[PATCH 54/54] drm/i915/execlists: Preempt-to-busy

Chris Wilson chris at chris-wilson.co.uk
Tue Apr 16 12:29:12 UTC 2019


When using a global seqno, we required a precise stop-the-workd event to
handle preemption and unwind the global seqno counter. To accomplish
this, we would preempt to a special out-of-band context and wait for the
machine to report that it was idle. Given an idle machine, we could very
precisely see which requests had completed and which we needed to feed
back into the run queue.

However, now that we have scrapped the global seqno, we no longer need
to precisely unwind the global counter and only track requests by their
per-context seqno. This allows us to loosely unwind inflight requests
while scheduling a preemption, with the enormous caveat that the
requests we put back on the run queue are still _inflight_ (until the
preemption request is complete). This makes request tracking much more
messy, as at any point then we can see a completed request that we
believe is not currently scheduled for execution. We also have to be
careful not to rewind RING_TAIL past RING_HEAD on preempting to the
running context, and for this we use a semaphore to prevent completion
of the request before continuing.

To accomplish this feat, we change how we track requests scheduled to
the HW. Instead of appending our requests onto a single list as we
submit, we track each submission to ELSP as its own block. Then upon
receiving the CS preemption event, we promote the pending block to the
inflight block (discarding what was previously being tracked). As normal
CS completion events arrive, we then remove stale entries from the
inflight tracker.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/gem/i915_gem_context.c   |   2 +-
 drivers/gpu/drm/i915/gt/intel_breadcrumbs.c   |  52 +-
 drivers/gpu/drm/i915/gt/intel_context_types.h |   5 +
 drivers/gpu/drm/i915/gt/intel_engine.h        |  61 +-
 drivers/gpu/drm/i915/gt/intel_engine_cs.c     |  58 +-
 drivers/gpu/drm/i915/gt/intel_engine_types.h  |  52 +-
 drivers/gpu/drm/i915/gt/intel_lrc.c           | 646 +++++++++---------
 drivers/gpu/drm/i915/i915_gpu_error.c         |  19 +-
 drivers/gpu/drm/i915/i915_request.c           |   7 +
 drivers/gpu/drm/i915/i915_request.h           |   1 +
 drivers/gpu/drm/i915/i915_scheduler.c         |  13 +-
 drivers/gpu/drm/i915/i915_utils.h             |  13 +
 drivers/gpu/drm/i915/intel_guc_submission.c   | 174 +++--
 drivers/gpu/drm/i915/selftests/i915_request.c |   8 +-
 14 files changed, 500 insertions(+), 611 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index c3a51fb7c950..ce4e4aa6a95c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -629,7 +629,7 @@ static void init_contexts(struct drm_i915_private *i915)
 
 static bool needs_preempt_context(struct drm_i915_private *i915)
 {
-	return HAS_EXECLISTS(i915);
+	return USES_GUC_SUBMISSION(i915);
 }
 
 int i915_gem_contexts_init(struct drm_i915_private *dev_priv)
diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
index 3cbffd400b1b..95b1b31df590 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -83,6 +83,7 @@ static inline bool __request_completed(const struct i915_request *rq)
 void intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	const ktime_t timestamp = ktime_get();
 	struct intel_context *ce, *cn;
 	struct list_head *pos, *next;
 	LIST_HEAD(signal);
@@ -104,6 +105,16 @@ void intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
 
 			GEM_BUG_ON(!test_bit(I915_FENCE_FLAG_SIGNAL,
 					     &rq->fence.flags));
+			clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
+
+			if (test_and_set_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+					     &rq->fence.flags)) {
+				list_del(&rq->signal_link);
+				continue;
+			}
+
+			rq->fence.timestamp = timestamp;
+			set_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT, &rq->fence.flags);
 
 			/*
 			 * Queue for execution after dropping the signaling
@@ -111,14 +122,6 @@ void intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
 			 * more signalers to the same context or engine.
 			 */
 			i915_request_get(rq);
-
-			/*
-			 * We may race with direct invocation of
-			 * dma_fence_signal(), e.g. i915_request_retire(),
-			 * so we need to acquire our reference to the request
-			 * before we cancel the breadcrumb.
-			 */
-			clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
 			list_add_tail(&rq->signal_link, &signal);
 		}
 
@@ -140,8 +143,17 @@ void intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
 	list_for_each_safe(pos, next, &signal) {
 		struct i915_request *rq =
 			list_entry(pos, typeof(*rq), signal_link);
+		struct dma_fence_cb *cur, *tmp;
+
+		/* trace_dma_fence_signaled(fence); */
+
+		spin_lock(&rq->lock);
+		list_for_each_entry_safe(cur, tmp, &rq->fence.cb_list, node) {
+			INIT_LIST_HEAD(&cur->node);
+			cur->func(&rq->fence, cur);
+		}
+		spin_unlock(&rq->lock);
 
-		dma_fence_signal(&rq->fence);
 		i915_request_put(rq);
 	}
 }
@@ -243,19 +255,17 @@ void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
 
 bool i915_request_enable_breadcrumb(struct i915_request *rq)
 {
-	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
-
-	GEM_BUG_ON(test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags));
-
-	if (!test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags))
-		return true;
+	lockdep_assert_held(&rq->lock);
+	lockdep_assert_irqs_disabled();
 
-	spin_lock(&b->irq_lock);
-	if (test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags) &&
-	    !__request_completed(rq)) {
+	if (test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags)) {
+		struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
 		struct intel_context *ce = rq->hw_context;
 		struct list_head *pos;
 
+		spin_lock(&b->irq_lock);
+		GEM_BUG_ON(test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags));
+
 		__intel_breadcrumbs_arm_irq(b);
 
 		/*
@@ -284,8 +294,8 @@ bool i915_request_enable_breadcrumb(struct i915_request *rq)
 			list_move_tail(&ce->signal_link, &b->signalers);
 
 		set_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
+		spin_unlock(&b->irq_lock);
 	}
-	spin_unlock(&b->irq_lock);
 
 	return !__request_completed(rq);
 }
@@ -294,8 +304,8 @@ void i915_request_cancel_breadcrumb(struct i915_request *rq)
 {
 	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
 
-	if (!test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags))
-		return;
+	lockdep_assert_held(&rq->lock);
+	lockdep_assert_irqs_disabled();
 
 	spin_lock(&b->irq_lock);
 	if (test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags)) {
diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h b/drivers/gpu/drm/i915/gt/intel_context_types.h
index 92401d25764b..05e984864098 100644
--- a/drivers/gpu/drm/i915/gt/intel_context_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
@@ -13,6 +13,7 @@
 #include <linux/types.h>
 
 #include "i915_active_types.h"
+#include "i915_utils.h"
 #include "intel_sseu.h"
 
 struct i915_gem_context;
@@ -37,6 +38,10 @@ struct intel_context {
 	struct i915_gem_context *gem_context;
 	struct intel_engine_cs *engine;
 	struct intel_engine_cs *inflight;
+#define intel_context_inflight(ce) ptr_mask_bits((ce)->inflight, 2)
+#define intel_context_inflight_count(ce)  ptr_unmask_bits((ce)->inflight, 2)
+#define intel_context_inflight_inc(ce) ptr_count_inc(&(ce)->inflight)
+#define intel_context_inflight_dec(ce) ptr_count_dec(&(ce)->inflight)
 
 	struct list_head signal_link;
 	struct list_head signals;
diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h
index d8eed9455fc2..2c8c01876ba7 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine.h
@@ -124,71 +124,26 @@ static inline bool __execlists_need_preempt(int prio, int last)
 	return prio > max(I915_PRIORITY_NORMAL - 1, last);
 }
 
-static inline void
-execlists_set_active(struct intel_engine_execlists *execlists,
-		     unsigned int bit)
-{
-	__set_bit(bit, (unsigned long *)&execlists->active);
-}
-
-static inline bool
-execlists_set_active_once(struct intel_engine_execlists *execlists,
-			  unsigned int bit)
-{
-	return !__test_and_set_bit(bit, (unsigned long *)&execlists->active);
-}
-
-static inline void
-execlists_clear_active(struct intel_engine_execlists *execlists,
-		       unsigned int bit)
-{
-	__clear_bit(bit, (unsigned long *)&execlists->active);
-}
-
-static inline void
-execlists_clear_all_active(struct intel_engine_execlists *execlists)
+static inline unsigned int
+execlists_num_ports(const struct intel_engine_execlists * const execlists)
 {
-	execlists->active = 0;
+	return execlists->port_mask + 1;
 }
 
-static inline bool
-execlists_is_active(const struct intel_engine_execlists *execlists,
-		    unsigned int bit)
+static inline struct i915_request *
+execlists_active(const struct intel_engine_execlists *execlists)
 {
-	return test_bit(bit, (unsigned long *)&execlists->active);
+	GEM_BUG_ON(execlists->active - execlists->inflight >
+		   execlists_num_ports(execlists));
+	return READ_ONCE(*execlists->active);
 }
 
