[Intel-gfx] [PATCH 5/9] drm/i915: Derive GEM requests from dma-fence

Chris Wilson chris at chris-wilson.co.uk
Fri Jul 15 10:11:17 UTC 2016


dma-buf provides a generic fence class for interoperation between
drivers. Internally we use the request structure as a fence, and so with
only a little bit of interfacing we can rebase those requests on top of
dma-buf fences. This will allow us, in the future, to pass those fences
back to userspace or between drivers.

v2: The fence_context needs to be globally unique, not just unique to
this device.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes at virtuousgeek.org>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |   2 +-
 drivers/gpu/drm/i915/i915_gem_request.c    | 113 ++++++++++++++++++++++++++---
 drivers/gpu/drm/i915/i915_gem_request.h    |  43 +++++++----
 drivers/gpu/drm/i915/i915_gpu_error.c      |   2 +-
 drivers/gpu/drm/i915/i915_guc_submission.c |   4 +-
 drivers/gpu/drm/i915/i915_trace.h          |  10 +--
 drivers/gpu/drm/i915/intel_breadcrumbs.c   |   7 +-
 drivers/gpu/drm/i915/intel_engine_cs.c     |   2 +
 drivers/gpu/drm/i915/intel_lrc.c           |   2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c    |  10 +--
 drivers/gpu/drm/i915/intel_ringbuffer.h    |   1 +
 11 files changed, 150 insertions(+), 46 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 90aef4540193..55fd3d9cc448 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -768,7 +768,7 @@ static int i915_gem_request_info(struct seq_file *m, void *data)
 			if (req->pid)
 				task = pid_task(req->pid, PIDTYPE_PID);
 			seq_printf(m, "    %x @ %d: %s [%d]\n",
-				   req->seqno,
+				   req->fence.seqno,
 				   (int) (jiffies - req->emitted_jiffies),
 				   task ? task->comm : "<unknown>",
 				   task ? task->pid : -1);
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 5cbb11ece60a..6528536878f2 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -24,6 +24,95 @@
 
 #include "i915_drv.h"
 
+static const char *i915_fence_get_driver_name(struct fence *fence)
+{
+	return "i915";
+}
+
+static const char *i915_fence_get_timeline_name(struct fence *fence)
+{
+	/* Timelines are bound by eviction to a VM. However, since
+	 * we only have a global seqno at the moment, we only have
+	 * a single timeline. Note that each timeline will have
+	 * multiple execution contexts (fence contexts) as we allow
+	 * engines within a single timeline to execute in parallel.
+	 */
+	return "global";
+}
+
+static bool i915_fence_signaled(struct fence *fence)
+{
+	return i915_gem_request_completed(to_request(fence));
+}
+
+static bool i915_fence_enable_signaling(struct fence *fence)
+{
+	if (i915_fence_signaled(fence))
+		return false;
+
+	intel_engine_enable_signaling(to_request(fence));
+	return true;
+}
+
+static signed long i915_fence_wait(struct fence *fence,
+				   bool interruptible,
+				   signed long timeout_jiffies)
+{
+	s64 timeout_ns, *timeout;
+	int ret;
+
+	if (timeout_jiffies != MAX_SCHEDULE_TIMEOUT) {
+		timeout_ns = jiffies_to_nsecs(timeout_jiffies);
+		timeout = &timeout_ns;
+	} else {
+		timeout = NULL;
+	}
+
+	ret = __i915_wait_request(to_request(fence),
+				  interruptible, timeout,
+				  NULL);
+	if (ret == -ETIME)
+		return 0;
+
+	if (ret < 0)
+		return ret;
+
+	if (timeout_jiffies != MAX_SCHEDULE_TIMEOUT)
+		timeout_jiffies = nsecs_to_jiffies(timeout_ns);
+
+	return timeout_jiffies;
+}
+
+static void i915_fence_value_str(struct fence *fence, char *str, int size)
+{
+	snprintf(str, size, "%u", fence->seqno);
+}
+
+static void i915_fence_timeline_value_str(struct fence *fence, char *str,
+					  int size)
+{
+	snprintf(str, size, "%u",
+		 intel_engine_get_seqno(to_request(fence)->engine));
+}
+
+static void i915_fence_release(struct fence *fence)
+{
+	struct drm_i915_gem_request *req = to_request(fence);
+
+	kmem_cache_free(req->i915->requests, req);
+}
+
+const struct fence_ops i915_fence_ops = {
+	.get_driver_name = i915_fence_get_driver_name,
+	.get_timeline_name = i915_fence_get_timeline_name,
+	.enable_signaling = i915_fence_enable_signaling,
+	.signaled = i915_fence_signaled,
+	.wait = i915_fence_wait,
+	.release = i915_fence_release,
+	.fence_value_str = i915_fence_value_str,
+	.timeline_value_str = i915_fence_timeline_value_str,
+};
+
 int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
 				   struct drm_file *file)
 {
@@ -211,6 +300,7 @@ __i915_gem_request_alloc(struct intel_engine_cs *engine,
 	struct drm_i915_private *dev_priv = engine->i915;
 	unsigned int reset_counter = i915_reset_counter(&dev_priv->gpu_error);
 	struct drm_i915_gem_request *req;
+	u32 seqno;
 	int ret;
 
