785b257..6f0adcd2a058 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -406,63 +406,6 @@ static void guc_stage_desc_fini(struct intel_guc *guc,
 	memset(desc, 0, sizeof(*desc));
 }
 
-/**
- * i915_guc_wq_reserve() - reserve space in the GuC's workqueue
- * @request:	request associated with the commands
- *
- * Return:	0 if space is available
- *		-EAGAIN if space is not currently available
- *
- * This function must be called (and must return 0) before a request
- * is submitted to the GuC via i915_guc_submit() below. Once a result
- * of 0 has been returned, it must be balanced by a corresponding
- * call to submit().
- *
- * Reservation allows the caller to determine in advance that space
- * will be available for the next submission before committing resources
- * to it, and helps avoid late failures with complicated recovery paths.
- */
-int i915_guc_wq_reserve(struct drm_i915_gem_request *request)
-{
-	const size_t wqi_size = sizeof(struct guc_wq_item);
-	struct i915_guc_client *client = request->i915->guc.execbuf_client;
-	struct guc_process_desc *desc = __get_process_desc(client);
-	u32 freespace;
-	int ret;
-
-	spin_lock_irq(&client->wq_lock);
-	freespace = CIRC_SPACE(client->wq_tail, desc->head, client->wq_size);
-	freespace -= client->wq_rsvd;
-	if (likely(freespace >= wqi_size)) {
-		client->wq_rsvd += wqi_size;
-		ret = 0;
-	} else {
-		client->no_wq_space++;
-		ret = -EAGAIN;
-	}
-	spin_unlock_irq(&client->wq_lock);
-
-	return ret;
-}
-
-static void guc_client_update_wq_rsvd(struct i915_guc_client *client, int size)
-{
-	unsigned long flags;
-
-	spin_lock_irqsave(&client->wq_lock, flags);
-	client->wq_rsvd += size;
-	spin_unlock_irqrestore(&client->wq_lock, flags);
-}
-
-void i915_guc_wq_unreserve(struct drm_i915_gem_request *request)
-{
-	const int wqi_size = sizeof(struct guc_wq_item);
-	struct i915_guc_client *client = request->i915->guc.execbuf_client;
-
-	GEM_BUG_ON(READ_ONCE(client->wq_rsvd) < wqi_size);
-	guc_client_update_wq_rsvd(client, -wqi_size);
-}
-
 /* Construct a Work Item and append it to the GuC's Work Queue */
 static void guc_wq_item_append(struct i915_guc_client *client,
 			       struct drm_i915_gem_request *rq)
@@ -475,7 +418,7 @@ static void guc_wq_item_append(struct i915_guc_client *client,
 	struct guc_wq_item *wqi;
 	u32 freespace, tail, wq_off;
 
-	/* Free space is guaranteed, see i915_guc_wq_reserve() above */
+	/* Free space is guaranteed */
 	freespace = CIRC_SPACE(client->wq_tail, desc->head, client->wq_size);
 	GEM_BUG_ON(freespace < wqi_size);
 
@@ -491,14 +434,12 @@ static void guc_wq_item_append(struct i915_guc_client *client,
 	 * workqueue buffer dw by dw.
 	 */
 	BUILD_BUG_ON(wqi_size != 16);
-	GEM_BUG_ON(client->wq_rsvd < wqi_size);
 
 	/* postincrement WQ tail for next time */
 	wq_off = client->wq_tail;
 	GEM_BUG_ON(wq_off & (wqi_size - 1));
 	client->wq_tail += wqi_size;
 	client->wq_tail &= client->wq_size - 1;
-	client->wq_rsvd -= wqi_size;
 
 	/* WQ starts from the page after doorbell / process_desc */
 	wqi = client->vaddr + wq_off + GUC_DB_SIZE;
@@ -580,47 +521,43 @@ static int guc_ring_doorbell(struct i915_guc_client *client)
 }
 