-void execlists_user_begin(struct intel_engine_execlists *execlists,
-			  const struct execlist_port *port);
-void execlists_user_end(struct intel_engine_execlists *execlists);
-
 void
 execlists_cancel_port_requests(struct intel_engine_execlists * const execlists);
 
 struct i915_request *
 execlists_unwind_incomplete_requests(struct intel_engine_execlists *execlists);
 
-static inline unsigned int
-execlists_num_ports(const struct intel_engine_execlists * const execlists)
-{
-	return execlists->port_mask + 1;
-}
-
-static inline struct execlist_port *
-execlists_port_complete(struct intel_engine_execlists * const execlists,
-			struct execlist_port * const port)
-{
-	const unsigned int m = execlists->port_mask;
-
-	GEM_BUG_ON(port_index(port, execlists) != 0);
-	GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_USER));
-
-	memmove(port, port + 1, m * sizeof(struct execlist_port));
-	memset(port + m, 0, sizeof(struct execlist_port));
-
-	return port;
-}
-
 static inline u32
 intel_read_status_page(const struct intel_engine_cs *engine, int reg)
 {
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 8d768f817fad..72c96d1d078d 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -442,6 +442,7 @@ void intel_engine_init_execlists(struct intel_engine_cs *engine)
 	execlists->port_mask = 1;
 	GEM_BUG_ON(!is_power_of_2(execlists_num_ports(execlists)));
 	GEM_BUG_ON(execlists_num_ports(execlists) > EXECLIST_MAX_PORTS);
+	execlists->active = &execlists->inflight[0];
 
 	execlists->queue_priority_hint = INT_MIN;
 	execlists->queue = RB_ROOT_CACHED;
@@ -1085,7 +1086,7 @@ bool intel_engine_is_idle(struct intel_engine_cs *engine)
 		return true;
 
 	/* Waiting to drain ELSP? */
-	if (READ_ONCE(engine->execlists.active)) {
+	if (execlists_active(&engine->execlists)) {
 		struct tasklet_struct *t = &engine->execlists.tasklet;
 
 		local_bh_disable();
@@ -1100,7 +1101,7 @@ bool intel_engine_is_idle(struct intel_engine_cs *engine)
 		/* Otherwise flush the tasklet if it was on another cpu */
 		tasklet_unlock_wait(t);
 
-		if (READ_ONCE(engine->execlists.active))
+		if (execlists_active(&engine->execlists))
 			return false;
 	}
 
@@ -1294,6 +1295,7 @@ static void intel_engine_print_registers(const struct intel_engine_cs *engine,
 	}
 
 	if (HAS_EXECLISTS(dev_priv)) {
+		struct i915_request * const *port, *rq;
 		const u32 *hws =
 			&engine->status_page.addr[I915_HWS_CSB_BUF0_INDEX];
 		const u8 num_entries = execlists->csb_size;
@@ -1326,26 +1328,28 @@ static void intel_engine_print_registers(const struct intel_engine_cs *engine,
 		}
 
 		rcu_read_lock();
-		for (idx = 0; idx < execlists_num_ports(execlists); idx++) {
-			struct i915_request *rq;
-			unsigned int count;
-
-			rq = port_unpack(&execlists->port[idx], &count);
-			if (rq) {
-				char hdr[80];
-
-				snprintf(hdr, sizeof(hdr),
-					 "\t\tELSP[%d] count=%d, ring:{start:%08x, hwsp:%08x, seqno:%08x}, rq: ",
-					 idx, count,
-					 i915_ggtt_offset(rq->ring->vma),
-					 rq->timeline->hwsp_offset,
-					 hwsp_seqno(rq));
-				print_request(m, rq, hdr);
-			} else {
-				drm_printf(m, "\t\tELSP[%d] idle\n", idx);
-			}
+		for (port = execlists->active; (rq = *port); port++) {
+			char hdr[80];
+
+			snprintf(hdr, sizeof(hdr),
+				 "\t\tActive[%d] ring:{start:%08x, hwsp:%08x, seqno:%08x}, rq: ",
+				 (int)(port - execlists->active),
+				 i915_ggtt_offset(rq->ring->vma),
+				 rq->timeline->hwsp_offset,
+				 hwsp_seqno(rq));
+			print_request(m, rq, hdr);
+		}
+		for (port = execlists->pending; (rq = *port); port++) {
+			char hdr[80];
+
+			snprintf(hdr, sizeof(hdr),
+				 "\t\tPending[%d] ring:{start:%08x, hwsp:%08x, seqno:%08x}, rq: ",
+				 (int)(port - execlists->pending),
+				 i915_ggtt_offset(rq->ring->vma),
+				 rq->timeline->hwsp_offset,
+				 hwsp_seqno(rq));
+			print_request(m, rq, hdr);
 		}
-		drm_printf(m, "\t\tHW active? 0x%x\n", execlists->active);
 		rcu_read_unlock();
 	} else if (INTEL_GEN(dev_priv) > 6) {
 		drm_printf(m, "\tPP_DIR_BASE: 0x%08x\n",
@@ -1511,15 +1515,19 @@ int intel_enable_engine_stats(struct intel_engine_cs *engine)
 	}
 
 	if (engine->stats.enabled++ == 0) {
-		const struct execlist_port *port = execlists->port;
-		unsigned int num_ports = execlists_num_ports(execlists);
+		struct i915_request * const *port;
+		struct i915_request *rq;
 
 		engine->stats.enabled_at = ktime_get();
 
 		/* XXX submission method oblivious? */
-		while (num_ports-- && port_isset(port)) {
+		for (port = execlists->active; (rq = *port); port++)
 			engine->stats.active++;
-			port++;
+
+		for (port = execlists->pending; (rq = *port); port++) {
+			/* Exclude any contexts already counted in active */
+			if (intel_context_inflight_count(rq->hw_context) == 1)
+				engine->stats.active++;
 		}
 
 		if (engine->stats.active)
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 4f4f796bc0b8..1a3c8e2bf93e 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -160,51 +160,10 @@ struct intel_engine_execlists {
 	 */
 	u32 __iomem *ctrl_reg;
 
-	/**
-	 * @port: execlist port states
-	 *
-	 * For each hardware ELSP (ExecList Submission Port) we keep
-	 * track of the last request and the number of times we submitted
-	 * that port to hw. We then count the number of times the hw reports
-	 * a context completion or preemption. As only one context can
-	 * be active on hw, we limit resubmission of context to port[0]. This
-	 * is called Lite Restore, of the context.
-	 */
-	struct execlist_port {
-		/**
-		 * @request_count: combined request and submission count
-		 */
-		struct i915_request *request_count;
-#define EXECLIST_COUNT_BITS 2
-#define port_request(p) ptr_mask_bits((p)->request_count, EXECLIST_COUNT_BITS)
-#define port_count(p) ptr_unmask_bits((p)->request_count, EXECLIST_COUNT_BITS)
-#define port_pack(rq, count) ptr_pack_bits(rq, count, EXECLIST_COUNT_BITS)
-#define port_unpack(p, count) ptr_unpack_bits((p)->request_count, count, EXECLIST_COUNT_BITS)
-#define port_set(p, packed) ((p)->request_count = (packed))
-#define port_isset(p) ((p)->request_count)
-#define port_index(p, execlists) ((p) - (execlists)->port)
-
-		/**
-		 * @context_id: context ID for port
-		 */
-		GEM_DEBUG_DECL(u32 context_id);
-
 #define EXECLIST_MAX_PORTS 2
-	} port[EXECLIST_MAX_PORTS];
-
-	/**
-	 * @active: is the HW active? We consider the HW as active after
-	 * submitting any context for execution and until we have seen the
-	 * last context completion event. After that, we do not expect any
-	 * more events until we submit, and so can park the HW.
-	 *
-	 * As we have a small number of different sources from which we feed
-	 * the HW, we track the state of each inside a single bitfield.
-	 */
-	unsigned int active;
-#define EXECLISTS_ACTIVE_USER 0
-#define EXECLISTS_ACTIVE_PREEMPT 1
-#define EXECLISTS_ACTIVE_HWACK 2
+	struct i915_request * const *active;
+	struct i915_request *inflight[EXECLIST_MAX_PORTS + 1 /* sentinel */];
+	struct i915_request *pending[EXECLIST_MAX_PORTS + 1];
 
 	/**
 	 * @port_mask: number of execlist ports - 1
@@ -245,11 +204,6 @@ struct intel_engine_execlists {
 	 */
 	u32 *csb_status;
 
-	/**
-	 * @preempt_complete_status: expected CSB upon completing preemption
-	 */
-	u32 preempt_complete_status;
-
 	/**
 	 * @csb_size: context status buffer FIFO size
 	 */
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
index 710343c67501..16eb75e749df 100644
--- a/drivers/gpu/drm/i915/gt/intel_lrc.c
+++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
@@ -161,6 +161,8 @@
 #define GEN8_CTX_STATUS_COMPLETED_MASK \
 	 (GEN8_CTX_STATUS_COMPLETE | GEN8_CTX_STATUS_PREEMPTED)
 