 	if (!req_out)
@@ -238,11 +328,17 @@ __i915_gem_request_alloc(struct intel_engine_cs *engine,
 	if (!req)
 		return -ENOMEM;
 
-	ret = i915_gem_get_seqno(dev_priv, &req->seqno);
+	ret = i915_gem_get_seqno(dev_priv, &seqno);
 	if (ret)
 		goto err;
 
-	kref_init(&req->ref);
+	spin_lock_init(&req->lock);
+	fence_init(&req->fence,
+		   &i915_fence_ops,
+		   &req->lock,
+		   engine->fence_context,
+		   seqno);
+
 	req->i915 = dev_priv;
 	req->engine = engine;
 	req->ctx = ctx;
@@ -385,7 +481,7 @@ void __i915_add_request(struct drm_i915_gem_request *request,
 	 */
 	request->emitted_jiffies = jiffies;
 	request->previous_seqno = engine->last_submitted_seqno;
-	smp_store_mb(engine->last_submitted_seqno, request->seqno);
+	smp_store_mb(engine->last_submitted_seqno, request->fence.seqno);
 	list_add_tail(&request->list, &engine->request_list);
 
 	/* Record the position of the start of the request so that
@@ -556,7 +652,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 	set_current_state(state);
 	add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
-	intel_wait_init(&wait, req->seqno);
+	intel_wait_init(&wait, req->fence.seqno);
 	if (intel_engine_add_wait(req->engine, &wait))
 		/* In order to check that we haven't missed the interrupt
 		 * as we enabled it, we need to kick ourselves to do a
@@ -617,7 +713,7 @@ complete:
 			*timeout = 0;
 	}
 
-	if (rps && req->seqno == req->engine->last_submitted_seqno) {
+	if (rps && req->fence.seqno == req->engine->last_submitted_seqno) {
 		/* The GPU is now idle and this client has stalled.
 		 * Since no other client has submitted a request in the
 		 * meantime, assume that this client is the only one
@@ -657,10 +753,3 @@ int i915_wait_request(struct drm_i915_gem_request *req)
 
 	return 0;
 }
-
-void i915_gem_request_free(struct kref *req_ref)
-{
-	struct drm_i915_gem_request *req =
-		container_of(req_ref, typeof(*req), ref);
-	kmem_cache_free(req->i915->requests, req);
-}
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index ea700befcc28..6f2c820785f3 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -25,6 +25,10 @@
 #ifndef I915_GEM_REQUEST_H
 #define I915_GEM_REQUEST_H
 