 /**
- * __i915_guc_submit() - Submit commands through GuC
+ * i915_guc_submit() - Submit commands through GuC
  * @rq:		request associated with the commands
  *
- * The caller must have already called i915_guc_wq_reserve() above with
- * a result of 0 (success), guaranteeing that there is space in the work
- * queue for the new request, so enqueuing the item cannot fail.
- *
- * Bad Things Will Happen if the caller violates this protocol e.g. calls
- * submit() when _reserve() says there's no space, or calls _submit()
- * a different number of times from (successful) calls to _reserve().
- *
  * The only error here arises if the doorbell hardware isn't functioning
  * as expected, which really shouln't happen.
  */
-static void __i915_guc_submit(struct drm_i915_gem_request *rq)
+static void i915_guc_submit(struct intel_engine_cs *engine)
 {
-	struct drm_i915_private *dev_priv = rq->i915;
-	struct intel_engine_cs *engine = rq->engine;
-	unsigned int engine_id = engine->id;
-	struct intel_guc *guc = &rq->i915->guc;
+	struct drm_i915_private *dev_priv = engine->i915;
+	struct intel_guc *guc = &dev_priv->guc;
 	struct i915_guc_client *client = guc->execbuf_client;
+	struct execlist_port *port = engine->execlist_port;
+	unsigned int engine_id = engine->id;
+	unsigned int n;
 	unsigned long flags;
 
-	/* WA to flush out the pending GMADR writes to ring buffer. */
-	if (i915_vma_is_map_and_fenceable(rq->ring->vma))
-		POSTING_READ_FW(GUC_STATUS);
+	for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++) {
+		struct drm_i915_gem_request *rq;
+		unsigned int count;
 
-	spin_lock_irqsave(&client->wq_lock, flags);
+		rq = port_unpack(&port[n], &count);
+		if (rq && count == 0) {
+			port_set(&port[n], port_pack(rq, ++count));
 
-	guc_wq_item_append(client, rq);
-	WARN_ON(guc_ring_doorbell(client));
+			if (i915_vma_is_map_and_fenceable(rq->ring->vma))
+				POSTING_READ_FW(GUC_STATUS);
 
-	client->submissions[engine_id] += 1;
+			spin_lock_irqsave(&client->wq_lock, flags);
 
-	spin_unlock_irqrestore(&client->wq_lock, flags);
-}
+			guc_wq_item_append(client, rq);
+			WARN_ON(guc_ring_doorbell(client));
 
-static void i915_guc_submit(struct drm_i915_gem_request *rq)
-{
-	__i915_gem_request_submit(rq);
-	__i915_guc_submit(rq);
+			client->submissions[engine_id] += 1;
+
+			spin_unlock_irqrestore(&client->wq_lock, flags);
+		}
+	}
 }
 
 static void nested_enable_signaling(struct drm_i915_gem_request *rq)
@@ -654,16 +591,19 @@ static void port_assign(struct execlist_port *port,
 	if (port_isset(port))
 		i915_gem_request_put(port_request(port));
 
-	port_set(port, i915_gem_request_get(rq));
+	port_set(port, port_pack(i915_gem_request_get(rq), port_count(port)));
 	nested_enable_signaling(rq);
 }
 
-static bool i915_guc_dequeue(struct intel_engine_cs *engine)
+static void i915_guc_dequeue(struct intel_engine_cs *engine)
 {
 	struct execlist_port *port = engine->execlist_port;
-	struct drm_i915_gem_request *last = port_request(port);
-	struct rb_node *rb;
+	struct drm_i915_gem_request *last = NULL;
 	bool submit = false;
+	struct rb_node *rb;
+
+	if (port_isset(port))
+		port++;
 
 	spin_lock_irq(&engine->timeline->lock);
 	rb = engine->execlist_first;
@@ -688,7 +628,7 @@ static bool i915_guc_dequeue(struct intel_engine_cs *engine)
 			INIT_LIST_HEAD(&rq->priotree.link);
 			rq->priotree.priority = INT_MAX;
 
-			i915_guc_submit(rq);
+			__i915_gem_request_submit(rq);
 			trace_i915_gem_request_in(rq, port_index(port, engine));
 			last = rq;
 			submit = true;
@@ -702,11 +642,11 @@ static bool i915_guc_dequeue(struct intel_engine_cs *engine)
 	}
 done:
 	engine->execlist_first = rb;
-	if (submit)
+	if (submit) {
 		port_assign(port, last);
+		i915_guc_submit(engine);
+	}
 	spin_unlock_irq(&engine->timeline->lock);
-
-	return submit;
 }
 
 static void i915_guc_irq_handler(unsigned long data)
@@ -714,24 +654,20 @@ static void i915_guc_irq_handler(unsigned long data)
 	struct intel_engine_cs *engine = (struct intel_engine_cs *)data;
 	struct execlist_port *port = engine->execlist_port;
 	struct drm_i915_gem_request *rq;
-	bool submit;
 
-	do {
-		rq = port_request(&port[0]);
-		while (rq && i915_gem_request_completed(rq)) {
-			trace_i915_gem_request_out(rq);
-			i915_gem_request_put(rq);
+	rq = port_request(&port[0]);
+	while (rq && i915_gem_request_completed(rq)) {
+		trace_i915_gem_request_out(rq);
+		i915_gem_request_put(rq);
 