+#define CTX_DESC_FORCE_RESTORE BIT_ULL(2)
+
 /* Typical size of the average request (2 pipecontrols and a MI_BB) */
 #define EXECLISTS_REQUEST_SIZE 64 /* bytes */
 #define WA_TAIL_DWORDS 2
@@ -222,6 +224,14 @@ static void execlists_init_reg_state(u32 *reg_state,
 				     struct intel_engine_cs *engine,
 				     struct intel_ring *ring);
 
+static inline u32 intel_hws_preempt_address(struct intel_engine_cs *engine)
+{
+	return (i915_ggtt_offset(engine->status_page.vma) +
+		I915_GEM_HWS_PREEMPT_ADDR);
+}
+
+#define ring_suspend(E) ((E)->status_page.addr[I915_GEM_HWS_PREEMPT])
+
 static inline struct i915_priolist *to_priolist(struct rb_node *rb)
 {
 	return rb_entry(rb, struct i915_priolist, node);
@@ -283,12 +293,6 @@ static inline bool need_preempt(const struct intel_engine_cs *engine,
 {
 	int last_prio;
 
-	if (!engine->preempt_context)
-		return false;
-
-	if (i915_request_completed(rq))
-		return false;
-
 	/*
 	 * Check if the current priority hint merits a preemption attempt.
 	 *
@@ -347,9 +351,6 @@ __maybe_unused static inline bool
 assert_priority_queue(const struct i915_request *prev,
 		      const struct i915_request *next)
 {
-	const struct intel_engine_execlists *execlists =
-		&prev->engine->execlists;
-
 	/*
 	 * Without preemption, the prev may refer to the still active element
 	 * which we refuse to let go.
@@ -357,7 +358,7 @@ assert_priority_queue(const struct i915_request *prev,
 	 * Even with preemption, there are times when we think it is better not
 	 * to preempt and leave an ostensibly lower priority request in flight.
 	 */
-	if (port_request(execlists->port) == prev)
+	if (i915_request_is_active(prev))
 		return true;
 
 	return rq_prio(prev) >= rq_prio(next);
@@ -451,13 +452,11 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine)
 		struct intel_engine_cs *owner;
 
 		if (i915_request_completed(rq))
-			break;
+			continue; /* XXX */
 
 		__i915_request_unsubmit(rq);
 		unwind_wa_tail(rq);
 
-		GEM_BUG_ON(rq->hw_context->inflight);
-
 		/*
 		 * Push the request back into the queue for later resubmission.
 		 * If this request is not native to this physical engine (i.e.
@@ -506,6 +505,7 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine)
 	 */
 	if (~prio & ACTIVE_PRIORITY && __i915_request_has_started(active)) {
 		prio |= ACTIVE_PRIORITY;
+		GEM_BUG_ON(active->sched.attr.priority >= prio);
 		active->sched.attr.priority = prio;
 		list_move_tail(&active->sched.link,
 			       i915_sched_lookup_priolist(engine, prio));
@@ -537,53 +537,66 @@ execlists_context_status_change(struct i915_request *rq, unsigned long status)
 				   status, rq);
 }
 
-inline void
-execlists_user_begin(struct intel_engine_execlists *execlists,
-		     const struct execlist_port *port)
+static inline struct i915_request *
+execlists_schedule_in(struct i915_request *rq, int idx)
 {
-	execlists_set_active_once(execlists, EXECLISTS_ACTIVE_USER);
-}
+	struct intel_context *ce = rq->hw_context;
+	int count;
 
-inline void
-execlists_user_end(struct intel_engine_execlists *execlists)
-{
-	execlists_clear_active(execlists, EXECLISTS_ACTIVE_USER);
-}
+	trace_i915_request_in(rq, idx);
 
-static inline void
-execlists_context_schedule_in(struct i915_request *rq)
-{
-	GEM_BUG_ON(rq->hw_context->inflight);
+	count = intel_context_inflight_count(ce);
+	if (!count) {
+		intel_context_get(ce);
+		ce->inflight = rq->engine;
 
-	execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_IN);
-	intel_engine_context_in(rq->engine);
-	rq->hw_context->inflight = rq->engine;
+		execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_IN);
+		intel_engine_context_in(ce->inflight);
+	}
+
+	intel_context_inflight_inc(ce);
+	GEM_BUG_ON(intel_context_inflight(ce) != rq->engine);
+
+	return i915_request_get(rq);
 }
 
 static inline void
-execlists_context_schedule_out(struct i915_request *rq, unsigned long status)
+execlists_schedule_out(struct i915_request *rq)
 {
-	rq->hw_context->inflight = NULL;
-	intel_engine_context_out(rq->engine);
-	execlists_context_status_change(rq, status);
+	struct intel_context *ce = rq->hw_context;
+
+	GEM_BUG_ON(!intel_context_inflight_count(ce));
+
 	trace_i915_request_out(rq);
 
-	/*
-	 * If this is part of a virtual engine, its next request may have
-	 * been blocked waiting for access to the active context. We have
-	 * to kick all the siblings again in case we need to switch (e.g.
-	 * the next request is not runnable on this engine). Hopefully,
-	 * we will already have submitted the next request before the
-	 * tasklet runs and do not need to rebuild each virtual tree
-	 * and kick everyone again.
-	 */
-	if (rq->engine != rq->hw_context->engine)
-		tasklet_schedule(&rq->hw_context->engine->execlists.tasklet);
+	intel_context_inflight_dec(ce);
+	if (!intel_context_inflight_count(ce)) {
+		intel_engine_context_out(ce->inflight);
+		execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT);
+
+		ce->inflight = NULL;
+		intel_context_put(ce);
+
+		/*
+		 * If this is part of a virtual engine, its next request may
+		 * have been blocked waiting for access to the active context.
+		 * We have to kick all the siblings again in case we need to
+		 * switch (e.g. the next request is not runnable on this
+		 * engine). Hopefully, we will already have submitted the next
+		 * request before the tasklet runs and do not need to rebuild
+		 * each virtual tree and kick everyone again.
+		 */
+		if (rq->engine != ce->engine)
+			tasklet_schedule(&ce->engine->execlists.tasklet);
+	}
+
+	i915_request_put(rq);
 }
 
-static u64 execlists_update_context(struct i915_request *rq)
+static u64 execlists_update_context(const struct i915_request *rq)
 {
 	struct intel_context *ce = rq->hw_context;
+	u64 desc;
 
 	ce->lrc_reg_state[CTX_RING_TAIL + 1] =
 		intel_ring_set_tail(rq->ring, rq->tail);
@@ -604,7 +617,11 @@ static u64 execlists_update_context(struct i915_request *rq)
 	 * wmb).
 	 */
 	mb();
-	return ce->lrc_desc;
+
+	desc = ce->lrc_desc;
+	ce->lrc_desc &= ~CTX_DESC_FORCE_RESTORE;
+
+	return desc;
 }
 
 static inline void write_desc(struct intel_engine_execlists *execlists, u64 desc, u32 port)
@@ -618,12 +635,54 @@ static inline void write_desc(struct intel_engine_execlists *execlists, u64 desc
 	}
 }
 
+static __maybe_unused void
+trace_ports(const struct intel_engine_execlists *execlists,
+	    const char *msg,
+	    struct i915_request * const *ports)
+{
+	const struct intel_engine_cs *engine =
+		container_of(execlists, typeof(*engine), execlists);
+
+	GEM_TRACE("%s: %s { %llx:%lld%s, %llx:%lld }\n",
+		  engine->name, msg,
+		  ports[0]->fence.context,
+		  ports[0]->fence.seqno,
+		  i915_request_completed(ports[0]) ? "!" :
+		  i915_request_started(ports[0]) ? "*" :
+		  "",
+		  ports[1] ? ports[1]->fence.context : 0,
+		  ports[1] ? ports[1]->fence.seqno : 0);
+}
+
+static __maybe_unused bool
+assert_pending_valid(const struct intel_engine_execlists *execlists,
+		     const char *msg)
+{
+	struct i915_request * const *port, *rq;
+	struct intel_context *ce = NULL;
+
+	trace_ports(execlists, msg, execlists->pending);
+
+	if (execlists->pending[execlists_num_ports(execlists)])
+		return false;
+
+	for (port = execlists->pending; (rq = *port); port++) {
+		if (ce == rq->hw_context)
+			return false;
+
+		ce = rq->hw_context;
+	}
+
+	return ce;
+}
+
 static void execlists_submit_ports(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists *execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
 	unsigned int n;
 