+#include <linux/fence.h>
+
+#include "i915_gem.h"
+
 /**
  * Request queue structure.
  *
@@ -36,11 +40,11 @@
  * emission time to be associated with the request for tracking how far ahead
  * of the GPU the submission is.
  *
- * The requests are reference counted, so upon creation they should have an
- * initial reference taken using kref_init
+ * The requests are reference counted.
  */
 struct drm_i915_gem_request {
-	struct kref ref;
+	struct fence fence;
+	spinlock_t lock;
 
 	/** On Which ring this request was generated */
 	struct drm_i915_private *i915;
@@ -66,12 +70,6 @@ struct drm_i915_gem_request {
 	 */
 	u32 previous_seqno;
 
-	/** GEM sequence number associated with this request,
-	 * when the HWS breadcrumb is equal or greater than this the GPU
-	 * has finished processing this request.
-	 */
-	u32 seqno;
-
 	/** Position in the ringbuffer of the start of the request */
 	u32 head;
 
@@ -140,10 +138,16 @@ struct drm_i915_gem_request {
 	unsigned int ctx_hw_id;
 };
 
+extern const struct fence_ops i915_fence_ops;
+
+static inline bool fence_is_i915(struct fence *fence)
+{
+	return fence->ops == &i915_fence_ops;
+}
+
 struct drm_i915_gem_request * __must_check
 i915_gem_request_alloc(struct intel_engine_cs *engine,
 		       struct i915_gem_context *ctx);
-void i915_gem_request_free(struct kref *req_ref);
 int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
 				   struct drm_file *file);
 void i915_gem_request_retire_upto(struct drm_i915_gem_request *req);
@@ -151,7 +155,7 @@ void i915_gem_request_retire_upto(struct drm_i915_gem_request *req);
 static inline u32
 i915_gem_request_get_seqno(struct drm_i915_gem_request *req)
 {
-	return req ? req->seqno : 0;
+	return req ? req->fence.seqno : 0;
 }
 
 static inline struct intel_engine_cs *
@@ -161,17 +165,24 @@ i915_gem_request_get_engine(struct drm_i915_gem_request *req)
 }
 
 static inline struct drm_i915_gem_request *
+to_request(struct 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));
+	return container_of(fence, struct drm_i915_gem_request, fence);
+}
+
+static inline struct drm_i915_gem_request *
 i915_gem_request_reference(struct drm_i915_gem_request *req)
 {
-	if (req)
-		kref_get(&req->ref);
-	return req;
+	return to_request(fence_get(&req->fence));
 }
 
 static inline void
 i915_gem_request_unreference(struct drm_i915_gem_request *req)
 {
-	kref_put(&req->ref, i915_gem_request_free);
+	fence_put(&req->fence);
 }
 
 static inline void i915_gem_request_assign(struct drm_i915_gem_request **pdst,
@@ -223,7 +234,7 @@ static inline bool
 i915_gem_request_completed(const struct drm_i915_gem_request *req)
 {
 	return i915_seqno_passed(intel_engine_get_seqno(req->engine),
-				 req->seqno);
+				 req->fence.seqno);
 }
 
 bool __i915_spin_request(const struct drm_i915_gem_request *request,
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 9d73d2216adc..6daaf4ecd2da 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -1182,7 +1182,7 @@ static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
 			}
 
 			erq = &error->ring[i].requests[count++];
-			erq->seqno = request->seqno;
+			erq->seqno = request->fence.seqno;
 			erq->jiffies = request->emitted_jiffies;
 			erq->tail = request->postfix;
 		}
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 2112e029db6a..1cc5de129e76 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -506,7 +506,7 @@ static void guc_add_workqueue_item(struct i915_guc_client *gc,
 							     rq->engine);
 
 	wqi->ring_tail = tail << WQ_RING_TAIL_SHIFT;
-	wqi->fence_id = rq->seqno;
+	wqi->fence_id = rq->fence.seqno;
 
 	kunmap_atomic(base);
 }
@@ -601,7 +601,7 @@ int i915_guc_submit(struct drm_i915_gem_request *rq)
 		client->b_fail += 1;
 