-			port[0] = port[1];
-			memset(&port[1], 0, sizeof(port[1]));
+		port[0] = port[1];
+		memset(&port[1], 0, sizeof(port[1]));
 
-			rq = port_request(&port[0]);
-		}
+		rq = port_request(&port[0]);
+	}
 
-		submit = false;
-		if (!port_count(&port[1]))
-			submit = i915_guc_dequeue(engine);
-	} while (submit);
+	if (!port_isset(&port[1]))
+		i915_guc_dequeue(engine);
 }
 
 /*
@@ -1220,6 +1156,19 @@ int i915_guc_submission_enable(struct drm_i915_private *dev_priv)
 	enum intel_engine_id id;
 	int err;
 
+	/*
+	 * We're using GuC work items for submitting work through GuC. Since
+	 * we're coalescing multiple requests from a single context into a
+	 * single work item prior to assigning it to execlist_port, we can
+	 * never have more work items than the total number of ports (for all
+	 * engines). The GuC firmware is controlling the HEAD of work queue,
+	 * and it is guaranteed that it will remove the work item from the
+	 * queue before our request is completed.
+	 */
+	BUILD_BUG_ON(ARRAY_SIZE(engine->execlist_port) *
+		     sizeof(struct guc_wq_item) *
+		     I915_NUM_ENGINES > GUC_WQ_SIZE);
+
 	if (!client) {
 		client = guc_client_alloc(dev_priv,
 					  INTEL_INFO(dev_priv)->ring_mask,
@@ -1247,9 +1196,6 @@ int i915_guc_submission_enable(struct drm_i915_private *dev_priv)
 	guc_interrupts_capture(dev_priv);
 
 	for_each_engine(engine, dev_priv, id) {
-		const int wqi_size = sizeof(struct guc_wq_item);
-		struct drm_i915_gem_request *rq;
-
 		/* The tasklet was initialised by execlists, and may be in
 		 * a state of flux (across a reset) and so we just want to
 		 * take over the callback without changing any other state
@@ -1257,14 +1203,7 @@ int i915_guc_submission_enable(struct drm_i915_private *dev_priv)
 		 */
 		engine->irq_tasklet.func = i915_guc_irq_handler;
 		clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted);
-
-		/* Replay the current set of previously submitted requests */
-		spin_lock_irq(&engine->timeline->lock);
-		list_for_each_entry(rq, &engine->timeline->requests, link) {
-			guc_client_update_wq_rsvd(client, wqi_size);
-			__i915_guc_submit(rq);
-		}
-		spin_unlock_irq(&engine->timeline->lock);
+		i915_guc_submit(engine);
 	}
 
 	return 0;
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index d89e1b8e1cc5..5837b33f9705 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -914,27 +914,14 @@ static int execlists_request_alloc(struct drm_i915_gem_request *request)
 	 */
 	request->reserved_space += EXECLISTS_REQUEST_SIZE;
 
-	if (i915.enable_guc_submission) {
-		/*
-		 * Check that the GuC has space for the request before
-		 * going any further, as the i915_add_request() call
-		 * later on mustn't fail ...
-		 */
-		ret = i915_guc_wq_reserve(request);
-		if (ret)
-			goto err;
-	}
-
 	cs = intel_ring_begin(request, 0);
-	if (IS_ERR(cs)) {
-		ret = PTR_ERR(cs);
-		goto err_unreserve;
-	}
+	if (IS_ERR(cs))
+		return PTR_ERR(cs);
 
 	if (!ce->initialised) {
 		ret = engine->init_context(request);
 		if (ret)
-			goto err_unreserve;
+			return ret;
 
 		ce->initialised = true;
 	}
@@ -948,12 +935,6 @@ static int execlists_request_alloc(struct drm_i915_gem_request *request)
 
 	request->reserved_space -= EXECLISTS_REQUEST_SIZE;
 	return 0;
-
-err_unreserve:
-	if (i915.enable_guc_submission)
-		i915_guc_wq_unreserve(request);
-err:
-	return ret;
 }
 