+	GEM_BUG_ON(!assert_pending_valid(execlists, "submit"));
+
 	/*
 	 * We can skip acquiring intel_runtime_pm_get() here as it was taken
 	 * on our behalf by the request (see i915_gem_mark_busy()) and it will
@@ -641,38 +700,16 @@ static void execlists_submit_ports(struct intel_engine_cs *engine)
 	 * of elsq entries, keep this in mind before changing the loop below.
 	 */
 	for (n = execlists_num_ports(execlists); n--; ) {
-		struct i915_request *rq;
-		unsigned int count;
-		u64 desc;
-
-		rq = port_unpack(&port[n], &count);
-		if (rq) {
-			GEM_BUG_ON(count > !n);
-			if (!count++)
-				execlists_context_schedule_in(rq);
-			port_set(&port[n], port_pack(rq, count));
-			desc = execlists_update_context(rq);
-			GEM_DEBUG_EXEC(port[n].context_id = upper_32_bits(desc));
-
-			GEM_TRACE("%s in[%d]:  ctx=%d.%d, fence %llx:%lld (current %d), prio=%d\n",
-				  engine->name, n,
-				  port[n].context_id, count,
-				  rq->fence.context, rq->fence.seqno,
-				  hwsp_seqno(rq),
-				  rq_prio(rq));
-		} else {
-			GEM_BUG_ON(!n);
-			desc = 0;
-		}
+		struct i915_request *rq = execlists->pending[n];
 
-		write_desc(execlists, desc, n);
+		write_desc(execlists,
+			   rq ? execlists_update_context(rq) : 0,
+			   n);
 	}
 
 	/* we need to manually load the submit queue */
 	if (execlists->ctrl_reg)
 		writel(EL_CTRL_LOAD, execlists->ctrl_reg);
-
-	execlists_clear_active(execlists, EXECLISTS_ACTIVE_HWACK);
 }
 
 static bool ctx_single_port_submission(const struct intel_context *ce)
@@ -696,6 +733,7 @@ static bool can_merge_ctx(const struct intel_context *prev,
 static bool can_merge_rq(const struct i915_request *prev,
 			 const struct i915_request *next)
 {
+	GEM_BUG_ON(prev == next);
 	GEM_BUG_ON(!assert_priority_queue(prev, next));
 
 	if (!can_merge_ctx(prev->hw_context, next->hw_context))
@@ -704,58 +742,6 @@ static bool can_merge_rq(const struct i915_request *prev,
 	return true;
 }
 
-static void port_assign(struct execlist_port *port, struct i915_request *rq)
-{
-	GEM_BUG_ON(rq == port_request(port));
-
-	if (port_isset(port))
-		i915_request_put(port_request(port));
-
-	port_set(port, port_pack(i915_request_get(rq), port_count(port)));
-}
-
-static void inject_preempt_context(struct intel_engine_cs *engine)
-{
-	struct intel_engine_execlists *execlists = &engine->execlists;
-	struct intel_context *ce = engine->preempt_context;
-	unsigned int n;
-
-	GEM_BUG_ON(execlists->preempt_complete_status !=
-		   upper_32_bits(ce->lrc_desc));
-
-	/*
-	 * Switch to our empty preempt context so
-	 * the state of the GPU is known (idle).
-	 */
-	GEM_TRACE("%s\n", engine->name);
-	for (n = execlists_num_ports(execlists); --n; )
-		write_desc(execlists, 0, n);
-
-	write_desc(execlists, ce->lrc_desc, n);
-
-	/* we need to manually load the submit queue */
-	if (execlists->ctrl_reg)
-		writel(EL_CTRL_LOAD, execlists->ctrl_reg);
-
-	execlists_clear_active(execlists, EXECLISTS_ACTIVE_HWACK);
-	execlists_set_active(execlists, EXECLISTS_ACTIVE_PREEMPT);
-
-	(void)I915_SELFTEST_ONLY(execlists->preempt_hang.count++);
-}
-
-static void complete_preempt_context(struct intel_engine_execlists *execlists)
-{
-	GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT));
-
-	if (inject_preempt_hang(execlists))
-		return;
-
-	execlists_cancel_port_requests(execlists);
-	__unwind_incomplete_requests(container_of(execlists,
-						  struct intel_engine_cs,
-						  execlists));
-}
-
 static void virtual_update_register_offsets(u32 *regs,
 					    struct intel_engine_cs *engine)
 {
@@ -818,20 +804,30 @@ static bool virtual_matches(const struct virtual_engine *ve,
 	 * we reuse the register offsets). This is a very small
 	 * hystersis on the greedy seelction algorithm.
 	 */
-	inflight = READ_ONCE(ve->context.inflight);
+	inflight = intel_context_inflight(&ve->context);
 	if (inflight && inflight != engine)
 		return false;
 
 	return true;
 }
 
-static void execlists_dequeue(struct intel_engine_cs *engine)
+static struct i915_request *
+last_active(const struct intel_engine_execlists *execlists)
+{
+	struct i915_request * const *last = execlists->active;
+
+	while (*last && i915_request_completed(*last))
+		last++;
+
+	return *last;
+}
+
+static bool execlists_dequeue(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
-	const struct execlist_port * const last_port =
-		&execlists->port[execlists->port_mask];
-	struct i915_request *last = port_request(port);
+	struct i915_request **port = execlists->pending;
+	struct i915_request ** const last_port = port + execlists->port_mask;
+	struct i915_request *last;
 	struct rb_node *rb;
 	bool submit = false;
 
@@ -877,65 +873,75 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		break;
 	}
 
+	/*
+	 * If the queue is higher priority than the last
+	 * request in the currently active context, submit afresh.
+	 * We will resubmit again afterwards in case we need to split
+	 * the active context to interject the preemption request,
+	 * i.e. we will retrigger preemption following the ack in case
+	 * of trouble.
+	 */
+	last = last_active(execlists);
 	if (last) {
-		/*
-		 * Don't resubmit or switch until all outstanding
-		 * preemptions (lite-restore) are seen. Then we
-		 * know the next preemption status we see corresponds
-		 * to this ELSP update.
-		 */
-		GEM_BUG_ON(!execlists_is_active(execlists,
-						EXECLISTS_ACTIVE_USER));
-		GEM_BUG_ON(!port_count(&port[0]));
+		if (need_preempt(engine, last, rb)) {
+			GEM_TRACE("%s: preempting last=%llx:%lld, prio=%d, hint=%d\n",
+				  engine->name,
+				  last->fence.context,
+				  last->fence.seqno,
+				  last->sched.attr.priority,
+				  execlists->queue_priority_hint);
+			/*
+			 * Don't let the RING_HEAD advance past the breadcrumb
+			 * as we unwind (and until we resubmit) so that we do
+			 * not accidentally tell it to go backwards.
+			 */
+			ring_suspend(engine) = 1;
 
-		/*
-		 * If we write to ELSP a second time before the HW has had
-		 * a chance to respond to the previous write, we can confuse
-		 * the HW and hit "undefined behaviour". After writing to ELSP,
-		 * we must then wait until we see a context-switch event from
-		 * the HW to indicate that it has had a chance to respond.
-		 */
-		if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_HWACK))
-			return;
+			/*
+			 * If we need to return to the preempted context, we
+			 * need to skip the lite-restore and force it to
+			 * reload the RING_TAIL. Otherwise, the HW has a
+			 * tendency to ignore us rewinding the TAIL to the
+			 * end of an earlier request.
+			 */
+			last->hw_context->lrc_desc |= CTX_DESC_FORCE_RESTORE;
 
-		if (need_preempt(engine, last, rb)) {
-			inject_preempt_context(engine);
-			return;
-		}
+			/*
+			 * Note that we have not stopped the GPU at this point,
+			 * so we are unwinding the incomplete requests as they
+			 * remain inflight and so by the time we do complete
+			 * the preemption, some of the unwound requests may
+			 * complete!
+			 */
+			__unwind_incomplete_requests(engine);
+			last = NULL;
+		} else {
+			/*
+			 * Otherwise if we already have a request pending
+			 * for execution after the current one, we can
+			 * just wait until the next CS event before
+			 * queuing more. In either case we will force a
+			 * lite-restore preemption event, but if we wait
+			 * we hopefully coalesce several updates into a single
+			 * submission.
+			 */
+			if (!list_is_last(&last->sched.link,
+					  &engine->active.requests))
+				return false;
 
-		/*
-		 * In theory, we could coalesce more requests onto
-		 * the second port (the first port is active, with
-		 * no preemptions pending). However, that means we
-		 * then have to deal with the possible lite-restore
-		 * of the second port (as we submit the ELSP, there
-		 * may be a context-switch) but also we may complete
-		 * the resubmission before the context-switch. Ergo,
-		 * coalescing onto the second port will cause a
-		 * preemption event, but we cannot predict whether
-		 * that will affect port[0] or port[1].
-		 *
-		 * If the second port is already active, we can wait
-		 * until the next context-switch before contemplating
-		 * new requests. The GPU will be busy and we should be
-		 * able to resubmit the new ELSP before it idles,
-		 * avoiding pipeline bubbles (momentary pauses where
-		 * the driver is unable to keep up the supply of new
-		 * work). However, we have to double check that the
-		 * priorities of the ports haven't been switch.
-		 */
-		if (port_count(&port[1]))
-			return;
+			if (i915_request_started(last))
+				return false;
 