 	guc->submissions[engine_id] += 1;
-	guc->last_seqno[engine_id] = rq->seqno;
+	guc->last_seqno[engine_id] = rq->fence.seqno;
 
 	return b_ret;
 }
diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
index 534154e05fbe..007112d1e049 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -465,7 +465,7 @@ TRACE_EVENT(i915_gem_ring_sync_to,
 			   __entry->dev = from->i915->drm.primary->index;
 			   __entry->sync_from = from->id;
 			   __entry->sync_to = to_req->engine->id;
-			   __entry->seqno = i915_gem_request_get_seqno(req);
+			   __entry->seqno = req->fence.seqno;
 			   ),
 
 	    TP_printk("dev=%u, sync-from=%u, sync-to=%u, seqno=%u",
@@ -488,9 +488,9 @@ TRACE_EVENT(i915_gem_ring_dispatch,
 	    TP_fast_assign(
 			   __entry->dev = req->i915->drm.primary->index;
 			   __entry->ring = req->engine->id;
-			   __entry->seqno = req->seqno;
+			   __entry->seqno = req->fence.seqno;
 			   __entry->flags = flags;
-			   intel_engine_enable_signaling(req);
+			   fence_enable_sw_signaling(&req->fence);
 			   ),
 
 	    TP_printk("dev=%u, ring=%u, seqno=%u, flags=%x",
@@ -533,7 +533,7 @@ DECLARE_EVENT_CLASS(i915_gem_request,
 	    TP_fast_assign(
 			   __entry->dev = req->i915->drm.primary->index;
 			   __entry->ring = req->engine->id;
-			   __entry->seqno = req->seqno;
+			   __entry->seqno = req->fence.seqno;
 			   ),
 
 	    TP_printk("dev=%u, ring=%u, seqno=%u",
@@ -595,7 +595,7 @@ TRACE_EVENT(i915_gem_request_wait_begin,
 	    TP_fast_assign(
 			   __entry->dev = req->i915->drm.primary->index;
 			   __entry->ring = req->engine->id;
-			   __entry->seqno = req->seqno;
+			   __entry->seqno = req->fence.seqno;
 			   __entry->blocking =
 				     mutex_is_locked(&req->i915->drm.struct_mutex);
 			   ),
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index b074f3d6d127..32ada41dedfb 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -436,6 +436,7 @@ static int intel_breadcrumbs_signaler(void *arg)
 			 */
 			intel_engine_remove_wait(engine,
 						 &request->signaling.wait);
+			fence_signal(&request->fence);
 
 			/* Find the next oldest signal. Note that as we have
 			 * not been holding the lock, another client may
@@ -482,7 +483,7 @@ void intel_engine_enable_signaling(struct drm_i915_gem_request *request)
 	}
 
 	request->signaling.wait.tsk = b->signaler;
-	request->signaling.wait.seqno = request->seqno;
+	request->signaling.wait.seqno = request->fence.seqno;
 	i915_gem_request_reference(request);
 
 	/* First add ourselves into the list of waiters, but register our
@@ -504,8 +505,8 @@ void intel_engine_enable_signaling(struct drm_i915_gem_request *request)
 	p = &b->signals.rb_node;
 	while (*p) {
 		parent = *p;
-		if (i915_seqno_passed(request->seqno,
-				      to_signaler(parent)->seqno)) {
+		if (i915_seqno_passed(request->fence.seqno,
+				      to_signaler(parent)->fence.seqno)) {
 			p = &parent->rb_right;
 			first = false;
 		} else {
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index e3c9f04ea51e..f4a35ec78481 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -182,6 +182,8 @@ void intel_engine_setup_common(struct intel_engine_cs *engine)
 	INIT_LIST_HEAD(&engine->execlist_queue);
 	spin_lock_init(&engine->execlist_lock);
 