 /*
diff --git a/drivers/gpu/drm/i915/intel_uc.h b/drivers/gpu/drm/i915/intel_uc.h
index 69daf4c01cd0..d41051688221 100644
--- a/drivers/gpu/drm/i915/intel_uc.h
+++ b/drivers/gpu/drm/i915/intel_uc.h
@@ -52,13 +52,6 @@ struct drm_i915_gem_request;
  * GuC). The subsequent  pages of the client object constitute the work
  * queue (a circular array of work items), again described in the process
  * descriptor. Work queue pages are mapped momentarily as required.
- *
- * We also keep a few statistics on failures. Ideally, these should all
- * be zero!
- *   no_wq_space: times that the submission pre-check found no space was
- *                available in the work queue (note, the queue is shared,
- *                not per-engine). It is OK for this to be nonzero, but
- *                it should not be huge!
  */
 struct i915_guc_client {
 	struct i915_vma *vma;
@@ -79,8 +72,6 @@ struct i915_guc_client {
 	uint32_t wq_offset;
 	uint32_t wq_size;
 	uint32_t wq_tail;
-	uint32_t wq_rsvd;
-	uint32_t no_wq_space;
 
 	/* Per-engine counts of GuC submissions */
 	uint64_t submissions[I915_NUM_ENGINES];
@@ -246,8 +237,6 @@ u32 intel_guc_wopcm_size(struct drm_i915_private *dev_priv);
 /* i915_guc_submission.c */
 int i915_guc_submission_init(struct drm_i915_private *dev_priv);
 int i915_guc_submission_enable(struct drm_i915_private *dev_priv);
-int i915_guc_wq_reserve(struct drm_i915_gem_request *rq);
-void i915_guc_wq_unreserve(struct drm_i915_gem_request *request);
 void i915_guc_submission_disable(struct drm_i915_private *dev_priv);
 void i915_guc_submission_fini(struct drm_i915_private *dev_priv);
 struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size);
-- 
2.13.5

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

[PATCH 3/5] drm/i915/guc: Simplify GuC doorbell logic

[Michał Winiarski]

All we're really doing is incrementing a simple counter in a
doorbell_info struct. We can do without extra variables and a separate
counter kept in guc_client. Since it's gone, we're also removing its
debugfs.
The only functional change here, is that we're no longer treating 0 as a
special value. GuC doesn't seem to care, why should we?

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Oscar Mateo <oscar.mateo@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |  4 +-
 drivers/gpu/drm/i915/i915_guc_submission.c | 64 ++++++------------------------
 drivers/gpu/drm/i915/intel_uc.h            |  1 -
 3 files changed, 15 insertions(+), 54 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index f5fd00cfb3b0..40a315fc0e31 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -2445,8 +2445,8 @@ static void i915_guc_client_info(struct seq_file *m,
 
 	seq_printf(m, "\tPriority %d, GuC stage index: %u, PD offset 0x%x\n",
 		client->priority, client->stage_id, client->proc_desc_offset);
-	seq_printf(m, "\tDoorbell id %d, offset: 0x%lx, cookie 0x%x\n",
-		client->doorbell_id, client->doorbell_offset, client->doorbell_cookie);
+	seq_printf(m, "\tDoorbell id %d, offset: 0x%lx\n",
+		client->doorbell_id, client->doorbell_offset);
 	seq_printf(m, "\tWQ size %d, offset: 0x%x, tail %d\n",
 		client->wq_size, client->wq_offset, client->wq_tail);
 
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 6f0adcd2a058..674f6a4035e6 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -192,13 +192,12 @@ static int __create_doorbell(struct i915_guc_client *client)
 
 	doorbell = __get_doorbell(client);
 	doorbell->db_status = GUC_DOORBELL_ENABLED;
-	doorbell->cookie = client->doorbell_cookie;
+	doorbell->cookie = 0;
 
 	err = __guc_allocate_doorbell(client->guc, client->stage_id);
-	if (err) {
+	if (err)
 		doorbell->db_status = GUC_DOORBELL_DISABLED;
-		doorbell->cookie = 0;
-	}
+
 	return err;
 }
 
@@ -467,57 +466,20 @@ static void guc_reset_wq(struct i915_guc_client *client)
 	client->wq_tail = 0;
 }
 
-static int guc_ring_doorbell(struct i915_guc_client *client)
+static void guc_ring_doorbell(struct i915_guc_client *client)
 {
-	struct guc_process_desc *desc = __get_process_desc(client);
-	union guc_doorbell_qw db_cmp, db_exc, db_ret;
-	union guc_doorbell_qw *db;
-	int attempt = 2, ret = -EAGAIN;
-
-	/* Update the tail so it is visible to GuC */
-	desc->tail = client->wq_tail;
-
-	/* current cookie */
-	db_cmp.db_status = GUC_DOORBELL_ENABLED;
-	db_cmp.cookie = client->doorbell_cookie;
-
-	/* cookie to be updated */
-	db_exc.db_status = GUC_DOORBELL_ENABLED;
-	db_exc.cookie = client->doorbell_cookie + 1;
-	if (db_exc.cookie == 0)
-		db_exc.cookie = 1;
+	struct guc_doorbell_info *db;
+	u32 cookie;
 