-		/*
-		 * WaIdleLiteRestore:bdw,skl
-		 * Apply the wa NOOPs to prevent
-		 * ring:HEAD == rq:TAIL as we resubmit the
-		 * request. See gen8_emit_fini_breadcrumb() for
-		 * where we prepare the padding after the
-		 * end of the request.
-		 */
-		last->tail = last->wa_tail;
+			/*
+			 * WaIdleLiteRestore:bdw,skl
+			 * Apply the wa NOOPs to prevent
+			 * ring:HEAD == rq:TAIL as we resubmit the
+			 * request. See gen8_emit_fini_breadcrumb() for
+			 * where we prepare the padding after the
+			 * end of the request.
+			 */
+			last->tail = last->wa_tail;
+		}
 	}
 
 	while (rb) { /* XXX virtual is always taking precedence */
@@ -965,9 +971,24 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 				continue;
 			}
 
+			if (i915_request_completed(rq)) {
+				ve->request = NULL;
+				ve->base.execlists.queue_priority_hint = INT_MIN;
+				rb_erase_cached(rb, &execlists->virtual);
+				RB_CLEAR_NODE(rb);
+
+				rq->engine = engine;
+				__i915_request_submit(rq);
+
+				spin_unlock(&ve->base.active.lock);
+
+				rb = rb_first_cached(&execlists->virtual);
+				continue;
+			}
+
 			if (last && !can_merge_rq(last, rq)) {
 				spin_unlock(&ve->base.active.lock);
-				return; /* leave this rq for another engine */
+				return false; /* leave this for another */
 			}
 
 			GEM_TRACE("%s: virtual rq=%llx:%lld%s, new engine? %s\n",
@@ -1013,9 +1034,10 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 			}
 
 			__i915_request_submit(rq);
-			trace_i915_request_in(rq, port_index(port, execlists));
-			submit = true;
-			last = rq;
+			if (!i915_request_completed(rq)) {
+				submit = true;
+				last = rq;
+			}
 		}
 
 		spin_unlock(&ve->base.active.lock);
@@ -1028,6 +1050,9 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		int i;
 
 		priolist_for_each_request_consume(rq, rn, p, i) {
+			if (i915_request_completed(rq))
+				goto skip;
+
 			/*
 			 * Can we combine this request with the current port?
 			 * It has to be the same context/ringbuffer and not
@@ -1067,19 +1092,14 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 				    ctx_single_port_submission(rq->hw_context))
 					goto done;
 
-
-				if (submit)
-					port_assign(port, last);
+				*port = execlists_schedule_in(last, port - execlists->pending);
 				port++;
-
-				GEM_BUG_ON(port_isset(port));
 			}
 
-			__i915_request_submit(rq);
-			trace_i915_request_in(rq, port_index(port, execlists));
-
 			last = rq;
 			submit = true;
+skip:
+			__i915_request_submit(rq);
 		}
 
 		rb_erase_cached(&p->node, &execlists->queue);
@@ -1104,54 +1124,32 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 	 * interrupt for secondary ports).
 	 */
 	execlists->queue_priority_hint = queue_prio(execlists);
+	GEM_TRACE("%s: queue_priority_hint:%d, submit:%s\n",
+		  engine->name, execlists->queue_priority_hint,
+		  yesno(submit));
 
 	if (submit) {
-		port_assign(port, last);
+		*port = execlists_schedule_in(last, port - execlists->pending);
+		memset(port + 1, 0, (last_port - port) * sizeof(*port));
 		execlists_submit_ports(engine);
 	}
 
-	/* We must always keep the beast fed if we have work piled up */
-	GEM_BUG_ON(rb_first_cached(&execlists->queue) &&
-		   !port_isset(execlists->port));
-
-	/* Re-evaluate the executing context setup after each preemptive kick */
-	if (last)
-		execlists_user_begin(execlists, execlists->port);
-
-	/* If the engine is now idle, so should be the flag; and vice versa. */
-	GEM_BUG_ON(execlists_is_active(&engine->execlists,
-				       EXECLISTS_ACTIVE_USER) ==
-		   !port_isset(engine->execlists.port));
+	return submit;
 }
 
 void
 execlists_cancel_port_requests(struct intel_engine_execlists * const execlists)
 {
-	struct execlist_port *port = execlists->port;
-	unsigned int num_ports = execlists_num_ports(execlists);
-
-	while (num_ports-- && port_isset(port)) {
-		struct i915_request *rq = port_request(port);
-
-		GEM_TRACE("%s:port%u fence %llx:%lld, (current %d)\n",
-			  rq->engine->name,
-			  (unsigned int)(port - execlists->port),
-			  rq->fence.context, rq->fence.seqno,
-			  hwsp_seqno(rq));
-
-		GEM_BUG_ON(!execlists->active);
-		execlists_context_schedule_out(rq,
-					       i915_request_completed(rq) ?
-					       INTEL_CONTEXT_SCHEDULE_OUT :
-					       INTEL_CONTEXT_SCHEDULE_PREEMPTED);
-
-		i915_request_put(rq);
+	struct i915_request * const *port, *rq;
 
-		memset(port, 0, sizeof(*port));
-		port++;
-	}
+	for (port = execlists->pending; (rq = *port); port++)
+		execlists_schedule_out(rq);
+	memset(execlists->pending, 0, sizeof(execlists->pending));
 
-	execlists_clear_all_active(execlists);
+	for (port = execlists->active; (rq = *port); port++)
+		execlists_schedule_out(rq);
+	execlists->active =
+		memset(execlists->inflight, 0, sizeof(execlists->inflight));
 }
 
 static inline void
@@ -1170,7 +1168,6 @@ reset_in_progress(const struct intel_engine_execlists *execlists)
 static void process_csb(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
 	const u32 * const buf = execlists->csb_status;
 	const u8 num_entries = execlists->csb_size;
 	u8 head, tail;
@@ -1204,9 +1201,7 @@ static void process_csb(struct intel_engine_cs *engine)
 	rmb();
 
 	do {
-		struct i915_request *rq;
 		unsigned int status;
-		unsigned int count;
 
 		if (++head == num_entries)
 			head = 0;
@@ -1229,68 +1224,37 @@ static void process_csb(struct intel_engine_cs *engine)
 		 * status notifier.
 		 */
 
-		GEM_TRACE("%s csb[%d]: status=0x%08x:0x%08x, active=0x%x\n",
+		GEM_TRACE("%s csb[%d]: status=0x%08x:0x%08x\n",
 			  engine->name, head,
-			  buf[2 * head + 0], buf[2 * head + 1],
-			  execlists->active);
+			  buf[2 * head + 0], buf[2 * head + 1]);
 
 		status = buf[2 * head];
-		if (status & (GEN8_CTX_STATUS_IDLE_ACTIVE |
-			      GEN8_CTX_STATUS_PREEMPTED))
-			execlists_set_active(execlists,
-					     EXECLISTS_ACTIVE_HWACK);
-		if (status & GEN8_CTX_STATUS_ACTIVE_IDLE)
-			execlists_clear_active(execlists,
-					       EXECLISTS_ACTIVE_HWACK);
-
-		if (!(status & GEN8_CTX_STATUS_COMPLETED_MASK))
-			continue;
+		if (status & GEN8_CTX_STATUS_IDLE_ACTIVE) {
+promote:
+			GEM_BUG_ON(!assert_pending_valid(execlists, "promote"));
+			execlists->active =
+				memcpy(execlists->inflight,
+				       execlists->pending,
+				       execlists_num_ports(execlists) *
+				       sizeof(*execlists->pending));
+			execlists->pending[0] = NULL;
 
-		/* We should never get a COMPLETED | IDLE_ACTIVE! */
-		GEM_BUG_ON(status & GEN8_CTX_STATUS_IDLE_ACTIVE);
+			if (!inject_preempt_hang(execlists))
+				ring_suspend(engine) = 0;
+		} else if (status & GEN8_CTX_STATUS_PREEMPTED) {
+			struct i915_request * const *port = execlists->active;
 
-		if (status & GEN8_CTX_STATUS_COMPLETE &&
-		    buf[2*head + 1] == execlists->preempt_complete_status) {
-			GEM_TRACE("%s preempt-idle\n", engine->name);
-			complete_preempt_context(execlists);
-			continue;
-		}
+			trace_ports(execlists, "preempted", execlists->active);
 
-		if (status & GEN8_CTX_STATUS_PREEMPTED &&
-		    execlists_is_active(execlists,
-					EXECLISTS_ACTIVE_PREEMPT))
-			continue;
+			while (*port)
+				execlists_schedule_out(*port++);
 
-		GEM_BUG_ON(!execlists_is_active(execlists,
-						EXECLISTS_ACTIVE_USER));
+			goto promote;
+		} else if (*execlists->active) {
+			struct i915_request *rq = *execlists->active++;
 