+	engine->fence_context = fence_context_alloc(1);
+
 	intel_engine_init_hangcheck(engine);
 	i915_gem_batch_pool_init(&engine->i915->drm, &engine->batch_pool);
 }
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index b6af635e3a0f..b82c12e38e4b 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1827,7 +1827,7 @@ static int gen8_emit_request(struct drm_i915_gem_request *request)
 				intel_hws_seqno_address(request->engine) |
 				MI_FLUSH_DW_USE_GTT);
 	intel_logical_ring_emit(ringbuf, 0);
-	intel_logical_ring_emit(ringbuf, request->seqno);
+	intel_logical_ring_emit(ringbuf, request->fence.seqno);
 	intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT);
 	intel_logical_ring_emit(ringbuf, MI_NOOP);
 	return intel_logical_ring_advance_and_submit(request);
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 94c8ef461721..af0bd71e3a36 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1348,7 +1348,7 @@ static int gen8_rcs_signal(struct drm_i915_gem_request *signaller_req,
 					   PIPE_CONTROL_CS_STALL);
 		intel_ring_emit(signaller, lower_32_bits(gtt_offset));
 		intel_ring_emit(signaller, upper_32_bits(gtt_offset));
-		intel_ring_emit(signaller, signaller_req->seqno);
+		intel_ring_emit(signaller, signaller_req->fence.seqno);
 		intel_ring_emit(signaller, 0);
 		intel_ring_emit(signaller, MI_SEMAPHORE_SIGNAL |
 					   MI_SEMAPHORE_TARGET(waiter->hw_id));
@@ -1386,7 +1386,7 @@ static int gen8_xcs_signal(struct drm_i915_gem_request *signaller_req,
 		intel_ring_emit(signaller, lower_32_bits(gtt_offset) |
 					   MI_FLUSH_DW_USE_GTT);
 		intel_ring_emit(signaller, upper_32_bits(gtt_offset));
-		intel_ring_emit(signaller, signaller_req->seqno);
+		intel_ring_emit(signaller, signaller_req->fence.seqno);
 		intel_ring_emit(signaller, MI_SEMAPHORE_SIGNAL |
 					   MI_SEMAPHORE_TARGET(waiter->hw_id));
 		intel_ring_emit(signaller, 0);
@@ -1419,7 +1419,7 @@ static int gen6_signal(struct drm_i915_gem_request *signaller_req,
 		if (i915_mmio_reg_valid(mbox_reg)) {
 			intel_ring_emit(signaller, MI_LOAD_REGISTER_IMM(1));
 			intel_ring_emit_reg(signaller, mbox_reg);
-			intel_ring_emit(signaller, signaller_req->seqno);
+			intel_ring_emit(signaller, signaller_req->fence.seqno);
 		}
 	}
 
@@ -1455,7 +1455,7 @@ gen6_add_request(struct drm_i915_gem_request *req)
 	intel_ring_emit(engine, MI_STORE_DWORD_INDEX);
 	intel_ring_emit(engine,
 			I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
-	intel_ring_emit(engine, req->seqno);
+	intel_ring_emit(engine, req->fence.seqno);
 	intel_ring_emit(engine, MI_USER_INTERRUPT);
 	__intel_ring_advance(engine);
 
@@ -1704,7 +1704,7 @@ i9xx_add_request(struct drm_i915_gem_request *req)
 	intel_ring_emit(engine, MI_STORE_DWORD_INDEX);
 	intel_ring_emit(engine,
 			I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
-	intel_ring_emit(engine, req->seqno);
+	intel_ring_emit(engine, req->fence.seqno);
 	intel_ring_emit(engine, MI_USER_INTERRUPT);
 	__intel_ring_advance(engine);
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index df7587ab1ba7..5cbafc00bfd5 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -146,6 +146,7 @@ struct intel_engine_cs {
 	unsigned int exec_id;
 	unsigned int hw_id;
 	unsigned int guc_id; /* XXX same as hw_id? */
+	u64 fence_context;
 	u32		mmio_base;
 	unsigned int irq_shift;
 	struct intel_ringbuffer *buffer;
-- 
2.8.1



More information about the Intel-gfx mailing list