 	/* pointer of current doorbell cacheline */
-	db = (union guc_doorbell_qw *)__get_doorbell(client);
-
-	while (attempt--) {
-		/* lets ring the doorbell */
-		db_ret.value_qw = atomic64_cmpxchg((atomic64_t *)db,
-			db_cmp.value_qw, db_exc.value_qw);
-
-		/* if the exchange was successfully executed */
-		if (db_ret.value_qw == db_cmp.value_qw) {
-			/* db was successfully rung */
-			client->doorbell_cookie = db_exc.cookie;
-			ret = 0;
-			break;
-		}
-
-		/* XXX: doorbell was lost and need to acquire it again */
-		if (db_ret.db_status == GUC_DOORBELL_DISABLED)
-			break;
+	db = __get_doorbell(client);
 
-		DRM_WARN("Cookie mismatch. Expected %d, found %d\n",
-			 db_cmp.cookie, db_ret.cookie);
+	do {
+		cookie = READ_ONCE(db->cookie);
+	} while (cmpxchg(&db->cookie, cookie, cookie + 1) != cookie);
 
-		/* update the cookie to newly read cookie from GuC */
-		db_cmp.cookie = db_ret.cookie;
-		db_exc.cookie = db_ret.cookie + 1;
-		if (db_exc.cookie == 0)
-			db_exc.cookie = 1;
-	}
-
-	return ret;
+	/* XXX: doorbell was lost and need to acquire it again */
+	GEM_BUG_ON(db->db_status != GUC_DOORBELL_ENABLED);
 }
 
 /**
@@ -551,7 +513,7 @@ static void i915_guc_submit(struct intel_engine_cs *engine)
 			spin_lock_irqsave(&client->wq_lock, flags);
 
 			guc_wq_item_append(client, rq);
-			WARN_ON(guc_ring_doorbell(client));
+			guc_ring_doorbell(client);
 
 			client->submissions[engine_id] += 1;
 
diff --git a/drivers/gpu/drm/i915/intel_uc.h b/drivers/gpu/drm/i915/intel_uc.h
index d41051688221..f505dcccd613 100644
--- a/drivers/gpu/drm/i915/intel_uc.h
+++ b/drivers/gpu/drm/i915/intel_uc.h
@@ -66,7 +66,6 @@ struct i915_guc_client {
 
 	u16 doorbell_id;
 	unsigned long doorbell_offset;
-	u32 doorbell_cookie;
 
 	spinlock_t wq_lock;
 	uint32_t wq_offset;
-- 
2.13.5

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

Re: [PATCH 3/5] drm/i915/guc: Simplify GuC doorbell logic

[Chris Wilson]

Quoting Michał Winiarski (2017-09-12 18:40:54)
> -static int guc_ring_doorbell(struct i915_guc_client *client)
> +static void guc_ring_doorbell(struct i915_guc_client *client)
>  {
> -       struct guc_process_desc *desc = __get_process_desc(client);
> -       union guc_doorbell_qw db_cmp, db_exc, db_ret;
> -       union guc_doorbell_qw *db;
> -       int attempt = 2, ret = -EAGAIN;
> -
> -       /* Update the tail so it is visible to GuC */
> -       desc->tail = client->wq_tail;

At this point in the series, guc_ring_doorbell() is still responsible
for the desc->tail update.

Hmm. Twice now that kbl has behaved unexpectedly. Do you have one to
hand?
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v2 4/5] drm/i915/guc: Make adding GuC work items lockless

[Michał Winiarski]

We can get rid of a spinlock by updating the tail directly using
cmpxchg. We can also put guc client on a diet by removing some constants
from the struct.
This causes a small change in one of GuC debugfs files.
We're no longer reporting constant values (which I don't think is a
problem), but we're also no longer reporting the tail (does anyone care?).