-		rq = port_unpack(port, &count);
-		GEM_TRACE("%s out[0]: ctx=%d.%d, fence %llx:%lld (current %d), prio=%d\n",
-			  engine->name,
-			  port->context_id, count,
-			  rq ? rq->fence.context : 0,
-			  rq ? rq->fence.seqno : 0,
-			  rq ? hwsp_seqno(rq) : 0,
-			  rq ? rq_prio(rq) : 0);
-
-		/* Check the context/desc id for this event matches */
-		GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id);
-
-		GEM_BUG_ON(count == 0);
-		if (--count == 0) {
-			/*
-			 * On the final event corresponding to the
-			 * submission of this context, we expect either
-			 * an element-switch event or a completion
-			 * event (and on completion, the active-idle
-			 * marker). No more preemptions, lite-restore
-			 * or otherwise.
-			 */
-			GEM_BUG_ON(status & GEN8_CTX_STATUS_PREEMPTED);
-			GEM_BUG_ON(port_isset(&port[1]) &&
-				   !(status & GEN8_CTX_STATUS_ELEMENT_SWITCH));
-			GEM_BUG_ON(!port_isset(&port[1]) &&
-				   !(status & GEN8_CTX_STATUS_ACTIVE_IDLE));
+			trace_ports(execlists, "completed",
+				    execlists->active - 1);
 
 			/*
 			 * We rely on the hardware being strongly
@@ -1299,21 +1263,10 @@ static void process_csb(struct intel_engine_cs *engine)
 			 * user interrupt and CSB is processed.
 			 */
 			GEM_BUG_ON(!i915_request_completed(rq));
+			execlists_schedule_out(rq);
 
-			execlists_context_schedule_out(rq,
-						       INTEL_CONTEXT_SCHEDULE_OUT);
-			i915_request_put(rq);
-
-			GEM_TRACE("%s completed ctx=%d\n",
-				  engine->name, port->context_id);
-
-			port = execlists_port_complete(execlists, port);
-			if (port_isset(port))
-				execlists_user_begin(execlists, port);
-			else
-				execlists_user_end(execlists);
-		} else {
-			port_set(port, port_pack(rq, count));
+			GEM_BUG_ON(execlists->active - execlists->inflight >
+				   execlists_num_ports(execlists));
 		}
 	} while (head != tail);
 
@@ -1335,11 +1288,17 @@ static void process_csb(struct intel_engine_cs *engine)
 
 static void __execlists_submission_tasklet(struct intel_engine_cs *const engine)
 {
+	bool submit;
+
 	lockdep_assert_held(&engine->active.lock);
 
-	process_csb(engine);
-	if (!execlists_is_active(&engine->execlists, EXECLISTS_ACTIVE_PREEMPT))
-		execlists_dequeue(engine);
+	do {
+		process_csb(engine);
+
+		submit = false;
+		if (!engine->execlists.pending[0])
+			submit = execlists_dequeue(engine);
+	} while (submit);
 }
 
 /*
@@ -1351,11 +1310,6 @@ static void execlists_submission_tasklet(unsigned long data)
 	struct intel_engine_cs * const engine = (struct intel_engine_cs *)data;
 	unsigned long flags;
 
-	GEM_TRACE("%s awake?=%d, active=%x\n",
-		  engine->name,
-		  !!engine->i915->gt.awake,
-		  engine->execlists.active);
-
 	spin_lock_irqsave(&engine->active.lock, flags);
 	__execlists_submission_tasklet(engine);
 	spin_unlock_irqrestore(&engine->active.lock, flags);
@@ -1381,12 +1335,16 @@ static void __submit_queue_imm(struct intel_engine_cs *engine)
 		tasklet_hi_schedule(&execlists->tasklet);
 }
 
-static void submit_queue(struct intel_engine_cs *engine, int prio)
+static void submit_queue(struct intel_engine_cs *engine,
+			 const struct i915_request *rq)
 {
-	if (prio > engine->execlists.queue_priority_hint) {
-		engine->execlists.queue_priority_hint = prio;
-		__submit_queue_imm(engine);
-	}
+	struct intel_engine_execlists *execlists = &engine->execlists;
+
+	if (rq_prio(rq) <= execlists->queue_priority_hint)
+		return;
+
+	execlists->queue_priority_hint = rq_prio(rq);
+	__submit_queue_imm(engine);
 }
 
 static void execlists_submit_request(struct i915_request *request)
@@ -1402,7 +1360,7 @@ static void execlists_submit_request(struct i915_request *request)
 	GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root));
 	GEM_BUG_ON(list_empty(&request->sched.link));
 
-	submit_queue(engine, rq_prio(request));
+	submit_queue(engine, request);
 
 	spin_unlock_irqrestore(&engine->active.lock, flags);
 }
@@ -2076,10 +2034,13 @@ static bool lrc_regs_ok(const struct i915_request *rq)
 	return true;
 }
 
-static void reset_csb_pointers(struct intel_engine_execlists *execlists)
+static void reset_csb_pointers(struct intel_engine_cs *engine)
 {
+	struct intel_engine_execlists * const execlists = &engine->execlists;
 	const unsigned int reset_value = execlists->csb_size - 1;
 
+	ring_suspend(engine) = 0;
+
 	/*
 	 * After a reset, the HW starts writing into CSB entry [0]. We
 	 * therefore have to set our HEAD pointer back one entry so that
@@ -2107,17 +2068,18 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled)
 	process_csb(engine); /* drain preemption events */
 
 	/* Following the reset, we need to reload the CSB read/write pointers */
-	reset_csb_pointers(&engine->execlists);
+	reset_csb_pointers(engine);
 
 	/*
 	 * Save the currently executing context, even if we completed
 	 * its request, it was still running at the time of the
 	 * reset and will have been clobbered.
 	 */
-	if (!port_isset(execlists->port))
-		goto out_clear;
+	rq = execlists_active(execlists);
+	if (!rq)
+		return;
 
-	ce = port_request(execlists->port)->hw_context;
+	ce = rq->hw_context;
 
 	/*
 	 * Catch up with any missed context-switch interrupts.
@@ -2192,9 +2154,6 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled)
 		rq ? intel_ring_wrap(ce->ring, rq->head) : ce->ring->tail;
 	intel_ring_update_space(ce->ring);
 	__execlists_update_reg_state(ce, engine);
-
-out_clear:
-	execlists_clear_all_active(execlists);
 }
 
 static void execlists_reset(struct intel_engine_cs *engine, bool stalled)
@@ -2290,7 +2249,6 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine)
 
 	execlists->queue_priority_hint = INT_MIN;
 	execlists->queue = RB_ROOT_CACHED;
-	GEM_BUG_ON(port_isset(execlists->port));
 
 	GEM_BUG_ON(__tasklet_is_enabled(&execlists->tasklet));
 	execlists->tasklet.func = nop_submission_tasklet;
@@ -2508,6 +2466,19 @@ static u32 *gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
 	return cs;
 }
 
+static u32 *emit_preempt_busywait(struct i915_request *request, u32 *cs)
+{
+	*cs++ = MI_SEMAPHORE_WAIT |
+		MI_SEMAPHORE_GLOBAL_GTT |
+		MI_SEMAPHORE_POLL |
+		MI_SEMAPHORE_SAD_EQ_SDD;
+	*cs++ = 0;
+	*cs++ = intel_hws_preempt_address(request->engine);
+	*cs++ = 0;
+
+	return cs;
+}
+
 static u32 *gen8_emit_fini_breadcrumb(struct i915_request *request, u32 *cs)
 {
 	cs = gen8_emit_ggtt_write(cs,
@@ -2522,7 +2493,9 @@ static u32 *gen8_emit_fini_breadcrumb(struct i915_request *request, u32 *cs)
 
 
 	*cs++ = MI_USER_INTERRUPT;
+
 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+	cs = emit_preempt_busywait(request, cs);
 
 	request->tail = intel_ring_offset(request, cs);
 	assert_ring_tail_valid(request->ring, request->tail);
@@ -2547,7 +2520,9 @@ static u32 *gen8_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 				      PIPE_CONTROL_STORE_DATA_INDEX);
 
 	*cs++ = MI_USER_INTERRUPT;
+
 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+	cs = emit_preempt_busywait(request, cs);
 
 	request->tail = intel_ring_offset(request, cs);
 	assert_ring_tail_valid(request->ring, request->tail);
@@ -2625,8 +2600,7 @@ void intel_execlists_set_default_submission(struct intel_engine_cs *engine)
 	engine->flags |= I915_ENGINE_SUPPORTS_STATS;
 	if (!intel_vgpu_active(engine->i915))
 		engine->flags |= I915_ENGINE_HAS_SEMAPHORES;
-	if (engine->preempt_context &&
-	    HAS_LOGICAL_RING_PREEMPTION(engine->i915))
+	if (HAS_LOGICAL_RING_PREEMPTION(engine->i915))
 		engine->flags |= I915_ENGINE_HAS_PREEMPTION;
 }
 
@@ -2739,11 +2713,6 @@ int intel_execlists_submission_init(struct intel_engine_cs *engine)
 			i915_mmio_reg_offset(RING_ELSP(base));
 	}
 
-	execlists->preempt_complete_status = ~0u;
-	if (engine->preempt_context)
-		execlists->preempt_complete_status =
-			upper_32_bits(engine->preempt_context->lrc_desc);
-
 	execlists->csb_status =
 		&engine->status_page.addr[I915_HWS_CSB_BUF0_INDEX];
 
@@ -2755,7 +2724,7 @@ int intel_execlists_submission_init(struct intel_engine_cs *engine)
 	else
 		execlists->csb_size = GEN11_CSB_ENTRIES;
 
-	reset_csb_pointers(execlists);
+	reset_csb_pointers(engine);
 
 	return 0;
 }
@@ -2935,11 +2904,6 @@ populate_lr_context(struct intel_context *ce,
 	if (!engine->default_state)
 		regs[CTX_CONTEXT_CONTROL + 1] |=
 			_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
-	if (ce->gem_context == engine->i915->preempt_context &&
-	    INTEL_GEN(engine->i915) < 11)
-		regs[CTX_CONTEXT_CONTROL + 1] |=
-			_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT |
-					   CTX_CTRL_ENGINE_CTX_SAVE_INHIBIT);
 
 	ret = 0;
 err_unpin_ctx:
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 679882046527..1bc893ef2306 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -1238,10 +1238,10 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
 	}
 }
 
-static void record_request(struct i915_request *request,
+static void record_request(const struct i915_request *request,
 			   struct drm_i915_error_request *erq)
 {
-	struct i915_gem_context *ctx = request->gem_context;
+	const struct i915_gem_context *ctx = request->gem_context;
 
 	erq->flags = request->fence.flags;
 	erq->context = request->fence.context;
@@ -1305,20 +1305,15 @@ static void engine_record_requests(struct intel_engine_cs *engine,
 	ee->num_requests = count;
 }
 
-static void error_record_engine_execlists(struct intel_engine_cs *engine,
+static void error_record_engine_execlists(const struct intel_engine_cs *engine,
 					  struct drm_i915_error_engine *ee)
 {
 	const struct intel_engine_execlists * const execlists = &engine->execlists;
-	unsigned int n;
+	struct i915_request * const *port = execlists->active;
+	unsigned int n = 0;
 
-	for (n = 0; n < execlists_num_ports(execlists); n++) {
-		struct i915_request *rq = port_request(&execlists->port[n]);
-
-		if (!rq)
-			break;
-
-		record_request(rq, &ee->execlist[n]);
-	}
+	while (*port)
+		record_request(*port++, &ee->execlist[n++]);
 
 	ee->num_ports = n;
 }
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 34c597372c1e..7cd0a65c61b4 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -232,6 +232,12 @@ static bool i915_request_retire(struct i915_request *rq)
 
 	local_irq_disable();
 
+	/*
+	 * We only loosely track inflight requests across preemption,
+	 * and so we may find ourselves attempting to retire a _completed_
+	 * request that we have removed from the HW and put back on a run
+	 * queue.
+	 */
 	spin_lock(&rq->engine->active.lock);
 	list_del(&rq->sched.link);
 	spin_unlock(&rq->engine->active.lock);
@@ -398,6 +404,7 @@ void __i915_request_submit(struct i915_request *request)
 	set_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags);
 
 	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags) &&
+	    !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &request->fence.flags) &&
 	    !i915_request_enable_breadcrumb(request))
 		intel_engine_queue_breadcrumbs(engine);
 
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index edbbdfec24ab..bebc1e9b4a5e 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -28,6 +28,7 @@
 #include <linux/dma-fence.h>
 #include <linux/lockdep.h>
 
+#include "gt/intel_context_types.h"
 #include "gt/intel_engine_types.h"
 
 #include "i915_gem.h"
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 946b5eb443bf..601aae909491 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -268,16 +268,11 @@ sched_lock_engine(const struct i915_sched_node *node,
 	return locked;
 }
 
-static bool inflight(const struct i915_request *rq,
-		     const struct intel_engine_cs *engine)
+static bool kick_tasklet(const struct intel_engine_cs *engine, int prio)
 {
-	const struct i915_request *active;
+	struct i915_request *inflight = *engine->execlists.active;
 
-	if (!i915_request_is_active(rq))
-		return false;
-
-	active = port_request(engine->execlists.port);
-	return active->hw_context == rq->hw_context;
+	return inflight && prio > inflight->sched.attr.priority;
 }
 
 static void __i915_schedule(struct i915_request *rq,
@@ -413,7 +408,7 @@ static void __i915_schedule(struct i915_request *rq,
 		 * If we are already the currently executing context, don't
 		 * bother evaluating if we should preempt ourselves.
 		 */
-		if (inflight(node_to_request(node), engine))
+		if (!kick_tasklet(engine, prio))
 			continue;
 
 		/* Defer (tasklet) submission until after all of our updates. */
diff --git a/drivers/gpu/drm/i915/i915_utils.h b/drivers/gpu/drm/i915/i915_utils.h
index 1436fe2fb5f8..27ec164149c3 100644
--- a/drivers/gpu/drm/i915/i915_utils.h
+++ b/drivers/gpu/drm/i915/i915_utils.h
@@ -27,6 +27,7 @@
 
 #include <linux/kernel.h>
 #include <linux/overflow.h>
+#include <linux/workqueue.h>
 
 #undef WARN_ON
 /* Many gcc seem to no see through this and fall over :( */
@@ -128,6 +129,18 @@ __check_struct_size(size_t base, size_t arr, size_t count, size_t *size)
 	((typeof(ptr))((unsigned long)(ptr) | __bits));			\
 })
 
+#define ptr_count_dec(p_ptr) do {					\
+	typeof(p_ptr) __p = (p_ptr);					\
+	unsigned long __v = (unsigned long)(*__p);			\
+	*__p = (typeof(*p_ptr))(--__v);					\
+} while (0)
+
+#define ptr_count_inc(p_ptr) do {					\
+	typeof(p_ptr) __p = (p_ptr);					\
+	unsigned long __v = (unsigned long)(*__p);			\
+	*__p = (typeof(*p_ptr))(++__v);					\
+} while (0)
+
 #define page_mask_bits(ptr) ptr_mask_bits(ptr, PAGE_SHIFT)
 #define page_unmask_bits(ptr) ptr_unmask_bits(ptr, PAGE_SHIFT)
 #define page_pack_bits(ptr, bits) ptr_pack_bits(ptr, bits, PAGE_SHIFT)
diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index 0b10568f49ce..ccf54a5d18a1 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -32,7 +32,11 @@
 #include "intel_guc_submission.h"
 #include "i915_drv.h"
 
-#define GUC_PREEMPT_FINISHED		0x1
+enum {
+	GUC_PREEMPT_NONE = 0,
+	GUC_PREEMPT_INPROGRESS,
+	GUC_PREEMPT_FINISHED,
+};
 #define GUC_PREEMPT_BREADCRUMB_DWORDS	0x8
 #define GUC_PREEMPT_BREADCRUMB_BYTES	\
 	(sizeof(u32) * GUC_PREEMPT_BREADCRUMB_DWORDS)
@@ -537,15 +541,11 @@ static void guc_add_request(struct intel_guc *guc, struct i915_request *rq)
 	u32 ctx_desc = lower_32_bits(rq->hw_context->lrc_desc);
 	u32 ring_tail = intel_ring_set_tail(rq->ring, rq->tail) / sizeof(u64);
 
-	spin_lock(&client->wq_lock);
-
 	guc_wq_item_append(client, engine->guc_id, ctx_desc,
 			   ring_tail, rq->fence.seqno);
 	guc_ring_doorbell(client);
 
 	client->submissions[engine->id] += 1;
-
-	spin_unlock(&client->wq_lock);
 }
 
 /*
@@ -631,8 +631,9 @@ static void inject_preempt_context(struct work_struct *work)
 	data[6] = intel_guc_ggtt_offset(guc, guc->shared_data);
 
 	if (WARN_ON(intel_guc_send(guc, data, ARRAY_SIZE(data)))) {
-		execlists_clear_active(&engine->execlists,
-				       EXECLISTS_ACTIVE_PREEMPT);
+		intel_write_status_page(engine,
+					I915_GEM_HWS_PREEMPT,
+					GUC_PREEMPT_NONE);
 		tasklet_schedule(&engine->execlists.tasklet);
 	}
 
@@ -672,8 +673,6 @@ static void complete_preempt_context(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists *execlists = &engine->execlists;
 
-	GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT));
-
 	if (inject_preempt_hang(execlists))
 		return;
 
@@ -681,88 +680,90 @@ static void complete_preempt_context(struct intel_engine_cs *engine)
 	execlists_unwind_incomplete_requests(execlists);
 
 	wait_for_guc_preempt_report(engine);
-	intel_write_status_page(engine, I915_GEM_HWS_PREEMPT, 0);
+	intel_write_status_page(engine, I915_GEM_HWS_PREEMPT, GUC_PREEMPT_NONE);
 }
 
-/**
- * guc_submit() - Submit commands through GuC
- * @engine: engine associated with the commands
- *
- * The only error here arises if the doorbell hardware isn't functioning
- * as expected, which really shouln't happen.
- */
-static void guc_submit(struct intel_engine_cs *engine)
+static void guc_submit(struct intel_engine_cs *engine,
+		       struct i915_request **out,
+		       struct i915_request **end)
 {
 	struct intel_guc *guc = &engine->i915->guc;
-	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
-	unsigned int n;
+	struct intel_guc_client *client = guc->execbuf_client;
 