v2: Free space calculation inside the loop, compiler barriers. (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Oscar Mateo <oscar.mateo@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |  2 --
 drivers/gpu/drm/i915/i915_guc_submission.c | 41 ++++++++++++------------------
 drivers/gpu/drm/i915/intel_uc.h            |  5 ----
 3 files changed, 16 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 40a315fc0e31..0bd5cacd288b 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -2447,8 +2447,6 @@ static void i915_guc_client_info(struct seq_file *m,
 		client->priority, client->stage_id, client->proc_desc_offset);
 	seq_printf(m, "\tDoorbell id %d, offset: 0x%lx\n",
 		client->doorbell_id, client->doorbell_offset);
-	seq_printf(m, "\tWQ size %d, offset: 0x%x, tail %d\n",
-		client->wq_size, client->wq_offset, client->wq_tail);
 
 	for_each_engine(engine, dev_priv, id) {
 		u64 submissions = client->submissions[id];
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 674f6a4035e6..16d6bb13c2bf 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -305,7 +305,7 @@ static void guc_proc_desc_init(struct intel_guc *guc,
 	desc->db_base_addr = 0;
 
 	desc->stage_id = client->stage_id;
-	desc->wq_size_bytes = client->wq_size;
+	desc->wq_size_bytes = GUC_WQ_SIZE;
 	desc->wq_status = WQ_STATUS_ACTIVE;
 	desc->priority = client->priority;
 }
@@ -390,8 +390,8 @@ static void guc_stage_desc_init(struct intel_guc *guc,
 	desc->db_trigger_cpu = (uintptr_t)__get_doorbell(client);
 	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;
+	desc->wq_addr = gfx_addr + GUC_DB_SIZE;
+	desc->wq_size = GUC_WQ_SIZE;
 
 	desc->desc_private = (uintptr_t)client;
 }
@@ -415,15 +415,11 @@ static void guc_wq_item_append(struct i915_guc_client *client,
 	struct intel_engine_cs *engine = rq->engine;
 	struct guc_process_desc *desc = __get_process_desc(client);
 	struct guc_wq_item *wqi;
-	u32 freespace, tail, wq_off;
-
-	/* Free space is guaranteed */
-	freespace = CIRC_SPACE(client->wq_tail, desc->head, client->wq_size);
-	GEM_BUG_ON(freespace < wqi_size);
+	u32 ring_tail, wq_off, wq_next;
 
 	/* The GuC firmware wants the tail index in QWords, not bytes */
-	tail = intel_ring_set_tail(rq->ring, rq->tail) >> 3;
-	GEM_BUG_ON(tail > WQ_RING_TAIL_MAX);
+	ring_tail = intel_ring_set_tail(rq->ring, rq->tail) >> 3;
+	GEM_BUG_ON(ring_tail > WQ_RING_TAIL_MAX);
 
 	/* 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
@@ -434,11 +430,16 @@ static void guc_wq_item_append(struct i915_guc_client *client,
 	 */
 	BUILD_BUG_ON(wqi_size != 16);
 
-	/* postincrement WQ tail for next time */
-	wq_off = client->wq_tail;
+	/* Find our offset and postincrement WQ tail for next time, free space
+	 * is guaranteed.
+	 */
+	do {
+		wq_off = READ_ONCE(desc->tail);
+		wq_next = (wq_off + wqi_size) & (GUC_WQ_SIZE - 1);
+		GEM_BUG_ON(CIRC_SPACE(wq_off, READ_ONCE(desc->head),
+				      GUC_WQ_SIZE) < wqi_size);
+	} while (cmpxchg(&desc->tail, wq_off, wq_next) != wq_off);
 	GEM_BUG_ON(wq_off & (wqi_size - 1));
-	client->wq_tail += wqi_size;
-	client->wq_tail &= client->wq_size - 1;
 
 	/* WQ starts from the page after doorbell / process_desc */
 	wqi = client->vaddr + wq_off + GUC_DB_SIZE;
@@ -452,7 +453,7 @@ static void guc_wq_item_append(struct i915_guc_client *client,
 	/* The GuC wants only the low-order word of the context descriptor */
 	wqi->context_desc = (u32)intel_lr_context_descriptor(rq->ctx, engine);
 
-	wqi->submit_element_info = tail << WQ_RING_TAIL_SHIFT;
+	wqi->submit_element_info = ring_tail << WQ_RING_TAIL_SHIFT;
 	wqi->fence_id = rq->global_seqno;
 }
 