-	for (n = 0; n < execlists_num_ports(execlists); n++) {
-		struct i915_request *rq;
-		unsigned int count;
+	spin_lock(&client->wq_lock);
 
-		rq = port_unpack(&port[n], &count);
-		if (rq && count == 0) {
-			port_set(&port[n], port_pack(rq, ++count));
+	do {
+		struct i915_request *rq = *out++;
 
-			flush_ggtt_writes(rq->ring->vma);
+		flush_ggtt_writes(rq->ring->vma);
+		guc_add_request(guc, rq);
+	} while (out != end);
 
-			guc_add_request(guc, rq);
-		}
-	}
+	spin_unlock(&client->wq_lock);
 }
 
-static void port_assign(struct execlist_port *port, struct i915_request *rq)
+static inline int rq_prio(const struct i915_request *rq)
 {
-	GEM_BUG_ON(port_isset(port));
-
-	port_set(port, i915_request_get(rq));
+	return rq->sched.attr.priority | __NO_PREEMPTION;
 }
 
-static inline int rq_prio(const struct i915_request *rq)
+static struct i915_request *schedule_in(struct i915_request *rq, int idx)
 {
-	return rq->sched.attr.priority;
+	trace_i915_request_in(rq, idx);
+
+	if (!rq->hw_context->inflight)
+		rq->hw_context->inflight = rq->engine;
+	intel_context_inflight_inc(rq->hw_context);
+
+	return i915_request_get(rq);
 }
 
-static inline int port_prio(const struct execlist_port *port)
+static void schedule_out(struct i915_request *rq)
 {
-	return rq_prio(port_request(port)) | __NO_PREEMPTION;
+	trace_i915_request_out(rq);
+
+	intel_context_inflight_dec(rq->hw_context);
+	if (!intel_context_inflight_count(rq->hw_context))
+		rq->hw_context->inflight = NULL;
+
+	i915_request_put(rq);
 }
 
-static bool __guc_dequeue(struct intel_engine_cs *engine)
+static void __guc_dequeue(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
-	struct i915_request *last = NULL;
-	const struct execlist_port * const last_port =
-		&execlists->port[execlists->port_mask];
+	struct i915_request **first = execlists->inflight;
+	struct i915_request ** const last_port = first + execlists->port_mask;
+	struct i915_request *last = first[0];
+	struct i915_request **port;
 	bool submit = false;
 	struct rb_node *rb;
 
 	lockdep_assert_held(&engine->active.lock);
 
-	if (port_isset(port)) {
+	if (last) {
 		if (intel_engine_has_preemption(engine)) {
 			struct guc_preempt_work *preempt_work =
 				&engine->i915->guc.preempt_work[engine->id];
 			int prio = execlists->queue_priority_hint;
 
-			if (__execlists_need_preempt(prio, port_prio(port))) {
-				execlists_set_active(execlists,
-						     EXECLISTS_ACTIVE_PREEMPT);
+			if (__execlists_need_preempt(prio, rq_prio(last))) {
+				intel_write_status_page(engine,
+							I915_GEM_HWS_PREEMPT,
+							GUC_PREEMPT_INPROGRESS);
 				queue_work(engine->i915->guc.preempt_wq,
 					   &preempt_work->work);
-				return false;
+				return;
 			}
 		}
 
-		port++;
-		if (port_isset(port))
-			return false;
+		if (*++first)
+			return;
+
+		last = NULL;
 	}
-	GEM_BUG_ON(port_isset(port));
 
+	port = first;
 	while ((rb = rb_first_cached(&execlists->queue))) {
 		struct i915_priolist *p = to_priolist(rb);
 		struct i915_request *rq, *rn;
@@ -773,18 +774,15 @@ static bool __guc_dequeue(struct intel_engine_cs *engine)
 				if (port == last_port)
 					goto done;
 
-				if (submit)
-					port_assign(port, last);
+				*port = schedule_in(last,
+						    port - execlists->inflight);
 				port++;
 			}
 
 			list_del_init(&rq->sched.link);
-
 			__i915_request_submit(rq);
-			trace_i915_request_in(rq, port_index(port, execlists));
-
-			last = rq;
 			submit = true;
+			last = rq;
 		}
 
 		rb_erase_cached(&p->node, &execlists->queue);
@@ -793,58 +791,41 @@ static bool __guc_dequeue(struct intel_engine_cs *engine)
 done:
 	execlists->queue_priority_hint =
 		rb ? to_priolist(rb)->priority : INT_MIN;
-	if (submit)
-		port_assign(port, last);
-	if (last)
-		execlists_user_begin(execlists, execlists->port);
-
-	/* We must always keep the beast fed if we have work piled up */
-	GEM_BUG_ON(port_isset(execlists->port) &&
-		   !execlists_is_active(execlists, EXECLISTS_ACTIVE_USER));
-	GEM_BUG_ON(rb_first_cached(&execlists->queue) &&
-		   !port_isset(execlists->port));
-
-	return submit;
-}
-
-static void guc_dequeue(struct intel_engine_cs *engine)
-{
-	if (__guc_dequeue(engine))
-		guc_submit(engine);
+	if (submit) {
+		*port = schedule_in(last, port - execlists->inflight);
+		*++port = NULL;
+		guc_submit(engine, first, port);
+	}
+	execlists->active = execlists->inflight;
 }
 
 static void guc_submission_tasklet(unsigned long data)
 {
 	struct intel_engine_cs * const engine = (struct intel_engine_cs *)data;
 	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct execlist_port *port = execlists->port;
-	struct i915_request *rq;
+	struct i915_request **port, *rq;
 	unsigned long flags;
 
 	spin_lock_irqsave(&engine->active.lock, flags);
 
-	rq = port_request(port);
-	while (rq && i915_request_completed(rq)) {
-		trace_i915_request_out(rq);
-		i915_request_put(rq);
+	for (port = execlists->inflight; (rq = *port); port++) {
+		if (!i915_request_completed(rq))
+			break;
 
-		port = execlists_port_complete(execlists, port);
-		if (port_isset(port)) {
-			execlists_user_begin(execlists, port);
-			rq = port_request(port);
-		} else {
-			execlists_user_end(execlists);
-			rq = NULL;
-		}
+		schedule_out(rq);
+	}
+	if (port != execlists->inflight) {
+		int idx = port - execlists->inflight;
+		int rem = ARRAY_SIZE(execlists->inflight) - idx;
+		memmove(execlists->inflight, port, rem * sizeof(rq));
 	}
 
-	if (execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT) &&
-	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT) ==
+	if (intel_read_status_page(engine, I915_GEM_HWS_PREEMPT) ==
 	    GUC_PREEMPT_FINISHED)
 		complete_preempt_context(engine);
 
-	if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT))
-		guc_dequeue(engine);
+	if (!intel_read_status_page(engine, I915_GEM_HWS_PREEMPT))
+		__guc_dequeue(engine);
 
 	spin_unlock_irqrestore(&engine->active.lock, flags);
 }
@@ -958,7 +939,6 @@ static void guc_cancel_requests(struct intel_engine_cs *engine)
 
 	execlists->queue_priority_hint = INT_MIN;
 	execlists->queue = RB_ROOT_CACHED;
-	GEM_BUG_ON(port_isset(execlists->port));
 
 	spin_unlock_irqrestore(&engine->active.lock, flags);
 }
@@ -1420,7 +1400,7 @@ int intel_guc_submission_enable(struct intel_guc *guc)
 	 * 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->execlists.port) *
+	BUILD_BUG_ON(ARRAY_SIZE(engine->execlists.inflight) *
 		     sizeof(struct guc_wq_item) *
 		     I915_NUM_ENGINES > GUC_WQ_SIZE);
 
diff --git a/drivers/gpu/drm/i915/selftests/i915_request.c b/drivers/gpu/drm/i915/selftests/i915_request.c
index 693ea097c05f..218671b250d2 100644
--- a/drivers/gpu/drm/i915/selftests/i915_request.c
+++ b/drivers/gpu/drm/i915/selftests/i915_request.c
@@ -367,13 +367,15 @@ static int __igt_breadcrumbs_smoketest(void *arg)
 
 		if (!wait_event_timeout(wait->wait,
 					i915_sw_fence_done(wait),
-					HZ / 2)) {
+					5 * HZ)) {
 			struct i915_request *rq = requests[count - 1];
 
-			pr_err("waiting for %d fences (last %llx:%lld) on %s timed out!\n",
-			       count,
+			pr_err("waiting for %d/%d fences (last %llx:%lld) on %s timed out!\n",
+			       atomic_read(&wait->pending), count,
 			       rq->fence.context, rq->fence.seqno,
 			       t->engine->name);
+			GEM_TRACE_DUMP();
+
 			i915_gem_set_wedged(t->engine->i915);
 			GEM_BUG_ON(!i915_request_completed(rq));
 			i915_sw_fence_wait(wait);
-- 
2.20.1



More information about the Intel-gfx-trybot mailing list