@@ -462,8 +463,6 @@ static void guc_reset_wq(struct i915_guc_client *client)
 
 	desc->head = 0;
 	desc->tail = 0;
-
-	client->wq_tail = 0;
 }
 
 static void guc_ring_doorbell(struct i915_guc_client *client)
@@ -497,7 +496,6 @@ static void i915_guc_submit(struct intel_engine_cs *engine)
 	struct execlist_port *port = engine->execlist_port;
 	unsigned int engine_id = engine->id;
 	unsigned int n;
-	unsigned long flags;
 
 	for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++) {
 		struct drm_i915_gem_request *rq;
@@ -510,14 +508,10 @@ static void i915_guc_submit(struct intel_engine_cs *engine)
 			if (i915_vma_is_map_and_fenceable(rq->ring->vma))
 				POSTING_READ_FW(GUC_STATUS);
 
-			spin_lock_irqsave(&client->wq_lock, flags);
-
 			guc_wq_item_append(client, rq);
 			guc_ring_doorbell(client);
 
 			client->submissions[engine_id] += 1;
-
-			spin_unlock_irqrestore(&client->wq_lock, flags);
 		}
 	}
 }
@@ -810,9 +804,6 @@ guc_client_alloc(struct drm_i915_private *dev_priv,
 	client->engines = engines;
 	client->priority = priority;
 	client->doorbell_id = GUC_DOORBELL_INVALID;
-	client->wq_offset = GUC_DB_SIZE;
-	client->wq_size = GUC_WQ_SIZE;
-	spin_lock_init(&client->wq_lock);
 
 	ret = ida_simple_get(&guc->stage_ids, 0, GUC_MAX_STAGE_DESCRIPTORS,
 				GFP_KERNEL);
diff --git a/drivers/gpu/drm/i915/intel_uc.h b/drivers/gpu/drm/i915/intel_uc.h
index f505dcccd613..386e0bcd0cc3 100644
--- a/drivers/gpu/drm/i915/intel_uc.h
+++ b/drivers/gpu/drm/i915/intel_uc.h
@@ -67,11 +67,6 @@ struct i915_guc_client {
 	u16 doorbell_id;
 	unsigned long doorbell_offset;
 
-	spinlock_t wq_lock;
-	uint32_t wq_offset;
-	uint32_t wq_size;
-	uint32_t wq_tail;
-
 	/* Per-engine counts of GuC submissions */
 	uint64_t submissions[I915_NUM_ENGINES];
 };
-- 
2.13.5

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

Re: [PATCH v2 4/5] drm/i915/guc: Make adding GuC work items lockless

[Chris Wilson]

Quoting Michał Winiarski (2017-09-12 18:40:55)
> @@ -434,11 +430,16 @@ static void guc_wq_item_append(struct i915_guc_client *client,
>          */
>         BUILD_BUG_ON(wqi_size != 16);
>  
> -       /* postincrement WQ tail for next time */
> -       wq_off = client->wq_tail;
> +       /* Find our offset and postincrement WQ tail for next time, free space
> +        * is guaranteed.
> +        */
> +       do {
> +               wq_off = READ_ONCE(desc->tail);
> +               wq_next = (wq_off + wqi_size) & (GUC_WQ_SIZE - 1);
> +               GEM_BUG_ON(CIRC_SPACE(wq_off, READ_ONCE(desc->head),
> +                                     GUC_WQ_SIZE) < wqi_size);
> +       } while (cmpxchg(&desc->tail, wq_off, wq_next) != wq_off);
>         GEM_BUG_ON(wq_off & (wqi_size - 1));

I forgot the trick in cmpxchg to avoid the READ_ONCE in every loop:
	wq_next = READ_ONCE(desc->tail);
	do {
		wq_off = wq_next;
		wq_next = (wq_off + wqi_size) & (GUC_WQ_SIZE - 1);
		GEM_BUG_ON(...);
	} while ((wq_next = cmpxchg(&desc->tail, wq_off, wq_next) != wq_off);
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 5/5] HAX Enable GuC Submission for CI

[Michał Winiarski]

Also:
Revert "drm/i915/guc: Assert that we switch between known ggtt->invalidate functions"

This reverts commit 04f7b24eccdfae680a36e9825fe0d61dcd5ed528.
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 8 ++------
 drivers/gpu/drm/i915/i915_params.c  | 4 ++--
 2 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 09e524dbc090..478a8d42aeb0 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -3189,17 +3189,13 @@ int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv)
 
 void i915_ggtt_enable_guc(struct drm_i915_private *i915)
 {
-	GEM_BUG_ON(i915->ggtt.invalidate != gen6_ggtt_invalidate);
-
 	i915->ggtt.invalidate = guc_ggtt_invalidate;
 }
 
 void i915_ggtt_disable_guc(struct drm_i915_private *i915)
 {
-	/* We should only be called after i915_ggtt_enable_guc() */
-	GEM_BUG_ON(i915->ggtt.invalidate != guc_ggtt_invalidate);
-
-	i915->ggtt.invalidate = gen6_ggtt_invalidate;
+	if (i915->ggtt.invalidate == guc_ggtt_invalidate)
+		i915->ggtt.invalidate = gen6_ggtt_invalidate;
 }
 
 void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv)
diff --git a/drivers/gpu/drm/i915/i915_params.c b/drivers/gpu/drm/i915/i915_params.c
index 8ab003dca113..c9d72f1b8383 100644
--- a/drivers/gpu/drm/i915/i915_params.c
+++ b/drivers/gpu/drm/i915/i915_params.c
@@ -56,8 +56,8 @@ struct i915_params i915 __read_mostly = {
 	.verbose_state_checks = 1,
 	.nuclear_pageflip = 0,
 	.edp_vswing = 0,
-	.enable_guc_loading = 0,
-	.enable_guc_submission = 0,
+	.enable_guc_loading = 2,
+	.enable_guc_submission = 2,
 	.guc_log_level = -1,
 	.guc_firmware_path = NULL,
 	.huc_firmware_path = NULL,
-- 
2.13.5

_______________________________________________
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 [1/5] drm/i915/guc: Remove obsolete comments and remove unused variable

[Patchwork]

== Series Details ==

Series: series starting with [1/5] drm/i915/guc: Remove obsolete comments and remove unused variable
URL   : https://patchwork.freedesktop.org/series/30230/
State : failure

== Summary ==

Series 30230v1 series starting with [1/5] drm/i915/guc: Remove obsolete comments and remove unused variable
https://patchwork.freedesktop.org/api/1.0/series/30230/revisions/1/mbox/

Test chamelium:
        Subgroup dp-crc-fast:
                pass       -> INCOMPLETE (fi-kbl-7500u) fdo#102514
Test debugfs_test:
        Subgroup read_all_entries:
                pass       -> SKIP       (fi-glk-2a)
Test drv_hangman:
        Subgroup error-state-basic:
                pass       -> SKIP       (fi-glk-2a)
Test gem_busy:
        Subgroup basic-busy-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-hang-default:
                pass       -> SKIP       (fi-glk-2a)
Test gem_close_race:
        Subgroup basic-process:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-threads:
                pass       -> SKIP       (fi-glk-2a)
Test gem_cpu_reloc:
        Subgroup basic:
                pass       -> SKIP       (fi-glk-2a)
Test gem_cs_tlb:
        Subgroup basic-default:
                pass       -> SKIP       (fi-glk-2a)
Test gem_ctx_create:
        Subgroup basic:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-files:
                pass       -> SKIP       (fi-glk-2a)
Test gem_ctx_exec:
        Subgroup basic:
                pass       -> SKIP       (fi-glk-2a)
Test gem_ctx_switch:
        Subgroup basic-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-default-heavy:
                pass       -> SKIP       (fi-glk-2a)
Test gem_exec_basic:
        Subgroup basic-blt:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-bsd:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-render:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-vebox:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup gtt-blt:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup gtt-bsd:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup gtt-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup gtt-render:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup gtt-vebox:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup readonly-blt:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup readonly-bsd:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup readonly-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup readonly-render:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup readonly-vebox:
                pass       -> SKIP       (fi-glk-2a)
Test gem_exec_create:
        Subgroup basic:
                pass       -> SKIP       (fi-glk-2a)
Test gem_exec_fence:
        Subgroup basic-busy-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-wait-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-await-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup await-hang-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup nb-await-default:
                pass       -> SKIP       (fi-glk-2a)
Test gem_exec_flush:
        Subgroup basic-batch-kernel-default-uc:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-batch-kernel-default-wb:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-uc-pro-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-uc-prw-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-uc-ro-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-uc-rw-default:
                pass       -> SKIP       (fi-glk-2a)
        Subgroup basic-uc-set-default:
WARNING: Long output truncated

c422dc1ff5f3ab3d3bea7556b542d415a9bd93eb drm-tip: 2017y-09m-12d-17h-33m-37s UTC integration manifest
62ac81736b4f HAX Enable GuC Submission for CI
74cb87023934 drm/i915/guc: Make adding GuC work items lockless
ef4d46942c6f drm/i915/guc: Simplify GuC doorbell logic
660070ff4ef6 drm/i915/guc: Submit GuC workitems containing coalesced requests
7fb18f07c186 drm/i915/guc: Remove obsolete comments and remove unused variable

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5666/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx