[PATCH 01/33] drm/i915/execlists: Move RPCS setup to context pin

[PATCH 01/33] drm/i915/execlists: Move RPCS setup to context pin

[Chris Wilson]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Configuring RPCS in context image just before pin is sufficient and will
come extra handy in one of the following patches.

v2:
 * Split image setup a bit differently. (Chris Wilson)

v3:
 * Update context image after reset as well - otherwise the application
   of pinned default state clears the RPCS.

v4:
 * Use local variable throughout the function. (Chris Wilson)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/i915/intel_lrc.c | 45 ++++++++++++++++++++------------
 1 file changed, 28 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 8aa8a4862543..9155cc675924 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1173,6 +1173,24 @@ static int __context_pin(struct i915_gem_context *ctx, struct i915_vma *vma)
 	return i915_vma_pin(vma, 0, 0, flags);
 }
 
+static u32 make_rpcs(struct drm_i915_private *dev_priv);
+
+static void
+__execlists_update_reg_state(struct intel_engine_cs *engine,
+			     struct intel_context *ce)
+{
+	u32 *regs = ce->lrc_reg_state;
+	struct intel_ring *ring = ce->ring;
+
+	regs[CTX_RING_BUFFER_START + 1] = i915_ggtt_offset(ring->vma);
+	regs[CTX_RING_HEAD + 1] = ring->head;
+	regs[CTX_RING_TAIL + 1] = ring->tail;
+
+	/* RPCS */
+	if (engine->class == RENDER_CLASS)
+		regs[CTX_R_PWR_CLK_STATE + 1] = make_rpcs(engine->i915);
+}
+
 static struct intel_context *
 __execlists_context_pin(struct intel_engine_cs *engine,
 			struct i915_gem_context *ctx,
@@ -1211,10 +1229,8 @@ __execlists_context_pin(struct intel_engine_cs *engine,
 	GEM_BUG_ON(!intel_ring_offset_valid(ce->ring, ce->ring->head));
 
 	ce->lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE;
-	ce->lrc_reg_state[CTX_RING_BUFFER_START+1] =
-		i915_ggtt_offset(ce->ring->vma);
-	ce->lrc_reg_state[CTX_RING_HEAD + 1] = ce->ring->head;
-	ce->lrc_reg_state[CTX_RING_TAIL + 1] = ce->ring->tail;
+
+	__execlists_update_reg_state(engine, ce);
 
 	ce->state->obj->pin_global++;
 	i915_gem_context_get(ctx);
@@ -1838,14 +1854,14 @@ static void execlists_reset(struct intel_engine_cs *engine,
 		       engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE,
 		       engine->context_size - PAGE_SIZE);
 	}
-	execlists_init_reg_state(regs,
-				 request->gem_context, engine, request->ring);
 
 	/* Move the RING_HEAD onto the breadcrumb, past the hanging batch */
-	regs[CTX_RING_BUFFER_START + 1] = i915_ggtt_offset(request->ring->vma);
-
 	request->ring->head = intel_ring_wrap(request->ring, request->postfix);
-	regs[CTX_RING_HEAD + 1] = request->ring->head;
+
+	execlists_init_reg_state(regs, request->gem_context, engine,
+				 request->ring);
+
+	__execlists_update_reg_state(engine, request->hw_context);
 
 	intel_ring_update_space(request->ring);
 
@@ -2534,8 +2550,7 @@ static void execlists_init_reg_state(u32 *regs,
 
 	if (rcs) {
 		regs[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1);
-		CTX_REG(regs, CTX_R_PWR_CLK_STATE, GEN8_R_PWR_CLK_STATE,
-			make_rpcs(dev_priv));
+		CTX_REG(regs, CTX_R_PWR_CLK_STATE, GEN8_R_PWR_CLK_STATE, 0);
 
 		i915_oa_init_reg_state(engine, ctx, regs);
 	}
@@ -2696,12 +2711,8 @@ void intel_lr_context_resume(struct drm_i915_private *i915)
 
 			intel_ring_reset(ce->ring, 0);
 
-			if (ce->pin_count) { /* otherwise done in context_pin */
-				u32 *regs = ce->lrc_reg_state;
-
-				regs[CTX_RING_HEAD + 1] = ce->ring->head;
-				regs[CTX_RING_TAIL + 1] = ce->ring->tail;
-			}
+			if (ce->pin_count) /* otherwise done in context_pin */
+				__execlists_update_reg_state(engine, ce);
 		}
 	}
 }
-- 
2.20.1

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

[PATCH 02/33] drm/i915: Measure the required reserved size for request emission

[Chris Wilson]

Instead of tediously and fragilely counting up the number of dwords
required to emit the breadcrumb to seal a request, fake a request and
measure it automatically once during engine setup.

The downside is that this requires a fair amount of mocking to create a
proper breadcrumb. Still, should be less error prone in future as the
breadcrumb size fluctuates!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_engine_cs.c       | 49 ++++++++++++++++++++
 drivers/gpu/drm/i915/intel_lrc.c             | 12 +++--
 drivers/gpu/drm/i915/intel_ringbuffer.c      | 24 +++++++---
 drivers/gpu/drm/i915/intel_ringbuffer.h      |  2 +-
 drivers/gpu/drm/i915/selftests/mock_engine.c |  4 +-
 5 files changed, 77 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 2f3c71f6d313..883ba208d1c2 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -604,6 +604,47 @@ static void __intel_context_unpin(struct i915_gem_context *ctx,
 	intel_context_unpin(to_intel_context(ctx, engine));
 }
 
+struct measure_breadcrumb {
+	struct i915_request rq;
+	struct i915_timeline timeline;
+	struct intel_ring ring;
+	u32 cs[1024];
+};
+
+static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
+{
+	struct measure_breadcrumb *frame;
+	unsigned int dw;
+
+	GEM_BUG_ON(!engine->i915->gt.scratch);
+
+	frame = kzalloc(sizeof(*frame), GFP_KERNEL);
+	if (!frame)
+		return -ENOMEM;
+
+	i915_timeline_init(engine->i915, &frame->timeline, engine->name);
+
+	frame->ring.timeline = &frame->timeline;
+	frame->ring.vaddr = frame->cs;
+	frame->ring.size = sizeof(frame->cs);
+	frame->ring.effective_size = frame->ring.size;
+	frame->ring.space = frame->ring.size - 8;
+	INIT_LIST_HEAD(&frame->ring.request_list);
+
+	frame->rq.i915 = engine->i915;
+	frame->rq.engine = engine;
+	frame->rq.ring = &frame->ring;
+	frame->rq.timeline = &frame->timeline;
+
+	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
+	GEM_BUG_ON(dw != engine->emit_breadcrumb_sz);
+
+	i915_timeline_fini(&frame->timeline);
+	kfree(frame);
+
+	return dw;
+}
+
 /**
  * intel_engines_init_common - initialize cengine state which might require hw access
  * @engine: Engine to initialize.
@@ -657,8 +698,16 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 	if (ret)
 		goto err_breadcrumbs;
 
+	ret = measure_breadcrumb_sz(engine);
+	if (ret < 0)
+		goto err_status_page;
+
+	engine->emit_breadcrumb_sz = ret;
+
 	return 0;
 
+err_status_page:
+	cleanup_status_page(engine);
 err_breadcrumbs:
 	intel_engine_fini_breadcrumbs(engine);
 err_unpin_preempt:
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 9155cc675924..d2299425cf2f 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -2051,15 +2051,17 @@ static int gen8_emit_flush_render(struct i915_request *request,
  * used as a workaround for not being allowed to do lite
  * restore with HEAD==TAIL (WaIdleLiteRestore).
  */
-static void gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
+static u32 *gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
 {
 	/* Ensure there's always at least one preemption point per-request. */
 	*cs++ = MI_ARB_CHECK;
 	*cs++ = MI_NOOP;
 	request->wa_tail = intel_ring_offset(request, cs);
+
+	return cs;
 }
 
-static void gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
+static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 {
 	/* w/a: bit 5 needs to be zero for MI_FLUSH_DW address. */
 	BUILD_BUG_ON(I915_GEM_HWS_INDEX_ADDR & (1 << 5));
@@ -2071,11 +2073,11 @@ static void gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 	request->tail = intel_ring_offset(request, cs);
 	assert_ring_tail_valid(request->ring, request->tail);
 
-	gen8_emit_wa_tail(request, cs);
+	return gen8_emit_wa_tail(request, cs);
 }
 static const int gen8_emit_breadcrumb_sz = 6 + WA_TAIL_DWORDS;
 
-static void gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
+static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
 	/* We're using qword write, seqno should be aligned to 8 bytes. */
 	BUILD_BUG_ON(I915_GEM_HWS_INDEX & 1);
@@ -2095,7 +2097,7 @@ static void gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 	request->tail = intel_ring_offset(request, cs);
 	assert_ring_tail_valid(request->ring, request->tail);
 
-	gen8_emit_wa_tail(request, cs);
+	return gen8_emit_wa_tail(request, cs);
 }
 static const int gen8_emit_breadcrumb_rcs_sz = 8 + WA_TAIL_DWORDS;
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index e39e483d8d16..107c4934e2fa 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -299,7 +299,7 @@ gen6_render_ring_flush(struct i915_request *rq, u32 mode)
 	return 0;
 }
 
-static void gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	/* First we do the gen6_emit_post_sync_nonzero_flush w/a */
 	*cs++ = GFX_OP_PIPE_CONTROL(4);
@@ -327,6 +327,8 @@ static void gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int gen6_rcs_emit_breadcrumb_sz = 14;
 
@@ -409,7 +411,7 @@ gen7_render_ring_flush(struct i915_request *rq, u32 mode)
 	return 0;
 }
 
-static void gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	*cs++ = GFX_OP_PIPE_CONTROL(4);
 	*cs++ = (PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
@@ -427,10 +429,12 @@ static void gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int gen7_rcs_emit_breadcrumb_sz = 6;
 
-static void gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW;
 	*cs++ = intel_hws_seqno_address(rq->engine) | MI_FLUSH_DW_USE_GTT;
@@ -439,11 +443,13 @@ static void gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int gen6_xcs_emit_breadcrumb_sz = 4;
 
 #define GEN7_XCS_WA 32
-static void gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	int i;
 
@@ -466,6 +472,8 @@ static void gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int gen7_xcs_emit_breadcrumb_sz = 8 + GEN7_XCS_WA * 3;
 #undef GEN7_XCS_WA
@@ -861,7 +869,7 @@ static void i9xx_submit_request(struct i915_request *request)
 			intel_ring_set_tail(request->ring, request->tail));
 }
 
-static void i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	*cs++ = MI_FLUSH;
 
@@ -874,11 +882,13 @@ static void i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int i9xx_emit_breadcrumb_sz = 6;
 
 #define GEN5_WA_STORES 8 /* must be at least 1! */
-static void gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
+static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	int i;
 
@@ -895,6 +905,8 @@ static void gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
+
+	return cs;
 }
 static const int gen5_emit_breadcrumb_sz = GEN5_WA_STORES * 3 + 2;
 #undef GEN5_WA_STORES
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index c3ef0f9bf321..479bd53d4ac6 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -470,7 +470,7 @@ struct intel_engine_cs {
 					 unsigned int dispatch_flags);
 #define I915_DISPATCH_SECURE BIT(0)
 #define I915_DISPATCH_PINNED BIT(1)
-	void		(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
+	u32		*(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
 	int		emit_breadcrumb_sz;
 
 	/* Pass the request to the hardware queue (e.g. directly into
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 442ec2aeec81..905318b7ae18 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -159,9 +159,9 @@ static int mock_emit_flush(struct i915_request *request,
 	return 0;
 }
 
-static void mock_emit_breadcrumb(struct i915_request *request,
-				 u32 *flags)
+static u32 *mock_emit_breadcrumb(struct i915_request *request, u32 *cs)
 {
+	return cs;
 }
 
 static void mock_submit_request(struct i915_request *request)
-- 
2.20.1

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

Re: [PATCH 02/33] drm/i915: Measure the required reserved size for request emission

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Instead of tediously and fragilely counting up the number of dwords
> required to emit the breadcrumb to seal a request, fake a request and
> measure it automatically once during engine setup.
>
> The downside is that this requires a fair amount of mocking to create a
> proper breadcrumb. Still, should be less error prone in future as the
> breadcrumb size fluctuates!

We are quick to notice, but this method saves brains and time,
review time.

>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/intel_engine_cs.c       | 49 ++++++++++++++++++++
>  drivers/gpu/drm/i915/intel_lrc.c             | 12 +++--
>  drivers/gpu/drm/i915/intel_ringbuffer.c      | 24 +++++++---
>  drivers/gpu/drm/i915/intel_ringbuffer.h      |  2 +-
>  drivers/gpu/drm/i915/selftests/mock_engine.c |  4 +-
>  5 files changed, 77 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index 2f3c71f6d313..883ba208d1c2 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -604,6 +604,47 @@ static void __intel_context_unpin(struct i915_gem_context *ctx,
>  	intel_context_unpin(to_intel_context(ctx, engine));
>  }
>  
> +struct measure_breadcrumb {
> +	struct i915_request rq;
> +	struct i915_timeline timeline;
> +	struct intel_ring ring;
> +	u32 cs[1024];
> +};
> +
> +static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
> +{
> +	struct measure_breadcrumb *frame;
> +	unsigned int dw;
> +
> +	GEM_BUG_ON(!engine->i915->gt.scratch);
> +
> +	frame = kzalloc(sizeof(*frame), GFP_KERNEL);
> +	if (!frame)
> +		return -ENOMEM;
> +
> +	i915_timeline_init(engine->i915, &frame->timeline, engine->name);

You could init with null name. This is so short lived
and we dont expect debugs. If it ever leaks into wild,
blowout would be instant. Now the name is the same as
the real deal.

> +
> +	frame->ring.timeline = &frame->timeline;
> +	frame->ring.vaddr = frame->cs;
> +	frame->ring.size = sizeof(frame->cs);
> +	frame->ring.effective_size = frame->ring.size;
> +	frame->ring.space = frame->ring.size - 8;

Why 2 dwords short? Just curious as it doesn't seem
to matter in this use case.

> +	INIT_LIST_HEAD(&frame->ring.request_list);
> +
> +	frame->rq.i915 = engine->i915;
> +	frame->rq.engine = engine;
> +	frame->rq.ring = &frame->ring;
> +	frame->rq.timeline = &frame->timeline;
> +
> +	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
> +	GEM_BUG_ON(dw != engine->emit_breadcrumb_sz);

Peace of mind provided =)

Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

> +
> +	i915_timeline_fini(&frame->timeline);
> +	kfree(frame);
> +
> +	return dw;
> +}
> +
>  /**
>   * intel_engines_init_common - initialize cengine state which might require hw access
>   * @engine: Engine to initialize.
> @@ -657,8 +698,16 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
>  	if (ret)
>  		goto err_breadcrumbs;
>  
> +	ret = measure_breadcrumb_sz(engine);
> +	if (ret < 0)
> +		goto err_status_page;
> +
> +	engine->emit_breadcrumb_sz = ret;
> +
>  	return 0;
>  
> +err_status_page:
> +	cleanup_status_page(engine);
>  err_breadcrumbs:
>  	intel_engine_fini_breadcrumbs(engine);
>  err_unpin_preempt:
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index 9155cc675924..d2299425cf2f 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -2051,15 +2051,17 @@ static int gen8_emit_flush_render(struct i915_request *request,
>   * used as a workaround for not being allowed to do lite
>   * restore with HEAD==TAIL (WaIdleLiteRestore).
>   */
> -static void gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
> +static u32 *gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
>  {
>  	/* Ensure there's always at least one preemption point per-request. */
>  	*cs++ = MI_ARB_CHECK;
>  	*cs++ = MI_NOOP;
>  	request->wa_tail = intel_ring_offset(request, cs);
> +
> +	return cs;
>  }
>  
> -static void gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
> +static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
>  {
>  	/* w/a: bit 5 needs to be zero for MI_FLUSH_DW address. */
>  	BUILD_BUG_ON(I915_GEM_HWS_INDEX_ADDR & (1 << 5));
> @@ -2071,11 +2073,11 @@ static void gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
>  	request->tail = intel_ring_offset(request, cs);
>  	assert_ring_tail_valid(request->ring, request->tail);
>  
> -	gen8_emit_wa_tail(request, cs);
> +	return gen8_emit_wa_tail(request, cs);
>  }
>  static const int gen8_emit_breadcrumb_sz = 6 + WA_TAIL_DWORDS;
>  
> -static void gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
> +static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
>  {
>  	/* We're using qword write, seqno should be aligned to 8 bytes. */
>  	BUILD_BUG_ON(I915_GEM_HWS_INDEX & 1);
> @@ -2095,7 +2097,7 @@ static void gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
>  	request->tail = intel_ring_offset(request, cs);
>  	assert_ring_tail_valid(request->ring, request->tail);
>  
> -	gen8_emit_wa_tail(request, cs);
> +	return gen8_emit_wa_tail(request, cs);
>  }
>  static const int gen8_emit_breadcrumb_rcs_sz = 8 + WA_TAIL_DWORDS;
>  
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> index e39e483d8d16..107c4934e2fa 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> @@ -299,7 +299,7 @@ gen6_render_ring_flush(struct i915_request *rq, u32 mode)
>  	return 0;
>  }
>  
> -static void gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	/* First we do the gen6_emit_post_sync_nonzero_flush w/a */
>  	*cs++ = GFX_OP_PIPE_CONTROL(4);
> @@ -327,6 +327,8 @@ static void gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int gen6_rcs_emit_breadcrumb_sz = 14;
>  
> @@ -409,7 +411,7 @@ gen7_render_ring_flush(struct i915_request *rq, u32 mode)
>  	return 0;
>  }
>  
> -static void gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	*cs++ = GFX_OP_PIPE_CONTROL(4);
>  	*cs++ = (PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
> @@ -427,10 +429,12 @@ static void gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int gen7_rcs_emit_breadcrumb_sz = 6;
>  
> -static void gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW;
>  	*cs++ = intel_hws_seqno_address(rq->engine) | MI_FLUSH_DW_USE_GTT;
> @@ -439,11 +443,13 @@ static void gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int gen6_xcs_emit_breadcrumb_sz = 4;
>  
>  #define GEN7_XCS_WA 32
> -static void gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	int i;
>  
> @@ -466,6 +472,8 @@ static void gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int gen7_xcs_emit_breadcrumb_sz = 8 + GEN7_XCS_WA * 3;
>  #undef GEN7_XCS_WA
> @@ -861,7 +869,7 @@ static void i9xx_submit_request(struct i915_request *request)
>  			intel_ring_set_tail(request->ring, request->tail));
>  }
>  
> -static void i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	*cs++ = MI_FLUSH;
>  
> @@ -874,11 +882,13 @@ static void i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int i9xx_emit_breadcrumb_sz = 6;
>  
>  #define GEN5_WA_STORES 8 /* must be at least 1! */
> -static void gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> +static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
>  	int i;
>  
> @@ -895,6 +905,8 @@ static void gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	rq->tail = intel_ring_offset(rq, cs);
>  	assert_ring_tail_valid(rq->ring, rq->tail);
> +
> +	return cs;
>  }
>  static const int gen5_emit_breadcrumb_sz = GEN5_WA_STORES * 3 + 2;
>  #undef GEN5_WA_STORES
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index c3ef0f9bf321..479bd53d4ac6 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -470,7 +470,7 @@ struct intel_engine_cs {
>  					 unsigned int dispatch_flags);
>  #define I915_DISPATCH_SECURE BIT(0)
>  #define I915_DISPATCH_PINNED BIT(1)
> -	void		(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
> +	u32		*(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
>  	int		emit_breadcrumb_sz;
>  
>  	/* Pass the request to the hardware queue (e.g. directly into
> diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
> index 442ec2aeec81..905318b7ae18 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_engine.c
> +++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
> @@ -159,9 +159,9 @@ static int mock_emit_flush(struct i915_request *request,
>  	return 0;
>  }
>  
> -static void mock_emit_breadcrumb(struct i915_request *request,
> -				 u32 *flags)
> +static u32 *mock_emit_breadcrumb(struct i915_request *request, u32 *cs)
>  {
> +	return cs;
>  }
>  
>  static void mock_submit_request(struct i915_request *request)
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 02/33] drm/i915: Measure the required reserved size for request emission

[Chris Wilson]

Quoting Mika Kuoppala (2019-01-25 08:34:37)
> Chris Wilson <chris@chris-wilson.co.uk> writes:
> 
> > Instead of tediously and fragilely counting up the number of dwords
> > required to emit the breadcrumb to seal a request, fake a request and
> > measure it automatically once during engine setup.
> >
> > The downside is that this requires a fair amount of mocking to create a
> > proper breadcrumb. Still, should be less error prone in future as the
> > breadcrumb size fluctuates!
> 
> We are quick to notice, but this method saves brains and time,
> review time.
> 
> >
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > ---
> >  drivers/gpu/drm/i915/intel_engine_cs.c       | 49 ++++++++++++++++++++
> >  drivers/gpu/drm/i915/intel_lrc.c             | 12 +++--
> >  drivers/gpu/drm/i915/intel_ringbuffer.c      | 24 +++++++---
> >  drivers/gpu/drm/i915/intel_ringbuffer.h      |  2 +-
> >  drivers/gpu/drm/i915/selftests/mock_engine.c |  4 +-
> >  5 files changed, 77 insertions(+), 14 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> > index 2f3c71f6d313..883ba208d1c2 100644
> > --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> > +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> > @@ -604,6 +604,47 @@ static void __intel_context_unpin(struct i915_gem_context *ctx,
> >       intel_context_unpin(to_intel_context(ctx, engine));
> >  }
> >  
> > +struct measure_breadcrumb {
> > +     struct i915_request rq;
> > +     struct i915_timeline timeline;
> > +     struct intel_ring ring;
> > +     u32 cs[1024];
> > +};
> > +
> > +static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
> > +{
> > +     struct measure_breadcrumb *frame;
> > +     unsigned int dw;
> > +
> > +     GEM_BUG_ON(!engine->i915->gt.scratch);
> > +
> > +     frame = kzalloc(sizeof(*frame), GFP_KERNEL);
> > +     if (!frame)
> > +             return -ENOMEM;
> > +
> > +     i915_timeline_init(engine->i915, &frame->timeline, engine->name);
> 
> You could init with null name. This is so short lived
> and we dont expect debugs. If it ever leaks into wild,
> blowout would be instant. Now the name is the same as
> the real deal.

Good point. I was just cutting and pasting for convenience, but having
the same name may be doubly confusing for the strange bugs where it might
matter :)

> > +     frame->ring.timeline = &frame->timeline;
> > +     frame->ring.vaddr = frame->cs;
> > +     frame->ring.size = sizeof(frame->cs);
> > +     frame->ring.effective_size = frame->ring.size;
> > +     frame->ring.space = frame->ring.size - 8;
> 
> Why 2 dwords short? Just curious as it doesn't seem
> to matter in this use case.

Just rules of the HW. Head steps in qword jumps. But should just
intel_ring_update_space ftw.

> > +     INIT_LIST_HEAD(&frame->ring.request_list);
> > +
> > +     frame->rq.i915 = engine->i915;
> > +     frame->rq.engine = engine;
> > +     frame->rq.ring = &frame->ring;
> > +     frame->rq.timeline = &frame->timeline;
> > +
> > +     dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
> > +     GEM_BUG_ON(dw != engine->emit_breadcrumb_sz);
> 
> Peace of mind provided =)

For full peace of mind, see the earlier runs with the BUG active.
https://patchwork.freedesktop.org/series/55683/
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 03/33] drm/i915: Remove manual breadcumb counting

[Chris Wilson]

Now that we know we measure the size of the engine->emit_breadcrumb()
correctly, we can remove the previous manual counting.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_request.c     |  4 ++--
 drivers/gpu/drm/i915/intel_engine_cs.c  |  7 +++----
 drivers/gpu/drm/i915/intel_lrc.c        |  4 ----
 drivers/gpu/drm/i915/intel_ringbuffer.c | 28 +++++--------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h |  2 +-
 5 files changed, 11 insertions(+), 34 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index f941e40fd373..ddc35e9dc0c0 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -650,7 +650,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	 * around inside i915_request_add() there is sufficient space at
 	 * the beginning of the ring as well.
 	 */
-	rq->reserved_space = 2 * engine->emit_breadcrumb_sz * sizeof(u32);
+	rq->reserved_space = 2 * engine->emit_breadcrumb_dw * sizeof(u32);
 
 	/*
 	 * Record the position of the start of the request so that
@@ -901,7 +901,7 @@ void i915_request_add(struct i915_request *request)
 	 * GPU processing the request, we never over-estimate the
 	 * position of the ring's HEAD.
 	 */
-	cs = intel_ring_begin(request, engine->emit_breadcrumb_sz);
+	cs = intel_ring_begin(request, engine->emit_breadcrumb_dw);
 	GEM_BUG_ON(IS_ERR(cs));
 	request->postfix = intel_ring_offset(request, cs);
 
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 883ba208d1c2..235a2d70d671 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -611,7 +611,7 @@ struct measure_breadcrumb {
 	u32 cs[1024];
 };
 
-static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
+static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 {
 	struct measure_breadcrumb *frame;
 	unsigned int dw;
@@ -637,7 +637,6 @@ static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
 	frame->rq.timeline = &frame->timeline;
 
 	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
-	GEM_BUG_ON(dw != engine->emit_breadcrumb_sz);
 
 	i915_timeline_fini(&frame->timeline);
 	kfree(frame);
@@ -698,11 +697,11 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 	if (ret)
 		goto err_breadcrumbs;
 
-	ret = measure_breadcrumb_sz(engine);
+	ret = measure_breadcrumb_dw(engine);
 	if (ret < 0)
 		goto err_status_page;
 
-	engine->emit_breadcrumb_sz = ret;
+	engine->emit_breadcrumb_dw = ret;
 
 	return 0;
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index d2299425cf2f..5551dd2ec0e6 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -2075,7 +2075,6 @@ static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 
 	return gen8_emit_wa_tail(request, cs);
 }
-static const int gen8_emit_breadcrumb_sz = 6 + WA_TAIL_DWORDS;
 
 static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
@@ -2099,7 +2098,6 @@ static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 
 	return gen8_emit_wa_tail(request, cs);
 }
-static const int gen8_emit_breadcrumb_rcs_sz = 8 + WA_TAIL_DWORDS;
 
 static int gen8_init_rcs_context(struct i915_request *rq)
 {
@@ -2192,7 +2190,6 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
 
 	engine->emit_flush = gen8_emit_flush;
 	engine->emit_breadcrumb = gen8_emit_breadcrumb;
-	engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_sz;
 
 	engine->set_default_submission = intel_execlists_set_default_submission;
 
@@ -2298,7 +2295,6 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
 	engine->init_context = gen8_init_rcs_context;
 	engine->emit_flush = gen8_emit_flush_render;
 	engine->emit_breadcrumb = gen8_emit_breadcrumb_rcs;
-	engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_rcs_sz;
 
 	ret = logical_ring_init(engine);
 	if (ret)
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 107c4934e2fa..09c90475168a 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -330,7 +330,6 @@ static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int gen6_rcs_emit_breadcrumb_sz = 14;
 
 static int
 gen7_render_ring_cs_stall_wa(struct i915_request *rq)
@@ -432,7 +431,6 @@ static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int gen7_rcs_emit_breadcrumb_sz = 6;
 
 static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
@@ -446,7 +444,6 @@ static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int gen6_xcs_emit_breadcrumb_sz = 4;
 
 #define GEN7_XCS_WA 32
 static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
@@ -475,7 +472,6 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int gen7_xcs_emit_breadcrumb_sz = 8 + GEN7_XCS_WA * 3;
 #undef GEN7_XCS_WA
 
 static void set_hwstam(struct intel_engine_cs *engine, u32 mask)
@@ -885,7 +881,6 @@ static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int i9xx_emit_breadcrumb_sz = 6;
 
 #define GEN5_WA_STORES 8 /* must be at least 1! */
 static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
@@ -908,7 +903,6 @@ static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	return cs;
 }
-static const int gen5_emit_breadcrumb_sz = GEN5_WA_STORES * 3 + 2;
 #undef GEN5_WA_STORES
 
 static void
@@ -2206,11 +2200,8 @@ static void intel_ring_default_vfuncs(struct drm_i915_private *dev_priv,
 	engine->request_alloc = ring_request_alloc;
 
 	engine->emit_breadcrumb = i9xx_emit_breadcrumb;
-	engine->emit_breadcrumb_sz = i9xx_emit_breadcrumb_sz;
-	if (IS_GEN(dev_priv, 5)) {
+	if (IS_GEN(dev_priv, 5))
 		engine->emit_breadcrumb = gen5_emit_breadcrumb;
-		engine->emit_breadcrumb_sz = gen5_emit_breadcrumb_sz;
-	}
 
 	engine->set_default_submission = i9xx_set_default_submission;
 
@@ -2240,12 +2231,10 @@ int intel_init_render_ring_buffer(struct intel_engine_cs *engine)
 		engine->init_context = intel_rcs_ctx_init;
 		engine->emit_flush = gen7_render_ring_flush;
 		engine->emit_breadcrumb = gen7_rcs_emit_breadcrumb;
-		engine->emit_breadcrumb_sz = gen7_rcs_emit_breadcrumb_sz;
 	} else if (IS_GEN(dev_priv, 6)) {
 		engine->init_context = intel_rcs_ctx_init;
 		engine->emit_flush = gen6_render_ring_flush;
 		engine->emit_breadcrumb = gen6_rcs_emit_breadcrumb;
-		engine->emit_breadcrumb_sz = gen6_rcs_emit_breadcrumb_sz;
 	} else if (IS_GEN(dev_priv, 5)) {
 		engine->emit_flush = gen4_render_ring_flush;
 	} else {
@@ -2281,13 +2270,10 @@ int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine)
 		engine->emit_flush = gen6_bsd_ring_flush;
 		engine->irq_enable_mask = GT_BSD_USER_INTERRUPT;
 
-		if (IS_GEN(dev_priv, 6)) {
+		if (IS_GEN(dev_priv, 6))
 			engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
-			engine->emit_breadcrumb_sz = gen6_xcs_emit_breadcrumb_sz;
-		} else {
+		else
 			engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
-			engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
-		}
 	} else {
 		engine->emit_flush = bsd_ring_flush;
 		if (IS_GEN(dev_priv, 5))
@@ -2310,13 +2296,10 @@ int intel_init_blt_ring_buffer(struct intel_engine_cs *engine)
 	engine->emit_flush = gen6_ring_flush;
 	engine->irq_enable_mask = GT_BLT_USER_INTERRUPT;
 
-	if (IS_GEN(dev_priv, 6)) {
+	if (IS_GEN(dev_priv, 6))
 		engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
-		engine->emit_breadcrumb_sz = gen6_xcs_emit_breadcrumb_sz;
-	} else {
+	else
 		engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
-		engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
-	}
 
 	return intel_init_ring_buffer(engine);
 }
@@ -2335,7 +2318,6 @@ int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine)
 	engine->irq_disable = hsw_vebox_irq_disable;
 
 	engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
-	engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
 
 	return intel_init_ring_buffer(engine);
 }
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 479bd53d4ac6..0834e91d4ace 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -471,7 +471,7 @@ struct intel_engine_cs {
 #define I915_DISPATCH_SECURE BIT(0)
 #define I915_DISPATCH_PINNED BIT(1)
 	u32		*(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
-	int		emit_breadcrumb_sz;
+	int		emit_breadcrumb_dw;
 
 	/* Pass the request to the hardware queue (e.g. directly into
 	 * the legacy ringbuffer or to the end of an execlist).
-- 
2.20.1

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

Re: [PATCH 03/33] drm/i915: Remove manual breadcumb counting

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Now that we know we measure the size of the engine->emit_breadcrumb()
> correctly, we can remove the previous manual counting.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/i915_request.c     |  4 ++--
>  drivers/gpu/drm/i915/intel_engine_cs.c  |  7 +++----
>  drivers/gpu/drm/i915/intel_lrc.c        |  4 ----
>  drivers/gpu/drm/i915/intel_ringbuffer.c | 28 +++++--------------------
>  drivers/gpu/drm/i915/intel_ringbuffer.h |  2 +-
>  5 files changed, 11 insertions(+), 34 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index f941e40fd373..ddc35e9dc0c0 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -650,7 +650,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
>  	 * around inside i915_request_add() there is sufficient space at
>  	 * the beginning of the ring as well.
>  	 */
> -	rq->reserved_space = 2 * engine->emit_breadcrumb_sz * sizeof(u32);
> +	rq->reserved_space = 2 * engine->emit_breadcrumb_dw * sizeof(u32);

*_sz got me startled, raising doubts about the previous patch.
But it was misnamed all the way.

Obviously merge the previous patch before this one :)
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

>  
>  	/*
>  	 * Record the position of the start of the request so that
> @@ -901,7 +901,7 @@ void i915_request_add(struct i915_request *request)
>  	 * GPU processing the request, we never over-estimate the
>  	 * position of the ring's HEAD.
>  	 */
> -	cs = intel_ring_begin(request, engine->emit_breadcrumb_sz);
> +	cs = intel_ring_begin(request, engine->emit_breadcrumb_dw);
>  	GEM_BUG_ON(IS_ERR(cs));
>  	request->postfix = intel_ring_offset(request, cs);
>  
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index 883ba208d1c2..235a2d70d671 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -611,7 +611,7 @@ struct measure_breadcrumb {
>  	u32 cs[1024];
>  };
>  
> -static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
> +static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
>  {
>  	struct measure_breadcrumb *frame;
>  	unsigned int dw;
> @@ -637,7 +637,6 @@ static int measure_breadcrumb_sz(struct intel_engine_cs *engine)
>  	frame->rq.timeline = &frame->timeline;
>  
>  	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
> -	GEM_BUG_ON(dw != engine->emit_breadcrumb_sz);
>  
>  	i915_timeline_fini(&frame->timeline);
>  	kfree(frame);
> @@ -698,11 +697,11 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
>  	if (ret)
>  		goto err_breadcrumbs;
>  
> -	ret = measure_breadcrumb_sz(engine);
> +	ret = measure_breadcrumb_dw(engine);
>  	if (ret < 0)
>  		goto err_status_page;
>  
> -	engine->emit_breadcrumb_sz = ret;
> +	engine->emit_breadcrumb_dw = ret;
>  
>  	return 0;
>  
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index d2299425cf2f..5551dd2ec0e6 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -2075,7 +2075,6 @@ static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
>  
>  	return gen8_emit_wa_tail(request, cs);
>  }
> -static const int gen8_emit_breadcrumb_sz = 6 + WA_TAIL_DWORDS;
>  
>  static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
>  {
> @@ -2099,7 +2098,6 @@ static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
>  
>  	return gen8_emit_wa_tail(request, cs);
>  }
> -static const int gen8_emit_breadcrumb_rcs_sz = 8 + WA_TAIL_DWORDS;
>  
>  static int gen8_init_rcs_context(struct i915_request *rq)
>  {
> @@ -2192,7 +2190,6 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
>  
>  	engine->emit_flush = gen8_emit_flush;
>  	engine->emit_breadcrumb = gen8_emit_breadcrumb;
> -	engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_sz;
>  
>  	engine->set_default_submission = intel_execlists_set_default_submission;
>  
> @@ -2298,7 +2295,6 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
>  	engine->init_context = gen8_init_rcs_context;
>  	engine->emit_flush = gen8_emit_flush_render;
>  	engine->emit_breadcrumb = gen8_emit_breadcrumb_rcs;
> -	engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_rcs_sz;
>  
>  	ret = logical_ring_init(engine);
>  	if (ret)
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> index 107c4934e2fa..09c90475168a 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> @@ -330,7 +330,6 @@ static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int gen6_rcs_emit_breadcrumb_sz = 14;
>  
>  static int
>  gen7_render_ring_cs_stall_wa(struct i915_request *rq)
> @@ -432,7 +431,6 @@ static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int gen7_rcs_emit_breadcrumb_sz = 6;
>  
>  static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  {
> @@ -446,7 +444,6 @@ static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int gen6_xcs_emit_breadcrumb_sz = 4;
>  
>  #define GEN7_XCS_WA 32
>  static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> @@ -475,7 +472,6 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int gen7_xcs_emit_breadcrumb_sz = 8 + GEN7_XCS_WA * 3;
>  #undef GEN7_XCS_WA
>  
>  static void set_hwstam(struct intel_engine_cs *engine, u32 mask)
> @@ -885,7 +881,6 @@ static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int i9xx_emit_breadcrumb_sz = 6;
>  
>  #define GEN5_WA_STORES 8 /* must be at least 1! */
>  static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> @@ -908,7 +903,6 @@ static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>  
>  	return cs;
>  }
> -static const int gen5_emit_breadcrumb_sz = GEN5_WA_STORES * 3 + 2;
>  #undef GEN5_WA_STORES
>  
>  static void
> @@ -2206,11 +2200,8 @@ static void intel_ring_default_vfuncs(struct drm_i915_private *dev_priv,
>  	engine->request_alloc = ring_request_alloc;
>  
>  	engine->emit_breadcrumb = i9xx_emit_breadcrumb;
> -	engine->emit_breadcrumb_sz = i9xx_emit_breadcrumb_sz;
> -	if (IS_GEN(dev_priv, 5)) {
> +	if (IS_GEN(dev_priv, 5))
>  		engine->emit_breadcrumb = gen5_emit_breadcrumb;
> -		engine->emit_breadcrumb_sz = gen5_emit_breadcrumb_sz;
> -	}
>  
>  	engine->set_default_submission = i9xx_set_default_submission;
>  
> @@ -2240,12 +2231,10 @@ int intel_init_render_ring_buffer(struct intel_engine_cs *engine)
>  		engine->init_context = intel_rcs_ctx_init;
>  		engine->emit_flush = gen7_render_ring_flush;
>  		engine->emit_breadcrumb = gen7_rcs_emit_breadcrumb;
> -		engine->emit_breadcrumb_sz = gen7_rcs_emit_breadcrumb_sz;
>  	} else if (IS_GEN(dev_priv, 6)) {
>  		engine->init_context = intel_rcs_ctx_init;
>  		engine->emit_flush = gen6_render_ring_flush;
>  		engine->emit_breadcrumb = gen6_rcs_emit_breadcrumb;
> -		engine->emit_breadcrumb_sz = gen6_rcs_emit_breadcrumb_sz;
>  	} else if (IS_GEN(dev_priv, 5)) {
>  		engine->emit_flush = gen4_render_ring_flush;
>  	} else {
> @@ -2281,13 +2270,10 @@ int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine)
>  		engine->emit_flush = gen6_bsd_ring_flush;
>  		engine->irq_enable_mask = GT_BSD_USER_INTERRUPT;
>  
> -		if (IS_GEN(dev_priv, 6)) {
> +		if (IS_GEN(dev_priv, 6))
>  			engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
> -			engine->emit_breadcrumb_sz = gen6_xcs_emit_breadcrumb_sz;
> -		} else {
> +		else
>  			engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
> -			engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
> -		}
>  	} else {
>  		engine->emit_flush = bsd_ring_flush;
>  		if (IS_GEN(dev_priv, 5))
> @@ -2310,13 +2296,10 @@ int intel_init_blt_ring_buffer(struct intel_engine_cs *engine)
>  	engine->emit_flush = gen6_ring_flush;
>  	engine->irq_enable_mask = GT_BLT_USER_INTERRUPT;
>  
> -	if (IS_GEN(dev_priv, 6)) {
> +	if (IS_GEN(dev_priv, 6))
>  		engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
> -		engine->emit_breadcrumb_sz = gen6_xcs_emit_breadcrumb_sz;
> -	} else {
> +	else
>  		engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
> -		engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
> -	}
>  
>  	return intel_init_ring_buffer(engine);
>  }
> @@ -2335,7 +2318,6 @@ int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine)
>  	engine->irq_disable = hsw_vebox_irq_disable;
>  
>  	engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
> -	engine->emit_breadcrumb_sz = gen7_xcs_emit_breadcrumb_sz;
>  
>  	return intel_init_ring_buffer(engine);
>  }
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 479bd53d4ac6..0834e91d4ace 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -471,7 +471,7 @@ struct intel_engine_cs {
>  #define I915_DISPATCH_SECURE BIT(0)
>  #define I915_DISPATCH_PINNED BIT(1)
>  	u32		*(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
> -	int		emit_breadcrumb_sz;
> +	int		emit_breadcrumb_dw;
>  
>  	/* Pass the request to the hardware queue (e.g. directly into
>  	 * the legacy ringbuffer or to the end of an execlist).
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 04/33] drm/i915: Compute the HWS offsets explicitly

[Chris Wilson]

Simplify by using sizeof(u32) to convert from the index inside the HWSP
to the byte offset. This has the advantage of not only being shorter
(and so not upsetting checkpatch!) but that it matches use where we are
writing to byte addresses using other commands than MI_STORE_DWORD_IMM.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_guc_submission.c |  4 ++--
 drivers/gpu/drm/i915/intel_ringbuffer.h     | 10 +++++-----
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index ab1c49b106f2..349ae5844f24 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -666,7 +666,7 @@ 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_INDEX, 0);
+	intel_write_status_page(engine, I915_GEM_HWS_PREEMPT, 0);
 }
 
 /**
@@ -824,7 +824,7 @@ static void guc_submission_tasklet(unsigned long data)
 	}
 
 	if (execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT) &&
-	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT_INDEX) ==
+	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT) ==
 	    GUC_PREEMPT_FINISHED)
 		complete_preempt_context(engine);
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 0834e91d4ace..5ad46c2fbc0f 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -716,11 +716,11 @@ intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
  * The area from dword 0x30 to 0x3ff is available for driver usage.
  */
 #define I915_GEM_HWS_INDEX		0x30
-#define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
-#define I915_GEM_HWS_PREEMPT_INDEX	0x32
-#define I915_GEM_HWS_PREEMPT_ADDR (I915_GEM_HWS_PREEMPT_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
-#define I915_GEM_HWS_SCRATCH_INDEX	0x40
-#define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
+#define I915_GEM_HWS_INDEX_ADDR		(I915_GEM_HWS_INDEX * sizeof(u32))
+#define I915_GEM_HWS_PREEMPT		0x32
+#define I915_GEM_HWS_PREEMPT_ADDR	(I915_GEM_HWS_PREEMPT * sizeof(u32))
+#define I915_GEM_HWS_SCRATCH		0x40
+#define I915_GEM_HWS_SCRATCH_ADDR	(I915_GEM_HWS_SCRATCH * sizeof(u32))
 
 #define I915_HWS_CSB_BUF0_INDEX		0x10
 #define I915_HWS_CSB_WRITE_INDEX	0x1f
-- 
2.20.1

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

Re: [PATCH 04/33] drm/i915: Compute the HWS offsets explicitly

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Simplify by using sizeof(u32) to convert from the index inside the HWSP
> to the byte offset. This has the advantage of not only being shorter
> (and so not upsetting checkpatch!) but that it matches use where we are
> writing to byte addresses using other commands than MI_STORE_DWORD_IMM.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/intel_guc_submission.c |  4 ++--
>  drivers/gpu/drm/i915/intel_ringbuffer.h     | 10 +++++-----
>  2 files changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
> index ab1c49b106f2..349ae5844f24 100644
> --- a/drivers/gpu/drm/i915/intel_guc_submission.c
> +++ b/drivers/gpu/drm/i915/intel_guc_submission.c
> @@ -666,7 +666,7 @@ 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_INDEX, 0);
> +	intel_write_status_page(engine, I915_GEM_HWS_PREEMPT, 0);
>  }
>  
>  /**
> @@ -824,7 +824,7 @@ static void guc_submission_tasklet(unsigned long data)
>  	}
>  
>  	if (execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT) &&
> -	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT_INDEX) ==
> +	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT) ==
>  	    GUC_PREEMPT_FINISHED)
>  		complete_preempt_context(engine);
>  
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 0834e91d4ace..5ad46c2fbc0f 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -716,11 +716,11 @@ intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
>   * The area from dword 0x30 to 0x3ff is available for driver usage.
>   */
>  #define I915_GEM_HWS_INDEX		0x30
> -#define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
> -#define I915_GEM_HWS_PREEMPT_INDEX	0x32
> -#define I915_GEM_HWS_PREEMPT_ADDR (I915_GEM_HWS_PREEMPT_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
> -#define I915_GEM_HWS_SCRATCH_INDEX	0x40
> -#define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)

I don't find usage nor reasoning for MI_STORE_DWORD_INDEX_SHIFT in the
intel_gpu_commands.h. So it should go too.

Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

> +#define I915_GEM_HWS_INDEX_ADDR		(I915_GEM_HWS_INDEX * sizeof(u32))
> +#define I915_GEM_HWS_PREEMPT		0x32
> +#define I915_GEM_HWS_PREEMPT_ADDR	(I915_GEM_HWS_PREEMPT * sizeof(u32))
> +#define I915_GEM_HWS_SCRATCH		0x40
> +#define I915_GEM_HWS_SCRATCH_ADDR	(I915_GEM_HWS_SCRATCH * sizeof(u32))
>  
>  #define I915_HWS_CSB_BUF0_INDEX		0x10
>  #define I915_HWS_CSB_WRITE_INDEX	0x1f
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 05/33] drm/i915/execlists: Suppress preempting self

[Chris Wilson]

In order to avoid preempting ourselves, we currently refuse to schedule
the tasklet if we reschedule an inflight context. However, this glosses
over a few issues such as what happens after a CS completion event and
we then preempt the newly executing context with itself, or if something
else causes a tasklet_schedule triggering the same evaluation to
preempt the active context with itself.

To avoid the extra complications, after deciding that we have
potentially queued a request with higher priority than the currently
executing request, inspect the head of the queue to see if it is indeed
higher priority from another context.

v2: We can simplify a bunch of tests based on the knowledge that PI will
ensure that earlier requests along the same context will have the highest
priority.

References: a2bf92e8cc16 ("drm/i915/execlists: Avoid kicking priority on the current context")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_scheduler.c | 20 ++++--
 drivers/gpu/drm/i915/intel_lrc.c      | 91 ++++++++++++++++++++++++---
 2 files changed, 100 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 340faea6c08a..fb5d953430e5 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -239,6 +239,18 @@ sched_lock_engine(struct i915_sched_node *node, struct intel_engine_cs *locked)
 	return engine;
 }
 
+static bool inflight(const struct i915_request *rq,
+		     const struct intel_engine_cs *engine)
+{
+	const struct i915_request *active;
+
+	if (!rq->global_seqno)
+		return false;
+
+	active = port_request(engine->execlists.port);
+	return active->hw_context == rq->hw_context;
+}
+
 static void __i915_schedule(struct i915_request *rq,
 			    const struct i915_sched_attr *attr)
 {
@@ -328,6 +340,7 @@ static void __i915_schedule(struct i915_request *rq,
 		INIT_LIST_HEAD(&dep->dfs_link);
 
 		engine = sched_lock_engine(node, engine);
+		lockdep_assert_held(&engine->timeline.lock);
 
 		/* Recheck after acquiring the engine->timeline.lock */
 		if (prio <= node->attr.priority || node_signaled(node))
@@ -356,17 +369,16 @@ static void __i915_schedule(struct i915_request *rq,
 		if (prio <= engine->execlists.queue_priority)
 			continue;
 
+		engine->execlists.queue_priority = prio;
+
 		/*
 		 * If we are already the currently executing context, don't
 		 * bother evaluating if we should preempt ourselves.
 		 */
-		if (node_to_request(node)->global_seqno &&
-		    i915_seqno_passed(port_request(engine->execlists.port)->global_seqno,
-				      node_to_request(node)->global_seqno))
+		if (inflight(node_to_request(node), engine))
 			continue;
 
 		/* Defer (tasklet) submission until after all of our updates. */
-		engine->execlists.queue_priority = prio;
 		tasklet_hi_schedule(&engine->execlists.tasklet);
 	}
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 5551dd2ec0e6..37defd14c198 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -181,13 +181,89 @@ static inline int rq_prio(const struct i915_request *rq)
 	return rq->sched.attr.priority;
 }
 
+static int queue_prio(const struct intel_engine_execlists *execlists)
+{
+	struct i915_priolist *p;
+	struct rb_node *rb;
+
+	rb = rb_first_cached(&execlists->queue);
+	if (!rb)
+		return INT_MIN;
+
+	/*
+	 * As the priolist[] are inverted, with the highest priority in [0],
+	 * we have to flip the index value to become priority.
+	 */
+	p = to_priolist(rb);
+	return ((p->priority + 1) << I915_USER_PRIORITY_SHIFT) - ffs(p->used);
+}
+
 static inline bool need_preempt(const struct intel_engine_cs *engine,
-				const struct i915_request *last,
-				int prio)
+				const struct i915_request *rq,
+				int q_prio)
 {
-	return (intel_engine_has_preemption(engine) &&
-		__execlists_need_preempt(prio, rq_prio(last)) &&
-		!i915_request_completed(last));
+	const struct intel_context *ctx = rq->hw_context;
+	const int last_prio = rq_prio(rq);
+
+	if (!intel_engine_has_preemption(engine))
+		return false;
+
+	if (i915_request_completed(rq))
+		return false;
+
+	/*
+	 * Check if the current queue_priority merits a preemption attempt.
+	 *
+	 * However, the queue_priority is a mere hint that we may need to
+	 * preempt. If that hint is stale or we may be trying to preempt
+	 * ourselves, ignore the request.
+	 */
+	if (!__execlists_need_preempt(q_prio, last_prio))
+		return false;
+
+	/*
+	 * Check against the first request in ELSP[1], it will, thanks to the
+	 * power of PI, be the highest priority of that context.
+	 */
+	if (!list_is_last(&rq->link, &engine->timeline.requests)) {
+		rq = list_next_entry(rq, link);
+		GEM_BUG_ON(rq->hw_context == ctx);
+		if (rq_prio(rq) > last_prio)
+			return true;
+	}
+
+	/*
+	 * If the inflight context did not trigger the preemption, then maybe
+	 * it was the set of queued requests? Pick the highest priority in
+	 * the queue (the first active priolist) and see if it deserves to be
+	 * running instead of ELSP[0].
+	 *
+	 * The highest priority request in the queue can not be either
+	 * ELSP[0] or ELSP[1] as, thanks again to PI, if it was the same
+	 * context, it's priority would not exceed ELSP[0] aka last_prio.
+	 */
+	return queue_prio(&engine->execlists) > last_prio;
+}
+
+__maybe_unused static inline bool
+assert_priority_queue(const struct intel_engine_execlists *execlists,
+		      const struct i915_request *prev,
+		      const struct i915_request *next)
+{
+	if (!prev)
+		return true;
+
+	/*
+	 * Without preemption, the prev may refer to the still active element
+	 * which we refuse to let go.
+	 *
+	 * 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)
+		return true;
+
+	return rq_prio(prev) >= rq_prio(next);
 }
 
 /*
@@ -626,8 +702,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		int i;
 
 		priolist_for_each_request_consume(rq, rn, p, i) {
-			GEM_BUG_ON(last &&
-				   need_preempt(engine, last, rq_prio(rq)));
+			GEM_BUG_ON(!assert_priority_queue(execlists, last, rq));
 
 			/*
 			 * Can we combine this request with the current port?
@@ -872,6 +947,8 @@ static void process_csb(struct intel_engine_cs *engine)
 	const u32 * const buf = execlists->csb_status;
 	u8 head, tail;
 
+	lockdep_assert_held(&engine->timeline.lock);
+
 	/*
 	 * Note that csb_write, csb_status may be either in HWSP or mmio.
 	 * When reading from the csb_write mmio register, we have to be
-- 
2.20.1

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

[PATCH 06/33] drm/i915/execlists: Suppress redundant preemption

[Chris Wilson]

On unwinding the active request we give it a small (limited to internal
priority levels) boost to prevent it from being gazumped a second time.
However, this means that it can be promoted to above the request that
triggered the preemption request, causing a preempt-to-idle cycle for no
change. We can avoid this if we take the boost into account when
checking if the preemption request is valid.

v2: After preemption the active request will be after the preemptee if
they end up with equal priority.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_lrc.c | 39 ++++++++++++++++++++++++++++----
 1 file changed, 35 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 37defd14c198..6cef458f6b43 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -163,6 +163,8 @@
 #define WA_TAIL_DWORDS 2
 #define WA_TAIL_BYTES (sizeof(u32) * WA_TAIL_DWORDS)
 
+#define ACTIVE_PRIORITY (I915_PRIORITY_NEWCLIENT)
+
 static int execlists_context_deferred_alloc(struct i915_gem_context *ctx,
 					    struct intel_engine_cs *engine,
 					    struct intel_context *ce);
@@ -181,6 +183,34 @@ static inline int rq_prio(const struct i915_request *rq)
 	return rq->sched.attr.priority;
 }
 
+static inline int active_prio(const struct i915_request *rq)
+{
+	int prio = rq_prio(rq);
+
+	/*
+	 * On unwinding the active request, we give it a priority bump
+	 * equivalent to a freshly submitted request. This protects it from
+	 * being gazumped again, but it would be preferable if we didn't
+	 * let it be gazumped in the first place!
+	 *
+	 * See __unwind_incomplete_requests()
+	 */
+	if ((prio & ACTIVE_PRIORITY) != ACTIVE_PRIORITY &&
+	    i915_request_started(rq)) {
+		/*
+		 * After preemption, we insert the active request at the
+		 * end of the new priority level. This means that we will be
+		 * _lower_ priority than the preemptee all things equal (and
+		 * so the preemption is valid), so adjust our comparison
+		 * accordingly.
+		 */
+		prio |= ACTIVE_PRIORITY;
+		prio--;
+	}
+
+	return prio;
+}
+
 static int queue_prio(const struct intel_engine_execlists *execlists)
 {
 	struct i915_priolist *p;
@@ -203,7 +233,7 @@ static inline bool need_preempt(const struct intel_engine_cs *engine,
 				int q_prio)
 {
 	const struct intel_context *ctx = rq->hw_context;
-	const int last_prio = rq_prio(rq);
+	int last_prio;
 
 	if (!intel_engine_has_preemption(engine))
 		return false;
@@ -218,6 +248,7 @@ static inline bool need_preempt(const struct intel_engine_cs *engine,
 	 * preempt. If that hint is stale or we may be trying to preempt
 	 * ourselves, ignore the request.
 	 */
+	last_prio = active_prio(rq);
 	if (!__execlists_need_preempt(q_prio, last_prio))
 		return false;
 
@@ -344,7 +375,7 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
 {
 	struct i915_request *rq, *rn, *active = NULL;
 	struct list_head *uninitialized_var(pl);
-	int prio = I915_PRIORITY_INVALID | I915_PRIORITY_NEWCLIENT;
+	int prio = I915_PRIORITY_INVALID | ACTIVE_PRIORITY;
 
 	lockdep_assert_held(&engine->timeline.lock);
 
@@ -376,8 +407,8 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
 	 * stream, so give it the equivalent small priority bump to prevent
 	 * it being gazumped a second time by another peer.
 	 */
-	if (!(prio & I915_PRIORITY_NEWCLIENT)) {
-		prio |= I915_PRIORITY_NEWCLIENT;
+	if ((prio & ACTIVE_PRIORITY) != ACTIVE_PRIORITY) {
+		prio |= ACTIVE_PRIORITY;
 		active->sched.attr.priority = prio;
 		list_move_tail(&active->sched.link,
 			       i915_sched_lookup_priolist(engine, prio));
-- 
2.20.1

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

[PATCH 07/33] drm/i915/selftests: Apply a subtest filter

[Chris Wilson]

In bringup on simulated HW even rudimentary tests are slow, and so many
may fail that we want to be able to filter out the noise to focus on the
specific problem. Even just the tests groups provided for igt is not
specific enough, and we would like to isolate one particular subtest
(and probably subsubtests!). For simplicity, allow the user to provide a
command line parameter such as

	i915.st_filter=i915_timeline_mock_selftests/igt_sync

to restrict ourselves to only running on subtest. The exact name to use
is given during a normal run, highlighted as an error if it failed,
debug otherwise. The test group is optional, and then all subtests are
compared for an exact match with the filter (most subtests have unique
names). The filter can be negated, e.g. i915.st_filter=!igt_sync and
then all tests but those that match will be run. More than one match can
be supplied separated by a comma, e.g.

	i915.st_filter=igt_vma_create,igt_vma_pin1

to only run those specified, or

	i915.st_filter=!igt_vma_create,!igt_vma_pin1

to run all but those named. Mixing a blacklist and whitelist will only
execute those subtests matching the whitelist so long as they are
previously excluded in the blacklist.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_selftest.h          |  1 +
 .../gpu/drm/i915/selftests/i915_selftest.c    | 44 +++++++++++++++++++
 2 files changed, 45 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_selftest.h b/drivers/gpu/drm/i915/i915_selftest.h
index a73472dd12fd..207e21b478f2 100644
--- a/drivers/gpu/drm/i915/i915_selftest.h
+++ b/drivers/gpu/drm/i915/i915_selftest.h
@@ -31,6 +31,7 @@ struct i915_selftest {
 	unsigned long timeout_jiffies;
 	unsigned int timeout_ms;
 	unsigned int random_seed;
+	char *filter;
 	int mock;
 	int live;
 };
diff --git a/drivers/gpu/drm/i915/selftests/i915_selftest.c b/drivers/gpu/drm/i915/selftests/i915_selftest.c
index 86c54ea37f48..1b174ac65ff9 100644
--- a/drivers/gpu/drm/i915/selftests/i915_selftest.c
+++ b/drivers/gpu/drm/i915/selftests/i915_selftest.c
@@ -197,6 +197,46 @@ int i915_live_selftests(struct pci_dev *pdev)
 	return 0;
 }
 
+static bool apply_subtest_filter(const char *caller, const char *name)
+{
+	char *filter, *sep, *tok;
+	bool result = true;
+
+	filter = kstrdup(i915_selftest.filter, GFP_KERNEL);
+	for (sep = filter; (tok = strsep(&sep, ","));) {
+		bool allow = true;
+		char *sl;
+
+		if (*tok == '!') {
+			allow = false;
+			tok++;
+		}
+
+		sl = strchr(tok, '/');
+		if (sl) {
+			*sl++ = '\0';
+			if (strcmp(tok, caller)) {
+				if (allow)
+					result = false;
+				continue;
+			}
+			tok = sl;
+		}
+
+		if (strcmp(tok, name)) {
+			if (allow)
+				result = false;
+			continue;
+		}
+
+		result = allow;
+		break;
+	}
+	kfree(filter);
+
+	return result;
+}
+
 int __i915_subtests(const char *caller,
 		    const struct i915_subtest *st,
 		    unsigned int count,
@@ -209,6 +249,9 @@ int __i915_subtests(const char *caller,
 		if (signal_pending(current))
 			return -EINTR;
 
+		if (!apply_subtest_filter(caller, st->name))
+			continue;
+
 		pr_debug(DRIVER_NAME ": Running %s/%s\n", caller, st->name);
 		GEM_TRACE("Running %s/%s\n", caller, st->name);
 
@@ -244,6 +287,7 @@ bool __igt_timeout(unsigned long timeout, const char *fmt, ...)
 
 module_param_named(st_random_seed, i915_selftest.random_seed, uint, 0400);
 module_param_named(st_timeout, i915_selftest.timeout_ms, uint, 0400);
+module_param_named(st_filter, i915_selftest.filter, charp, 0400);
 
 module_param_named_unsafe(mock_selftests, i915_selftest.mock, int, 0400);
 MODULE_PARM_DESC(mock_selftests, "Run selftests before loading, using mock hardware (0:disabled [default], 1:run tests then load driver, -1:run tests then exit module)");
-- 
2.20.1

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

Re: [PATCH 07/33] drm/i915/selftests: Apply a subtest filter

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> In bringup on simulated HW even rudimentary tests are slow, and so many
> may fail that we want to be able to filter out the noise to focus on the
> specific problem. Even just the tests groups provided for igt is not
> specific enough, and we would like to isolate one particular subtest
> (and probably subsubtests!). For simplicity, allow the user to provide a
> command line parameter such as
>
> 	i915.st_filter=i915_timeline_mock_selftests/igt_sync
>
> to restrict ourselves to only running on subtest. The exact name to use
> is given during a normal run, highlighted as an error if it failed,
> debug otherwise. The test group is optional, and then all subtests are
> compared for an exact match with the filter (most subtests have unique
> names). The filter can be negated, e.g. i915.st_filter=!igt_sync and
> then all tests but those that match will be run. More than one match can
> be supplied separated by a comma, e.g.
>
> 	i915.st_filter=igt_vma_create,igt_vma_pin1
>
> to only run those specified, or
>
> 	i915.st_filter=!igt_vma_create,!igt_vma_pin1
>
> to run all but those named. Mixing a blacklist and whitelist will only
> execute those subtests matching the whitelist so long as they are
> previously excluded in the blacklist.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
> ---
>  drivers/gpu/drm/i915/i915_selftest.h          |  1 +
>  .../gpu/drm/i915/selftests/i915_selftest.c    | 44 +++++++++++++++++++
>  2 files changed, 45 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_selftest.h b/drivers/gpu/drm/i915/i915_selftest.h
> index a73472dd12fd..207e21b478f2 100644
> --- a/drivers/gpu/drm/i915/i915_selftest.h
> +++ b/drivers/gpu/drm/i915/i915_selftest.h
> @@ -31,6 +31,7 @@ struct i915_selftest {
>  	unsigned long timeout_jiffies;
>  	unsigned int timeout_ms;
>  	unsigned int random_seed;
> +	char *filter;
>  	int mock;
>  	int live;
>  };
> diff --git a/drivers/gpu/drm/i915/selftests/i915_selftest.c b/drivers/gpu/drm/i915/selftests/i915_selftest.c
> index 86c54ea37f48..1b174ac65ff9 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_selftest.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_selftest.c
> @@ -197,6 +197,46 @@ int i915_live_selftests(struct pci_dev *pdev)
>  	return 0;
>  }
>  
> +static bool apply_subtest_filter(const char *caller, const char *name)
> +{
> +	char *filter, *sep, *tok;
> +	bool result = true;
> +
> +	filter = kstrdup(i915_selftest.filter, GFP_KERNEL);

Was going to say that bailout if !filter.
But apparently it is ok this way.

-Mika

> +	for (sep = filter; (tok = strsep(&sep, ","));) {
> +		bool allow = true;
> +		char *sl;
> +
> +		if (*tok == '!') {
> +			allow = false;
> +			tok++;
> +		}
> +
> +		sl = strchr(tok, '/');
> +		if (sl) {
> +			*sl++ = '\0';
> +			if (strcmp(tok, caller)) {
> +				if (allow)
> +					result = false;
> +				continue;
> +			}
> +			tok = sl;
> +		}
> +
> +		if (strcmp(tok, name)) {
> +			if (allow)
> +				result = false;
> +			continue;
> +		}
> +
> +		result = allow;
> +		break;
> +	}
> +	kfree(filter);
> +
> +	return result;
> +}
> +
>  int __i915_subtests(const char *caller,
>  		    const struct i915_subtest *st,
>  		    unsigned int count,
> @@ -209,6 +249,9 @@ int __i915_subtests(const char *caller,
>  		if (signal_pending(current))
>  			return -EINTR;
>  
> +		if (!apply_subtest_filter(caller, st->name))
> +			continue;
> +
>  		pr_debug(DRIVER_NAME ": Running %s/%s\n", caller, st->name);
>  		GEM_TRACE("Running %s/%s\n", caller, st->name);
>  
> @@ -244,6 +287,7 @@ bool __igt_timeout(unsigned long timeout, const char *fmt, ...)
>  
>  module_param_named(st_random_seed, i915_selftest.random_seed, uint, 0400);
>  module_param_named(st_timeout, i915_selftest.timeout_ms, uint, 0400);
> +module_param_named(st_filter, i915_selftest.filter, charp, 0400);
>  
>  module_param_named_unsafe(mock_selftests, i915_selftest.mock, int, 0400);
>  MODULE_PARM_DESC(mock_selftests, "Run selftests before loading, using mock hardware (0:disabled [default], 1:run tests then load driver, -1:run tests then exit module)");
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 07/33] drm/i915/selftests: Apply a subtest filter

[Chris Wilson]

Quoting Mika Kuoppala (2019-01-25 11:44:30)
> Chris Wilson <chris@chris-wilson.co.uk> writes:
> > +static bool apply_subtest_filter(const char *caller, const char *name)
> > +{
> > +     char *filter, *sep, *tok;
> > +     bool result = true;
> > +
> > +     filter = kstrdup(i915_selftest.filter, GFP_KERNEL);
> 
> Was going to say that bailout if !filter.
> But apparently it is ok this way.

I thought it was nifty, "Look ma, no hands!" :)
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 07/33] drm/i915/selftests: Apply a subtest filter

[Joonas Lahtinen]

Quoting Chris Wilson (2019-01-25 04:29:39)
> In bringup on simulated HW even rudimentary tests are slow, and so many
> may fail that we want to be able to filter out the noise to focus on the
> specific problem. Even just the tests groups provided for igt is not
> specific enough, and we would like to isolate one particular subtest
> (and probably subsubtests!). For simplicity, allow the user to provide a
> command line parameter such as
> 
>         i915.st_filter=i915_timeline_mock_selftests/igt_sync
> 
> to restrict ourselves to only running on subtest. The exact name to use
> is given during a normal run, highlighted as an error if it failed,
> debug otherwise. The test group is optional, and then all subtests are
> compared for an exact match with the filter (most subtests have unique
> names). The filter can be negated, e.g. i915.st_filter=!igt_sync and
> then all tests but those that match will be run. More than one match can
> be supplied separated by a comma, e.g.
> 
>         i915.st_filter=igt_vma_create,igt_vma_pin1
> 
> to only run those specified, or
> 
>         i915.st_filter=!igt_vma_create,!igt_vma_pin1
> 
> to run all but those named. Mixing a blacklist and whitelist will only
> execute those subtests matching the whitelist so long as they are
> previously excluded in the blacklist.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

I don't think we have time budger pressure for O(1) solution, so
this is:

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

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

[PATCH 08/33] drm/i915: Make all GPU resets atomic

[Chris Wilson]

In preparation for the next few commits, make resetting the GPU atomic.
Currently, we have prepared gen6+ for atomic resetting of individual
engines, but now there is a requirement to perform the whole device
level reset (just the register poking) from inside an atomic context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_reset.c | 90 ++++++++++++++-----------------
 1 file changed, 39 insertions(+), 51 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index 342d9ee42601..2f840858572c 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -12,6 +12,8 @@
 
 #include "intel_guc.h"
 
+#define RESET_MAX_RETRIES 3
+
 static void engine_skip_context(struct i915_request *rq)
 {
 	struct intel_engine_cs *engine = rq->engine;
@@ -144,14 +146,14 @@ static int i915_do_reset(struct drm_i915_private *i915,
 
 	/* Assert reset for at least 20 usec, and wait for acknowledgement. */
 	pci_write_config_byte(pdev, I915_GDRST, GRDOM_RESET_ENABLE);
-	usleep_range(50, 200);
-	err = wait_for(i915_in_reset(pdev), 500);
+	udelay(50);
+	err = wait_for_atomic(i915_in_reset(pdev), 50);
 
 	/* Clear the reset request. */
 	pci_write_config_byte(pdev, I915_GDRST, 0);
-	usleep_range(50, 200);
+	udelay(50);
 	if (!err)
-		err = wait_for(!i915_in_reset(pdev), 500);
+		err = wait_for_atomic(!i915_in_reset(pdev), 50);
 
 	return err;
 }
@@ -171,7 +173,7 @@ static int g33_do_reset(struct drm_i915_private *i915,
 	struct pci_dev *pdev = i915->drm.pdev;
 
 	pci_write_config_byte(pdev, I915_GDRST, GRDOM_RESET_ENABLE);
-	return wait_for(g4x_reset_complete(pdev), 500);
+	return wait_for_atomic(g4x_reset_complete(pdev), 50);
 }
 
 static int g4x_do_reset(struct drm_i915_private *dev_priv,
@@ -182,13 +184,13 @@ static int g4x_do_reset(struct drm_i915_private *dev_priv,
 	int ret;
 
 	/* WaVcpClkGateDisableForMediaReset:ctg,elk */
-	I915_WRITE(VDECCLK_GATE_D,
-		   I915_READ(VDECCLK_GATE_D) | VCP_UNIT_CLOCK_GATE_DISABLE);
-	POSTING_READ(VDECCLK_GATE_D);
+	I915_WRITE_FW(VDECCLK_GATE_D,
+		      I915_READ(VDECCLK_GATE_D) | VCP_UNIT_CLOCK_GATE_DISABLE);
+	POSTING_READ_FW(VDECCLK_GATE_D);
 
 	pci_write_config_byte(pdev, I915_GDRST,
 			      GRDOM_MEDIA | GRDOM_RESET_ENABLE);
-	ret =  wait_for(g4x_reset_complete(pdev), 500);
+	ret =  wait_for_atomic(g4x_reset_complete(pdev), 50);
 	if (ret) {
 		DRM_DEBUG_DRIVER("Wait for media reset failed\n");
 		goto out;
@@ -196,7 +198,7 @@ static int g4x_do_reset(struct drm_i915_private *dev_priv,
 
 	pci_write_config_byte(pdev, I915_GDRST,
 			      GRDOM_RENDER | GRDOM_RESET_ENABLE);
-	ret =  wait_for(g4x_reset_complete(pdev), 500);
+	ret =  wait_for_atomic(g4x_reset_complete(pdev), 50);
 	if (ret) {
 		DRM_DEBUG_DRIVER("Wait for render reset failed\n");
 		goto out;
@@ -205,9 +207,9 @@ static int g4x_do_reset(struct drm_i915_private *dev_priv,
 out:
 	pci_write_config_byte(pdev, I915_GDRST, 0);
 
-	I915_WRITE(VDECCLK_GATE_D,
-		   I915_READ(VDECCLK_GATE_D) & ~VCP_UNIT_CLOCK_GATE_DISABLE);
-	POSTING_READ(VDECCLK_GATE_D);
+	I915_WRITE_FW(VDECCLK_GATE_D,
+		      I915_READ(VDECCLK_GATE_D) & ~VCP_UNIT_CLOCK_GATE_DISABLE);
+	POSTING_READ_FW(VDECCLK_GATE_D);
 
 	return ret;
 }
@@ -218,27 +220,29 @@ static int ironlake_do_reset(struct drm_i915_private *dev_priv,
 {
 	int ret;
 
-	I915_WRITE(ILK_GDSR, ILK_GRDOM_RENDER | ILK_GRDOM_RESET_ENABLE);
-	ret = intel_wait_for_register(dev_priv,
-				      ILK_GDSR, ILK_GRDOM_RESET_ENABLE, 0,
-				      500);
+	I915_WRITE_FW(ILK_GDSR, ILK_GRDOM_RENDER | ILK_GRDOM_RESET_ENABLE);
+	ret = __intel_wait_for_register_fw(dev_priv, ILK_GDSR,
+					   ILK_GRDOM_RESET_ENABLE, 0,
+					   5000, 0,
+					   NULL);
 	if (ret) {
 		DRM_DEBUG_DRIVER("Wait for render reset failed\n");
 		goto out;
 	}
 
-	I915_WRITE(ILK_GDSR, ILK_GRDOM_MEDIA | ILK_GRDOM_RESET_ENABLE);
-	ret = intel_wait_for_register(dev_priv,
-				      ILK_GDSR, ILK_GRDOM_RESET_ENABLE, 0,
-				      500);
+	I915_WRITE_FW(ILK_GDSR, ILK_GRDOM_MEDIA | ILK_GRDOM_RESET_ENABLE);
+	ret = __intel_wait_for_register_fw(dev_priv, ILK_GDSR,
+					   ILK_GRDOM_RESET_ENABLE, 0,
+					   5000, 0,
+					   NULL);
 	if (ret) {
 		DRM_DEBUG_DRIVER("Wait for media reset failed\n");
 		goto out;
 	}
 
 out:
-	I915_WRITE(ILK_GDSR, 0);
-	POSTING_READ(ILK_GDSR);
+	I915_WRITE_FW(ILK_GDSR, 0);
+	POSTING_READ_FW(ILK_GDSR);
 	return ret;
 }
 
@@ -527,32 +531,21 @@ static reset_func intel_get_gpu_reset(struct drm_i915_private *i915)
 
 int intel_gpu_reset(struct drm_i915_private *i915, unsigned int engine_mask)
 {
-	reset_func reset = intel_get_gpu_reset(i915);
+	const int retries = engine_mask == ALL_ENGINES ? RESET_MAX_RETRIES : 1;
+	reset_func reset;
+	int ret = -ETIMEDOUT;
 	int retry;
-	int ret;
 
-	/*
-	 * We want to perform per-engine reset from atomic context (e.g.
-	 * softirq), which imposes the constraint that we cannot sleep.
-	 * However, experience suggests that spending a bit of time waiting
-	 * for a reset helps in various cases, so for a full-device reset
-	 * we apply the opposite rule and wait if we want to. As we should
-	 * always follow up a failed per-engine reset with a full device reset,
-	 * being a little faster, stricter and more error prone for the
-	 * atomic case seems an acceptable compromise.
-	 *
-	 * Unfortunately this leads to a bimodal routine, when the goal was
-	 * to have a single reset function that worked for resetting any
-	 * number of engines simultaneously.
-	 */
-	might_sleep_if(engine_mask == ALL_ENGINES);
+	reset = intel_get_gpu_reset(i915);
+	if (!reset)
+		return -ENODEV;
 
 	/*
 	 * If the power well sleeps during the reset, the reset
 	 * request may be dropped and never completes (causing -EIO).
 	 */
 	intel_uncore_forcewake_get(i915, FORCEWAKE_ALL);
-	for (retry = 0; retry < 3; retry++) {
+	for (retry = 0; ret == -ETIMEDOUT && retry < retries; retry++) {
 		/*
 		 * We stop engines, otherwise we might get failed reset and a
 		 * dead gpu (on elk). Also as modern gpu as kbl can suffer
@@ -569,15 +562,10 @@ int intel_gpu_reset(struct drm_i915_private *i915, unsigned int engine_mask)
 		 */
 		i915_stop_engines(i915, engine_mask);
 
-		ret = -ENODEV;
-		if (reset) {
-			GEM_TRACE("engine_mask=%x\n", engine_mask);
-			ret = reset(i915, engine_mask, retry);
-		}
-		if (ret != -ETIMEDOUT || engine_mask != ALL_ENGINES)
-			break;
-
-		cond_resched();
+		GEM_TRACE("engine_mask=%x\n", engine_mask);
+		preempt_disable();
+		ret = reset(i915, engine_mask, retry);
+		preempt_enable();
 	}
 	intel_uncore_forcewake_put(i915, FORCEWAKE_ALL);
 
@@ -1014,7 +1002,7 @@ void i915_reset(struct drm_i915_private *i915,
 		goto error;
 	}
 
-	for (i = 0; i < 3; i++) {
+	for (i = 0; i < RESET_MAX_RETRIES; i++) {
 		ret = intel_gpu_reset(i915, ALL_ENGINES);
 		if (ret == 0)
 			break;
-- 
2.20.1

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

[PATCH 09/33] drm/i915/guc: Disable global reset

[Chris Wilson]

The guc (and huc) currently inexcruitably depend on struct_mutex for
device reinitialisation from inside the reset, and indeed taking any
mutex here is verboten (as we must be able to reset from underneath any
of our mutexes). That makes recovering the guc unviable without, for
example, reserving contiguous vma space and pages for it to use.

The plan to re-enable global reset for the GuC centres around reusing the
WOPM reserved space at the top of the aperture (that we know we can
populate a contiguous range large enough to dma xfer the fw image).

In the meantime, hopefully no one even notices as the device-reset is
only used as a backup to the per-engine resets for handling GPU hangs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
---
 drivers/gpu/drm/i915/i915_reset.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index 2f840858572c..33408c4e6358 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -574,6 +574,9 @@ int intel_gpu_reset(struct drm_i915_private *i915, unsigned int engine_mask)
 
 bool intel_has_gpu_reset(struct drm_i915_private *i915)
 {
+	if (USES_GUC(i915))
+		return false;
+
 	return intel_get_gpu_reset(i915);
 }
 
-- 
2.20.1

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

[PATCH 10/33] drm/i915: Remove GPU reset dependence on struct_mutex

[Chris Wilson]

Now that the submission backends are controlled via their own spinlocks,
with a wave of a magic wand we can lift the struct_mutex requirement
around GPU reset. That is we allow the submission frontend (userspace)
to keep on submitting while we process the GPU reset as we can suspend
the backend independently.

The major change is around the backoff/handoff strategy for performing
the reset. With no mutex deadlock, we no longer have to coordinate with
any waiter, and just perform the reset immediately.

Testcase: igt/gem_mmap_gtt/hang # regresses
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c           |  38 +-
 drivers/gpu/drm/i915/i915_drv.h               |   5 -
 drivers/gpu/drm/i915/i915_gem.c               |  18 +-
 drivers/gpu/drm/i915/i915_gem_fence_reg.h     |   1 -
 drivers/gpu/drm/i915/i915_gem_gtt.h           |   1 +
 drivers/gpu/drm/i915/i915_gpu_error.c         | 104 +++--
 drivers/gpu/drm/i915/i915_gpu_error.h         |  28 +-
 drivers/gpu/drm/i915/i915_request.c           |  47 ---
 drivers/gpu/drm/i915/i915_reset.c             | 383 ++++++++----------
 drivers/gpu/drm/i915/i915_reset.h             |   3 +
 drivers/gpu/drm/i915/intel_engine_cs.c        |   6 +-
 drivers/gpu/drm/i915/intel_guc_submission.c   |   5 +-
 drivers/gpu/drm/i915/intel_hangcheck.c        |  28 +-
 drivers/gpu/drm/i915/intel_lrc.c              |  91 ++---
 drivers/gpu/drm/i915/intel_overlay.c          |   2 -
 drivers/gpu/drm/i915/intel_ringbuffer.c       |  91 +++--
 drivers/gpu/drm/i915/intel_ringbuffer.h       |  17 +-
 .../gpu/drm/i915/selftests/intel_hangcheck.c  |  57 +--
 .../drm/i915/selftests/intel_workarounds.c    |   3 -
 .../gpu/drm/i915/selftests/mock_gem_device.c  |   4 +-
 20 files changed, 395 insertions(+), 537 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index e80903114ca8..3b995f9fdc06 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1285,8 +1285,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 		seq_puts(m, "Wedged\n");
 	if (test_bit(I915_RESET_BACKOFF, &dev_priv->gpu_error.flags))
 		seq_puts(m, "Reset in progress: struct_mutex backoff\n");
-	if (test_bit(I915_RESET_HANDOFF, &dev_priv->gpu_error.flags))
-		seq_puts(m, "Reset in progress: reset handoff to waiter\n");
 	if (waitqueue_active(&dev_priv->gpu_error.wait_queue))
 		seq_puts(m, "Waiter holding struct mutex\n");
 	if (waitqueue_active(&dev_priv->gpu_error.reset_queue))
@@ -1322,15 +1320,15 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 		struct rb_node *rb;
 
 		seq_printf(m, "%s:\n", engine->name);
-		seq_printf(m, "\tseqno = %x [current %x, last %x]\n",
+		seq_printf(m, "\tseqno = %x [current %x, last %x], %dms ago\n",
 			   engine->hangcheck.seqno, seqno[id],
-			   intel_engine_last_submit(engine));
-		seq_printf(m, "\twaiters? %s, fake irq active? %s, stalled? %s, wedged? %s\n",
+			   intel_engine_last_submit(engine),
+			   jiffies_to_msecs(jiffies -
+					    engine->hangcheck.action_timestamp));
+		seq_printf(m, "\twaiters? %s, fake irq active? %s\n",
 			   yesno(intel_engine_has_waiter(engine)),
 			   yesno(test_bit(engine->id,
-					  &dev_priv->gpu_error.missed_irq_rings)),
-			   yesno(engine->hangcheck.stalled),
-			   yesno(engine->hangcheck.wedged));
+					  &dev_priv->gpu_error.missed_irq_rings)));
 
 		spin_lock_irq(&b->rb_lock);
 		for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
@@ -1344,11 +1342,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
 			   (long long)engine->hangcheck.acthd,
 			   (long long)acthd[id]);
-		seq_printf(m, "\taction = %s(%d) %d ms ago\n",
-			   hangcheck_action_to_str(engine->hangcheck.action),
-			   engine->hangcheck.action,
-			   jiffies_to_msecs(jiffies -
-					    engine->hangcheck.action_timestamp));
 
 		if (engine->id == RCS) {
 			seq_puts(m, "\tinstdone read =\n");
@@ -3912,8 +3905,6 @@ static int
 i915_wedged_set(void *data, u64 val)
 {
 	struct drm_i915_private *i915 = data;
-	struct intel_engine_cs *engine;
-	unsigned int tmp;
 
 	/*
 	 * There is no safeguard against this debugfs entry colliding
@@ -3926,18 +3917,8 @@ i915_wedged_set(void *data, u64 val)
 	if (i915_reset_backoff(&i915->gpu_error))
 		return -EAGAIN;
 
-	for_each_engine_masked(engine, i915, val, tmp) {
-		engine->hangcheck.seqno = intel_engine_get_seqno(engine);
-		engine->hangcheck.stalled = true;
-	}
-
 	i915_handle_error(i915, val, I915_ERROR_CAPTURE,
 			  "Manually set wedged engine mask = %llx", val);
-
-	wait_on_bit(&i915->gpu_error.flags,
-		    I915_RESET_HANDOFF,
-		    TASK_UNINTERRUPTIBLE);
-
 	return 0;
 }
 
@@ -4092,13 +4073,8 @@ i915_drop_caches_set(void *data, u64 val)
 		mutex_unlock(&i915->drm.struct_mutex);
 	}
 
-	if (val & DROP_RESET_ACTIVE &&
-	    i915_terminally_wedged(&i915->gpu_error)) {
+	if (val & DROP_RESET_ACTIVE && i915_terminally_wedged(&i915->gpu_error))
 		i915_handle_error(i915, ALL_ENGINES, 0, NULL);
-		wait_on_bit(&i915->gpu_error.flags,
-			    I915_RESET_HANDOFF,
-			    TASK_UNINTERRUPTIBLE);
-	}
 
 	fs_reclaim_acquire(GFP_KERNEL);
 	if (val & DROP_BOUND)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 3c111ad09922..0133d1da3d3c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3001,11 +3001,6 @@ static inline bool i915_reset_backoff(struct i915_gpu_error *error)
 	return unlikely(test_bit(I915_RESET_BACKOFF, &error->flags));
 }
 
-static inline bool i915_reset_handoff(struct i915_gpu_error *error)
-{
-	return unlikely(test_bit(I915_RESET_HANDOFF, &error->flags));
-}
-
 static inline bool i915_terminally_wedged(struct i915_gpu_error *error)
 {
 	return unlikely(test_bit(I915_WEDGED, &error->flags));
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 44c82a6b9934..dcbe644869b3 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -659,11 +659,6 @@ i915_gem_object_wait(struct drm_i915_gem_object *obj,
 		     struct intel_rps_client *rps_client)
 {
 	might_sleep();
-#if IS_ENABLED(CONFIG_LOCKDEP)
-	GEM_BUG_ON(debug_locks &&
-		   !!lockdep_is_held(&obj->base.dev->struct_mutex) !=
-		   !!(flags & I915_WAIT_LOCKED));
-#endif
 	GEM_BUG_ON(timeout < 0);
 
 	timeout = i915_gem_object_wait_reservation(obj->resv,
@@ -4495,8 +4490,6 @@ void i915_gem_sanitize(struct drm_i915_private *i915)
 
 	GEM_TRACE("\n");
 
-	mutex_lock(&i915->drm.struct_mutex);
-
 	wakeref = intel_runtime_pm_get(i915);
 	intel_uncore_forcewake_get(i915, FORCEWAKE_ALL);
 
@@ -4522,6 +4515,7 @@ void i915_gem_sanitize(struct drm_i915_private *i915)
 	intel_uncore_forcewake_put(i915, FORCEWAKE_ALL);
 	intel_runtime_pm_put(i915, wakeref);
 
+	mutex_lock(&i915->drm.struct_mutex);
 	i915_gem_contexts_lost(i915);
 	mutex_unlock(&i915->drm.struct_mutex);
 }
@@ -4536,6 +4530,8 @@ int i915_gem_suspend(struct drm_i915_private *i915)
 	wakeref = intel_runtime_pm_get(i915);
 	intel_suspend_gt_powersave(i915);
 
+	flush_workqueue(i915->wq);
+
 	mutex_lock(&i915->drm.struct_mutex);
 
 	/*
@@ -4565,11 +4561,9 @@ int i915_gem_suspend(struct drm_i915_private *i915)
 	i915_retire_requests(i915); /* ensure we flush after wedging */
 
 	mutex_unlock(&i915->drm.struct_mutex);
+	i915_reset_flush(i915);
 
-	intel_uc_suspend(i915);
-
-	cancel_delayed_work_sync(&i915->gpu_error.hangcheck_work);
-	cancel_delayed_work_sync(&i915->gt.retire_work);
+	drain_delayed_work(&i915->gt.retire_work);
 
 	/*
 	 * As the idle_work is rearming if it detects a race, play safe and
@@ -4577,6 +4571,8 @@ int i915_gem_suspend(struct drm_i915_private *i915)
 	 */
 	drain_delayed_work(&i915->gt.idle_work);
 
+	intel_uc_suspend(i915);
+
 	/*
 	 * Assert that we successfully flushed all the work and
 	 * reset the GPU back to its idle, low power state.
diff --git a/drivers/gpu/drm/i915/i915_gem_fence_reg.h b/drivers/gpu/drm/i915/i915_gem_fence_reg.h
index 99a31ded4dfd..09dcaf14121b 100644
--- a/drivers/gpu/drm/i915/i915_gem_fence_reg.h
+++ b/drivers/gpu/drm/i915/i915_gem_fence_reg.h
@@ -50,4 +50,3 @@ struct drm_i915_fence_reg {
 };
 
 #endif
-
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 9229b03d629b..a0039ea97cdc 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -39,6 +39,7 @@
 #include <linux/pagevec.h>
 
 #include "i915_request.h"
+#include "i915_reset.h"
 #include "i915_selftest.h"
 #include "i915_timeline.h"
 
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 1f8e80e31b49..4eef0462489c 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -533,10 +533,7 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
 	err_printf(m, "  waiting: %s\n", yesno(ee->waiting));
 	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
 	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
-	err_printf(m, "  hangcheck stall: %s\n", yesno(ee->hangcheck_stalled));
-	err_printf(m, "  hangcheck action: %s\n",
-		   hangcheck_action_to_str(ee->hangcheck_action));
-	err_printf(m, "  hangcheck action timestamp: %dms (%lu%s)\n",
+	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
 		   jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
 		   ee->hangcheck_timestamp,
 		   ee->hangcheck_timestamp == epoch ? "; epoch" : "");
@@ -684,15 +681,15 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 		   jiffies_to_msecs(error->capture - error->epoch));
 
 	for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
-		if (error->engine[i].hangcheck_stalled &&
-		    error->engine[i].context.pid) {
-			err_printf(m, "Active process (on ring %s): %s [%d], score %d%s\n",
-				   engine_name(m->i915, i),
-				   error->engine[i].context.comm,
-				   error->engine[i].context.pid,
-				   error->engine[i].context.ban_score,
-				   bannable(&error->engine[i].context));
-		}
+		if (!error->engine[i].context.pid)
+			continue;
+
+		err_printf(m, "Active process (on ring %s): %s [%d], score %d%s\n",
+			   engine_name(m->i915, i),
+			   error->engine[i].context.comm,
+			   error->engine[i].context.pid,
+			   error->engine[i].context.ban_score,
+			   bannable(&error->engine[i].context));
 	}
 	err_printf(m, "Reset count: %u\n", error->reset_count);
 	err_printf(m, "Suspend count: %u\n", error->suspend_count);
@@ -1144,7 +1141,8 @@ static u32 capture_error_bo(struct drm_i915_error_buffer *err,
 	return i;
 }
 
-/* Generate a semi-unique error code. The code is not meant to have meaning, The
+/*
+ * Generate a semi-unique error code. The code is not meant to have meaning, The
  * code's only purpose is to try to prevent false duplicated bug reports by
  * grossly estimating a GPU error state.
  *
@@ -1153,29 +1151,23 @@ static u32 capture_error_bo(struct drm_i915_error_buffer *err,
  *
  * It's only a small step better than a random number in its current form.
  */
-static u32 i915_error_generate_code(struct drm_i915_private *dev_priv,
-				    struct i915_gpu_state *error,
-				    int *engine_id)
+static u32 i915_error_generate_code(struct i915_gpu_state *error,
+				    unsigned long engine_mask)
 {
-	u32 error_code = 0;
-	int i;
-
-	/* IPEHR would be an ideal way to detect errors, as it's the gross
+	/*
+	 * IPEHR would be an ideal way to detect errors, as it's the gross
 	 * measure of "the command that hung." However, has some very common
 	 * synchronization commands which almost always appear in the case
 	 * strictly a client bug. Use instdone to differentiate those some.
 	 */
-	for (i = 0; i < I915_NUM_ENGINES; i++) {
-		if (error->engine[i].hangcheck_stalled) {
-			if (engine_id)
-				*engine_id = i;
+	if (engine_mask) {
+		struct drm_i915_error_engine *ee =
+			&error->engine[ffs(engine_mask)];
 
-			return error->engine[i].ipehr ^
-			       error->engine[i].instdone.instdone;
-		}
+		return ee->ipehr ^ ee->instdone.instdone;
 	}
 
-	return error_code;
+	return 0;
 }
 
 static void gem_record_fences(struct i915_gpu_state *error)
@@ -1338,9 +1330,8 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
 	}
 
 	ee->idle = intel_engine_is_idle(engine);
-	ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
-	ee->hangcheck_action = engine->hangcheck.action;
-	ee->hangcheck_stalled = engine->hangcheck.stalled;
+	if (!ee->idle)
+		ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
 	ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
 						  engine);
 
@@ -1783,31 +1774,35 @@ static void capture_reg_state(struct i915_gpu_state *error)
 	error->pgtbl_er = I915_READ(PGTBL_ER);
 }
 
-static void i915_error_capture_msg(struct drm_i915_private *dev_priv,
-				   struct i915_gpu_state *error,
-				   u32 engine_mask,
-				   const char *error_msg)
+static const char *
+error_msg(struct i915_gpu_state *error, unsigned long engines, const char *msg)
 {
-	u32 ecode;
-	int engine_id = -1, len;
+	int len;
+	int i;
 
-	ecode = i915_error_generate_code(dev_priv, error, &engine_id);
+	for (i = 0; i < ARRAY_SIZE(error->engine); i++)
+		if (!error->engine[i].context.pid)
+			engines &= ~BIT(i);
 
 	len = scnprintf(error->error_msg, sizeof(error->error_msg),
-			"GPU HANG: ecode %d:%d:0x%08x",
-			INTEL_GEN(dev_priv), engine_id, ecode);
-
-	if (engine_id != -1 && error->engine[engine_id].context.pid)
+			"GPU HANG: ecode %d:%lx:0x%08x",
+			INTEL_GEN(error->i915), engines,
+			i915_error_generate_code(error, engines));
+	if (engines) {
+		/* Just show the first executing process, more is confusing */
+		i = ffs(engines);
 		len += scnprintf(error->error_msg + len,
 				 sizeof(error->error_msg) - len,
 				 ", in %s [%d]",
-				 error->engine[engine_id].context.comm,
-				 error->engine[engine_id].context.pid);
+				 error->engine[i].context.comm,
+				 error->engine[i].context.pid);
+	}
+	if (msg)
+		len += scnprintf(error->error_msg + len,
+				 sizeof(error->error_msg) - len,
+				 ", %s", msg);
 
-	scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
-		  ", reason: %s, action: %s",
-		  error_msg,
-		  engine_mask ? "reset" : "continue");
+	return error->error_msg;
 }
 
 static void capture_gen_state(struct i915_gpu_state *error)
@@ -1847,7 +1842,7 @@ static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
 	for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
 		const struct drm_i915_error_engine *ee = &error->engine[i];
 
-		if (ee->hangcheck_stalled &&
+		if (ee->hangcheck_timestamp &&
 		    time_before(ee->hangcheck_timestamp, epoch))
 			epoch = ee->hangcheck_timestamp;
 	}
@@ -1921,7 +1916,7 @@ i915_capture_gpu_state(struct drm_i915_private *i915)
  * i915_capture_error_state - capture an error record for later analysis
  * @i915: i915 device
  * @engine_mask: the mask of engines triggering the hang
- * @error_msg: a message to insert into the error capture header
+ * @msg: a message to insert into the error capture header
  *
  * Should be called when an error is detected (either a hang or an error
  * interrupt) to capture error state from the time of the error.  Fills
@@ -1929,8 +1924,8 @@ i915_capture_gpu_state(struct drm_i915_private *i915)
  * to pick up.
  */
 void i915_capture_error_state(struct drm_i915_private *i915,
-			      u32 engine_mask,
-			      const char *error_msg)
+			      unsigned long engine_mask,
+			      const char *msg)
 {
 	static bool warned;
 	struct i915_gpu_state *error;
@@ -1946,8 +1941,7 @@ void i915_capture_error_state(struct drm_i915_private *i915,
 	if (IS_ERR(error))
 		return;
 
-	i915_error_capture_msg(i915, error, engine_mask, error_msg);
-	DRM_INFO("%s\n", error->error_msg);
+	dev_info(i915->drm.dev, "%s\n", error_msg(error, engine_mask, msg));
 
 	if (!error->simulated) {
 		spin_lock_irqsave(&i915->gpu_error.lock, flags);
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
index 604291f7762d..231173786eae 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.h
+++ b/drivers/gpu/drm/i915/i915_gpu_error.h
@@ -85,8 +85,6 @@ struct i915_gpu_state {
 		bool waiting;
 		int num_waiters;
 		unsigned long hangcheck_timestamp;
-		bool hangcheck_stalled;
-		enum intel_engine_hangcheck_action hangcheck_action;
 		struct i915_address_space *vm;
 		int num_requests;
 		u32 reset_count;
@@ -197,6 +195,8 @@ struct i915_gpu_state {
 	struct scatterlist *sgl, *fit;
 };
 
+struct i915_gpu_restart;
+
 struct i915_gpu_error {
 	/* For hangcheck timer */
 #define DRM_I915_HANGCHECK_PERIOD 1500 /* in ms */
@@ -247,15 +247,6 @@ struct i915_gpu_error {
 	 * i915_mutex_lock_interruptible()?). I915_RESET_BACKOFF serves a
 	 * secondary role in preventing two concurrent global reset attempts.
 	 *
-	 * #I915_RESET_HANDOFF - To perform the actual GPU reset, we need the
-	 * struct_mutex. We try to acquire the struct_mutex in the reset worker,
-	 * but it may be held by some long running waiter (that we cannot
-	 * interrupt without causing trouble). Once we are ready to do the GPU
-	 * reset, we set the I915_RESET_HANDOFF bit and wakeup any waiters. If
-	 * they already hold the struct_mutex and want to participate they can
-	 * inspect the bit and do the reset directly, otherwise the worker
-	 * waits for the struct_mutex.
-	 *
 	 * #I915_RESET_ENGINE[num_engines] - Since the driver doesn't need to
 	 * acquire the struct_mutex to reset an engine, we need an explicit
 	 * flag to prevent two concurrent reset attempts in the same engine.
@@ -269,20 +260,13 @@ struct i915_gpu_error {
 	 */
 	unsigned long flags;
 #define I915_RESET_BACKOFF	0
-#define I915_RESET_HANDOFF	1
-#define I915_RESET_MODESET	2
-#define I915_RESET_ENGINE	3
+#define I915_RESET_MODESET	1
+#define I915_RESET_ENGINE	2
 #define I915_WEDGED		(BITS_PER_LONG - 1)
 
 	/** Number of times an engine has been reset */
 	u32 reset_engine_count[I915_NUM_ENGINES];
 
-	/** Set of stalled engines with guilty requests, in the current reset */
-	u32 stalled_mask;
-
-	/** Reason for the current *global* reset */
-	const char *reason;
-
 	struct mutex wedge_mutex; /* serialises wedging/unwedging */
 
 	/**
@@ -299,6 +283,8 @@ struct i915_gpu_error {
 
 	/* For missed irq/seqno simulation. */
 	unsigned long test_irq_rings;
+
+	struct i915_gpu_restart *restart;
 };
 
 struct drm_i915_error_state_buf {
@@ -320,7 +306,7 @@ void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...);
 
 struct i915_gpu_state *i915_capture_gpu_state(struct drm_i915_private *i915);
 void i915_capture_error_state(struct drm_i915_private *dev_priv,
-			      u32 engine_mask,
+			      unsigned long engine_mask,
 			      const char *error_msg);
 
 static inline struct i915_gpu_state *
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index ddc35e9dc0c0..f4241a17e2ad 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1083,18 +1083,6 @@ static bool __i915_spin_request(const struct i915_request *rq,
 	return false;
 }
 
-static bool __i915_wait_request_check_and_reset(struct i915_request *request)
-{
-	struct i915_gpu_error *error = &request->i915->gpu_error;
-
-	if (likely(!i915_reset_handoff(error)))
-		return false;
-
-	__set_current_state(TASK_RUNNING);
-	i915_reset(request->i915, error->stalled_mask, error->reason);
-	return true;
-}
-
 /**
  * i915_request_wait - wait until execution of request has finished
  * @rq: the request to wait upon
@@ -1120,17 +1108,10 @@ long i915_request_wait(struct i915_request *rq,
 {
 	const int state = flags & I915_WAIT_INTERRUPTIBLE ?
 		TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
-	wait_queue_head_t *errq = &rq->i915->gpu_error.wait_queue;
-	DEFINE_WAIT_FUNC(reset, default_wake_function);
 	DEFINE_WAIT_FUNC(exec, default_wake_function);
 	struct intel_wait wait;
 
 	might_sleep();
-#if IS_ENABLED(CONFIG_LOCKDEP)
-	GEM_BUG_ON(debug_locks &&
-		   !!lockdep_is_held(&rq->i915->drm.struct_mutex) !=
-		   !!(flags & I915_WAIT_LOCKED));
-#endif
 	GEM_BUG_ON(timeout < 0);
 
 	if (i915_request_completed(rq))
@@ -1140,11 +1121,7 @@ long i915_request_wait(struct i915_request *rq,
 		return -ETIME;
 
 	trace_i915_request_wait_begin(rq, flags);
-
 	add_wait_queue(&rq->execute, &exec);
-	if (flags & I915_WAIT_LOCKED)
-		add_wait_queue(errq, &reset);
-
 	intel_wait_init(&wait);
 	if (flags & I915_WAIT_PRIORITY)
 		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
@@ -1155,10 +1132,6 @@ long i915_request_wait(struct i915_request *rq,
 		if (intel_wait_update_request(&wait, rq))
 			break;
 
-		if (flags & I915_WAIT_LOCKED &&
-		    __i915_wait_request_check_and_reset(rq))
-			continue;
-
 		if (signal_pending_state(state, current)) {
 			timeout = -ERESTARTSYS;
 			goto complete;
@@ -1188,9 +1161,6 @@ long i915_request_wait(struct i915_request *rq,
 		 */
 		goto wakeup;
 
-	if (flags & I915_WAIT_LOCKED)
-		__i915_wait_request_check_and_reset(rq);
-
 	for (;;) {
 		if (signal_pending_state(state, current)) {
 			timeout = -ERESTARTSYS;
@@ -1214,21 +1184,6 @@ long i915_request_wait(struct i915_request *rq,
 		if (i915_request_completed(rq))
 			break;
 
-		/*
-		 * If the GPU is hung, and we hold the lock, reset the GPU
-		 * and then check for completion. On a full reset, the engine's
-		 * HW seqno will be advanced passed us and we are complete.
-		 * If we do a partial reset, we have to wait for the GPU to
-		 * resume and update the breadcrumb.
-		 *
-		 * If we don't hold the mutex, we can just wait for the worker
-		 * to come along and update the breadcrumb (either directly
-		 * itself, or indirectly by recovering the GPU).
-		 */
-		if (flags & I915_WAIT_LOCKED &&
-		    __i915_wait_request_check_and_reset(rq))
-			continue;
-
 		/* Only spin if we know the GPU is processing this request */
 		if (__i915_spin_request(rq, wait.seqno, state, 2))
 			break;
@@ -1242,8 +1197,6 @@ long i915_request_wait(struct i915_request *rq,
 	intel_engine_remove_wait(rq->engine, &wait);
 complete:
 	__set_current_state(TASK_RUNNING);
-	if (flags & I915_WAIT_LOCKED)
-		remove_wait_queue(errq, &reset);
 	remove_wait_queue(&rq->execute, &exec);
 	trace_i915_request_wait_end(rq);
 
diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index 33408c4e6358..c01acc832518 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -5,6 +5,7 @@
  */
 
 #include <linux/sched/mm.h>
+#include <linux/stop_machine.h>
 
 #include "i915_drv.h"
 #include "i915_gpu_error.h"
@@ -14,27 +15,31 @@
 
 #define RESET_MAX_RETRIES 3
 
+/* XXX How to handle concurrent GGTT updates using tiling registers? */
+#define RESET_UNDER_STOP_MACHINE 0
+
 static void engine_skip_context(struct i915_request *rq)
 {
 	struct intel_engine_cs *engine = rq->engine;
 	struct i915_gem_context *hung_ctx = rq->gem_context;
 	struct i915_timeline *timeline = rq->timeline;
-	unsigned long flags;
 
+	lockdep_assert_held(&engine->timeline.lock);
 	GEM_BUG_ON(timeline == &engine->timeline);
 
-	spin_lock_irqsave(&engine->timeline.lock, flags);
 	spin_lock(&timeline->lock);
 
-	list_for_each_entry_continue(rq, &engine->timeline.requests, link)
-		if (rq->gem_context == hung_ctx)
-			i915_request_skip(rq, -EIO);
+	if (rq->global_seqno) {
+		list_for_each_entry_continue(rq,
+					     &engine->timeline.requests, link)
+			if (rq->gem_context == hung_ctx)
+				i915_request_skip(rq, -EIO);
+	}
 
 	list_for_each_entry(rq, &timeline->requests, link)
 		i915_request_skip(rq, -EIO);
 
 	spin_unlock(&timeline->lock);
-	spin_unlock_irqrestore(&engine->timeline.lock, flags);
 }
 
 static void client_mark_guilty(struct drm_i915_file_private *file_priv,
@@ -61,7 +66,7 @@ static void client_mark_guilty(struct drm_i915_file_private *file_priv,
 	}
 }
 
-static void context_mark_guilty(struct i915_gem_context *ctx)
+static bool context_mark_guilty(struct i915_gem_context *ctx)
 {
 	unsigned int score;
 	bool banned, bannable;
@@ -74,7 +79,7 @@ static void context_mark_guilty(struct i915_gem_context *ctx)
 
 	/* Cool contexts don't accumulate client ban score */
 	if (!bannable)
-		return;
+		return false;
 
 	if (banned) {
 		DRM_DEBUG_DRIVER("context %s: guilty %d, score %u, banned\n",
@@ -85,6 +90,8 @@ static void context_mark_guilty(struct i915_gem_context *ctx)
 
 	if (!IS_ERR_OR_NULL(ctx->file_priv))
 		client_mark_guilty(ctx->file_priv, ctx);
+
+	return banned;
 }
 
 static void context_mark_innocent(struct i915_gem_context *ctx)
@@ -92,6 +99,21 @@ static void context_mark_innocent(struct i915_gem_context *ctx)
 	atomic_inc(&ctx->active_count);
 }
 
+void i915_reset_request(struct i915_request *rq, bool guilty)
+{
+	lockdep_assert_held(&rq->engine->timeline.lock);
+	GEM_BUG_ON(i915_request_completed(rq));
+
+	if (guilty) {
+		i915_request_skip(rq, -EIO);
+		if (context_mark_guilty(rq->gem_context))
+			engine_skip_context(rq);
+	} else {
+		dma_fence_set_error(&rq->fence, -EAGAIN);
+		context_mark_innocent(rq->gem_context);
+	}
+}
+
 static void gen3_stop_engine(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
@@ -604,11 +626,8 @@ int intel_reset_guc(struct drm_i915_private *i915)
  * Ensure irq handler finishes, and not run again.
  * Also return the active request so that we only search for it once.
  */
-static struct i915_request *
-reset_prepare_engine(struct intel_engine_cs *engine)
+static void reset_prepare_engine(struct intel_engine_cs *engine)
 {
-	struct i915_request *rq;
-
 	/*
 	 * During the reset sequence, we must prevent the engine from
 	 * entering RC6. As the context state is undefined until we restart
@@ -617,162 +636,85 @@ reset_prepare_engine(struct intel_engine_cs *engine)
 	 * GPU state upon resume, i.e. fail to restart after a reset.
 	 */
 	intel_uncore_forcewake_get(engine->i915, FORCEWAKE_ALL);
-
-	rq = engine->reset.prepare(engine);
-	if (rq && rq->fence.error == -EIO)
-		rq = ERR_PTR(-EIO); /* Previous reset failed! */
-
-	return rq;
+	engine->reset.prepare(engine);
 }
 
-static int reset_prepare(struct drm_i915_private *i915)
+static void reset_prepare(struct drm_i915_private *i915)
 {
 	struct intel_engine_cs *engine;
-	struct i915_request *rq;
 	enum intel_engine_id id;
-	int err = 0;
 
-	for_each_engine(engine, i915, id) {
-		rq = reset_prepare_engine(engine);
-		if (IS_ERR(rq)) {
-			err = PTR_ERR(rq);
-			continue;
-		}
-
-		engine->hangcheck.active_request = rq;
-	}
+	for_each_engine(engine, i915, id)
+		reset_prepare_engine(engine);
 
-	i915_gem_revoke_fences(i915);
 	intel_uc_sanitize(i915);
-
-	return err;
 }
 
-/* Returns the request if it was guilty of the hang */
-static struct i915_request *
-reset_request(struct intel_engine_cs *engine,
-	      struct i915_request *rq,
-	      bool stalled)
+static int gt_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
 {
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	int err;
+
 	/*
-	 * The guilty request will get skipped on a hung engine.
-	 *
-	 * Users of client default contexts do not rely on logical
-	 * state preserved between batches so it is safe to execute
-	 * queued requests following the hang. Non default contexts
-	 * rely on preserved state, so skipping a batch loses the
-	 * evolution of the state and it needs to be considered corrupted.
-	 * Executing more queued batches on top of corrupted state is
-	 * risky. But we take the risk by trying to advance through
-	 * the queued requests in order to make the client behaviour
-	 * more predictable around resets, by not throwing away random
-	 * amount of batches it has prepared for execution. Sophisticated
-	 * clients can use gem_reset_stats_ioctl and dma fence status
-	 * (exported via sync_file info ioctl on explicit fences) to observe
-	 * when it loses the context state and should rebuild accordingly.
-	 *
-	 * The context ban, and ultimately the client ban, mechanism are safety
-	 * valves if client submission ends up resulting in nothing more than
-	 * subsequent hangs.
+	 * Everything depends on having the GTT running, so we need to start
+	 * there.
 	 */
+	err = i915_ggtt_enable_hw(i915);
+	if (err)
+		return err;
 
-	if (i915_request_completed(rq)) {
-		GEM_TRACE("%s pardoned global=%d (fence %llx:%lld), current %d\n",
-			  engine->name, rq->global_seqno,
-			  rq->fence.context, rq->fence.seqno,
-			  intel_engine_get_seqno(engine));
-		stalled = false;
-	}
-
-	if (stalled) {
-		context_mark_guilty(rq->gem_context);
-		i915_request_skip(rq, -EIO);
+	for_each_engine(engine, i915, id)
+		intel_engine_reset(engine, stalled_mask & ENGINE_MASK(id));
 
-		/* If this context is now banned, skip all pending requests. */
-		if (i915_gem_context_is_banned(rq->gem_context))
-			engine_skip_context(rq);
-	} else {
-		/*
-		 * Since this is not the hung engine, it may have advanced
-		 * since the hang declaration. Double check by refinding
-		 * the active request at the time of the reset.
-		 */
-		rq = i915_gem_find_active_request(engine);
-		if (rq) {
-			unsigned long flags;
-
-			context_mark_innocent(rq->gem_context);
-			dma_fence_set_error(&rq->fence, -EAGAIN);
-
-			/* Rewind the engine to replay the incomplete rq */
-			spin_lock_irqsave(&engine->timeline.lock, flags);
-			rq = list_prev_entry(rq, link);
-			if (&rq->link == &engine->timeline.requests)
-				rq = NULL;
-			spin_unlock_irqrestore(&engine->timeline.lock, flags);
-		}
-	}
+	i915_gem_restore_fences(i915);
 
-	return rq;
+	return err;
 }
 
-static void reset_engine(struct intel_engine_cs *engine,
-			 struct i915_request *rq,
-			 bool stalled)
+static void reset_finish_engine(struct intel_engine_cs *engine)
 {
-	if (rq)
-		rq = reset_request(engine, rq, stalled);
-
-	/* Setup the CS to resume from the breadcrumb of the hung request */
-	engine->reset.reset(engine, rq);
+	engine->reset.finish(engine);
+	intel_uncore_forcewake_put(engine->i915, FORCEWAKE_ALL);
 }
 
-static void gt_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
+struct i915_gpu_restart {
+	struct work_struct work;
+	struct drm_i915_private *i915;
+};
+
+static void restart_work(struct work_struct *work)
 {
+	struct i915_gpu_restart *arg = container_of(work, typeof(*arg), work);
+	struct drm_i915_private *i915 = arg->i915;
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
+	intel_wakeref_t wakeref;
 
-	lockdep_assert_held(&i915->drm.struct_mutex);
-
-	i915_retire_requests(i915);
+	wakeref = intel_runtime_pm_get(i915);
+	mutex_lock(&i915->drm.struct_mutex);
+	WRITE_ONCE(i915->gpu_error.restart, NULL);
 
 	for_each_engine(engine, i915, id) {
-		struct intel_context *ce;
-
-		reset_engine(engine,
-			     engine->hangcheck.active_request,
-			     stalled_mask & ENGINE_MASK(id));
-		ce = fetch_and_zero(&engine->last_retired_context);
-		if (ce)
-			intel_context_unpin(ce);
+		struct i915_request *rq;
 
 		/*
 		 * Ostensibily, we always want a context loaded for powersaving,
 		 * so if the engine is idle after the reset, send a request
 		 * to load our scratch kernel_context.
-		 *
-		 * More mysteriously, if we leave the engine idle after a reset,
-		 * the next userspace batch may hang, with what appears to be
-		 * an incoherent read by the CS (presumably stale TLB). An
-		 * empty request appears sufficient to paper over the glitch.
 		 */
-		if (intel_engine_is_idle(engine)) {
-			struct i915_request *rq;
+		if (!intel_engine_is_idle(engine))
+			continue;
 
-			rq = i915_request_alloc(engine, i915->kernel_context);
-			if (!IS_ERR(rq))
-				i915_request_add(rq);
-		}
+		rq = i915_request_alloc(engine, i915->kernel_context);
+		if (!IS_ERR(rq))
+			i915_request_add(rq);
 	}
 
-	i915_gem_restore_fences(i915);
-}
-
-static void reset_finish_engine(struct intel_engine_cs *engine)
-{
-	engine->reset.finish(engine);
+	mutex_unlock(&i915->drm.struct_mutex);
+	intel_runtime_pm_put(i915, wakeref);
 
-	intel_uncore_forcewake_put(engine->i915, FORCEWAKE_ALL);
+	kfree(arg);
 }
 
 static void reset_finish(struct drm_i915_private *i915)
@@ -780,11 +722,30 @@ static void reset_finish(struct drm_i915_private *i915)
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 
-	lockdep_assert_held(&i915->drm.struct_mutex);
-
-	for_each_engine(engine, i915, id) {
-		engine->hangcheck.active_request = NULL;
+	for_each_engine(engine, i915, id)
 		reset_finish_engine(engine);
+}
+
+static void reset_restart(struct drm_i915_private *i915)
+{
+	struct i915_gpu_restart *arg;
+
+	/*
+	 * Following the reset, ensure that we always reload context for
+	 * powersaving, and to correct engine->last_retired_context. Since
+	 * this requires us to submit a request, queue a worker to do that
+	 * task for us to evade any locking here.
+	 */
+	if (READ_ONCE(i915->gpu_error.restart))
+		return;
+
+	arg = kmalloc(sizeof(*arg), GFP_KERNEL);
+	if (arg) {
+		arg->i915 = i915;
+		INIT_WORK(&arg->work, restart_work);
+
+		WRITE_ONCE(i915->gpu_error.restart, arg);
+		queue_work(i915->wq, &arg->work);
 	}
 }
 
@@ -873,8 +834,6 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	struct i915_timeline *tl;
 	bool ret = false;
 
-	lockdep_assert_held(&i915->drm.struct_mutex);
-
 	if (!test_bit(I915_WEDGED, &error->flags))
 		return true;
 
@@ -897,9 +856,9 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	 */
 	list_for_each_entry(tl, &i915->gt.timelines, link) {
 		struct i915_request *rq;
+		long timeout;
 
-		rq = i915_gem_active_peek(&tl->last_request,
-					  &i915->drm.struct_mutex);
+		rq = i915_gem_active_get_unlocked(&tl->last_request);
 		if (!rq)
 			continue;
 
@@ -914,12 +873,12 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 		 * and when the seqno passes the fence, the signaler
 		 * then signals the fence waking us up).
 		 */
-		if (dma_fence_default_wait(&rq->fence, true,
-					   MAX_SCHEDULE_TIMEOUT) < 0)
+		timeout = dma_fence_default_wait(&rq->fence, true,
+						 MAX_SCHEDULE_TIMEOUT);
+		i915_request_put(rq);
+		if (timeout < 0)
 			goto unlock;
 	}
-	i915_retire_requests(i915);
-	GEM_BUG_ON(i915->gt.active_requests);
 
 	intel_engines_sanitize(i915, false);
 
@@ -933,7 +892,6 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	 * context and do not require stop_machine().
 	 */
 	intel_engines_reset_default_submission(i915);
-	i915_gem_contexts_lost(i915);
 
 	GEM_TRACE("end\n");
 
@@ -946,6 +904,43 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	return ret;
 }
 
+struct __i915_reset {
+	struct drm_i915_private *i915;
+	unsigned int stalled_mask;
+};
+
+static int __i915_reset__BKL(void *data)
+{
+	struct __i915_reset *arg = data;
+	int err;
+
+	err = intel_gpu_reset(arg->i915, ALL_ENGINES);
+	if (err)
+		return err;
+
+	return gt_reset(arg->i915, arg->stalled_mask);
+}
+
+#if RESET_UNDER_STOP_MACHINE
+#define __do_reset(fn, arg) stop_machine(fn, arg, NULL)
+#else
+#define __do_reset(fn, arg) fn(arg)
+#endif
+
+static int do_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
+{
+	struct __i915_reset arg = { i915, stalled_mask };
+	int err, i;
+
+	err = __do_reset(__i915_reset__BKL, &arg);
+	for (i = 0; err && i < RESET_MAX_RETRIES; i++) {
+		msleep(100);
+		err = __do_reset(__i915_reset__BKL, &arg);
+	}
+
+	return err;
+}
+
 /**
  * i915_reset - reset chip after a hang
  * @i915: #drm_i915_private to reset
@@ -971,31 +966,22 @@ void i915_reset(struct drm_i915_private *i915,
 {
 	struct i915_gpu_error *error = &i915->gpu_error;
 	int ret;
-	int i;
 
 	GEM_TRACE("flags=%lx\n", error->flags);
 
 	might_sleep();
-	lockdep_assert_held(&i915->drm.struct_mutex);
 	assert_rpm_wakelock_held(i915);
 	GEM_BUG_ON(!test_bit(I915_RESET_BACKOFF, &error->flags));
 
-	if (!test_bit(I915_RESET_HANDOFF, &error->flags))
-		return;
-
 	/* Clear any previous failed attempts at recovery. Time to try again. */
 	if (!i915_gem_unset_wedged(i915))
-		goto wakeup;
+		return;
 
 	if (reason)
 		dev_notice(i915->drm.dev, "Resetting chip for %s\n", reason);
 	error->reset_count++;
 
-	ret = reset_prepare(i915);
-	if (ret) {
-		dev_err(i915->drm.dev, "GPU recovery failed\n");
-		goto taint;
-	}
+	reset_prepare(i915);
 
 	if (!intel_has_gpu_reset(i915)) {
 		if (i915_modparams.reset)
@@ -1005,32 +991,11 @@ void i915_reset(struct drm_i915_private *i915,
 		goto error;
 	}
 
-	for (i = 0; i < RESET_MAX_RETRIES; i++) {
-		ret = intel_gpu_reset(i915, ALL_ENGINES);
-		if (ret == 0)
-			break;
-
-		msleep(100);
-	}
-	if (ret) {
+	if (do_reset(i915, stalled_mask)) {
 		dev_err(i915->drm.dev, "Failed to reset chip\n");
 		goto taint;
 	}
 
-	/* Ok, now get things going again... */
-
-	/*
-	 * Everything depends on having the GTT running, so we need to start
-	 * there.
-	 */
-	ret = i915_ggtt_enable_hw(i915);
-	if (ret) {
-		DRM_ERROR("Failed to re-enable GGTT following reset (%d)\n",
-			  ret);
-		goto error;
-	}
-
-	gt_reset(i915, stalled_mask);
 	intel_overlay_reset(i915);
 
 	/*
@@ -1052,9 +1017,8 @@ void i915_reset(struct drm_i915_private *i915,
 
 finish:
 	reset_finish(i915);
-wakeup:
-	clear_bit(I915_RESET_HANDOFF, &error->flags);
-	wake_up_bit(&error->flags, I915_RESET_HANDOFF);
+	if (!i915_terminally_wedged(error))
+		reset_restart(i915);
 	return;
 
 taint:
@@ -1073,7 +1037,6 @@ void i915_reset(struct drm_i915_private *i915,
 	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
 error:
 	i915_gem_set_wedged(i915);
-	i915_retire_requests(i915);
 	goto finish;
 }
 
@@ -1099,18 +1062,16 @@ static inline int intel_gt_reset_engine(struct drm_i915_private *i915,
 int i915_reset_engine(struct intel_engine_cs *engine, const char *msg)
 {
 	struct i915_gpu_error *error = &engine->i915->gpu_error;
-	struct i915_request *active_request;
 	int ret;
 
 	GEM_TRACE("%s flags=%lx\n", engine->name, error->flags);
 	GEM_BUG_ON(!test_bit(I915_RESET_ENGINE + engine->id, &error->flags));
 
-	active_request = reset_prepare_engine(engine);
-	if (IS_ERR_OR_NULL(active_request)) {
-		/* Either the previous reset failed, or we pardon the reset. */
-		ret = PTR_ERR(active_request);
-		goto out;
-	}
+	if (i915_seqno_passed(intel_engine_get_seqno(engine),
+			      intel_engine_last_submit(engine)))
+		return 0;
+
+	reset_prepare_engine(engine);
 
 	if (msg)
 		dev_notice(engine->i915->drm.dev,
@@ -1134,7 +1095,7 @@ int i915_reset_engine(struct intel_engine_cs *engine, const char *msg)
 	 * active request and can drop it, adjust head to skip the offending
 	 * request to resume executing remaining requests in the queue.
 	 */
-	reset_engine(engine, active_request, true);
+	intel_engine_reset(engine, true);
 
 	/*
 	 * The engine and its registers (and workarounds in case of render)
@@ -1171,30 +1132,7 @@ static void i915_reset_device(struct drm_i915_private *i915,
 	i915_wedge_on_timeout(&w, i915, 5 * HZ) {
 		intel_prepare_reset(i915);
 
-		error->reason = reason;
-		error->stalled_mask = engine_mask;
-
-		/* Signal that locked waiters should reset the GPU */
-		smp_mb__before_atomic();
-		set_bit(I915_RESET_HANDOFF, &error->flags);
-		wake_up_all(&error->wait_queue);
-
-		/*
-		 * Wait for anyone holding the lock to wakeup, without
-		 * blocking indefinitely on struct_mutex.
-		 */
-		do {
-			if (mutex_trylock(&i915->drm.struct_mutex)) {
-				i915_reset(i915, engine_mask, reason);
-				mutex_unlock(&i915->drm.struct_mutex);
-			}
-		} while (wait_on_bit_timeout(&error->flags,
-					     I915_RESET_HANDOFF,
-					     TASK_UNINTERRUPTIBLE,
-					     1));
-
-		error->stalled_mask = 0;
-		error->reason = NULL;
+		i915_reset(i915, engine_mask, reason);
 
 		intel_finish_reset(i915);
 	}
@@ -1350,6 +1288,25 @@ void i915_handle_error(struct drm_i915_private *i915,
 	intel_runtime_pm_put(i915, wakeref);
 }
 
+bool i915_reset_flush(struct drm_i915_private *i915)
+{
+	int err;
+
+	cancel_delayed_work_sync(&i915->gpu_error.hangcheck_work);
+
+	flush_workqueue(i915->wq);
+	GEM_BUG_ON(READ_ONCE(i915->gpu_error.restart));
+
+	mutex_lock(&i915->drm.struct_mutex);
+	err = i915_gem_wait_for_idle(i915,
+				     I915_WAIT_LOCKED |
+				     I915_WAIT_FOR_IDLE_BOOST,
+				     MAX_SCHEDULE_TIMEOUT);
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	return !err;
+}
+
 static void i915_wedge_me(struct work_struct *work)
 {
 	struct i915_wedge_me *w = container_of(work, typeof(*w), work.work);
diff --git a/drivers/gpu/drm/i915/i915_reset.h b/drivers/gpu/drm/i915/i915_reset.h
index b6a519bde67d..f2d347f319df 100644
--- a/drivers/gpu/drm/i915/i915_reset.h
+++ b/drivers/gpu/drm/i915/i915_reset.h
@@ -29,6 +29,9 @@ void i915_reset(struct drm_i915_private *i915,
 int i915_reset_engine(struct intel_engine_cs *engine,
 		      const char *reason);
 
+void i915_reset_request(struct i915_request *rq, bool guilty);
+bool i915_reset_flush(struct drm_i915_private *i915);
+
 bool intel_has_gpu_reset(struct drm_i915_private *i915);
 bool intel_has_reset_engine(struct drm_i915_private *i915);
 
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 235a2d70d671..815a05fe0e1e 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -1119,10 +1119,8 @@ void intel_engines_sanitize(struct drm_i915_private *i915, bool force)
 	if (!reset_engines(i915) && !force)
 		return;
 
-	for_each_engine(engine, i915, id) {
-		if (engine->reset.reset)
-			engine->reset.reset(engine, NULL);
-	}
+	for_each_engine(engine, i915, id)
+		intel_engine_reset(engine, false);
 }
 
 /**
diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index 349ae5844f24..45e2db683fe5 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -834,8 +834,7 @@ static void guc_submission_tasklet(unsigned long data)
 	spin_unlock_irqrestore(&engine->timeline.lock, flags);
 }
 
-static struct i915_request *
-guc_reset_prepare(struct intel_engine_cs *engine)
+static void guc_reset_prepare(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
 
@@ -861,8 +860,6 @@ guc_reset_prepare(struct intel_engine_cs *engine)
 	 */
 	if (engine->i915->guc.preempt_wq)
 		flush_workqueue(engine->i915->guc.preempt_wq);
-
-	return i915_gem_find_active_request(engine);
 }
 
 /*
diff --git a/drivers/gpu/drm/i915/intel_hangcheck.c b/drivers/gpu/drm/i915/intel_hangcheck.c
index 741441daae32..5662d6fed523 100644
--- a/drivers/gpu/drm/i915/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/intel_hangcheck.c
@@ -25,6 +25,17 @@
 #include "i915_drv.h"
 #include "i915_reset.h"
 
+struct hangcheck {
+	u64 acthd;
+	u32 seqno;
+	enum intel_engine_hangcheck_action action;
+	unsigned long action_timestamp;
+	int deadlock;
+	struct intel_instdone instdone;
+	bool wedged:1;
+	bool stalled:1;
+};
+
 static bool instdone_unchanged(u32 current_instdone, u32 *old_instdone)
 {
 	u32 tmp = current_instdone | *old_instdone;
@@ -119,25 +130,22 @@ engine_stuck(struct intel_engine_cs *engine, u64 acthd)
 }
 
 static void hangcheck_load_sample(struct intel_engine_cs *engine,
-				  struct intel_engine_hangcheck *hc)
+				  struct hangcheck *hc)
 {
 	hc->acthd = intel_engine_get_active_head(engine);
 	hc->seqno = intel_engine_get_seqno(engine);
 }
 
 static void hangcheck_store_sample(struct intel_engine_cs *engine,
-				   const struct intel_engine_hangcheck *hc)
+				   const struct hangcheck *hc)
 {
 	engine->hangcheck.acthd = hc->acthd;
 	engine->hangcheck.seqno = hc->seqno;
-	engine->hangcheck.action = hc->action;
-	engine->hangcheck.stalled = hc->stalled;
-	engine->hangcheck.wedged = hc->wedged;
 }
 
 static enum intel_engine_hangcheck_action
 hangcheck_get_action(struct intel_engine_cs *engine,
-		     const struct intel_engine_hangcheck *hc)
+		     const struct hangcheck *hc)
 {
 	if (engine->hangcheck.seqno != hc->seqno)
 		return ENGINE_ACTIVE_SEQNO;
@@ -149,7 +157,7 @@ hangcheck_get_action(struct intel_engine_cs *engine,
 }
 
 static void hangcheck_accumulate_sample(struct intel_engine_cs *engine,
-					struct intel_engine_hangcheck *hc)
+					struct hangcheck *hc)
 {
 	unsigned long timeout = I915_ENGINE_DEAD_TIMEOUT;
 
@@ -265,19 +273,19 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 	intel_uncore_arm_unclaimed_mmio_detection(dev_priv);
 
 	for_each_engine(engine, dev_priv, id) {
-		struct intel_engine_hangcheck hc;
+		struct hangcheck hc;
 
 		hangcheck_load_sample(engine, &hc);
 		hangcheck_accumulate_sample(engine, &hc);
 		hangcheck_store_sample(engine, &hc);
 
-		if (engine->hangcheck.stalled) {
+		if (hc.stalled) {
 			hung |= intel_engine_flag(engine);
 			if (hc.action != ENGINE_DEAD)
 				stuck |= intel_engine_flag(engine);
 		}
 
-		if (engine->hangcheck.wedged)
+		if (hc.wedged)
 			wedged |= intel_engine_flag(engine);
 	}
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 6cef458f6b43..10354bdf8237 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -136,6 +136,7 @@
 #include <drm/i915_drm.h>
 #include "i915_drv.h"
 #include "i915_gem_render_state.h"
+#include "i915_reset.h"
 #include "i915_vgpu.h"
 #include "intel_lrc_reg.h"
 #include "intel_mocs.h"
@@ -371,7 +372,8 @@ static void unwind_wa_tail(struct i915_request *rq)
 	assert_ring_tail_valid(rq->ring, rq->tail);
 }
 
-static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
+static struct i915_request *
+__unwind_incomplete_requests(struct intel_engine_cs *engine)
 {
 	struct i915_request *rq, *rn, *active = NULL;
 	struct list_head *uninitialized_var(pl);
@@ -413,6 +415,8 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
 		list_move_tail(&active->sched.link,
 			       i915_sched_lookup_priolist(engine, prio));
 	}
+
+	return active;
 }
 
 void
@@ -1840,11 +1844,9 @@ static int gen8_init_common_ring(struct intel_engine_cs *engine)
 	return 0;
 }
 
-static struct i915_request *
-execlists_reset_prepare(struct intel_engine_cs *engine)
+static void execlists_reset_prepare(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
-	struct i915_request *request, *active;
 	unsigned long flags;
 
 	GEM_TRACE("%s: depth<-%d\n", engine->name,
@@ -1860,59 +1862,21 @@ execlists_reset_prepare(struct intel_engine_cs *engine)
 	 * prevents the race.
 	 */
 	__tasklet_disable_sync_once(&execlists->tasklet);
+	GEM_BUG_ON(!reset_in_progress(execlists));
 
+	/* And flush any current direct submission. */
 	spin_lock_irqsave(&engine->timeline.lock, flags);
-
-	/*
-	 * We want to flush the pending context switches, having disabled
-	 * the tasklet above, we can assume exclusive access to the execlists.
-	 * For this allows us to catch up with an inflight preemption event,
-	 * and avoid blaming an innocent request if the stall was due to the
-	 * preemption itself.
-	 */
-	process_csb(engine);
-
-	/*
-	 * The last active request can then be no later than the last request
-	 * now in ELSP[0]. So search backwards from there, so that if the GPU
-	 * has advanced beyond the last CSB update, it will be pardoned.
-	 */
-	active = NULL;
-	request = port_request(execlists->port);
-	if (request) {
-		/*
-		 * Prevent the breadcrumb from advancing before we decide
-		 * which request is currently active.
-		 */
-		intel_engine_stop_cs(engine);
-
-		list_for_each_entry_from_reverse(request,
-						 &engine->timeline.requests,
-						 link) {
-			if (__i915_request_completed(request,
-						     request->global_seqno))
-				break;
-
-			active = request;
-		}
-	}
-
+	process_csb(engine); /* drain preemption events */
 	spin_unlock_irqrestore(&engine->timeline.lock, flags);
-
-	return active;
 }
 
-static void execlists_reset(struct intel_engine_cs *engine,
-			    struct i915_request *request)
+static void execlists_reset(struct intel_engine_cs *engine, bool stalled)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
+	struct i915_request *rq;
 	unsigned long flags;
 	u32 *regs;
 
-	GEM_TRACE("%s request global=%d, current=%d\n",
-		  engine->name, request ? request->global_seqno : 0,
-		  intel_engine_get_seqno(engine));
-
 	spin_lock_irqsave(&engine->timeline.lock, flags);
 
 	/*
@@ -1927,12 +1891,18 @@ static void execlists_reset(struct intel_engine_cs *engine,
 	execlists_cancel_port_requests(execlists);
 
 	/* Push back any incomplete requests for replay after the reset. */
-	__unwind_incomplete_requests(engine);
+	rq = __unwind_incomplete_requests(engine);
 
 	/* Following the reset, we need to reload the CSB read/write pointers */
 	reset_csb_pointers(&engine->execlists);
 
-	spin_unlock_irqrestore(&engine->timeline.lock, flags);
+	GEM_TRACE("%s seqno=%d, current=%d, stalled? %s\n",
+		  engine->name,
+		  rq ? rq->global_seqno : 0,
+		  intel_engine_get_seqno(engine),
+		  yesno(stalled));
+	if (!rq)
+		goto out_unlock;
 
 	/*
 	 * If the request was innocent, we leave the request in the ELSP
@@ -1945,8 +1915,9 @@ static void execlists_reset(struct intel_engine_cs *engine,
 	 * and have to at least restore the RING register in the context
 	 * image back to the expected values to skip over the guilty request.
 	 */
-	if (!request || request->fence.error != -EIO)
-		return;
+	i915_reset_request(rq, stalled);
+	if (!stalled)
+		goto out_unlock;
 
 	/*
 	 * We want a simple context + ring to execute the breadcrumb update.
@@ -1956,7 +1927,7 @@ static void execlists_reset(struct intel_engine_cs *engine,
 	 * future request will be after userspace has had the opportunity
 	 * to recreate its own state.
 	 */
-	regs = request->hw_context->lrc_reg_state;
+	regs = rq->hw_context->lrc_reg_state;
 	if (engine->pinned_default_state) {
 		memcpy(regs, /* skip restoring the vanilla PPHWSP */
 		       engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE,
@@ -1964,17 +1935,14 @@ static void execlists_reset(struct intel_engine_cs *engine,
 	}
 
 	/* Move the RING_HEAD onto the breadcrumb, past the hanging batch */
-	request->ring->head = intel_ring_wrap(request->ring, request->postfix);
+	rq->ring->head = intel_ring_wrap(rq->ring, rq->postfix);
+	intel_ring_update_space(rq->ring);
 
-	execlists_init_reg_state(regs, request->gem_context, engine,
-				 request->ring);
+	execlists_init_reg_state(regs, rq->gem_context, engine, rq->ring);
+	__execlists_update_reg_state(engine, rq->hw_context);
 
-	__execlists_update_reg_state(engine, request->hw_context);
-
-	intel_ring_update_space(request->ring);
-
-	/* Reset WaIdleLiteRestore:bdw,skl as well */
-	unwind_wa_tail(request);
+out_unlock:
+	spin_unlock_irqrestore(&engine->timeline.lock, flags);
 }
 
 static void execlists_reset_finish(struct intel_engine_cs *engine)
@@ -1987,6 +1955,7 @@ static void execlists_reset_finish(struct intel_engine_cs *engine)
 	 * to sleep before we restart and reload a context.
 	 *
 	 */
+	GEM_BUG_ON(!reset_in_progress(execlists));
 	if (!RB_EMPTY_ROOT(&execlists->queue.rb_root))
 		execlists->tasklet.func(execlists->tasklet.data);
 
diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
index c300e5787b3c..a9238fd07e30 100644
--- a/drivers/gpu/drm/i915/intel_overlay.c
+++ b/drivers/gpu/drm/i915/intel_overlay.c
@@ -480,8 +480,6 @@ void intel_overlay_reset(struct drm_i915_private *dev_priv)
 	if (!overlay)
 		return;
 
-	intel_overlay_release_old_vid(overlay);
-
 	overlay->old_xscale = 0;
 	overlay->old_yscale = 0;
 	overlay->crtc = NULL;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 09c90475168a..a9efc8c71254 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -33,6 +33,7 @@
 
 #include "i915_drv.h"
 #include "i915_gem_render_state.h"
+#include "i915_reset.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
 #include "intel_workarounds.h"
@@ -711,52 +712,80 @@ static int init_ring_common(struct intel_engine_cs *engine)
 	return ret;
 }
 
-static struct i915_request *reset_prepare(struct intel_engine_cs *engine)
+static void reset_prepare(struct intel_engine_cs *engine)
 {
 	intel_engine_stop_cs(engine);
-	return i915_gem_find_active_request(engine);
 }
 
-static void skip_request(struct i915_request *rq)
+static void reset_ring(struct intel_engine_cs *engine, bool stalled)
 {
-	void *vaddr = rq->ring->vaddr;
+	struct i915_timeline *tl = &engine->timeline;
+	struct i915_request *pos, *rq;
+	unsigned long flags;
 	u32 head;
 
-	head = rq->infix;
-	if (rq->postfix < head) {
-		memset32(vaddr + head, MI_NOOP,
-			 (rq->ring->size - head) / sizeof(u32));
-		head = 0;
+	rq = NULL;
+	spin_lock_irqsave(&tl->lock, flags);
+	list_for_each_entry(pos, &tl->requests, link) {
+		if (!__i915_request_completed(pos, pos->global_seqno)) {
+			rq = pos;
+			break;
+		}
 	}
-	memset32(vaddr + head, MI_NOOP, (rq->postfix - head) / sizeof(u32));
-}
-
-static void reset_ring(struct intel_engine_cs *engine, struct i915_request *rq)
-{
-	GEM_TRACE("%s request global=%d, current=%d\n",
-		  engine->name, rq ? rq->global_seqno : 0,
-		  intel_engine_get_seqno(engine));
 
+	GEM_TRACE("%s seqno=%d, current=%d, stalled? %s\n",
+		  engine->name,
+		  rq ? rq->global_seqno : 0,
+		  intel_engine_get_seqno(engine),
+		  yesno(stalled));
 	/*
-	 * Try to restore the logical GPU state to match the continuation
-	 * of the request queue. If we skip the context/PD restore, then
-	 * the next request may try to execute assuming that its context
-	 * is valid and loaded on the GPU and so may try to access invalid
-	 * memory, prompting repeated GPU hangs.
+	 * The guilty request will get skipped on a hung engine.
 	 *
-	 * If the request was guilty, we still restore the logical state
-	 * in case the next request requires it (e.g. the aliasing ppgtt),
-	 * but skip over the hung batch.
+	 * Users of client default contexts do not rely on logical
+	 * state preserved between batches so it is safe to execute
+	 * queued requests following the hang. Non default contexts
+	 * rely on preserved state, so skipping a batch loses the
+	 * evolution of the state and it needs to be considered corrupted.
+	 * Executing more queued batches on top of corrupted state is
+	 * risky. But we take the risk by trying to advance through
+	 * the queued requests in order to make the client behaviour
+	 * more predictable around resets, by not throwing away random
+	 * amount of batches it has prepared for execution. Sophisticated
+	 * clients can use gem_reset_stats_ioctl and dma fence status
+	 * (exported via sync_file info ioctl on explicit fences) to observe
+	 * when it loses the context state and should rebuild accordingly.
 	 *
-	 * If the request was innocent, we try to replay the request with
-	 * the restored context.
+	 * The context ban, and ultimately the client ban, mechanism are safety
+	 * valves if client submission ends up resulting in nothing more than
+	 * subsequent hangs.
 	 */
+
 	if (rq) {
-		/* If the rq hung, jump to its breadcrumb and skip the batch */
-		rq->ring->head = intel_ring_wrap(rq->ring, rq->head);
-		if (rq->fence.error == -EIO)
-			skip_request(rq);
+		/*
+		 * Try to restore the logical GPU state to match the
+		 * continuation of the request queue. If we skip the
+		 * context/PD restore, then the next request may try to execute
+		 * assuming that its context is valid and loaded on the GPU and
+		 * so may try to access invalid memory, prompting repeated GPU
+		 * hangs.
+		 *
+		 * If the request was guilty, we still restore the logical
+		 * state in case the next request requires it (e.g. the
+		 * aliasing ppgtt), but skip over the hung batch.
+		 *
+		 * If the request was innocent, we try to replay the request
+		 * with the restored context.
+		 */
+		i915_reset_request(rq, stalled);
+
+		GEM_BUG_ON(rq->ring != engine->buffer);
+		head = rq->head;
+	} else {
+		head = engine->buffer->tail;
 	}
+	engine->buffer->head = intel_ring_wrap(engine->buffer, head);
+
+	spin_unlock_irqrestore(&tl->lock, flags);
 }
 
 static void reset_finish(struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 5ad46c2fbc0f..f2effd001540 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -120,13 +120,8 @@ struct intel_instdone {
 struct intel_engine_hangcheck {
 	u64 acthd;
 	u32 seqno;
-	enum intel_engine_hangcheck_action action;
 	unsigned long action_timestamp;
-	int deadlock;
 	struct intel_instdone instdone;
-	struct i915_request *active_request;
-	bool stalled:1;
-	bool wedged:1;
 };
 
 struct intel_ring {
@@ -444,9 +439,8 @@ struct intel_engine_cs {
 	int		(*init_hw)(struct intel_engine_cs *engine);
 
 	struct {
-		struct i915_request *(*prepare)(struct intel_engine_cs *engine);
-		void (*reset)(struct intel_engine_cs *engine,
-			      struct i915_request *rq);
+		void (*prepare)(struct intel_engine_cs *engine);
+		void (*reset)(struct intel_engine_cs *engine, bool stalled);
 		void (*finish)(struct intel_engine_cs *engine);
 	} reset;
 
@@ -1018,6 +1012,13 @@ gen8_emit_ggtt_write(u32 *cs, u32 value, u32 gtt_offset)
 	return cs;
 }
 
+static inline void intel_engine_reset(struct intel_engine_cs *engine,
+				      bool stalled)
+{
+	if (engine->reset.reset)
+		engine->reset.reset(engine, stalled);
+}
+
 void intel_engines_sanitize(struct drm_i915_private *i915, bool force);
 
 bool intel_engine_is_idle(struct intel_engine_cs *engine);
diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 12550b55c42f..67431355cd6e 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -363,9 +363,7 @@ static int igt_global_reset(void *arg)
 	/* Check that we can issue a global GPU reset */
 
 	igt_global_reset_lock(i915);
-	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
 
-	mutex_lock(&i915->drm.struct_mutex);
 	reset_count = i915_reset_count(&i915->gpu_error);
 
 	i915_reset(i915, ALL_ENGINES, NULL);
@@ -374,9 +372,7 @@ static int igt_global_reset(void *arg)
 		pr_err("No GPU reset recorded!\n");
 		err = -EINVAL;
 	}
-	mutex_unlock(&i915->drm.struct_mutex);
 
-	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
 	igt_global_reset_unlock(i915);
 
 	if (i915_terminally_wedged(&i915->gpu_error))
@@ -399,9 +395,7 @@ static int igt_wedged_reset(void *arg)
 	i915_gem_set_wedged(i915);
 	GEM_BUG_ON(!i915_terminally_wedged(&i915->gpu_error));
 
-	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
 	i915_reset(i915, ALL_ENGINES, NULL);
-	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
 
 	intel_runtime_pm_put(i915, wakeref);
 	mutex_unlock(&i915->drm.struct_mutex);
@@ -511,7 +505,7 @@ static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
 				break;
 			}
 
-			if (!wait_for_idle(engine)) {
+			if (!i915_reset_flush(i915)) {
 				struct drm_printer p =
 					drm_info_printer(i915->drm.dev);
 
@@ -903,20 +897,13 @@ static int igt_reset_engines(void *arg)
 	return 0;
 }
 
-static u32 fake_hangcheck(struct i915_request *rq, u32 mask)
+static u32 fake_hangcheck(struct drm_i915_private *i915, u32 mask)
 {
-	struct i915_gpu_error *error = &rq->i915->gpu_error;
-	u32 reset_count = i915_reset_count(error);
-
-	error->stalled_mask = mask;
-
-	/* set_bit() must be after we have setup the backchannel (mask) */
-	smp_mb__before_atomic();
-	set_bit(I915_RESET_HANDOFF, &error->flags);
+	u32 count = i915_reset_count(&i915->gpu_error);
 
-	wake_up_all(&error->wait_queue);
+	i915_reset(i915, mask, NULL);
 
-	return reset_count;
+	return count;
 }
 
 static int igt_reset_wait(void *arg)
@@ -962,7 +949,7 @@ static int igt_reset_wait(void *arg)
 		goto out_rq;
 	}
 
-	reset_count = fake_hangcheck(rq, ALL_ENGINES);
+	reset_count = fake_hangcheck(i915, ALL_ENGINES);
 
 	timeout = i915_request_wait(rq, I915_WAIT_LOCKED, 10);
 	if (timeout < 0) {
@@ -972,7 +959,6 @@ static int igt_reset_wait(void *arg)
 		goto out_rq;
 	}
 
-	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
 	if (i915_reset_count(&i915->gpu_error) == reset_count) {
 		pr_err("No GPU reset recorded!\n");
 		err = -EINVAL;
@@ -1162,7 +1148,7 @@ static int __igt_reset_evict_vma(struct drm_i915_private *i915,
 	}
 
 out_reset:
-	fake_hangcheck(rq, intel_engine_flag(rq->engine));
+	fake_hangcheck(rq->i915, intel_engine_flag(rq->engine));
 
 	if (tsk) {
 		struct igt_wedge_me w;
@@ -1341,12 +1327,7 @@ static int igt_reset_queue(void *arg)
 				goto fini;
 			}
 
-			reset_count = fake_hangcheck(prev, ENGINE_MASK(id));
-
-			i915_reset(i915, ENGINE_MASK(id), NULL);
-
-			GEM_BUG_ON(test_bit(I915_RESET_HANDOFF,
-					    &i915->gpu_error.flags));
+			reset_count = fake_hangcheck(i915, ENGINE_MASK(id));
 
 			if (prev->fence.error != -EIO) {
 				pr_err("GPU reset not recorded on hanging request [fence.error=%d]!\n",
@@ -1565,6 +1546,7 @@ static int igt_atomic_reset_engine(struct intel_engine_cs *engine,
 		pr_err("%s(%s): Failed to start request %llx, at %x\n",
 		       __func__, engine->name,
 		       rq->fence.seqno, hws_seqno(&h, rq));
+		i915_gem_set_wedged(i915);
 		err = -EIO;
 	}
 
@@ -1588,7 +1570,6 @@ static int igt_atomic_reset_engine(struct intel_engine_cs *engine,
 static void force_reset(struct drm_i915_private *i915)
 {
 	i915_gem_set_wedged(i915);
-	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
 	i915_reset(i915, 0, NULL);
 }
 
@@ -1618,6 +1599,26 @@ static int igt_atomic_reset(void *arg)
 	if (i915_terminally_wedged(&i915->gpu_error))
 		goto unlock;
 
+	if (intel_has_gpu_reset(i915)) {
+		const typeof(*phases) *p;
+
+		for (p = phases; p->name; p++) {
+			GEM_TRACE("intel_gpu_reset under %s\n", p->name);
+
+			p->critical_section_begin();
+			err = intel_gpu_reset(i915, ALL_ENGINES);
+			p->critical_section_end();
+
+			if (err) {
+				pr_err("intel_gpu_reset failed under %s\n",
+				       p->name);
+				goto out;
+			}
+		}
+
+		force_reset(i915);
+	}
+
 	if (intel_has_reset_engine(i915)) {
 		struct intel_engine_cs *engine;
 		enum intel_engine_id id;
diff --git a/drivers/gpu/drm/i915/selftests/intel_workarounds.c b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
index a8cac56be835..b15c4f26c593 100644
--- a/drivers/gpu/drm/i915/selftests/intel_workarounds.c
+++ b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
@@ -214,7 +214,6 @@ static int check_whitelist(struct i915_gem_context *ctx,
 
 static int do_device_reset(struct intel_engine_cs *engine)
 {
-	set_bit(I915_RESET_HANDOFF, &engine->i915->gpu_error.flags);
 	i915_reset(engine->i915, ENGINE_MASK(engine->id), "live_workarounds");
 	return 0;
 }
@@ -394,7 +393,6 @@ static int
 live_gpu_reset_gt_engine_workarounds(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
-	struct i915_gpu_error *error = &i915->gpu_error;
 	intel_wakeref_t wakeref;
 	struct wa_lists lists;
 	bool ok;
@@ -413,7 +411,6 @@ live_gpu_reset_gt_engine_workarounds(void *arg)
 	if (!ok)
 		goto out;
 
-	set_bit(I915_RESET_HANDOFF, &error->flags);
 	i915_reset(i915, ALL_ENGINES, "live_workarounds");
 
 	ok = verify_gt_engine_wa(i915, &lists, "after reset");
diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
index 5477ad4a7e7d..8ab5a2688a0c 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
@@ -58,8 +58,8 @@ static void mock_device_release(struct drm_device *dev)
 	i915_gem_contexts_lost(i915);
 	mutex_unlock(&i915->drm.struct_mutex);
 
-	cancel_delayed_work_sync(&i915->gt.retire_work);
-	cancel_delayed_work_sync(&i915->gt.idle_work);
+	drain_delayed_work(&i915->gt.retire_work);
+	drain_delayed_work(&i915->gt.idle_work);
 	i915_gem_drain_workqueue(i915);
 
 	mutex_lock(&i915->drm.struct_mutex);
-- 
2.20.1

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

Re: [PATCH 10/33] drm/i915: Remove GPU reset dependence on struct_mutex

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Now that the submission backends are controlled via their own spinlocks,
> with a wave of a magic wand we can lift the struct_mutex requirement
> around GPU reset. That is we allow the submission frontend (userspace)
> to keep on submitting while we process the GPU reset as we can suspend
> the backend independently.
>
> The major change is around the backoff/handoff strategy for performing
> the reset. With no mutex deadlock, we no longer have to coordinate with
> any waiter, and just perform the reset immediately.
>
> Testcase: igt/gem_mmap_gtt/hang # regresses
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/i915_debugfs.c           |  38 +-
>  drivers/gpu/drm/i915/i915_drv.h               |   5 -
>  drivers/gpu/drm/i915/i915_gem.c               |  18 +-
>  drivers/gpu/drm/i915/i915_gem_fence_reg.h     |   1 -
>  drivers/gpu/drm/i915/i915_gem_gtt.h           |   1 +
>  drivers/gpu/drm/i915/i915_gpu_error.c         | 104 +++--
>  drivers/gpu/drm/i915/i915_gpu_error.h         |  28 +-
>  drivers/gpu/drm/i915/i915_request.c           |  47 ---
>  drivers/gpu/drm/i915/i915_reset.c             | 383 ++++++++----------
>  drivers/gpu/drm/i915/i915_reset.h             |   3 +
>  drivers/gpu/drm/i915/intel_engine_cs.c        |   6 +-
>  drivers/gpu/drm/i915/intel_guc_submission.c   |   5 +-
>  drivers/gpu/drm/i915/intel_hangcheck.c        |  28 +-
>  drivers/gpu/drm/i915/intel_lrc.c              |  91 ++---
>  drivers/gpu/drm/i915/intel_overlay.c          |   2 -
>  drivers/gpu/drm/i915/intel_ringbuffer.c       |  91 +++--
>  drivers/gpu/drm/i915/intel_ringbuffer.h       |  17 +-
>  .../gpu/drm/i915/selftests/intel_hangcheck.c  |  57 +--
>  .../drm/i915/selftests/intel_workarounds.c    |   3 -
>  .../gpu/drm/i915/selftests/mock_gem_device.c  |   4 +-
>  20 files changed, 395 insertions(+), 537 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index e80903114ca8..3b995f9fdc06 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -1285,8 +1285,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
>  		seq_puts(m, "Wedged\n");
>  	if (test_bit(I915_RESET_BACKOFF, &dev_priv->gpu_error.flags))
>  		seq_puts(m, "Reset in progress: struct_mutex backoff\n");
> -	if (test_bit(I915_RESET_HANDOFF, &dev_priv->gpu_error.flags))
> -		seq_puts(m, "Reset in progress: reset handoff to waiter\n");
>  	if (waitqueue_active(&dev_priv->gpu_error.wait_queue))
>  		seq_puts(m, "Waiter holding struct mutex\n");
>  	if (waitqueue_active(&dev_priv->gpu_error.reset_queue))
> @@ -1322,15 +1320,15 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
>  		struct rb_node *rb;
>  
>  		seq_printf(m, "%s:\n", engine->name);
> -		seq_printf(m, "\tseqno = %x [current %x, last %x]\n",
> +		seq_printf(m, "\tseqno = %x [current %x, last %x], %dms ago\n",
>  			   engine->hangcheck.seqno, seqno[id],
> -			   intel_engine_last_submit(engine));
> -		seq_printf(m, "\twaiters? %s, fake irq active? %s, stalled? %s, wedged? %s\n",
> +			   intel_engine_last_submit(engine),
> +			   jiffies_to_msecs(jiffies -
> +					    engine->hangcheck.action_timestamp));
> +		seq_printf(m, "\twaiters? %s, fake irq active? %s\n",
>  			   yesno(intel_engine_has_waiter(engine)),
>  			   yesno(test_bit(engine->id,
> -					  &dev_priv->gpu_error.missed_irq_rings)),
> -			   yesno(engine->hangcheck.stalled),
> -			   yesno(engine->hangcheck.wedged));
> +					  &dev_priv->gpu_error.missed_irq_rings)));
>  
>  		spin_lock_irq(&b->rb_lock);
>  		for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
> @@ -1344,11 +1342,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
>  		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
>  			   (long long)engine->hangcheck.acthd,
>  			   (long long)acthd[id]);
> -		seq_printf(m, "\taction = %s(%d) %d ms ago\n",
> -			   hangcheck_action_to_str(engine->hangcheck.action),
> -			   engine->hangcheck.action,
> -			   jiffies_to_msecs(jiffies -
> -					    engine->hangcheck.action_timestamp));
>  
>  		if (engine->id == RCS) {
>  			seq_puts(m, "\tinstdone read =\n");
> @@ -3912,8 +3905,6 @@ static int
>  i915_wedged_set(void *data, u64 val)
>  {
>  	struct drm_i915_private *i915 = data;
> -	struct intel_engine_cs *engine;
> -	unsigned int tmp;
>  
>  	/*
>  	 * There is no safeguard against this debugfs entry colliding
> @@ -3926,18 +3917,8 @@ i915_wedged_set(void *data, u64 val)
>  	if (i915_reset_backoff(&i915->gpu_error))
>  		return -EAGAIN;
>  
> -	for_each_engine_masked(engine, i915, val, tmp) {
> -		engine->hangcheck.seqno = intel_engine_get_seqno(engine);
> -		engine->hangcheck.stalled = true;
> -	}
> -
>  	i915_handle_error(i915, val, I915_ERROR_CAPTURE,
>  			  "Manually set wedged engine mask = %llx", val);
> -
> -	wait_on_bit(&i915->gpu_error.flags,
> -		    I915_RESET_HANDOFF,
> -		    TASK_UNINTERRUPTIBLE);
> -
>  	return 0;
>  }
>  
> @@ -4092,13 +4073,8 @@ i915_drop_caches_set(void *data, u64 val)
>  		mutex_unlock(&i915->drm.struct_mutex);
>  	}
>  
> -	if (val & DROP_RESET_ACTIVE &&
> -	    i915_terminally_wedged(&i915->gpu_error)) {
> +	if (val & DROP_RESET_ACTIVE && i915_terminally_wedged(&i915->gpu_error))
>  		i915_handle_error(i915, ALL_ENGINES, 0, NULL);
> -		wait_on_bit(&i915->gpu_error.flags,
> -			    I915_RESET_HANDOFF,
> -			    TASK_UNINTERRUPTIBLE);
> -	}
>  
>  	fs_reclaim_acquire(GFP_KERNEL);
>  	if (val & DROP_BOUND)
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 3c111ad09922..0133d1da3d3c 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -3001,11 +3001,6 @@ static inline bool i915_reset_backoff(struct i915_gpu_error *error)
>  	return unlikely(test_bit(I915_RESET_BACKOFF, &error->flags));
>  }
>  
> -static inline bool i915_reset_handoff(struct i915_gpu_error *error)
> -{
> -	return unlikely(test_bit(I915_RESET_HANDOFF, &error->flags));
> -}
> -
>  static inline bool i915_terminally_wedged(struct i915_gpu_error *error)
>  {
>  	return unlikely(test_bit(I915_WEDGED, &error->flags));
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 44c82a6b9934..dcbe644869b3 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -659,11 +659,6 @@ i915_gem_object_wait(struct drm_i915_gem_object *obj,
>  		     struct intel_rps_client *rps_client)
>  {
>  	might_sleep();
> -#if IS_ENABLED(CONFIG_LOCKDEP)
> -	GEM_BUG_ON(debug_locks &&
> -		   !!lockdep_is_held(&obj->base.dev->struct_mutex) !=
> -		   !!(flags & I915_WAIT_LOCKED));
> -#endif
>  	GEM_BUG_ON(timeout < 0);
>  
>  	timeout = i915_gem_object_wait_reservation(obj->resv,
> @@ -4495,8 +4490,6 @@ void i915_gem_sanitize(struct drm_i915_private *i915)
>  
>  	GEM_TRACE("\n");
>  
> -	mutex_lock(&i915->drm.struct_mutex);
> -
>  	wakeref = intel_runtime_pm_get(i915);
>  	intel_uncore_forcewake_get(i915, FORCEWAKE_ALL);
>  
> @@ -4522,6 +4515,7 @@ void i915_gem_sanitize(struct drm_i915_private *i915)
>  	intel_uncore_forcewake_put(i915, FORCEWAKE_ALL);
>  	intel_runtime_pm_put(i915, wakeref);
>  
> +	mutex_lock(&i915->drm.struct_mutex);
>  	i915_gem_contexts_lost(i915);
>  	mutex_unlock(&i915->drm.struct_mutex);
>  }
> @@ -4536,6 +4530,8 @@ int i915_gem_suspend(struct drm_i915_private *i915)
>  	wakeref = intel_runtime_pm_get(i915);
>  	intel_suspend_gt_powersave(i915);
>  
> +	flush_workqueue(i915->wq);
> +
>  	mutex_lock(&i915->drm.struct_mutex);
>  
>  	/*
> @@ -4565,11 +4561,9 @@ int i915_gem_suspend(struct drm_i915_private *i915)
>  	i915_retire_requests(i915); /* ensure we flush after wedging */
>  
>  	mutex_unlock(&i915->drm.struct_mutex);
> +	i915_reset_flush(i915);
>  
> -	intel_uc_suspend(i915);
> -
> -	cancel_delayed_work_sync(&i915->gpu_error.hangcheck_work);
> -	cancel_delayed_work_sync(&i915->gt.retire_work);
> +	drain_delayed_work(&i915->gt.retire_work);
>  
>  	/*
>  	 * As the idle_work is rearming if it detects a race, play safe and
> @@ -4577,6 +4571,8 @@ int i915_gem_suspend(struct drm_i915_private *i915)
>  	 */
>  	drain_delayed_work(&i915->gt.idle_work);
>  
> +	intel_uc_suspend(i915);
> +
>  	/*
>  	 * Assert that we successfully flushed all the work and
>  	 * reset the GPU back to its idle, low power state.
> diff --git a/drivers/gpu/drm/i915/i915_gem_fence_reg.h b/drivers/gpu/drm/i915/i915_gem_fence_reg.h
> index 99a31ded4dfd..09dcaf14121b 100644
> --- a/drivers/gpu/drm/i915/i915_gem_fence_reg.h
> +++ b/drivers/gpu/drm/i915/i915_gem_fence_reg.h
> @@ -50,4 +50,3 @@ struct drm_i915_fence_reg {
>  };
>  
>  #endif
> -
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
> index 9229b03d629b..a0039ea97cdc 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
> @@ -39,6 +39,7 @@
>  #include <linux/pagevec.h>
>  
>  #include "i915_request.h"
> +#include "i915_reset.h"
>  #include "i915_selftest.h"
>  #include "i915_timeline.h"
>  
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
> index 1f8e80e31b49..4eef0462489c 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
> @@ -533,10 +533,7 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
>  	err_printf(m, "  waiting: %s\n", yesno(ee->waiting));
>  	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
>  	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
> -	err_printf(m, "  hangcheck stall: %s\n", yesno(ee->hangcheck_stalled));
> -	err_printf(m, "  hangcheck action: %s\n",
> -		   hangcheck_action_to_str(ee->hangcheck_action));
> -	err_printf(m, "  hangcheck action timestamp: %dms (%lu%s)\n",
> +	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
>  		   jiffies_to_msecs(ee->hangcheck_timestamp - epoch),
>  		   ee->hangcheck_timestamp,
>  		   ee->hangcheck_timestamp == epoch ? "; epoch" : "");
> @@ -684,15 +681,15 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
>  		   jiffies_to_msecs(error->capture - error->epoch));
>  
>  	for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
> -		if (error->engine[i].hangcheck_stalled &&
> -		    error->engine[i].context.pid) {
> -			err_printf(m, "Active process (on ring %s): %s [%d], score %d%s\n",
> -				   engine_name(m->i915, i),
> -				   error->engine[i].context.comm,
> -				   error->engine[i].context.pid,
> -				   error->engine[i].context.ban_score,
> -				   bannable(&error->engine[i].context));
> -		}
> +		if (!error->engine[i].context.pid)
> +			continue;
> +
> +		err_printf(m, "Active process (on ring %s): %s [%d], score %d%s\n",
> +			   engine_name(m->i915, i),
> +			   error->engine[i].context.comm,
> +			   error->engine[i].context.pid,
> +			   error->engine[i].context.ban_score,
> +			   bannable(&error->engine[i].context));
>  	}
>  	err_printf(m, "Reset count: %u\n", error->reset_count);
>  	err_printf(m, "Suspend count: %u\n", error->suspend_count);
> @@ -1144,7 +1141,8 @@ static u32 capture_error_bo(struct drm_i915_error_buffer *err,
>  	return i;
>  }
>  
> -/* Generate a semi-unique error code. The code is not meant to have meaning, The
> +/*
> + * Generate a semi-unique error code. The code is not meant to have meaning, The
>   * code's only purpose is to try to prevent false duplicated bug reports by
>   * grossly estimating a GPU error state.
>   *
> @@ -1153,29 +1151,23 @@ static u32 capture_error_bo(struct drm_i915_error_buffer *err,
>   *
>   * It's only a small step better than a random number in its current form.
>   */
> -static u32 i915_error_generate_code(struct drm_i915_private *dev_priv,
> -				    struct i915_gpu_state *error,
> -				    int *engine_id)
> +static u32 i915_error_generate_code(struct i915_gpu_state *error,
> +				    unsigned long engine_mask)
>  {
> -	u32 error_code = 0;
> -	int i;
> -
> -	/* IPEHR would be an ideal way to detect errors, as it's the gross
> +	/*
> +	 * IPEHR would be an ideal way to detect errors, as it's the gross
>  	 * measure of "the command that hung." However, has some very common
>  	 * synchronization commands which almost always appear in the case
>  	 * strictly a client bug. Use instdone to differentiate those some.
>  	 */
> -	for (i = 0; i < I915_NUM_ENGINES; i++) {
> -		if (error->engine[i].hangcheck_stalled) {
> -			if (engine_id)
> -				*engine_id = i;
> +	if (engine_mask) {
> +		struct drm_i915_error_engine *ee =
> +			&error->engine[ffs(engine_mask)];
>  
> -			return error->engine[i].ipehr ^
> -			       error->engine[i].instdone.instdone;
> -		}
> +		return ee->ipehr ^ ee->instdone.instdone;
>  	}
>  
> -	return error_code;
> +	return 0;
>  }
>  
>  static void gem_record_fences(struct i915_gpu_state *error)
> @@ -1338,9 +1330,8 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
>  	}
>  
>  	ee->idle = intel_engine_is_idle(engine);
> -	ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
> -	ee->hangcheck_action = engine->hangcheck.action;
> -	ee->hangcheck_stalled = engine->hangcheck.stalled;
> +	if (!ee->idle)
> +		ee->hangcheck_timestamp = engine->hangcheck.action_timestamp;
>  	ee->reset_count = i915_reset_engine_count(&dev_priv->gpu_error,
>  						  engine);
>  
> @@ -1783,31 +1774,35 @@ static void capture_reg_state(struct i915_gpu_state *error)
>  	error->pgtbl_er = I915_READ(PGTBL_ER);
>  }
>  
> -static void i915_error_capture_msg(struct drm_i915_private *dev_priv,
> -				   struct i915_gpu_state *error,
> -				   u32 engine_mask,
> -				   const char *error_msg)
> +static const char *
> +error_msg(struct i915_gpu_state *error, unsigned long engines, const char *msg)
>  {
> -	u32 ecode;
> -	int engine_id = -1, len;
> +	int len;
> +	int i;
>  
> -	ecode = i915_error_generate_code(dev_priv, error, &engine_id);
> +	for (i = 0; i < ARRAY_SIZE(error->engine); i++)
> +		if (!error->engine[i].context.pid)
> +			engines &= ~BIT(i);
>  
>  	len = scnprintf(error->error_msg, sizeof(error->error_msg),
> -			"GPU HANG: ecode %d:%d:0x%08x",
> -			INTEL_GEN(dev_priv), engine_id, ecode);
> -
> -	if (engine_id != -1 && error->engine[engine_id].context.pid)
> +			"GPU HANG: ecode %d:%lx:0x%08x",
> +			INTEL_GEN(error->i915), engines,
> +			i915_error_generate_code(error, engines));
> +	if (engines) {
> +		/* Just show the first executing process, more is confusing */
> +		i = ffs(engines);
>  		len += scnprintf(error->error_msg + len,
>  				 sizeof(error->error_msg) - len,
>  				 ", in %s [%d]",
> -				 error->engine[engine_id].context.comm,
> -				 error->engine[engine_id].context.pid);
> +				 error->engine[i].context.comm,
> +				 error->engine[i].context.pid);
> +	}
> +	if (msg)
> +		len += scnprintf(error->error_msg + len,
> +				 sizeof(error->error_msg) - len,
> +				 ", %s", msg);
>  
> -	scnprintf(error->error_msg + len, sizeof(error->error_msg) - len,
> -		  ", reason: %s, action: %s",
> -		  error_msg,
> -		  engine_mask ? "reset" : "continue");
> +	return error->error_msg;
>  }
>  
>  static void capture_gen_state(struct i915_gpu_state *error)
> @@ -1847,7 +1842,7 @@ static unsigned long capture_find_epoch(const struct i915_gpu_state *error)
>  	for (i = 0; i < ARRAY_SIZE(error->engine); i++) {
>  		const struct drm_i915_error_engine *ee = &error->engine[i];
>  
> -		if (ee->hangcheck_stalled &&
> +		if (ee->hangcheck_timestamp &&
>  		    time_before(ee->hangcheck_timestamp, epoch))
>  			epoch = ee->hangcheck_timestamp;
>  	}
> @@ -1921,7 +1916,7 @@ i915_capture_gpu_state(struct drm_i915_private *i915)
>   * i915_capture_error_state - capture an error record for later analysis
>   * @i915: i915 device
>   * @engine_mask: the mask of engines triggering the hang
> - * @error_msg: a message to insert into the error capture header
> + * @msg: a message to insert into the error capture header
>   *
>   * Should be called when an error is detected (either a hang or an error
>   * interrupt) to capture error state from the time of the error.  Fills
> @@ -1929,8 +1924,8 @@ i915_capture_gpu_state(struct drm_i915_private *i915)
>   * to pick up.
>   */
>  void i915_capture_error_state(struct drm_i915_private *i915,
> -			      u32 engine_mask,
> -			      const char *error_msg)
> +			      unsigned long engine_mask,
> +			      const char *msg)
>  {
>  	static bool warned;
>  	struct i915_gpu_state *error;
> @@ -1946,8 +1941,7 @@ void i915_capture_error_state(struct drm_i915_private *i915,
>  	if (IS_ERR(error))
>  		return;
>  
> -	i915_error_capture_msg(i915, error, engine_mask, error_msg);
> -	DRM_INFO("%s\n", error->error_msg);
> +	dev_info(i915->drm.dev, "%s\n", error_msg(error, engine_mask, msg));
>  
>  	if (!error->simulated) {
>  		spin_lock_irqsave(&i915->gpu_error.lock, flags);
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
> index 604291f7762d..231173786eae 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.h
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.h
> @@ -85,8 +85,6 @@ struct i915_gpu_state {
>  		bool waiting;
>  		int num_waiters;
>  		unsigned long hangcheck_timestamp;
> -		bool hangcheck_stalled;
> -		enum intel_engine_hangcheck_action hangcheck_action;
>  		struct i915_address_space *vm;
>  		int num_requests;
>  		u32 reset_count;
> @@ -197,6 +195,8 @@ struct i915_gpu_state {
>  	struct scatterlist *sgl, *fit;
>  };
>  
> +struct i915_gpu_restart;
> +
>  struct i915_gpu_error {
>  	/* For hangcheck timer */
>  #define DRM_I915_HANGCHECK_PERIOD 1500 /* in ms */
> @@ -247,15 +247,6 @@ struct i915_gpu_error {
>  	 * i915_mutex_lock_interruptible()?). I915_RESET_BACKOFF serves a
>  	 * secondary role in preventing two concurrent global reset attempts.
>  	 *
> -	 * #I915_RESET_HANDOFF - To perform the actual GPU reset, we need the
> -	 * struct_mutex. We try to acquire the struct_mutex in the reset worker,
> -	 * but it may be held by some long running waiter (that we cannot
> -	 * interrupt without causing trouble). Once we are ready to do the GPU
> -	 * reset, we set the I915_RESET_HANDOFF bit and wakeup any waiters. If
> -	 * they already hold the struct_mutex and want to participate they can
> -	 * inspect the bit and do the reset directly, otherwise the worker
> -	 * waits for the struct_mutex.
> -	 *
>  	 * #I915_RESET_ENGINE[num_engines] - Since the driver doesn't need to
>  	 * acquire the struct_mutex to reset an engine, we need an explicit
>  	 * flag to prevent two concurrent reset attempts in the same engine.
> @@ -269,20 +260,13 @@ struct i915_gpu_error {
>  	 */
>  	unsigned long flags;
>  #define I915_RESET_BACKOFF	0
> -#define I915_RESET_HANDOFF	1
> -#define I915_RESET_MODESET	2
> -#define I915_RESET_ENGINE	3
> +#define I915_RESET_MODESET	1
> +#define I915_RESET_ENGINE	2
>  #define I915_WEDGED		(BITS_PER_LONG - 1)
>  
>  	/** Number of times an engine has been reset */
>  	u32 reset_engine_count[I915_NUM_ENGINES];
>  
> -	/** Set of stalled engines with guilty requests, in the current reset */
> -	u32 stalled_mask;
> -
> -	/** Reason for the current *global* reset */
> -	const char *reason;
> -
>  	struct mutex wedge_mutex; /* serialises wedging/unwedging */
>  
>  	/**
> @@ -299,6 +283,8 @@ struct i915_gpu_error {
>  
>  	/* For missed irq/seqno simulation. */
>  	unsigned long test_irq_rings;
> +
> +	struct i915_gpu_restart *restart;
>  };
>  
>  struct drm_i915_error_state_buf {
> @@ -320,7 +306,7 @@ void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...);
>  
>  struct i915_gpu_state *i915_capture_gpu_state(struct drm_i915_private *i915);
>  void i915_capture_error_state(struct drm_i915_private *dev_priv,
> -			      u32 engine_mask,
> +			      unsigned long engine_mask,
>  			      const char *error_msg);
>  
>  static inline struct i915_gpu_state *
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index ddc35e9dc0c0..f4241a17e2ad 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -1083,18 +1083,6 @@ static bool __i915_spin_request(const struct i915_request *rq,
>  	return false;
>  }
>  
> -static bool __i915_wait_request_check_and_reset(struct i915_request *request)
> -{
> -	struct i915_gpu_error *error = &request->i915->gpu_error;
> -
> -	if (likely(!i915_reset_handoff(error)))
> -		return false;
> -
> -	__set_current_state(TASK_RUNNING);
> -	i915_reset(request->i915, error->stalled_mask, error->reason);
> -	return true;
> -}
> -
>  /**
>   * i915_request_wait - wait until execution of request has finished
>   * @rq: the request to wait upon
> @@ -1120,17 +1108,10 @@ long i915_request_wait(struct i915_request *rq,
>  {
>  	const int state = flags & I915_WAIT_INTERRUPTIBLE ?
>  		TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
> -	wait_queue_head_t *errq = &rq->i915->gpu_error.wait_queue;
> -	DEFINE_WAIT_FUNC(reset, default_wake_function);
>  	DEFINE_WAIT_FUNC(exec, default_wake_function);
>  	struct intel_wait wait;
>  
>  	might_sleep();
> -#if IS_ENABLED(CONFIG_LOCKDEP)
> -	GEM_BUG_ON(debug_locks &&
> -		   !!lockdep_is_held(&rq->i915->drm.struct_mutex) !=
> -		   !!(flags & I915_WAIT_LOCKED));
> -#endif
>  	GEM_BUG_ON(timeout < 0);
>  
>  	if (i915_request_completed(rq))
> @@ -1140,11 +1121,7 @@ long i915_request_wait(struct i915_request *rq,
>  		return -ETIME;
>  
>  	trace_i915_request_wait_begin(rq, flags);
> -
>  	add_wait_queue(&rq->execute, &exec);
> -	if (flags & I915_WAIT_LOCKED)
> -		add_wait_queue(errq, &reset);
> -
>  	intel_wait_init(&wait);
>  	if (flags & I915_WAIT_PRIORITY)
>  		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
> @@ -1155,10 +1132,6 @@ long i915_request_wait(struct i915_request *rq,
>  		if (intel_wait_update_request(&wait, rq))
>  			break;
>  
> -		if (flags & I915_WAIT_LOCKED &&
> -		    __i915_wait_request_check_and_reset(rq))
> -			continue;
> -
>  		if (signal_pending_state(state, current)) {
>  			timeout = -ERESTARTSYS;
>  			goto complete;
> @@ -1188,9 +1161,6 @@ long i915_request_wait(struct i915_request *rq,
>  		 */
>  		goto wakeup;
>  
> -	if (flags & I915_WAIT_LOCKED)
> -		__i915_wait_request_check_and_reset(rq);
> -
>  	for (;;) {
>  		if (signal_pending_state(state, current)) {
>  			timeout = -ERESTARTSYS;
> @@ -1214,21 +1184,6 @@ long i915_request_wait(struct i915_request *rq,
>  		if (i915_request_completed(rq))
>  			break;
>  
> -		/*
> -		 * If the GPU is hung, and we hold the lock, reset the GPU
> -		 * and then check for completion. On a full reset, the engine's
> -		 * HW seqno will be advanced passed us and we are complete.
> -		 * If we do a partial reset, we have to wait for the GPU to
> -		 * resume and update the breadcrumb.
> -		 *
> -		 * If we don't hold the mutex, we can just wait for the worker
> -		 * to come along and update the breadcrumb (either directly
> -		 * itself, or indirectly by recovering the GPU).
> -		 */
> -		if (flags & I915_WAIT_LOCKED &&
> -		    __i915_wait_request_check_and_reset(rq))
> -			continue;
> -
>  		/* Only spin if we know the GPU is processing this request */
>  		if (__i915_spin_request(rq, wait.seqno, state, 2))
>  			break;
> @@ -1242,8 +1197,6 @@ long i915_request_wait(struct i915_request *rq,
>  	intel_engine_remove_wait(rq->engine, &wait);
>  complete:
>  	__set_current_state(TASK_RUNNING);
> -	if (flags & I915_WAIT_LOCKED)
> -		remove_wait_queue(errq, &reset);
>  	remove_wait_queue(&rq->execute, &exec);
>  	trace_i915_request_wait_end(rq);
>  
> diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
> index 33408c4e6358..c01acc832518 100644
> --- a/drivers/gpu/drm/i915/i915_reset.c
> +++ b/drivers/gpu/drm/i915/i915_reset.c
> @@ -5,6 +5,7 @@
>   */
>  
>  #include <linux/sched/mm.h>
> +#include <linux/stop_machine.h>
>  
>  #include "i915_drv.h"
>  #include "i915_gpu_error.h"
> @@ -14,27 +15,31 @@
>  
>  #define RESET_MAX_RETRIES 3
>  
> +/* XXX How to handle concurrent GGTT updates using tiling registers? */
> +#define RESET_UNDER_STOP_MACHINE 0

We discussed in irc kind of stop_machine_pids(pid_t *, fn)
type of way to exclude only the currently tiled clients
out of the way. 

For this patch, can't find anything obvious codewise.
For the nuances of how the hw will interact with the
changes, only the tests and time will tell. 

Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

> +
>  static void engine_skip_context(struct i915_request *rq)
>  {
>  	struct intel_engine_cs *engine = rq->engine;
>  	struct i915_gem_context *hung_ctx = rq->gem_context;
>  	struct i915_timeline *timeline = rq->timeline;
> -	unsigned long flags;
>  
> +	lockdep_assert_held(&engine->timeline.lock);
>  	GEM_BUG_ON(timeline == &engine->timeline);
>  
> -	spin_lock_irqsave(&engine->timeline.lock, flags);
>  	spin_lock(&timeline->lock);
>  
> -	list_for_each_entry_continue(rq, &engine->timeline.requests, link)
> -		if (rq->gem_context == hung_ctx)
> -			i915_request_skip(rq, -EIO);
> +	if (rq->global_seqno) {
> +		list_for_each_entry_continue(rq,
> +					     &engine->timeline.requests, link)
> +			if (rq->gem_context == hung_ctx)
> +				i915_request_skip(rq, -EIO);
> +	}
>  
>  	list_for_each_entry(rq, &timeline->requests, link)
>  		i915_request_skip(rq, -EIO);
>  
>  	spin_unlock(&timeline->lock);
> -	spin_unlock_irqrestore(&engine->timeline.lock, flags);
>  }
>  
>  static void client_mark_guilty(struct drm_i915_file_private *file_priv,
> @@ -61,7 +66,7 @@ static void client_mark_guilty(struct drm_i915_file_private *file_priv,
>  	}
>  }
>  
> -static void context_mark_guilty(struct i915_gem_context *ctx)
> +static bool context_mark_guilty(struct i915_gem_context *ctx)
>  {
>  	unsigned int score;
>  	bool banned, bannable;
> @@ -74,7 +79,7 @@ static void context_mark_guilty(struct i915_gem_context *ctx)
>  
>  	/* Cool contexts don't accumulate client ban score */
>  	if (!bannable)
> -		return;
> +		return false;
>  
>  	if (banned) {
>  		DRM_DEBUG_DRIVER("context %s: guilty %d, score %u, banned\n",
> @@ -85,6 +90,8 @@ static void context_mark_guilty(struct i915_gem_context *ctx)
>  
>  	if (!IS_ERR_OR_NULL(ctx->file_priv))
>  		client_mark_guilty(ctx->file_priv, ctx);
> +
> +	return banned;
>  }
>  
>  static void context_mark_innocent(struct i915_gem_context *ctx)
> @@ -92,6 +99,21 @@ static void context_mark_innocent(struct i915_gem_context *ctx)
>  	atomic_inc(&ctx->active_count);
>  }
>  
> +void i915_reset_request(struct i915_request *rq, bool guilty)
> +{
> +	lockdep_assert_held(&rq->engine->timeline.lock);
> +	GEM_BUG_ON(i915_request_completed(rq));
> +
> +	if (guilty) {
> +		i915_request_skip(rq, -EIO);
> +		if (context_mark_guilty(rq->gem_context))
> +			engine_skip_context(rq);
> +	} else {
> +		dma_fence_set_error(&rq->fence, -EAGAIN);
> +		context_mark_innocent(rq->gem_context);
> +	}
> +}
> +
>  static void gen3_stop_engine(struct intel_engine_cs *engine)
>  {
>  	struct drm_i915_private *dev_priv = engine->i915;
> @@ -604,11 +626,8 @@ int intel_reset_guc(struct drm_i915_private *i915)
>   * Ensure irq handler finishes, and not run again.
>   * Also return the active request so that we only search for it once.
>   */
> -static struct i915_request *
> -reset_prepare_engine(struct intel_engine_cs *engine)
> +static void reset_prepare_engine(struct intel_engine_cs *engine)
>  {
> -	struct i915_request *rq;
> -
>  	/*
>  	 * During the reset sequence, we must prevent the engine from
>  	 * entering RC6. As the context state is undefined until we restart
> @@ -617,162 +636,85 @@ reset_prepare_engine(struct intel_engine_cs *engine)
>  	 * GPU state upon resume, i.e. fail to restart after a reset.
>  	 */
>  	intel_uncore_forcewake_get(engine->i915, FORCEWAKE_ALL);
> -
> -	rq = engine->reset.prepare(engine);
> -	if (rq && rq->fence.error == -EIO)
> -		rq = ERR_PTR(-EIO); /* Previous reset failed! */
> -
> -	return rq;
> +	engine->reset.prepare(engine);
>  }
>  
> -static int reset_prepare(struct drm_i915_private *i915)
> +static void reset_prepare(struct drm_i915_private *i915)
>  {
>  	struct intel_engine_cs *engine;
> -	struct i915_request *rq;
>  	enum intel_engine_id id;
> -	int err = 0;
>  
> -	for_each_engine(engine, i915, id) {
> -		rq = reset_prepare_engine(engine);
> -		if (IS_ERR(rq)) {
> -			err = PTR_ERR(rq);
> -			continue;
> -		}
> -
> -		engine->hangcheck.active_request = rq;
> -	}
> +	for_each_engine(engine, i915, id)
> +		reset_prepare_engine(engine);
>  
> -	i915_gem_revoke_fences(i915);
>  	intel_uc_sanitize(i915);
> -
> -	return err;
>  }
>  
> -/* Returns the request if it was guilty of the hang */
> -static struct i915_request *
> -reset_request(struct intel_engine_cs *engine,
> -	      struct i915_request *rq,
> -	      bool stalled)
> +static int gt_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
>  {
> +	struct intel_engine_cs *engine;
> +	enum intel_engine_id id;
> +	int err;
> +
>  	/*
> -	 * The guilty request will get skipped on a hung engine.
> -	 *
> -	 * Users of client default contexts do not rely on logical
> -	 * state preserved between batches so it is safe to execute
> -	 * queued requests following the hang. Non default contexts
> -	 * rely on preserved state, so skipping a batch loses the
> -	 * evolution of the state and it needs to be considered corrupted.
> -	 * Executing more queued batches on top of corrupted state is
> -	 * risky. But we take the risk by trying to advance through
> -	 * the queued requests in order to make the client behaviour
> -	 * more predictable around resets, by not throwing away random
> -	 * amount of batches it has prepared for execution. Sophisticated
> -	 * clients can use gem_reset_stats_ioctl and dma fence status
> -	 * (exported via sync_file info ioctl on explicit fences) to observe
> -	 * when it loses the context state and should rebuild accordingly.
> -	 *
> -	 * The context ban, and ultimately the client ban, mechanism are safety
> -	 * valves if client submission ends up resulting in nothing more than
> -	 * subsequent hangs.
> +	 * Everything depends on having the GTT running, so we need to start
> +	 * there.
>  	 */
> +	err = i915_ggtt_enable_hw(i915);
> +	if (err)
> +		return err;
>  
> -	if (i915_request_completed(rq)) {
> -		GEM_TRACE("%s pardoned global=%d (fence %llx:%lld), current %d\n",
> -			  engine->name, rq->global_seqno,
> -			  rq->fence.context, rq->fence.seqno,
> -			  intel_engine_get_seqno(engine));
> -		stalled = false;
> -	}
> -
> -	if (stalled) {
> -		context_mark_guilty(rq->gem_context);
> -		i915_request_skip(rq, -EIO);
> +	for_each_engine(engine, i915, id)
> +		intel_engine_reset(engine, stalled_mask & ENGINE_MASK(id));
>  
> -		/* If this context is now banned, skip all pending requests. */
> -		if (i915_gem_context_is_banned(rq->gem_context))
> -			engine_skip_context(rq);
> -	} else {
> -		/*
> -		 * Since this is not the hung engine, it may have advanced
> -		 * since the hang declaration. Double check by refinding
> -		 * the active request at the time of the reset.
> -		 */
> -		rq = i915_gem_find_active_request(engine);
> -		if (rq) {
> -			unsigned long flags;
> -
> -			context_mark_innocent(rq->gem_context);
> -			dma_fence_set_error(&rq->fence, -EAGAIN);
> -
> -			/* Rewind the engine to replay the incomplete rq */
> -			spin_lock_irqsave(&engine->timeline.lock, flags);
> -			rq = list_prev_entry(rq, link);
> -			if (&rq->link == &engine->timeline.requests)
> -				rq = NULL;
> -			spin_unlock_irqrestore(&engine->timeline.lock, flags);
> -		}
> -	}
> +	i915_gem_restore_fences(i915);
>  
> -	return rq;
> +	return err;
>  }
>  
> -static void reset_engine(struct intel_engine_cs *engine,
> -			 struct i915_request *rq,
> -			 bool stalled)
> +static void reset_finish_engine(struct intel_engine_cs *engine)
>  {
> -	if (rq)
> -		rq = reset_request(engine, rq, stalled);
> -
> -	/* Setup the CS to resume from the breadcrumb of the hung request */
> -	engine->reset.reset(engine, rq);
> +	engine->reset.finish(engine);
> +	intel_uncore_forcewake_put(engine->i915, FORCEWAKE_ALL);
>  }
>  
> -static void gt_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
> +struct i915_gpu_restart {
> +	struct work_struct work;
> +	struct drm_i915_private *i915;
> +};
> +
> +static void restart_work(struct work_struct *work)
>  {
> +	struct i915_gpu_restart *arg = container_of(work, typeof(*arg), work);
> +	struct drm_i915_private *i915 = arg->i915;
>  	struct intel_engine_cs *engine;
>  	enum intel_engine_id id;
> +	intel_wakeref_t wakeref;
>  
> -	lockdep_assert_held(&i915->drm.struct_mutex);
> -
> -	i915_retire_requests(i915);
> +	wakeref = intel_runtime_pm_get(i915);
> +	mutex_lock(&i915->drm.struct_mutex);
> +	WRITE_ONCE(i915->gpu_error.restart, NULL);
>  
>  	for_each_engine(engine, i915, id) {
> -		struct intel_context *ce;
> -
> -		reset_engine(engine,
> -			     engine->hangcheck.active_request,
> -			     stalled_mask & ENGINE_MASK(id));
> -		ce = fetch_and_zero(&engine->last_retired_context);
> -		if (ce)
> -			intel_context_unpin(ce);
> +		struct i915_request *rq;
>  
>  		/*
>  		 * Ostensibily, we always want a context loaded for powersaving,
>  		 * so if the engine is idle after the reset, send a request
>  		 * to load our scratch kernel_context.
> -		 *
> -		 * More mysteriously, if we leave the engine idle after a reset,
> -		 * the next userspace batch may hang, with what appears to be
> -		 * an incoherent read by the CS (presumably stale TLB). An
> -		 * empty request appears sufficient to paper over the glitch.
>  		 */
> -		if (intel_engine_is_idle(engine)) {
> -			struct i915_request *rq;
> +		if (!intel_engine_is_idle(engine))
> +			continue;
>  
> -			rq = i915_request_alloc(engine, i915->kernel_context);
> -			if (!IS_ERR(rq))
> -				i915_request_add(rq);
> -		}
> +		rq = i915_request_alloc(engine, i915->kernel_context);
> +		if (!IS_ERR(rq))
> +			i915_request_add(rq);
>  	}
>  
> -	i915_gem_restore_fences(i915);
> -}
> -
> -static void reset_finish_engine(struct intel_engine_cs *engine)
> -{
> -	engine->reset.finish(engine);
> +	mutex_unlock(&i915->drm.struct_mutex);
> +	intel_runtime_pm_put(i915, wakeref);
>  
> -	intel_uncore_forcewake_put(engine->i915, FORCEWAKE_ALL);
> +	kfree(arg);
>  }
>  
>  static void reset_finish(struct drm_i915_private *i915)
> @@ -780,11 +722,30 @@ static void reset_finish(struct drm_i915_private *i915)
>  	struct intel_engine_cs *engine;
>  	enum intel_engine_id id;
>  
> -	lockdep_assert_held(&i915->drm.struct_mutex);
> -
> -	for_each_engine(engine, i915, id) {
> -		engine->hangcheck.active_request = NULL;
> +	for_each_engine(engine, i915, id)
>  		reset_finish_engine(engine);
> +}
> +
> +static void reset_restart(struct drm_i915_private *i915)
> +{
> +	struct i915_gpu_restart *arg;
> +
> +	/*
> +	 * Following the reset, ensure that we always reload context for
> +	 * powersaving, and to correct engine->last_retired_context. Since
> +	 * this requires us to submit a request, queue a worker to do that
> +	 * task for us to evade any locking here.
> +	 */
> +	if (READ_ONCE(i915->gpu_error.restart))
> +		return;
> +
> +	arg = kmalloc(sizeof(*arg), GFP_KERNEL);
> +	if (arg) {
> +		arg->i915 = i915;
> +		INIT_WORK(&arg->work, restart_work);
> +
> +		WRITE_ONCE(i915->gpu_error.restart, arg);
> +		queue_work(i915->wq, &arg->work);
>  	}
>  }
>  
> @@ -873,8 +834,6 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
>  	struct i915_timeline *tl;
>  	bool ret = false;
>  
> -	lockdep_assert_held(&i915->drm.struct_mutex);
> -
>  	if (!test_bit(I915_WEDGED, &error->flags))
>  		return true;
>  
> @@ -897,9 +856,9 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
>  	 */
>  	list_for_each_entry(tl, &i915->gt.timelines, link) {
>  		struct i915_request *rq;
> +		long timeout;
>  
> -		rq = i915_gem_active_peek(&tl->last_request,
> -					  &i915->drm.struct_mutex);
> +		rq = i915_gem_active_get_unlocked(&tl->last_request);
>  		if (!rq)
>  			continue;
>  
> @@ -914,12 +873,12 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
>  		 * and when the seqno passes the fence, the signaler
>  		 * then signals the fence waking us up).
>  		 */
> -		if (dma_fence_default_wait(&rq->fence, true,
> -					   MAX_SCHEDULE_TIMEOUT) < 0)
> +		timeout = dma_fence_default_wait(&rq->fence, true,
> +						 MAX_SCHEDULE_TIMEOUT);
> +		i915_request_put(rq);
> +		if (timeout < 0)
>  			goto unlock;
>  	}
> -	i915_retire_requests(i915);
> -	GEM_BUG_ON(i915->gt.active_requests);
>  
>  	intel_engines_sanitize(i915, false);
>  
> @@ -933,7 +892,6 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
>  	 * context and do not require stop_machine().
>  	 */
>  	intel_engines_reset_default_submission(i915);
> -	i915_gem_contexts_lost(i915);
>  
>  	GEM_TRACE("end\n");
>  
> @@ -946,6 +904,43 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
>  	return ret;
>  }
>  
> +struct __i915_reset {
> +	struct drm_i915_private *i915;
> +	unsigned int stalled_mask;
> +};
> +
> +static int __i915_reset__BKL(void *data)
> +{
> +	struct __i915_reset *arg = data;
> +	int err;
> +
> +	err = intel_gpu_reset(arg->i915, ALL_ENGINES);
> +	if (err)
> +		return err;
> +
> +	return gt_reset(arg->i915, arg->stalled_mask);
> +}
> +
> +#if RESET_UNDER_STOP_MACHINE
> +#define __do_reset(fn, arg) stop_machine(fn, arg, NULL)
> +#else
> +#define __do_reset(fn, arg) fn(arg)
> +#endif
> +
> +static int do_reset(struct drm_i915_private *i915, unsigned int stalled_mask)
> +{
> +	struct __i915_reset arg = { i915, stalled_mask };
> +	int err, i;
> +
> +	err = __do_reset(__i915_reset__BKL, &arg);
> +	for (i = 0; err && i < RESET_MAX_RETRIES; i++) {
> +		msleep(100);
> +		err = __do_reset(__i915_reset__BKL, &arg);
> +	}
> +
> +	return err;
> +}
> +
>  /**
>   * i915_reset - reset chip after a hang
>   * @i915: #drm_i915_private to reset
> @@ -971,31 +966,22 @@ void i915_reset(struct drm_i915_private *i915,
>  {
>  	struct i915_gpu_error *error = &i915->gpu_error;
>  	int ret;
> -	int i;
>  
>  	GEM_TRACE("flags=%lx\n", error->flags);
>  
>  	might_sleep();
> -	lockdep_assert_held(&i915->drm.struct_mutex);
>  	assert_rpm_wakelock_held(i915);
>  	GEM_BUG_ON(!test_bit(I915_RESET_BACKOFF, &error->flags));
>  
> -	if (!test_bit(I915_RESET_HANDOFF, &error->flags))
> -		return;
> -
>  	/* Clear any previous failed attempts at recovery. Time to try again. */
>  	if (!i915_gem_unset_wedged(i915))
> -		goto wakeup;
> +		return;
>  
>  	if (reason)
>  		dev_notice(i915->drm.dev, "Resetting chip for %s\n", reason);
>  	error->reset_count++;
>  
> -	ret = reset_prepare(i915);
> -	if (ret) {
> -		dev_err(i915->drm.dev, "GPU recovery failed\n");
> -		goto taint;
> -	}
> +	reset_prepare(i915);
>  
>  	if (!intel_has_gpu_reset(i915)) {
>  		if (i915_modparams.reset)
> @@ -1005,32 +991,11 @@ void i915_reset(struct drm_i915_private *i915,
>  		goto error;
>  	}
>  
> -	for (i = 0; i < RESET_MAX_RETRIES; i++) {
> -		ret = intel_gpu_reset(i915, ALL_ENGINES);
> -		if (ret == 0)
> -			break;
> -
> -		msleep(100);
> -	}
> -	if (ret) {
> +	if (do_reset(i915, stalled_mask)) {
>  		dev_err(i915->drm.dev, "Failed to reset chip\n");
>  		goto taint;
>  	}
>  
> -	/* Ok, now get things going again... */
> -
> -	/*
> -	 * Everything depends on having the GTT running, so we need to start
> -	 * there.
> -	 */
> -	ret = i915_ggtt_enable_hw(i915);
> -	if (ret) {
> -		DRM_ERROR("Failed to re-enable GGTT following reset (%d)\n",
> -			  ret);
> -		goto error;
> -	}
> -
> -	gt_reset(i915, stalled_mask);
>  	intel_overlay_reset(i915);
>  
>  	/*
> @@ -1052,9 +1017,8 @@ void i915_reset(struct drm_i915_private *i915,
>  
>  finish:
>  	reset_finish(i915);
> -wakeup:
> -	clear_bit(I915_RESET_HANDOFF, &error->flags);
> -	wake_up_bit(&error->flags, I915_RESET_HANDOFF);
> +	if (!i915_terminally_wedged(error))
> +		reset_restart(i915);
>  	return;
>  
>  taint:
> @@ -1073,7 +1037,6 @@ void i915_reset(struct drm_i915_private *i915,
>  	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
>  error:
>  	i915_gem_set_wedged(i915);
> -	i915_retire_requests(i915);
>  	goto finish;
>  }
>  
> @@ -1099,18 +1062,16 @@ static inline int intel_gt_reset_engine(struct drm_i915_private *i915,
>  int i915_reset_engine(struct intel_engine_cs *engine, const char *msg)
>  {
>  	struct i915_gpu_error *error = &engine->i915->gpu_error;
> -	struct i915_request *active_request;
>  	int ret;
>  
>  	GEM_TRACE("%s flags=%lx\n", engine->name, error->flags);
>  	GEM_BUG_ON(!test_bit(I915_RESET_ENGINE + engine->id, &error->flags));
>  
> -	active_request = reset_prepare_engine(engine);
> -	if (IS_ERR_OR_NULL(active_request)) {
> -		/* Either the previous reset failed, or we pardon the reset. */
> -		ret = PTR_ERR(active_request);
> -		goto out;
> -	}
> +	if (i915_seqno_passed(intel_engine_get_seqno(engine),
> +			      intel_engine_last_submit(engine)))
> +		return 0;
> +
> +	reset_prepare_engine(engine);
>  
>  	if (msg)
>  		dev_notice(engine->i915->drm.dev,
> @@ -1134,7 +1095,7 @@ int i915_reset_engine(struct intel_engine_cs *engine, const char *msg)
>  	 * active request and can drop it, adjust head to skip the offending
>  	 * request to resume executing remaining requests in the queue.
>  	 */
> -	reset_engine(engine, active_request, true);
> +	intel_engine_reset(engine, true);
>  
>  	/*
>  	 * The engine and its registers (and workarounds in case of render)
> @@ -1171,30 +1132,7 @@ static void i915_reset_device(struct drm_i915_private *i915,
>  	i915_wedge_on_timeout(&w, i915, 5 * HZ) {
>  		intel_prepare_reset(i915);
>  
> -		error->reason = reason;
> -		error->stalled_mask = engine_mask;
> -
> -		/* Signal that locked waiters should reset the GPU */
> -		smp_mb__before_atomic();
> -		set_bit(I915_RESET_HANDOFF, &error->flags);
> -		wake_up_all(&error->wait_queue);
> -
> -		/*
> -		 * Wait for anyone holding the lock to wakeup, without
> -		 * blocking indefinitely on struct_mutex.
> -		 */
> -		do {
> -			if (mutex_trylock(&i915->drm.struct_mutex)) {
> -				i915_reset(i915, engine_mask, reason);
> -				mutex_unlock(&i915->drm.struct_mutex);
> -			}
> -		} while (wait_on_bit_timeout(&error->flags,
> -					     I915_RESET_HANDOFF,
> -					     TASK_UNINTERRUPTIBLE,
> -					     1));
> -
> -		error->stalled_mask = 0;
> -		error->reason = NULL;
> +		i915_reset(i915, engine_mask, reason);
>  
>  		intel_finish_reset(i915);
>  	}
> @@ -1350,6 +1288,25 @@ void i915_handle_error(struct drm_i915_private *i915,
>  	intel_runtime_pm_put(i915, wakeref);
>  }
>  
> +bool i915_reset_flush(struct drm_i915_private *i915)
> +{
> +	int err;
> +
> +	cancel_delayed_work_sync(&i915->gpu_error.hangcheck_work);
> +
> +	flush_workqueue(i915->wq);
> +	GEM_BUG_ON(READ_ONCE(i915->gpu_error.restart));
> +
> +	mutex_lock(&i915->drm.struct_mutex);
> +	err = i915_gem_wait_for_idle(i915,
> +				     I915_WAIT_LOCKED |
> +				     I915_WAIT_FOR_IDLE_BOOST,
> +				     MAX_SCHEDULE_TIMEOUT);
> +	mutex_unlock(&i915->drm.struct_mutex);
> +
> +	return !err;
> +}
> +
>  static void i915_wedge_me(struct work_struct *work)
>  {
>  	struct i915_wedge_me *w = container_of(work, typeof(*w), work.work);
> diff --git a/drivers/gpu/drm/i915/i915_reset.h b/drivers/gpu/drm/i915/i915_reset.h
> index b6a519bde67d..f2d347f319df 100644
> --- a/drivers/gpu/drm/i915/i915_reset.h
> +++ b/drivers/gpu/drm/i915/i915_reset.h
> @@ -29,6 +29,9 @@ void i915_reset(struct drm_i915_private *i915,
>  int i915_reset_engine(struct intel_engine_cs *engine,
>  		      const char *reason);
>  
> +void i915_reset_request(struct i915_request *rq, bool guilty);
> +bool i915_reset_flush(struct drm_i915_private *i915);
> +
>  bool intel_has_gpu_reset(struct drm_i915_private *i915);
>  bool intel_has_reset_engine(struct drm_i915_private *i915);
>  
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index 235a2d70d671..815a05fe0e1e 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -1119,10 +1119,8 @@ void intel_engines_sanitize(struct drm_i915_private *i915, bool force)
>  	if (!reset_engines(i915) && !force)
>  		return;
>  
> -	for_each_engine(engine, i915, id) {
> -		if (engine->reset.reset)
> -			engine->reset.reset(engine, NULL);
> -	}
> +	for_each_engine(engine, i915, id)
> +		intel_engine_reset(engine, false);
>  }
>  
>  /**
> diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
> index 349ae5844f24..45e2db683fe5 100644
> --- a/drivers/gpu/drm/i915/intel_guc_submission.c
> +++ b/drivers/gpu/drm/i915/intel_guc_submission.c
> @@ -834,8 +834,7 @@ static void guc_submission_tasklet(unsigned long data)
>  	spin_unlock_irqrestore(&engine->timeline.lock, flags);
>  }
>  
> -static struct i915_request *
> -guc_reset_prepare(struct intel_engine_cs *engine)
> +static void guc_reset_prepare(struct intel_engine_cs *engine)
>  {
>  	struct intel_engine_execlists * const execlists = &engine->execlists;
>  
> @@ -861,8 +860,6 @@ guc_reset_prepare(struct intel_engine_cs *engine)
>  	 */
>  	if (engine->i915->guc.preempt_wq)
>  		flush_workqueue(engine->i915->guc.preempt_wq);
> -
> -	return i915_gem_find_active_request(engine);
>  }
>  
>  /*
> diff --git a/drivers/gpu/drm/i915/intel_hangcheck.c b/drivers/gpu/drm/i915/intel_hangcheck.c
> index 741441daae32..5662d6fed523 100644
> --- a/drivers/gpu/drm/i915/intel_hangcheck.c
> +++ b/drivers/gpu/drm/i915/intel_hangcheck.c
> @@ -25,6 +25,17 @@
>  #include "i915_drv.h"
>  #include "i915_reset.h"
>  
> +struct hangcheck {
> +	u64 acthd;
> +	u32 seqno;
> +	enum intel_engine_hangcheck_action action;
> +	unsigned long action_timestamp;
> +	int deadlock;
> +	struct intel_instdone instdone;
> +	bool wedged:1;
> +	bool stalled:1;
> +};
> +
>  static bool instdone_unchanged(u32 current_instdone, u32 *old_instdone)
>  {
>  	u32 tmp = current_instdone | *old_instdone;
> @@ -119,25 +130,22 @@ engine_stuck(struct intel_engine_cs *engine, u64 acthd)
>  }
>  
>  static void hangcheck_load_sample(struct intel_engine_cs *engine,
> -				  struct intel_engine_hangcheck *hc)
> +				  struct hangcheck *hc)
>  {
>  	hc->acthd = intel_engine_get_active_head(engine);
>  	hc->seqno = intel_engine_get_seqno(engine);
>  }
>  
>  static void hangcheck_store_sample(struct intel_engine_cs *engine,
> -				   const struct intel_engine_hangcheck *hc)
> +				   const struct hangcheck *hc)
>  {
>  	engine->hangcheck.acthd = hc->acthd;
>  	engine->hangcheck.seqno = hc->seqno;
> -	engine->hangcheck.action = hc->action;
> -	engine->hangcheck.stalled = hc->stalled;
> -	engine->hangcheck.wedged = hc->wedged;
>  }
>  
>  static enum intel_engine_hangcheck_action
>  hangcheck_get_action(struct intel_engine_cs *engine,
> -		     const struct intel_engine_hangcheck *hc)
> +		     const struct hangcheck *hc)
>  {
>  	if (engine->hangcheck.seqno != hc->seqno)
>  		return ENGINE_ACTIVE_SEQNO;
> @@ -149,7 +157,7 @@ hangcheck_get_action(struct intel_engine_cs *engine,
>  }
>  
>  static void hangcheck_accumulate_sample(struct intel_engine_cs *engine,
> -					struct intel_engine_hangcheck *hc)
> +					struct hangcheck *hc)
>  {
>  	unsigned long timeout = I915_ENGINE_DEAD_TIMEOUT;
>  
> @@ -265,19 +273,19 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
>  	intel_uncore_arm_unclaimed_mmio_detection(dev_priv);
>  
>  	for_each_engine(engine, dev_priv, id) {
> -		struct intel_engine_hangcheck hc;
> +		struct hangcheck hc;
>  
>  		hangcheck_load_sample(engine, &hc);
>  		hangcheck_accumulate_sample(engine, &hc);
>  		hangcheck_store_sample(engine, &hc);
>  
> -		if (engine->hangcheck.stalled) {
> +		if (hc.stalled) {
>  			hung |= intel_engine_flag(engine);
>  			if (hc.action != ENGINE_DEAD)
>  				stuck |= intel_engine_flag(engine);
>  		}
>  
> -		if (engine->hangcheck.wedged)
> +		if (hc.wedged)
>  			wedged |= intel_engine_flag(engine);
>  	}
>  
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index 6cef458f6b43..10354bdf8237 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -136,6 +136,7 @@
>  #include <drm/i915_drm.h>
>  #include "i915_drv.h"
>  #include "i915_gem_render_state.h"
> +#include "i915_reset.h"
>  #include "i915_vgpu.h"
>  #include "intel_lrc_reg.h"
>  #include "intel_mocs.h"
> @@ -371,7 +372,8 @@ static void unwind_wa_tail(struct i915_request *rq)
>  	assert_ring_tail_valid(rq->ring, rq->tail);
>  }
>  
> -static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
> +static struct i915_request *
> +__unwind_incomplete_requests(struct intel_engine_cs *engine)
>  {
>  	struct i915_request *rq, *rn, *active = NULL;
>  	struct list_head *uninitialized_var(pl);
> @@ -413,6 +415,8 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine)
>  		list_move_tail(&active->sched.link,
>  			       i915_sched_lookup_priolist(engine, prio));
>  	}
> +
> +	return active;
>  }
>  
>  void
> @@ -1840,11 +1844,9 @@ static int gen8_init_common_ring(struct intel_engine_cs *engine)
>  	return 0;
>  }
>  
> -static struct i915_request *
> -execlists_reset_prepare(struct intel_engine_cs *engine)
> +static void execlists_reset_prepare(struct intel_engine_cs *engine)
>  {
>  	struct intel_engine_execlists * const execlists = &engine->execlists;
> -	struct i915_request *request, *active;
>  	unsigned long flags;
>  
>  	GEM_TRACE("%s: depth<-%d\n", engine->name,
> @@ -1860,59 +1862,21 @@ execlists_reset_prepare(struct intel_engine_cs *engine)
>  	 * prevents the race.
>  	 */
>  	__tasklet_disable_sync_once(&execlists->tasklet);
> +	GEM_BUG_ON(!reset_in_progress(execlists));
>  
> +	/* And flush any current direct submission. */
>  	spin_lock_irqsave(&engine->timeline.lock, flags);
> -
> -	/*
> -	 * We want to flush the pending context switches, having disabled
> -	 * the tasklet above, we can assume exclusive access to the execlists.
> -	 * For this allows us to catch up with an inflight preemption event,
> -	 * and avoid blaming an innocent request if the stall was due to the
> -	 * preemption itself.
> -	 */
> -	process_csb(engine);
> -
> -	/*
> -	 * The last active request can then be no later than the last request
> -	 * now in ELSP[0]. So search backwards from there, so that if the GPU
> -	 * has advanced beyond the last CSB update, it will be pardoned.
> -	 */
> -	active = NULL;
> -	request = port_request(execlists->port);
> -	if (request) {
> -		/*
> -		 * Prevent the breadcrumb from advancing before we decide
> -		 * which request is currently active.
> -		 */
> -		intel_engine_stop_cs(engine);
> -
> -		list_for_each_entry_from_reverse(request,
> -						 &engine->timeline.requests,
> -						 link) {
> -			if (__i915_request_completed(request,
> -						     request->global_seqno))
> -				break;
> -
> -			active = request;
> -		}
> -	}
> -
> +	process_csb(engine); /* drain preemption events */
>  	spin_unlock_irqrestore(&engine->timeline.lock, flags);
> -
> -	return active;
>  }
>  
> -static void execlists_reset(struct intel_engine_cs *engine,
> -			    struct i915_request *request)
> +static void execlists_reset(struct intel_engine_cs *engine, bool stalled)
>  {
>  	struct intel_engine_execlists * const execlists = &engine->execlists;
> +	struct i915_request *rq;
>  	unsigned long flags;
>  	u32 *regs;
>  
> -	GEM_TRACE("%s request global=%d, current=%d\n",
> -		  engine->name, request ? request->global_seqno : 0,
> -		  intel_engine_get_seqno(engine));
> -
>  	spin_lock_irqsave(&engine->timeline.lock, flags);
>  
>  	/*
> @@ -1927,12 +1891,18 @@ static void execlists_reset(struct intel_engine_cs *engine,
>  	execlists_cancel_port_requests(execlists);
>  
>  	/* Push back any incomplete requests for replay after the reset. */
> -	__unwind_incomplete_requests(engine);
> +	rq = __unwind_incomplete_requests(engine);
>  
>  	/* Following the reset, we need to reload the CSB read/write pointers */
>  	reset_csb_pointers(&engine->execlists);
>  
> -	spin_unlock_irqrestore(&engine->timeline.lock, flags);
> +	GEM_TRACE("%s seqno=%d, current=%d, stalled? %s\n",
> +		  engine->name,
> +		  rq ? rq->global_seqno : 0,
> +		  intel_engine_get_seqno(engine),
> +		  yesno(stalled));
> +	if (!rq)
> +		goto out_unlock;
>  
>  	/*
>  	 * If the request was innocent, we leave the request in the ELSP
> @@ -1945,8 +1915,9 @@ static void execlists_reset(struct intel_engine_cs *engine,
>  	 * and have to at least restore the RING register in the context
>  	 * image back to the expected values to skip over the guilty request.
>  	 */
> -	if (!request || request->fence.error != -EIO)
> -		return;
> +	i915_reset_request(rq, stalled);
> +	if (!stalled)
> +		goto out_unlock;
>  
>  	/*
>  	 * We want a simple context + ring to execute the breadcrumb update.
> @@ -1956,7 +1927,7 @@ static void execlists_reset(struct intel_engine_cs *engine,
>  	 * future request will be after userspace has had the opportunity
>  	 * to recreate its own state.
>  	 */
> -	regs = request->hw_context->lrc_reg_state;
> +	regs = rq->hw_context->lrc_reg_state;
>  	if (engine->pinned_default_state) {
>  		memcpy(regs, /* skip restoring the vanilla PPHWSP */
>  		       engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE,
> @@ -1964,17 +1935,14 @@ static void execlists_reset(struct intel_engine_cs *engine,
>  	}
>  
>  	/* Move the RING_HEAD onto the breadcrumb, past the hanging batch */
> -	request->ring->head = intel_ring_wrap(request->ring, request->postfix);
> +	rq->ring->head = intel_ring_wrap(rq->ring, rq->postfix);
> +	intel_ring_update_space(rq->ring);
>  
> -	execlists_init_reg_state(regs, request->gem_context, engine,
> -				 request->ring);
> +	execlists_init_reg_state(regs, rq->gem_context, engine, rq->ring);
> +	__execlists_update_reg_state(engine, rq->hw_context);
>  
> -	__execlists_update_reg_state(engine, request->hw_context);
> -
> -	intel_ring_update_space(request->ring);
> -
> -	/* Reset WaIdleLiteRestore:bdw,skl as well */
> -	unwind_wa_tail(request);
> +out_unlock:
> +	spin_unlock_irqrestore(&engine->timeline.lock, flags);
>  }
>  
>  static void execlists_reset_finish(struct intel_engine_cs *engine)
> @@ -1987,6 +1955,7 @@ static void execlists_reset_finish(struct intel_engine_cs *engine)
>  	 * to sleep before we restart and reload a context.
>  	 *
>  	 */
> +	GEM_BUG_ON(!reset_in_progress(execlists));
>  	if (!RB_EMPTY_ROOT(&execlists->queue.rb_root))
>  		execlists->tasklet.func(execlists->tasklet.data);
>  
> diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
> index c300e5787b3c..a9238fd07e30 100644
> --- a/drivers/gpu/drm/i915/intel_overlay.c
> +++ b/drivers/gpu/drm/i915/intel_overlay.c
> @@ -480,8 +480,6 @@ void intel_overlay_reset(struct drm_i915_private *dev_priv)
>  	if (!overlay)
>  		return;
>  
> -	intel_overlay_release_old_vid(overlay);
> -
>  	overlay->old_xscale = 0;
>  	overlay->old_yscale = 0;
>  	overlay->crtc = NULL;
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> index 09c90475168a..a9efc8c71254 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> @@ -33,6 +33,7 @@
>  
>  #include "i915_drv.h"
>  #include "i915_gem_render_state.h"
> +#include "i915_reset.h"
>  #include "i915_trace.h"
>  #include "intel_drv.h"
>  #include "intel_workarounds.h"
> @@ -711,52 +712,80 @@ static int init_ring_common(struct intel_engine_cs *engine)
>  	return ret;
>  }
>  
> -static struct i915_request *reset_prepare(struct intel_engine_cs *engine)
> +static void reset_prepare(struct intel_engine_cs *engine)
>  {
>  	intel_engine_stop_cs(engine);
> -	return i915_gem_find_active_request(engine);
>  }
>  
> -static void skip_request(struct i915_request *rq)
> +static void reset_ring(struct intel_engine_cs *engine, bool stalled)
>  {
> -	void *vaddr = rq->ring->vaddr;
> +	struct i915_timeline *tl = &engine->timeline;
> +	struct i915_request *pos, *rq;
> +	unsigned long flags;
>  	u32 head;
>  
> -	head = rq->infix;
> -	if (rq->postfix < head) {
> -		memset32(vaddr + head, MI_NOOP,
> -			 (rq->ring->size - head) / sizeof(u32));
> -		head = 0;
> +	rq = NULL;
> +	spin_lock_irqsave(&tl->lock, flags);
> +	list_for_each_entry(pos, &tl->requests, link) {
> +		if (!__i915_request_completed(pos, pos->global_seqno)) {
> +			rq = pos;
> +			break;
> +		}
>  	}
> -	memset32(vaddr + head, MI_NOOP, (rq->postfix - head) / sizeof(u32));
> -}
> -
> -static void reset_ring(struct intel_engine_cs *engine, struct i915_request *rq)
> -{
> -	GEM_TRACE("%s request global=%d, current=%d\n",
> -		  engine->name, rq ? rq->global_seqno : 0,
> -		  intel_engine_get_seqno(engine));
>  
> +	GEM_TRACE("%s seqno=%d, current=%d, stalled? %s\n",
> +		  engine->name,
> +		  rq ? rq->global_seqno : 0,
> +		  intel_engine_get_seqno(engine),
> +		  yesno(stalled));
>  	/*
> -	 * Try to restore the logical GPU state to match the continuation
> -	 * of the request queue. If we skip the context/PD restore, then
> -	 * the next request may try to execute assuming that its context
> -	 * is valid and loaded on the GPU and so may try to access invalid
> -	 * memory, prompting repeated GPU hangs.
> +	 * The guilty request will get skipped on a hung engine.
>  	 *
> -	 * If the request was guilty, we still restore the logical state
> -	 * in case the next request requires it (e.g. the aliasing ppgtt),
> -	 * but skip over the hung batch.
> +	 * Users of client default contexts do not rely on logical
> +	 * state preserved between batches so it is safe to execute
> +	 * queued requests following the hang. Non default contexts
> +	 * rely on preserved state, so skipping a batch loses the
> +	 * evolution of the state and it needs to be considered corrupted.
> +	 * Executing more queued batches on top of corrupted state is
> +	 * risky. But we take the risk by trying to advance through
> +	 * the queued requests in order to make the client behaviour
> +	 * more predictable around resets, by not throwing away random
> +	 * amount of batches it has prepared for execution. Sophisticated
> +	 * clients can use gem_reset_stats_ioctl and dma fence status
> +	 * (exported via sync_file info ioctl on explicit fences) to observe
> +	 * when it loses the context state and should rebuild accordingly.
>  	 *
> -	 * If the request was innocent, we try to replay the request with
> -	 * the restored context.
> +	 * The context ban, and ultimately the client ban, mechanism are safety
> +	 * valves if client submission ends up resulting in nothing more than
> +	 * subsequent hangs.
>  	 */
> +
>  	if (rq) {
> -		/* If the rq hung, jump to its breadcrumb and skip the batch */
> -		rq->ring->head = intel_ring_wrap(rq->ring, rq->head);
> -		if (rq->fence.error == -EIO)
> -			skip_request(rq);
> +		/*
> +		 * Try to restore the logical GPU state to match the
> +		 * continuation of the request queue. If we skip the
> +		 * context/PD restore, then the next request may try to execute
> +		 * assuming that its context is valid and loaded on the GPU and
> +		 * so may try to access invalid memory, prompting repeated GPU
> +		 * hangs.
> +		 *
> +		 * If the request was guilty, we still restore the logical
> +		 * state in case the next request requires it (e.g. the
> +		 * aliasing ppgtt), but skip over the hung batch.
> +		 *
> +		 * If the request was innocent, we try to replay the request
> +		 * with the restored context.
> +		 */
> +		i915_reset_request(rq, stalled);
> +
> +		GEM_BUG_ON(rq->ring != engine->buffer);
> +		head = rq->head;
> +	} else {
> +		head = engine->buffer->tail;
>  	}
> +	engine->buffer->head = intel_ring_wrap(engine->buffer, head);
> +
> +	spin_unlock_irqrestore(&tl->lock, flags);
>  }
>  
>  static void reset_finish(struct intel_engine_cs *engine)
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 5ad46c2fbc0f..f2effd001540 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -120,13 +120,8 @@ struct intel_instdone {
>  struct intel_engine_hangcheck {
>  	u64 acthd;
>  	u32 seqno;
> -	enum intel_engine_hangcheck_action action;
>  	unsigned long action_timestamp;
> -	int deadlock;
>  	struct intel_instdone instdone;
> -	struct i915_request *active_request;
> -	bool stalled:1;
> -	bool wedged:1;
>  };
>  
>  struct intel_ring {
> @@ -444,9 +439,8 @@ struct intel_engine_cs {
>  	int		(*init_hw)(struct intel_engine_cs *engine);
>  
>  	struct {
> -		struct i915_request *(*prepare)(struct intel_engine_cs *engine);
> -		void (*reset)(struct intel_engine_cs *engine,
> -			      struct i915_request *rq);
> +		void (*prepare)(struct intel_engine_cs *engine);
> +		void (*reset)(struct intel_engine_cs *engine, bool stalled);
>  		void (*finish)(struct intel_engine_cs *engine);
>  	} reset;
>  
> @@ -1018,6 +1012,13 @@ gen8_emit_ggtt_write(u32 *cs, u32 value, u32 gtt_offset)
>  	return cs;
>  }
>  
> +static inline void intel_engine_reset(struct intel_engine_cs *engine,
> +				      bool stalled)
> +{
> +	if (engine->reset.reset)
> +		engine->reset.reset(engine, stalled);
> +}
> +
>  void intel_engines_sanitize(struct drm_i915_private *i915, bool force);
>  
>  bool intel_engine_is_idle(struct intel_engine_cs *engine);
> diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> index 12550b55c42f..67431355cd6e 100644
> --- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> +++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> @@ -363,9 +363,7 @@ static int igt_global_reset(void *arg)
>  	/* Check that we can issue a global GPU reset */
>  
>  	igt_global_reset_lock(i915);
> -	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
>  
> -	mutex_lock(&i915->drm.struct_mutex);
>  	reset_count = i915_reset_count(&i915->gpu_error);
>  
>  	i915_reset(i915, ALL_ENGINES, NULL);
> @@ -374,9 +372,7 @@ static int igt_global_reset(void *arg)
>  		pr_err("No GPU reset recorded!\n");
>  		err = -EINVAL;
>  	}
> -	mutex_unlock(&i915->drm.struct_mutex);
>  
> -	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
>  	igt_global_reset_unlock(i915);
>  
>  	if (i915_terminally_wedged(&i915->gpu_error))
> @@ -399,9 +395,7 @@ static int igt_wedged_reset(void *arg)
>  	i915_gem_set_wedged(i915);
>  	GEM_BUG_ON(!i915_terminally_wedged(&i915->gpu_error));
>  
> -	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
>  	i915_reset(i915, ALL_ENGINES, NULL);
> -	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
>  
>  	intel_runtime_pm_put(i915, wakeref);
>  	mutex_unlock(&i915->drm.struct_mutex);
> @@ -511,7 +505,7 @@ static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
>  				break;
>  			}
>  
> -			if (!wait_for_idle(engine)) {
> +			if (!i915_reset_flush(i915)) {
>  				struct drm_printer p =
>  					drm_info_printer(i915->drm.dev);
>  
> @@ -903,20 +897,13 @@ static int igt_reset_engines(void *arg)
>  	return 0;
>  }
>  
> -static u32 fake_hangcheck(struct i915_request *rq, u32 mask)
> +static u32 fake_hangcheck(struct drm_i915_private *i915, u32 mask)
>  {
> -	struct i915_gpu_error *error = &rq->i915->gpu_error;
> -	u32 reset_count = i915_reset_count(error);
> -
> -	error->stalled_mask = mask;
> -
> -	/* set_bit() must be after we have setup the backchannel (mask) */
> -	smp_mb__before_atomic();
> -	set_bit(I915_RESET_HANDOFF, &error->flags);
> +	u32 count = i915_reset_count(&i915->gpu_error);
>  
> -	wake_up_all(&error->wait_queue);
> +	i915_reset(i915, mask, NULL);
>  
> -	return reset_count;
> +	return count;
>  }
>  
>  static int igt_reset_wait(void *arg)
> @@ -962,7 +949,7 @@ static int igt_reset_wait(void *arg)
>  		goto out_rq;
>  	}
>  
> -	reset_count = fake_hangcheck(rq, ALL_ENGINES);
> +	reset_count = fake_hangcheck(i915, ALL_ENGINES);
>  
>  	timeout = i915_request_wait(rq, I915_WAIT_LOCKED, 10);
>  	if (timeout < 0) {
> @@ -972,7 +959,6 @@ static int igt_reset_wait(void *arg)
>  		goto out_rq;
>  	}
>  
> -	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
>  	if (i915_reset_count(&i915->gpu_error) == reset_count) {
>  		pr_err("No GPU reset recorded!\n");
>  		err = -EINVAL;
> @@ -1162,7 +1148,7 @@ static int __igt_reset_evict_vma(struct drm_i915_private *i915,
>  	}
>  
>  out_reset:
> -	fake_hangcheck(rq, intel_engine_flag(rq->engine));
> +	fake_hangcheck(rq->i915, intel_engine_flag(rq->engine));
>  
>  	if (tsk) {
>  		struct igt_wedge_me w;
> @@ -1341,12 +1327,7 @@ static int igt_reset_queue(void *arg)
>  				goto fini;
>  			}
>  
> -			reset_count = fake_hangcheck(prev, ENGINE_MASK(id));
> -
> -			i915_reset(i915, ENGINE_MASK(id), NULL);
> -
> -			GEM_BUG_ON(test_bit(I915_RESET_HANDOFF,
> -					    &i915->gpu_error.flags));
> +			reset_count = fake_hangcheck(i915, ENGINE_MASK(id));
>  
>  			if (prev->fence.error != -EIO) {
>  				pr_err("GPU reset not recorded on hanging request [fence.error=%d]!\n",
> @@ -1565,6 +1546,7 @@ static int igt_atomic_reset_engine(struct intel_engine_cs *engine,
>  		pr_err("%s(%s): Failed to start request %llx, at %x\n",
>  		       __func__, engine->name,
>  		       rq->fence.seqno, hws_seqno(&h, rq));
> +		i915_gem_set_wedged(i915);
>  		err = -EIO;
>  	}
>  
> @@ -1588,7 +1570,6 @@ static int igt_atomic_reset_engine(struct intel_engine_cs *engine,
>  static void force_reset(struct drm_i915_private *i915)
>  {
>  	i915_gem_set_wedged(i915);
> -	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
>  	i915_reset(i915, 0, NULL);
>  }
>  
> @@ -1618,6 +1599,26 @@ static int igt_atomic_reset(void *arg)
>  	if (i915_terminally_wedged(&i915->gpu_error))
>  		goto unlock;
>  
> +	if (intel_has_gpu_reset(i915)) {
> +		const typeof(*phases) *p;
> +
> +		for (p = phases; p->name; p++) {
> +			GEM_TRACE("intel_gpu_reset under %s\n", p->name);
> +
> +			p->critical_section_begin();
> +			err = intel_gpu_reset(i915, ALL_ENGINES);
> +			p->critical_section_end();
> +
> +			if (err) {
> +				pr_err("intel_gpu_reset failed under %s\n",
> +				       p->name);
> +				goto out;
> +			}
> +		}
> +
> +		force_reset(i915);
> +	}
> +
>  	if (intel_has_reset_engine(i915)) {
>  		struct intel_engine_cs *engine;
>  		enum intel_engine_id id;
> diff --git a/drivers/gpu/drm/i915/selftests/intel_workarounds.c b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
> index a8cac56be835..b15c4f26c593 100644
> --- a/drivers/gpu/drm/i915/selftests/intel_workarounds.c
> +++ b/drivers/gpu/drm/i915/selftests/intel_workarounds.c
> @@ -214,7 +214,6 @@ static int check_whitelist(struct i915_gem_context *ctx,
>  
>  static int do_device_reset(struct intel_engine_cs *engine)
>  {
> -	set_bit(I915_RESET_HANDOFF, &engine->i915->gpu_error.flags);
>  	i915_reset(engine->i915, ENGINE_MASK(engine->id), "live_workarounds");
>  	return 0;
>  }
> @@ -394,7 +393,6 @@ static int
>  live_gpu_reset_gt_engine_workarounds(void *arg)
>  {
>  	struct drm_i915_private *i915 = arg;
> -	struct i915_gpu_error *error = &i915->gpu_error;
>  	intel_wakeref_t wakeref;
>  	struct wa_lists lists;
>  	bool ok;
> @@ -413,7 +411,6 @@ live_gpu_reset_gt_engine_workarounds(void *arg)
>  	if (!ok)
>  		goto out;
>  
> -	set_bit(I915_RESET_HANDOFF, &error->flags);
>  	i915_reset(i915, ALL_ENGINES, "live_workarounds");
>  
>  	ok = verify_gt_engine_wa(i915, &lists, "after reset");
> diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> index 5477ad4a7e7d..8ab5a2688a0c 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> +++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> @@ -58,8 +58,8 @@ static void mock_device_release(struct drm_device *dev)
>  	i915_gem_contexts_lost(i915);
>  	mutex_unlock(&i915->drm.struct_mutex);
>  
> -	cancel_delayed_work_sync(&i915->gt.retire_work);
> -	cancel_delayed_work_sync(&i915->gt.idle_work);
> +	drain_delayed_work(&i915->gt.retire_work);
> +	drain_delayed_work(&i915->gt.idle_work);
>  	i915_gem_drain_workqueue(i915);
>  
>  	mutex_lock(&i915->drm.struct_mutex);
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 11/33] drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest

[Chris Wilson]

Trim the struct_mutex hold and exclude the call to i915_gem_set_wedged()
as a reminder that it must be callable without struct_mutex held.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
---
 drivers/gpu/drm/i915/selftests/intel_hangcheck.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 67431355cd6e..8025c7e0bf6c 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -389,16 +389,16 @@ static int igt_wedged_reset(void *arg)
 	/* Check that we can recover a wedged device with a GPU reset */
 
 	igt_global_reset_lock(i915);
-	mutex_lock(&i915->drm.struct_mutex);
 	wakeref = intel_runtime_pm_get(i915);
 
 	i915_gem_set_wedged(i915);
-	GEM_BUG_ON(!i915_terminally_wedged(&i915->gpu_error));
 
+	mutex_lock(&i915->drm.struct_mutex);
+	GEM_BUG_ON(!i915_terminally_wedged(&i915->gpu_error));
 	i915_reset(i915, ALL_ENGINES, NULL);
+	mutex_unlock(&i915->drm.struct_mutex);
 
 	intel_runtime_pm_put(i915, wakeref);
-	mutex_unlock(&i915->drm.struct_mutex);
 	igt_global_reset_unlock(i915);
 
 	return i915_terminally_wedged(&i915->gpu_error) ? -EIO : 0;
@@ -1675,6 +1675,7 @@ int intel_hangcheck_live_selftests(struct drm_i915_private *i915)
 
 	wakeref = intel_runtime_pm_get(i915);
 	saved_hangcheck = fetch_and_zero(&i915_modparams.enable_hangcheck);
+	drain_delayed_work(&i915->gpu_error.hangcheck_work); /* flush param */
 
 	err = i915_subtests(tests, i915);
 
-- 
2.20.1

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

[PATCH 12/33] drm/i915: Issue engine resets onto idle engines

[Chris Wilson]

Always perform the requested reset, even if we believe the engine is
idle. Presumably there was a reason the caller wanted the reset, and in
the near future we lose the easy tracking for whether the engine is
idle.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
---
 drivers/gpu/drm/i915/i915_reset.c             |  4 ----
 .../gpu/drm/i915/selftests/intel_hangcheck.c  | 22 +++++--------------
 2 files changed, 6 insertions(+), 20 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index c01acc832518..7299b4f0d6d7 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -1067,10 +1067,6 @@ int i915_reset_engine(struct intel_engine_cs *engine, const char *msg)
 	GEM_TRACE("%s flags=%lx\n", engine->name, error->flags);
 	GEM_BUG_ON(!test_bit(I915_RESET_ENGINE + engine->id, &error->flags));
 
-	if (i915_seqno_passed(intel_engine_get_seqno(engine),
-			      intel_engine_last_submit(engine)))
-		return 0;
-
 	reset_prepare_engine(engine);
 
 	if (msg)
diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 8025c7e0bf6c..2c38ea5892d9 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -449,8 +449,6 @@ static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
 
 		set_bit(I915_RESET_ENGINE + id, &i915->gpu_error.flags);
 		do {
-			u32 seqno = intel_engine_get_seqno(engine);
-
 			if (active) {
 				struct i915_request *rq;
 
@@ -479,8 +477,6 @@ static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
 					break;
 				}
 
-				GEM_BUG_ON(!rq->global_seqno);
-				seqno = rq->global_seqno - 1;
 				i915_request_put(rq);
 			}
 
@@ -496,11 +492,10 @@ static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
 				break;
 			}
 
-			reset_engine_count += active;
 			if (i915_reset_engine_count(&i915->gpu_error, engine) !=
-			    reset_engine_count) {
-				pr_err("%s engine reset %srecorded!\n",
-				       engine->name, active ? "not " : "");
+			    ++reset_engine_count) {
+				pr_err("%s engine reset not recorded!\n",
+				       engine->name);
 				err = -EINVAL;
 				break;
 			}
@@ -728,7 +723,6 @@ static int __igt_reset_engines(struct drm_i915_private *i915,
 
 		set_bit(I915_RESET_ENGINE + id, &i915->gpu_error.flags);
 		do {
-			u32 seqno = intel_engine_get_seqno(engine);
 			struct i915_request *rq = NULL;
 
 			if (flags & TEST_ACTIVE) {
@@ -756,9 +750,6 @@ static int __igt_reset_engines(struct drm_i915_private *i915,
 					err = -EIO;
 					break;
 				}
-
-				GEM_BUG_ON(!rq->global_seqno);
-				seqno = rq->global_seqno - 1;
 			}
 
 			err = i915_reset_engine(engine, NULL);
@@ -795,10 +786,9 @@ static int __igt_reset_engines(struct drm_i915_private *i915,
 
 		reported = i915_reset_engine_count(&i915->gpu_error, engine);
 		reported -= threads[engine->id].resets;
-		if (reported != (flags & TEST_ACTIVE ? count : 0)) {
-			pr_err("i915_reset_engine(%s:%s): reset %lu times, but reported %lu, expected %lu reported\n",
-			       engine->name, test_name, count, reported,
-			       (flags & TEST_ACTIVE ? count : 0));
+		if (reported != count) {
+			pr_err("i915_reset_engine(%s:%s): reset %lu times, but reported %lu\n",
+			       engine->name, test_name, count, reported);
 			if (!err)
 				err = -EINVAL;
 		}
-- 
2.20.1

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

[PATCH 13/33] drm/i915: Stop tracking MRU activity on VMA

[Chris Wilson]

Our goal is to remove struct_mutex and replace it with fine grained
locking. One of the thorny issues is our eviction logic for reclaiming
space for an execbuffer (or GTT mmaping, among a few other examples).
While eviction itself is easy to move under a per-VM mutex, performing
the activity tracking is less agreeable. One solution is not to do any
MRU tracking and do a simple coarse evaluation during eviction of
active/inactive, with a loose temporal ordering of last
insertion/evaluation. That keeps all the locking constrained to when we
are manipulating the VM itself, neatly avoiding the tricky handling of
possible recursive locking during execbuf and elsewhere.

Note that discarding the MRU (currently implemented as a pair of lists,
to avoid scanning the active list for a NONBLOCKING search) is unlikely
to impact upon our efficiency to reclaim VM space (where we think a LRU
model is best) as our current strategy is to use random idle replacement
first before doing a search, and over time the use of softpinned 48b
per-ppGTT is growing (thereby eliminating any need to perform any eviction
searches, in theory at least) with the remaining users being found on
much older devices (gen2-gen6).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c               | 10 +--
 drivers/gpu/drm/i915/i915_gem_evict.c         | 71 ++++++++++++-------
 drivers/gpu/drm/i915/i915_gem_gtt.c           | 15 ++--
 drivers/gpu/drm/i915/i915_gem_gtt.h           | 26 +------
 drivers/gpu/drm/i915/i915_gem_shrinker.c      |  8 ++-
 drivers/gpu/drm/i915/i915_gem_stolen.c        |  3 +-
 drivers/gpu/drm/i915/i915_gpu_error.c         | 37 +++++-----
 drivers/gpu/drm/i915/i915_vma.c               |  9 +--
 .../gpu/drm/i915/selftests/i915_gem_evict.c   |  4 +-
 drivers/gpu/drm/i915/selftests/i915_gem_gtt.c |  2 +-
 10 files changed, 84 insertions(+), 101 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index dcbe644869b3..12a0a80bc989 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -255,10 +255,7 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 
 	pinned = ggtt->vm.reserved;
 	mutex_lock(&dev->struct_mutex);
-	list_for_each_entry(vma, &ggtt->vm.active_list, vm_link)
-		if (i915_vma_is_pinned(vma))
-			pinned += vma->node.size;
-	list_for_each_entry(vma, &ggtt->vm.inactive_list, vm_link)
+	list_for_each_entry(vma, &ggtt->vm.bound_list, vm_link)
 		if (i915_vma_is_pinned(vma))
 			pinned += vma->node.size;
 	mutex_unlock(&dev->struct_mutex);
@@ -1541,13 +1538,10 @@ static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj)
 	GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
 
 	for_each_ggtt_vma(vma, obj) {
-		if (i915_vma_is_active(vma))
-			continue;
-
 		if (!drm_mm_node_allocated(&vma->node))
 			continue;
 
-		list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
+		list_move_tail(&vma->vm_link, &vma->vm->bound_list);
 	}
 
 	i915 = to_i915(obj->base.dev);
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index f6855401f247..5cfe4b75e7d6 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -126,14 +126,10 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	struct drm_i915_private *dev_priv = vm->i915;
 	struct drm_mm_scan scan;
 	struct list_head eviction_list;
-	struct list_head *phases[] = {
-		&vm->inactive_list,
-		&vm->active_list,
-		NULL,
-	}, **phase;
 	struct i915_vma *vma, *next;
 	struct drm_mm_node *node;
 	enum drm_mm_insert_mode mode;
+	struct i915_vma *active;
 	int ret;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
@@ -169,17 +165,46 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 */
 	if (!(flags & PIN_NONBLOCK))
 		i915_retire_requests(dev_priv);
-	else
-		phases[1] = NULL;
 
 search_again:
+	active = NULL;
 	INIT_LIST_HEAD(&eviction_list);
-	phase = phases;
-	do {
-		list_for_each_entry(vma, *phase, vm_link)
-			if (mark_free(&scan, vma, flags, &eviction_list))
-				goto found;
-	} while (*++phase);
+	list_for_each_entry_safe(vma, next, &vm->bound_list, vm_link) {
+		/*
+		 * We keep this list in a rough least-recently scanned order
+		 * of active elements (inactive elements are cheap to reap).
+		 * New entries are added to the end, and we move anything we
+		 * scan to the end. The assumption is that the working set
+		 * of applications is either steady state (and thanks to the
+		 * userspace bo cache it almost always is) or volatile and
+		 * frequently replaced after a frame, which are self-evicting!
+		 * Given that assumption, the MRU order of the scan list is
+		 * fairly static, and keeping it in least-recently scan order
+		 * is suitable.
+		 *
+		 * To notice when we complete one full cycle, we record the
+		 * first active element seen, before moving it to the tail.
+		 */
+		if (i915_vma_is_active(vma)) {
+			if (vma == active) {
+				if (flags & PIN_NONBLOCK)
+					break;
+
+				active = ERR_PTR(-EAGAIN);
+			}
+
+			if (active != ERR_PTR(-EAGAIN)) {
+				if (!active)
+					active = vma;
+
+				list_move_tail(&vma->vm_link, &vm->bound_list);
+				continue;
+			}
+		}
+
+		if (mark_free(&scan, vma, flags, &eviction_list))
+			goto found;
+	}
 
 	/* Nothing found, clean up and bail out! */
 	list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
@@ -388,11 +413,6 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
  */
 int i915_gem_evict_vm(struct i915_address_space *vm)
 {
-	struct list_head *phases[] = {
-		&vm->inactive_list,
-		&vm->active_list,
-		NULL
-	}, **phase;
 	struct list_head eviction_list;
 	struct i915_vma *vma, *next;
 	int ret;
@@ -412,16 +432,13 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
 	}
 
 	INIT_LIST_HEAD(&eviction_list);
-	phase = phases;
-	do {
-		list_for_each_entry(vma, *phase, vm_link) {
-			if (i915_vma_is_pinned(vma))
-				continue;
+	list_for_each_entry(vma, &vm->bound_list, vm_link) {
+		if (i915_vma_is_pinned(vma))
+			continue;
 
-			__i915_vma_pin(vma);
-			list_add(&vma->evict_link, &eviction_list);
-		}
-	} while (*++phase);
+		__i915_vma_pin(vma);
+		list_add(&vma->evict_link, &eviction_list);
+	}
 
 	ret = 0;
 	list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 9081e3bc5a59..2ad9070a54c1 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -491,9 +491,8 @@ static void i915_address_space_init(struct i915_address_space *vm, int subclass)
 
 	stash_init(&vm->free_pages);
 
-	INIT_LIST_HEAD(&vm->active_list);
-	INIT_LIST_HEAD(&vm->inactive_list);
 	INIT_LIST_HEAD(&vm->unbound_list);
+	INIT_LIST_HEAD(&vm->bound_list);
 }
 
 static void i915_address_space_fini(struct i915_address_space *vm)
@@ -2111,8 +2110,7 @@ void i915_ppgtt_close(struct i915_address_space *vm)
 static void ppgtt_destroy_vma(struct i915_address_space *vm)
 {
 	struct list_head *phases[] = {
-		&vm->active_list,
-		&vm->inactive_list,
+		&vm->bound_list,
 		&vm->unbound_list,
 		NULL,
 	}, **phase;
@@ -2135,8 +2133,7 @@ void i915_ppgtt_release(struct kref *kref)
 
 	ppgtt_destroy_vma(&ppgtt->vm);
 
-	GEM_BUG_ON(!list_empty(&ppgtt->vm.active_list));
-	GEM_BUG_ON(!list_empty(&ppgtt->vm.inactive_list));
+	GEM_BUG_ON(!list_empty(&ppgtt->vm.bound_list));
 	GEM_BUG_ON(!list_empty(&ppgtt->vm.unbound_list));
 
 	ppgtt->vm.cleanup(&ppgtt->vm);
@@ -2801,8 +2798,7 @@ void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv)
 	mutex_lock(&dev_priv->drm.struct_mutex);
 	i915_gem_fini_aliasing_ppgtt(dev_priv);
 
-	GEM_BUG_ON(!list_empty(&ggtt->vm.active_list));
-	list_for_each_entry_safe(vma, vn, &ggtt->vm.inactive_list, vm_link)
+	list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link)
 		WARN_ON(i915_vma_unbind(vma));
 
 	if (drm_mm_node_allocated(&ggtt->error_capture))
@@ -3514,8 +3510,7 @@ void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv)
 	ggtt->vm.closed = true; /* skip rewriting PTE on VMA unbind */
 
 	/* clflush objects bound into the GGTT and rebind them. */
-	GEM_BUG_ON(!list_empty(&ggtt->vm.active_list));
-	list_for_each_entry_safe(vma, vn, &ggtt->vm.inactive_list, vm_link) {
+	list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link) {
 		struct drm_i915_gem_object *obj = vma->obj;
 
 		if (!(vma->flags & I915_VMA_GLOBAL_BIND))
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index a0039ea97cdc..bd679c8c56dd 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -299,32 +299,12 @@ struct i915_address_space {
 	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
 
 	/**
-	 * List of objects currently involved in rendering.
-	 *
-	 * Includes buffers having the contents of their GPU caches
-	 * flushed, not necessarily primitives. last_read_req
-	 * represents when the rendering involved will be completed.
-	 *
-	 * A reference is held on the buffer while on this list.
+	 * List of vma currently bound.
 	 */
-	struct list_head active_list;
+	struct list_head bound_list;
 
 	/**
-	 * LRU list of objects which are not in the ringbuffer and
-	 * are ready to unbind, but are still in the GTT.
-	 *
-	 * last_read_req is NULL while an object is in this list.
-	 *
-	 * A reference is not held on the buffer while on this list,
-	 * as merely being GTT-bound shouldn't prevent its being
-	 * freed, and we'll pull it off the list in the free path.
-	 */
-	struct list_head inactive_list;
-
-	/**
-	 * List of vma that have been unbound.
-	 *
-	 * A reference is not held on the buffer while on this list.
+	 * List of vma that are not unbound.
 	 */
 	struct list_head unbound_list;
 
diff --git a/drivers/gpu/drm/i915/i915_gem_shrinker.c b/drivers/gpu/drm/i915/i915_gem_shrinker.c
index 8ceecb026910..a76d6c95c824 100644
--- a/drivers/gpu/drm/i915/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/i915_gem_shrinker.c
@@ -462,9 +462,13 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
 
 	/* We also want to clear any cached iomaps as they wrap vmap */
 	list_for_each_entry_safe(vma, next,
-				 &i915->ggtt.vm.inactive_list, vm_link) {
+				 &i915->ggtt.vm.bound_list, vm_link) {
 		unsigned long count = vma->node.size >> PAGE_SHIFT;
-		if (vma->iomap && i915_vma_unbind(vma) == 0)
+
+		if (!vma->iomap || i915_vma_is_active(vma))
+			continue;
+
+		if (i915_vma_unbind(vma) == 0)
 			freed_pages += count;
 	}
 
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index 9df615eea2d8..a9e365789686 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -701,7 +701,8 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv
 	vma->pages = obj->mm.pages;
 	vma->flags |= I915_VMA_GLOBAL_BIND;
 	__i915_vma_set_map_and_fenceable(vma);
-	list_move_tail(&vma->vm_link, &ggtt->vm.inactive_list);
+
+	list_move_tail(&vma->vm_link, &ggtt->vm.bound_list);
 
 	spin_lock(&dev_priv->mm.obj_lock);
 	list_move_tail(&obj->mm.link, &dev_priv->mm.bound_list);
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 4eef0462489c..96d1d634a29d 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -1121,7 +1121,7 @@ static void capture_bo(struct drm_i915_error_buffer *err,
 
 static u32 capture_error_bo(struct drm_i915_error_buffer *err,
 			    int count, struct list_head *head,
-			    bool pinned_only)
+			    bool active_only, bool pinned_only)
 {
 	struct i915_vma *vma;
 	int i = 0;
@@ -1130,6 +1130,9 @@ static u32 capture_error_bo(struct drm_i915_error_buffer *err,
 		if (!vma->obj)
 			continue;
 
+		if (active_only && !i915_vma_is_active(vma))
+			continue;
+
 		if (pinned_only && !i915_vma_is_pinned(vma))
 			continue;
 
@@ -1601,14 +1604,16 @@ static void gem_capture_vm(struct i915_gpu_state *error,
 	int count;
 
 	count = 0;
-	list_for_each_entry(vma, &vm->active_list, vm_link)
-		count++;
+	list_for_each_entry(vma, &vm->bound_list, vm_link)
+		if (i915_vma_is_active(vma))
+			count++;
 
 	active_bo = NULL;
 	if (count)
 		active_bo = kcalloc(count, sizeof(*active_bo), GFP_ATOMIC);
 	if (active_bo)
-		count = capture_error_bo(active_bo, count, &vm->active_list, false);
+		count = capture_error_bo(active_bo, count, &vm->bound_list,
+					 true, false);
 	else
 		count = 0;
 
@@ -1646,28 +1651,20 @@ static void capture_pinned_buffers(struct i915_gpu_state *error)
 	struct i915_address_space *vm = &error->i915->ggtt.vm;
 	struct drm_i915_error_buffer *bo;
 	struct i915_vma *vma;
-	int count_inactive, count_active;
-
-	count_inactive = 0;
-	list_for_each_entry(vma, &vm->inactive_list, vm_link)
-		count_inactive++;
+	int count;
 
-	count_active = 0;
-	list_for_each_entry(vma, &vm->active_list, vm_link)
-		count_active++;
+	count = 0;
+	list_for_each_entry(vma, &vm->bound_list, vm_link)
+		count++;
 
 	bo = NULL;
-	if (count_inactive + count_active)
-		bo = kcalloc(count_inactive + count_active,
-			     sizeof(*bo), GFP_ATOMIC);
+	if (count)
+		bo = kcalloc(count, sizeof(*bo), GFP_ATOMIC);
 	if (!bo)
 		return;
 
-	count_inactive = capture_error_bo(bo, count_inactive,
-					  &vm->active_list, true);
-	count_active = capture_error_bo(bo + count_inactive, count_active,
-					&vm->inactive_list, true);
-	error->pinned_bo_count = count_inactive + count_active;
+	error->pinned_bo_count =
+		capture_error_bo(bo, count, &vm->bound_list, false, true);
 	error->pinned_bo = bo;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 5b4d78cdb4ca..7de28baffb8f 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -79,9 +79,6 @@ __i915_vma_retire(struct i915_vma *vma, struct i915_request *rq)
 	if (--vma->active_count)
 		return;
 
-	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
-	list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
-
 	GEM_BUG_ON(!i915_gem_object_is_active(obj));
 	if (--obj->active_count)
 		return;
@@ -659,7 +656,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 	GEM_BUG_ON(!i915_gem_valid_gtt_space(vma, cache_level));
 
-	list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
+	list_move_tail(&vma->vm_link, &vma->vm->bound_list);
 
 	if (vma->obj) {
 		struct drm_i915_gem_object *obj = vma->obj;
@@ -1003,10 +1000,8 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 	 * add the active reference first and queue for it to be dropped
 	 * *last*.
 	 */
-	if (!i915_gem_active_isset(active) && !vma->active_count++) {
-		list_move_tail(&vma->vm_link, &vma->vm->active_list);
+	if (!i915_gem_active_isset(active) && !vma->active_count++)
 		obj->active_count++;
-	}
 	i915_gem_active_set(active, rq);
 	GEM_BUG_ON(!i915_vma_is_active(vma));
 	GEM_BUG_ON(!obj->active_count);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index d0553bc69705..af9b85cb8639 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -84,7 +84,7 @@ static int populate_ggtt(struct drm_i915_private *i915,
 		return -EINVAL;
 	}
 
-	if (list_empty(&i915->ggtt.vm.inactive_list)) {
+	if (list_empty(&i915->ggtt.vm.bound_list)) {
 		pr_err("No objects on the GGTT inactive list!\n");
 		return -EINVAL;
 	}
@@ -96,7 +96,7 @@ static void unpin_ggtt(struct drm_i915_private *i915)
 {
 	struct i915_vma *vma;
 
-	list_for_each_entry(vma, &i915->ggtt.vm.inactive_list, vm_link)
+	list_for_each_entry(vma, &i915->ggtt.vm.bound_list, vm_link)
 		if (vma->obj->mm.quirked)
 			i915_vma_unpin(vma);
 }
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 06bde4a273cb..8feb4af308ff 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1237,7 +1237,7 @@ static void track_vma_bind(struct i915_vma *vma)
 	__i915_gem_object_pin_pages(obj);
 
 	vma->pages = obj->mm.pages;
-	list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
+	list_move_tail(&vma->vm_link, &vma->vm->bound_list);
 }
 
 static int exercise_mock(struct drm_i915_private *i915,
-- 
2.20.1

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

[PATCH 14/33] drm/i915: Pull VM lists under the VM mutex.

[Chris Wilson]

A starting point to counter the pervasive struct_mutex. For the goal of
avoiding (or at least blocking under them!) global locks during user
request submission, a simple but important step is being able to manage
each clients GTT separately. For which, we want to replace using the
struct_mutex as the guard for all things GTT/VM and switch instead to a
specific mutex inside i915_address_space.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c                 | 14 ++++++++------
 drivers/gpu/drm/i915/i915_gem_evict.c           |  2 ++
 drivers/gpu/drm/i915/i915_gem_gtt.c             | 15 +++++++++++++--
 drivers/gpu/drm/i915/i915_gem_shrinker.c        |  4 ++++
 drivers/gpu/drm/i915/i915_gem_stolen.c          |  2 ++
 drivers/gpu/drm/i915/i915_vma.c                 | 11 +++++++++++
 drivers/gpu/drm/i915/selftests/i915_gem_evict.c |  3 +++
 drivers/gpu/drm/i915/selftests/i915_gem_gtt.c   |  3 +++
 8 files changed, 46 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 12a0a80bc989..111a047a45b7 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -247,18 +247,19 @@ int
 i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 			    struct drm_file *file)
 {
-	struct drm_i915_private *dev_priv = to_i915(dev);
-	struct i915_ggtt *ggtt = &dev_priv->ggtt;
+	struct i915_ggtt *ggtt = &to_i915(dev)->ggtt;
 	struct drm_i915_gem_get_aperture *args = data;
 	struct i915_vma *vma;
 	u64 pinned;
 
+	mutex_lock(&ggtt->vm.mutex);
+
 	pinned = ggtt->vm.reserved;
-	mutex_lock(&dev->struct_mutex);
 	list_for_each_entry(vma, &ggtt->vm.bound_list, vm_link)
 		if (i915_vma_is_pinned(vma))
 			pinned += vma->node.size;
-	mutex_unlock(&dev->struct_mutex);
+
+	mutex_unlock(&ggtt->vm.mutex);
 
 	args->aper_size = ggtt->vm.total;
 	args->aper_available_size = args->aper_size - pinned;
@@ -1531,20 +1532,21 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 
 static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj)
 {
-	struct drm_i915_private *i915;
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
 	struct list_head *list;
 	struct i915_vma *vma;
 
 	GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
 
+	mutex_lock(&i915->ggtt.vm.mutex);
 	for_each_ggtt_vma(vma, obj) {
 		if (!drm_mm_node_allocated(&vma->node))
 			continue;
 
 		list_move_tail(&vma->vm_link, &vma->vm->bound_list);
 	}
+	mutex_unlock(&i915->ggtt.vm.mutex);
 
-	i915 = to_i915(obj->base.dev);
 	spin_lock(&i915->mm.obj_lock);
 	list = obj->bind_count ? &i915->mm.bound_list : &i915->mm.unbound_list;
 	list_move_tail(&obj->mm.link, list);
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 5cfe4b75e7d6..dc137701acb8 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -432,6 +432,7 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
 	}
 
 	INIT_LIST_HEAD(&eviction_list);
+	mutex_lock(&vm->mutex);
 	list_for_each_entry(vma, &vm->bound_list, vm_link) {
 		if (i915_vma_is_pinned(vma))
 			continue;
@@ -439,6 +440,7 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
 		__i915_vma_pin(vma);
 		list_add(&vma->evict_link, &eviction_list);
 	}
+	mutex_unlock(&vm->mutex);
 
 	ret = 0;
 	list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 2ad9070a54c1..49b00996a15e 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -1931,7 +1931,10 @@ static struct i915_vma *pd_vma_create(struct gen6_hw_ppgtt *ppgtt, int size)
 	vma->ggtt_view.type = I915_GGTT_VIEW_ROTATED; /* prevent fencing */
 
 	INIT_LIST_HEAD(&vma->obj_link);
+
+	mutex_lock(&vma->vm->mutex);
 	list_add(&vma->vm_link, &vma->vm->unbound_list);
+	mutex_unlock(&vma->vm->mutex);
 
 	return vma;
 }
@@ -3504,9 +3507,10 @@ void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv)
 
 	i915_check_and_clear_faults(dev_priv);
 
+	mutex_lock(&ggtt->vm.mutex);
+
 	/* First fill our portion of the GTT with scratch pages */
 	ggtt->vm.clear_range(&ggtt->vm, 0, ggtt->vm.total);
-
 	ggtt->vm.closed = true; /* skip rewriting PTE on VMA unbind */
 
 	/* clflush objects bound into the GGTT and rebind them. */
@@ -3516,19 +3520,26 @@ void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv)
 		if (!(vma->flags & I915_VMA_GLOBAL_BIND))
 			continue;
 
+		mutex_unlock(&ggtt->vm.mutex);
+
 		if (!i915_vma_unbind(vma))
-			continue;
+			goto lock;
 
 		WARN_ON(i915_vma_bind(vma,
 				      obj ? obj->cache_level : 0,
 				      PIN_UPDATE));
 		if (obj)
 			WARN_ON(i915_gem_object_set_to_gtt_domain(obj, false));
+
+lock:
+		mutex_lock(&ggtt->vm.mutex);
 	}
 
 	ggtt->vm.closed = false;
 	i915_ggtt_invalidate(dev_priv);
 
+	mutex_unlock(&ggtt->vm.mutex);
+
 	if (INTEL_GEN(dev_priv) >= 8) {
 		struct intel_ppat *ppat = &dev_priv->ppat;
 
diff --git a/drivers/gpu/drm/i915/i915_gem_shrinker.c b/drivers/gpu/drm/i915/i915_gem_shrinker.c
index a76d6c95c824..6da795c7e62e 100644
--- a/drivers/gpu/drm/i915/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/i915_gem_shrinker.c
@@ -461,6 +461,7 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
 					       I915_SHRINK_VMAPS);
 
 	/* We also want to clear any cached iomaps as they wrap vmap */
+	mutex_lock(&i915->ggtt.vm.mutex);
 	list_for_each_entry_safe(vma, next,
 				 &i915->ggtt.vm.bound_list, vm_link) {
 		unsigned long count = vma->node.size >> PAGE_SHIFT;
@@ -468,9 +469,12 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
 		if (!vma->iomap || i915_vma_is_active(vma))
 			continue;
 
+		mutex_unlock(&i915->ggtt.vm.mutex);
 		if (i915_vma_unbind(vma) == 0)
 			freed_pages += count;
+		mutex_lock(&i915->ggtt.vm.mutex);
 	}
+	mutex_unlock(&i915->ggtt.vm.mutex);
 
 out:
 	shrinker_unlock(i915, unlock);
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index a9e365789686..74a9661479ca 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -702,7 +702,9 @@ i915_gem_object_create_stolen_for_preallocated(struct drm_i915_private *dev_priv
 	vma->flags |= I915_VMA_GLOBAL_BIND;
 	__i915_vma_set_map_and_fenceable(vma);
 
+	mutex_lock(&ggtt->vm.mutex);
 	list_move_tail(&vma->vm_link, &ggtt->vm.bound_list);
+	mutex_unlock(&ggtt->vm.mutex);
 
 	spin_lock(&dev_priv->mm.obj_lock);
 	list_move_tail(&obj->mm.link, &dev_priv->mm.bound_list);
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 7de28baffb8f..dcbd0d345c72 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -213,7 +213,10 @@ vma_create(struct drm_i915_gem_object *obj,
 	}
 	rb_link_node(&vma->obj_node, rb, p);
 	rb_insert_color(&vma->obj_node, &obj->vma_tree);
+
+	mutex_lock(&vm->mutex);
 	list_add(&vma->vm_link, &vm->unbound_list);
+	mutex_unlock(&vm->mutex);
 
 	return vma;
 
@@ -656,7 +659,9 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 	GEM_BUG_ON(!i915_gem_valid_gtt_space(vma, cache_level));
 
+	mutex_lock(&vma->vm->mutex);
 	list_move_tail(&vma->vm_link, &vma->vm->bound_list);
+	mutex_unlock(&vma->vm->mutex);
 
 	if (vma->obj) {
 		struct drm_i915_gem_object *obj = vma->obj;
@@ -689,8 +694,10 @@ i915_vma_remove(struct i915_vma *vma)
 
 	vma->ops->clear_pages(vma);
 
+	mutex_lock(&vma->vm->mutex);
 	drm_mm_remove_node(&vma->node);
 	list_move_tail(&vma->vm_link, &vma->vm->unbound_list);
+	mutex_unlock(&vma->vm->mutex);
 
 	/*
 	 * Since the unbound list is global, only move to that list if
@@ -802,7 +809,11 @@ static void __i915_vma_destroy(struct i915_vma *vma)
 	GEM_BUG_ON(i915_gem_active_isset(&vma->last_fence));
 
 	list_del(&vma->obj_link);
+
+	mutex_lock(&vma->vm->mutex);
 	list_del(&vma->vm_link);
+	mutex_unlock(&vma->vm->mutex);
+
 	if (vma->obj)
 		rb_erase(&vma->obj_node, &vma->obj->vma_tree);
 
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index af9b85cb8639..32dce7176f63 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -94,11 +94,14 @@ static int populate_ggtt(struct drm_i915_private *i915,
 
 static void unpin_ggtt(struct drm_i915_private *i915)
 {
+	struct i915_ggtt *ggtt = &i915->ggtt;
 	struct i915_vma *vma;
 
+	mutex_lock(&ggtt->vm.mutex);
 	list_for_each_entry(vma, &i915->ggtt.vm.bound_list, vm_link)
 		if (vma->obj->mm.quirked)
 			i915_vma_unpin(vma);
+	mutex_unlock(&ggtt->vm.mutex);
 }
 
 static void cleanup_objects(struct drm_i915_private *i915,
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 8feb4af308ff..3850ef4a5ec8 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1237,7 +1237,10 @@ static void track_vma_bind(struct i915_vma *vma)
 	__i915_gem_object_pin_pages(obj);
 
 	vma->pages = obj->mm.pages;
+
+	mutex_lock(&vma->vm->mutex);
 	list_move_tail(&vma->vm_link, &vma->vm->bound_list);
+	mutex_unlock(&vma->vm->mutex);
 }
 
 static int exercise_mock(struct drm_i915_private *i915,
-- 
2.20.1

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

[PATCH 15/33] drm/i915: Move vma lookup to its own lock

[Chris Wilson]

Remove the struct_mutex requirement for looking up the vma for an
object.

v2: Highlight how the race for duplicate vma creation is resolved on
reacquiring the lock with a short comment.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c       |  6 +--
 drivers/gpu/drm/i915/i915_gem.c           | 33 +++++++-----
 drivers/gpu/drm/i915/i915_gem_object.h    | 45 +++++++++-------
 drivers/gpu/drm/i915/i915_vma.c           | 66 ++++++++++++++++-------
 drivers/gpu/drm/i915/i915_vma.h           |  2 +-
 drivers/gpu/drm/i915/selftests/i915_vma.c |  4 +-
 6 files changed, 98 insertions(+), 58 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 3b995f9fdc06..b64f758803f3 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -160,14 +160,14 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 		   obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");
 	if (obj->base.name)
 		seq_printf(m, " (name: %d)", obj->base.name);
-	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+	list_for_each_entry(vma, &obj->vma.list, obj_link) {
 		if (i915_vma_is_pinned(vma))
 			pin_count++;
 	}
 	seq_printf(m, " (pinned x %d)", pin_count);
 	if (obj->pin_global)
 		seq_printf(m, " (global)");
-	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+	list_for_each_entry(vma, &obj->vma.list, obj_link) {
 		if (!drm_mm_node_allocated(&vma->node))
 			continue;
 
@@ -323,7 +323,7 @@ static int per_file_stats(int id, void *ptr, void *data)
 	if (obj->base.name || obj->base.dma_buf)
 		stats->shared += obj->base.size;
 
-	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+	list_for_each_entry(vma, &obj->vma.list, obj_link) {
 		if (!drm_mm_node_allocated(&vma->node))
 			continue;
 
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 111a047a45b7..653c7ba4c69f 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -439,15 +439,19 @@ int i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 	if (ret)
 		return ret;
 
-	while ((vma = list_first_entry_or_null(&obj->vma_list,
-					       struct i915_vma,
-					       obj_link))) {
+	spin_lock(&obj->vma.lock);
+	while (!ret && (vma = list_first_entry_or_null(&obj->vma.list,
+						       struct i915_vma,
+						       obj_link))) {
 		list_move_tail(&vma->obj_link, &still_in_list);
+		spin_unlock(&obj->vma.lock);
+
 		ret = i915_vma_unbind(vma);
-		if (ret)
-			break;
+
+		spin_lock(&obj->vma.lock);
 	}
-	list_splice(&still_in_list, &obj->vma_list);
+	list_splice(&still_in_list, &obj->vma.list);
+	spin_unlock(&obj->vma.lock);
 
 	return ret;
 }
@@ -3491,7 +3495,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,
 	 * reading an invalid PTE on older architectures.
 	 */
 restart:
-	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+	list_for_each_entry(vma, &obj->vma.list, obj_link) {
 		if (!drm_mm_node_allocated(&vma->node))
 			continue;
 
@@ -3569,7 +3573,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,
 			 */
 		}
 
-		list_for_each_entry(vma, &obj->vma_list, obj_link) {
+		list_for_each_entry(vma, &obj->vma.list, obj_link) {
 			if (!drm_mm_node_allocated(&vma->node))
 				continue;
 
@@ -3579,7 +3583,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,
 		}
 	}
 
-	list_for_each_entry(vma, &obj->vma_list, obj_link)
+	list_for_each_entry(vma, &obj->vma.list, obj_link)
 		vma->node.color = cache_level;
 	i915_gem_object_set_cache_coherency(obj, cache_level);
 	obj->cache_dirty = true; /* Always invalidate stale cachelines */
@@ -4155,7 +4159,9 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj,
 {
 	mutex_init(&obj->mm.lock);
 
-	INIT_LIST_HEAD(&obj->vma_list);
+	spin_lock_init(&obj->vma.lock);
+	INIT_LIST_HEAD(&obj->vma.list);
+
 	INIT_LIST_HEAD(&obj->lut_list);
 	INIT_LIST_HEAD(&obj->batch_pool_link);
 
@@ -4321,14 +4327,13 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915,
 		mutex_lock(&i915->drm.struct_mutex);
 
 		GEM_BUG_ON(i915_gem_object_is_active(obj));
-		list_for_each_entry_safe(vma, vn,
-					 &obj->vma_list, obj_link) {
+		list_for_each_entry_safe(vma, vn, &obj->vma.list, obj_link) {
 			GEM_BUG_ON(i915_vma_is_active(vma));
 			vma->flags &= ~I915_VMA_PIN_MASK;
 			i915_vma_destroy(vma);
 		}
-		GEM_BUG_ON(!list_empty(&obj->vma_list));
-		GEM_BUG_ON(!RB_EMPTY_ROOT(&obj->vma_tree));
+		GEM_BUG_ON(!list_empty(&obj->vma.list));
+		GEM_BUG_ON(!RB_EMPTY_ROOT(&obj->vma.tree));
 
 		/* This serializes freeing with the shrinker. Since the free
 		 * is delayed, first by RCU then by the workqueue, we want the
diff --git a/drivers/gpu/drm/i915/i915_gem_object.h b/drivers/gpu/drm/i915/i915_gem_object.h
index cb1b0144d274..5a33b6d9f942 100644
--- a/drivers/gpu/drm/i915/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/i915_gem_object.h
@@ -87,24 +87,33 @@ struct drm_i915_gem_object {
 
 	const struct drm_i915_gem_object_ops *ops;
 
-	/**
-	 * @vma_list: List of VMAs backed by this object
-	 *
-	 * The VMA on this list are ordered by type, all GGTT vma are placed
-	 * at the head and all ppGTT vma are placed at the tail. The different
-	 * types of GGTT vma are unordered between themselves, use the
-	 * @vma_tree (which has a defined order between all VMA) to find an
-	 * exact match.
-	 */
-	struct list_head vma_list;
-	/**
-	 * @vma_tree: Ordered tree of VMAs backed by this object
-	 *
-	 * All VMA created for this object are placed in the @vma_tree for
-	 * fast retrieval via a binary search in i915_vma_instance().
-	 * They are also added to @vma_list for easy iteration.
-	 */
-	struct rb_root vma_tree;
+	struct {
+		/**
+		 * @vma.lock: protect the list/tree of vmas
+		 */
+		struct spinlock lock;
+
+		/**
+		 * @vma.list: List of VMAs backed by this object
+		 *
+		 * The VMA on this list are ordered by type, all GGTT vma are
+		 * placed at the head and all ppGTT vma are placed at the tail.
+		 * The different types of GGTT vma are unordered between
+		 * themselves, use the @vma.tree (which has a defined order
+		 * between all VMA) to quickly find an exact match.
+		 */
+		struct list_head list;
+
+		/**
+		 * @vma.tree: Ordered tree of VMAs backed by this object
+		 *
+		 * All VMA created for this object are placed in the @vma.tree
+		 * for fast retrieval via a binary search in
+		 * i915_vma_instance(). They are also added to @vma.list for
+		 * easy iteration.
+		 */
+		struct rb_root tree;
+	} vma;
 
 	/**
 	 * @lut_list: List of vma lookup entries in use for this object.
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index dcbd0d345c72..d83b8ad5f859 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -187,32 +187,52 @@ vma_create(struct drm_i915_gem_object *obj,
 								i915_gem_object_get_stride(obj));
 		GEM_BUG_ON(!is_power_of_2(vma->fence_alignment));
 
-		/*
-		 * We put the GGTT vma at the start of the vma-list, followed
-		 * by the ppGGTT vma. This allows us to break early when
-		 * iterating over only the GGTT vma for an object, see
-		 * for_each_ggtt_vma()
-		 */
 		vma->flags |= I915_VMA_GGTT;
-		list_add(&vma->obj_link, &obj->vma_list);
-	} else {
-		list_add_tail(&vma->obj_link, &obj->vma_list);
 	}
 
+	spin_lock(&obj->vma.lock);
+
 	rb = NULL;
-	p = &obj->vma_tree.rb_node;
+	p = &obj->vma.tree.rb_node;
 	while (*p) {
 		struct i915_vma *pos;
+		long cmp;
 
 		rb = *p;
 		pos = rb_entry(rb, struct i915_vma, obj_node);
-		if (i915_vma_compare(pos, vm, view) < 0)
+
+		/*
+		 * If the view already exists in the tree, another thread
+		 * already created a matching vma, so return the older instance
+		 * and dispose of ours.
+		 */
+		cmp = i915_vma_compare(pos, vm, view);
+		if (cmp == 0) {
+			spin_unlock(&obj->vma.lock);
+			kmem_cache_free(vm->i915->vmas, vma);
+			return pos;
+		}
+
+		if (cmp < 0)
 			p = &rb->rb_right;
 		else
 			p = &rb->rb_left;
 	}
 	rb_link_node(&vma->obj_node, rb, p);
-	rb_insert_color(&vma->obj_node, &obj->vma_tree);
+	rb_insert_color(&vma->obj_node, &obj->vma.tree);
+
+	if (i915_vma_is_ggtt(vma))
+		/*
+		 * We put the GGTT vma at the start of the vma-list, followed
+		 * by the ppGGTT vma. This allows us to break early when
+		 * iterating over only the GGTT vma for an object, see
+		 * for_each_ggtt_vma()
+		 */
+		list_add(&vma->obj_link, &obj->vma.list);
+	else
+		list_add_tail(&vma->obj_link, &obj->vma.list);
+
+	spin_unlock(&obj->vma.lock);
 
 	mutex_lock(&vm->mutex);
 	list_add(&vma->vm_link, &vm->unbound_list);
@@ -232,7 +252,7 @@ vma_lookup(struct drm_i915_gem_object *obj,
 {
 	struct rb_node *rb;
 
-	rb = obj->vma_tree.rb_node;
+	rb = obj->vma.tree.rb_node;
 	while (rb) {
 		struct i915_vma *vma = rb_entry(rb, struct i915_vma, obj_node);
 		long cmp;
@@ -272,16 +292,18 @@ i915_vma_instance(struct drm_i915_gem_object *obj,
 {
 	struct i915_vma *vma;
 
-	lockdep_assert_held(&obj->base.dev->struct_mutex);
 	GEM_BUG_ON(view && !i915_is_ggtt(vm));
 	GEM_BUG_ON(vm->closed);
 
+	spin_lock(&obj->vma.lock);
 	vma = vma_lookup(obj, vm, view);
-	if (!vma)
+	spin_unlock(&obj->vma.lock);
+
+	/* vma_create() will resolve the race if another creates the vma */
+	if (unlikely(!vma))
 		vma = vma_create(obj, vm, view);
 
 	GEM_BUG_ON(!IS_ERR(vma) && i915_vma_compare(vma, vm, view));
-	GEM_BUG_ON(!IS_ERR(vma) && vma_lookup(obj, vm, view) != vma);
 	return vma;
 }
 
@@ -808,14 +830,18 @@ static void __i915_vma_destroy(struct i915_vma *vma)
 
 	GEM_BUG_ON(i915_gem_active_isset(&vma->last_fence));
 
-	list_del(&vma->obj_link);
-
 	mutex_lock(&vma->vm->mutex);
 	list_del(&vma->vm_link);
 	mutex_unlock(&vma->vm->mutex);
 
-	if (vma->obj)
-		rb_erase(&vma->obj_node, &vma->obj->vma_tree);
+	if (vma->obj) {
+		struct drm_i915_gem_object *obj = vma->obj;
+
+		spin_lock(&obj->vma.lock);
+		list_del(&vma->obj_link);
+		rb_erase(&vma->obj_node, &vma->obj->vma.tree);
+		spin_unlock(&obj->vma.lock);
+	}
 
 	rbtree_postorder_for_each_entry_safe(iter, n, &vma->active, node) {
 		GEM_BUG_ON(i915_gem_active_isset(&iter->base));
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 4f7c1c7599f4..7252abc73d3e 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -425,7 +425,7 @@ void i915_vma_parked(struct drm_i915_private *i915);
  * or the list is empty ofc.
  */
 #define for_each_ggtt_vma(V, OBJ) \
-	list_for_each_entry(V, &(OBJ)->vma_list, obj_link)		\
+	list_for_each_entry(V, &(OBJ)->vma.list, obj_link)		\
 		for_each_until(!i915_vma_is_ggtt(V))
 
 #endif
diff --git a/drivers/gpu/drm/i915/selftests/i915_vma.c b/drivers/gpu/drm/i915/selftests/i915_vma.c
index f0a32edfb9b1..cf1de82741fa 100644
--- a/drivers/gpu/drm/i915/selftests/i915_vma.c
+++ b/drivers/gpu/drm/i915/selftests/i915_vma.c
@@ -672,7 +672,7 @@ static int igt_vma_partial(void *arg)
 		}
 
 		count = 0;
-		list_for_each_entry(vma, &obj->vma_list, obj_link)
+		list_for_each_entry(vma, &obj->vma.list, obj_link)
 			count++;
 		if (count != nvma) {
 			pr_err("(%s) All partial vma were not recorded on the obj->vma_list: found %u, expected %u\n",
@@ -701,7 +701,7 @@ static int igt_vma_partial(void *arg)
 		i915_vma_unpin(vma);
 
 		count = 0;
-		list_for_each_entry(vma, &obj->vma_list, obj_link)
+		list_for_each_entry(vma, &obj->vma.list, obj_link)
 			count++;
 		if (count != nvma) {
 			pr_err("(%s) allocated an extra full vma!\n", p->name);
-- 
2.20.1

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

[PATCH 16/33] drm/i915: Always allocate an object/vma for the HWSP

[Chris Wilson]

Currently we only allocate an object and vma if we are using a GGTT
virtual HWSP, and a plain struct page for a physical HWSP. For
convenience later on with global timelines, it will be useful to always
have the status page being tracked by a struct i915_vma. Make it so.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/intel_engine_cs.c       | 109 ++++++++++---------
 drivers/gpu/drm/i915/intel_guc_submission.c  |   6 +
 drivers/gpu/drm/i915/intel_lrc.c             |  12 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c      |  21 +++-
 drivers/gpu/drm/i915/intel_ringbuffer.h      |  23 +---
 drivers/gpu/drm/i915/selftests/mock_engine.c |   2 +-
 6 files changed, 93 insertions(+), 80 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 815a05fe0e1e..c1dc1769e337 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -506,27 +506,61 @@ void intel_engine_setup_common(struct intel_engine_cs *engine)
 
 static void cleanup_status_page(struct intel_engine_cs *engine)
 {
+	struct i915_vma *vma;
+
 	/* Prevent writes into HWSP after returning the page to the system */
 	intel_engine_set_hwsp_writemask(engine, ~0u);
 
-	if (HWS_NEEDS_PHYSICAL(engine->i915)) {
-		void *addr = fetch_and_zero(&engine->status_page.page_addr);
+	vma = fetch_and_zero(&engine->status_page.vma);
+	if (!vma)
+		return;
 
-		__free_page(virt_to_page(addr));
-	}
+	if (!HWS_NEEDS_PHYSICAL(engine->i915))
+		i915_vma_unpin(vma);
+
+	i915_gem_object_unpin_map(vma->obj);
+	__i915_gem_object_release_unless_active(vma->obj);
+}
+
+static int pin_ggtt_status_page(struct intel_engine_cs *engine,
+				struct i915_vma *vma)
+{
+	unsigned int flags;
+
+	flags = PIN_GLOBAL;
+	if (!HAS_LLC(engine->i915))
+		/*
+		 * On g33, we cannot place HWS above 256MiB, so
+		 * restrict its pinning to the low mappable arena.
+		 * Though this restriction is not documented for
+		 * gen4, gen5, or byt, they also behave similarly
+		 * and hang if the HWS is placed at the top of the
+		 * GTT. To generalise, it appears that all !llc
+		 * platforms have issues with us placing the HWS
+		 * above the mappable region (even though we never
+		 * actually map it).
+		 */
+		flags |= PIN_MAPPABLE;
+	else
+		flags |= PIN_HIGH;
 
-	i915_vma_unpin_and_release(&engine->status_page.vma,
-				   I915_VMA_RELEASE_MAP);
+	return i915_vma_pin(vma, 0, 0, flags);
 }
 
 static int init_status_page(struct intel_engine_cs *engine)
 {
 	struct drm_i915_gem_object *obj;
 	struct i915_vma *vma;
-	unsigned int flags;
 	void *vaddr;
 	int ret;
 
+	/*
+	 * Though the HWS register does support 36bit addresses, historically
+	 * we have had hangs and corruption reported due to wild writes if
+	 * the HWS is placed above 4G. We only allow objects to be allocated
+	 * in GFP_DMA32 for i965, and no earlier physical address users had
+	 * access to more than 4G.
+	 */
 	obj = i915_gem_object_create_internal(engine->i915, PAGE_SIZE);
 	if (IS_ERR(obj)) {
 		DRM_ERROR("Failed to allocate status page\n");
@@ -543,61 +577,30 @@ static int init_status_page(struct intel_engine_cs *engine)
 		goto err;
 	}
 
-	flags = PIN_GLOBAL;
-	if (!HAS_LLC(engine->i915))
-		/* On g33, we cannot place HWS above 256MiB, so
-		 * restrict its pinning to the low mappable arena.
-		 * Though this restriction is not documented for
-		 * gen4, gen5, or byt, they also behave similarly
-		 * and hang if the HWS is placed at the top of the
-		 * GTT. To generalise, it appears that all !llc
-		 * platforms have issues with us placing the HWS
-		 * above the mappable region (even though we never
-		 * actually map it).
-		 */
-		flags |= PIN_MAPPABLE;
-	else
-		flags |= PIN_HIGH;
-	ret = i915_vma_pin(vma, 0, 0, flags);
-	if (ret)
-		goto err;
-
 	vaddr = i915_gem_object_pin_map(obj, I915_MAP_WB);
 	if (IS_ERR(vaddr)) {
 		ret = PTR_ERR(vaddr);
-		goto err_unpin;
+		goto err;
 	}
 
+	engine->status_page.addr = memset(vaddr, 0, PAGE_SIZE);
 	engine->status_page.vma = vma;
-	engine->status_page.ggtt_offset = i915_ggtt_offset(vma);
-	engine->status_page.page_addr = memset(vaddr, 0, PAGE_SIZE);
+
+	if (!HWS_NEEDS_PHYSICAL(engine->i915)) {
+		ret = pin_ggtt_status_page(engine, vma);
+		if (ret)
+			goto err_unpin;
+	}
+
 	return 0;
 
 err_unpin:
-	i915_vma_unpin(vma);
+	i915_gem_object_unpin_map(obj);
 err:
 	i915_gem_object_put(obj);
 	return ret;
 }
 
-static int init_phys_status_page(struct intel_engine_cs *engine)
-{
-	struct page *page;
-
-	/*
-	 * Though the HWS register does support 36bit addresses, historically
-	 * we have had hangs and corruption reported due to wild writes if
-	 * the HWS is placed above 4G.
-	 */
-	page = alloc_page(GFP_KERNEL | __GFP_DMA32 | __GFP_ZERO);
-	if (!page)
-		return -ENOMEM;
-
-	engine->status_page.page_addr = page_address(page);
-
-	return 0;
-}
-
 static void __intel_context_unpin(struct i915_gem_context *ctx,
 				  struct intel_engine_cs *engine)
 {
@@ -690,10 +693,7 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 	if (ret)
 		goto err_unpin_preempt;
 
-	if (HWS_NEEDS_PHYSICAL(i915))
-		ret = init_phys_status_page(engine);
-	else
-		ret = init_status_page(engine);
+	ret = init_status_page(engine);
 	if (ret)
 		goto err_breadcrumbs;
 
@@ -1366,7 +1366,8 @@ static void intel_engine_print_registers(const struct intel_engine_cs *engine,
 	}
 
 	if (HAS_EXECLISTS(dev_priv)) {
-		const u32 *hws = &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX];
+		const u32 *hws =
+			&engine->status_page.addr[I915_HWS_CSB_BUF0_INDEX];
 		unsigned int idx;
 		u8 read, write;
 
@@ -1549,7 +1550,7 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 	spin_unlock_irqrestore(&b->rb_lock, flags);
 
 	drm_printf(m, "HWSP:\n");
-	hexdump(m, engine->status_page.page_addr, PAGE_SIZE);
+	hexdump(m, engine->status_page.addr, PAGE_SIZE);
 
 	drm_printf(m, "Idle? %s\n", yesno(intel_engine_is_idle(engine)));
 }
diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index 45e2db683fe5..4295ade0d613 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -81,6 +81,12 @@
  *
  */
 
+static inline u32 intel_hws_preempt_done_address(struct intel_engine_cs *engine)
+{
+	return (i915_ggtt_offset(engine->status_page.vma) +
+		I915_GEM_HWS_PREEMPT_ADDR);
+}
+
 static inline struct i915_priolist *to_priolist(struct rb_node *rb)
 {
 	return rb_entry(rb, struct i915_priolist, node);
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 10354bdf8237..891878ab93a4 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -174,6 +174,12 @@ static void execlists_init_reg_state(u32 *reg_state,
 				     struct intel_engine_cs *engine,
 				     struct intel_ring *ring);
 
+static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
+{
+	return (i915_ggtt_offset(engine->status_page.vma) +
+		I915_GEM_HWS_INDEX_ADDR);
+}
+
 static inline struct i915_priolist *to_priolist(struct rb_node *rb)
 {
 	return rb_entry(rb, struct i915_priolist, node);
@@ -1807,7 +1813,7 @@ static void enable_execlists(struct intel_engine_cs *engine)
 		   _MASKED_BIT_DISABLE(STOP_RING));
 
 	I915_WRITE(RING_HWS_PGA(engine->mmio_base),
-		   engine->status_page.ggtt_offset);
+		   i915_ggtt_offset(engine->status_page.vma));
 	POSTING_READ(RING_HWS_PGA(engine->mmio_base));
 }
 
@@ -2352,10 +2358,10 @@ static int logical_ring_init(struct intel_engine_cs *engine)
 	}
 
 	execlists->csb_status =
-		&engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX];
+		&engine->status_page.addr[I915_HWS_CSB_BUF0_INDEX];
 
 	execlists->csb_write =
-		&engine->status_page.page_addr[intel_hws_csb_write_index(i915)];
+		&engine->status_page.addr[intel_hws_csb_write_index(i915)];
 
 	reset_csb_pointers(execlists);
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index a9efc8c71254..cb6d2aa2a829 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -43,6 +43,12 @@
  */
 #define LEGACY_REQUEST_SIZE 200
 
+static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
+{
+	return (i915_ggtt_offset(engine->status_page.vma) +
+		I915_GEM_HWS_INDEX_ADDR);
+}
+
 static unsigned int __intel_ring_space(unsigned int head,
 				       unsigned int tail,
 				       unsigned int size)
@@ -503,12 +509,17 @@ static void set_hws_pga(struct intel_engine_cs *engine, phys_addr_t phys)
 	I915_WRITE(HWS_PGA, addr);
 }
 
-static void ring_setup_phys_status_page(struct intel_engine_cs *engine)
+static struct page *status_page(struct intel_engine_cs *engine)
 {
-	struct page *page = virt_to_page(engine->status_page.page_addr);
-	phys_addr_t phys = PFN_PHYS(page_to_pfn(page));
+	struct drm_i915_gem_object *obj = engine->status_page.vma->obj;
 
-	set_hws_pga(engine, phys);
+	GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
+	return sg_page(obj->mm.pages->sgl);
+}
+
+static void ring_setup_phys_status_page(struct intel_engine_cs *engine)
+{
+	set_hws_pga(engine, PFN_PHYS(page_to_pfn(status_page(engine))));
 	set_hwstam(engine, ~0u);
 }
 
@@ -575,7 +586,7 @@ static void flush_cs_tlb(struct intel_engine_cs *engine)
 
 static void ring_setup_status_page(struct intel_engine_cs *engine)
 {
-	set_hwsp(engine, engine->status_page.ggtt_offset);
+	set_hwsp(engine, i915_ggtt_offset(engine->status_page.vma));
 	set_hwstam(engine, ~0u);
 
 	flush_cs_tlb(engine);
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index f2effd001540..32371ae67f24 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -32,8 +32,7 @@ struct i915_sched_attr;
 
 struct intel_hw_status_page {
 	struct i915_vma *vma;
-	u32 *page_addr;
-	u32 ggtt_offset;
+	u32 *addr;
 };
 
 #define I915_READ_TAIL(engine) I915_READ(RING_TAIL((engine)->mmio_base))
@@ -671,7 +670,7 @@ static inline u32
 intel_read_status_page(const struct intel_engine_cs *engine, int reg)
 {
 	/* Ensure that the compiler doesn't optimize away the load. */
-	return READ_ONCE(engine->status_page.page_addr[reg]);
+	return READ_ONCE(engine->status_page.addr[reg]);
 }
 
 static inline void
@@ -684,12 +683,12 @@ intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
 	 */
 	if (static_cpu_has(X86_FEATURE_CLFLUSH)) {
 		mb();
-		clflush(&engine->status_page.page_addr[reg]);
-		engine->status_page.page_addr[reg] = value;
-		clflush(&engine->status_page.page_addr[reg]);
+		clflush(&engine->status_page.addr[reg]);
+		engine->status_page.addr[reg] = value;
+		clflush(&engine->status_page.addr[reg]);
 		mb();
 	} else {
-		WRITE_ONCE(engine->status_page.page_addr[reg], value);
+		WRITE_ONCE(engine->status_page.addr[reg], value);
 	}
 }
 
@@ -877,16 +876,6 @@ static inline bool intel_engine_has_started(struct intel_engine_cs *engine,
 void intel_engine_get_instdone(struct intel_engine_cs *engine,
 			       struct intel_instdone *instdone);
 
-static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
-{
-	return engine->status_page.ggtt_offset + I915_GEM_HWS_INDEX_ADDR;
-}
-
-static inline u32 intel_hws_preempt_done_address(struct intel_engine_cs *engine)
-{
-	return engine->status_page.ggtt_offset + I915_GEM_HWS_PREEMPT_ADDR;
-}
-
 /* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
 
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 905318b7ae18..4e5b4dc6df0f 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -200,7 +200,7 @@ struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
 	engine->base.i915 = i915;
 	snprintf(engine->base.name, sizeof(engine->base.name), "%s", name);
 	engine->base.id = id;
-	engine->base.status_page.page_addr = (void *)(engine + 1);
+	engine->base.status_page.addr = (void *)(engine + 1);
 
 	engine->base.context_pin = mock_context_pin;
 	engine->base.request_alloc = mock_request_alloc;
-- 
2.20.1

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

[PATCH 17/33] drm/i915: Add timeline barrier support

[Chris Wilson]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Timeline barrier allows serialization between different timelines.

After calling i915_timeline_set_barrier with a request, all following
submissions on this timeline will be set up as depending on this request,
or barrier. Once the barrier has been completed it automatically gets
cleared and things continue as normal.

This facility will be used by the upcoming context SSEU code.

v2:
 * Assert barrier has been retired on timeline_fini. (Chris Wilson)
 * Fix mock_timeline.

v3:
 * Improved comment language. (Chris Wilson)

v4:
 * Maintain ordering with previous barriers set on the timeline.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_request.c           | 17 ++++++++++++++
 drivers/gpu/drm/i915/i915_timeline.c          | 21 ++++++++++++++++++
 drivers/gpu/drm/i915/i915_timeline.h          | 22 +++++++++++++++++++
 .../gpu/drm/i915/selftests/mock_timeline.c    |  1 +
 4 files changed, 61 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index f4241a17e2ad..894b32258340 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -517,6 +517,19 @@ i915_request_alloc_slow(struct intel_context *ce)
 	return kmem_cache_alloc(ce->gem_context->i915->requests, GFP_KERNEL);
 }
 
+static int add_barrier(struct i915_request *rq, struct i915_gem_active *active)
+{
+	struct i915_request *barrier =
+		i915_gem_active_raw(active, &rq->i915->drm.struct_mutex);
+
+	return barrier ? i915_request_await_dma_fence(rq, &barrier->fence) : 0;
+}
+
+static int add_timeline_barrier(struct i915_request *rq)
+{
+	return add_barrier(rq, &rq->timeline->barrier);
+}
+
 /**
  * i915_request_alloc - allocate a request structure
  *
@@ -660,6 +673,10 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	 */
 	rq->head = rq->ring->emit;
 
+	ret = add_timeline_barrier(rq);
+	if (ret)
+		goto err_unwind;
+
 	ret = engine->request_alloc(rq);
 	if (ret)
 		goto err_unwind;
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 4667cc08c416..6d5774cb8504 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -33,6 +33,7 @@ void i915_timeline_init(struct drm_i915_private *i915,
 
 	spin_lock_init(&timeline->lock);
 
+	init_request_active(&timeline->barrier, NULL);
 	init_request_active(&timeline->last_request, NULL);
 	INIT_LIST_HEAD(&timeline->requests);
 
@@ -69,6 +70,7 @@ void i915_timelines_park(struct drm_i915_private *i915)
 void i915_timeline_fini(struct i915_timeline *timeline)
 {
 	GEM_BUG_ON(!list_empty(&timeline->requests));
+	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
 	i915_syncmap_free(&timeline->sync);
 
@@ -90,6 +92,25 @@ i915_timeline_create(struct drm_i915_private *i915, const char *name)
 	return timeline;
 }
 
+int i915_timeline_set_barrier(struct i915_timeline *tl, struct i915_request *rq)
+{
+	struct i915_request *old;
+	int err;
+
+	lockdep_assert_held(&rq->i915->drm.struct_mutex);
+
+	/* Must maintain ordering wrt existing barriers */
+	old = i915_gem_active_raw(&tl->barrier, &rq->i915->drm.struct_mutex);
+	if (old) {
+		err = i915_request_await_dma_fence(rq, &old->fence);
+		if (err)
+			return err;
+	}
+
+	i915_gem_active_set(&tl->barrier, rq);
+	return 0;
+}
+
 void __i915_timeline_free(struct kref *kref)
 {
 	struct i915_timeline *timeline =
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index 38c1e15e927a..c8d7117bb205 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -64,6 +64,16 @@ struct i915_timeline {
 	 */
 	struct i915_syncmap *sync;
 
+	/**
+	 * Barrier provides the ability to serialize ordering between different
+	 * timelines.
+	 *
+	 * Users can call i915_timeline_set_barrier which will make all
+	 * subsequent submissions to this timeline be executed only after the
+	 * barrier has been completed.
+	 */
+	struct i915_gem_active barrier;
+
 	struct list_head link;
 	const char *name;
 
@@ -136,4 +146,16 @@ static inline bool i915_timeline_sync_is_later(struct i915_timeline *tl,
 
 void i915_timelines_park(struct drm_i915_private *i915);
 
+/**
+ * i915_timeline_set_barrier - orders submission between different timelines
+ * @timeline: timeline to set the barrier on
+ * @rq: request after which new submissions can proceed
+ *
+ * Sets the passed in request as the serialization point for all subsequent
+ * submissions on @timeline. Subsequent requests will not be submitted to GPU
+ * until the barrier has been completed.
+ */
+int i915_timeline_set_barrier(struct i915_timeline *timeline,
+			      struct i915_request *rq);
+
 #endif
diff --git a/drivers/gpu/drm/i915/selftests/mock_timeline.c b/drivers/gpu/drm/i915/selftests/mock_timeline.c
index dcf3b16f5a07..408113c1cc63 100644
--- a/drivers/gpu/drm/i915/selftests/mock_timeline.c
+++ b/drivers/gpu/drm/i915/selftests/mock_timeline.c
@@ -14,6 +14,7 @@ void mock_timeline_init(struct i915_timeline *timeline, u64 context)
 
 	spin_lock_init(&timeline->lock);
 
+	init_request_active(&timeline->barrier, NULL);
 	init_request_active(&timeline->last_request, NULL);
 	INIT_LIST_HEAD(&timeline->requests);
 
-- 
2.20.1

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

[PATCH 18/33] drm/i915: Move list of timelines under its own lock

[Chris Wilson]

Currently, the list of timelines is serialised by the struct_mutex, but
to alleviate difficulties with using that mutex in future, move the
list management under its own dedicated mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h               |   5 +-
 drivers/gpu/drm/i915/i915_gem.c               | 103 ++++++++++--------
 drivers/gpu/drm/i915/i915_reset.c             |   8 +-
 drivers/gpu/drm/i915/i915_timeline.c          |  38 ++++++-
 drivers/gpu/drm/i915/i915_timeline.h          |   3 +
 .../gpu/drm/i915/selftests/mock_gem_device.c  |   7 +-
 .../gpu/drm/i915/selftests/mock_timeline.c    |   3 +-
 7 files changed, 109 insertions(+), 58 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 0133d1da3d3c..8a181b455197 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1975,7 +1975,10 @@ struct drm_i915_private {
 		void (*resume)(struct drm_i915_private *);
 		void (*cleanup_engine)(struct intel_engine_cs *engine);
 
-		struct list_head timelines;
+		struct i915_gt_timelines {
+			struct mutex mutex; /* protects list, tainted by GPU */
+			struct list_head list;
+		} timelines;
 
 		struct list_head active_rings;
 		struct list_head closed_vma;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 653c7ba4c69f..d68f3fdd8a8e 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3224,33 +3224,6 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	return ret;
 }
 
-static long wait_for_timeline(struct i915_timeline *tl,
-			      unsigned int flags, long timeout)
-{
-	struct i915_request *rq;
-
-	rq = i915_gem_active_get_unlocked(&tl->last_request);
-	if (!rq)
-		return timeout;
-
-	/*
-	 * "Race-to-idle".
-	 *
-	 * Switching to the kernel context is often used a synchronous
-	 * step prior to idling, e.g. in suspend for flushing all
-	 * current operations to memory before sleeping. These we
-	 * want to complete as quickly as possible to avoid prolonged
-	 * stalls, so allow the gpu to boost to maximum clocks.
-	 */
-	if (flags & I915_WAIT_FOR_IDLE_BOOST)
-		gen6_rps_boost(rq, NULL);
-
-	timeout = i915_request_wait(rq, flags, timeout);
-	i915_request_put(rq);
-
-	return timeout;
-}
-
 static int wait_for_engines(struct drm_i915_private *i915)
 {
 	if (wait_for(intel_engines_are_idle(i915), I915_IDLE_ENGINES_TIMEOUT)) {
@@ -3264,6 +3237,52 @@ static int wait_for_engines(struct drm_i915_private *i915)
 	return 0;
 }
 
+static long
+wait_for_timelines(struct drm_i915_private *i915,
+		   unsigned int flags, long timeout)
+{
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
+	struct i915_timeline *tl;
+
+	if (!READ_ONCE(i915->gt.active_requests))
+		return timeout;
+
+	mutex_lock(&gt->mutex);
+	list_for_each_entry(tl, &gt->list, link) {
+		struct i915_request *rq;
+
+		rq = i915_gem_active_get_unlocked(&tl->last_request);
+		if (!rq)
+			continue;
+
+		mutex_unlock(&gt->mutex);
+
+		/*
+		 * "Race-to-idle".
+		 *
+		 * Switching to the kernel context is often used a synchronous
+		 * step prior to idling, e.g. in suspend for flushing all
+		 * current operations to memory before sleeping. These we
+		 * want to complete as quickly as possible to avoid prolonged
+		 * stalls, so allow the gpu to boost to maximum clocks.
+		 */
+		if (flags & I915_WAIT_FOR_IDLE_BOOST)
+			gen6_rps_boost(rq, NULL);
+
+		timeout = i915_request_wait(rq, flags, timeout);
+		i915_request_put(rq);
+		if (timeout < 0)
+			return timeout;
+
+		/* restart after reacquiring the lock */
+		mutex_lock(&gt->mutex);
+		tl = list_entry(&gt->list, typeof(*tl), link);
+	}
+	mutex_unlock(&gt->mutex);
+
+	return timeout;
+}
+
 int i915_gem_wait_for_idle(struct drm_i915_private *i915,
 			   unsigned int flags, long timeout)
 {
@@ -3275,17 +3294,15 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915,
 	if (!READ_ONCE(i915->gt.awake))
 		return 0;
 
+	timeout = wait_for_timelines(i915, flags, timeout);
+	if (timeout < 0)
+		return timeout;
+
 	if (flags & I915_WAIT_LOCKED) {
-		struct i915_timeline *tl;
 		int err;
 
 		lockdep_assert_held(&i915->drm.struct_mutex);
 
-		list_for_each_entry(tl, &i915->gt.timelines, link) {
-			timeout = wait_for_timeline(tl, flags, timeout);
-			if (timeout < 0)
-				return timeout;
-		}
 		if (GEM_SHOW_DEBUG() && !timeout) {
 			/* Presume that timeout was non-zero to begin with! */
 			dev_warn(&i915->drm.pdev->dev,
@@ -3299,17 +3316,6 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915,
 
 		i915_retire_requests(i915);
 		GEM_BUG_ON(i915->gt.active_requests);
-	} else {
-		struct intel_engine_cs *engine;
-		enum intel_engine_id id;
-
-		for_each_engine(engine, i915, id) {
-			struct i915_timeline *tl = &engine->timeline;
-
-			timeout = wait_for_timeline(tl, flags, timeout);
-			if (timeout < 0)
-				return timeout;
-		}
 	}
 
 	return 0;
@@ -5010,6 +5016,8 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 		dev_priv->gt.cleanup_engine = intel_engine_cleanup;
 	}
 
+	i915_timelines_init(dev_priv);
+
 	ret = i915_gem_init_userptr(dev_priv);
 	if (ret)
 		return ret;
@@ -5132,8 +5140,10 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 err_uc_misc:
 	intel_uc_fini_misc(dev_priv);
 
-	if (ret != -EIO)
+	if (ret != -EIO) {
 		i915_gem_cleanup_userptr(dev_priv);
+		i915_timelines_fini(dev_priv);
+	}
 
 	if (ret == -EIO) {
 		mutex_lock(&dev_priv->drm.struct_mutex);
@@ -5184,6 +5194,7 @@ void i915_gem_fini(struct drm_i915_private *dev_priv)
 
 	intel_uc_fini_misc(dev_priv);
 	i915_gem_cleanup_userptr(dev_priv);
+	i915_timelines_fini(dev_priv);
 
 	i915_gem_drain_freed_objects(dev_priv);
 
@@ -5286,7 +5297,6 @@ int i915_gem_init_early(struct drm_i915_private *dev_priv)
 	if (!dev_priv->priorities)
 		goto err_dependencies;
 
-	INIT_LIST_HEAD(&dev_priv->gt.timelines);
 	INIT_LIST_HEAD(&dev_priv->gt.active_rings);
 	INIT_LIST_HEAD(&dev_priv->gt.closed_vma);
 
@@ -5330,7 +5340,6 @@ void i915_gem_cleanup_early(struct drm_i915_private *dev_priv)
 	GEM_BUG_ON(!llist_empty(&dev_priv->mm.free_list));
 	GEM_BUG_ON(atomic_read(&dev_priv->mm.free_count));
 	WARN_ON(dev_priv->mm.object_count);
-	WARN_ON(!list_empty(&dev_priv->gt.timelines));
 
 	kmem_cache_destroy(dev_priv->priorities);
 	kmem_cache_destroy(dev_priv->dependencies);
diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index 7299b4f0d6d7..c57b76aebd8f 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -854,7 +854,8 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	 *
 	 * No more can be submitted until we reset the wedged bit.
 	 */
-	list_for_each_entry(tl, &i915->gt.timelines, link) {
+	mutex_lock(&i915->gt.timelines.mutex);
+	list_for_each_entry(tl, &i915->gt.timelines.list, link) {
 		struct i915_request *rq;
 		long timeout;
 
@@ -876,9 +877,12 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 		timeout = dma_fence_default_wait(&rq->fence, true,
 						 MAX_SCHEDULE_TIMEOUT);
 		i915_request_put(rq);
-		if (timeout < 0)
+		if (timeout < 0) {
+			mutex_unlock(&i915->gt.timelines.mutex);
 			goto unlock;
+		}
 	}
+	mutex_unlock(&i915->gt.timelines.mutex);
 
 	intel_engines_sanitize(i915, false);
 
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 6d5774cb8504..79ab03a0fdfe 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -13,7 +13,7 @@ void i915_timeline_init(struct drm_i915_private *i915,
 			struct i915_timeline *timeline,
 			const char *name)
 {
-	lockdep_assert_held(&i915->drm.struct_mutex);
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
 
 	/*
 	 * Ideally we want a set of engines on a single leaf as we expect
@@ -23,9 +23,12 @@ void i915_timeline_init(struct drm_i915_private *i915,
 	 */
 	BUILD_BUG_ON(KSYNCMAP < I915_NUM_ENGINES);
 
+	timeline->i915 = i915;
 	timeline->name = name;
 
-	list_add(&timeline->link, &i915->gt.timelines);
+	mutex_lock(&gt->mutex);
+	list_add(&timeline->link, &gt->list);
+	mutex_unlock(&gt->mutex);
 
 	/* Called during early_init before we know how many engines there are */
 
@@ -40,6 +43,17 @@ void i915_timeline_init(struct drm_i915_private *i915,
 	i915_syncmap_init(&timeline->sync);
 }
 
+void i915_timelines_init(struct drm_i915_private *i915)
+{
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
+
+	mutex_init(&gt->mutex);
+	INIT_LIST_HEAD(&gt->list);
+
+	/* via i915_gem_wait_for_idle() */
+	i915_gem_shrinker_taints_mutex(i915, &gt->mutex);
+}
+
 /**
  * i915_timelines_park - called when the driver idles
  * @i915: the drm_i915_private device
@@ -52,11 +66,11 @@ void i915_timeline_init(struct drm_i915_private *i915,
  */
 void i915_timelines_park(struct drm_i915_private *i915)
 {
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
 	struct i915_timeline *timeline;
 
-	lockdep_assert_held(&i915->drm.struct_mutex);
-
-	list_for_each_entry(timeline, &i915->gt.timelines, link) {
+	mutex_lock(&gt->mutex);
+	list_for_each_entry(timeline, &gt->list, link) {
 		/*
 		 * All known fences are completed so we can scrap
 		 * the current sync point tracking and start afresh,
@@ -65,16 +79,21 @@ void i915_timelines_park(struct drm_i915_private *i915)
 		 */
 		i915_syncmap_free(&timeline->sync);
 	}
+	mutex_unlock(&gt->mutex);
 }
 
 void i915_timeline_fini(struct i915_timeline *timeline)
 {
+	struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
+
 	GEM_BUG_ON(!list_empty(&timeline->requests));
 	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
 	i915_syncmap_free(&timeline->sync);
 
+	mutex_lock(&gt->mutex);
 	list_del(&timeline->link);
+	mutex_unlock(&gt->mutex);
 }
 
 struct i915_timeline *
@@ -120,6 +139,15 @@ void __i915_timeline_free(struct kref *kref)
 	kfree(timeline);
 }
 
+void i915_timelines_fini(struct drm_i915_private *i915)
+{
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
+
+	GEM_BUG_ON(!list_empty(&gt->list));
+
+	mutex_destroy(&gt->mutex);
+}
+
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/mock_timeline.c"
 #include "selftests/i915_timeline.c"
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index c8d7117bb205..b0df513b6ca3 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -76,6 +76,7 @@ struct i915_timeline {
 
 	struct list_head link;
 	const char *name;
+	struct drm_i915_private *i915;
 
 	struct kref kref;
 };
@@ -144,7 +145,9 @@ static inline bool i915_timeline_sync_is_later(struct i915_timeline *tl,
 	return __i915_timeline_sync_is_later(tl, fence->context, fence->seqno);
 }
 
+void i915_timelines_init(struct drm_i915_private *i915);
 void i915_timelines_park(struct drm_i915_private *i915);
+void i915_timelines_fini(struct drm_i915_private *i915);
 
 /**
  * i915_timeline_set_barrier - orders submission between different timelines
diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
index 8ab5a2688a0c..14ae46fda49f 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
@@ -68,13 +68,14 @@ static void mock_device_release(struct drm_device *dev)
 	i915_gem_contexts_fini(i915);
 	mutex_unlock(&i915->drm.struct_mutex);
 
+	i915_timelines_fini(i915);
+
 	drain_workqueue(i915->wq);
 	i915_gem_drain_freed_objects(i915);
 
 	mutex_lock(&i915->drm.struct_mutex);
 	mock_fini_ggtt(&i915->ggtt);
 	mutex_unlock(&i915->drm.struct_mutex);
-	WARN_ON(!list_empty(&i915->gt.timelines));
 
 	destroy_workqueue(i915->wq);
 
@@ -226,7 +227,8 @@ struct drm_i915_private *mock_gem_device(void)
 	if (!i915->priorities)
 		goto err_dependencies;
 
-	INIT_LIST_HEAD(&i915->gt.timelines);
+	i915_timelines_init(i915);
+
 	INIT_LIST_HEAD(&i915->gt.active_rings);
 	INIT_LIST_HEAD(&i915->gt.closed_vma);
 
@@ -253,6 +255,7 @@ struct drm_i915_private *mock_gem_device(void)
 	i915_gem_contexts_fini(i915);
 err_unlock:
 	mutex_unlock(&i915->drm.struct_mutex);
+	i915_timelines_fini(i915);
 	kmem_cache_destroy(i915->priorities);
 err_dependencies:
 	kmem_cache_destroy(i915->dependencies);
diff --git a/drivers/gpu/drm/i915/selftests/mock_timeline.c b/drivers/gpu/drm/i915/selftests/mock_timeline.c
index 408113c1cc63..e5659aaa856d 100644
--- a/drivers/gpu/drm/i915/selftests/mock_timeline.c
+++ b/drivers/gpu/drm/i915/selftests/mock_timeline.c
@@ -10,6 +10,7 @@
 
 void mock_timeline_init(struct i915_timeline *timeline, u64 context)
 {
+	timeline->i915 = NULL;
 	timeline->fence_context = context;
 
 	spin_lock_init(&timeline->lock);
@@ -25,5 +26,5 @@ void mock_timeline_init(struct i915_timeline *timeline, u64 context)
 
 void mock_timeline_fini(struct i915_timeline *timeline)
 {
-	i915_timeline_fini(timeline);
+	i915_syncmap_free(&timeline->sync);
 }
-- 
2.20.1

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

[PATCH 19/33] drm/i915: Introduce concept of per-timeline (context) HWSP

[Chris Wilson]

Supplement the per-engine HWSP with a per-timeline HWSP. That is a
per-request pointer through which we can check a local seqno,
abstracting away the presumption of a global seqno. In this first step,
we point each request back into the engine's HWSP so everything
continues to work with the global timeline.

v2: s/i915_request_hwsp/hwsp_seqno/ to emphasis that this is the current
HW value and that we are accessing it via i915_request merely as a
convenience.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
---
 drivers/gpu/drm/i915/i915_request.c | 16 ++++++----
 drivers/gpu/drm/i915/i915_request.h | 45 ++++++++++++++++++++++++-----
 drivers/gpu/drm/i915/intel_lrc.c    |  9 ++++--
 3 files changed, 55 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 894b32258340..79eb1957cf99 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -182,10 +182,11 @@ static void free_capture_list(struct i915_request *request)
 static void __retire_engine_request(struct intel_engine_cs *engine,
 				    struct i915_request *rq)
 {
-	GEM_TRACE("%s(%s) fence %llx:%lld, global=%d, current %d\n",
+	GEM_TRACE("%s(%s) fence %llx:%lld, global=%d, current %d:%d\n",
 		  __func__, engine->name,
 		  rq->fence.context, rq->fence.seqno,
 		  rq->global_seqno,
+		  hwsp_seqno(rq),
 		  intel_engine_get_seqno(engine));
 
 	GEM_BUG_ON(!i915_request_completed(rq));
@@ -244,10 +245,11 @@ static void i915_request_retire(struct i915_request *request)
 {
 	struct i915_gem_active *active, *next;
 
-	GEM_TRACE("%s fence %llx:%lld, global=%d, current %d\n",
+	GEM_TRACE("%s fence %llx:%lld, global=%d, current %d:%d\n",
 		  request->engine->name,
 		  request->fence.context, request->fence.seqno,
 		  request->global_seqno,
+		  hwsp_seqno(request),
 		  intel_engine_get_seqno(request->engine));
 
 	lockdep_assert_held(&request->i915->drm.struct_mutex);
@@ -307,10 +309,11 @@ void i915_request_retire_upto(struct i915_request *rq)
 	struct intel_ring *ring = rq->ring;
 	struct i915_request *tmp;
 
-	GEM_TRACE("%s fence %llx:%lld, global=%d, current %d\n",
+	GEM_TRACE("%s fence %llx:%lld, global=%d, current %d:%d\n",
 		  rq->engine->name,
 		  rq->fence.context, rq->fence.seqno,
 		  rq->global_seqno,
+		  hwsp_seqno(rq),
 		  intel_engine_get_seqno(rq->engine));
 
 	lockdep_assert_held(&rq->i915->drm.struct_mutex);
@@ -355,10 +358,11 @@ void __i915_request_submit(struct i915_request *request)
 	struct intel_engine_cs *engine = request->engine;
 	u32 seqno;
 
-	GEM_TRACE("%s fence %llx:%lld -> global=%d, current %d\n",
+	GEM_TRACE("%s fence %llx:%lld -> global=%d, current %d:%d\n",
 		  engine->name,
 		  request->fence.context, request->fence.seqno,
 		  engine->timeline.seqno + 1,
+		  hwsp_seqno(request),
 		  intel_engine_get_seqno(engine));
 
 	GEM_BUG_ON(!irqs_disabled());
@@ -405,10 +409,11 @@ void __i915_request_unsubmit(struct i915_request *request)
 {
 	struct intel_engine_cs *engine = request->engine;
 
-	GEM_TRACE("%s fence %llx:%lld <- global=%d, current %d\n",
+	GEM_TRACE("%s fence %llx:%lld <- global=%d, current %d:%d\n",
 		  engine->name,
 		  request->fence.context, request->fence.seqno,
 		  request->global_seqno,
+		  hwsp_seqno(request),
 		  intel_engine_get_seqno(engine));
 
 	GEM_BUG_ON(!irqs_disabled());
@@ -629,6 +634,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	rq->ring = ce->ring;
 	rq->timeline = ce->ring->timeline;
 	GEM_BUG_ON(rq->timeline == &engine->timeline);
+	rq->hwsp_seqno = &engine->status_page.addr[I915_GEM_HWS_INDEX];
 
 	spin_lock_init(&rq->lock);
 	dma_fence_init(&rq->fence,
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index c0f084ca4f29..ade010fe6e26 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -130,6 +130,13 @@ struct i915_request {
 	struct i915_sched_node sched;
 	struct i915_dependency dep;
 
+	/*
+	 * A convenience pointer to the current breadcrumb value stored in
+	 * the HW status page (or our timeline's local equivalent). The full
+	 * path would be rq->hw_context->ring->timeline->hwsp_seqno.
+	 */
+	const u32 *hwsp_seqno;
+
 	/**
 	 * GEM sequence number associated with this request on the
 	 * global execution timeline. It is zero when the request is not
@@ -285,11 +292,6 @@ static inline bool i915_request_signaled(const struct i915_request *rq)
 	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &rq->fence.flags);
 }
 
-static inline bool intel_engine_has_started(struct intel_engine_cs *engine,
-					    u32 seqno);
-static inline bool intel_engine_has_completed(struct intel_engine_cs *engine,
-					      u32 seqno);
-
 /**
  * Returns true if seq1 is later than seq2.
  */
@@ -298,6 +300,35 @@ static inline bool i915_seqno_passed(u32 seq1, u32 seq2)
 	return (s32)(seq1 - seq2) >= 0;
 }
 
+static inline u32 __hwsp_seqno(const struct i915_request *rq)
+{
+	return READ_ONCE(*rq->hwsp_seqno);
+}
+
+/**
+ * hwsp_seqno - the current breadcrumb value in the HW status page
+ * @rq: the request, to chase the relevant HW status page
+ *
+ * The emphasis in naming here is that hwsp_seqno() is not a property of the
+ * request, but an indication of the current HW state (associated with this
+ * request). Its value will change as the GPU executes more requests.
+ *
+ * Returns the current breadcrumb value in the associated HW status page (or
+ * the local timeline's equivalent) for this request. The request itself
+ * has the associated breadcrumb value of rq->fence.seqno, when the HW
+ * status page has that breadcrumb or later, this request is complete.
+ */
+static inline u32 hwsp_seqno(const struct i915_request *rq)
+{
+	u32 seqno;
+
+	rcu_read_lock(); /* the HWSP may be freed at runtime */
+	seqno = __hwsp_seqno(rq);
+	rcu_read_unlock();
+
+	return seqno;
+}
+
 /**
  * i915_request_started - check if the request has begun being executed
  * @rq: the request
@@ -315,14 +346,14 @@ static inline bool i915_request_started(const struct i915_request *rq)
 	if (!seqno) /* not yet submitted to HW */
 		return false;
 
-	return intel_engine_has_started(rq->engine, seqno);
+	return i915_seqno_passed(hwsp_seqno(rq), seqno - 1);
 }
 
 static inline bool
 __i915_request_completed(const struct i915_request *rq, u32 seqno)
 {
 	GEM_BUG_ON(!seqno);
-	return intel_engine_has_completed(rq->engine, seqno) &&
+	return i915_seqno_passed(hwsp_seqno(rq), seqno) &&
 		seqno == i915_request_global_seqno(rq);
 }
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 891878ab93a4..4f4dfb6e8854 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -553,11 +553,12 @@ static void execlists_submit_ports(struct intel_engine_cs *engine)
 			desc = execlists_update_context(rq);
 			GEM_DEBUG_EXEC(port[n].context_id = upper_32_bits(desc));
 
-			GEM_TRACE("%s in[%d]:  ctx=%d.%d, global=%d (fence %llx:%lld) (current %d), prio=%d\n",
+			GEM_TRACE("%s in[%d]:  ctx=%d.%d, global=%d (fence %llx:%lld) (current %d:%d), prio=%d\n",
 				  engine->name, n,
 				  port[n].context_id, count,
 				  rq->global_seqno,
 				  rq->fence.context, rq->fence.seqno,
+				  hwsp_seqno(rq),
 				  intel_engine_get_seqno(engine),
 				  rq_prio(rq));
 		} else {
@@ -848,11 +849,12 @@ execlists_cancel_port_requests(struct intel_engine_execlists * const execlists)
 	while (num_ports-- && port_isset(port)) {
 		struct i915_request *rq = port_request(port);
 
-		GEM_TRACE("%s:port%u global=%d (fence %llx:%lld), (current %d)\n",
+		GEM_TRACE("%s:port%u global=%d (fence %llx:%lld), (current %d:%d)\n",
 			  rq->engine->name,
 			  (unsigned int)(port - execlists->port),
 			  rq->global_seqno,
 			  rq->fence.context, rq->fence.seqno,
+			  hwsp_seqno(rq),
 			  intel_engine_get_seqno(rq->engine));
 
 		GEM_BUG_ON(!execlists->active);
@@ -1078,12 +1080,13 @@ static void process_csb(struct intel_engine_cs *engine)
 						EXECLISTS_ACTIVE_USER));
 
 		rq = port_unpack(port, &count);
-		GEM_TRACE("%s out[0]: ctx=%d.%d, global=%d (fence %llx:%lld) (current %d), prio=%d\n",
+		GEM_TRACE("%s out[0]: ctx=%d.%d, global=%d (fence %llx:%lld) (current %d:%d), prio=%d\n",
 			  engine->name,
 			  port->context_id, count,
 			  rq ? rq->global_seqno : 0,
 			  rq ? rq->fence.context : 0,
 			  rq ? rq->fence.seqno : 0,
+			  rq ? hwsp_seqno(rq) : 0,
 			  intel_engine_get_seqno(engine),
 			  rq ? rq_prio(rq) : 0);
 
-- 
2.20.1

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

[PATCH 20/33] drm/i915: Enlarge vma->pin_count

[Chris Wilson]

Previously we only accommodated having a vma pinned by a small number of
users, with the maximum being pinned for use by the display engine. As
such, we used a small bitfield only large enough to allow the vma to
be pinned twice (for back/front buffers) in each scanout plane. Keeping
the maximum permissible pin_count small allows us to quickly catch a
potential leak. However, as we want to split a 4096B page into 64
different cachelines and pin each cacheline for use by a different
timeline, we will exceed the current maximum permissible vma->pin_count
and so time has come to enlarge it.

Whilst we are here, try to pull together the similar bits:

Address/layout specification:
 - bias, mappable, zone_4g: address limit specifiers
 - fixed: address override, limits still apply though
 - high: not strictly an address limit, but an address direction to search

Search controls:
 - nonblock, nonfault, noevict

v2: Rewrite the guideline comment on bit consumption.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: John Harrison <john.C.Harrison@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.h | 26 ++++++++---------
 drivers/gpu/drm/i915/i915_vma.h     | 45 +++++++++++++++++++----------
 2 files changed, 42 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index bd679c8c56dd..03ade71b8d9a 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -642,19 +642,19 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 
 /* Flags used by pin/bind&friends. */
 #define PIN_NONBLOCK		BIT_ULL(0)
-#define PIN_MAPPABLE		BIT_ULL(1)
-#define PIN_ZONE_4G		BIT_ULL(2)
-#define PIN_NONFAULT		BIT_ULL(3)
-#define PIN_NOEVICT		BIT_ULL(4)
-
-#define PIN_MBZ			BIT_ULL(5) /* I915_VMA_PIN_OVERFLOW */
-#define PIN_GLOBAL		BIT_ULL(6) /* I915_VMA_GLOBAL_BIND */
-#define PIN_USER		BIT_ULL(7) /* I915_VMA_LOCAL_BIND */
-#define PIN_UPDATE		BIT_ULL(8)
-
-#define PIN_HIGH		BIT_ULL(9)
-#define PIN_OFFSET_BIAS		BIT_ULL(10)
-#define PIN_OFFSET_FIXED	BIT_ULL(11)
+#define PIN_NONFAULT		BIT_ULL(1)
+#define PIN_NOEVICT		BIT_ULL(2)
+#define PIN_MAPPABLE		BIT_ULL(3)
+#define PIN_ZONE_4G		BIT_ULL(4)
+#define PIN_HIGH		BIT_ULL(5)
+#define PIN_OFFSET_BIAS		BIT_ULL(6)
+#define PIN_OFFSET_FIXED	BIT_ULL(7)
+
+#define PIN_MBZ			BIT_ULL(8) /* I915_VMA_PIN_OVERFLOW */
+#define PIN_GLOBAL		BIT_ULL(9) /* I915_VMA_GLOBAL_BIND */
+#define PIN_USER		BIT_ULL(10) /* I915_VMA_LOCAL_BIND */
+#define PIN_UPDATE		BIT_ULL(11)
+
 #define PIN_OFFSET_MASK		(-I915_GTT_PAGE_SIZE)
 
 #endif
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 7252abc73d3e..5793abe509a2 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -71,29 +71,42 @@ struct i915_vma {
 	unsigned int open_count;
 	unsigned long flags;
 	/**
-	 * How many users have pinned this object in GTT space. The following
-	 * users can each hold at most one reference: pwrite/pread, execbuffer
-	 * (objects are not allowed multiple times for the same batchbuffer),
-	 * and the framebuffer code. When switching/pageflipping, the
-	 * framebuffer code has at most two buffers pinned per crtc.
+	 * How many users have pinned this object in GTT space.
 	 *
-	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
-	 * bits with absolutely no headroom. So use 4 bits.
+	 * This is a tightly bound, fairly small number of users, so we
+	 * stuff inside the flags field so that we can both check for overflow
+	 * and detect a no-op i915_vma_pin() in a single check, while also
+	 * pinning the vma.
+	 *
+	 * The worst case display setup would have the same vma pinned for
+	 * use on each plane on each crtc, while also building the next atomic
+	 * state and holding a pin for the length of the cleanup queue. In the
+	 * future, the flip queue may be increased from 1.
+	 * Estimated worst case: 3 [qlen] * 4 [max crtcs] * 7 [max planes] = 84
+	 *
+	 * For GEM, the number of concurrent users for pwrite/pread is
+	 * unbounded. For execbuffer, it is currently one but will in future
+	 * be extended to allow multiple clients to pin vma concurrently.
+	 *
+	 * We also use suballocated pages, with each suballocation claiming
+	 * its own pin on the shared vma. At present, this is limited to
+	 * exclusive cachelines of a single page, so a maximum of 64 possible
+	 * users.
 	 */
-#define I915_VMA_PIN_MASK 0xf
-#define I915_VMA_PIN_OVERFLOW	BIT(5)
+#define I915_VMA_PIN_MASK 0xff
+#define I915_VMA_PIN_OVERFLOW	BIT(8)
 
 	/** Flags and address space this VMA is bound to */
-#define I915_VMA_GLOBAL_BIND	BIT(6)
-#define I915_VMA_LOCAL_BIND	BIT(7)
+#define I915_VMA_GLOBAL_BIND	BIT(9)
+#define I915_VMA_LOCAL_BIND	BIT(10)
 #define I915_VMA_BIND_MASK (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND | I915_VMA_PIN_OVERFLOW)
 
-#define I915_VMA_GGTT		BIT(8)
-#define I915_VMA_CAN_FENCE	BIT(9)
-#define I915_VMA_CLOSED		BIT(10)
-#define I915_VMA_USERFAULT_BIT	11
+#define I915_VMA_GGTT		BIT(11)
+#define I915_VMA_CAN_FENCE	BIT(12)
+#define I915_VMA_CLOSED		BIT(13)
+#define I915_VMA_USERFAULT_BIT	14
 #define I915_VMA_USERFAULT	BIT(I915_VMA_USERFAULT_BIT)
-#define I915_VMA_GGTT_WRITE	BIT(12)
+#define I915_VMA_GGTT_WRITE	BIT(15)
 
 	unsigned int active_count;
 	struct rb_root active;
-- 
2.20.1

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

[PATCH 21/33] drm/i915: Allocate a status page for each timeline

[Chris Wilson]

Allocate a page for use as a status page by a group of timelines, as we
only need a dword of storage for each (rounded up to the cacheline for
safety) we can pack multiple timelines into the same page. Each timeline
will then be able to track its own HW seqno.

v2: Reuse the common per-engine HWSP for the solitary ringbuffer
timeline, so that we do not have to emit (using per-gen specialised
vfuncs) the breadcrumb into the distinct timeline HWSP and instead can
keep on using the common MI_STORE_DWORD_INDEX. However, to maintain the
sleight-of-hand for the global/per-context seqno switchover, we will
store both temporarily (and so use a custom offset for the shared timeline
HWSP until the switch over).

v3: Keep things simple and allocate a page for each timeline, page
sharing comes next.

v4: I was caught repeating the same MI_STORE_DWORD_IMM over and over
again in selftests.

v5: And caught red handed copying create timeline + check.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_timeline.c          | 121 ++++++-
 drivers/gpu/drm/i915/i915_timeline.h          |  21 +-
 drivers/gpu/drm/i915/intel_engine_cs.c        |  76 ++--
 drivers/gpu/drm/i915/intel_lrc.c              |  22 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c       |  10 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h       |   6 +-
 .../drm/i915/selftests/i915_live_selftests.h  |   1 +
 .../drm/i915/selftests/i915_mock_selftests.h  |   2 +-
 .../gpu/drm/i915/selftests/i915_timeline.c    | 326 +++++++++++++++++-
 drivers/gpu/drm/i915/selftests/mock_engine.c  |  14 +-
 10 files changed, 543 insertions(+), 56 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 79ab03a0fdfe..5b5f9dacfce9 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -9,28 +9,78 @@
 #include "i915_timeline.h"
 #include "i915_syncmap.h"
 
-void i915_timeline_init(struct drm_i915_private *i915,
-			struct i915_timeline *timeline,
-			const char *name)
+static struct i915_vma *__hwsp_alloc(struct drm_i915_private *i915)
+{
+	struct drm_i915_gem_object *obj;
+	struct i915_vma *vma;
+
+	obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		return ERR_CAST(obj);
+
+	i915_gem_object_set_cache_coherency(obj, I915_CACHE_LLC);
+
+	vma = i915_vma_instance(obj, &i915->ggtt.vm, NULL);
+	if (IS_ERR(vma))
+		i915_gem_object_put(obj);
+
+	return vma;
+}
+
+static int hwsp_alloc(struct i915_timeline *timeline)
+{
+	struct i915_vma *vma;
+
+	vma = __hwsp_alloc(timeline->i915);
+	if (IS_ERR(vma))
+		return PTR_ERR(vma);
+
+	timeline->hwsp_ggtt = vma;
+	timeline->hwsp_offset = 0;
+
+	return 0;
+}
+
+int i915_timeline_init(struct drm_i915_private *i915,
+		       struct i915_timeline *timeline,
+		       const char *name,
+		       struct i915_vma *global_hwsp)
 {
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
+	void *vaddr;
+	int err;
 
 	/*
 	 * Ideally we want a set of engines on a single leaf as we expect
 	 * to mostly be tracking synchronisation between engines. It is not
 	 * a huge issue if this is not the case, but we may want to mitigate
 	 * any page crossing penalties if they become an issue.
+	 *
+	 * Called during early_init before we know how many engines there are.
 	 */
 	BUILD_BUG_ON(KSYNCMAP < I915_NUM_ENGINES);
 
 	timeline->i915 = i915;
 	timeline->name = name;
+	timeline->pin_count = 0;
 
-	mutex_lock(&gt->mutex);
-	list_add(&timeline->link, &gt->list);
-	mutex_unlock(&gt->mutex);
+	if (global_hwsp) {
+		timeline->hwsp_ggtt = i915_vma_get(global_hwsp);
+		timeline->hwsp_offset = I915_GEM_HWS_SEQNO_ADDR;
+	} else {
+		err = hwsp_alloc(timeline);
+		if (err)
+			return err;
+	}
+
+	vaddr = i915_gem_object_pin_map(timeline->hwsp_ggtt->obj, I915_MAP_WB);
+	if (IS_ERR(vaddr)) {
+		i915_vma_put(timeline->hwsp_ggtt);
+		return PTR_ERR(vaddr);
+	}
 
-	/* Called during early_init before we know how many engines there are */
+	timeline->hwsp_seqno =
+		memset(vaddr + timeline->hwsp_offset, 0, CACHELINE_BYTES);
 
 	timeline->fence_context = dma_fence_context_alloc(1);
 
@@ -41,6 +91,12 @@ void i915_timeline_init(struct drm_i915_private *i915,
 	INIT_LIST_HEAD(&timeline->requests);
 
 	i915_syncmap_init(&timeline->sync);
+
+	mutex_lock(&gt->mutex);
+	list_add(&timeline->link, &gt->list);
+	mutex_unlock(&gt->mutex);
+
+	return 0;
 }
 
 void i915_timelines_init(struct drm_i915_private *i915)
@@ -86,6 +142,7 @@ void i915_timeline_fini(struct i915_timeline *timeline)
 {
 	struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
 
+	GEM_BUG_ON(timeline->pin_count);
 	GEM_BUG_ON(!list_empty(&timeline->requests));
 	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
@@ -94,18 +151,29 @@ void i915_timeline_fini(struct i915_timeline *timeline)
 	mutex_lock(&gt->mutex);
 	list_del(&timeline->link);
 	mutex_unlock(&gt->mutex);
+
+	i915_gem_object_unpin_map(timeline->hwsp_ggtt->obj);
+	i915_vma_put(timeline->hwsp_ggtt);
 }
 
 struct i915_timeline *
-i915_timeline_create(struct drm_i915_private *i915, const char *name)
+i915_timeline_create(struct drm_i915_private *i915,
+		     const char *name,
+		     struct i915_vma *global_hwsp)
 {
 	struct i915_timeline *timeline;
+	int err;
 
 	timeline = kzalloc(sizeof(*timeline), GFP_KERNEL);
 	if (!timeline)
 		return ERR_PTR(-ENOMEM);
 
-	i915_timeline_init(i915, timeline, name);
+	err = i915_timeline_init(i915, timeline, name, global_hwsp);
+	if (err) {
+		kfree(timeline);
+		return ERR_PTR(err);
+	}
+
 	kref_init(&timeline->kref);
 
 	return timeline;
@@ -130,6 +198,41 @@ int i915_timeline_set_barrier(struct i915_timeline *tl, struct i915_request *rq)
 	return 0;
 }
 
+int i915_timeline_pin(struct i915_timeline *tl)
+{
+	int err;
+
+	if (tl->pin_count++)
+		return 0;
+	GEM_BUG_ON(!tl->pin_count);
+
+	err = i915_vma_pin(tl->hwsp_ggtt, 0, 0, PIN_GLOBAL | PIN_HIGH);
+	if (err)
+		goto unpin;
+
+	return 0;
+
+unpin:
+	tl->pin_count = 0;
+	return err;
+}
+
+void i915_timeline_unpin(struct i915_timeline *tl)
+{
+	GEM_BUG_ON(!tl->pin_count);
+	if (--tl->pin_count)
+		return;
+
+	/*
+	 * Since this timeline is idle, all bariers upon which we were waiting
+	 * must also be complete and so we can discard the last used barriers
+	 * without loss of information.
+	 */
+	i915_syncmap_free(&tl->sync);
+
+	__i915_vma_unpin(tl->hwsp_ggtt);
+}
+
 void __i915_timeline_free(struct kref *kref)
 {
 	struct i915_timeline *timeline =
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index b0df513b6ca3..8e29182a2584 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -32,6 +32,8 @@
 #include "i915_syncmap.h"
 #include "i915_utils.h"
 
+struct i915_vma;
+
 struct i915_timeline {
 	u64 fence_context;
 	u32 seqno;
@@ -40,6 +42,11 @@ struct i915_timeline {
 #define TIMELINE_CLIENT 0 /* default subclass */
 #define TIMELINE_ENGINE 1
 
+	unsigned int pin_count;
+	const u32 *hwsp_seqno;
+	struct i915_vma *hwsp_ggtt;
+	u32 hwsp_offset;
+
 	/**
 	 * List of breadcrumbs associated with GPU requests currently
 	 * outstanding.
@@ -81,9 +88,10 @@ struct i915_timeline {
 	struct kref kref;
 };
 
-void i915_timeline_init(struct drm_i915_private *i915,
-			struct i915_timeline *tl,
-			const char *name);
+int i915_timeline_init(struct drm_i915_private *i915,
+		       struct i915_timeline *tl,
+		       const char *name,
+		       struct i915_vma *hwsp);
 void i915_timeline_fini(struct i915_timeline *tl);
 
 static inline void
@@ -106,7 +114,9 @@ i915_timeline_set_subclass(struct i915_timeline *timeline,
 }
 
 struct i915_timeline *
-i915_timeline_create(struct drm_i915_private *i915, const char *name);
+i915_timeline_create(struct drm_i915_private *i915,
+		     const char *name,
+		     struct i915_vma *global_hwsp);
 
 static inline struct i915_timeline *
 i915_timeline_get(struct i915_timeline *timeline)
@@ -145,6 +155,9 @@ static inline bool i915_timeline_sync_is_later(struct i915_timeline *tl,
 	return __i915_timeline_sync_is_later(tl, fence->context, fence->seqno);
 }
 
+int i915_timeline_pin(struct i915_timeline *tl);
+void i915_timeline_unpin(struct i915_timeline *tl);
+
 void i915_timelines_init(struct drm_i915_private *i915);
 void i915_timelines_park(struct drm_i915_private *i915);
 void i915_timelines_fini(struct drm_i915_private *i915);
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index c1dc1769e337..28a66242d698 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -484,26 +484,6 @@ static void intel_engine_init_execlist(struct intel_engine_cs *engine)
 	execlists->queue = RB_ROOT_CACHED;
 }
 
-/**
- * intel_engines_setup_common - setup engine state not requiring hw access
- * @engine: Engine to setup.
- *
- * Initializes @engine@ structure members shared between legacy and execlists
- * submission modes which do not require hardware access.
- *
- * Typically done early in the submission mode specific engine setup stage.
- */
-void intel_engine_setup_common(struct intel_engine_cs *engine)
-{
-	i915_timeline_init(engine->i915, &engine->timeline, engine->name);
-	i915_timeline_set_subclass(&engine->timeline, TIMELINE_ENGINE);
-
-	intel_engine_init_execlist(engine);
-	intel_engine_init_hangcheck(engine);
-	intel_engine_init_batch_pool(engine);
-	intel_engine_init_cmd_parser(engine);
-}
-
 static void cleanup_status_page(struct intel_engine_cs *engine)
 {
 	struct i915_vma *vma;
@@ -601,6 +581,44 @@ static int init_status_page(struct intel_engine_cs *engine)
 	return ret;
 }
 
+/**
+ * intel_engines_setup_common - setup engine state not requiring hw access
+ * @engine: Engine to setup.
+ *
+ * Initializes @engine@ structure members shared between legacy and execlists
+ * submission modes which do not require hardware access.
+ *
+ * Typically done early in the submission mode specific engine setup stage.
+ */
+int intel_engine_setup_common(struct intel_engine_cs *engine)
+{
+	int err;
+
+	err = init_status_page(engine);
+	if (err)
+		return err;
+
+	err = i915_timeline_init(engine->i915,
+				 &engine->timeline,
+				 engine->name,
+				 engine->status_page.vma);
+	if (err)
+		goto err_hwsp;
+
+	i915_timeline_set_subclass(&engine->timeline, TIMELINE_ENGINE);
+
+	intel_engine_init_execlist(engine);
+	intel_engine_init_hangcheck(engine);
+	intel_engine_init_batch_pool(engine);
+	intel_engine_init_cmd_parser(engine);
+
+	return 0;
+
+err_hwsp:
+	cleanup_status_page(engine);
+	return err;
+}
+
 static void __intel_context_unpin(struct i915_gem_context *ctx,
 				  struct intel_engine_cs *engine)
 {
@@ -617,7 +635,7 @@ struct measure_breadcrumb {
 static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 {
 	struct measure_breadcrumb *frame;
-	unsigned int dw;
+	int dw = -ENOMEM;
 
 	GEM_BUG_ON(!engine->i915->gt.scratch);
 
@@ -625,7 +643,10 @@ static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 	if (!frame)
 		return -ENOMEM;
 
-	i915_timeline_init(engine->i915, &frame->timeline, engine->name);
+	if (i915_timeline_init(engine->i915,
+			       &frame->timeline, engine->name,
+			       engine->status_page.vma))
+		goto out_frame;
 
 	frame->ring.timeline = &frame->timeline;
 	frame->ring.vaddr = frame->cs;
@@ -642,8 +663,9 @@ static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
 
 	i915_timeline_fini(&frame->timeline);
-	kfree(frame);
 
+out_frame:
+	kfree(frame);
 	return dw;
 }
 
@@ -693,20 +715,14 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 	if (ret)
 		goto err_unpin_preempt;
 
-	ret = init_status_page(engine);
-	if (ret)
-		goto err_breadcrumbs;
-
 	ret = measure_breadcrumb_dw(engine);
 	if (ret < 0)
-		goto err_status_page;
+		goto err_breadcrumbs;
 
 	engine->emit_breadcrumb_dw = ret;
 
 	return 0;
 
-err_status_page:
-	cleanup_status_page(engine);
 err_breadcrumbs:
 	intel_engine_fini_breadcrumbs(engine);
 err_unpin_preempt:
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 4f4dfb6e8854..f4c52b9bdfca 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -2314,10 +2314,14 @@ logical_ring_default_irqs(struct intel_engine_cs *engine)
 	engine->irq_keep_mask = GT_CONTEXT_SWITCH_INTERRUPT << shift;
 }
 
-static void
+static int
 logical_ring_setup(struct intel_engine_cs *engine)
 {
-	intel_engine_setup_common(engine);
+	int err;
+
+	err = intel_engine_setup_common(engine);
+	if (err)
+		return err;
 
 	/* Intentionally left blank. */
 	engine->buffer = NULL;
@@ -2327,6 +2331,8 @@ logical_ring_setup(struct intel_engine_cs *engine)
 
 	logical_ring_default_vfuncs(engine);
 	logical_ring_default_irqs(engine);
+
+	return 0;
 }
 
 static int logical_ring_init(struct intel_engine_cs *engine)
@@ -2375,7 +2381,9 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
 {
 	int ret;
 
-	logical_ring_setup(engine);
+	ret = logical_ring_setup(engine);
+	if (ret)
+		return ret;
 
 	/* Override some for render ring. */
 	engine->init_context = gen8_init_rcs_context;
@@ -2404,7 +2412,11 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
 
 int logical_xcs_ring_init(struct intel_engine_cs *engine)
 {
-	logical_ring_setup(engine);
+	int err;
+
+	err = logical_ring_setup(engine);
+	if (err)
+		return err;
 
 	return logical_ring_init(engine);
 }
@@ -2737,7 +2749,7 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx,
 		goto error_deref_obj;
 	}
 
-	timeline = i915_timeline_create(ctx->i915, ctx->name);
+	timeline = i915_timeline_create(ctx->i915, ctx->name, NULL);
 	if (IS_ERR(timeline)) {
 		ret = PTR_ERR(timeline);
 		goto error_deref_obj;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index cb6d2aa2a829..174795622eb1 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1545,9 +1545,13 @@ static int intel_init_ring_buffer(struct intel_engine_cs *engine)
 	struct intel_ring *ring;
 	int err;
 
-	intel_engine_setup_common(engine);
+	err = intel_engine_setup_common(engine);
+	if (err)
+		return err;
 
-	timeline = i915_timeline_create(engine->i915, engine->name);
+	timeline = i915_timeline_create(engine->i915,
+					engine->name,
+					engine->status_page.vma);
 	if (IS_ERR(timeline)) {
 		err = PTR_ERR(timeline);
 		goto err;
@@ -1571,6 +1575,8 @@ static int intel_init_ring_buffer(struct intel_engine_cs *engine)
 	if (err)
 		goto err_unpin;
 
+	GEM_BUG_ON(ring->timeline->hwsp_ggtt != engine->status_page.vma);
+
 	return 0;
 
 err_unpin:
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 32371ae67f24..2927b712b973 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -712,7 +712,9 @@ intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
 #define I915_GEM_HWS_INDEX_ADDR		(I915_GEM_HWS_INDEX * sizeof(u32))
 #define I915_GEM_HWS_PREEMPT		0x32
 #define I915_GEM_HWS_PREEMPT_ADDR	(I915_GEM_HWS_PREEMPT * sizeof(u32))
-#define I915_GEM_HWS_SCRATCH		0x40
+#define I915_GEM_HWS_SEQNO		0x40
+#define I915_GEM_HWS_SEQNO_ADDR		(I915_GEM_HWS_SEQNO * sizeof(u32))
+#define I915_GEM_HWS_SCRATCH		0x80
 #define I915_GEM_HWS_SCRATCH_ADDR	(I915_GEM_HWS_SCRATCH * sizeof(u32))
 
 #define I915_HWS_CSB_BUF0_INDEX		0x10
@@ -818,7 +820,7 @@ intel_ring_set_tail(struct intel_ring *ring, unsigned int tail)
 
 void intel_engine_write_global_seqno(struct intel_engine_cs *engine, u32 seqno);
 
-void intel_engine_setup_common(struct intel_engine_cs *engine);
+int intel_engine_setup_common(struct intel_engine_cs *engine);
 int intel_engine_init_common(struct intel_engine_cs *engine);
 void intel_engine_cleanup_common(struct intel_engine_cs *engine);
 
diff --git a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
index a15713cae3b3..76b4f87fc853 100644
--- a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
+++ b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
@@ -13,6 +13,7 @@ selftest(sanitycheck, i915_live_sanitycheck) /* keep first (igt selfcheck) */
 selftest(uncore, intel_uncore_live_selftests)
 selftest(workarounds, intel_workarounds_live_selftests)
 selftest(requests, i915_request_live_selftests)
+selftest(timelines, i915_timeline_live_selftests)
 selftest(objects, i915_gem_object_live_selftests)
 selftest(dmabuf, i915_gem_dmabuf_live_selftests)
 selftest(coherency, i915_gem_coherency_live_selftests)
diff --git a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
index 1b70208eeea7..4a83a1c6c406 100644
--- a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
+++ b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
@@ -16,7 +16,7 @@ selftest(syncmap, i915_syncmap_mock_selftests)
 selftest(uncore, intel_uncore_mock_selftests)
 selftest(engine, intel_engine_cs_mock_selftests)
 selftest(breadcrumbs, intel_breadcrumbs_mock_selftests)
-selftest(timelines, i915_gem_timeline_mock_selftests)
+selftest(timelines, i915_timeline_mock_selftests)
 selftest(requests, i915_request_mock_selftests)
 selftest(objects, i915_gem_object_mock_selftests)
 selftest(dmabuf, i915_gem_dmabuf_mock_selftests)
diff --git a/drivers/gpu/drm/i915/selftests/i915_timeline.c b/drivers/gpu/drm/i915/selftests/i915_timeline.c
index 19f1c6a5c8fb..1585b614510d 100644
--- a/drivers/gpu/drm/i915/selftests/i915_timeline.c
+++ b/drivers/gpu/drm/i915/selftests/i915_timeline.c
@@ -7,6 +7,7 @@
 #include "../i915_selftest.h"
 #include "i915_random.h"
 
+#include "igt_flush_test.h"
 #include "mock_gem_device.h"
 #include "mock_timeline.h"
 
@@ -256,7 +257,7 @@ static int bench_sync(void *arg)
 	return 0;
 }
 
-int i915_gem_timeline_mock_selftests(void)
+int i915_timeline_mock_selftests(void)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(igt_sync),
@@ -265,3 +266,326 @@ int i915_gem_timeline_mock_selftests(void)
 
 	return i915_subtests(tests, NULL);
 }
+
+static int emit_ggtt_store_dw(struct i915_request *rq, u32 addr, u32 value)
+{
+	u32 *cs;
+
+	cs = intel_ring_begin(rq, 4);
+	if (IS_ERR(cs))
+		return PTR_ERR(cs);
+
+	if (INTEL_GEN(rq->i915) >= 8) {
+		*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
+		*cs++ = addr;
+		*cs++ = 0;
+		*cs++ = value;
+	} else if (INTEL_GEN(rq->i915) >= 4) {
+		*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
+		*cs++ = 0;
+		*cs++ = addr;
+		*cs++ = value;
+	} else {
+		*cs++ = MI_STORE_DWORD_IMM | MI_MEM_VIRTUAL;
+		*cs++ = addr;
+		*cs++ = value;
+		*cs++ = MI_NOOP;
+	}
+
+	intel_ring_advance(rq, cs);
+
+	return 0;
+}
+
+static u32 hwsp_address(const struct i915_timeline *tl)
+{
+	return i915_ggtt_offset(tl->hwsp_ggtt) + tl->hwsp_offset;
+}
+
+static struct i915_request *
+tl_write(struct i915_timeline *tl, struct intel_engine_cs *engine, u32 value)
+{
+	struct i915_request *rq;
+	int err;
+
+	lockdep_assert_held(&tl->i915->drm.struct_mutex); /* lazy rq refs */
+
+	err = i915_timeline_pin(tl);
+	if (err) {
+		rq = ERR_PTR(err);
+		goto out;
+	}
+
+	rq = i915_request_alloc(engine, engine->i915->kernel_context);
+	if (IS_ERR(rq))
+		goto out_unpin;
+
+	err = emit_ggtt_store_dw(rq, hwsp_address(tl), value);
+	i915_request_add(rq);
+	if (err)
+		rq = ERR_PTR(err);
+
+out_unpin:
+	i915_timeline_unpin(tl);
+out:
+	if (IS_ERR(rq))
+		pr_err("Failed to write to timeline!\n");
+	return rq;
+}
+
+static struct i915_timeline *
+checked_i915_timeline_create(struct drm_i915_private *i915)
+{
+	struct i915_timeline *tl;
+
+	tl = i915_timeline_create(i915, "live", NULL);
+	if (IS_ERR(tl))
+		return tl;
+
+	if (*tl->hwsp_seqno != tl->seqno) {
+		pr_err("Timeline created with incorrect breadcrumb, found %x, expected %x\n",
+		       *tl->hwsp_seqno, tl->seqno);
+		i915_timeline_put(tl);
+		return ERR_PTR(-EINVAL);
+	}
+
+	return tl;
+}
+
+static int live_hwsp_engine(void *arg)
+{
+#define NUM_TIMELINES 4096
+	struct drm_i915_private *i915 = arg;
+	struct i915_timeline **timelines;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	intel_wakeref_t wakeref;
+	unsigned long count, n;
+	int err = 0;
+
+	/*
+	 * Create a bunch of timelines and check we can write
+	 * independently to each of their breadcrumb slots.
+	 */
+
+	timelines = kvmalloc_array(NUM_TIMELINES * I915_NUM_ENGINES,
+				   sizeof(*timelines),
+				   GFP_KERNEL);
+	if (!timelines)
+		return -ENOMEM;
+
+	mutex_lock(&i915->drm.struct_mutex);
+	wakeref = intel_runtime_pm_get(i915);
+
+	count = 0;
+	for_each_engine(engine, i915, id) {
+		if (!intel_engine_can_store_dword(engine))
+			continue;
+
+		for (n = 0; n < NUM_TIMELINES; n++) {
+			struct i915_timeline *tl;
+			struct i915_request *rq;
+
+			tl = checked_i915_timeline_create(i915);
+			if (IS_ERR(tl)) {
+				err = PTR_ERR(tl);
+				goto out;
+			}
+
+			rq = tl_write(tl, engine, count);
+			if (IS_ERR(rq)) {
+				i915_timeline_put(tl);
+				err = PTR_ERR(rq);
+				goto out;
+			}
+
+			timelines[count++] = tl;
+		}
+	}
+
+out:
+	if (igt_flush_test(i915, I915_WAIT_LOCKED))
+		err = -EIO;
+
+	for (n = 0; n < count; n++) {
+		struct i915_timeline *tl = timelines[n];
+
+		if (!err && *tl->hwsp_seqno != n) {
+			pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
+			       n, *tl->hwsp_seqno);
+			err = -EINVAL;
+		}
+		i915_timeline_put(tl);
+	}
+
+	intel_runtime_pm_put(i915, wakeref);
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	kvfree(timelines);
+
+	return err;
+#undef NUM_TIMELINES
+}
+
+static int live_hwsp_alternate(void *arg)
+{
+#define NUM_TIMELINES 4096
+	struct drm_i915_private *i915 = arg;
+	struct i915_timeline **timelines;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	intel_wakeref_t wakeref;
+	unsigned long count, n;
+	int err = 0;
+
+	/*
+	 * Create a bunch of timelines and check we can write
+	 * independently to each of their breadcrumb slots with adjacent
+	 * engines.
+	 */
+
+	timelines = kvmalloc_array(NUM_TIMELINES * I915_NUM_ENGINES,
+				   sizeof(*timelines),
+				   GFP_KERNEL);
+	if (!timelines)
+		return -ENOMEM;
+
+	mutex_lock(&i915->drm.struct_mutex);
+	wakeref = intel_runtime_pm_get(i915);
+
+	count = 0;
+	for (n = 0; n < NUM_TIMELINES; n++) {
+		for_each_engine(engine, i915, id) {
+			struct i915_timeline *tl;
+			struct i915_request *rq;
+
+			if (!intel_engine_can_store_dword(engine))
+				continue;
+
+			tl = checked_i915_timeline_create(i915);
+			if (IS_ERR(tl)) {
+				err = PTR_ERR(tl);
+				goto out;
+			}
+
+			rq = tl_write(tl, engine, count);
+			if (IS_ERR(rq)) {
+				i915_timeline_put(tl);
+				err = PTR_ERR(rq);
+				goto out;
+			}
+
+			timelines[count++] = tl;
+		}
+	}
+
+out:
+	if (igt_flush_test(i915, I915_WAIT_LOCKED))
+		err = -EIO;
+
+	for (n = 0; n < count; n++) {
+		struct i915_timeline *tl = timelines[n];
+
+		if (!err && *tl->hwsp_seqno != n) {
+			pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
+			       n, *tl->hwsp_seqno);
+			err = -EINVAL;
+		}
+		i915_timeline_put(tl);
+	}
+
+	intel_runtime_pm_put(i915, wakeref);
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	kvfree(timelines);
+
+	return err;
+#undef NUM_TIMELINES
+}
+
+static int live_hwsp_recycle(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	intel_wakeref_t wakeref;
+	unsigned long count;
+	int err = 0;
+
+	/*
+	 * Check seqno writes into one timeline at a time. We expect to
+	 * recycle the breadcrumb slot between iterations and neither
+	 * want to confuse ourselves or the GPU.
+	 */
+
+	mutex_lock(&i915->drm.struct_mutex);
+	wakeref = intel_runtime_pm_get(i915);
+
+	count = 0;
+	for_each_engine(engine, i915, id) {
+		IGT_TIMEOUT(end_time);
+
+		if (!intel_engine_can_store_dword(engine))
+			continue;
+
+		do {
+			struct i915_timeline *tl;
+			struct i915_request *rq;
+
+			tl = checked_i915_timeline_create(i915);
+			if (IS_ERR(tl)) {
+				err = PTR_ERR(tl);
+				goto out;
+			}
+
+			rq = tl_write(tl, engine, count);
+			if (IS_ERR(rq)) {
+				i915_timeline_put(tl);
+				err = PTR_ERR(rq);
+				goto out;
+			}
+
+			if (i915_request_wait(rq,
+					      I915_WAIT_LOCKED,
+					      HZ / 5) < 0) {
+				pr_err("Wait for timeline writes timed out!\n");
+				i915_timeline_put(tl);
+				err = -EIO;
+				goto out;
+			}
+
+			if (*tl->hwsp_seqno != count) {
+				pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
+				       count, *tl->hwsp_seqno);
+				err = -EINVAL;
+			}
+
+			i915_timeline_put(tl);
+			count++;
+
+			if (err)
+				goto out;
+
+			i915_timelines_park(i915); /* Encourage recycling! */
+		} while (!__igt_timeout(end_time, NULL));
+	}
+
+out:
+	if (igt_flush_test(i915, I915_WAIT_LOCKED))
+		err = -EIO;
+	intel_runtime_pm_put(i915, wakeref);
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	return err;
+}
+
+int i915_timeline_live_selftests(struct drm_i915_private *i915)
+{
+	static const struct i915_subtest tests[] = {
+		SUBTEST(live_hwsp_recycle),
+		SUBTEST(live_hwsp_engine),
+		SUBTEST(live_hwsp_alternate),
+	};
+
+	return i915_subtests(tests, i915);
+}
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 4e5b4dc6df0f..919c89fd6ee5 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -39,7 +39,12 @@ static struct intel_ring *mock_ring(struct intel_engine_cs *engine)
 	if (!ring)
 		return NULL;
 
-	i915_timeline_init(engine->i915, &ring->timeline, engine->name);
+	if (i915_timeline_init(engine->i915,
+			       &ring->timeline, engine->name,
+			       NULL)) {
+		kfree(ring);
+		return NULL;
+	}
 
 	ring->base.size = sz;
 	ring->base.effective_size = sz;
@@ -208,7 +213,11 @@ struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
 	engine->base.emit_breadcrumb = mock_emit_breadcrumb;
 	engine->base.submit_request = mock_submit_request;
 
-	i915_timeline_init(i915, &engine->base.timeline, engine->base.name);
+	if (i915_timeline_init(i915,
+			       &engine->base.timeline,
+			       engine->base.name,
+			       NULL))
+		goto err_free;
 	i915_timeline_set_subclass(&engine->base.timeline, TIMELINE_ENGINE);
 
 	intel_engine_init_breadcrumbs(&engine->base);
@@ -226,6 +235,7 @@ struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
 err_breadcrumbs:
 	intel_engine_fini_breadcrumbs(&engine->base);
 	i915_timeline_fini(&engine->base.timeline);
+err_free:
 	kfree(engine);
 	return NULL;
 }
-- 
2.20.1

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

[PATCH 22/33] drm/i915: Share per-timeline HWSP using a slab suballocator

[Chris Wilson]

If we restrict ourselves to only using a cacheline for each timeline's
HWSP (we could go smaller, but want to avoid needless polluting
cachelines on different engines between different contexts), then we can
suballocate a single 4k page into 64 different timeline HWSP. By
treating each fresh allocation as a slab of 64 entries, we can keep it
around for the next 64 allocation attempts until we need to refresh the
slab cache.

John Harrison noted the issue of fragmentation leading to the same worst
case performance of one page per timeline as before, which can be
mitigated by adopting a freelist.

v2: Keep all partially allocated HWSP on a freelist

This is still without migration, so it is possible for the system to end
up with each timeline in its own page, but we ensure that no new
allocation would needless allocate a fresh page!

v3: Throw a selftest at the allocator to try and catch invalid cacheline
reuse.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h               |   4 +
 drivers/gpu/drm/i915/i915_timeline.c          | 116 ++++++++++++---
 drivers/gpu/drm/i915/i915_timeline.h          |   1 +
 drivers/gpu/drm/i915/i915_vma.h               |  12 ++
 drivers/gpu/drm/i915/selftests/i915_random.c  |  33 ++++-
 drivers/gpu/drm/i915/selftests/i915_random.h  |   3 +
 .../gpu/drm/i915/selftests/i915_timeline.c    | 140 ++++++++++++++++++
 7 files changed, 281 insertions(+), 28 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8a181b455197..6a051381f535 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1978,6 +1978,10 @@ struct drm_i915_private {
 		struct i915_gt_timelines {
 			struct mutex mutex; /* protects list, tainted by GPU */
 			struct list_head list;
+
+			/* Pack multiple timelines' seqnos into the same page */
+			spinlock_t hwsp_lock;
+			struct list_head hwsp_free_list;
 		} timelines;
 
 		struct list_head active_rings;
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 5b5f9dacfce9..95c689826c8c 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -9,6 +9,12 @@
 #include "i915_timeline.h"
 #include "i915_syncmap.h"
 
+struct i915_timeline_hwsp {
+	struct i915_vma *vma;
+	struct list_head free_link;
+	u64 free_bitmap;
+};
+
 static struct i915_vma *__hwsp_alloc(struct drm_i915_private *i915)
 {
 	struct drm_i915_gem_object *obj;
@@ -27,28 +33,91 @@ static struct i915_vma *__hwsp_alloc(struct drm_i915_private *i915)
 	return vma;
 }
 
-static int hwsp_alloc(struct i915_timeline *timeline)
+static struct i915_vma *hwsp_alloc(struct i915_timeline *timeline, int *offset)
 {
-	struct i915_vma *vma;
+	struct drm_i915_private *i915 = timeline->i915;
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
+	struct i915_timeline_hwsp *hwsp;
+	int cacheline;
 
-	vma = __hwsp_alloc(timeline->i915);
-	if (IS_ERR(vma))
-		return PTR_ERR(vma);
+	BUILD_BUG_ON(BITS_PER_TYPE(u64) * CACHELINE_BYTES > PAGE_SIZE);
 
-	timeline->hwsp_ggtt = vma;
-	timeline->hwsp_offset = 0;
+	spin_lock(&gt->hwsp_lock);
 
-	return 0;
+	/* hwsp_free_list only contains HWSP that have available cachelines */
+	hwsp = list_first_entry_or_null(&gt->hwsp_free_list,
+					typeof(*hwsp), free_link);
+	if (!hwsp) {
+		struct i915_vma *vma;
+
+		spin_unlock(&gt->hwsp_lock);
+
+		hwsp = kmalloc(sizeof(*hwsp), GFP_KERNEL);
+		if (!hwsp)
+			return ERR_PTR(-ENOMEM);
+
+		vma = __hwsp_alloc(i915);
+		if (IS_ERR(vma)) {
+			kfree(hwsp);
+			return vma;
+		}
+
+		vma->private = hwsp;
+		hwsp->vma = vma;
+		hwsp->free_bitmap = ~0ull;
+
+		spin_lock(&gt->hwsp_lock);
+		list_add(&hwsp->free_link, &gt->hwsp_free_list);
+	}
+
+	GEM_BUG_ON(!hwsp->free_bitmap);
+	cacheline = __ffs64(hwsp->free_bitmap);
+	hwsp->free_bitmap &= ~BIT_ULL(cacheline);
+	if (!hwsp->free_bitmap)
+		list_del(&hwsp->free_link);
+
+	spin_unlock(&gt->hwsp_lock);
+
+	GEM_BUG_ON(hwsp->vma->private != hwsp);
+
+	*offset = cacheline * CACHELINE_BYTES;
+	return hwsp->vma;
+}
+
+static void hwsp_free(struct i915_timeline *timeline)
+{
+	struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
+	struct i915_timeline_hwsp *hwsp;
+
+	hwsp = i915_timeline_hwsp(timeline);
+	if (!hwsp) /* leave global HWSP alone! */
+		return;
+
+	spin_lock(&gt->hwsp_lock);
+
+	/* As a cacheline becomes available, publish the HWSP on the freelist */
+	if (!hwsp->free_bitmap)
+		list_add_tail(&hwsp->free_link, &gt->hwsp_free_list);
+
+	hwsp->free_bitmap |= BIT_ULL(timeline->hwsp_offset / CACHELINE_BYTES);
+
+	/* And if no one is left using it, give the page back to the system */
+	if (hwsp->free_bitmap == ~0ull) {
+		i915_vma_put(hwsp->vma);
+		list_del(&hwsp->free_link);
+		kfree(hwsp);
+	}
+
+	spin_unlock(&gt->hwsp_lock);
 }
 
 int i915_timeline_init(struct drm_i915_private *i915,
 		       struct i915_timeline *timeline,
 		       const char *name,
-		       struct i915_vma *global_hwsp)
+		       struct i915_vma *hwsp)
 {
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
 	void *vaddr;
-	int err;
 
 	/*
 	 * Ideally we want a set of engines on a single leaf as we expect
@@ -64,18 +133,18 @@ int i915_timeline_init(struct drm_i915_private *i915,
 	timeline->name = name;
 	timeline->pin_count = 0;
 
-	if (global_hwsp) {
-		timeline->hwsp_ggtt = i915_vma_get(global_hwsp);
-		timeline->hwsp_offset = I915_GEM_HWS_SEQNO_ADDR;
-	} else {
-		err = hwsp_alloc(timeline);
-		if (err)
-			return err;
+	timeline->hwsp_offset = I915_GEM_HWS_SEQNO_ADDR;
+	if (!hwsp) {
+		hwsp = hwsp_alloc(timeline, &timeline->hwsp_offset);
+		if (IS_ERR(hwsp))
+			return PTR_ERR(hwsp);
 	}
+	timeline->hwsp_ggtt = i915_vma_get(hwsp);
 
-	vaddr = i915_gem_object_pin_map(timeline->hwsp_ggtt->obj, I915_MAP_WB);
+	vaddr = i915_gem_object_pin_map(hwsp->obj, I915_MAP_WB);
 	if (IS_ERR(vaddr)) {
-		i915_vma_put(timeline->hwsp_ggtt);
+		hwsp_free(timeline);
+		i915_vma_put(hwsp);
 		return PTR_ERR(vaddr);
 	}
 
@@ -106,6 +175,9 @@ void i915_timelines_init(struct drm_i915_private *i915)
 	mutex_init(&gt->mutex);
 	INIT_LIST_HEAD(&gt->list);
 
+	spin_lock_init(&gt->hwsp_lock);
+	INIT_LIST_HEAD(&gt->hwsp_free_list);
+
 	/* via i915_gem_wait_for_idle() */
 	i915_gem_shrinker_taints_mutex(i915, &gt->mutex);
 }
@@ -146,12 +218,13 @@ void i915_timeline_fini(struct i915_timeline *timeline)
 	GEM_BUG_ON(!list_empty(&timeline->requests));
 	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
-	i915_syncmap_free(&timeline->sync);
-
 	mutex_lock(&gt->mutex);
 	list_del(&timeline->link);
 	mutex_unlock(&gt->mutex);
 
+	i915_syncmap_free(&timeline->sync);
+	hwsp_free(timeline);
+
 	i915_gem_object_unpin_map(timeline->hwsp_ggtt->obj);
 	i915_vma_put(timeline->hwsp_ggtt);
 }
@@ -247,6 +320,7 @@ void i915_timelines_fini(struct drm_i915_private *i915)
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
 
 	GEM_BUG_ON(!list_empty(&gt->list));
+	GEM_BUG_ON(!list_empty(&gt->hwsp_free_list));
 
 	mutex_destroy(&gt->mutex);
 }
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index 8e29182a2584..42ef4830dd95 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -33,6 +33,7 @@
 #include "i915_utils.h"
 
 struct i915_vma;
+struct i915_timeline_hwsp;
 
 struct i915_timeline {
 	u64 fence_context;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 5793abe509a2..46eb818ed309 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -226,6 +226,18 @@ static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
 	return i915_vm_to_ggtt(vma->vm)->pin_bias;
 }
 
+/* XXX inline spaghetti */
+static inline struct i915_timeline_hwsp *
+i915_timeline_hwsp(const struct i915_timeline *tl)
+{
+	return tl->hwsp_ggtt->private;
+}
+
+static inline bool i915_timeline_is_global(const struct i915_timeline *tl)
+{
+	return !i915_timeline_hwsp(tl);
+}
+
 static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
 {
 	i915_gem_object_get(vma->obj);
diff --git a/drivers/gpu/drm/i915/selftests/i915_random.c b/drivers/gpu/drm/i915/selftests/i915_random.c
index 1f415ce47018..716a3f19f030 100644
--- a/drivers/gpu/drm/i915/selftests/i915_random.c
+++ b/drivers/gpu/drm/i915/selftests/i915_random.c
@@ -41,18 +41,37 @@ u64 i915_prandom_u64_state(struct rnd_state *rnd)
 	return x;
 }
 
-void i915_random_reorder(unsigned int *order, unsigned int count,
-			 struct rnd_state *state)
+void i915_prandom_shuffle(void *arr, size_t elsz, size_t count,
+			  struct rnd_state *state)
 {
-	unsigned int i, j;
+	char stack[128];
+
+	if (WARN_ON(elsz > sizeof(stack) || count > U32_MAX))
+		return;
+
+	if (!elsz || !count)
+		return;
+
+	/* Fisher-Yates shuffle courtesy of Knuth */
+	while (--count) {
+		size_t swp;
+
+		swp = i915_prandom_u32_max_state(count + 1, state);
+		if (swp == count)
+			continue;
 
-	for (i = 0; i < count; i++) {
-		BUILD_BUG_ON(sizeof(unsigned int) > sizeof(u32));
-		j = i915_prandom_u32_max_state(count, state);
-		swap(order[i], order[j]);
+		memcpy(stack, arr + count * elsz, elsz);
+		memcpy(arr + count * elsz, arr + swp * elsz, elsz);
+		memcpy(arr + swp * elsz, stack, elsz);
 	}
 }
 
+void i915_random_reorder(unsigned int *order, unsigned int count,
+			 struct rnd_state *state)
+{
+	i915_prandom_shuffle(order, sizeof(*order), count, state);
+}
+
 unsigned int *i915_random_order(unsigned int count, struct rnd_state *state)
 {
 	unsigned int *order, i;
diff --git a/drivers/gpu/drm/i915/selftests/i915_random.h b/drivers/gpu/drm/i915/selftests/i915_random.h
index 7dffedc501ca..8e1ff9c105b6 100644
--- a/drivers/gpu/drm/i915/selftests/i915_random.h
+++ b/drivers/gpu/drm/i915/selftests/i915_random.h
@@ -54,4 +54,7 @@ void i915_random_reorder(unsigned int *order,
 			 unsigned int count,
 			 struct rnd_state *state);
 
+void i915_prandom_shuffle(void *arr, size_t elsz, size_t count,
+			  struct rnd_state *state);
+
 #endif /* !__I915_SELFTESTS_RANDOM_H__ */
diff --git a/drivers/gpu/drm/i915/selftests/i915_timeline.c b/drivers/gpu/drm/i915/selftests/i915_timeline.c
index 1585b614510d..1cecc71fba74 100644
--- a/drivers/gpu/drm/i915/selftests/i915_timeline.c
+++ b/drivers/gpu/drm/i915/selftests/i915_timeline.c
@@ -4,6 +4,8 @@
  * Copyright © 2017-2018 Intel Corporation
  */
 
+#include <linux/prime_numbers.h>
+
 #include "../i915_selftest.h"
 #include "i915_random.h"
 
@@ -11,6 +13,143 @@
 #include "mock_gem_device.h"
 #include "mock_timeline.h"
 
+static struct page *hwsp_page(struct i915_timeline *tl)
+{
+	struct drm_i915_gem_object *obj = tl->hwsp_ggtt->obj;
+
+	GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
+	return sg_page(obj->mm.pages->sgl);
+}
+
+static unsigned long hwsp_cacheline(struct i915_timeline *tl)
+{
+	unsigned long address = (unsigned long)page_address(hwsp_page(tl));
+
+	return (address + tl->hwsp_offset) / CACHELINE_BYTES;
+}
+
+#define CACHELINES_PER_PAGE (PAGE_SIZE / CACHELINE_BYTES)
+
+struct mock_hwsp_freelist {
+	struct drm_i915_private *i915;
+	struct radix_tree_root cachelines;
+	struct i915_timeline **history;
+	unsigned long count, max;
+	struct rnd_state prng;
+};
+
+enum {
+	SHUFFLE = BIT(0),
+};
+
+static void __mock_hwsp_record(struct mock_hwsp_freelist *state,
+			       unsigned int idx,
+			       struct i915_timeline *tl)
+{
+	tl = xchg(&state->history[idx], tl);
+	if (tl) {
+		radix_tree_delete(&state->cachelines, hwsp_cacheline(tl));
+		i915_timeline_put(tl);
+	}
+}
+
+static int __mock_hwsp_timeline(struct mock_hwsp_freelist *state,
+				unsigned int count,
+				unsigned int flags)
+{
+	struct i915_timeline *tl;
+	unsigned int idx;
+
+	while (count--) {
+		unsigned long cacheline;
+		int err;
+
+		tl = i915_timeline_create(state->i915, "mock", NULL);
+		if (IS_ERR(tl))
+			return PTR_ERR(tl);
+
+		cacheline = hwsp_cacheline(tl);
+		err = radix_tree_insert(&state->cachelines, cacheline, tl);
+		if (err) {
+			if (err == -EEXIST) {
+				pr_err("HWSP cacheline %lu already used; duplicate allocation!\n",
+				       cacheline);
+			}
+			i915_timeline_put(tl);
+			return err;
+		}
+
+		idx = state->count++ % state->max;
+		__mock_hwsp_record(state, idx, tl);
+	}
+
+	if (flags & SHUFFLE)
+		i915_prandom_shuffle(state->history,
+				     sizeof(*state->history),
+				     min(state->count, state->max),
+				     &state->prng);
+
+	count = i915_prandom_u32_max_state(min(state->count, state->max),
+					   &state->prng);
+	while (count--) {
+		idx = --state->count % state->max;
+		__mock_hwsp_record(state, idx, NULL);
+	}
+
+	return 0;
+}
+
+static int mock_hwsp_freelist(void *arg)
+{
+	struct mock_hwsp_freelist state;
+	const struct {
+		const char *name;
+		unsigned int flags;
+	} phases[] = {
+		{ "linear", 0 },
+		{ "shuffled", SHUFFLE },
+		{ },
+	}, *p;
+	unsigned int na;
+	int err = 0;
+
+	INIT_RADIX_TREE(&state.cachelines, GFP_KERNEL);
+	state.prng = I915_RND_STATE_INITIALIZER(i915_selftest.random_seed);
+
+	state.i915 = mock_gem_device();
+	if (!state.i915)
+		return -ENOMEM;
+
+	/*
+	 * Create a bunch of timelines and check that their HWSP do not overlap.
+	 * Free some, and try again.
+	 */
+
+	state.max = PAGE_SIZE / sizeof(*state.history);
+	state.count = 0;
+	state.history = kcalloc(state.max, sizeof(*state.history), GFP_KERNEL);
+	if (!state.history) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	for (p = phases; p->name; p++) {
+		pr_debug("%s(%s)\n", __func__, p->name);
+		for_each_prime_number_from(na, 1, 2 * CACHELINES_PER_PAGE) {
+			err = __mock_hwsp_timeline(&state, na, p->flags);
+			if (err)
+				goto out;
+		}
+	}
+
+out:
+	for (na = 0; na < state.max; na++)
+		__mock_hwsp_record(&state, na, NULL);
+	kfree(state.history);
+	drm_dev_put(&state.i915->drm);
+	return err;
+}
+
 struct __igt_sync {
 	const char *name;
 	u32 seqno;
@@ -260,6 +399,7 @@ static int bench_sync(void *arg)
 int i915_timeline_mock_selftests(void)
 {
 	static const struct i915_subtest tests[] = {
+		SUBTEST(mock_hwsp_freelist),
 		SUBTEST(igt_sync),
 		SUBTEST(bench_sync),
 	};
-- 
2.20.1

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

[PATCH 23/33] drm/i915: Track the context's seqno in its own timeline HWSP

[Chris Wilson]

Now that we have allocated ourselves a cacheline to store a breadcrumb,
we can emit a write from the GPU into the timeline's HWSP of the
per-context seqno as we complete each request. This drops the mirroring
of the per-engine HWSP and allows each context to operate independently.
We do not need to unwind the per-context timeline, and so requests are
always consistent with the timeline breadcrumb, greatly simplifying the
completion checks as we no longer need to be concerned about the
global_seqno changing mid check.

One complication though is that we have to be wary that the request may
outlive the HWSP and so avoid touching the potentially danging pointer
after we have retired the fence. We also have to guard our access of the
HWSP with RCU, the release of the obj->mm.pages should already be RCU-safe.

At this point, we are emitting both per-context and global seqno and
still using the single per-engine execution timeline for resolving
interrupts.

v2: s/fake_complete/mark_complete/

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c              |  2 +-
 drivers/gpu/drm/i915/i915_request.c          |  3 +-
 drivers/gpu/drm/i915/i915_request.h          | 30 +++-----
 drivers/gpu/drm/i915/i915_reset.c            |  1 +
 drivers/gpu/drm/i915/i915_vma.h              |  6 ++
 drivers/gpu/drm/i915/intel_engine_cs.c       | 15 +++-
 drivers/gpu/drm/i915/intel_lrc.c             | 31 +++++---
 drivers/gpu/drm/i915/intel_ringbuffer.c      | 76 +++++++++++++++-----
 drivers/gpu/drm/i915/selftests/mock_engine.c | 19 ++++-
 9 files changed, 131 insertions(+), 52 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index d68f3fdd8a8e..1bd724d663d9 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2892,7 +2892,7 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
 	 */
 	spin_lock_irqsave(&engine->timeline.lock, flags);
 	list_for_each_entry(request, &engine->timeline.requests, link) {
-		if (__i915_request_completed(request, request->global_seqno))
+		if (i915_request_completed(request))
 			continue;
 
 		active = request;
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 79eb1957cf99..cdbdbcff28ec 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -199,6 +199,7 @@ static void __retire_engine_request(struct intel_engine_cs *engine,
 	spin_unlock(&engine->timeline.lock);
 
 	spin_lock(&rq->lock);
+	i915_request_mark_complete(rq);
 	if (!i915_request_signaled(rq))
 		dma_fence_signal_locked(&rq->fence);
 	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &rq->fence.flags))
@@ -634,7 +635,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	rq->ring = ce->ring;
 	rq->timeline = ce->ring->timeline;
 	GEM_BUG_ON(rq->timeline == &engine->timeline);
-	rq->hwsp_seqno = &engine->status_page.addr[I915_GEM_HWS_INDEX];
+	rq->hwsp_seqno = rq->timeline->hwsp_seqno;
 
 	spin_lock_init(&rq->lock);
 	dma_fence_init(&rq->fence,
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index ade010fe6e26..96c586d6ff4d 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -289,6 +289,7 @@ long i915_request_wait(struct i915_request *rq,
 
 static inline bool i915_request_signaled(const struct i915_request *rq)
 {
+	/* The request may live longer than its HWSP, so check flags first! */
 	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &rq->fence.flags);
 }
 
@@ -340,32 +341,23 @@ static inline u32 hwsp_seqno(const struct i915_request *rq)
  */
 static inline bool i915_request_started(const struct i915_request *rq)
 {
-	u32 seqno;
-
-	seqno = i915_request_global_seqno(rq);
-	if (!seqno) /* not yet submitted to HW */
-		return false;
+	if (i915_request_signaled(rq))
+		return true;
 
-	return i915_seqno_passed(hwsp_seqno(rq), seqno - 1);
-}
-
-static inline bool
-__i915_request_completed(const struct i915_request *rq, u32 seqno)
-{
-	GEM_BUG_ON(!seqno);
-	return i915_seqno_passed(hwsp_seqno(rq), seqno) &&
-		seqno == i915_request_global_seqno(rq);
+	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
 }
 
 static inline bool i915_request_completed(const struct i915_request *rq)
 {
-	u32 seqno;
+	if (i915_request_signaled(rq))
+		return true;
 
-	seqno = i915_request_global_seqno(rq);
-	if (!seqno)
-		return false;
+	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno);
+}
 
-	return __i915_request_completed(rq, seqno);
+static inline void i915_request_mark_complete(struct i915_request *rq)
+{
+	rq->hwsp_seqno = (u32 *)&rq->fence.seqno; /* decouple from HWSP */
 }
 
 void i915_retire_requests(struct drm_i915_private *i915);
diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index c57b76aebd8f..e88691da60c0 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -760,6 +760,7 @@ static void nop_submit_request(struct i915_request *request)
 
 	spin_lock_irqsave(&request->engine->timeline.lock, flags);
 	__i915_request_submit(request);
+	i915_request_mark_complete(request);
 	intel_engine_write_global_seqno(request->engine, request->global_seqno);
 	spin_unlock_irqrestore(&request->engine->timeline.lock, flags);
 }
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 46eb818ed309..b0f6b1d904a5 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -227,6 +227,12 @@ static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
 }
 
 /* XXX inline spaghetti */
+static inline u32 i915_timeline_seqno_address(const struct i915_timeline *tl)
+{
+	GEM_BUG_ON(!tl->pin_count);
+	return i915_ggtt_offset(tl->hwsp_ggtt) + tl->hwsp_offset;
+}
+
 static inline struct i915_timeline_hwsp *
 i915_timeline_hwsp(const struct i915_timeline *tl)
 {
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 28a66242d698..8ca062815ff9 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -660,10 +660,16 @@ static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 	frame->rq.ring = &frame->ring;
 	frame->rq.timeline = &frame->timeline;
 
+	dw = i915_timeline_pin(&frame->timeline);
+	if (dw < 0)
+		goto out_timeline;
+
 	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
 
-	i915_timeline_fini(&frame->timeline);
+	i915_timeline_unpin(&frame->timeline);
 
+out_timeline:
+	i915_timeline_fini(&frame->timeline);
 out_frame:
 	kfree(frame);
 	return dw;
@@ -1426,9 +1432,10 @@ static void intel_engine_print_registers(const struct intel_engine_cs *engine,
 				char hdr[80];
 
 				snprintf(hdr, sizeof(hdr),
-					 "\t\tELSP[%d] count=%d, ring->start=%08x, rq: ",
+					 "\t\tELSP[%d] count=%d, ring: {start:%08x, hwsp:%08x}, rq: ",
 					 idx, count,
-					 i915_ggtt_offset(rq->ring->vma));
+					 i915_ggtt_offset(rq->ring->vma),
+					 i915_timeline_seqno_address(rq->timeline));
 				print_request(m, rq, hdr);
 			} else {
 				drm_printf(m, "\t\tELSP[%d] idle\n", idx);
@@ -1538,6 +1545,8 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 			   rq->ring->emit);
 		drm_printf(m, "\t\tring->space:  0x%08x\n",
 			   rq->ring->space);
+		drm_printf(m, "\t\tring->hwsp:   0x%08x\n",
+			   i915_timeline_seqno_address(rq->timeline));
 
 		print_request_ring(m, rq);
 	}
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index f4c52b9bdfca..a75918d50788 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -938,10 +938,10 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine)
 	list_for_each_entry(rq, &engine->timeline.requests, link) {
 		GEM_BUG_ON(!rq->global_seqno);
 
-		if (i915_request_signaled(rq))
-			continue;
+		if (!i915_request_signaled(rq))
+			dma_fence_set_error(&rq->fence, -EIO);
 
-		dma_fence_set_error(&rq->fence, -EIO);
+		i915_request_mark_complete(rq);
 	}
 
 	/* Flush the queued requests to the timeline list (for retiring). */
@@ -951,9 +951,9 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine)
 
 		priolist_for_each_request_consume(rq, rn, p, i) {
 			list_del_init(&rq->sched.link);
-
-			dma_fence_set_error(&rq->fence, -EIO);
 			__i915_request_submit(rq);
+			dma_fence_set_error(&rq->fence, -EIO);
+			i915_request_mark_complete(rq);
 		}
 
 		rb_erase_cached(&p->node, &execlists->queue);
@@ -2152,10 +2152,17 @@ static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 	/* w/a: bit 5 needs to be zero for MI_FLUSH_DW address. */
 	BUILD_BUG_ON(I915_GEM_HWS_INDEX_ADDR & (1 << 5));
 
-	cs = gen8_emit_ggtt_write(cs, request->global_seqno,
+	cs = gen8_emit_ggtt_write(cs,
+				  request->fence.seqno,
+				  i915_timeline_seqno_address(request->timeline));
+
+	cs = gen8_emit_ggtt_write(cs,
+				  request->global_seqno,
 				  intel_hws_seqno_address(request->engine));
+
 	*cs++ = MI_USER_INTERRUPT;
 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+
 	request->tail = intel_ring_offset(request, cs);
 	assert_ring_tail_valid(request->ring, request->tail);
 
@@ -2164,18 +2171,20 @@ static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 
 static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
-	/* We're using qword write, seqno should be aligned to 8 bytes. */
-	BUILD_BUG_ON(I915_GEM_HWS_INDEX & 1);
-
 	cs = gen8_emit_ggtt_write_rcs(cs,
-				      request->global_seqno,
-				      intel_hws_seqno_address(request->engine),
+				      request->fence.seqno,
+				      i915_timeline_seqno_address(request->timeline),
 				      PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
 				      PIPE_CONTROL_DEPTH_CACHE_FLUSH |
 				      PIPE_CONTROL_DC_FLUSH_ENABLE |
 				      PIPE_CONTROL_FLUSH_ENABLE |
 				      PIPE_CONTROL_CS_STALL);
 
+	cs = gen8_emit_ggtt_write_rcs(cs,
+				      request->global_seqno,
+				      intel_hws_seqno_address(request->engine),
+				      PIPE_CONTROL_CS_STALL);
+
 	*cs++ = MI_USER_INTERRUPT;
 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 174795622eb1..71a51a129c98 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -326,6 +326,12 @@ static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 		 PIPE_CONTROL_DC_FLUSH_ENABLE |
 		 PIPE_CONTROL_QW_WRITE |
 		 PIPE_CONTROL_CS_STALL);
+	*cs++ = i915_timeline_seqno_address(rq->timeline) |
+		PIPE_CONTROL_GLOBAL_GTT;
+	*cs++ = rq->fence.seqno;
+
+	*cs++ = GFX_OP_PIPE_CONTROL(4);
+	*cs++ = PIPE_CONTROL_QW_WRITE | PIPE_CONTROL_CS_STALL;
 	*cs++ = intel_hws_seqno_address(rq->engine) | PIPE_CONTROL_GLOBAL_GTT;
 	*cs++ = rq->global_seqno;
 
@@ -427,6 +433,13 @@ static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 		 PIPE_CONTROL_QW_WRITE |
 		 PIPE_CONTROL_GLOBAL_GTT_IVB |
 		 PIPE_CONTROL_CS_STALL);
+	*cs++ = i915_timeline_seqno_address(rq->timeline);
+	*cs++ = rq->fence.seqno;
+
+	*cs++ = GFX_OP_PIPE_CONTROL(4);
+	*cs++ = (PIPE_CONTROL_QW_WRITE |
+		 PIPE_CONTROL_GLOBAL_GTT_IVB |
+		 PIPE_CONTROL_CS_STALL);
 	*cs++ = intel_hws_seqno_address(rq->engine);
 	*cs++ = rq->global_seqno;
 
@@ -441,10 +454,16 @@ static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
-	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW;
-	*cs++ = intel_hws_seqno_address(rq->engine) | MI_FLUSH_DW_USE_GTT;
+	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
+	*cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT;
+	*cs++ = rq->fence.seqno;
+
+	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
+	*cs++ = I915_GEM_HWS_INDEX_ADDR | MI_FLUSH_DW_USE_GTT;
 	*cs++ = rq->global_seqno;
+
 	*cs++ = MI_USER_INTERRUPT;
+	*cs++ = MI_NOOP;
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
@@ -457,8 +476,12 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	int i;
 
-	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW;
-	*cs++ = intel_hws_seqno_address(rq->engine) | MI_FLUSH_DW_USE_GTT;
+	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
+	*cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT;
+	*cs++ = rq->fence.seqno;
+
+	*cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
+	*cs++ = I915_GEM_HWS_INDEX_ADDR | MI_FLUSH_DW_USE_GTT;
 	*cs++ = rq->global_seqno;
 
 	for (i = 0; i < GEN7_XCS_WA; i++) {
@@ -472,7 +495,6 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 	*cs++ = 0;
 
 	*cs++ = MI_USER_INTERRUPT;
-	*cs++ = MI_NOOP;
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
@@ -738,7 +760,7 @@ static void reset_ring(struct intel_engine_cs *engine, bool stalled)
 	rq = NULL;
 	spin_lock_irqsave(&tl->lock, flags);
 	list_for_each_entry(pos, &tl->requests, link) {
-		if (!__i915_request_completed(pos, pos->global_seqno)) {
+		if (!i915_request_completed(pos)) {
 			rq = pos;
 			break;
 		}
@@ -880,10 +902,10 @@ static void cancel_requests(struct intel_engine_cs *engine)
 	list_for_each_entry(request, &engine->timeline.requests, link) {
 		GEM_BUG_ON(!request->global_seqno);
 
-		if (i915_request_signaled(request))
-			continue;
+		if (!i915_request_signaled(request))
+			dma_fence_set_error(&request->fence, -EIO);
 
-		dma_fence_set_error(&request->fence, -EIO);
+		i915_request_mark_complete(request);
 	}
 
 	intel_write_status_page(engine,
@@ -907,14 +929,19 @@ static void i9xx_submit_request(struct i915_request *request)
 
 static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
+	GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma);
+
 	*cs++ = MI_FLUSH;
 
+	*cs++ = MI_STORE_DWORD_INDEX;
+	*cs++ = I915_GEM_HWS_SEQNO_ADDR;
+	*cs++ = rq->fence.seqno;
+
 	*cs++ = MI_STORE_DWORD_INDEX;
 	*cs++ = I915_GEM_HWS_INDEX_ADDR;
 	*cs++ = rq->global_seqno;
 
 	*cs++ = MI_USER_INTERRUPT;
-	*cs++ = MI_NOOP;
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
@@ -927,8 +954,14 @@ static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 {
 	int i;
 
+	GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma);
+
 	*cs++ = MI_FLUSH;
 
+	*cs++ = MI_STORE_DWORD_INDEX;
+	*cs++ = I915_GEM_HWS_SEQNO_ADDR;
+	*cs++ = rq->fence.seqno;
+
 	BUILD_BUG_ON(GEN5_WA_STORES < 1);
 	for (i = 0; i < GEN5_WA_STORES; i++) {
 		*cs++ = MI_STORE_DWORD_INDEX;
@@ -937,6 +970,7 @@ static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 	}
 
 	*cs++ = MI_USER_INTERRUPT;
+	*cs++ = MI_NOOP;
 
 	rq->tail = intel_ring_offset(rq, cs);
 	assert_ring_tail_valid(rq->ring, rq->tail);
@@ -1169,6 +1203,10 @@ int intel_ring_pin(struct intel_ring *ring)
 
 	GEM_BUG_ON(ring->vaddr);
 
+	ret = i915_timeline_pin(ring->timeline);
+	if (ret)
+		return ret;
+
 	flags = PIN_GLOBAL;
 
 	/* Ring wraparound at offset 0 sometimes hangs. No idea why. */
@@ -1185,28 +1223,32 @@ int intel_ring_pin(struct intel_ring *ring)
 		else
 			ret = i915_gem_object_set_to_cpu_domain(vma->obj, true);
 		if (unlikely(ret))
-			return ret;
+			goto unpin_timeline;
 	}
 
 	ret = i915_vma_pin(vma, 0, 0, flags);
 	if (unlikely(ret))
-		return ret;
+		goto unpin_timeline;
 
 	if (i915_vma_is_map_and_fenceable(vma))
 		addr = (void __force *)i915_vma_pin_iomap(vma);
 	else
 		addr = i915_gem_object_pin_map(vma->obj, map);
-	if (IS_ERR(addr))
-		goto err;
+	if (IS_ERR(addr)) {
+		ret = PTR_ERR(addr);
+		goto unpin_ring;
+	}
 
 	vma->obj->pin_global++;
 
 	ring->vaddr = addr;
 	return 0;
 
-err:
+unpin_ring:
 	i915_vma_unpin(vma);
-	return PTR_ERR(addr);
+unpin_timeline:
+	i915_timeline_unpin(ring->timeline);
+	return ret;
 }
 
 void intel_ring_reset(struct intel_ring *ring, u32 tail)
@@ -1235,6 +1277,8 @@ void intel_ring_unpin(struct intel_ring *ring)
 
 	ring->vma->obj->pin_global--;
 	i915_vma_unpin(ring->vma);
+
+	i915_timeline_unpin(ring->timeline);
 }
 
 static struct i915_vma *
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 919c89fd6ee5..95e890d7f58b 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -30,6 +30,17 @@ struct mock_ring {
 	struct i915_timeline timeline;
 };
 
+static void mock_timeline_pin(struct i915_timeline *tl)
+{
+	tl->pin_count++;
+}
+
+static void mock_timeline_unpin(struct i915_timeline *tl)
+{
+	GEM_BUG_ON(!tl->pin_count);
+	tl->pin_count--;
+}
+
 static struct intel_ring *mock_ring(struct intel_engine_cs *engine)
 {
 	const unsigned long sz = PAGE_SIZE / 2;
@@ -76,6 +87,8 @@ static void advance(struct mock_request *request)
 {
 	list_del_init(&request->link);
 	mock_seqno_advance(request->base.engine, request->base.global_seqno);
+	i915_request_mark_complete(&request->base);
+	GEM_BUG_ON(!i915_request_completed(&request->base));
 }
 
 static void hw_delay_complete(struct timer_list *t)
@@ -108,6 +121,7 @@ static void hw_delay_complete(struct timer_list *t)
 
 static void mock_context_unpin(struct intel_context *ce)
 {
+	mock_timeline_unpin(ce->ring->timeline);
 	i915_gem_context_put(ce->gem_context);
 }
 
@@ -129,6 +143,7 @@ mock_context_pin(struct intel_engine_cs *engine,
 		 struct i915_gem_context *ctx)
 {
 	struct intel_context *ce = to_intel_context(ctx, engine);
+	int err = -ENOMEM;
 
 	if (ce->pin_count++)
 		return ce;
@@ -139,13 +154,15 @@ mock_context_pin(struct intel_engine_cs *engine,
 			goto err;
 	}
 
+	mock_timeline_pin(ce->ring->timeline);
+
 	ce->ops = &mock_context_ops;
 	i915_gem_context_get(ctx);
 	return ce;
 
 err:
 	ce->pin_count = 0;
-	return ERR_PTR(-ENOMEM);
+	return ERR_PTR(err);
 }
 
 static int mock_request_alloc(struct i915_request *request)
-- 
2.20.1

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

[PATCH 24/33] drm/i915: Track active timelines

[Chris Wilson]

Now that we pin timelines around use, we have a clearly defined lifetime
and convenient points at which we can track only the active timelines.
This allows us to reduce the list iteration to only consider those
active timelines and not all.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h      |  2 +-
 drivers/gpu/drm/i915/i915_gem.c      |  4 +--
 drivers/gpu/drm/i915/i915_reset.c    |  2 +-
 drivers/gpu/drm/i915/i915_timeline.c | 39 ++++++++++++++++++----------
 4 files changed, 29 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 6a051381f535..d072f3369ee1 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1977,7 +1977,7 @@ struct drm_i915_private {
 
 		struct i915_gt_timelines {
 			struct mutex mutex; /* protects list, tainted by GPU */
-			struct list_head list;
+			struct list_head active_list;
 
 			/* Pack multiple timelines' seqnos into the same page */
 			spinlock_t hwsp_lock;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 1bd724d663d9..05627000b77d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3248,7 +3248,7 @@ wait_for_timelines(struct drm_i915_private *i915,
 		return timeout;
 
 	mutex_lock(&gt->mutex);
-	list_for_each_entry(tl, &gt->list, link) {
+	list_for_each_entry(tl, &gt->active_list, link) {
 		struct i915_request *rq;
 
 		rq = i915_gem_active_get_unlocked(&tl->last_request);
@@ -3276,7 +3276,7 @@ wait_for_timelines(struct drm_i915_private *i915,
 
 		/* restart after reacquiring the lock */
 		mutex_lock(&gt->mutex);
-		tl = list_entry(&gt->list, typeof(*tl), link);
+		tl = list_entry(&gt->active_list, typeof(*tl), link);
 	}
 	mutex_unlock(&gt->mutex);
 
diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index e88691da60c0..c3861a0219a9 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -856,7 +856,7 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	 * No more can be submitted until we reset the wedged bit.
 	 */
 	mutex_lock(&i915->gt.timelines.mutex);
-	list_for_each_entry(tl, &i915->gt.timelines.list, link) {
+	list_for_each_entry(tl, &i915->gt.timelines.active_list, link) {
 		struct i915_request *rq;
 		long timeout;
 
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 95c689826c8c..548c99bbf126 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -116,7 +116,6 @@ int i915_timeline_init(struct drm_i915_private *i915,
 		       const char *name,
 		       struct i915_vma *hwsp)
 {
-	struct i915_gt_timelines *gt = &i915->gt.timelines;
 	void *vaddr;
 
 	/*
@@ -161,10 +160,6 @@ int i915_timeline_init(struct drm_i915_private *i915,
 
 	i915_syncmap_init(&timeline->sync);
 
-	mutex_lock(&gt->mutex);
-	list_add(&timeline->link, &gt->list);
-	mutex_unlock(&gt->mutex);
-
 	return 0;
 }
 
@@ -173,7 +168,7 @@ void i915_timelines_init(struct drm_i915_private *i915)
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
 
 	mutex_init(&gt->mutex);
-	INIT_LIST_HEAD(&gt->list);
+	INIT_LIST_HEAD(&gt->active_list);
 
 	spin_lock_init(&gt->hwsp_lock);
 	INIT_LIST_HEAD(&gt->hwsp_free_list);
@@ -182,6 +177,24 @@ void i915_timelines_init(struct drm_i915_private *i915)
 	i915_gem_shrinker_taints_mutex(i915, &gt->mutex);
 }
 
+static void timeline_add_to_active(struct i915_timeline *tl)
+{
+	struct i915_gt_timelines *gt = &tl->i915->gt.timelines;
+
+	mutex_lock(&gt->mutex);
+	list_add(&tl->link, &gt->active_list);
+	mutex_unlock(&gt->mutex);
+}
+
+static void timeline_remove_from_active(struct i915_timeline *tl)
+{
+	struct i915_gt_timelines *gt = &tl->i915->gt.timelines;
+
+	mutex_lock(&gt->mutex);
+	list_del(&tl->link);
+	mutex_unlock(&gt->mutex);
+}
+
 /**
  * i915_timelines_park - called when the driver idles
  * @i915: the drm_i915_private device
@@ -198,7 +211,7 @@ void i915_timelines_park(struct drm_i915_private *i915)
 	struct i915_timeline *timeline;
 
 	mutex_lock(&gt->mutex);
-	list_for_each_entry(timeline, &gt->list, link) {
+	list_for_each_entry(timeline, &gt->active_list, link) {
 		/*
 		 * All known fences are completed so we can scrap
 		 * the current sync point tracking and start afresh,
@@ -212,16 +225,10 @@ void i915_timelines_park(struct drm_i915_private *i915)
 
 void i915_timeline_fini(struct i915_timeline *timeline)
 {
-	struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
-
 	GEM_BUG_ON(timeline->pin_count);
 	GEM_BUG_ON(!list_empty(&timeline->requests));
 	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
-	mutex_lock(&gt->mutex);
-	list_del(&timeline->link);
-	mutex_unlock(&gt->mutex);
-
 	i915_syncmap_free(&timeline->sync);
 	hwsp_free(timeline);
 
@@ -283,6 +290,8 @@ int i915_timeline_pin(struct i915_timeline *tl)
 	if (err)
 		goto unpin;
 
+	timeline_add_to_active(tl);
+
 	return 0;
 
 unpin:
@@ -296,6 +305,8 @@ void i915_timeline_unpin(struct i915_timeline *tl)
 	if (--tl->pin_count)
 		return;
 
+	timeline_remove_from_active(tl);
+
 	/*
 	 * Since this timeline is idle, all bariers upon which we were waiting
 	 * must also be complete and so we can discard the last used barriers
@@ -319,7 +330,7 @@ void i915_timelines_fini(struct drm_i915_private *i915)
 {
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
 
-	GEM_BUG_ON(!list_empty(&gt->list));
+	GEM_BUG_ON(!list_empty(&gt->active_list));
 	GEM_BUG_ON(!list_empty(&gt->hwsp_free_list));
 
 	mutex_destroy(&gt->mutex);
-- 
2.20.1

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

[PATCH 25/33] drm/i915: Identify active requests

[Chris Wilson]

To allow requests to forgo a common execution timeline, one question we
need to be able to answer is "is this request running?". To track
whether a request has started on HW, we can emit a breadcrumb at the
beginning of the request and check its timeline's HWSP to see if the
breadcrumb has advanced past the start of this request. (This is in
contrast to the global timeline where we need only ask if we are on the
global timeline and if the timeline has advanced past the end of the
previous request.)

There is still confusion from a preempted request, which has already
started but relinquished the HW to a high priority request. For the
common case, this discrepancy should be negligible. However, for
identification of hung requests, knowing which one was running at the
time of the hang will be much more important.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c              |  3 ++
 drivers/gpu/drm/i915/i915_gem_execbuffer.c   | 12 ++++++
 drivers/gpu/drm/i915/i915_request.c          | 10 ++---
 drivers/gpu/drm/i915/i915_request.h          |  1 +
 drivers/gpu/drm/i915/i915_timeline.c         |  1 +
 drivers/gpu/drm/i915/i915_timeline.h         |  2 +
 drivers/gpu/drm/i915/intel_engine_cs.c       |  8 ++--
 drivers/gpu/drm/i915/intel_lrc.c             | 39 +++++++++++++++++---
 drivers/gpu/drm/i915/intel_ringbuffer.c      | 25 ++++++++-----
 drivers/gpu/drm/i915/intel_ringbuffer.h      |  7 +++-
 drivers/gpu/drm/i915/selftests/mock_engine.c |  2 +-
 11 files changed, 85 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 05627000b77d..101a0f644787 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2895,6 +2895,9 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
 		if (i915_request_completed(request))
 			continue;
 
+		if (!i915_request_started(request))
+			break;
+
 		active = request;
 		break;
 	}
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index f250109e1f66..8eedf7cac493 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1976,6 +1976,18 @@ static int eb_submit(struct i915_execbuffer *eb)
 			return err;
 	}
 
+	/*
+	 * After we completed waiting for other engines (using HW semaphores)
+	 * then we can signal that this request/batch is ready to run. This
+	 * allows us to determine if the batch is still waiting on the GPU
+	 * or actually running by checking the breadcrumb.
+	 */
+	if (eb->engine->emit_init_breadcrumb) {
+		err = eb->engine->emit_init_breadcrumb(eb->request);
+		if (err)
+			return err;
+	}
+
 	err = eb->engine->emit_bb_start(eb->request,
 					eb->batch->node.start +
 					eb->batch_start_offset,
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index cdbdbcff28ec..2171df2d3019 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -333,7 +333,7 @@ void i915_request_retire_upto(struct i915_request *rq)
 
 static u32 timeline_get_seqno(struct i915_timeline *tl)
 {
-	return ++tl->seqno;
+	return tl->seqno += 1 + tl->has_initial_breadcrumb;
 }
 
 static void move_to_timeline(struct i915_request *request,
@@ -382,8 +382,8 @@ void __i915_request_submit(struct i915_request *request)
 		intel_engine_enable_signaling(request, false);
 	spin_unlock(&request->lock);
 
-	engine->emit_breadcrumb(request,
-				request->ring->vaddr + request->postfix);
+	engine->emit_fini_breadcrumb(request,
+				     request->ring->vaddr + request->postfix);
 
 	/* Transfer from per-context onto the global per-engine timeline */
 	move_to_timeline(request, &engine->timeline);
@@ -670,7 +670,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	 * around inside i915_request_add() there is sufficient space at
 	 * the beginning of the ring as well.
 	 */
-	rq->reserved_space = 2 * engine->emit_breadcrumb_dw * sizeof(u32);
+	rq->reserved_space = 2 * engine->emit_fini_breadcrumb_dw * sizeof(u32);
 
 	/*
 	 * Record the position of the start of the request so that
@@ -925,7 +925,7 @@ void i915_request_add(struct i915_request *request)
 	 * GPU processing the request, we never over-estimate the
 	 * position of the ring's HEAD.
 	 */
-	cs = intel_ring_begin(request, engine->emit_breadcrumb_dw);
+	cs = intel_ring_begin(request, engine->emit_fini_breadcrumb_dw);
 	GEM_BUG_ON(IS_ERR(cs));
 	request->postfix = intel_ring_offset(request, cs);
 
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 96c586d6ff4d..340d6216791c 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -344,6 +344,7 @@ static inline bool i915_request_started(const struct i915_request *rq)
 	if (i915_request_signaled(rq))
 		return true;
 
+	/* Remember: started but may have since been preempted! */
 	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
 }
 
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 548c99bbf126..8f5c57304064 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -131,6 +131,7 @@ int i915_timeline_init(struct drm_i915_private *i915,
 	timeline->i915 = i915;
 	timeline->name = name;
 	timeline->pin_count = 0;
+	timeline->has_initial_breadcrumb = !hwsp;
 
 	timeline->hwsp_offset = I915_GEM_HWS_SEQNO_ADDR;
 	if (!hwsp) {
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index 42ef4830dd95..d167e04073c5 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -48,6 +48,8 @@ struct i915_timeline {
 	struct i915_vma *hwsp_ggtt;
 	u32 hwsp_offset;
 
+	bool has_initial_breadcrumb;
+
 	/**
 	 * List of breadcrumbs associated with GPU requests currently
 	 * outstanding.
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index 8ca062815ff9..b35ba4f03079 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -664,7 +664,7 @@ static int measure_breadcrumb_dw(struct intel_engine_cs *engine)
 	if (dw < 0)
 		goto out_timeline;
 
-	dw = engine->emit_breadcrumb(&frame->rq, frame->cs) - frame->cs;
+	dw = engine->emit_fini_breadcrumb(&frame->rq, frame->cs) - frame->cs;
 
 	i915_timeline_unpin(&frame->timeline);
 
@@ -725,7 +725,7 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 	if (ret < 0)
 		goto err_breadcrumbs;
 
-	engine->emit_breadcrumb_dw = ret;
+	engine->emit_fini_breadcrumb_dw = ret;
 
 	return 0;
 
@@ -1297,7 +1297,9 @@ static void print_request(struct drm_printer *m,
 	drm_printf(m, "%s%x%s [%llx:%llx]%s @ %dms: %s\n",
 		   prefix,
 		   rq->global_seqno,
-		   i915_request_completed(rq) ? "!" : "",
+		   i915_request_completed(rq) ? "!" :
+		   i915_request_started(rq) ? "*" :
+		   "",
 		   rq->fence.context, rq->fence.seqno,
 		   buf,
 		   jiffies_to_msecs(jiffies - rq->emitted_jiffies),
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index a75918d50788..cbf20273e9f8 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -731,7 +731,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		 * WaIdleLiteRestore:bdw,skl
 		 * Apply the wa NOOPs to prevent
 		 * ring:HEAD == rq:TAIL as we resubmit the
-		 * request. See gen8_emit_breadcrumb() for
+		 * request. See gen8_emit_fini_breadcrumb() for
 		 * where we prepare the padding after the
 		 * end of the request.
 		 */
@@ -1391,6 +1391,34 @@ execlists_context_pin(struct intel_engine_cs *engine,
 	return __execlists_context_pin(engine, ctx, ce);
 }
 
+static int gen8_emit_init_breadcrumb(struct i915_request *rq)
+{
+	u32 *cs;
+
+	GEM_BUG_ON(!rq->timeline->has_initial_breadcrumb);
+
+	cs = intel_ring_begin(rq, 6);
+	if (IS_ERR(cs))
+		return PTR_ERR(cs);
+
+	/*
+	 * Check if we have been preempted before we even get started.
+	 *
+	 * After this point i915_request_started() reports true, even if
+	 * we get preempted and so are no longer running.
+	 */
+	*cs++ = MI_ARB_CHECK;
+	*cs++ = MI_NOOP;
+
+	*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
+	*cs++ = i915_timeline_seqno_address(rq->timeline);
+	*cs++ = 0;
+	*cs++ = rq->fence.seqno - 1;
+
+	intel_ring_advance(rq, cs);
+	return 0;
+}
+
 static int emit_pdps(struct i915_request *rq)
 {
 	const struct intel_engine_cs * const engine = rq->engine;
@@ -2147,7 +2175,7 @@ static u32 *gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
 	return cs;
 }
 
-static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
+static u32 *gen8_emit_fini_breadcrumb(struct i915_request *request, u32 *cs)
 {
 	/* w/a: bit 5 needs to be zero for MI_FLUSH_DW address. */
 	BUILD_BUG_ON(I915_GEM_HWS_INDEX_ADDR & (1 << 5));
@@ -2169,7 +2197,7 @@ static u32 *gen8_emit_breadcrumb(struct i915_request *request, u32 *cs)
 	return gen8_emit_wa_tail(request, cs);
 }
 
-static u32 *gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs)
+static u32 *gen8_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
 	cs = gen8_emit_ggtt_write_rcs(cs,
 				      request->fence.seqno,
@@ -2284,7 +2312,8 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
 	engine->request_alloc = execlists_request_alloc;
 
 	engine->emit_flush = gen8_emit_flush;
-	engine->emit_breadcrumb = gen8_emit_breadcrumb;
+	engine->emit_init_breadcrumb = gen8_emit_init_breadcrumb;
+	engine->emit_fini_breadcrumb = gen8_emit_fini_breadcrumb;
 
 	engine->set_default_submission = intel_execlists_set_default_submission;
 
@@ -2397,7 +2426,7 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
 	/* Override some for render ring. */
 	engine->init_context = gen8_init_rcs_context;
 	engine->emit_flush = gen8_emit_flush_render;
-	engine->emit_breadcrumb = gen8_emit_breadcrumb_rcs;
+	engine->emit_fini_breadcrumb = gen8_emit_fini_breadcrumb_rcs;
 
 	ret = logical_ring_init(engine);
 	if (ret)
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 71a51a129c98..3f9f0a5a0f44 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1600,6 +1600,7 @@ static int intel_init_ring_buffer(struct intel_engine_cs *engine)
 		err = PTR_ERR(timeline);
 		goto err;
 	}
+	GEM_BUG_ON(timeline->has_initial_breadcrumb);
 
 	ring = intel_engine_create_ring(engine, timeline, 32 * PAGE_SIZE);
 	i915_timeline_put(timeline);
@@ -1953,6 +1954,7 @@ static int ring_request_alloc(struct i915_request *request)
 	int ret;
 
 	GEM_BUG_ON(!request->hw_context->pin_count);
+	GEM_BUG_ON(request->timeline->has_initial_breadcrumb);
 
 	/*
 	 * Flush enough space to reduce the likelihood of waiting after
@@ -2289,9 +2291,14 @@ static void intel_ring_default_vfuncs(struct drm_i915_private *dev_priv,
 	engine->context_pin = intel_ring_context_pin;
 	engine->request_alloc = ring_request_alloc;
 
-	engine->emit_breadcrumb = i9xx_emit_breadcrumb;
+	/*
+	 * Using a global execution timeline; the previous final breadcrumb is
+	 * equivalent to our next initial bread so we can elide
+	 * engine->emit_init_breadcrumb().
+	 */
+	engine->emit_fini_breadcrumb = i9xx_emit_breadcrumb;
 	if (IS_GEN(dev_priv, 5))
-		engine->emit_breadcrumb = gen5_emit_breadcrumb;
+		engine->emit_fini_breadcrumb = gen5_emit_breadcrumb;
 
 	engine->set_default_submission = i9xx_set_default_submission;
 
@@ -2320,11 +2327,11 @@ int intel_init_render_ring_buffer(struct intel_engine_cs *engine)
 	if (INTEL_GEN(dev_priv) >= 7) {
 		engine->init_context = intel_rcs_ctx_init;
 		engine->emit_flush = gen7_render_ring_flush;
-		engine->emit_breadcrumb = gen7_rcs_emit_breadcrumb;
+		engine->emit_fini_breadcrumb = gen7_rcs_emit_breadcrumb;
 	} else if (IS_GEN(dev_priv, 6)) {
 		engine->init_context = intel_rcs_ctx_init;
 		engine->emit_flush = gen6_render_ring_flush;
-		engine->emit_breadcrumb = gen6_rcs_emit_breadcrumb;
+		engine->emit_fini_breadcrumb = gen6_rcs_emit_breadcrumb;
 	} else if (IS_GEN(dev_priv, 5)) {
 		engine->emit_flush = gen4_render_ring_flush;
 	} else {
@@ -2361,9 +2368,9 @@ int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine)
 		engine->irq_enable_mask = GT_BSD_USER_INTERRUPT;
 
 		if (IS_GEN(dev_priv, 6))
-			engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
+			engine->emit_fini_breadcrumb = gen6_xcs_emit_breadcrumb;
 		else
-			engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
+			engine->emit_fini_breadcrumb = gen7_xcs_emit_breadcrumb;
 	} else {
 		engine->emit_flush = bsd_ring_flush;
 		if (IS_GEN(dev_priv, 5))
@@ -2387,9 +2394,9 @@ int intel_init_blt_ring_buffer(struct intel_engine_cs *engine)
 	engine->irq_enable_mask = GT_BLT_USER_INTERRUPT;
 
 	if (IS_GEN(dev_priv, 6))
-		engine->emit_breadcrumb = gen6_xcs_emit_breadcrumb;
+		engine->emit_fini_breadcrumb = gen6_xcs_emit_breadcrumb;
 	else
-		engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
+		engine->emit_fini_breadcrumb = gen7_xcs_emit_breadcrumb;
 
 	return intel_init_ring_buffer(engine);
 }
@@ -2407,7 +2414,7 @@ int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine)
 	engine->irq_enable = hsw_vebox_irq_enable;
 	engine->irq_disable = hsw_vebox_irq_disable;
 
-	engine->emit_breadcrumb = gen7_xcs_emit_breadcrumb;
+	engine->emit_fini_breadcrumb = gen7_xcs_emit_breadcrumb;
 
 	return intel_init_ring_buffer(engine);
 }
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 2927b712b973..5ed06d4230f6 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -463,8 +463,11 @@ struct intel_engine_cs {
 					 unsigned int dispatch_flags);
 #define I915_DISPATCH_SECURE BIT(0)
 #define I915_DISPATCH_PINNED BIT(1)
-	u32		*(*emit_breadcrumb)(struct i915_request *rq, u32 *cs);
-	int		emit_breadcrumb_dw;
+	int		(*emit_init_breadcrumb)(struct i915_request *rq);
+	u32		*(*emit_fini_breadcrumb)(struct i915_request *rq,
+						 u32 *cs);
+	unsigned int	emit_fini_breadcrumb_dw;
+	unsigned int	emit_fini_breadcrumb_sz;
 
 	/* Pass the request to the hardware queue (e.g. directly into
 	 * the legacy ringbuffer or to the end of an execlist).
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 95e890d7f58b..3b226ebc6bc4 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -227,7 +227,7 @@ struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
 	engine->base.context_pin = mock_context_pin;
 	engine->base.request_alloc = mock_request_alloc;
 	engine->base.emit_flush = mock_emit_flush;
-	engine->base.emit_breadcrumb = mock_emit_breadcrumb;
+	engine->base.emit_fini_breadcrumb = mock_emit_breadcrumb;
 	engine->base.submit_request = mock_submit_request;
 
 	if (i915_timeline_init(i915,
-- 
2.20.1

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

[PATCH 26/33] drm/i915: Remove the intel_engine_notify tracepoint

[Chris Wilson]

The global seqno is defunct and so we have no meaningful indicator of
forward progress for an engine. You need to listen to the request
signaling tracepoints instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c   |  2 --
 drivers/gpu/drm/i915/i915_trace.h | 25 -------------------------
 2 files changed, 27 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 7e56611b3d60..bedac6c14928 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -1211,8 +1211,6 @@ static void notify_ring(struct intel_engine_cs *engine)
 		wake_up_process(tsk);
 
 	rcu_read_unlock();
-
-	trace_intel_engine_notify(engine, wait);
 }
 
 static void vlv_c0_read(struct drm_i915_private *dev_priv,
diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
index 43da14f08dc0..eab313c3163c 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -752,31 +752,6 @@ trace_i915_request_out(struct i915_request *rq)
 #endif
 #endif
 
-TRACE_EVENT(intel_engine_notify,
-	    TP_PROTO(struct intel_engine_cs *engine, bool waiters),
-	    TP_ARGS(engine, waiters),
-
-	    TP_STRUCT__entry(
-			     __field(u32, dev)
-			     __field(u16, class)
-			     __field(u16, instance)
-			     __field(u32, seqno)
-			     __field(bool, waiters)
-			     ),
-
-	    TP_fast_assign(
-			   __entry->dev = engine->i915->drm.primary->index;
-			   __entry->class = engine->uabi_class;
-			   __entry->instance = engine->instance;
-			   __entry->seqno = intel_engine_get_seqno(engine);
-			   __entry->waiters = waiters;
-			   ),
-
-	    TP_printk("dev=%u, engine=%u:%u, seqno=%u, waiters=%u",
-		      __entry->dev, __entry->class, __entry->instance,
-		      __entry->seqno, __entry->waiters)
-);
-
 DEFINE_EVENT(i915_request, i915_request_retire,
 	    TP_PROTO(struct i915_request *rq),
 	    TP_ARGS(rq)
-- 
2.20.1

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

Re: [PATCH 26/33] drm/i915: Remove the intel_engine_notify tracepoint

[Tvrtko Ursulin]

On 25/01/2019 02:29, Chris Wilson wrote:
> The global seqno is defunct and so we have no meaningful indicator of
> forward progress for an engine. You need to listen to the request
> signaling tracepoints instead.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_irq.c   |  2 --
>   drivers/gpu/drm/i915/i915_trace.h | 25 -------------------------
>   2 files changed, 27 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 7e56611b3d60..bedac6c14928 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -1211,8 +1211,6 @@ static void notify_ring(struct intel_engine_cs *engine)
>   		wake_up_process(tsk);
>   
>   	rcu_read_unlock();
> -
> -	trace_intel_engine_notify(engine, wait);
>   }
>   
>   static void vlv_c0_read(struct drm_i915_private *dev_priv,
> diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
> index 43da14f08dc0..eab313c3163c 100644
> --- a/drivers/gpu/drm/i915/i915_trace.h
> +++ b/drivers/gpu/drm/i915/i915_trace.h
> @@ -752,31 +752,6 @@ trace_i915_request_out(struct i915_request *rq)
>   #endif
>   #endif
>   
> -TRACE_EVENT(intel_engine_notify,
> -	    TP_PROTO(struct intel_engine_cs *engine, bool waiters),
> -	    TP_ARGS(engine, waiters),
> -
> -	    TP_STRUCT__entry(
> -			     __field(u32, dev)
> -			     __field(u16, class)
> -			     __field(u16, instance)
> -			     __field(u32, seqno)
> -			     __field(bool, waiters)
> -			     ),
> -
> -	    TP_fast_assign(
> -			   __entry->dev = engine->i915->drm.primary->index;
> -			   __entry->class = engine->uabi_class;
> -			   __entry->instance = engine->instance;
> -			   __entry->seqno = intel_engine_get_seqno(engine);
> -			   __entry->waiters = waiters;
> -			   ),
> -
> -	    TP_printk("dev=%u, engine=%u:%u, seqno=%u, waiters=%u",
> -		      __entry->dev, __entry->class, __entry->instance,
> -		      __entry->seqno, __entry->waiters)
> -);
> -
>   DEFINE_EVENT(i915_request, i915_request_retire,
>   	    TP_PROTO(struct i915_request *rq),
>   	    TP_ARGS(rq)
> 

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

I think I checked that trace_dma_fence_singal will allow me to fix 
trace.pl.. I think I have.

Regards,

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

[PATCH 27/33] drm/i915: Replace global breadcrumbs with per-context interrupt tracking

[Chris Wilson]

A few years ago, see commit 688e6c725816 ("drm/i915: Slaughter the
thundering i915_wait_request herd"), the issue of handling multiple
clients waiting in parallel was brought to our attention. The
requirement was that every client should be woken immediately upon its
request being signaled, without incurring any cpu overhead.

To handle certain fragility of our hw meant that we could not do a
simple check inside the irq handler (some generations required almost
unbounded delays before we could be sure of seqno coherency) and so
request completion checking required delegation.

Before commit 688e6c725816, the solution was simple. Every client waking
on a request would be woken on every interrupt and each would do a
heavyweight check to see if their request was complete. Commit
688e6c725816 introduced an rbtree so that only the earliest waiter on
the global timeline would woken, and would wake the next and so on.
(Along with various complications to handle requests being reordered
along the global timeline, and also a requirement for kthread to provide
a delegate for fence signaling that had no process context.)

The global rbtree depends on knowing the execution timeline (and global
seqno). Without knowing that order, we must instead check all contexts
queued to the HW to see which may have advanced. We trim that list by
only checking queued contexts that are being waited on, but still we
keep a list of all active contexts and their active signalers that we
inspect from inside the irq handler. By moving the waiters onto the fence
signal list, we can combine the client wakeup with the dma_fence
signaling (a dramatic reduction in complexity, but does require the HW
being coherent, the seqno must be visible from the cpu before the
interrupt is raised - we keep a timer backup just in case).

Having previously fixed all the issues with irq-seqno serialisation (by
inserting delays onto the GPU after each request instead of random delays
on the CPU after each interrupt), we can rely on the seqno state to
perfom direct wakeups from the interrupt handler. This allows us to
preserve our single context switch behaviour of the current routine,
with the only downside that we lose the RT priority sorting of wakeups.
In general, direct wakeup latency of multiple clients is about the same
(about 10% better in most cases) with a reduction in total CPU time spent
in the waiter (about 20-50% depending on gen). Average herd behaviour is
improved, but at the cost of not delegating wakeups on task_prio.

v2: Capture fence signaling state for error state and add comments to
warm even the most cold of hearts.
v3: Check if the request is still active before busywaiting
v4: Reduce the amount of pointer misdirection with list_for_each_safe
and using a local i915_request variable inside the loops

References: 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c           |  28 +-
 drivers/gpu/drm/i915/i915_gem_context.c       |   3 +
 drivers/gpu/drm/i915/i915_gem_context.h       |   2 +
 drivers/gpu/drm/i915/i915_gpu_error.c         |  83 +-
 drivers/gpu/drm/i915/i915_gpu_error.h         |   9 +-
 drivers/gpu/drm/i915/i915_irq.c               |  88 +-
 drivers/gpu/drm/i915/i915_request.c           | 140 +--
 drivers/gpu/drm/i915/i915_request.h           |  72 +-
 drivers/gpu/drm/i915/i915_reset.c             |  13 +-
 drivers/gpu/drm/i915/intel_breadcrumbs.c      | 818 +++++-------------
 drivers/gpu/drm/i915/intel_engine_cs.c        |  35 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c       |   6 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h       |  94 +-
 .../drm/i915/selftests/i915_mock_selftests.h  |   1 -
 drivers/gpu/drm/i915/selftests/i915_request.c | 420 +++++++++
 drivers/gpu/drm/i915/selftests/igt_spinner.c  |   5 -
 .../drm/i915/selftests/intel_breadcrumbs.c    | 470 ----------
 .../gpu/drm/i915/selftests/intel_hangcheck.c  |   2 +-
 drivers/gpu/drm/i915/selftests/lib_sw_fence.c |  54 ++
 drivers/gpu/drm/i915/selftests/lib_sw_fence.h |   3 +
 drivers/gpu/drm/i915/selftests/mock_engine.c  |  17 +-
 drivers/gpu/drm/i915/selftests/mock_engine.h  |   6 -
 22 files changed, 886 insertions(+), 1483 deletions(-)
 delete mode 100644 drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index b64f758803f3..e0ca3987eed6 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1316,29 +1316,16 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 	seq_printf(m, "GT active? %s\n", yesno(dev_priv->gt.awake));
 
 	for_each_engine(engine, dev_priv, id) {
-		struct intel_breadcrumbs *b = &engine->breadcrumbs;
-		struct rb_node *rb;
-
 		seq_printf(m, "%s:\n", engine->name);
 		seq_printf(m, "\tseqno = %x [current %x, last %x], %dms ago\n",
 			   engine->hangcheck.seqno, seqno[id],
 			   intel_engine_last_submit(engine),
 			   jiffies_to_msecs(jiffies -
 					    engine->hangcheck.action_timestamp));
-		seq_printf(m, "\twaiters? %s, fake irq active? %s\n",
-			   yesno(intel_engine_has_waiter(engine)),
+		seq_printf(m, "\tfake irq active? %s\n",
 			   yesno(test_bit(engine->id,
 					  &dev_priv->gpu_error.missed_irq_rings)));
 
-		spin_lock_irq(&b->rb_lock);
-		for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
-			struct intel_wait *w = rb_entry(rb, typeof(*w), node);
-
-			seq_printf(m, "\t%s [%d] waiting for %x\n",
-				   w->tsk->comm, w->tsk->pid, w->seqno);
-		}
-		spin_unlock_irq(&b->rb_lock);
-
 		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
 			   (long long)engine->hangcheck.acthd,
 			   (long long)acthd[id]);
@@ -2022,18 +2009,6 @@ static int i915_swizzle_info(struct seq_file *m, void *data)
 	return 0;
 }
 
-static int count_irq_waiters(struct drm_i915_private *i915)
-{
-	struct intel_engine_cs *engine;
-	enum intel_engine_id id;
-	int count = 0;
-
-	for_each_engine(engine, i915, id)
-		count += intel_engine_has_waiter(engine);
-
-	return count;
-}
-
 static const char *rps_power_to_str(unsigned int power)
 {
 	static const char * const strings[] = {
@@ -2073,7 +2048,6 @@ static int i915_rps_boost_info(struct seq_file *m, void *data)
 	seq_printf(m, "RPS enabled? %d\n", rps->enabled);
 	seq_printf(m, "GPU busy? %s [%d requests]\n",
 		   yesno(dev_priv->gt.awake), dev_priv->gt.active_requests);
-	seq_printf(m, "CPU waiting? %d\n", count_irq_waiters(dev_priv));
 	seq_printf(m, "Boosts outstanding? %d\n",
 		   atomic_read(&rps->num_waiters));
 	seq_printf(m, "Interactive? %d\n", READ_ONCE(rps->power.interactive));
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 93e84751370f..6faf1f6faab5 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -327,6 +327,9 @@ intel_context_init(struct intel_context *ce,
 		   struct intel_engine_cs *engine)
 {
 	ce->gem_context = ctx;
+
+	INIT_LIST_HEAD(&ce->signal_link);
+	INIT_LIST_HEAD(&ce->signals);
 }
 
 static struct i915_gem_context *
diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
index 3769438228f6..6ba40ff6b91f 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.h
+++ b/drivers/gpu/drm/i915/i915_gem_context.h
@@ -164,6 +164,8 @@ struct i915_gem_context {
 	struct intel_context {
 		struct i915_gem_context *gem_context;
 		struct intel_engine_cs *active;
+		struct list_head signal_link;
+		struct list_head signals;
 		struct i915_vma *state;
 		struct intel_ring *ring;
 		u32 *lrc_reg_state;
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 96d1d634a29d..420b94341433 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -447,9 +447,14 @@ static void error_print_request(struct drm_i915_error_state_buf *m,
 	if (!erq->seqno)
 		return;
 
-	err_printf(m, "%s pid %d, ban score %d, seqno %8x:%08x, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
+	err_printf(m, "%s pid %d, ban score %d, seqno %8x:%08x%s%s, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
 		   prefix, erq->pid, erq->ban_score,
-		   erq->context, erq->seqno, erq->sched_attr.priority,
+		   erq->context, erq->seqno,
+		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+			    &erq->flags) ? "!" : "",
+		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
+			    &erq->flags) ? "+" : "",
+		   erq->sched_attr.priority,
 		   jiffies_to_msecs(erq->jiffies - epoch),
 		   erq->start, erq->head, erq->tail);
 }
@@ -530,7 +535,6 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
 	}
 	err_printf(m, "  seqno: 0x%08x\n", ee->seqno);
 	err_printf(m, "  last_seqno: 0x%08x\n", ee->last_seqno);
-	err_printf(m, "  waiting: %s\n", yesno(ee->waiting));
 	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
 	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
 	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
@@ -804,21 +808,6 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 						    error->epoch);
 		}
 
-		if (IS_ERR(ee->waiters)) {
-			err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n",
-				   m->i915->engine[i]->name);
-		} else if (ee->num_waiters) {
-			err_printf(m, "%s --- %d waiters\n",
-				   m->i915->engine[i]->name,
-				   ee->num_waiters);
-			for (j = 0; j < ee->num_waiters; j++) {
-				err_printf(m, " seqno 0x%08x for %s [%d]\n",
-					   ee->waiters[j].seqno,
-					   ee->waiters[j].comm,
-					   ee->waiters[j].pid);
-			}
-		}
-
 		print_error_obj(m, m->i915->engine[i],
 				"ringbuffer", ee->ringbuffer);
 
@@ -1000,8 +989,6 @@ void __i915_gpu_state_free(struct kref *error_ref)
 		i915_error_object_free(ee->wa_ctx);
 
 		kfree(ee->requests);
-		if (!IS_ERR_OR_NULL(ee->waiters))
-			kfree(ee->waiters);
 	}
 
 	for (i = 0; i < ARRAY_SIZE(error->active_bo); i++)
@@ -1203,59 +1190,6 @@ static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
 			I915_READ(RING_SYNC_2(engine->mmio_base));
 }
 
-static void error_record_engine_waiters(struct intel_engine_cs *engine,
-					struct drm_i915_error_engine *ee)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct drm_i915_error_waiter *waiter;
-	struct rb_node *rb;
-	int count;
-
-	ee->num_waiters = 0;
-	ee->waiters = NULL;
-
-	if (RB_EMPTY_ROOT(&b->waiters))
-		return;
-
-	if (!spin_trylock_irq(&b->rb_lock)) {
-		ee->waiters = ERR_PTR(-EDEADLK);
-		return;
-	}
-
-	count = 0;
-	for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
-		count++;
-	spin_unlock_irq(&b->rb_lock);
-
-	waiter = NULL;
-	if (count)
-		waiter = kmalloc_array(count,
-				       sizeof(struct drm_i915_error_waiter),
-				       GFP_ATOMIC);
-	if (!waiter)
-		return;
-
-	if (!spin_trylock_irq(&b->rb_lock)) {
-		kfree(waiter);
-		ee->waiters = ERR_PTR(-EDEADLK);
-		return;
-	}
-
-	ee->waiters = waiter;
-	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
-		struct intel_wait *w = rb_entry(rb, typeof(*w), node);
-
-		strcpy(waiter->comm, w->tsk->comm);
-		waiter->pid = w->tsk->pid;
-		waiter->seqno = w->seqno;
-		waiter++;
-
-		if (++ee->num_waiters == count)
-			break;
-	}
-	spin_unlock_irq(&b->rb_lock);
-}
-
 static void error_record_engine_registers(struct i915_gpu_state *error,
 					  struct intel_engine_cs *engine,
 					  struct drm_i915_error_engine *ee)
@@ -1291,7 +1225,6 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
 
 	intel_engine_get_instdone(engine, &ee->instdone);
 
-	ee->waiting = intel_engine_has_waiter(engine);
 	ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
 	ee->acthd = intel_engine_get_active_head(engine);
 	ee->seqno = intel_engine_get_seqno(engine);
@@ -1365,6 +1298,7 @@ static void record_request(struct i915_request *request,
 {
 	struct i915_gem_context *ctx = request->gem_context;
 
+	erq->flags = request->fence.flags;
 	erq->context = ctx->hw_id;
 	erq->sched_attr = request->sched.attr;
 	erq->ban_score = atomic_read(&ctx->ban_score);
@@ -1540,7 +1474,6 @@ static void gem_record_rings(struct i915_gpu_state *error)
 		ee->engine_id = i;
 
 		error_record_engine_registers(error, engine, ee);
-		error_record_engine_waiters(engine, ee);
 		error_record_engine_execlists(engine, ee);
 
 		request = i915_gem_find_active_request(engine);
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
index 231173786eae..74757c424aab 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.h
+++ b/drivers/gpu/drm/i915/i915_gpu_error.h
@@ -82,8 +82,6 @@ struct i915_gpu_state {
 		int engine_id;
 		/* Software tracked state */
 		bool idle;
-		bool waiting;
-		int num_waiters;
 		unsigned long hangcheck_timestamp;
 		struct i915_address_space *vm;
 		int num_requests;
@@ -147,6 +145,7 @@ struct i915_gpu_state {
 		struct drm_i915_error_object *default_state;
 
 		struct drm_i915_error_request {
+			unsigned long flags;
 			long jiffies;
 			pid_t pid;
 			u32 context;
@@ -159,12 +158,6 @@ struct i915_gpu_state {
 		} *requests, execlist[EXECLIST_MAX_PORTS];
 		unsigned int num_ports;
 
-		struct drm_i915_error_waiter {
-			char comm[TASK_COMM_LEN];
-			pid_t pid;
-			u32 seqno;
-		} *waiters;
-
 		struct {
 			u32 gfx_mode;
 			union {
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index bedac6c14928..f09ea93e1037 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -28,12 +28,14 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
-#include <linux/sysrq.h>
-#include <linux/slab.h>
 #include <linux/circ_buf.h>
+#include <linux/slab.h>
+#include <linux/sysrq.h>
+
 #include <drm/drm_irq.h>
 #include <drm/drm_drv.h>
 #include <drm/i915_drm.h>
+
 #include "i915_drv.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
@@ -1153,66 +1155,6 @@ static void ironlake_rps_change_irq_handler(struct drm_i915_private *dev_priv)
 	return;
 }
 
-static void notify_ring(struct intel_engine_cs *engine)
-{
-	const u32 seqno = intel_engine_get_seqno(engine);
-	struct i915_request *rq = NULL;
-	struct task_struct *tsk = NULL;
-	struct intel_wait *wait;
-
-	if (unlikely(!engine->breadcrumbs.irq_armed))
-		return;
-
-	rcu_read_lock();
-
-	spin_lock(&engine->breadcrumbs.irq_lock);
-	wait = engine->breadcrumbs.irq_wait;
-	if (wait) {
-		/*
-		 * We use a callback from the dma-fence to submit
-		 * requests after waiting on our own requests. To
-		 * ensure minimum delay in queuing the next request to
-		 * hardware, signal the fence now rather than wait for
-		 * the signaler to be woken up. We still wake up the
-		 * waiter in order to handle the irq-seqno coherency
-		 * issues (we may receive the interrupt before the
-		 * seqno is written, see __i915_request_irq_complete())
-		 * and to handle coalescing of multiple seqno updates
-		 * and many waiters.
-		 */
-		if (i915_seqno_passed(seqno, wait->seqno)) {
-			struct i915_request *waiter = wait->request;
-
-			if (waiter &&
-			    !i915_request_signaled(waiter) &&
-			    intel_wait_check_request(wait, waiter))
-				rq = i915_request_get(waiter);
-
-			tsk = wait->tsk;
-		}
-
-		engine->breadcrumbs.irq_count++;
-	} else {
-		if (engine->breadcrumbs.irq_armed)
-			__intel_engine_disarm_breadcrumbs(engine);
-	}
-	spin_unlock(&engine->breadcrumbs.irq_lock);
-
-	if (rq) {
-		spin_lock(&rq->lock);
-		dma_fence_signal_locked(&rq->fence);
-		GEM_BUG_ON(!i915_request_completed(rq));
-		spin_unlock(&rq->lock);
-
-		i915_request_put(rq);
-	}
-
-	if (tsk && tsk->state & TASK_NORMAL)
-		wake_up_process(tsk);
-
-	rcu_read_unlock();
-}
-
 static void vlv_c0_read(struct drm_i915_private *dev_priv,
 			struct intel_rps_ei *ei)
 {
@@ -1457,20 +1399,20 @@ static void ilk_gt_irq_handler(struct drm_i915_private *dev_priv,
 			       u32 gt_iir)
 {
 	if (gt_iir & GT_RENDER_USER_INTERRUPT)
-		notify_ring(dev_priv->engine[RCS]);
+		intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
 	if (gt_iir & ILK_BSD_USER_INTERRUPT)
-		notify_ring(dev_priv->engine[VCS]);
+		intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
 }
 
 static void snb_gt_irq_handler(struct drm_i915_private *dev_priv,
 			       u32 gt_iir)
 {
 	if (gt_iir & GT_RENDER_USER_INTERRUPT)
-		notify_ring(dev_priv->engine[RCS]);
+		intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
 	if (gt_iir & GT_BSD_USER_INTERRUPT)
-		notify_ring(dev_priv->engine[VCS]);
+		intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
 	if (gt_iir & GT_BLT_USER_INTERRUPT)
-		notify_ring(dev_priv->engine[BCS]);
+		intel_engine_breadcrumbs_irq(dev_priv->engine[BCS]);
 
 	if (gt_iir & (GT_BLT_CS_ERROR_INTERRUPT |
 		      GT_BSD_CS_ERROR_INTERRUPT |
@@ -1490,7 +1432,7 @@ gen8_cs_irq_handler(struct intel_engine_cs *engine, u32 iir)
 		tasklet = true;
 
 	if (iir & GT_RENDER_USER_INTERRUPT) {
-		notify_ring(engine);
+		intel_engine_breadcrumbs_irq(engine);
 		tasklet |= USES_GUC_SUBMISSION(engine->i915);
 	}
 
@@ -1836,7 +1778,7 @@ static void gen6_rps_irq_handler(struct drm_i915_private *dev_priv, u32 pm_iir)
 
 	if (HAS_VEBOX(dev_priv)) {
 		if (pm_iir & PM_VEBOX_USER_INTERRUPT)
-			notify_ring(dev_priv->engine[VECS]);
+			intel_engine_breadcrumbs_irq(dev_priv->engine[VECS]);
 
 		if (pm_iir & PM_VEBOX_CS_ERROR_INTERRUPT)
 			DRM_DEBUG("Command parser error, pm_iir 0x%08x\n", pm_iir);
@@ -4260,7 +4202,7 @@ static irqreturn_t i8xx_irq_handler(int irq, void *arg)
 		I915_WRITE16(IIR, iir);
 
 		if (iir & I915_USER_INTERRUPT)
-			notify_ring(dev_priv->engine[RCS]);
+			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
 
 		if (iir & I915_MASTER_ERROR_INTERRUPT)
 			i8xx_error_irq_handler(dev_priv, eir, eir_stuck);
@@ -4368,7 +4310,7 @@ static irqreturn_t i915_irq_handler(int irq, void *arg)
 		I915_WRITE(IIR, iir);
 
 		if (iir & I915_USER_INTERRUPT)
-			notify_ring(dev_priv->engine[RCS]);
+			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
 
 		if (iir & I915_MASTER_ERROR_INTERRUPT)
 			i9xx_error_irq_handler(dev_priv, eir, eir_stuck);
@@ -4513,10 +4455,10 @@ static irqreturn_t i965_irq_handler(int irq, void *arg)
 		I915_WRITE(IIR, iir);
 
 		if (iir & I915_USER_INTERRUPT)
-			notify_ring(dev_priv->engine[RCS]);
+			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
 
 		if (iir & I915_BSD_USER_INTERRUPT)
-			notify_ring(dev_priv->engine[VCS]);
+			intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
 
 		if (iir & I915_MASTER_ERROR_INTERRUPT)
 			i9xx_error_irq_handler(dev_priv, eir, eir_stuck);
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 2171df2d3019..c09a6644a2ab 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -60,7 +60,7 @@ static bool i915_fence_signaled(struct dma_fence *fence)
 
 static bool i915_fence_enable_signaling(struct dma_fence *fence)
 {
-	return intel_engine_enable_signaling(to_request(fence), true);
+	return i915_request_enable_breadcrumb(to_request(fence));
 }
 
 static signed long i915_fence_wait(struct dma_fence *fence,
@@ -203,7 +203,7 @@ static void __retire_engine_request(struct intel_engine_cs *engine,
 	if (!i915_request_signaled(rq))
 		dma_fence_signal_locked(&rq->fence);
 	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &rq->fence.flags))
-		intel_engine_cancel_signaling(rq);
+		i915_request_cancel_breadcrumb(rq);
 	if (rq->waitboost) {
 		GEM_BUG_ON(!atomic_read(&rq->i915->gt_pm.rps.num_waiters));
 		atomic_dec(&rq->i915->gt_pm.rps.num_waiters);
@@ -377,9 +377,11 @@ void __i915_request_submit(struct i915_request *request)
 
 	/* We may be recursing from the signal callback of another i915 fence */
 	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
+	set_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags);
 	request->global_seqno = seqno;
-	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags))
-		intel_engine_enable_signaling(request, false);
+	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags) &&
+	    !i915_request_enable_breadcrumb(request))
+		intel_engine_queue_breadcrumbs(engine);
 	spin_unlock(&request->lock);
 
 	engine->emit_fini_breadcrumb(request,
@@ -389,8 +391,6 @@ void __i915_request_submit(struct i915_request *request)
 	move_to_timeline(request, &engine->timeline);
 
 	trace_i915_request_execute(request);
-
-	wake_up_all(&request->execute);
 }
 
 void i915_request_submit(struct i915_request *request)
@@ -433,7 +433,8 @@ void __i915_request_unsubmit(struct i915_request *request)
 	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
 	request->global_seqno = 0;
 	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags))
-		intel_engine_cancel_signaling(request);
+		i915_request_cancel_breadcrumb(request);
+	clear_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags);
 	spin_unlock(&request->lock);
 
 	/* Transfer back from the global per-engine timeline to per-context */
@@ -646,13 +647,11 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 
 	/* We bump the ref for the fence chain */
 	i915_sw_fence_init(&i915_request_get(rq)->submit, submit_notify);
-	init_waitqueue_head(&rq->execute);
 
 	i915_sched_node_init(&rq->sched);
 
 	/* No zalloc, must clear what we need by hand */
 	rq->global_seqno = 0;
-	rq->signaling.wait.seqno = 0;
 	rq->file_priv = NULL;
 	rq->batch = NULL;
 	rq->capture_list = NULL;
@@ -1047,13 +1046,10 @@ static bool busywait_stop(unsigned long timeout, unsigned int cpu)
 	return this_cpu != cpu;
 }
 
-static bool __i915_spin_request(const struct i915_request *rq,
-				u32 seqno, int state, unsigned long timeout_us)
+static bool __i915_spin_request(const struct i915_request * const rq,
+				int state, unsigned long timeout_us)
 {
-	struct intel_engine_cs *engine = rq->engine;
-	unsigned int irq, cpu;
-
-	GEM_BUG_ON(!seqno);
+	unsigned int cpu;
 
 	/*
 	 * Only wait for the request if we know it is likely to complete.
@@ -1061,12 +1057,12 @@ static bool __i915_spin_request(const struct i915_request *rq,
 	 * We don't track the timestamps around requests, nor the average
 	 * request length, so we do not have a good indicator that this
 	 * request will complete within the timeout. What we do know is the
-	 * order in which requests are executed by the engine and so we can
-	 * tell if the request has started. If the request hasn't started yet,
-	 * it is a fair assumption that it will not complete within our
-	 * relatively short timeout.
+	 * order in which requests are executed by the context and so we can
+	 * tell if the request has been started. If the request is not even
+	 * running yet, it is a fair assumption that it will not complete
+	 * within our relatively short timeout.
 	 */
-	if (!intel_engine_has_started(engine, seqno))
+	if (!i915_request_is_running(rq))
 		return false;
 
 	/*
@@ -1080,20 +1076,10 @@ static bool __i915_spin_request(const struct i915_request *rq,
 	 * takes to sleep on a request, on the order of a microsecond.
 	 */
 
-	irq = READ_ONCE(engine->breadcrumbs.irq_count);
 	timeout_us += local_clock_us(&cpu);
 	do {
-		if (intel_engine_has_completed(engine, seqno))
-			return seqno == i915_request_global_seqno(rq);
-
-		/*
-		 * Seqno are meant to be ordered *before* the interrupt. If
-		 * we see an interrupt without a corresponding seqno advance,
-		 * assume we won't see one in the near future but require
-		 * the engine->seqno_barrier() to fixup coherency.
-		 */
-		if (READ_ONCE(engine->breadcrumbs.irq_count) != irq)
-			break;
+		if (i915_request_completed(rq))
+			return true;
 
 		if (signal_pending_state(state, current))
 			break;
@@ -1107,6 +1093,18 @@ static bool __i915_spin_request(const struct i915_request *rq,
 	return false;
 }
 
+struct request_wait {
+	struct dma_fence_cb cb;
+	struct task_struct *tsk;
+};
+
+static void request_wait_wake(struct dma_fence *fence, struct dma_fence_cb *cb)
+{
+	struct request_wait *wait = container_of(cb, typeof(*wait), cb);
+
+	wake_up_process(wait->tsk);
+}
+
 /**
  * i915_request_wait - wait until execution of request has finished
  * @rq: the request to wait upon
@@ -1132,8 +1130,7 @@ long i915_request_wait(struct i915_request *rq,
 {
 	const int state = flags & I915_WAIT_INTERRUPTIBLE ?
 		TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
-	DEFINE_WAIT_FUNC(exec, default_wake_function);
-	struct intel_wait wait;
+	struct request_wait wait;
 
 	might_sleep();
 	GEM_BUG_ON(timeout < 0);
@@ -1145,47 +1142,24 @@ long i915_request_wait(struct i915_request *rq,
 		return -ETIME;
 
 	trace_i915_request_wait_begin(rq, flags);
-	add_wait_queue(&rq->execute, &exec);
-	intel_wait_init(&wait);
-	if (flags & I915_WAIT_PRIORITY)
-		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
-
-restart:
-	do {
-		set_current_state(state);
-		if (intel_wait_update_request(&wait, rq))
-			break;
-
-		if (signal_pending_state(state, current)) {
-			timeout = -ERESTARTSYS;
-			goto complete;
-		}
 
-		if (!timeout) {
-			timeout = -ETIME;
-			goto complete;
-		}
+	/* Optimistic short spin before touching IRQs */
+	if (__i915_spin_request(rq, state, 5))
+		goto out;
 
-		timeout = io_schedule_timeout(timeout);
-	} while (1);
+	if (flags & I915_WAIT_PRIORITY)
+		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
 
-	GEM_BUG_ON(!intel_wait_has_seqno(&wait));
-	GEM_BUG_ON(!i915_sw_fence_signaled(&rq->submit));
+	wait.tsk = current;
+	if (dma_fence_add_callback(&rq->fence, &wait.cb, request_wait_wake))
+		goto out;
 
-	/* Optimistic short spin before touching IRQs */
-	if (__i915_spin_request(rq, wait.seqno, state, 5))
-		goto complete;
+	for (;;) {
+		set_current_state(state);
 
-	set_current_state(state);
-	if (intel_engine_add_wait(rq->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
-		 * coherent check on the seqno before we sleep.
-		 */
-		goto wakeup;
+		if (i915_request_completed(rq))
+			break;
 
-	for (;;) {
 		if (signal_pending_state(state, current)) {
 			timeout = -ERESTARTSYS;
 			break;
@@ -1197,33 +1171,13 @@ long i915_request_wait(struct i915_request *rq,
 		}
 
 		timeout = io_schedule_timeout(timeout);
-
-		if (intel_wait_complete(&wait) &&
-		    intel_wait_check_request(&wait, rq))
-			break;
-
-		set_current_state(state);
-
-wakeup:
-		if (i915_request_completed(rq))
-			break;
-
-		/* Only spin if we know the GPU is processing this request */
-		if (__i915_spin_request(rq, wait.seqno, state, 2))
-			break;
-
-		if (!intel_wait_check_request(&wait, rq)) {
-			intel_engine_remove_wait(rq->engine, &wait);
-			goto restart;
-		}
 	}
-
-	intel_engine_remove_wait(rq->engine, &wait);
-complete:
 	__set_current_state(TASK_RUNNING);
-	remove_wait_queue(&rq->execute, &exec);
-	trace_i915_request_wait_end(rq);
 
+	dma_fence_remove_callback(&rq->fence, &wait.cb);
+
+out:
+	trace_i915_request_wait_end(rq);
 	return timeout;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 340d6216791c..3cffb96203b9 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -38,23 +38,34 @@ struct drm_i915_gem_object;
 struct i915_request;
 struct i915_timeline;
 
-struct intel_wait {
-	struct rb_node node;
-	struct task_struct *tsk;
-	struct i915_request *request;
-	u32 seqno;
-};
-
-struct intel_signal_node {
-	struct intel_wait wait;
-	struct list_head link;
-};
-
 struct i915_capture_list {
 	struct i915_capture_list *next;
 	struct i915_vma *vma;
 };
 
+enum {
+	/*
+	 * I915_FENCE_FLAG_ACTIVE - this request is currently submitted to HW.
+	 *
+	 * Set by __i915_request_submit() on handing over to HW, and cleared
+	 * by __i915_request_unsubmit() if we preempt this request.
+	 *
+	 * Finally cleared for consistency on retiring the request, when
+	 * we know the HW is no longer running this request.
+	 *
+	 * See i915_request_is_active()
+	 */
+	I915_FENCE_FLAG_ACTIVE = DMA_FENCE_FLAG_USER_BITS,
+
+	/*
+	 * I915_FENCE_FLAG_SIGNAL - this request is currently on signal_list
+	 *
+	 * Internal bookkeeping used by the breadcrumb code to track when
+	 * a request is on the various signal_list.
+	 */
+	I915_FENCE_FLAG_SIGNAL,
+};
+
 /**
  * Request queue structure.
  *
@@ -97,7 +108,7 @@ struct i915_request {
 	struct intel_context *hw_context;
 	struct intel_ring *ring;
 	struct i915_timeline *timeline;
-	struct intel_signal_node signaling;
+	struct list_head signal_link;
 
 	/*
 	 * The rcu epoch of when this request was allocated. Used to judiciously
@@ -116,7 +127,6 @@ struct i915_request {
 	 */
 	struct i915_sw_fence submit;
 	wait_queue_entry_t submitq;
-	wait_queue_head_t execute;
 
 	/*
 	 * A list of everyone we wait upon, and everyone who waits upon us.
@@ -255,7 +265,7 @@ i915_request_put(struct i915_request *rq)
  * that it has passed the global seqno and the global seqno is unchanged
  * after the read, it is indeed complete).
  */
-static u32
+static inline u32
 i915_request_global_seqno(const struct i915_request *request)
 {
 	return READ_ONCE(request->global_seqno);
@@ -277,6 +287,10 @@ void i915_request_skip(struct i915_request *request, int error);
 void __i915_request_unsubmit(struct i915_request *request);
 void i915_request_unsubmit(struct i915_request *request);
 
+/* Note: part of the intel_breadcrumbs family */
+bool i915_request_enable_breadcrumb(struct i915_request *request);
+void i915_request_cancel_breadcrumb(struct i915_request *request);
+
 long i915_request_wait(struct i915_request *rq,
 		       unsigned int flags,
 		       long timeout)
@@ -293,6 +307,11 @@ static inline bool i915_request_signaled(const struct i915_request *rq)
 	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &rq->fence.flags);
 }
 
+static inline bool i915_request_is_active(const struct i915_request *rq)
+{
+	return test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
+}
+
 /**
  * Returns true if seq1 is later than seq2.
  */
@@ -330,6 +349,11 @@ static inline u32 hwsp_seqno(const struct i915_request *rq)
 	return seqno;
 }
 
+static inline bool __i915_request_has_started(const struct i915_request *rq)
+{
+	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
+}
+
 /**
  * i915_request_started - check if the request has begun being executed
  * @rq: the request
@@ -345,7 +369,23 @@ static inline bool i915_request_started(const struct i915_request *rq)
 		return true;
 
 	/* Remember: started but may have since been preempted! */
-	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
+	return __i915_request_has_started(rq);
+}
+
+/**
+ * i915_request_is_running - check if the request may actually be executing
+ * @rq: the request
+ *
+ * Returns true if the request is currently submitted to hardware, has passed
+ * its start point (i.e. the context is setup and not busywaiting). Note that
+ * it may no longer be running by the time the function returns!
+ */
+static inline bool i915_request_is_running(const struct i915_request *rq)
+{
+	if (!i915_request_is_active(rq))
+		return false;
+
+	return __i915_request_has_started(rq);
 }
 
 static inline bool i915_request_completed(const struct i915_request *rq)
diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
index c3861a0219a9..5d0507d88e41 100644
--- a/drivers/gpu/drm/i915/i915_reset.c
+++ b/drivers/gpu/drm/i915/i915_reset.c
@@ -751,18 +751,19 @@ static void reset_restart(struct drm_i915_private *i915)
 
 static void nop_submit_request(struct i915_request *request)
 {
+	struct intel_engine_cs *engine = request->engine;
 	unsigned long flags;
 
 	GEM_TRACE("%s fence %llx:%lld -> -EIO\n",
-		  request->engine->name,
-		  request->fence.context, request->fence.seqno);
+		  engine->name, request->fence.context, request->fence.seqno);
 	dma_fence_set_error(&request->fence, -EIO);
 
-	spin_lock_irqsave(&request->engine->timeline.lock, flags);
+	spin_lock_irqsave(&engine->timeline.lock, flags);
 	__i915_request_submit(request);
 	i915_request_mark_complete(request);
-	intel_engine_write_global_seqno(request->engine, request->global_seqno);
-	spin_unlock_irqrestore(&request->engine->timeline.lock, flags);
+	spin_unlock_irqrestore(&engine->timeline.lock, flags);
+
+	intel_engine_queue_breadcrumbs(engine);
 }
 
 void i915_gem_set_wedged(struct drm_i915_private *i915)
@@ -817,7 +818,7 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 
 	for_each_engine(engine, i915, id) {
 		reset_finish_engine(engine);
-		intel_engine_wakeup(engine);
+		intel_engine_signal_breadcrumbs(engine);
 	}
 
 	smp_mb__before_atomic();
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index b58915b8708b..b0795b0ad227 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -29,48 +29,149 @@
 
 #define task_asleep(tsk) ((tsk)->state & TASK_NORMAL && !(tsk)->on_rq)
 
-static unsigned int __intel_breadcrumbs_wakeup(struct intel_breadcrumbs *b)
+static void irq_enable(struct intel_engine_cs *engine)
+{
+	if (!engine->irq_enable)
+		return;
+
+	/* Caller disables interrupts */
+	spin_lock(&engine->i915->irq_lock);
+	engine->irq_enable(engine);
+	spin_unlock(&engine->i915->irq_lock);
+}
+
+static void irq_disable(struct intel_engine_cs *engine)
 {
-	struct intel_wait *wait;
-	unsigned int result = 0;
+	if (!engine->irq_disable)
+		return;
+
+	/* Caller disables interrupts */
+	spin_lock(&engine->i915->irq_lock);
+	engine->irq_disable(engine);
+	spin_unlock(&engine->i915->irq_lock);
+}
 
+static void __intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
+{
 	lockdep_assert_held(&b->irq_lock);
 
-	wait = b->irq_wait;
-	if (wait) {
+	GEM_BUG_ON(!b->irq_enabled);
+	if (!--b->irq_enabled)
+		irq_disable(container_of(b,
+					 struct intel_engine_cs,
+					 breadcrumbs));
+
+	b->irq_armed = false;
+}
+
+void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+	if (!b->irq_armed)
+		return;
+
+	spin_lock_irq(&b->irq_lock);
+	if (b->irq_armed)
+		__intel_breadcrumbs_disarm_irq(b);
+	spin_unlock_irq(&b->irq_lock);
+}
+
+static inline bool __request_completed(const struct i915_request *rq)
+{
+	return i915_seqno_passed(__hwsp_seqno(rq), rq->fence.seqno);
+}
+
+bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct intel_context *ce, *cn;
+	struct list_head *pos, *next;
+	LIST_HEAD(signal);
+
+	spin_lock(&b->irq_lock);
+
+	b->irq_fired = true;
+	if (b->irq_armed && list_empty(&b->signalers))
+		__intel_breadcrumbs_disarm_irq(b);
+
+	list_for_each_entry_safe(ce, cn, &b->signalers, signal_link) {
+		GEM_BUG_ON(list_empty(&ce->signals));
+
+		list_for_each_safe(pos, next, &ce->signals) {
+			struct i915_request *rq =
+				list_entry(pos, typeof(*rq), signal_link);
+
+			if (!__request_completed(rq))
+				break;
+
+			GEM_BUG_ON(!test_bit(I915_FENCE_FLAG_SIGNAL,
+					     &rq->fence.flags));
+			clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
+
+			/*
+			 * We may race with direct invocation of
+			 * dma_fence_signal(), e.g. i915_request_retire(),
+			 * in which case we can skip processing it ourselves.
+			 */
+			if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+				     &rq->fence.flags))
+				continue;
+
+			/*
+			 * Queue for execution after dropping the signaling
+			 * spinlock as the callback chain may end up adding
+			 * more signalers to the same context or engine.
+			 */
+			i915_request_get(rq);
+			list_add_tail(&rq->signal_link, &signal);
+		}
+
 		/*
-		 * N.B. Since task_asleep() and ttwu are not atomic, the
-		 * waiter may actually go to sleep after the check, causing
-		 * us to suppress a valid wakeup. We prefer to reduce the
-		 * number of false positive missed_breadcrumb() warnings
-		 * at the expense of a few false negatives, as it it easy
-		 * to trigger a false positive under heavy load. Enough
-		 * signal should remain from genuine missed_breadcrumb()
-		 * for us to detect in CI.
+		 * We process the list deletion in bulk, only using a list_add
+		 * (not list_move) above but keeping the status of
+		 * rq->signal_link known with the I915_FENCE_FLAG_SIGNAL bit.
 		 */
-		bool was_asleep = task_asleep(wait->tsk);
+		if (!list_is_first(pos, &ce->signals)) {
+			/* Advance the list to the first incomplete request */
+			__list_del_many(&ce->signals, pos);
+			if (&ce->signals == pos) /* now empty */
+				list_del_init(&ce->signal_link);
+		}
+	}
+
+	spin_unlock(&b->irq_lock);
+
+	list_for_each_safe(pos, next, &signal) {
+		struct i915_request *rq =
+			list_entry(pos, typeof(*rq), signal_link);
 
-		result = ENGINE_WAKEUP_WAITER;
-		if (wake_up_process(wait->tsk) && was_asleep)
-			result |= ENGINE_WAKEUP_ASLEEP;
+		dma_fence_signal(&rq->fence);
+		i915_request_put(rq);
 	}
 
-	return result;
+	return !list_empty(&signal);
 }
 
-unsigned int intel_engine_wakeup(struct intel_engine_cs *engine)
+bool intel_engine_signal_breadcrumbs(struct intel_engine_cs *engine)
 {
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	unsigned long flags;
-	unsigned int result;
+	bool result;
 
-	spin_lock_irqsave(&b->irq_lock, flags);
-	result = __intel_breadcrumbs_wakeup(b);
-	spin_unlock_irqrestore(&b->irq_lock, flags);
+	local_irq_disable();
+	result = intel_engine_breadcrumbs_irq(engine);
+	local_irq_enable();
 
 	return result;
 }
 
+static void signal_irq_work(struct irq_work *work)
+{
+	struct intel_engine_cs *engine =
+		container_of(work, typeof(*engine), breadcrumbs.irq_work);
+
+	intel_engine_breadcrumbs_irq(engine);
+}
+
 static unsigned long wait_timeout(void)
 {
 	return round_jiffies_up(jiffies + DRM_I915_HANGCHECK_JIFFIES);
@@ -94,19 +195,15 @@ static void intel_breadcrumbs_hangcheck(struct timer_list *t)
 	struct intel_engine_cs *engine =
 		from_timer(engine, t, breadcrumbs.hangcheck);
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	unsigned int irq_count;
 
 	if (!b->irq_armed)
 		return;
 
-	irq_count = READ_ONCE(b->irq_count);
-	if (b->hangcheck_interrupts != irq_count) {
-		b->hangcheck_interrupts = irq_count;
-		mod_timer(&b->hangcheck, wait_timeout());
-		return;
-	}
+	if (b->irq_fired)
+		goto rearm;
 
-	/* We keep the hangcheck timer alive until we disarm the irq, even
+	/*
+	 * We keep the hangcheck timer alive until we disarm the irq, even
 	 * if there are no waiters at present.
 	 *
 	 * If the waiter was currently running, assume it hasn't had a chance
@@ -118,10 +215,13 @@ static void intel_breadcrumbs_hangcheck(struct timer_list *t)
 	 * but we still have a waiter. Assuming all batches complete within
 	 * DRM_I915_HANGCHECK_JIFFIES [1.5s]!
 	 */
-	if (intel_engine_wakeup(engine) & ENGINE_WAKEUP_ASLEEP) {
+	synchronize_hardirq(engine->i915->drm.irq);
+	if (intel_engine_signal_breadcrumbs(engine)) {
 		missed_breadcrumb(engine);
 		mod_timer(&b->fake_irq, jiffies + 1);
 	} else {
+rearm:
+		b->irq_fired = false;
 		mod_timer(&b->hangcheck, wait_timeout());
 	}
 }
@@ -140,11 +240,7 @@ static void intel_breadcrumbs_fake_irq(struct timer_list *t)
 	 * oldest waiter to do the coherent seqno check.
 	 */
 
-	spin_lock_irq(&b->irq_lock);
-	if (b->irq_armed && !__intel_breadcrumbs_wakeup(b))
-		__intel_engine_disarm_breadcrumbs(engine);
-	spin_unlock_irq(&b->irq_lock);
-	if (!b->irq_armed)
+	if (!intel_engine_signal_breadcrumbs(engine) && !b->irq_armed)
 		return;
 
 	/* If the user has disabled the fake-irq, restore the hangchecking */
@@ -156,43 +252,6 @@ static void intel_breadcrumbs_fake_irq(struct timer_list *t)
 	mod_timer(&b->fake_irq, jiffies + 1);
 }
 
-static void irq_enable(struct intel_engine_cs *engine)
-{
-	if (!engine->irq_enable)
-		return;
-
-	/* Caller disables interrupts */
-	spin_lock(&engine->i915->irq_lock);
-	engine->irq_enable(engine);
-	spin_unlock(&engine->i915->irq_lock);
-}
-
-static void irq_disable(struct intel_engine_cs *engine)
-{
-	if (!engine->irq_disable)
-		return;
-
-	/* Caller disables interrupts */
-	spin_lock(&engine->i915->irq_lock);
-	engine->irq_disable(engine);
-	spin_unlock(&engine->i915->irq_lock);
-}
-
-void __intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	lockdep_assert_held(&b->irq_lock);
-	GEM_BUG_ON(b->irq_wait);
-	GEM_BUG_ON(!b->irq_armed);
-
-	GEM_BUG_ON(!b->irq_enabled);
-	if (!--b->irq_enabled)
-		irq_disable(engine);
-
-	b->irq_armed = false;
-}
-
 void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
@@ -215,40 +274,6 @@ void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine)
 	spin_unlock_irq(&b->irq_lock);
 }
 
-void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct intel_wait *wait, *n;
-
-	if (!b->irq_armed)
-		return;
-
-	/*
-	 * We only disarm the irq when we are idle (all requests completed),
-	 * so if the bottom-half remains asleep, it missed the request
-	 * completion.
-	 */
-	if (intel_engine_wakeup(engine) & ENGINE_WAKEUP_ASLEEP)
-		missed_breadcrumb(engine);
-
-	spin_lock_irq(&b->rb_lock);
-
-	spin_lock(&b->irq_lock);
-	b->irq_wait = NULL;
-	if (b->irq_armed)
-		__intel_engine_disarm_breadcrumbs(engine);
-	spin_unlock(&b->irq_lock);
-
-	rbtree_postorder_for_each_entry_safe(wait, n, &b->waiters, node) {
-		GEM_BUG_ON(!intel_engine_signaled(engine, wait->seqno));
-		RB_CLEAR_NODE(&wait->node);
-		wake_up_process(wait->tsk);
-	}
-	b->waiters = RB_ROOT;
-
-	spin_unlock_irq(&b->rb_lock);
-}
-
 static bool use_fake_irq(const struct intel_breadcrumbs *b)
 {
 	const struct intel_engine_cs *engine =
@@ -264,7 +289,7 @@ static bool use_fake_irq(const struct intel_breadcrumbs *b)
 	 * engine->seqno_barrier(), a timing error that should be transient
 	 * and unlikely to reoccur.
 	 */
-	return READ_ONCE(b->irq_count) == b->hangcheck_interrupts;
+	return !b->irq_fired;
 }
 
 static void enable_fake_irq(struct intel_breadcrumbs *b)
@@ -276,7 +301,7 @@ static void enable_fake_irq(struct intel_breadcrumbs *b)
 		mod_timer(&b->hangcheck, wait_timeout());
 }
 
-static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
+static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
 {
 	struct intel_engine_cs *engine =
 		container_of(b, struct intel_engine_cs, breadcrumbs);
@@ -315,536 +340,149 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
 	return enabled;
 }
 
-static inline struct intel_wait *to_wait(struct rb_node *node)
+void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
 {
-	return rb_entry(node, struct intel_wait, node);
-}
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
 
-static inline void __intel_breadcrumbs_finish(struct intel_breadcrumbs *b,
-					      struct intel_wait *wait)
-{
-	lockdep_assert_held(&b->rb_lock);
-	GEM_BUG_ON(b->irq_wait == wait);
+	spin_lock_init(&b->irq_lock);
+	INIT_LIST_HEAD(&b->signalers);
 
-	/*
-	 * This request is completed, so remove it from the tree, mark it as
-	 * complete, and *then* wake up the associated task. N.B. when the
-	 * task wakes up, it will find the empty rb_node, discern that it
-	 * has already been removed from the tree and skip the serialisation
-	 * of the b->rb_lock and b->irq_lock. This means that the destruction
-	 * of the intel_wait is not serialised with the interrupt handler
-	 * by the waiter - it must instead be serialised by the caller.
-	 */
-	rb_erase(&wait->node, &b->waiters);
-	RB_CLEAR_NODE(&wait->node);
+	init_irq_work(&b->irq_work, signal_irq_work);
 
-	if (wait->tsk->state != TASK_RUNNING)
-		wake_up_process(wait->tsk); /* implicit smp_wmb() */
+	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
+	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
 }
 
-static inline void __intel_breadcrumbs_next(struct intel_engine_cs *engine,
-					    struct rb_node *next)
+static void cancel_fake_irq(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
 
-	spin_lock(&b->irq_lock);
-	GEM_BUG_ON(!b->irq_armed);
-	GEM_BUG_ON(!b->irq_wait);
-	b->irq_wait = to_wait(next);
-	spin_unlock(&b->irq_lock);
-
-	/* We always wake up the next waiter that takes over as the bottom-half
-	 * as we may delegate not only the irq-seqno barrier to the next waiter
-	 * but also the task of waking up concurrent waiters.
-	 */
-	if (next)
-		wake_up_process(to_wait(next)->tsk);
+	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
+	del_timer_sync(&b->hangcheck);
+	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
 }
 
-static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
-				    struct intel_wait *wait)
+void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct rb_node **p, *parent, *completed;
-	bool first, armed;
-	u32 seqno;
+	unsigned long flags;
 
-	GEM_BUG_ON(!wait->seqno);
+	spin_lock_irqsave(&b->irq_lock, flags);
 
-	/* Insert the request into the retirement ordered list
-	 * of waiters by walking the rbtree. If we are the oldest
-	 * seqno in the tree (the first to be retired), then
-	 * set ourselves as the bottom-half.
-	 *
-	 * As we descend the tree, prune completed branches since we hold the
-	 * spinlock we know that the first_waiter must be delayed and can
-	 * reduce some of the sequential wake up latency if we take action
-	 * ourselves and wake up the completed tasks in parallel. Also, by
-	 * removing stale elements in the tree, we may be able to reduce the
-	 * ping-pong between the old bottom-half and ourselves as first-waiter.
+	/*
+	 * Leave the fake_irq timer enabled (if it is running), but clear the
+	 * bit so that it turns itself off on its next wake up and goes back
+	 * to the long hangcheck interval if still required.
 	 */
-	armed = false;
-	first = true;
-	parent = NULL;
-	completed = NULL;
-	seqno = intel_engine_get_seqno(engine);
-
-	 /* If the request completed before we managed to grab the spinlock,
-	  * return now before adding ourselves to the rbtree. We let the
-	  * current bottom-half handle any pending wakeups and instead
-	  * try and get out of the way quickly.
-	  */
-	if (i915_seqno_passed(seqno, wait->seqno)) {
-		RB_CLEAR_NODE(&wait->node);
-		return first;
-	}
-
-	p = &b->waiters.rb_node;
-	while (*p) {
-		parent = *p;
-		if (wait->seqno == to_wait(parent)->seqno) {
-			/* We have multiple waiters on the same seqno, select
-			 * the highest priority task (that with the smallest
-			 * task->prio) to serve as the bottom-half for this
-			 * group.
-			 */
-			if (wait->tsk->prio > to_wait(parent)->tsk->prio) {
-				p = &parent->rb_right;
-				first = false;
-			} else {
-				p = &parent->rb_left;
-			}
-		} else if (i915_seqno_passed(wait->seqno,
-					     to_wait(parent)->seqno)) {
-			p = &parent->rb_right;
-			if (i915_seqno_passed(seqno, to_wait(parent)->seqno))
-				completed = parent;
-			else
-				first = false;
-		} else {
-			p = &parent->rb_left;
-		}
-	}
-	rb_link_node(&wait->node, parent, p);
-	rb_insert_color(&wait->node, &b->waiters);
-
-	if (first) {
-		spin_lock(&b->irq_lock);
-		b->irq_wait = wait;
-		/* After assigning ourselves as the new bottom-half, we must
-		 * perform a cursory check to prevent a missed interrupt.
-		 * Either we miss the interrupt whilst programming the hardware,
-		 * or if there was a previous waiter (for a later seqno) they
-		 * may be woken instead of us (due to the inherent race
-		 * in the unlocked read of b->irq_seqno_bh in the irq handler)
-		 * and so we miss the wake up.
-		 */
-		armed = __intel_breadcrumbs_enable_irq(b);
-		spin_unlock(&b->irq_lock);
-	}
-
-	if (completed) {
-		/* Advance the bottom-half (b->irq_wait) before we wake up
-		 * the waiters who may scribble over their intel_wait
-		 * just as the interrupt handler is dereferencing it via
-		 * b->irq_wait.
-		 */
-		if (!first) {
-			struct rb_node *next = rb_next(completed);
-			GEM_BUG_ON(next == &wait->node);
-			__intel_breadcrumbs_next(engine, next);
-		}
-
-		do {
-			struct intel_wait *crumb = to_wait(completed);
-			completed = rb_prev(completed);
-			__intel_breadcrumbs_finish(b, crumb);
-		} while (completed);
-	}
-
-	GEM_BUG_ON(!b->irq_wait);
-	GEM_BUG_ON(!b->irq_armed);
-	GEM_BUG_ON(rb_first(&b->waiters) != &b->irq_wait->node);
-
-	return armed;
-}
-
-bool intel_engine_add_wait(struct intel_engine_cs *engine,
-			   struct intel_wait *wait)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	bool armed;
-
-	spin_lock_irq(&b->rb_lock);
-	armed = __intel_engine_add_wait(engine, wait);
-	spin_unlock_irq(&b->rb_lock);
-	if (armed)
-		return armed;
-
-	/* Make the caller recheck if its request has already started. */
-	return intel_engine_has_started(engine, wait->seqno);
-}
-
-static inline bool chain_wakeup(struct rb_node *rb, int priority)
-{
-	return rb && to_wait(rb)->tsk->prio <= priority;
-}
+	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
 
-static inline int wakeup_priority(struct intel_breadcrumbs *b,
-				  struct task_struct *tsk)
-{
-	if (tsk == b->signaler)
-		return INT_MIN;
+	if (b->irq_enabled)
+		irq_enable(engine);
 	else
-		return tsk->prio;
-}
-
-static void __intel_engine_remove_wait(struct intel_engine_cs *engine,
-				       struct intel_wait *wait)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	lockdep_assert_held(&b->rb_lock);
-
-	if (RB_EMPTY_NODE(&wait->node))
-		goto out;
-
-	if (b->irq_wait == wait) {
-		const int priority = wakeup_priority(b, wait->tsk);
-		struct rb_node *next;
-
-		/* We are the current bottom-half. Find the next candidate,
-		 * the first waiter in the queue on the remaining oldest
-		 * request. As multiple seqnos may complete in the time it
-		 * takes us to wake up and find the next waiter, we have to
-		 * wake up that waiter for it to perform its own coherent
-		 * completion check.
-		 */
-		next = rb_next(&wait->node);
-		if (chain_wakeup(next, priority)) {
-			/* If the next waiter is already complete,
-			 * wake it up and continue onto the next waiter. So
-			 * if have a small herd, they will wake up in parallel
-			 * rather than sequentially, which should reduce
-			 * the overall latency in waking all the completed
-			 * clients.
-			 *
-			 * However, waking up a chain adds extra latency to
-			 * the first_waiter. This is undesirable if that
-			 * waiter is a high priority task.
-			 */
-			u32 seqno = intel_engine_get_seqno(engine);
-
-			while (i915_seqno_passed(seqno, to_wait(next)->seqno)) {
-				struct rb_node *n = rb_next(next);
-
-				__intel_breadcrumbs_finish(b, to_wait(next));
-				next = n;
-				if (!chain_wakeup(next, priority))
-					break;
-			}
-		}
-
-		__intel_breadcrumbs_next(engine, next);
-	} else {
-		GEM_BUG_ON(rb_first(&b->waiters) == &wait->node);
-	}
-
-	GEM_BUG_ON(RB_EMPTY_NODE(&wait->node));
-	rb_erase(&wait->node, &b->waiters);
-	RB_CLEAR_NODE(&wait->node);
+		irq_disable(engine);
 
-out:
-	GEM_BUG_ON(b->irq_wait == wait);
-	GEM_BUG_ON(rb_first(&b->waiters) !=
-		   (b->irq_wait ? &b->irq_wait->node : NULL));
+	spin_unlock_irqrestore(&b->irq_lock, flags);
 }
 
-void intel_engine_remove_wait(struct intel_engine_cs *engine,
-			      struct intel_wait *wait)
+void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
 {
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	/* Quick check to see if this waiter was already decoupled from
-	 * the tree by the bottom-half to avoid contention on the spinlock
-	 * by the herd.
-	 */
-	if (RB_EMPTY_NODE(&wait->node)) {
-		GEM_BUG_ON(READ_ONCE(b->irq_wait) == wait);
-		return;
-	}
-
-	spin_lock_irq(&b->rb_lock);
-	__intel_engine_remove_wait(engine, wait);
-	spin_unlock_irq(&b->rb_lock);
+	cancel_fake_irq(engine);
 }
 
-static void signaler_set_rtpriority(void)
+bool i915_request_enable_breadcrumb(struct i915_request *rq)
 {
-	 struct sched_param param = { .sched_priority = 1 };
-
-	 sched_setscheduler_nocheck(current, SCHED_FIFO, &param);
-}
+	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
 
-static int intel_breadcrumbs_signaler(void *arg)
-{
-	struct intel_engine_cs *engine = arg;
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct i915_request *rq, *n;
+	GEM_BUG_ON(test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags));
 
-	/* Install ourselves with high priority to reduce signalling latency */
-	signaler_set_rtpriority();
+	if (!test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags))
+		return true;
 
-	do {
-		bool do_schedule = true;
-		LIST_HEAD(list);
-		u32 seqno;
+	spin_lock(&b->irq_lock);
+	if (test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags) &&
+	    !__request_completed(rq)) {
+		struct intel_context *ce = rq->hw_context;
+		struct list_head *pos;
 
-		set_current_state(TASK_INTERRUPTIBLE);
-		if (list_empty(&b->signals))
-			goto sleep;
+		__intel_breadcrumbs_arm_irq(b);
 
 		/*
-		 * We are either woken up by the interrupt bottom-half,
-		 * or by a client adding a new signaller. In both cases,
-		 * the GPU seqno may have advanced beyond our oldest signal.
-		 * If it has, propagate the signal, remove the waiter and
-		 * check again with the next oldest signal. Otherwise we
-		 * need to wait for a new interrupt from the GPU or for
-		 * a new client.
+		 * We keep the seqno in retirement order, so we can break
+		 * inside intel_engine_breadcrumbs_irq as soon as we've passed
+		 * the last completed request (or seen a request that hasn't
+		 * event started). We could iterate the timeline->requests list,
+		 * but keeping a separate signalers_list has the advantage of
+		 * hopefully being much smaller than the full list and so
+		 * provides faster iteration and detection when there are no
+		 * more interrupts required for this context.
+		 *
+		 * We typically expect to add new signalers in order, so we
+		 * start looking for our insertion point from the tail of
+		 * the list.
 		 */
-		seqno = intel_engine_get_seqno(engine);
-
-		spin_lock_irq(&b->rb_lock);
-		list_for_each_entry_safe(rq, n, &b->signals, signaling.link) {
-			u32 this = rq->signaling.wait.seqno;
+		list_for_each_prev(pos, &ce->signals) {
+			struct i915_request *it =
+				list_entry(pos, typeof(*it), signal_link);
 
-			GEM_BUG_ON(!rq->signaling.wait.seqno);
-
-			if (!i915_seqno_passed(seqno, this))
+			if (i915_seqno_passed(rq->fence.seqno, it->fence.seqno))
 				break;
-
-			if (likely(this == i915_request_global_seqno(rq))) {
-				__intel_engine_remove_wait(engine,
-							   &rq->signaling.wait);
-
-				rq->signaling.wait.seqno = 0;
-				__list_del_entry(&rq->signaling.link);
-
-				if (!i915_request_signaled(rq)) {
-					list_add_tail(&rq->signaling.link,
-						      &list);
-					i915_request_get(rq);
-				}
-			}
 		}
-		spin_unlock_irq(&b->rb_lock);
-
-		if (!list_empty(&list)) {
-			local_bh_disable();
-			list_for_each_entry_safe(rq, n, &list, signaling.link) {
-				dma_fence_signal(&rq->fence);
-				GEM_BUG_ON(!i915_request_completed(rq));
-				i915_request_put(rq);
-			}
-			local_bh_enable(); /* kick start the tasklets */
-
-			/*
-			 * If the engine is saturated we may be continually
-			 * processing completed requests. This angers the
-			 * NMI watchdog if we never let anything else
-			 * have access to the CPU. Let's pretend to be nice
-			 * and relinquish the CPU if we burn through the
-			 * entire RT timeslice!
-			 */
-			do_schedule = need_resched();
-		}
-
-		if (unlikely(do_schedule)) {
-sleep:
-			if (kthread_should_park())
-				kthread_parkme();
-
-			if (unlikely(kthread_should_stop()))
-				break;
-
-			schedule();
-		}
-	} while (1);
-	__set_current_state(TASK_RUNNING);
-
-	return 0;
-}
+		list_add(&rq->signal_link, pos);
+		if (pos == &ce->signals) /* catch transitions from empty list */
+			list_move_tail(&ce->signal_link, &b->signalers);
 
-static void insert_signal(struct intel_breadcrumbs *b,
-			  struct i915_request *request,
-			  const u32 seqno)
-{
-	struct i915_request *iter;
-
-	lockdep_assert_held(&b->rb_lock);
-
-	/*
-	 * A reasonable assumption is that we are called to add signals
-	 * in sequence, as the requests are submitted for execution and
-	 * assigned a global_seqno. This will be the case for the majority
-	 * of internally generated signals (inter-engine signaling).
-	 *
-	 * Out of order waiters triggering random signaling enabling will
-	 * be more problematic, but hopefully rare enough and the list
-	 * small enough that the O(N) insertion sort is not an issue.
-	 */
-
-	list_for_each_entry_reverse(iter, &b->signals, signaling.link)
-		if (i915_seqno_passed(seqno, iter->signaling.wait.seqno))
-			break;
-
-	list_add(&request->signaling.link, &iter->signaling.link);
-}
-
-bool intel_engine_enable_signaling(struct i915_request *request, bool wakeup)
-{
-	struct intel_engine_cs *engine = request->engine;
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct intel_wait *wait = &request->signaling.wait;
-	u32 seqno;
-
-	/*
-	 * Note that we may be called from an interrupt handler on another
-	 * device (e.g. nouveau signaling a fence completion causing us
-	 * to submit a request, and so enable signaling). As such,
-	 * we need to make sure that all other users of b->rb_lock protect
-	 * against interrupts, i.e. use spin_lock_irqsave.
-	 */
-
-	/* locked by dma_fence_enable_sw_signaling() (irqsafe fence->lock) */
-	GEM_BUG_ON(!irqs_disabled());
-	lockdep_assert_held(&request->lock);
-
-	seqno = i915_request_global_seqno(request);
-	if (!seqno) /* will be enabled later upon execution */
-		return true;
-
-	GEM_BUG_ON(wait->seqno);
-	wait->tsk = b->signaler;
-	wait->request = request;
-	wait->seqno = seqno;
-
-	/*
-	 * Add ourselves into the list of waiters, but registering our
-	 * bottom-half as the signaller thread. As per usual, only the oldest
-	 * waiter (not just signaller) is tasked as the bottom-half waking
-	 * up all completed waiters after the user interrupt.
-	 *
-	 * If we are the oldest waiter, enable the irq (after which we
-	 * must double check that the seqno did not complete).
-	 */
-	spin_lock(&b->rb_lock);
-	insert_signal(b, request, seqno);
-	wakeup &= __intel_engine_add_wait(engine, wait);
-	spin_unlock(&b->rb_lock);
-
-	if (wakeup) {
-		wake_up_process(b->signaler);
-		return !intel_wait_complete(wait);
+		set_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
 	}
+	spin_unlock(&b->irq_lock);
 
-	return true;
+	return !__request_completed(rq);
 }
 
-void intel_engine_cancel_signaling(struct i915_request *request)
+void i915_request_cancel_breadcrumb(struct i915_request *rq)
 {
-	struct intel_engine_cs *engine = request->engine;
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
 
-	GEM_BUG_ON(!irqs_disabled());
-	lockdep_assert_held(&request->lock);
-
-	if (!READ_ONCE(request->signaling.wait.seqno))
+	if (!test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags))
 		return;
 
-	spin_lock(&b->rb_lock);
-	__intel_engine_remove_wait(engine, &request->signaling.wait);
-	if (fetch_and_zero(&request->signaling.wait.seqno))
-		__list_del_entry(&request->signaling.link);
-	spin_unlock(&b->rb_lock);
-}
-
-int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct task_struct *tsk;
-
-	spin_lock_init(&b->rb_lock);
-	spin_lock_init(&b->irq_lock);
-
-	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
-	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
-
-	INIT_LIST_HEAD(&b->signals);
-
-	/* Spawn a thread to provide a common bottom-half for all signals.
-	 * As this is an asynchronous interface we cannot steal the current
-	 * task for handling the bottom-half to the user interrupt, therefore
-	 * we create a thread to do the coherent seqno dance after the
-	 * interrupt and then signal the waitqueue (via the dma-buf/fence).
-	 */
-	tsk = kthread_run(intel_breadcrumbs_signaler, engine,
-			  "i915/signal:%d", engine->id);
-	if (IS_ERR(tsk))
-		return PTR_ERR(tsk);
-
-	b->signaler = tsk;
-
-	return 0;
-}
+	spin_lock(&b->irq_lock);
+	if (test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags)) {
+		struct intel_context *ce = rq->hw_context;
 
-static void cancel_fake_irq(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+		list_del(&rq->signal_link);
+		if (list_empty(&ce->signals))
+			list_del_init(&ce->signal_link);
 
-	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
-	del_timer_sync(&b->hangcheck);
-	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
+		clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
+	}
+	spin_unlock(&b->irq_lock);
 }
 
-void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
+void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
+				    struct drm_printer *p)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	unsigned long flags;
+	struct intel_context *ce;
+	struct i915_request *rq;
 
-	spin_lock_irqsave(&b->irq_lock, flags);
-
-	/*
-	 * Leave the fake_irq timer enabled (if it is running), but clear the
-	 * bit so that it turns itself off on its next wake up and goes back
-	 * to the long hangcheck interval if still required.
-	 */
-	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
-
-	if (b->irq_enabled)
-		irq_enable(engine);
-	else
-		irq_disable(engine);
-
-	spin_unlock_irqrestore(&b->irq_lock, flags);
-}
-
-void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	if (list_empty(&b->signalers))
+		return;
 
-	/* The engines should be idle and all requests accounted for! */
-	WARN_ON(READ_ONCE(b->irq_wait));
-	WARN_ON(!RB_EMPTY_ROOT(&b->waiters));
-	WARN_ON(!list_empty(&b->signals));
+	drm_printf(p, "Signals:\n");
 
-	if (!IS_ERR_OR_NULL(b->signaler))
-		kthread_stop(b->signaler);
+	spin_lock_irq(&b->irq_lock);
+	list_for_each_entry(ce, &b->signalers, signal_link) {
+		list_for_each_entry(rq, &ce->signals, signal_link) {
+			drm_printf(p, "\t[%llx:%llx%s] @ %dms\n",
+				   rq->fence.context, rq->fence.seqno,
+				   i915_request_completed(rq) ? "!" :
+				   i915_request_started(rq) ? "*" :
+				   "",
+				   jiffies_to_msecs(jiffies - rq->emitted_jiffies));
+		}
+	}
+	spin_unlock_irq(&b->irq_lock);
 
-	cancel_fake_irq(engine);
+	if (test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
+		drm_printf(p, "Fake irq active\n");
 }
-
-#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
-#include "selftests/intel_breadcrumbs.c"
-#endif
diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index b35ba4f03079..d093e4552a4c 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -458,12 +458,6 @@ int intel_engines_init(struct drm_i915_private *dev_priv)
 void intel_engine_write_global_seqno(struct intel_engine_cs *engine, u32 seqno)
 {
 	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
-
-	/* After manually advancing the seqno, fake the interrupt in case
-	 * there are any waiters for that seqno.
-	 */
-	intel_engine_wakeup(engine);
-
 	GEM_BUG_ON(intel_engine_get_seqno(engine) != seqno);
 }
 
@@ -607,6 +601,7 @@ int intel_engine_setup_common(struct intel_engine_cs *engine)
 
 	i915_timeline_set_subclass(&engine->timeline, TIMELINE_ENGINE);
 
+	intel_engine_init_breadcrumbs(engine);
 	intel_engine_init_execlist(engine);
 	intel_engine_init_hangcheck(engine);
 	intel_engine_init_batch_pool(engine);
@@ -717,20 +712,14 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
 		}
 	}
 
-	ret = intel_engine_init_breadcrumbs(engine);
-	if (ret)
-		goto err_unpin_preempt;
-
 	ret = measure_breadcrumb_dw(engine);
 	if (ret < 0)
-		goto err_breadcrumbs;
+		goto err_unpin_preempt;
 
 	engine->emit_fini_breadcrumb_dw = ret;
 
 	return 0;
 
-err_breadcrumbs:
-	intel_engine_fini_breadcrumbs(engine);
 err_unpin_preempt:
 	if (i915->preempt_context)
 		__intel_context_unpin(i915->preempt_context, engine);
@@ -1294,12 +1283,14 @@ static void print_request(struct drm_printer *m,
 
 	x = print_sched_attr(rq->i915, &rq->sched.attr, buf, x, sizeof(buf));
 
-	drm_printf(m, "%s%x%s [%llx:%llx]%s @ %dms: %s\n",
+	drm_printf(m, "%s%x%s%s [%llx:%llx]%s @ %dms: %s\n",
 		   prefix,
 		   rq->global_seqno,
 		   i915_request_completed(rq) ? "!" :
 		   i915_request_started(rq) ? "*" :
 		   "",
+		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
+			    &rq->fence.flags) ?  "+" : "",
 		   rq->fence.context, rq->fence.seqno,
 		   buf,
 		   jiffies_to_msecs(jiffies - rq->emitted_jiffies),
@@ -1492,12 +1483,9 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 		       struct drm_printer *m,
 		       const char *header, ...)
 {
-	struct intel_breadcrumbs * const b = &engine->breadcrumbs;
 	struct i915_gpu_error * const error = &engine->i915->gpu_error;
 	struct i915_request *rq;
 	intel_wakeref_t wakeref;
-	unsigned long flags;
-	struct rb_node *rb;
 
 	if (header) {
 		va_list ap;
@@ -1565,21 +1553,12 @@ void intel_engine_dump(struct intel_engine_cs *engine,
 
 	intel_execlists_show_requests(engine, m, print_request, 8);
 
-	spin_lock_irqsave(&b->rb_lock, flags);
-	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
-		struct intel_wait *w = rb_entry(rb, typeof(*w), node);
-
-		drm_printf(m, "\t%s [%d:%c] waiting for %x\n",
-			   w->tsk->comm, w->tsk->pid,
-			   task_state_to_char(w->tsk),
-			   w->seqno);
-	}
-	spin_unlock_irqrestore(&b->rb_lock, flags);
-
 	drm_printf(m, "HWSP:\n");
 	hexdump(m, engine->status_page.addr, PAGE_SIZE);
 
 	drm_printf(m, "Idle? %s\n", yesno(intel_engine_is_idle(engine)));
+
+	intel_engine_print_breadcrumbs(engine, m);
 }
 
 static u8 user_class_map[] = {
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 3f9f0a5a0f44..d7c31ad1d0ea 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -486,8 +486,8 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
 
 	for (i = 0; i < GEN7_XCS_WA; i++) {
 		*cs++ = MI_STORE_DWORD_INDEX;
-		*cs++ = I915_GEM_HWS_INDEX_ADDR;
-		*cs++ = rq->global_seqno;
+		*cs++ = I915_GEM_HWS_SEQNO_ADDR;
+		*cs++ = rq->fence.seqno;
 	}
 
 	*cs++ = MI_FLUSH_DW;
@@ -738,7 +738,7 @@ static int init_ring_common(struct intel_engine_cs *engine)
 	}
 
 	/* Papering over lost _interrupts_ immediately following the restart */
-	intel_engine_wakeup(engine);
+	intel_engine_queue_breadcrumbs(engine);
 out:
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 5ed06d4230f6..eb0f8667a246 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -5,6 +5,7 @@
 #include <drm/drm_util.h>
 
 #include <linux/hashtable.h>
+#include <linux/irq_work.h>
 #include <linux/seqlock.h>
 
 #include "i915_gem_batch_pool.h"
@@ -376,22 +377,19 @@ struct intel_engine_cs {
 	 * the overhead of waking that client is much preferred.
 	 */
 	struct intel_breadcrumbs {
-		spinlock_t irq_lock; /* protects irq_*; irqsafe */
-		struct intel_wait *irq_wait; /* oldest waiter by retirement */
+		spinlock_t irq_lock;
+		struct list_head signalers;
 
-		spinlock_t rb_lock; /* protects the rb and wraps irq_lock */
-		struct rb_root waiters; /* sorted by retirement, priority */
-		struct list_head signals; /* sorted by retirement */
-		struct task_struct *signaler; /* used for fence signalling */
+		struct irq_work irq_work; /* for use from inside irq_lock */
 
 		struct timer_list fake_irq; /* used after a missed interrupt */
 		struct timer_list hangcheck; /* detect missed interrupts */
 
 		unsigned int hangcheck_interrupts;
 		unsigned int irq_enabled;
-		unsigned int irq_count;
 
-		bool irq_armed : 1;
+		bool irq_armed;
+		bool irq_fired;
 	} breadcrumbs;
 
 	struct {
@@ -881,83 +879,29 @@ static inline bool intel_engine_has_started(struct intel_engine_cs *engine,
 void intel_engine_get_instdone(struct intel_engine_cs *engine,
 			       struct intel_instdone *instdone);
 
-/* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
-int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
-
-static inline void intel_wait_init(struct intel_wait *wait)
-{
-	wait->tsk = current;
-	wait->request = NULL;
-}
-
-static inline void intel_wait_init_for_seqno(struct intel_wait *wait, u32 seqno)
-{
-	wait->tsk = current;
-	wait->seqno = seqno;
-}
-
-static inline bool intel_wait_has_seqno(const struct intel_wait *wait)
-{
-	return wait->seqno;
-}
-
-static inline bool
-intel_wait_update_seqno(struct intel_wait *wait, u32 seqno)
-{
-	wait->seqno = seqno;
-	return intel_wait_has_seqno(wait);
-}
-
-static inline bool
-intel_wait_update_request(struct intel_wait *wait,
-			  const struct i915_request *rq)
-{
-	return intel_wait_update_seqno(wait, i915_request_global_seqno(rq));
-}
-
-static inline bool
-intel_wait_check_seqno(const struct intel_wait *wait, u32 seqno)
-{
-	return wait->seqno == seqno;
-}
-
-static inline bool
-intel_wait_check_request(const struct intel_wait *wait,
-			 const struct i915_request *rq)
-{
-	return intel_wait_check_seqno(wait, i915_request_global_seqno(rq));
-}
+void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
+void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
 
-static inline bool intel_wait_complete(const struct intel_wait *wait)
-{
-	return RB_EMPTY_NODE(&wait->node);
-}
+void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine);
+void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine);
 
-bool intel_engine_add_wait(struct intel_engine_cs *engine,
-			   struct intel_wait *wait);
-void intel_engine_remove_wait(struct intel_engine_cs *engine,
-			      struct intel_wait *wait);
-bool intel_engine_enable_signaling(struct i915_request *request, bool wakeup);
-void intel_engine_cancel_signaling(struct i915_request *request);
+bool intel_engine_signal_breadcrumbs(struct intel_engine_cs *engine);
+void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
 
-static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
+static inline void
+intel_engine_queue_breadcrumbs(struct intel_engine_cs *engine)
 {
-	return READ_ONCE(engine->breadcrumbs.irq_wait);
+	irq_work_queue(&engine->breadcrumbs.irq_work);
 }
 
-unsigned int intel_engine_wakeup(struct intel_engine_cs *engine);
-#define ENGINE_WAKEUP_WAITER BIT(0)
-#define ENGINE_WAKEUP_ASLEEP BIT(1)
-
-void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine);
-void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine);
-
-void __intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
-void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
+bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine);
 
 void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine);
 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
 
+void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
+				    struct drm_printer *p);
+
 static inline u32 *gen8_emit_pipe_control(u32 *batch, u32 flags, u32 offset)
 {
 	memset(batch, 0, 6 * sizeof(u32));
diff --git a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
index 4a83a1c6c406..88e5ab586337 100644
--- a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
+++ b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
@@ -15,7 +15,6 @@ selftest(scatterlist, scatterlist_mock_selftests)
 selftest(syncmap, i915_syncmap_mock_selftests)
 selftest(uncore, intel_uncore_mock_selftests)
 selftest(engine, intel_engine_cs_mock_selftests)
-selftest(breadcrumbs, intel_breadcrumbs_mock_selftests)
 selftest(timelines, i915_timeline_mock_selftests)
 selftest(requests, i915_request_mock_selftests)
 selftest(objects, i915_gem_object_mock_selftests)
diff --git a/drivers/gpu/drm/i915/selftests/i915_request.c b/drivers/gpu/drm/i915/selftests/i915_request.c
index 4d4b86b5fa11..5e4702a55688 100644
--- a/drivers/gpu/drm/i915/selftests/i915_request.c
+++ b/drivers/gpu/drm/i915/selftests/i915_request.c
@@ -25,9 +25,12 @@
 #include <linux/prime_numbers.h>
 
 #include "../i915_selftest.h"
+#include "i915_random.h"
 #include "igt_live_test.h"
+#include "lib_sw_fence.h"
 
 #include "mock_context.h"
+#include "mock_drm.h"
 #include "mock_gem_device.h"
 
 static int igt_add_request(void *arg)
@@ -247,6 +250,253 @@ static int igt_request_rewind(void *arg)
 	return err;
 }
 
+struct smoketest {
+	struct intel_engine_cs *engine;
+	struct i915_gem_context **contexts;
+	unsigned int ncontexts, max_batch;
+	atomic_long_t num_waits, num_fences;
+	struct i915_request *(*request_alloc)(struct i915_gem_context *,
+					      struct intel_engine_cs *);
+};
+
+static struct i915_request *
+__mock_request_alloc(struct i915_gem_context *ctx,
+		     struct intel_engine_cs *engine)
+{
+	return mock_request(engine, ctx, 0);
+}
+
+static struct i915_request *
+__live_request_alloc(struct i915_gem_context *ctx,
+		     struct intel_engine_cs *engine)
+{
+	return i915_request_alloc(engine, ctx);
+}
+
+static int __igt_breadcrumbs_smoketest(void *arg)
+{
+	struct smoketest *t = arg;
+	struct mutex * const BKL = &t->engine->i915->drm.struct_mutex;
+	const unsigned int max_batch = min(t->ncontexts, t->max_batch) - 1;
+	const unsigned int total = 4 * t->ncontexts + 1;
+	unsigned int num_waits = 0, num_fences = 0;
+	struct i915_request **requests;
+	I915_RND_STATE(prng);
+	unsigned int *order;
+	int err = 0;
+
+	/*
+	 * A very simple test to catch the most egregious of list handling bugs.
+	 *
+	 * At its heart, we simple create oodles of requests running across
+	 * multiple kthreads and enable signaling on them, for the sole purpose
+	 * of stressing our breadcrumb handling. The only inspection we do is
+	 * that the fences were marked as signaled.
+	 */
+
+	requests = kmalloc_array(total, sizeof(*requests), GFP_KERNEL);
+	if (!requests)
+		return -ENOMEM;
+
+	order = i915_random_order(total, &prng);
+	if (!order) {
+		err = -ENOMEM;
+		goto out_requests;
+	}
+
+	while (!kthread_should_stop()) {
+		struct i915_sw_fence *submit, *wait;
+		unsigned int n, count;
+
+		submit = heap_fence_create(GFP_KERNEL);
+		if (!submit) {
+			err = -ENOMEM;
+			break;
+		}
+
+		wait = heap_fence_create(GFP_KERNEL);
+		if (!wait) {
+			i915_sw_fence_commit(submit);
+			heap_fence_put(submit);
+			err = ENOMEM;
+			break;
+		}
+
+		i915_random_reorder(order, total, &prng);
+		count = 1 + i915_prandom_u32_max_state(max_batch, &prng);
+
+		for (n = 0; n < count; n++) {
+			struct i915_gem_context *ctx =
+				t->contexts[order[n] % t->ncontexts];
+			struct i915_request *rq;
+
+			mutex_lock(BKL);
+
+			rq = t->request_alloc(ctx, t->engine);
+			if (IS_ERR(rq)) {
+				mutex_unlock(BKL);
+				err = PTR_ERR(rq);
+				count = n;
+				break;
+			}
+
+			err = i915_sw_fence_await_sw_fence_gfp(&rq->submit,
+							       submit,
+							       GFP_KERNEL);
+
+			requests[n] = i915_request_get(rq);
+			i915_request_add(rq);
+
+			mutex_unlock(BKL);
+
+			if (err >= 0)
+				err = i915_sw_fence_await_dma_fence(wait,
+								    &rq->fence,
+								    0,
+								    GFP_KERNEL);
+			if (err < 0) {
+				i915_request_put(rq);
+				count = n;
+				break;
+			}
+		}
+
+		i915_sw_fence_commit(submit);
+		i915_sw_fence_commit(wait);
+
+		if (!wait_event_timeout(wait->wait,
+					i915_sw_fence_done(wait),
+					HZ / 2)) {
+			struct i915_request *rq = requests[count - 1];
+
+			pr_err("waiting for %d fences (last %llx:%lld) on %s timed out!\n",
+			       count,
+			       rq->fence.context, rq->fence.seqno,
+			       t->engine->name);
+			i915_gem_set_wedged(t->engine->i915);
+			GEM_BUG_ON(!i915_request_completed(rq));
+			i915_sw_fence_wait(wait);
+			err = -EIO;
+		}
+
+		for (n = 0; n < count; n++) {
+			struct i915_request *rq = requests[n];
+
+			if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+				      &rq->fence.flags)) {
+				pr_err("%llu:%llu was not signaled!\n",
+				       rq->fence.context, rq->fence.seqno);
+				err = -EINVAL;
+			}
+
+			i915_request_put(rq);
+		}
+
+		heap_fence_put(wait);
+		heap_fence_put(submit);
+
+		if (err < 0)
+			break;
+
+		num_fences += count;
+		num_waits++;
+
+		cond_resched();
+	}
+
+	atomic_long_add(num_fences, &t->num_fences);
+	atomic_long_add(num_waits, &t->num_waits);
+
+	kfree(order);
+out_requests:
+	kfree(requests);
+	return err;
+}
+
+static int mock_breadcrumbs_smoketest(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct smoketest t = {
+		.engine = i915->engine[RCS],
+		.ncontexts = 1024,
+		.max_batch = 1024,
+		.request_alloc = __mock_request_alloc
+	};
+	unsigned int ncpus = num_online_cpus();
+	struct task_struct **threads;
+	unsigned int n;
+	int ret = 0;
+
+	/*
+	 * Smoketest our breadcrumb/signal handling for requests across multiple
+	 * threads. A very simple test to only catch the most egregious of bugs.
+	 * See __igt_breadcrumbs_smoketest();
+	 */
+
+	threads = kmalloc_array(ncpus, sizeof(*threads), GFP_KERNEL);
+	if (!threads)
+		return -ENOMEM;
+
+	t.contexts =
+		kmalloc_array(t.ncontexts, sizeof(*t.contexts), GFP_KERNEL);
+	if (!t.contexts) {
+		ret = -ENOMEM;
+		goto out_threads;
+	}
+
+	mutex_lock(&t.engine->i915->drm.struct_mutex);
+	for (n = 0; n < t.ncontexts; n++) {
+		t.contexts[n] = mock_context(t.engine->i915, "mock");
+		if (!t.contexts[n]) {
+			ret = -ENOMEM;
+			goto out_contexts;
+		}
+	}
+
+	for (n = 0; n < ncpus; n++) {
+		threads[n] = kthread_run(__igt_breadcrumbs_smoketest,
+					 &t, "igt/%d", n);
+		if (IS_ERR(threads[n])) {
+			ret = PTR_ERR(threads[n]);
+			ncpus = n;
+			break;
+		}
+
+		get_task_struct(threads[n]);
+	}
+	mutex_unlock(&t.engine->i915->drm.struct_mutex);
+
+	msleep(jiffies_to_msecs(i915_selftest.timeout_jiffies));
+
+	for (n = 0; n < ncpus; n++) {
+		int err;
+
+		err = kthread_stop(threads[n]);
+		if (err < 0 && !ret)
+			ret = err;
+
+		put_task_struct(threads[n]);
+	}
+	pr_info("Completed %lu waits for %lu fence across %d cpus\n",
+		atomic_long_read(&t.num_waits),
+		atomic_long_read(&t.num_fences),
+		ncpus);
+
+	mutex_lock(&t.engine->i915->drm.struct_mutex);
+out_contexts:
+	for (n = 0; n < t.ncontexts; n++) {
+		if (!t.contexts[n])
+			break;
+		mock_context_close(t.contexts[n]);
+	}
+	mutex_unlock(&t.engine->i915->drm.struct_mutex);
+	kfree(t.contexts);
+out_threads:
+	kfree(threads);
+
+	return ret;
+}
+
 int i915_request_mock_selftests(void)
 {
 	static const struct i915_subtest tests[] = {
@@ -254,6 +504,7 @@ int i915_request_mock_selftests(void)
 		SUBTEST(igt_wait_request),
 		SUBTEST(igt_fence_wait),
 		SUBTEST(igt_request_rewind),
+		SUBTEST(mock_breadcrumbs_smoketest),
 	};
 	struct drm_i915_private *i915;
 	intel_wakeref_t wakeref;
@@ -812,6 +1063,174 @@ static int live_sequential_engines(void *arg)
 	return err;
 }
 
+static int
+max_batches(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
+{
+	struct i915_request *rq;
+	int ret;
+
+	/*
+	 * Before execlists, all contexts share the same ringbuffer. With
+	 * execlists, each context/engine has a separate ringbuffer and
+	 * for the purposes of this test, inexhaustible.
+	 *
+	 * For the global ringbuffer though, we have to be very careful
+	 * that we do not wrap while preventing the execution of requests
+	 * with a unsignaled fence.
+	 */
+	if (HAS_EXECLISTS(ctx->i915))
+		return INT_MAX;
+
+	rq = i915_request_alloc(engine, ctx);
+	if (IS_ERR(rq)) {
+		ret = PTR_ERR(rq);
+	} else {
+		int sz;
+
+		ret = rq->ring->size - rq->reserved_space;
+		i915_request_add(rq);
+
+		sz = rq->ring->emit - rq->head;
+		if (sz < 0)
+			sz += rq->ring->size;
+		ret /= sz;
+		ret /= 2; /* leave half spare, in case of emergency! */
+
+		/* One ring interleaved between requests from all cpus */
+		ret /= num_online_cpus() + 1;
+	}
+
+	return ret;
+}
+
+static int live_breadcrumbs_smoketest(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct smoketest t[I915_NUM_ENGINES];
+	unsigned int ncpus = num_online_cpus();
+	unsigned long num_waits, num_fences;
+	struct intel_engine_cs *engine;
+	struct task_struct **threads;
+	struct igt_live_test live;
+	enum intel_engine_id id;
+	intel_wakeref_t wakeref;
+	struct drm_file *file;
+	unsigned int n;
+	int ret = 0;
+
+	/*
+	 * Smoketest our breadcrumb/signal handling for requests across multiple
+	 * threads. A very simple test to only catch the most egregious of bugs.
+	 * See __igt_breadcrumbs_smoketest();
+	 *
+	 * On real hardware this time.
+	 */
+
+	wakeref = intel_runtime_pm_get(i915);
+
+	file = mock_file(i915);
+	if (IS_ERR(file)) {
+		ret = PTR_ERR(file);
+		goto out_rpm;
+	}
+
+	threads = kcalloc(ncpus * I915_NUM_ENGINES,
+			  sizeof(*threads),
+			  GFP_KERNEL);
+	if (!threads)
+		return -ENOMEM;
+
+	memset(&t[0], 0, sizeof(t[0]));
+	t[0].request_alloc = __live_request_alloc;
+	t[0].ncontexts = 64;
+	t[0].contexts = kmalloc_array(t[0].ncontexts,
+				      sizeof(*t[0].contexts),
+				      GFP_KERNEL);
+	if (!t[0].contexts) {
+		ret = -ENOMEM;
+		goto out_threads;
+	}
+
+	mutex_lock(&i915->drm.struct_mutex);
+	for (n = 0; n < t[0].ncontexts; n++) {
+		t[0].contexts[n] = live_context(i915, file);
+		if (!t[0].contexts[n]) {
+			ret = -ENOMEM;
+			goto out_contexts;
+		}
+	}
+
+	ret = igt_live_test_begin(&live, i915, __func__, "");
+	if (ret)
+		goto out_contexts;
+
+	for_each_engine(engine, i915, id) {
+		t[id] = t[0];
+		t[id].engine = engine;
+		t[id].max_batch = max_batches(t[0].contexts[0], engine);
+		if (t[id].max_batch < 0) {
+			ret = t[id].max_batch;
+			goto out_flush;
+		}
+		pr_debug("Limiting batches to %d requests on %s\n",
+			 t[id].max_batch, engine->name);
+
+		for (n = 0; n < ncpus; n++) {
+			struct task_struct *tsk;
+
+			tsk = kthread_run(__igt_breadcrumbs_smoketest,
+					  &t[id], "igt/%d.%d", id, n);
+			if (IS_ERR(tsk)) {
+				ret = PTR_ERR(tsk);
+				goto out_flush;
+			}
+
+			get_task_struct(tsk);
+			threads[id * ncpus + n] = tsk;
+		}
+	}
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	msleep(jiffies_to_msecs(i915_selftest.timeout_jiffies));
+
+out_flush:
+	num_waits = 0;
+	num_fences = 0;
+	for_each_engine(engine, i915, id) {
+		for (n = 0; n < ncpus; n++) {
+			struct task_struct *tsk = threads[id * ncpus + n];
+			int err;
+
+			if (!tsk)
+				continue;
+
+			err = kthread_stop(tsk);
+			if (err < 0 && !ret)
+				ret = err;
+
+			put_task_struct(tsk);
+		}
+
+		num_waits += atomic_long_read(&t[id].num_waits);
+		num_fences += atomic_long_read(&t[id].num_fences);
+	}
+	pr_info("Completed %lu waits for %lu fence across %d engines and %d cpus\n",
+		num_waits, num_fences, RUNTIME_INFO(i915)->num_rings, ncpus);
+
+	mutex_lock(&i915->drm.struct_mutex);
+	ret = igt_live_test_end(&live) ?: ret;
+out_contexts:
+	mutex_unlock(&i915->drm.struct_mutex);
+	kfree(t[0].contexts);
+out_threads:
+	kfree(threads);
+	mock_file_free(i915, file);
+out_rpm:
+	intel_runtime_pm_put(i915, wakeref);
+
+	return ret;
+}
+
 int i915_request_live_selftests(struct drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
@@ -819,6 +1238,7 @@ int i915_request_live_selftests(struct drm_i915_private *i915)
 		SUBTEST(live_all_engines),
 		SUBTEST(live_sequential_engines),
 		SUBTEST(live_empty_request),
+		SUBTEST(live_breadcrumbs_smoketest),
 	};
 
 	if (i915_terminally_wedged(&i915->gpu_error))
diff --git a/drivers/gpu/drm/i915/selftests/igt_spinner.c b/drivers/gpu/drm/i915/selftests/igt_spinner.c
index 0e70df0230b8..9ebd9225684e 100644
--- a/drivers/gpu/drm/i915/selftests/igt_spinner.c
+++ b/drivers/gpu/drm/i915/selftests/igt_spinner.c
@@ -185,11 +185,6 @@ void igt_spinner_fini(struct igt_spinner *spin)
 
 bool igt_wait_for_spinner(struct igt_spinner *spin, struct i915_request *rq)
 {
-	if (!wait_event_timeout(rq->execute,
-				READ_ONCE(rq->global_seqno),
-				msecs_to_jiffies(10)))
-		return false;
-
 	return !(wait_for_us(i915_seqno_passed(hws_seqno(spin, rq),
 					       rq->fence.seqno),
 			     10) &&
diff --git a/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c b/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
deleted file mode 100644
index f03b407fdbe2..000000000000
--- a/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
+++ /dev/null
@@ -1,470 +0,0 @@
-/*
- * Copyright © 2016 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
- *
- */
-
-#include "../i915_selftest.h"
-#include "i915_random.h"
-
-#include "mock_gem_device.h"
-#include "mock_engine.h"
-
-static int check_rbtree(struct intel_engine_cs *engine,
-			const unsigned long *bitmap,
-			const struct intel_wait *waiters,
-			const int count)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct rb_node *rb;
-	int n;
-
-	if (&b->irq_wait->node != rb_first(&b->waiters)) {
-		pr_err("First waiter does not match first element of wait-tree\n");
-		return -EINVAL;
-	}
-
-	n = find_first_bit(bitmap, count);
-	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
-		struct intel_wait *w = container_of(rb, typeof(*w), node);
-		int idx = w - waiters;
-
-		if (!test_bit(idx, bitmap)) {
-			pr_err("waiter[%d, seqno=%d] removed but still in wait-tree\n",
-			       idx, w->seqno);
-			return -EINVAL;
-		}
-
-		if (n != idx) {
-			pr_err("waiter[%d, seqno=%d] does not match expected next element in tree [%d]\n",
-			       idx, w->seqno, n);
-			return -EINVAL;
-		}
-
-		n = find_next_bit(bitmap, count, n + 1);
-	}
-
-	return 0;
-}
-
-static int check_completion(struct intel_engine_cs *engine,
-			    const unsigned long *bitmap,
-			    const struct intel_wait *waiters,
-			    const int count)
-{
-	int n;
-
-	for (n = 0; n < count; n++) {
-		if (intel_wait_complete(&waiters[n]) != !!test_bit(n, bitmap))
-			continue;
-
-		pr_err("waiter[%d, seqno=%d] is %s, but expected %s\n",
-		       n, waiters[n].seqno,
-		       intel_wait_complete(&waiters[n]) ? "complete" : "active",
-		       test_bit(n, bitmap) ? "active" : "complete");
-		return -EINVAL;
-	}
-
-	return 0;
-}
-
-static int check_rbtree_empty(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	if (b->irq_wait) {
-		pr_err("Empty breadcrumbs still has a waiter\n");
-		return -EINVAL;
-	}
-
-	if (!RB_EMPTY_ROOT(&b->waiters)) {
-		pr_err("Empty breadcrumbs, but wait-tree not empty\n");
-		return -EINVAL;
-	}
-
-	return 0;
-}
-
-static int igt_random_insert_remove(void *arg)
-{
-	const u32 seqno_bias = 0x1000;
-	I915_RND_STATE(prng);
-	struct intel_engine_cs *engine = arg;
-	struct intel_wait *waiters;
-	const int count = 4096;
-	unsigned int *order;
-	unsigned long *bitmap;
-	int err = -ENOMEM;
-	int n;
-
-	mock_engine_reset(engine);
-
-	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
-	if (!waiters)
-		goto out_engines;
-
-	bitmap = kcalloc(DIV_ROUND_UP(count, BITS_PER_LONG), sizeof(*bitmap),
-			 GFP_KERNEL);
-	if (!bitmap)
-		goto out_waiters;
-
-	order = i915_random_order(count, &prng);
-	if (!order)
-		goto out_bitmap;
-
-	for (n = 0; n < count; n++)
-		intel_wait_init_for_seqno(&waiters[n], seqno_bias + n);
-
-	err = check_rbtree(engine, bitmap, waiters, count);
-	if (err)
-		goto out_order;
-
-	/* Add and remove waiters into the rbtree in random order. At each
-	 * step, we verify that the rbtree is correctly ordered.
-	 */
-	for (n = 0; n < count; n++) {
-		int i = order[n];
-
-		intel_engine_add_wait(engine, &waiters[i]);
-		__set_bit(i, bitmap);
-
-		err = check_rbtree(engine, bitmap, waiters, count);
-		if (err)
-			goto out_order;
-	}
-
-	i915_random_reorder(order, count, &prng);
-	for (n = 0; n < count; n++) {
-		int i = order[n];
-
-		intel_engine_remove_wait(engine, &waiters[i]);
-		__clear_bit(i, bitmap);
-
-		err = check_rbtree(engine, bitmap, waiters, count);
-		if (err)
-			goto out_order;
-	}
-
-	err = check_rbtree_empty(engine);
-out_order:
-	kfree(order);
-out_bitmap:
-	kfree(bitmap);
-out_waiters:
-	kvfree(waiters);
-out_engines:
-	mock_engine_flush(engine);
-	return err;
-}
-
-static int igt_insert_complete(void *arg)
-{
-	const u32 seqno_bias = 0x1000;
-	struct intel_engine_cs *engine = arg;
-	struct intel_wait *waiters;
-	const int count = 4096;
-	unsigned long *bitmap;
-	int err = -ENOMEM;
-	int n, m;
-
-	mock_engine_reset(engine);
-
-	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
-	if (!waiters)
-		goto out_engines;
-
-	bitmap = kcalloc(DIV_ROUND_UP(count, BITS_PER_LONG), sizeof(*bitmap),
-			 GFP_KERNEL);
-	if (!bitmap)
-		goto out_waiters;
-
-	for (n = 0; n < count; n++) {
-		intel_wait_init_for_seqno(&waiters[n], n + seqno_bias);
-		intel_engine_add_wait(engine, &waiters[n]);
-		__set_bit(n, bitmap);
-	}
-	err = check_rbtree(engine, bitmap, waiters, count);
-	if (err)
-		goto out_bitmap;
-
-	/* On each step, we advance the seqno so that several waiters are then
-	 * complete (we increase the seqno by increasingly larger values to
-	 * retire more and more waiters at once). All retired waiters should
-	 * be woken and removed from the rbtree, and so that we check.
-	 */
-	for (n = 0; n < count; n = m) {
-		int seqno = 2 * n;
-
-		GEM_BUG_ON(find_first_bit(bitmap, count) != n);
-
-		if (intel_wait_complete(&waiters[n])) {
-			pr_err("waiter[%d, seqno=%d] completed too early\n",
-			       n, waiters[n].seqno);
-			err = -EINVAL;
-			goto out_bitmap;
-		}
-
-		/* complete the following waiters */
-		mock_seqno_advance(engine, seqno + seqno_bias);
-		for (m = n; m <= seqno; m++) {
-			if (m == count)
-				break;
-
-			GEM_BUG_ON(!test_bit(m, bitmap));
-			__clear_bit(m, bitmap);
-		}
-
-		intel_engine_remove_wait(engine, &waiters[n]);
-		RB_CLEAR_NODE(&waiters[n].node);
-
-		err = check_rbtree(engine, bitmap, waiters, count);
-		if (err) {
-			pr_err("rbtree corrupt after seqno advance to %d\n",
-			       seqno + seqno_bias);
-			goto out_bitmap;
-		}
-
-		err = check_completion(engine, bitmap, waiters, count);
-		if (err) {
-			pr_err("completions after seqno advance to %d failed\n",
-			       seqno + seqno_bias);
-			goto out_bitmap;
-		}
-	}
-
-	err = check_rbtree_empty(engine);
-out_bitmap:
-	kfree(bitmap);
-out_waiters:
-	kvfree(waiters);
-out_engines:
-	mock_engine_flush(engine);
-	return err;
-}
-
-struct igt_wakeup {
-	struct task_struct *tsk;
-	atomic_t *ready, *set, *done;
-	struct intel_engine_cs *engine;
-	unsigned long flags;
-#define STOP 0
-#define IDLE 1
-	wait_queue_head_t *wq;
-	u32 seqno;
-};
-
-static bool wait_for_ready(struct igt_wakeup *w)
-{
-	DEFINE_WAIT(ready);
-
-	set_bit(IDLE, &w->flags);
-	if (atomic_dec_and_test(w->done))
-		wake_up_var(w->done);
-
-	if (test_bit(STOP, &w->flags))
-		goto out;
-
-	for (;;) {
-		prepare_to_wait(w->wq, &ready, TASK_INTERRUPTIBLE);
-		if (atomic_read(w->ready) == 0)
-			break;
-
-		schedule();
-	}
-	finish_wait(w->wq, &ready);
-
-out:
-	clear_bit(IDLE, &w->flags);
-	if (atomic_dec_and_test(w->set))
-		wake_up_var(w->set);
-
-	return !test_bit(STOP, &w->flags);
-}
-
-static int igt_wakeup_thread(void *arg)
-{
-	struct igt_wakeup *w = arg;
-	struct intel_wait wait;
-
-	while (wait_for_ready(w)) {
-		GEM_BUG_ON(kthread_should_stop());
-
-		intel_wait_init_for_seqno(&wait, w->seqno);
-		intel_engine_add_wait(w->engine, &wait);
-		for (;;) {
-			set_current_state(TASK_UNINTERRUPTIBLE);
-			if (i915_seqno_passed(intel_engine_get_seqno(w->engine),
-					      w->seqno))
-				break;
-
-			if (test_bit(STOP, &w->flags)) /* emergency escape */
-				break;
-
-			schedule();
-		}
-		intel_engine_remove_wait(w->engine, &wait);
-		__set_current_state(TASK_RUNNING);
-	}
-
-	return 0;
-}
-
-static void igt_wake_all_sync(atomic_t *ready,
-			      atomic_t *set,
-			      atomic_t *done,
-			      wait_queue_head_t *wq,
-			      int count)
-{
-	atomic_set(set, count);
-	atomic_set(ready, 0);
-	wake_up_all(wq);
-
-	wait_var_event(set, !atomic_read(set));
-	atomic_set(ready, count);
-	atomic_set(done, count);
-}
-
-static int igt_wakeup(void *arg)
-{
-	I915_RND_STATE(prng);
-	struct intel_engine_cs *engine = arg;
-	struct igt_wakeup *waiters;
-	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
-	const int count = 4096;
-	const u32 max_seqno = count / 4;
-	atomic_t ready, set, done;
-	int err = -ENOMEM;
-	int n, step;
-
-	mock_engine_reset(engine);
-
-	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
-	if (!waiters)
-		goto out_engines;
-
-	/* Create a large number of threads, each waiting on a random seqno.
-	 * Multiple waiters will be waiting for the same seqno.
-	 */
-	atomic_set(&ready, count);
-	for (n = 0; n < count; n++) {
-		waiters[n].wq = &wq;
-		waiters[n].ready = &ready;
-		waiters[n].set = &set;
-		waiters[n].done = &done;
-		waiters[n].engine = engine;
-		waiters[n].flags = BIT(IDLE);
-
-		waiters[n].tsk = kthread_run(igt_wakeup_thread, &waiters[n],
-					     "i915/igt:%d", n);
-		if (IS_ERR(waiters[n].tsk))
-			goto out_waiters;
-
-		get_task_struct(waiters[n].tsk);
-	}
-
-	for (step = 1; step <= max_seqno; step <<= 1) {
-		u32 seqno;
-
-		/* The waiter threads start paused as we assign them a random
-		 * seqno and reset the engine. Once the engine is reset,
-		 * we signal that the threads may begin their wait upon their
-		 * seqno.
-		 */
-		for (n = 0; n < count; n++) {
-			GEM_BUG_ON(!test_bit(IDLE, &waiters[n].flags));
-			waiters[n].seqno =
-				1 + prandom_u32_state(&prng) % max_seqno;
-		}
-		mock_seqno_advance(engine, 0);
-		igt_wake_all_sync(&ready, &set, &done, &wq, count);
-
-		/* Simulate the GPU doing chunks of work, with one or more
-		 * seqno appearing to finish at the same time. A random number
-		 * of threads will be waiting upon the update and hopefully be
-		 * woken.
-		 */
-		for (seqno = 1; seqno <= max_seqno + step; seqno += step) {
-			usleep_range(50, 500);
-			mock_seqno_advance(engine, seqno);
-		}
-		GEM_BUG_ON(intel_engine_get_seqno(engine) < 1 + max_seqno);
-
-		/* With the seqno now beyond any of the waiting threads, they
-		 * should all be woken, see that they are complete and signal
-		 * that they are ready for the next test. We wait until all
-		 * threads are complete and waiting for us (i.e. not a seqno).
-		 */
-		if (!wait_var_event_timeout(&done,
-					    !atomic_read(&done), 10 * HZ)) {
-			pr_err("Timed out waiting for %d remaining waiters\n",
-			       atomic_read(&done));
-			err = -ETIMEDOUT;
-			break;
-		}
-
-		err = check_rbtree_empty(engine);
-		if (err)
-			break;
-	}
-
-out_waiters:
-	for (n = 0; n < count; n++) {
-		if (IS_ERR(waiters[n].tsk))
-			break;
-
-		set_bit(STOP, &waiters[n].flags);
-	}
-	mock_seqno_advance(engine, INT_MAX); /* wakeup any broken waiters */
-	igt_wake_all_sync(&ready, &set, &done, &wq, n);
-
-	for (n = 0; n < count; n++) {
-		if (IS_ERR(waiters[n].tsk))
-			break;
-
-		kthread_stop(waiters[n].tsk);
-		put_task_struct(waiters[n].tsk);
-	}
-
-	kvfree(waiters);
-out_engines:
-	mock_engine_flush(engine);
-	return err;
-}
-
-int intel_breadcrumbs_mock_selftests(void)
-{
-	static const struct i915_subtest tests[] = {
-		SUBTEST(igt_random_insert_remove),
-		SUBTEST(igt_insert_complete),
-		SUBTEST(igt_wakeup),
-	};
-	struct drm_i915_private *i915;
-	int err;
-
-	i915 = mock_gem_device();
-	if (!i915)
-		return -ENOMEM;
-
-	err = i915_subtests(tests, i915->engine[RCS]);
-	drm_dev_put(&i915->drm);
-
-	return err;
-}
diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
index 2c38ea5892d9..7b6f3bea9ef8 100644
--- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
@@ -1127,7 +1127,7 @@ static int __igt_reset_evict_vma(struct drm_i915_private *i915,
 
 	wait_for_completion(&arg.completion);
 
-	if (wait_for(waitqueue_active(&rq->execute), 10)) {
+	if (wait_for(!list_empty(&rq->fence.cb_list), 10)) {
 		struct drm_printer p = drm_info_printer(i915->drm.dev);
 
 		pr_err("igt/evict_vma kthread did not wait\n");
diff --git a/drivers/gpu/drm/i915/selftests/lib_sw_fence.c b/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
index b26f07b55d86..2bfa72c1654b 100644
--- a/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
+++ b/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
@@ -76,3 +76,57 @@ void timed_fence_fini(struct timed_fence *tf)
 	destroy_timer_on_stack(&tf->timer);
 	i915_sw_fence_fini(&tf->fence);
 }
+
+struct heap_fence {
+	struct i915_sw_fence fence;
+	union {
+		struct kref ref;
+		struct rcu_head rcu;
+	};
+};
+
+static int __i915_sw_fence_call
+heap_fence_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
+{
+	struct heap_fence *h = container_of(fence, typeof(*h), fence);
+
+	switch (state) {
+	case FENCE_COMPLETE:
+		break;
+
+	case FENCE_FREE:
+		heap_fence_put(&h->fence);
+	}
+
+	return NOTIFY_DONE;
+}
+
+struct i915_sw_fence *heap_fence_create(gfp_t gfp)
+{
+	struct heap_fence *h;
+
+	h = kmalloc(sizeof(*h), gfp);
+	if (!h)
+		return NULL;
+
+	i915_sw_fence_init(&h->fence, heap_fence_notify);
+	refcount_set(&h->ref.refcount, 2);
+
+	return &h->fence;
+}
+
+static void heap_fence_release(struct kref *ref)
+{
+	struct heap_fence *h = container_of(ref, typeof(*h), ref);
+
+	i915_sw_fence_fini(&h->fence);
+
+	kfree_rcu(h, rcu);
+}
+
+void heap_fence_put(struct i915_sw_fence *fence)
+{
+	struct heap_fence *h = container_of(fence, typeof(*h), fence);
+
+	kref_put(&h->ref, heap_fence_release);
+}
diff --git a/drivers/gpu/drm/i915/selftests/lib_sw_fence.h b/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
index 474aafb92ae1..1f9927e10f3a 100644
--- a/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
+++ b/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
@@ -39,4 +39,7 @@ struct timed_fence {
 void timed_fence_init(struct timed_fence *tf, unsigned long expires);
 void timed_fence_fini(struct timed_fence *tf);
 
+struct i915_sw_fence *heap_fence_create(gfp_t gfp);
+void heap_fence_put(struct i915_sw_fence *fence);
+
 #endif /* _LIB_SW_FENCE_H_ */
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
index 3b226ebc6bc4..08f0cab02e0f 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.c
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
@@ -86,17 +86,21 @@ static struct mock_request *first_request(struct mock_engine *engine)
 static void advance(struct mock_request *request)
 {
 	list_del_init(&request->link);
-	mock_seqno_advance(request->base.engine, request->base.global_seqno);
+	intel_engine_write_global_seqno(request->base.engine,
+					request->base.global_seqno);
 	i915_request_mark_complete(&request->base);
 	GEM_BUG_ON(!i915_request_completed(&request->base));
+
+	intel_engine_queue_breadcrumbs(request->base.engine);
 }
 
 static void hw_delay_complete(struct timer_list *t)
 {
 	struct mock_engine *engine = from_timer(engine, t, hw_delay);
 	struct mock_request *request;
+	unsigned long flags;
 
-	spin_lock(&engine->hw_lock);
+	spin_lock_irqsave(&engine->hw_lock, flags);
 
 	/* Timer fired, first request is complete */
 	request = first_request(engine);
@@ -116,7 +120,7 @@ static void hw_delay_complete(struct timer_list *t)
 		advance(request);
 	}
 
-	spin_unlock(&engine->hw_lock);
+	spin_unlock_irqrestore(&engine->hw_lock, flags);
 }
 
 static void mock_context_unpin(struct intel_context *ce)
@@ -191,11 +195,12 @@ static void mock_submit_request(struct i915_request *request)
 	struct mock_request *mock = container_of(request, typeof(*mock), base);
 	struct mock_engine *engine =
 		container_of(request->engine, typeof(*engine), base);
+	unsigned long flags;
 
 	i915_request_submit(request);
 	GEM_BUG_ON(!request->global_seqno);
 
-	spin_lock_irq(&engine->hw_lock);
+	spin_lock_irqsave(&engine->hw_lock, flags);
 	list_add_tail(&mock->link, &engine->hw_queue);
 	if (mock->link.prev == &engine->hw_queue) {
 		if (mock->delay)
@@ -203,7 +208,7 @@ static void mock_submit_request(struct i915_request *request)
 		else
 			advance(mock);
 	}
-	spin_unlock_irq(&engine->hw_lock);
+	spin_unlock_irqrestore(&engine->hw_lock, flags);
 }
 
 struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
@@ -273,7 +278,7 @@ void mock_engine_flush(struct intel_engine_cs *engine)
 
 void mock_engine_reset(struct intel_engine_cs *engine)
 {
-	intel_write_status_page(engine, I915_GEM_HWS_INDEX, 0);
+	intel_engine_write_global_seqno(engine, 0);
 }
 
 void mock_engine_free(struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.h b/drivers/gpu/drm/i915/selftests/mock_engine.h
index 133d0c21790d..b9cc3a245f16 100644
--- a/drivers/gpu/drm/i915/selftests/mock_engine.h
+++ b/drivers/gpu/drm/i915/selftests/mock_engine.h
@@ -46,10 +46,4 @@ void mock_engine_flush(struct intel_engine_cs *engine);
 void mock_engine_reset(struct intel_engine_cs *engine);
 void mock_engine_free(struct intel_engine_cs *engine);
 
-static inline void mock_seqno_advance(struct intel_engine_cs *engine, u32 seqno)
-{
-	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
-	intel_engine_wakeup(engine);
-}
-
 #endif /* !__MOCK_ENGINE_H__ */
-- 
2.20.1

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

Re: [PATCH 27/33] drm/i915: Replace global breadcrumbs with per-context interrupt tracking

[Tvrtko Ursulin]

On 25/01/2019 02:29, Chris Wilson wrote:
> A few years ago, see commit 688e6c725816 ("drm/i915: Slaughter the
> thundering i915_wait_request herd"), the issue of handling multiple
> clients waiting in parallel was brought to our attention. The
> requirement was that every client should be woken immediately upon its
> request being signaled, without incurring any cpu overhead.
> 
> To handle certain fragility of our hw meant that we could not do a
> simple check inside the irq handler (some generations required almost
> unbounded delays before we could be sure of seqno coherency) and so
> request completion checking required delegation.
> 
> Before commit 688e6c725816, the solution was simple. Every client waking
> on a request would be woken on every interrupt and each would do a
> heavyweight check to see if their request was complete. Commit
> 688e6c725816 introduced an rbtree so that only the earliest waiter on
> the global timeline would woken, and would wake the next and so on.
> (Along with various complications to handle requests being reordered
> along the global timeline, and also a requirement for kthread to provide
> a delegate for fence signaling that had no process context.)
> 
> The global rbtree depends on knowing the execution timeline (and global
> seqno). Without knowing that order, we must instead check all contexts
> queued to the HW to see which may have advanced. We trim that list by
> only checking queued contexts that are being waited on, but still we
> keep a list of all active contexts and their active signalers that we
> inspect from inside the irq handler. By moving the waiters onto the fence
> signal list, we can combine the client wakeup with the dma_fence
> signaling (a dramatic reduction in complexity, but does require the HW
> being coherent, the seqno must be visible from the cpu before the
> interrupt is raised - we keep a timer backup just in case).
> 
> Having previously fixed all the issues with irq-seqno serialisation (by
> inserting delays onto the GPU after each request instead of random delays
> on the CPU after each interrupt), we can rely on the seqno state to
> perfom direct wakeups from the interrupt handler. This allows us to
> preserve our single context switch behaviour of the current routine,
> with the only downside that we lose the RT priority sorting of wakeups.
> In general, direct wakeup latency of multiple clients is about the same
> (about 10% better in most cases) with a reduction in total CPU time spent
> in the waiter (about 20-50% depending on gen). Average herd behaviour is
> improved, but at the cost of not delegating wakeups on task_prio.
> 
> v2: Capture fence signaling state for error state and add comments to
> warm even the most cold of hearts.
> v3: Check if the request is still active before busywaiting
> v4: Reduce the amount of pointer misdirection with list_for_each_safe
> and using a local i915_request variable inside the loops
> 
> References: 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> ---
>   drivers/gpu/drm/i915/i915_debugfs.c           |  28 +-
>   drivers/gpu/drm/i915/i915_gem_context.c       |   3 +
>   drivers/gpu/drm/i915/i915_gem_context.h       |   2 +
>   drivers/gpu/drm/i915/i915_gpu_error.c         |  83 +-
>   drivers/gpu/drm/i915/i915_gpu_error.h         |   9 +-
>   drivers/gpu/drm/i915/i915_irq.c               |  88 +-
>   drivers/gpu/drm/i915/i915_request.c           | 140 +--
>   drivers/gpu/drm/i915/i915_request.h           |  72 +-
>   drivers/gpu/drm/i915/i915_reset.c             |  13 +-
>   drivers/gpu/drm/i915/intel_breadcrumbs.c      | 818 +++++-------------
>   drivers/gpu/drm/i915/intel_engine_cs.c        |  35 +-
>   drivers/gpu/drm/i915/intel_ringbuffer.c       |   6 +-
>   drivers/gpu/drm/i915/intel_ringbuffer.h       |  94 +-
>   .../drm/i915/selftests/i915_mock_selftests.h  |   1 -
>   drivers/gpu/drm/i915/selftests/i915_request.c | 420 +++++++++
>   drivers/gpu/drm/i915/selftests/igt_spinner.c  |   5 -
>   .../drm/i915/selftests/intel_breadcrumbs.c    | 470 ----------
>   .../gpu/drm/i915/selftests/intel_hangcheck.c  |   2 +-
>   drivers/gpu/drm/i915/selftests/lib_sw_fence.c |  54 ++
>   drivers/gpu/drm/i915/selftests/lib_sw_fence.h |   3 +
>   drivers/gpu/drm/i915/selftests/mock_engine.c  |  17 +-
>   drivers/gpu/drm/i915/selftests/mock_engine.h  |   6 -
>   22 files changed, 886 insertions(+), 1483 deletions(-)
>   delete mode 100644 drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
> 
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index b64f758803f3..e0ca3987eed6 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -1316,29 +1316,16 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
>   	seq_printf(m, "GT active? %s\n", yesno(dev_priv->gt.awake));
>   
>   	for_each_engine(engine, dev_priv, id) {
> -		struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -		struct rb_node *rb;
> -
>   		seq_printf(m, "%s:\n", engine->name);
>   		seq_printf(m, "\tseqno = %x [current %x, last %x], %dms ago\n",
>   			   engine->hangcheck.seqno, seqno[id],
>   			   intel_engine_last_submit(engine),
>   			   jiffies_to_msecs(jiffies -
>   					    engine->hangcheck.action_timestamp));
> -		seq_printf(m, "\twaiters? %s, fake irq active? %s\n",
> -			   yesno(intel_engine_has_waiter(engine)),
> +		seq_printf(m, "\tfake irq active? %s\n",
>   			   yesno(test_bit(engine->id,
>   					  &dev_priv->gpu_error.missed_irq_rings)));
>   
> -		spin_lock_irq(&b->rb_lock);
> -		for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
> -			struct intel_wait *w = rb_entry(rb, typeof(*w), node);
> -
> -			seq_printf(m, "\t%s [%d] waiting for %x\n",
> -				   w->tsk->comm, w->tsk->pid, w->seqno);
> -		}
> -		spin_unlock_irq(&b->rb_lock);
> -
>   		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
>   			   (long long)engine->hangcheck.acthd,
>   			   (long long)acthd[id]);
> @@ -2022,18 +2009,6 @@ static int i915_swizzle_info(struct seq_file *m, void *data)
>   	return 0;
>   }
>   
> -static int count_irq_waiters(struct drm_i915_private *i915)
> -{
> -	struct intel_engine_cs *engine;
> -	enum intel_engine_id id;
> -	int count = 0;
> -
> -	for_each_engine(engine, i915, id)
> -		count += intel_engine_has_waiter(engine);
> -
> -	return count;
> -}
> -
>   static const char *rps_power_to_str(unsigned int power)
>   {
>   	static const char * const strings[] = {
> @@ -2073,7 +2048,6 @@ static int i915_rps_boost_info(struct seq_file *m, void *data)
>   	seq_printf(m, "RPS enabled? %d\n", rps->enabled);
>   	seq_printf(m, "GPU busy? %s [%d requests]\n",
>   		   yesno(dev_priv->gt.awake), dev_priv->gt.active_requests);
> -	seq_printf(m, "CPU waiting? %d\n", count_irq_waiters(dev_priv));
>   	seq_printf(m, "Boosts outstanding? %d\n",
>   		   atomic_read(&rps->num_waiters));
>   	seq_printf(m, "Interactive? %d\n", READ_ONCE(rps->power.interactive));
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
> index 93e84751370f..6faf1f6faab5 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/i915_gem_context.c
> @@ -327,6 +327,9 @@ intel_context_init(struct intel_context *ce,
>   		   struct intel_engine_cs *engine)
>   {
>   	ce->gem_context = ctx;
> +
> +	INIT_LIST_HEAD(&ce->signal_link);
> +	INIT_LIST_HEAD(&ce->signals);
>   }
>   
>   static struct i915_gem_context *
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
> index 3769438228f6..6ba40ff6b91f 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.h
> +++ b/drivers/gpu/drm/i915/i915_gem_context.h
> @@ -164,6 +164,8 @@ struct i915_gem_context {
>   	struct intel_context {
>   		struct i915_gem_context *gem_context;
>   		struct intel_engine_cs *active;
> +		struct list_head signal_link;
> +		struct list_head signals;
>   		struct i915_vma *state;
>   		struct intel_ring *ring;
>   		u32 *lrc_reg_state;
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
> index 96d1d634a29d..420b94341433 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
> @@ -447,9 +447,14 @@ static void error_print_request(struct drm_i915_error_state_buf *m,
>   	if (!erq->seqno)
>   		return;
>   
> -	err_printf(m, "%s pid %d, ban score %d, seqno %8x:%08x, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
> +	err_printf(m, "%s pid %d, ban score %d, seqno %8x:%08x%s%s, prio %d, emitted %dms, start %08x, head %08x, tail %08x\n",
>   		   prefix, erq->pid, erq->ban_score,
> -		   erq->context, erq->seqno, erq->sched_attr.priority,
> +		   erq->context, erq->seqno,
> +		   test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> +			    &erq->flags) ? "!" : "",
> +		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
> +			    &erq->flags) ? "+" : "",
> +		   erq->sched_attr.priority,
>   		   jiffies_to_msecs(erq->jiffies - epoch),
>   		   erq->start, erq->head, erq->tail);
>   }
> @@ -530,7 +535,6 @@ static void error_print_engine(struct drm_i915_error_state_buf *m,
>   	}
>   	err_printf(m, "  seqno: 0x%08x\n", ee->seqno);
>   	err_printf(m, "  last_seqno: 0x%08x\n", ee->last_seqno);
> -	err_printf(m, "  waiting: %s\n", yesno(ee->waiting));
>   	err_printf(m, "  ring->head: 0x%08x\n", ee->cpu_ring_head);
>   	err_printf(m, "  ring->tail: 0x%08x\n", ee->cpu_ring_tail);
>   	err_printf(m, "  hangcheck timestamp: %dms (%lu%s)\n",
> @@ -804,21 +808,6 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
>   						    error->epoch);
>   		}
>   
> -		if (IS_ERR(ee->waiters)) {
> -			err_printf(m, "%s --- ? waiters [unable to acquire spinlock]\n",
> -				   m->i915->engine[i]->name);
> -		} else if (ee->num_waiters) {
> -			err_printf(m, "%s --- %d waiters\n",
> -				   m->i915->engine[i]->name,
> -				   ee->num_waiters);
> -			for (j = 0; j < ee->num_waiters; j++) {
> -				err_printf(m, " seqno 0x%08x for %s [%d]\n",
> -					   ee->waiters[j].seqno,
> -					   ee->waiters[j].comm,
> -					   ee->waiters[j].pid);
> -			}
> -		}
> -
>   		print_error_obj(m, m->i915->engine[i],
>   				"ringbuffer", ee->ringbuffer);
>   
> @@ -1000,8 +989,6 @@ void __i915_gpu_state_free(struct kref *error_ref)
>   		i915_error_object_free(ee->wa_ctx);
>   
>   		kfree(ee->requests);
> -		if (!IS_ERR_OR_NULL(ee->waiters))
> -			kfree(ee->waiters);
>   	}
>   
>   	for (i = 0; i < ARRAY_SIZE(error->active_bo); i++)
> @@ -1203,59 +1190,6 @@ static void gen6_record_semaphore_state(struct intel_engine_cs *engine,
>   			I915_READ(RING_SYNC_2(engine->mmio_base));
>   }
>   
> -static void error_record_engine_waiters(struct intel_engine_cs *engine,
> -					struct drm_i915_error_engine *ee)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct drm_i915_error_waiter *waiter;
> -	struct rb_node *rb;
> -	int count;
> -
> -	ee->num_waiters = 0;
> -	ee->waiters = NULL;
> -
> -	if (RB_EMPTY_ROOT(&b->waiters))
> -		return;
> -
> -	if (!spin_trylock_irq(&b->rb_lock)) {
> -		ee->waiters = ERR_PTR(-EDEADLK);
> -		return;
> -	}
> -
> -	count = 0;
> -	for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
> -		count++;
> -	spin_unlock_irq(&b->rb_lock);
> -
> -	waiter = NULL;
> -	if (count)
> -		waiter = kmalloc_array(count,
> -				       sizeof(struct drm_i915_error_waiter),
> -				       GFP_ATOMIC);
> -	if (!waiter)
> -		return;
> -
> -	if (!spin_trylock_irq(&b->rb_lock)) {
> -		kfree(waiter);
> -		ee->waiters = ERR_PTR(-EDEADLK);
> -		return;
> -	}
> -
> -	ee->waiters = waiter;
> -	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
> -		struct intel_wait *w = rb_entry(rb, typeof(*w), node);
> -
> -		strcpy(waiter->comm, w->tsk->comm);
> -		waiter->pid = w->tsk->pid;
> -		waiter->seqno = w->seqno;
> -		waiter++;
> -
> -		if (++ee->num_waiters == count)
> -			break;
> -	}
> -	spin_unlock_irq(&b->rb_lock);
> -}
> -
>   static void error_record_engine_registers(struct i915_gpu_state *error,
>   					  struct intel_engine_cs *engine,
>   					  struct drm_i915_error_engine *ee)
> @@ -1291,7 +1225,6 @@ static void error_record_engine_registers(struct i915_gpu_state *error,
>   
>   	intel_engine_get_instdone(engine, &ee->instdone);
>   
> -	ee->waiting = intel_engine_has_waiter(engine);
>   	ee->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
>   	ee->acthd = intel_engine_get_active_head(engine);
>   	ee->seqno = intel_engine_get_seqno(engine);
> @@ -1365,6 +1298,7 @@ static void record_request(struct i915_request *request,
>   {
>   	struct i915_gem_context *ctx = request->gem_context;
>   
> +	erq->flags = request->fence.flags;
>   	erq->context = ctx->hw_id;
>   	erq->sched_attr = request->sched.attr;
>   	erq->ban_score = atomic_read(&ctx->ban_score);
> @@ -1540,7 +1474,6 @@ static void gem_record_rings(struct i915_gpu_state *error)
>   		ee->engine_id = i;
>   
>   		error_record_engine_registers(error, engine, ee);
> -		error_record_engine_waiters(engine, ee);
>   		error_record_engine_execlists(engine, ee);
>   
>   		request = i915_gem_find_active_request(engine);
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
> index 231173786eae..74757c424aab 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.h
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.h
> @@ -82,8 +82,6 @@ struct i915_gpu_state {
>   		int engine_id;
>   		/* Software tracked state */
>   		bool idle;
> -		bool waiting;
> -		int num_waiters;
>   		unsigned long hangcheck_timestamp;
>   		struct i915_address_space *vm;
>   		int num_requests;
> @@ -147,6 +145,7 @@ struct i915_gpu_state {
>   		struct drm_i915_error_object *default_state;
>   
>   		struct drm_i915_error_request {
> +			unsigned long flags;
>   			long jiffies;
>   			pid_t pid;
>   			u32 context;
> @@ -159,12 +158,6 @@ struct i915_gpu_state {
>   		} *requests, execlist[EXECLIST_MAX_PORTS];
>   		unsigned int num_ports;
>   
> -		struct drm_i915_error_waiter {
> -			char comm[TASK_COMM_LEN];
> -			pid_t pid;
> -			u32 seqno;
> -		} *waiters;
> -
>   		struct {
>   			u32 gfx_mode;
>   			union {
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index bedac6c14928..f09ea93e1037 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -28,12 +28,14 @@
>   
>   #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
>   
> -#include <linux/sysrq.h>
> -#include <linux/slab.h>
>   #include <linux/circ_buf.h>
> +#include <linux/slab.h>
> +#include <linux/sysrq.h>
> +
>   #include <drm/drm_irq.h>
>   #include <drm/drm_drv.h>
>   #include <drm/i915_drm.h>
> +
>   #include "i915_drv.h"
>   #include "i915_trace.h"
>   #include "intel_drv.h"
> @@ -1153,66 +1155,6 @@ static void ironlake_rps_change_irq_handler(struct drm_i915_private *dev_priv)
>   	return;
>   }
>   
> -static void notify_ring(struct intel_engine_cs *engine)
> -{
> -	const u32 seqno = intel_engine_get_seqno(engine);
> -	struct i915_request *rq = NULL;
> -	struct task_struct *tsk = NULL;
> -	struct intel_wait *wait;
> -
> -	if (unlikely(!engine->breadcrumbs.irq_armed))
> -		return;
> -
> -	rcu_read_lock();
> -
> -	spin_lock(&engine->breadcrumbs.irq_lock);
> -	wait = engine->breadcrumbs.irq_wait;
> -	if (wait) {
> -		/*
> -		 * We use a callback from the dma-fence to submit
> -		 * requests after waiting on our own requests. To
> -		 * ensure minimum delay in queuing the next request to
> -		 * hardware, signal the fence now rather than wait for
> -		 * the signaler to be woken up. We still wake up the
> -		 * waiter in order to handle the irq-seqno coherency
> -		 * issues (we may receive the interrupt before the
> -		 * seqno is written, see __i915_request_irq_complete())
> -		 * and to handle coalescing of multiple seqno updates
> -		 * and many waiters.
> -		 */
> -		if (i915_seqno_passed(seqno, wait->seqno)) {
> -			struct i915_request *waiter = wait->request;
> -
> -			if (waiter &&
> -			    !i915_request_signaled(waiter) &&
> -			    intel_wait_check_request(wait, waiter))
> -				rq = i915_request_get(waiter);
> -
> -			tsk = wait->tsk;
> -		}
> -
> -		engine->breadcrumbs.irq_count++;
> -	} else {
> -		if (engine->breadcrumbs.irq_armed)
> -			__intel_engine_disarm_breadcrumbs(engine);
> -	}
> -	spin_unlock(&engine->breadcrumbs.irq_lock);
> -
> -	if (rq) {
> -		spin_lock(&rq->lock);
> -		dma_fence_signal_locked(&rq->fence);
> -		GEM_BUG_ON(!i915_request_completed(rq));
> -		spin_unlock(&rq->lock);
> -
> -		i915_request_put(rq);
> -	}
> -
> -	if (tsk && tsk->state & TASK_NORMAL)
> -		wake_up_process(tsk);
> -
> -	rcu_read_unlock();
> -}
> -
>   static void vlv_c0_read(struct drm_i915_private *dev_priv,
>   			struct intel_rps_ei *ei)
>   {
> @@ -1457,20 +1399,20 @@ static void ilk_gt_irq_handler(struct drm_i915_private *dev_priv,
>   			       u32 gt_iir)
>   {
>   	if (gt_iir & GT_RENDER_USER_INTERRUPT)
> -		notify_ring(dev_priv->engine[RCS]);
> +		intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
>   	if (gt_iir & ILK_BSD_USER_INTERRUPT)
> -		notify_ring(dev_priv->engine[VCS]);
> +		intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
>   }
>   
>   static void snb_gt_irq_handler(struct drm_i915_private *dev_priv,
>   			       u32 gt_iir)
>   {
>   	if (gt_iir & GT_RENDER_USER_INTERRUPT)
> -		notify_ring(dev_priv->engine[RCS]);
> +		intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
>   	if (gt_iir & GT_BSD_USER_INTERRUPT)
> -		notify_ring(dev_priv->engine[VCS]);
> +		intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
>   	if (gt_iir & GT_BLT_USER_INTERRUPT)
> -		notify_ring(dev_priv->engine[BCS]);
> +		intel_engine_breadcrumbs_irq(dev_priv->engine[BCS]);
>   
>   	if (gt_iir & (GT_BLT_CS_ERROR_INTERRUPT |
>   		      GT_BSD_CS_ERROR_INTERRUPT |
> @@ -1490,7 +1432,7 @@ gen8_cs_irq_handler(struct intel_engine_cs *engine, u32 iir)
>   		tasklet = true;
>   
>   	if (iir & GT_RENDER_USER_INTERRUPT) {
> -		notify_ring(engine);
> +		intel_engine_breadcrumbs_irq(engine);
>   		tasklet |= USES_GUC_SUBMISSION(engine->i915);
>   	}
>   
> @@ -1836,7 +1778,7 @@ static void gen6_rps_irq_handler(struct drm_i915_private *dev_priv, u32 pm_iir)
>   
>   	if (HAS_VEBOX(dev_priv)) {
>   		if (pm_iir & PM_VEBOX_USER_INTERRUPT)
> -			notify_ring(dev_priv->engine[VECS]);
> +			intel_engine_breadcrumbs_irq(dev_priv->engine[VECS]);
>   
>   		if (pm_iir & PM_VEBOX_CS_ERROR_INTERRUPT)
>   			DRM_DEBUG("Command parser error, pm_iir 0x%08x\n", pm_iir);
> @@ -4260,7 +4202,7 @@ static irqreturn_t i8xx_irq_handler(int irq, void *arg)
>   		I915_WRITE16(IIR, iir);
>   
>   		if (iir & I915_USER_INTERRUPT)
> -			notify_ring(dev_priv->engine[RCS]);
> +			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
>   
>   		if (iir & I915_MASTER_ERROR_INTERRUPT)
>   			i8xx_error_irq_handler(dev_priv, eir, eir_stuck);
> @@ -4368,7 +4310,7 @@ static irqreturn_t i915_irq_handler(int irq, void *arg)
>   		I915_WRITE(IIR, iir);
>   
>   		if (iir & I915_USER_INTERRUPT)
> -			notify_ring(dev_priv->engine[RCS]);
> +			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
>   
>   		if (iir & I915_MASTER_ERROR_INTERRUPT)
>   			i9xx_error_irq_handler(dev_priv, eir, eir_stuck);
> @@ -4513,10 +4455,10 @@ static irqreturn_t i965_irq_handler(int irq, void *arg)
>   		I915_WRITE(IIR, iir);
>   
>   		if (iir & I915_USER_INTERRUPT)
> -			notify_ring(dev_priv->engine[RCS]);
> +			intel_engine_breadcrumbs_irq(dev_priv->engine[RCS]);
>   
>   		if (iir & I915_BSD_USER_INTERRUPT)
> -			notify_ring(dev_priv->engine[VCS]);
> +			intel_engine_breadcrumbs_irq(dev_priv->engine[VCS]);
>   
>   		if (iir & I915_MASTER_ERROR_INTERRUPT)
>   			i9xx_error_irq_handler(dev_priv, eir, eir_stuck);
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 2171df2d3019..c09a6644a2ab 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -60,7 +60,7 @@ static bool i915_fence_signaled(struct dma_fence *fence)
>   
>   static bool i915_fence_enable_signaling(struct dma_fence *fence)
>   {
> -	return intel_engine_enable_signaling(to_request(fence), true);
> +	return i915_request_enable_breadcrumb(to_request(fence));

enable_signaling would be better IMO, enable_breadcrumb sounds like it 
could do with breadcrumb emission. And it would align with 
i915_fence_enable_signaling.

>   }
>   
>   static signed long i915_fence_wait(struct dma_fence *fence,
> @@ -203,7 +203,7 @@ static void __retire_engine_request(struct intel_engine_cs *engine,
>   	if (!i915_request_signaled(rq))
>   		dma_fence_signal_locked(&rq->fence);
>   	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &rq->fence.flags))
> -		intel_engine_cancel_signaling(rq);
> +		i915_request_cancel_breadcrumb(rq);
>   	if (rq->waitboost) {
>   		GEM_BUG_ON(!atomic_read(&rq->i915->gt_pm.rps.num_waiters));
>   		atomic_dec(&rq->i915->gt_pm.rps.num_waiters);
> @@ -377,9 +377,11 @@ void __i915_request_submit(struct i915_request *request)
>   
>   	/* We may be recursing from the signal callback of another i915 fence */
>   	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
> +	set_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags);
>   	request->global_seqno = seqno;
> -	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags))
> -		intel_engine_enable_signaling(request, false);
> +	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags) &&
> +	    !i915_request_enable_breadcrumb(request))
> +		intel_engine_queue_breadcrumbs(engine);
>   	spin_unlock(&request->lock);
>   
>   	engine->emit_fini_breadcrumb(request,
> @@ -389,8 +391,6 @@ void __i915_request_submit(struct i915_request *request)
>   	move_to_timeline(request, &engine->timeline);
>   
>   	trace_i915_request_execute(request);
> -
> -	wake_up_all(&request->execute);
>   }
>   
>   void i915_request_submit(struct i915_request *request)
> @@ -433,7 +433,8 @@ void __i915_request_unsubmit(struct i915_request *request)
>   	spin_lock_nested(&request->lock, SINGLE_DEPTH_NESTING);
>   	request->global_seqno = 0;
>   	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags))
> -		intel_engine_cancel_signaling(request);
> +		i915_request_cancel_breadcrumb(request);
> +	clear_bit(I915_FENCE_FLAG_ACTIVE, &request->fence.flags);
>   	spin_unlock(&request->lock);
>   
>   	/* Transfer back from the global per-engine timeline to per-context */
> @@ -646,13 +647,11 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
>   
>   	/* We bump the ref for the fence chain */
>   	i915_sw_fence_init(&i915_request_get(rq)->submit, submit_notify);
> -	init_waitqueue_head(&rq->execute);
>   
>   	i915_sched_node_init(&rq->sched);
>   
>   	/* No zalloc, must clear what we need by hand */
>   	rq->global_seqno = 0;
> -	rq->signaling.wait.seqno = 0;
>   	rq->file_priv = NULL;
>   	rq->batch = NULL;
>   	rq->capture_list = NULL;
> @@ -1047,13 +1046,10 @@ static bool busywait_stop(unsigned long timeout, unsigned int cpu)
>   	return this_cpu != cpu;
>   }
>   
> -static bool __i915_spin_request(const struct i915_request *rq,
> -				u32 seqno, int state, unsigned long timeout_us)
> +static bool __i915_spin_request(const struct i915_request * const rq,
> +				int state, unsigned long timeout_us)
>   {
> -	struct intel_engine_cs *engine = rq->engine;
> -	unsigned int irq, cpu;
> -
> -	GEM_BUG_ON(!seqno);
> +	unsigned int cpu;
>   
>   	/*
>   	 * Only wait for the request if we know it is likely to complete.
> @@ -1061,12 +1057,12 @@ static bool __i915_spin_request(const struct i915_request *rq,
>   	 * We don't track the timestamps around requests, nor the average
>   	 * request length, so we do not have a good indicator that this
>   	 * request will complete within the timeout. What we do know is the
> -	 * order in which requests are executed by the engine and so we can
> -	 * tell if the request has started. If the request hasn't started yet,
> -	 * it is a fair assumption that it will not complete within our
> -	 * relatively short timeout.
> +	 * order in which requests are executed by the context and so we can
> +	 * tell if the request has been started. If the request is not even
> +	 * running yet, it is a fair assumption that it will not complete
> +	 * within our relatively short timeout.
>   	 */
> -	if (!intel_engine_has_started(engine, seqno))
> +	if (!i915_request_is_running(rq))
>   		return false;
>   
>   	/*
> @@ -1080,20 +1076,10 @@ static bool __i915_spin_request(const struct i915_request *rq,
>   	 * takes to sleep on a request, on the order of a microsecond.
>   	 */
>   
> -	irq = READ_ONCE(engine->breadcrumbs.irq_count);
>   	timeout_us += local_clock_us(&cpu);
>   	do {
> -		if (intel_engine_has_completed(engine, seqno))
> -			return seqno == i915_request_global_seqno(rq);
> -
> -		/*
> -		 * Seqno are meant to be ordered *before* the interrupt. If
> -		 * we see an interrupt without a corresponding seqno advance,
> -		 * assume we won't see one in the near future but require
> -		 * the engine->seqno_barrier() to fixup coherency.
> -		 */
> -		if (READ_ONCE(engine->breadcrumbs.irq_count) != irq)
> -			break;
> +		if (i915_request_completed(rq))
> +			return true;
>   
>   		if (signal_pending_state(state, current))
>   			break;
> @@ -1107,6 +1093,18 @@ static bool __i915_spin_request(const struct i915_request *rq,
>   	return false;
>   }
>   
> +struct request_wait {
> +	struct dma_fence_cb cb;
> +	struct task_struct *tsk;
> +};
> +
> +static void request_wait_wake(struct dma_fence *fence, struct dma_fence_cb *cb)
> +{
> +	struct request_wait *wait = container_of(cb, typeof(*wait), cb);
> +
> +	wake_up_process(wait->tsk);
> +}
> +
>   /**
>    * i915_request_wait - wait until execution of request has finished
>    * @rq: the request to wait upon
> @@ -1132,8 +1130,7 @@ long i915_request_wait(struct i915_request *rq,
>   {
>   	const int state = flags & I915_WAIT_INTERRUPTIBLE ?
>   		TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
> -	DEFINE_WAIT_FUNC(exec, default_wake_function);
> -	struct intel_wait wait;
> +	struct request_wait wait;
>   
>   	might_sleep();
>   	GEM_BUG_ON(timeout < 0);
> @@ -1145,47 +1142,24 @@ long i915_request_wait(struct i915_request *rq,
>   		return -ETIME;
>   
>   	trace_i915_request_wait_begin(rq, flags);
> -	add_wait_queue(&rq->execute, &exec);
> -	intel_wait_init(&wait);
> -	if (flags & I915_WAIT_PRIORITY)
> -		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
> -
> -restart:
> -	do {
> -		set_current_state(state);
> -		if (intel_wait_update_request(&wait, rq))
> -			break;
> -
> -		if (signal_pending_state(state, current)) {
> -			timeout = -ERESTARTSYS;
> -			goto complete;
> -		}
>   
> -		if (!timeout) {
> -			timeout = -ETIME;
> -			goto complete;
> -		}
> +	/* Optimistic short spin before touching IRQs */
> +	if (__i915_spin_request(rq, state, 5))
> +		goto out;
>   
> -		timeout = io_schedule_timeout(timeout);
> -	} while (1);
> +	if (flags & I915_WAIT_PRIORITY)
> +		i915_schedule_bump_priority(rq, I915_PRIORITY_WAIT);
>   
> -	GEM_BUG_ON(!intel_wait_has_seqno(&wait));
> -	GEM_BUG_ON(!i915_sw_fence_signaled(&rq->submit));
> +	wait.tsk = current;
> +	if (dma_fence_add_callback(&rq->fence, &wait.cb, request_wait_wake))
> +		goto out;
>   
> -	/* Optimistic short spin before touching IRQs */
> -	if (__i915_spin_request(rq, wait.seqno, state, 5))
> -		goto complete;
> +	for (;;) {
> +		set_current_state(state);
>   
> -	set_current_state(state);
> -	if (intel_engine_add_wait(rq->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
> -		 * coherent check on the seqno before we sleep.
> -		 */
> -		goto wakeup;
> +		if (i915_request_completed(rq))
> +			break;
>   
> -	for (;;) {
>   		if (signal_pending_state(state, current)) {
>   			timeout = -ERESTARTSYS;
>   			break;
> @@ -1197,33 +1171,13 @@ long i915_request_wait(struct i915_request *rq,
>   		}
>   
>   		timeout = io_schedule_timeout(timeout);
> -
> -		if (intel_wait_complete(&wait) &&
> -		    intel_wait_check_request(&wait, rq))
> -			break;
> -
> -		set_current_state(state);
> -
> -wakeup:
> -		if (i915_request_completed(rq))
> -			break;
> -
> -		/* Only spin if we know the GPU is processing this request */
> -		if (__i915_spin_request(rq, wait.seqno, state, 2))
> -			break;
> -
> -		if (!intel_wait_check_request(&wait, rq)) {
> -			intel_engine_remove_wait(rq->engine, &wait);
> -			goto restart;
> -		}
>   	}
> -
> -	intel_engine_remove_wait(rq->engine, &wait);
> -complete:
>   	__set_current_state(TASK_RUNNING);
> -	remove_wait_queue(&rq->execute, &exec);
> -	trace_i915_request_wait_end(rq);
>   
> +	dma_fence_remove_callback(&rq->fence, &wait.cb);
> +
> +out:
> +	trace_i915_request_wait_end(rq);
>   	return timeout;
>   }
>   
> diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
> index 340d6216791c..3cffb96203b9 100644
> --- a/drivers/gpu/drm/i915/i915_request.h
> +++ b/drivers/gpu/drm/i915/i915_request.h
> @@ -38,23 +38,34 @@ struct drm_i915_gem_object;
>   struct i915_request;
>   struct i915_timeline;
>   
> -struct intel_wait {
> -	struct rb_node node;
> -	struct task_struct *tsk;
> -	struct i915_request *request;
> -	u32 seqno;
> -};
> -
> -struct intel_signal_node {
> -	struct intel_wait wait;
> -	struct list_head link;
> -};
> -
>   struct i915_capture_list {
>   	struct i915_capture_list *next;
>   	struct i915_vma *vma;
>   };
>   
> +enum {
> +	/*
> +	 * I915_FENCE_FLAG_ACTIVE - this request is currently submitted to HW.
> +	 *
> +	 * Set by __i915_request_submit() on handing over to HW, and cleared
> +	 * by __i915_request_unsubmit() if we preempt this request.
> +	 *
> +	 * Finally cleared for consistency on retiring the request, when
> +	 * we know the HW is no longer running this request.
> +	 *
> +	 * See i915_request_is_active()
> +	 */
> +	I915_FENCE_FLAG_ACTIVE = DMA_FENCE_FLAG_USER_BITS,
> +
> +	/*
> +	 * I915_FENCE_FLAG_SIGNAL - this request is currently on signal_list
> +	 *
> +	 * Internal bookkeeping used by the breadcrumb code to track when
> +	 * a request is on the various signal_list.
> +	 */
> +	I915_FENCE_FLAG_SIGNAL,
> +};
> +
>   /**
>    * Request queue structure.
>    *
> @@ -97,7 +108,7 @@ struct i915_request {
>   	struct intel_context *hw_context;
>   	struct intel_ring *ring;
>   	struct i915_timeline *timeline;
> -	struct intel_signal_node signaling;
> +	struct list_head signal_link;
>   
>   	/*
>   	 * The rcu epoch of when this request was allocated. Used to judiciously
> @@ -116,7 +127,6 @@ struct i915_request {
>   	 */
>   	struct i915_sw_fence submit;
>   	wait_queue_entry_t submitq;
> -	wait_queue_head_t execute;
>   
>   	/*
>   	 * A list of everyone we wait upon, and everyone who waits upon us.
> @@ -255,7 +265,7 @@ i915_request_put(struct i915_request *rq)
>    * that it has passed the global seqno and the global seqno is unchanged
>    * after the read, it is indeed complete).
>    */
> -static u32
> +static inline u32
>   i915_request_global_seqno(const struct i915_request *request)
>   {
>   	return READ_ONCE(request->global_seqno);
> @@ -277,6 +287,10 @@ void i915_request_skip(struct i915_request *request, int error);
>   void __i915_request_unsubmit(struct i915_request *request);
>   void i915_request_unsubmit(struct i915_request *request);
>   
> +/* Note: part of the intel_breadcrumbs family */
> +bool i915_request_enable_breadcrumb(struct i915_request *request);
> +void i915_request_cancel_breadcrumb(struct i915_request *request);
> +
>   long i915_request_wait(struct i915_request *rq,
>   		       unsigned int flags,
>   		       long timeout)
> @@ -293,6 +307,11 @@ static inline bool i915_request_signaled(const struct i915_request *rq)
>   	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &rq->fence.flags);
>   }
>   
> +static inline bool i915_request_is_active(const struct i915_request *rq)
> +{
> +	return test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
> +}
> +
>   /**
>    * Returns true if seq1 is later than seq2.
>    */
> @@ -330,6 +349,11 @@ static inline u32 hwsp_seqno(const struct i915_request *rq)
>   	return seqno;
>   }
>   
> +static inline bool __i915_request_has_started(const struct i915_request *rq)
> +{
> +	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
> +}
> +
>   /**
>    * i915_request_started - check if the request has begun being executed
>    * @rq: the request
> @@ -345,7 +369,23 @@ static inline bool i915_request_started(const struct i915_request *rq)
>   		return true;
>   
>   	/* Remember: started but may have since been preempted! */
> -	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
> +	return __i915_request_has_started(rq);
> +}

Is it worth having both i915_request_started and 
__i915_request_has_started. AFAIU fence cannot be signaled unless the 
seqno has advanced, so if __i915_request_has_started would be dropped 
and the caller just use the former, I don't think there would be any 
issues. Callers always check for completed before hand. I guess it saves 
a redundant fence signaled check.

> +
> +/**
> + * i915_request_is_running - check if the request may actually be executing
> + * @rq: the request
> + *
> + * Returns true if the request is currently submitted to hardware, has passed
> + * its start point (i.e. the context is setup and not busywaiting). Note that
> + * it may no longer be running by the time the function returns!
> + */
> +static inline bool i915_request_is_running(const struct i915_request *rq)
> +{
> +	if (!i915_request_is_active(rq))
> +		return false;
> +
> +	return __i915_request_has_started(rq);

return i915_request_is_active(rq) && __i915_request_has_started(rq); 
probably doesn't fit in 80chars to be that attractive?

>   }
>   
>   static inline bool i915_request_completed(const struct i915_request *rq)
> diff --git a/drivers/gpu/drm/i915/i915_reset.c b/drivers/gpu/drm/i915/i915_reset.c
> index c3861a0219a9..5d0507d88e41 100644
> --- a/drivers/gpu/drm/i915/i915_reset.c
> +++ b/drivers/gpu/drm/i915/i915_reset.c
> @@ -751,18 +751,19 @@ static void reset_restart(struct drm_i915_private *i915)
>   
>   static void nop_submit_request(struct i915_request *request)
>   {
> +	struct intel_engine_cs *engine = request->engine;
>   	unsigned long flags;
>   
>   	GEM_TRACE("%s fence %llx:%lld -> -EIO\n",
> -		  request->engine->name,
> -		  request->fence.context, request->fence.seqno);
> +		  engine->name, request->fence.context, request->fence.seqno);
>   	dma_fence_set_error(&request->fence, -EIO);
>   
> -	spin_lock_irqsave(&request->engine->timeline.lock, flags);
> +	spin_lock_irqsave(&engine->timeline.lock, flags);
>   	__i915_request_submit(request);
>   	i915_request_mark_complete(request);
> -	intel_engine_write_global_seqno(request->engine, request->global_seqno);
> -	spin_unlock_irqrestore(&request->engine->timeline.lock, flags);
> +	spin_unlock_irqrestore(&engine->timeline.lock, flags);
> +
> +	intel_engine_queue_breadcrumbs(engine);
>   }
>   
>   void i915_gem_set_wedged(struct drm_i915_private *i915)
> @@ -817,7 +818,7 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
>   
>   	for_each_engine(engine, i915, id) {
>   		reset_finish_engine(engine);
> -		intel_engine_wakeup(engine);
> +		intel_engine_signal_breadcrumbs(engine);
>   	}
>   
>   	smp_mb__before_atomic();
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index b58915b8708b..b0795b0ad227 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -29,48 +29,149 @@
>   
>   #define task_asleep(tsk) ((tsk)->state & TASK_NORMAL && !(tsk)->on_rq)
>   
> -static unsigned int __intel_breadcrumbs_wakeup(struct intel_breadcrumbs *b)
> +static void irq_enable(struct intel_engine_cs *engine)
> +{
> +	if (!engine->irq_enable)
> +		return;
> +
> +	/* Caller disables interrupts */

GEM_BUG_ON(!irqs_disabled); ?

> +	spin_lock(&engine->i915->irq_lock);
> +	engine->irq_enable(engine);
> +	spin_unlock(&engine->i915->irq_lock);
> +}
> +
> +static void irq_disable(struct intel_engine_cs *engine)
>   {
> -	struct intel_wait *wait;
> -	unsigned int result = 0;
> +	if (!engine->irq_disable)
> +		return;
> +
> +	/* Caller disables interrupts */
> +	spin_lock(&engine->i915->irq_lock);
> +	engine->irq_disable(engine);
> +	spin_unlock(&engine->i915->irq_lock);
> +}
>   
> +static void __intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
> +{
>   	lockdep_assert_held(&b->irq_lock);
>   
> -	wait = b->irq_wait;
> -	if (wait) {
> +	GEM_BUG_ON(!b->irq_enabled);
> +	if (!--b->irq_enabled)
> +		irq_disable(container_of(b,
> +					 struct intel_engine_cs,
> +					 breadcrumbs));
> +
> +	b->irq_armed = false;
> +}
> +
> +void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
> +{
> +	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +
> +	if (!b->irq_armed)
> +		return;
> +
> +	spin_lock_irq(&b->irq_lock);
> +	if (b->irq_armed)
> +		__intel_breadcrumbs_disarm_irq(b);
> +	spin_unlock_irq(&b->irq_lock);
> +}
> +
> +static inline bool __request_completed(const struct i915_request *rq)
> +{
> +	return i915_seqno_passed(__hwsp_seqno(rq), rq->fence.seqno);
> +}
> +
> +bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
> +{
> +	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	struct intel_context *ce, *cn;
> +	struct list_head *pos, *next;
> +	LIST_HEAD(signal);
> +
> +	spin_lock(&b->irq_lock);
> +
> +	b->irq_fired = true;
> +	if (b->irq_armed && list_empty(&b->signalers))
> +		__intel_breadcrumbs_disarm_irq(b);
> +
> +	list_for_each_entry_safe(ce, cn, &b->signalers, signal_link) {
> +		GEM_BUG_ON(list_empty(&ce->signals));
> +
> +		list_for_each_safe(pos, next, &ce->signals) {
> +			struct i915_request *rq =
> +				list_entry(pos, typeof(*rq), signal_link);
> +
> +			if (!__request_completed(rq))
> +				break;
> +
> +			GEM_BUG_ON(!test_bit(I915_FENCE_FLAG_SIGNAL,
> +					     &rq->fence.flags));
> +			clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
> +
> +			/*
> +			 * We may race with direct invocation of
> +			 * dma_fence_signal(), e.g. i915_request_retire(),
> +			 * in which case we can skip processing it ourselves.
> +			 */
> +			if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> +				     &rq->fence.flags))
> +				continue;

Doesn't look this is worth it. We could then race a line below and still 
end up with the already signaled rq on the local list. So would be 
tempted to drop this for simplicity.

> +
> +			/*
> +			 * Queue for execution after dropping the signaling
> +			 * spinlock as the callback chain may end up adding
> +			 * more signalers to the same context or engine.
> +			 */
> +			i915_request_get(rq);
> +			list_add_tail(&rq->signal_link, &signal);
> +		}
> +
>   		/*
> -		 * N.B. Since task_asleep() and ttwu are not atomic, the
> -		 * waiter may actually go to sleep after the check, causing
> -		 * us to suppress a valid wakeup. We prefer to reduce the
> -		 * number of false positive missed_breadcrumb() warnings
> -		 * at the expense of a few false negatives, as it it easy
> -		 * to trigger a false positive under heavy load. Enough
> -		 * signal should remain from genuine missed_breadcrumb()
> -		 * for us to detect in CI.
> +		 * We process the list deletion in bulk, only using a list_add
> +		 * (not list_move) above but keeping the status of
> +		 * rq->signal_link known with the I915_FENCE_FLAG_SIGNAL bit.
>   		 */
> -		bool was_asleep = task_asleep(wait->tsk);
> +		if (!list_is_first(pos, &ce->signals)) {
> +			/* Advance the list to the first incomplete request */
> +			__list_del_many(&ce->signals, pos);
> +			if (&ce->signals == pos) /* now empty */
> +				list_del_init(&ce->signal_link);
> +		}
> +	}
> +
> +	spin_unlock(&b->irq_lock);
> +
> +	list_for_each_safe(pos, next, &signal) {
> +		struct i915_request *rq =
> +			list_entry(pos, typeof(*rq), signal_link);
>   
> -		result = ENGINE_WAKEUP_WAITER;
> -		if (wake_up_process(wait->tsk) && was_asleep)
> -			result |= ENGINE_WAKEUP_ASLEEP;
> +		dma_fence_signal(&rq->fence);
> +		i915_request_put(rq);
>   	}
>   
> -	return result;
> +	return !list_empty(&signal);
>   }
>   
> -unsigned int intel_engine_wakeup(struct intel_engine_cs *engine)
> +bool intel_engine_signal_breadcrumbs(struct intel_engine_cs *engine)
>   {
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	unsigned long flags;
> -	unsigned int result;
> +	bool result;
>   
> -	spin_lock_irqsave(&b->irq_lock, flags);
> -	result = __intel_breadcrumbs_wakeup(b);
> -	spin_unlock_irqrestore(&b->irq_lock, flags);
> +	local_irq_disable();
> +	result = intel_engine_breadcrumbs_irq(engine);
> +	local_irq_enable();
>   
>   	return result;
>   }
>   
> +static void signal_irq_work(struct irq_work *work)
> +{
> +	struct intel_engine_cs *engine =
> +		container_of(work, typeof(*engine), breadcrumbs.irq_work);
> +
> +	intel_engine_breadcrumbs_irq(engine);
> +}
> +
>   static unsigned long wait_timeout(void)
>   {
>   	return round_jiffies_up(jiffies + DRM_I915_HANGCHECK_JIFFIES);
> @@ -94,19 +195,15 @@ static void intel_breadcrumbs_hangcheck(struct timer_list *t)
>   	struct intel_engine_cs *engine =
>   		from_timer(engine, t, breadcrumbs.hangcheck);
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	unsigned int irq_count;
>   
>   	if (!b->irq_armed)
>   		return;
>   
> -	irq_count = READ_ONCE(b->irq_count);
> -	if (b->hangcheck_interrupts != irq_count) {
> -		b->hangcheck_interrupts = irq_count;
> -		mod_timer(&b->hangcheck, wait_timeout());
> -		return;
> -	}
> +	if (b->irq_fired)
> +		goto rearm;
>   
> -	/* We keep the hangcheck timer alive until we disarm the irq, even
> +	/*
> +	 * We keep the hangcheck timer alive until we disarm the irq, even
>   	 * if there are no waiters at present.
>   	 *
>   	 * If the waiter was currently running, assume it hasn't had a chance
> @@ -118,10 +215,13 @@ static void intel_breadcrumbs_hangcheck(struct timer_list *t)
>   	 * but we still have a waiter. Assuming all batches complete within
>   	 * DRM_I915_HANGCHECK_JIFFIES [1.5s]!
>   	 */
> -	if (intel_engine_wakeup(engine) & ENGINE_WAKEUP_ASLEEP) {
> +	synchronize_hardirq(engine->i915->drm.irq);
> +	if (intel_engine_signal_breadcrumbs(engine)) {
>   		missed_breadcrumb(engine);
>   		mod_timer(&b->fake_irq, jiffies + 1);
>   	} else {
> +rearm:
> +		b->irq_fired = false;
>   		mod_timer(&b->hangcheck, wait_timeout());
>   	}
>   }
> @@ -140,11 +240,7 @@ static void intel_breadcrumbs_fake_irq(struct timer_list *t)
>   	 * oldest waiter to do the coherent seqno check.
>   	 */
>   
> -	spin_lock_irq(&b->irq_lock);
> -	if (b->irq_armed && !__intel_breadcrumbs_wakeup(b))
> -		__intel_engine_disarm_breadcrumbs(engine);
> -	spin_unlock_irq(&b->irq_lock);
> -	if (!b->irq_armed)
> +	if (!intel_engine_signal_breadcrumbs(engine) && !b->irq_armed)
>   		return;
>   
>   	/* If the user has disabled the fake-irq, restore the hangchecking */
> @@ -156,43 +252,6 @@ static void intel_breadcrumbs_fake_irq(struct timer_list *t)
>   	mod_timer(&b->fake_irq, jiffies + 1);
>   }
>   
> -static void irq_enable(struct intel_engine_cs *engine)
> -{
> -	if (!engine->irq_enable)
> -		return;
> -
> -	/* Caller disables interrupts */
> -	spin_lock(&engine->i915->irq_lock);
> -	engine->irq_enable(engine);
> -	spin_unlock(&engine->i915->irq_lock);
> -}
> -
> -static void irq_disable(struct intel_engine_cs *engine)
> -{
> -	if (!engine->irq_disable)
> -		return;
> -
> -	/* Caller disables interrupts */
> -	spin_lock(&engine->i915->irq_lock);
> -	engine->irq_disable(engine);
> -	spin_unlock(&engine->i915->irq_lock);
> -}
> -
> -void __intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	lockdep_assert_held(&b->irq_lock);
> -	GEM_BUG_ON(b->irq_wait);
> -	GEM_BUG_ON(!b->irq_armed);
> -
> -	GEM_BUG_ON(!b->irq_enabled);
> -	if (!--b->irq_enabled)
> -		irq_disable(engine);
> -
> -	b->irq_armed = false;
> -}
> -
>   void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> @@ -215,40 +274,6 @@ void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine)
>   	spin_unlock_irq(&b->irq_lock);
>   }
>   
> -void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct intel_wait *wait, *n;
> -
> -	if (!b->irq_armed)
> -		return;
> -
> -	/*
> -	 * We only disarm the irq when we are idle (all requests completed),
> -	 * so if the bottom-half remains asleep, it missed the request
> -	 * completion.
> -	 */
> -	if (intel_engine_wakeup(engine) & ENGINE_WAKEUP_ASLEEP)
> -		missed_breadcrumb(engine);
> -
> -	spin_lock_irq(&b->rb_lock);
> -
> -	spin_lock(&b->irq_lock);
> -	b->irq_wait = NULL;
> -	if (b->irq_armed)
> -		__intel_engine_disarm_breadcrumbs(engine);
> -	spin_unlock(&b->irq_lock);
> -
> -	rbtree_postorder_for_each_entry_safe(wait, n, &b->waiters, node) {
> -		GEM_BUG_ON(!intel_engine_signaled(engine, wait->seqno));
> -		RB_CLEAR_NODE(&wait->node);
> -		wake_up_process(wait->tsk);
> -	}
> -	b->waiters = RB_ROOT;
> -
> -	spin_unlock_irq(&b->rb_lock);
> -}
> -
>   static bool use_fake_irq(const struct intel_breadcrumbs *b)
>   {
>   	const struct intel_engine_cs *engine =
> @@ -264,7 +289,7 @@ static bool use_fake_irq(const struct intel_breadcrumbs *b)
>   	 * engine->seqno_barrier(), a timing error that should be transient
>   	 * and unlikely to reoccur.
>   	 */
> -	return READ_ONCE(b->irq_count) == b->hangcheck_interrupts;
> +	return !b->irq_fired;
>   }
>   
>   static void enable_fake_irq(struct intel_breadcrumbs *b)
> @@ -276,7 +301,7 @@ static void enable_fake_irq(struct intel_breadcrumbs *b)
>   		mod_timer(&b->hangcheck, wait_timeout());
>   }
>   
> -static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
> +static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
>   {
>   	struct intel_engine_cs *engine =
>   		container_of(b, struct intel_engine_cs, breadcrumbs);
> @@ -315,536 +340,149 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
>   	return enabled;
>   }
>   
> -static inline struct intel_wait *to_wait(struct rb_node *node)
> +void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
>   {
> -	return rb_entry(node, struct intel_wait, node);
> -}
> +	struct intel_breadcrumbs *b = &engine->breadcrumbs;
>   
> -static inline void __intel_breadcrumbs_finish(struct intel_breadcrumbs *b,
> -					      struct intel_wait *wait)
> -{
> -	lockdep_assert_held(&b->rb_lock);
> -	GEM_BUG_ON(b->irq_wait == wait);
> +	spin_lock_init(&b->irq_lock);
> +	INIT_LIST_HEAD(&b->signalers);
>   
> -	/*
> -	 * This request is completed, so remove it from the tree, mark it as
> -	 * complete, and *then* wake up the associated task. N.B. when the
> -	 * task wakes up, it will find the empty rb_node, discern that it
> -	 * has already been removed from the tree and skip the serialisation
> -	 * of the b->rb_lock and b->irq_lock. This means that the destruction
> -	 * of the intel_wait is not serialised with the interrupt handler
> -	 * by the waiter - it must instead be serialised by the caller.
> -	 */
> -	rb_erase(&wait->node, &b->waiters);
> -	RB_CLEAR_NODE(&wait->node);
> +	init_irq_work(&b->irq_work, signal_irq_work);
>   
> -	if (wait->tsk->state != TASK_RUNNING)
> -		wake_up_process(wait->tsk); /* implicit smp_wmb() */
> +	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
> +	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
>   }
>   
> -static inline void __intel_breadcrumbs_next(struct intel_engine_cs *engine,
> -					    struct rb_node *next)
> +static void cancel_fake_irq(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
>   
> -	spin_lock(&b->irq_lock);
> -	GEM_BUG_ON(!b->irq_armed);
> -	GEM_BUG_ON(!b->irq_wait);
> -	b->irq_wait = to_wait(next);
> -	spin_unlock(&b->irq_lock);
> -
> -	/* We always wake up the next waiter that takes over as the bottom-half
> -	 * as we may delegate not only the irq-seqno barrier to the next waiter
> -	 * but also the task of waking up concurrent waiters.
> -	 */
> -	if (next)
> -		wake_up_process(to_wait(next)->tsk);
> +	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
> +	del_timer_sync(&b->hangcheck);
> +	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
>   }
>   
> -static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
> -				    struct intel_wait *wait)
> +void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct rb_node **p, *parent, *completed;
> -	bool first, armed;
> -	u32 seqno;
> +	unsigned long flags;
>   
> -	GEM_BUG_ON(!wait->seqno);
> +	spin_lock_irqsave(&b->irq_lock, flags);
>   
> -	/* Insert the request into the retirement ordered list
> -	 * of waiters by walking the rbtree. If we are the oldest
> -	 * seqno in the tree (the first to be retired), then
> -	 * set ourselves as the bottom-half.
> -	 *
> -	 * As we descend the tree, prune completed branches since we hold the
> -	 * spinlock we know that the first_waiter must be delayed and can
> -	 * reduce some of the sequential wake up latency if we take action
> -	 * ourselves and wake up the completed tasks in parallel. Also, by
> -	 * removing stale elements in the tree, we may be able to reduce the
> -	 * ping-pong between the old bottom-half and ourselves as first-waiter.
> +	/*
> +	 * Leave the fake_irq timer enabled (if it is running), but clear the
> +	 * bit so that it turns itself off on its next wake up and goes back
> +	 * to the long hangcheck interval if still required.
>   	 */
> -	armed = false;
> -	first = true;
> -	parent = NULL;
> -	completed = NULL;
> -	seqno = intel_engine_get_seqno(engine);
> -
> -	 /* If the request completed before we managed to grab the spinlock,
> -	  * return now before adding ourselves to the rbtree. We let the
> -	  * current bottom-half handle any pending wakeups and instead
> -	  * try and get out of the way quickly.
> -	  */
> -	if (i915_seqno_passed(seqno, wait->seqno)) {
> -		RB_CLEAR_NODE(&wait->node);
> -		return first;
> -	}
> -
> -	p = &b->waiters.rb_node;
> -	while (*p) {
> -		parent = *p;
> -		if (wait->seqno == to_wait(parent)->seqno) {
> -			/* We have multiple waiters on the same seqno, select
> -			 * the highest priority task (that with the smallest
> -			 * task->prio) to serve as the bottom-half for this
> -			 * group.
> -			 */
> -			if (wait->tsk->prio > to_wait(parent)->tsk->prio) {
> -				p = &parent->rb_right;
> -				first = false;
> -			} else {
> -				p = &parent->rb_left;
> -			}
> -		} else if (i915_seqno_passed(wait->seqno,
> -					     to_wait(parent)->seqno)) {
> -			p = &parent->rb_right;
> -			if (i915_seqno_passed(seqno, to_wait(parent)->seqno))
> -				completed = parent;
> -			else
> -				first = false;
> -		} else {
> -			p = &parent->rb_left;
> -		}
> -	}
> -	rb_link_node(&wait->node, parent, p);
> -	rb_insert_color(&wait->node, &b->waiters);
> -
> -	if (first) {
> -		spin_lock(&b->irq_lock);
> -		b->irq_wait = wait;
> -		/* After assigning ourselves as the new bottom-half, we must
> -		 * perform a cursory check to prevent a missed interrupt.
> -		 * Either we miss the interrupt whilst programming the hardware,
> -		 * or if there was a previous waiter (for a later seqno) they
> -		 * may be woken instead of us (due to the inherent race
> -		 * in the unlocked read of b->irq_seqno_bh in the irq handler)
> -		 * and so we miss the wake up.
> -		 */
> -		armed = __intel_breadcrumbs_enable_irq(b);
> -		spin_unlock(&b->irq_lock);
> -	}
> -
> -	if (completed) {
> -		/* Advance the bottom-half (b->irq_wait) before we wake up
> -		 * the waiters who may scribble over their intel_wait
> -		 * just as the interrupt handler is dereferencing it via
> -		 * b->irq_wait.
> -		 */
> -		if (!first) {
> -			struct rb_node *next = rb_next(completed);
> -			GEM_BUG_ON(next == &wait->node);
> -			__intel_breadcrumbs_next(engine, next);
> -		}
> -
> -		do {
> -			struct intel_wait *crumb = to_wait(completed);
> -			completed = rb_prev(completed);
> -			__intel_breadcrumbs_finish(b, crumb);
> -		} while (completed);
> -	}
> -
> -	GEM_BUG_ON(!b->irq_wait);
> -	GEM_BUG_ON(!b->irq_armed);
> -	GEM_BUG_ON(rb_first(&b->waiters) != &b->irq_wait->node);
> -
> -	return armed;
> -}
> -
> -bool intel_engine_add_wait(struct intel_engine_cs *engine,
> -			   struct intel_wait *wait)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	bool armed;
> -
> -	spin_lock_irq(&b->rb_lock);
> -	armed = __intel_engine_add_wait(engine, wait);
> -	spin_unlock_irq(&b->rb_lock);
> -	if (armed)
> -		return armed;
> -
> -	/* Make the caller recheck if its request has already started. */
> -	return intel_engine_has_started(engine, wait->seqno);
> -}
> -
> -static inline bool chain_wakeup(struct rb_node *rb, int priority)
> -{
> -	return rb && to_wait(rb)->tsk->prio <= priority;
> -}
> +	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
>   
> -static inline int wakeup_priority(struct intel_breadcrumbs *b,
> -				  struct task_struct *tsk)
> -{
> -	if (tsk == b->signaler)
> -		return INT_MIN;
> +	if (b->irq_enabled)
> +		irq_enable(engine);
>   	else
> -		return tsk->prio;
> -}
> -
> -static void __intel_engine_remove_wait(struct intel_engine_cs *engine,
> -				       struct intel_wait *wait)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	lockdep_assert_held(&b->rb_lock);
> -
> -	if (RB_EMPTY_NODE(&wait->node))
> -		goto out;
> -
> -	if (b->irq_wait == wait) {
> -		const int priority = wakeup_priority(b, wait->tsk);
> -		struct rb_node *next;
> -
> -		/* We are the current bottom-half. Find the next candidate,
> -		 * the first waiter in the queue on the remaining oldest
> -		 * request. As multiple seqnos may complete in the time it
> -		 * takes us to wake up and find the next waiter, we have to
> -		 * wake up that waiter for it to perform its own coherent
> -		 * completion check.
> -		 */
> -		next = rb_next(&wait->node);
> -		if (chain_wakeup(next, priority)) {
> -			/* If the next waiter is already complete,
> -			 * wake it up and continue onto the next waiter. So
> -			 * if have a small herd, they will wake up in parallel
> -			 * rather than sequentially, which should reduce
> -			 * the overall latency in waking all the completed
> -			 * clients.
> -			 *
> -			 * However, waking up a chain adds extra latency to
> -			 * the first_waiter. This is undesirable if that
> -			 * waiter is a high priority task.
> -			 */
> -			u32 seqno = intel_engine_get_seqno(engine);
> -
> -			while (i915_seqno_passed(seqno, to_wait(next)->seqno)) {
> -				struct rb_node *n = rb_next(next);
> -
> -				__intel_breadcrumbs_finish(b, to_wait(next));
> -				next = n;
> -				if (!chain_wakeup(next, priority))
> -					break;
> -			}
> -		}
> -
> -		__intel_breadcrumbs_next(engine, next);
> -	} else {
> -		GEM_BUG_ON(rb_first(&b->waiters) == &wait->node);
> -	}
> -
> -	GEM_BUG_ON(RB_EMPTY_NODE(&wait->node));
> -	rb_erase(&wait->node, &b->waiters);
> -	RB_CLEAR_NODE(&wait->node);
> +		irq_disable(engine);
>   
> -out:
> -	GEM_BUG_ON(b->irq_wait == wait);
> -	GEM_BUG_ON(rb_first(&b->waiters) !=
> -		   (b->irq_wait ? &b->irq_wait->node : NULL));
> +	spin_unlock_irqrestore(&b->irq_lock, flags);
>   }
>   
> -void intel_engine_remove_wait(struct intel_engine_cs *engine,
> -			      struct intel_wait *wait)
> +void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
>   {
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	/* Quick check to see if this waiter was already decoupled from
> -	 * the tree by the bottom-half to avoid contention on the spinlock
> -	 * by the herd.
> -	 */
> -	if (RB_EMPTY_NODE(&wait->node)) {
> -		GEM_BUG_ON(READ_ONCE(b->irq_wait) == wait);
> -		return;
> -	}
> -
> -	spin_lock_irq(&b->rb_lock);
> -	__intel_engine_remove_wait(engine, wait);
> -	spin_unlock_irq(&b->rb_lock);
> +	cancel_fake_irq(engine);
>   }
>   
> -static void signaler_set_rtpriority(void)
> +bool i915_request_enable_breadcrumb(struct i915_request *rq)
>   {
> -	 struct sched_param param = { .sched_priority = 1 };
> -
> -	 sched_setscheduler_nocheck(current, SCHED_FIFO, &param);
> -}
> +	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
>   
> -static int intel_breadcrumbs_signaler(void *arg)
> -{
> -	struct intel_engine_cs *engine = arg;
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct i915_request *rq, *n;
> +	GEM_BUG_ON(test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags));
>   
> -	/* Install ourselves with high priority to reduce signalling latency */
> -	signaler_set_rtpriority();
> +	if (!test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags))
> +		return true;
>   
> -	do {
> -		bool do_schedule = true;
> -		LIST_HEAD(list);
> -		u32 seqno;
> +	spin_lock(&b->irq_lock);
> +	if (test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags) &&
> +	    !__request_completed(rq)) {
> +		struct intel_context *ce = rq->hw_context;
> +		struct list_head *pos;
>   
> -		set_current_state(TASK_INTERRUPTIBLE);
> -		if (list_empty(&b->signals))
> -			goto sleep;
> +		__intel_breadcrumbs_arm_irq(b);
>   
>   		/*
> -		 * We are either woken up by the interrupt bottom-half,
> -		 * or by a client adding a new signaller. In both cases,
> -		 * the GPU seqno may have advanced beyond our oldest signal.
> -		 * If it has, propagate the signal, remove the waiter and
> -		 * check again with the next oldest signal. Otherwise we
> -		 * need to wait for a new interrupt from the GPU or for
> -		 * a new client.
> +		 * We keep the seqno in retirement order, so we can break
> +		 * inside intel_engine_breadcrumbs_irq as soon as we've passed
> +		 * the last completed request (or seen a request that hasn't
> +		 * event started). We could iterate the timeline->requests list,
> +		 * but keeping a separate signalers_list has the advantage of
> +		 * hopefully being much smaller than the full list and so
> +		 * provides faster iteration and detection when there are no
> +		 * more interrupts required for this context.
> +		 *
> +		 * We typically expect to add new signalers in order, so we
> +		 * start looking for our insertion point from the tail of
> +		 * the list.
>   		 */
> -		seqno = intel_engine_get_seqno(engine);
> -
> -		spin_lock_irq(&b->rb_lock);
> -		list_for_each_entry_safe(rq, n, &b->signals, signaling.link) {
> -			u32 this = rq->signaling.wait.seqno;
> +		list_for_each_prev(pos, &ce->signals) {
> +			struct i915_request *it =
> +				list_entry(pos, typeof(*it), signal_link);
>   
> -			GEM_BUG_ON(!rq->signaling.wait.seqno);
> -
> -			if (!i915_seqno_passed(seqno, this))
> +			if (i915_seqno_passed(rq->fence.seqno, it->fence.seqno))
>   				break;
> -
> -			if (likely(this == i915_request_global_seqno(rq))) {
> -				__intel_engine_remove_wait(engine,
> -							   &rq->signaling.wait);
> -
> -				rq->signaling.wait.seqno = 0;
> -				__list_del_entry(&rq->signaling.link);
> -
> -				if (!i915_request_signaled(rq)) {
> -					list_add_tail(&rq->signaling.link,
> -						      &list);
> -					i915_request_get(rq);
> -				}
> -			}
>   		}
> -		spin_unlock_irq(&b->rb_lock);
> -
> -		if (!list_empty(&list)) {
> -			local_bh_disable();
> -			list_for_each_entry_safe(rq, n, &list, signaling.link) {
> -				dma_fence_signal(&rq->fence);
> -				GEM_BUG_ON(!i915_request_completed(rq));
> -				i915_request_put(rq);
> -			}
> -			local_bh_enable(); /* kick start the tasklets */
> -
> -			/*
> -			 * If the engine is saturated we may be continually
> -			 * processing completed requests. This angers the
> -			 * NMI watchdog if we never let anything else
> -			 * have access to the CPU. Let's pretend to be nice
> -			 * and relinquish the CPU if we burn through the
> -			 * entire RT timeslice!
> -			 */
> -			do_schedule = need_resched();
> -		}
> -
> -		if (unlikely(do_schedule)) {
> -sleep:
> -			if (kthread_should_park())
> -				kthread_parkme();
> -
> -			if (unlikely(kthread_should_stop()))
> -				break;
> -
> -			schedule();
> -		}
> -	} while (1);
> -	__set_current_state(TASK_RUNNING);
> -
> -	return 0;
> -}
> +		list_add(&rq->signal_link, pos);
> +		if (pos == &ce->signals) /* catch transitions from empty list */
> +			list_move_tail(&ce->signal_link, &b->signalers);
>   
> -static void insert_signal(struct intel_breadcrumbs *b,
> -			  struct i915_request *request,
> -			  const u32 seqno)
> -{
> -	struct i915_request *iter;
> -
> -	lockdep_assert_held(&b->rb_lock);
> -
> -	/*
> -	 * A reasonable assumption is that we are called to add signals
> -	 * in sequence, as the requests are submitted for execution and
> -	 * assigned a global_seqno. This will be the case for the majority
> -	 * of internally generated signals (inter-engine signaling).
> -	 *
> -	 * Out of order waiters triggering random signaling enabling will
> -	 * be more problematic, but hopefully rare enough and the list
> -	 * small enough that the O(N) insertion sort is not an issue.
> -	 */
> -
> -	list_for_each_entry_reverse(iter, &b->signals, signaling.link)
> -		if (i915_seqno_passed(seqno, iter->signaling.wait.seqno))
> -			break;
> -
> -	list_add(&request->signaling.link, &iter->signaling.link);
> -}
> -
> -bool intel_engine_enable_signaling(struct i915_request *request, bool wakeup)
> -{
> -	struct intel_engine_cs *engine = request->engine;
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct intel_wait *wait = &request->signaling.wait;
> -	u32 seqno;
> -
> -	/*
> -	 * Note that we may be called from an interrupt handler on another
> -	 * device (e.g. nouveau signaling a fence completion causing us
> -	 * to submit a request, and so enable signaling). As such,
> -	 * we need to make sure that all other users of b->rb_lock protect
> -	 * against interrupts, i.e. use spin_lock_irqsave.
> -	 */
> -
> -	/* locked by dma_fence_enable_sw_signaling() (irqsafe fence->lock) */
> -	GEM_BUG_ON(!irqs_disabled());
> -	lockdep_assert_held(&request->lock);
> -
> -	seqno = i915_request_global_seqno(request);
> -	if (!seqno) /* will be enabled later upon execution */
> -		return true;
> -
> -	GEM_BUG_ON(wait->seqno);
> -	wait->tsk = b->signaler;
> -	wait->request = request;
> -	wait->seqno = seqno;
> -
> -	/*
> -	 * Add ourselves into the list of waiters, but registering our
> -	 * bottom-half as the signaller thread. As per usual, only the oldest
> -	 * waiter (not just signaller) is tasked as the bottom-half waking
> -	 * up all completed waiters after the user interrupt.
> -	 *
> -	 * If we are the oldest waiter, enable the irq (after which we
> -	 * must double check that the seqno did not complete).
> -	 */
> -	spin_lock(&b->rb_lock);
> -	insert_signal(b, request, seqno);
> -	wakeup &= __intel_engine_add_wait(engine, wait);
> -	spin_unlock(&b->rb_lock);
> -
> -	if (wakeup) {
> -		wake_up_process(b->signaler);
> -		return !intel_wait_complete(wait);
> +		set_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
>   	}
> +	spin_unlock(&b->irq_lock);
>   
> -	return true;
> +	return !__request_completed(rq);
>   }
>   
> -void intel_engine_cancel_signaling(struct i915_request *request)
> +void i915_request_cancel_breadcrumb(struct i915_request *rq)
>   {
> -	struct intel_engine_cs *engine = request->engine;
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	struct intel_breadcrumbs *b = &rq->engine->breadcrumbs;
>   
> -	GEM_BUG_ON(!irqs_disabled());
> -	lockdep_assert_held(&request->lock);
> -
> -	if (!READ_ONCE(request->signaling.wait.seqno))
> +	if (!test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags))
>   		return;
>   
> -	spin_lock(&b->rb_lock);
> -	__intel_engine_remove_wait(engine, &request->signaling.wait);
> -	if (fetch_and_zero(&request->signaling.wait.seqno))
> -		__list_del_entry(&request->signaling.link);
> -	spin_unlock(&b->rb_lock);
> -}
> -
> -int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct task_struct *tsk;
> -
> -	spin_lock_init(&b->rb_lock);
> -	spin_lock_init(&b->irq_lock);
> -
> -	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
> -	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
> -
> -	INIT_LIST_HEAD(&b->signals);
> -
> -	/* Spawn a thread to provide a common bottom-half for all signals.
> -	 * As this is an asynchronous interface we cannot steal the current
> -	 * task for handling the bottom-half to the user interrupt, therefore
> -	 * we create a thread to do the coherent seqno dance after the
> -	 * interrupt and then signal the waitqueue (via the dma-buf/fence).
> -	 */
> -	tsk = kthread_run(intel_breadcrumbs_signaler, engine,
> -			  "i915/signal:%d", engine->id);
> -	if (IS_ERR(tsk))
> -		return PTR_ERR(tsk);
> -
> -	b->signaler = tsk;
> -
> -	return 0;
> -}
> +	spin_lock(&b->irq_lock);
> +	if (test_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags)) {
> +		struct intel_context *ce = rq->hw_context;
>   
> -static void cancel_fake_irq(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +		list_del(&rq->signal_link);
> +		if (list_empty(&ce->signals))
> +			list_del_init(&ce->signal_link);
>   
> -	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
> -	del_timer_sync(&b->hangcheck);
> -	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
> +		clear_bit(I915_FENCE_FLAG_SIGNAL, &rq->fence.flags);
> +	}
> +	spin_unlock(&b->irq_lock);
>   }
>   
> -void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
> +void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
> +				    struct drm_printer *p)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	unsigned long flags;
> +	struct intel_context *ce;
> +	struct i915_request *rq;
>   
> -	spin_lock_irqsave(&b->irq_lock, flags);
> -
> -	/*
> -	 * Leave the fake_irq timer enabled (if it is running), but clear the
> -	 * bit so that it turns itself off on its next wake up and goes back
> -	 * to the long hangcheck interval if still required.
> -	 */
> -	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
> -
> -	if (b->irq_enabled)
> -		irq_enable(engine);
> -	else
> -		irq_disable(engine);
> -
> -	spin_unlock_irqrestore(&b->irq_lock, flags);
> -}
> -
> -void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	if (list_empty(&b->signalers))
> +		return;
>   
> -	/* The engines should be idle and all requests accounted for! */
> -	WARN_ON(READ_ONCE(b->irq_wait));
> -	WARN_ON(!RB_EMPTY_ROOT(&b->waiters));
> -	WARN_ON(!list_empty(&b->signals));
> +	drm_printf(p, "Signals:\n");
>   
> -	if (!IS_ERR_OR_NULL(b->signaler))
> -		kthread_stop(b->signaler);
> +	spin_lock_irq(&b->irq_lock);
> +	list_for_each_entry(ce, &b->signalers, signal_link) {
> +		list_for_each_entry(rq, &ce->signals, signal_link) {
> +			drm_printf(p, "\t[%llx:%llx%s] @ %dms\n",
> +				   rq->fence.context, rq->fence.seqno,
> +				   i915_request_completed(rq) ? "!" :
> +				   i915_request_started(rq) ? "*" :
> +				   "",
> +				   jiffies_to_msecs(jiffies - rq->emitted_jiffies));
> +		}
> +	}
> +	spin_unlock_irq(&b->irq_lock);
>   
> -	cancel_fake_irq(engine);
> +	if (test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
> +		drm_printf(p, "Fake irq active\n");
>   }
> -
> -#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
> -#include "selftests/intel_breadcrumbs.c"
> -#endif
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index b35ba4f03079..d093e4552a4c 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -458,12 +458,6 @@ int intel_engines_init(struct drm_i915_private *dev_priv)
>   void intel_engine_write_global_seqno(struct intel_engine_cs *engine, u32 seqno)
>   {
>   	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
> -
> -	/* After manually advancing the seqno, fake the interrupt in case
> -	 * there are any waiters for that seqno.
> -	 */
> -	intel_engine_wakeup(engine);
> -
>   	GEM_BUG_ON(intel_engine_get_seqno(engine) != seqno);
>   }
>   
> @@ -607,6 +601,7 @@ int intel_engine_setup_common(struct intel_engine_cs *engine)
>   
>   	i915_timeline_set_subclass(&engine->timeline, TIMELINE_ENGINE);
>   
> +	intel_engine_init_breadcrumbs(engine);
>   	intel_engine_init_execlist(engine);
>   	intel_engine_init_hangcheck(engine);
>   	intel_engine_init_batch_pool(engine);
> @@ -717,20 +712,14 @@ int intel_engine_init_common(struct intel_engine_cs *engine)
>   		}
>   	}
>   
> -	ret = intel_engine_init_breadcrumbs(engine);
> -	if (ret)
> -		goto err_unpin_preempt;
> -
>   	ret = measure_breadcrumb_dw(engine);
>   	if (ret < 0)
> -		goto err_breadcrumbs;
> +		goto err_unpin_preempt;
>   
>   	engine->emit_fini_breadcrumb_dw = ret;
>   
>   	return 0;
>   
> -err_breadcrumbs:
> -	intel_engine_fini_breadcrumbs(engine);
>   err_unpin_preempt:
>   	if (i915->preempt_context)
>   		__intel_context_unpin(i915->preempt_context, engine);
> @@ -1294,12 +1283,14 @@ static void print_request(struct drm_printer *m,
>   
>   	x = print_sched_attr(rq->i915, &rq->sched.attr, buf, x, sizeof(buf));
>   
> -	drm_printf(m, "%s%x%s [%llx:%llx]%s @ %dms: %s\n",
> +	drm_printf(m, "%s%x%s%s [%llx:%llx]%s @ %dms: %s\n",
>   		   prefix,
>   		   rq->global_seqno,
>   		   i915_request_completed(rq) ? "!" :
>   		   i915_request_started(rq) ? "*" :
>   		   "",
> +		   test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
> +			    &rq->fence.flags) ?  "+" : "",
>   		   rq->fence.context, rq->fence.seqno,
>   		   buf,
>   		   jiffies_to_msecs(jiffies - rq->emitted_jiffies),
> @@ -1492,12 +1483,9 @@ void intel_engine_dump(struct intel_engine_cs *engine,
>   		       struct drm_printer *m,
>   		       const char *header, ...)
>   {
> -	struct intel_breadcrumbs * const b = &engine->breadcrumbs;
>   	struct i915_gpu_error * const error = &engine->i915->gpu_error;
>   	struct i915_request *rq;
>   	intel_wakeref_t wakeref;
> -	unsigned long flags;
> -	struct rb_node *rb;
>   
>   	if (header) {
>   		va_list ap;
> @@ -1565,21 +1553,12 @@ void intel_engine_dump(struct intel_engine_cs *engine,
>   
>   	intel_execlists_show_requests(engine, m, print_request, 8);
>   
> -	spin_lock_irqsave(&b->rb_lock, flags);
> -	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
> -		struct intel_wait *w = rb_entry(rb, typeof(*w), node);
> -
> -		drm_printf(m, "\t%s [%d:%c] waiting for %x\n",
> -			   w->tsk->comm, w->tsk->pid,
> -			   task_state_to_char(w->tsk),
> -			   w->seqno);
> -	}
> -	spin_unlock_irqrestore(&b->rb_lock, flags);
> -
>   	drm_printf(m, "HWSP:\n");
>   	hexdump(m, engine->status_page.addr, PAGE_SIZE);
>   
>   	drm_printf(m, "Idle? %s\n", yesno(intel_engine_is_idle(engine)));
> +
> +	intel_engine_print_breadcrumbs(engine, m);
>   }
>   
>   static u8 user_class_map[] = {
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> index 3f9f0a5a0f44..d7c31ad1d0ea 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> @@ -486,8 +486,8 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
>   
>   	for (i = 0; i < GEN7_XCS_WA; i++) {
>   		*cs++ = MI_STORE_DWORD_INDEX;
> -		*cs++ = I915_GEM_HWS_INDEX_ADDR;
> -		*cs++ = rq->global_seqno;
> +		*cs++ = I915_GEM_HWS_SEQNO_ADDR;
> +		*cs++ = rq->fence.seqno;

Why this, and in this patch?

>   	}
>   
>   	*cs++ = MI_FLUSH_DW;
> @@ -738,7 +738,7 @@ static int init_ring_common(struct intel_engine_cs *engine)
>   	}
>   
>   	/* Papering over lost _interrupts_ immediately following the restart */
> -	intel_engine_wakeup(engine);
> +	intel_engine_queue_breadcrumbs(engine);
>   out:
>   	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
>   
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 5ed06d4230f6..eb0f8667a246 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -5,6 +5,7 @@
>   #include <drm/drm_util.h>
>   
>   #include <linux/hashtable.h>
> +#include <linux/irq_work.h>
>   #include <linux/seqlock.h>
>   
>   #include "i915_gem_batch_pool.h"
> @@ -376,22 +377,19 @@ struct intel_engine_cs {
>   	 * the overhead of waking that client is much preferred.
>   	 */
>   	struct intel_breadcrumbs {
> -		spinlock_t irq_lock; /* protects irq_*; irqsafe */
> -		struct intel_wait *irq_wait; /* oldest waiter by retirement */
> +		spinlock_t irq_lock;
> +		struct list_head signalers;
>   
> -		spinlock_t rb_lock; /* protects the rb and wraps irq_lock */
> -		struct rb_root waiters; /* sorted by retirement, priority */
> -		struct list_head signals; /* sorted by retirement */
> -		struct task_struct *signaler; /* used for fence signalling */
> +		struct irq_work irq_work; /* for use from inside irq_lock */
>   
>   		struct timer_list fake_irq; /* used after a missed interrupt */
>   		struct timer_list hangcheck; /* detect missed interrupts */
>   
>   		unsigned int hangcheck_interrupts;
>   		unsigned int irq_enabled;
> -		unsigned int irq_count;
>   
> -		bool irq_armed : 1;
> +		bool irq_armed;
> +		bool irq_fired;
>   	} breadcrumbs;
>   
>   	struct {
> @@ -881,83 +879,29 @@ static inline bool intel_engine_has_started(struct intel_engine_cs *engine,
>   void intel_engine_get_instdone(struct intel_engine_cs *engine,
>   			       struct intel_instdone *instdone);
>   
> -/* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
> -int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
> -
> -static inline void intel_wait_init(struct intel_wait *wait)
> -{
> -	wait->tsk = current;
> -	wait->request = NULL;
> -}
> -
> -static inline void intel_wait_init_for_seqno(struct intel_wait *wait, u32 seqno)
> -{
> -	wait->tsk = current;
> -	wait->seqno = seqno;
> -}
> -
> -static inline bool intel_wait_has_seqno(const struct intel_wait *wait)
> -{
> -	return wait->seqno;
> -}
> -
> -static inline bool
> -intel_wait_update_seqno(struct intel_wait *wait, u32 seqno)
> -{
> -	wait->seqno = seqno;
> -	return intel_wait_has_seqno(wait);
> -}
> -
> -static inline bool
> -intel_wait_update_request(struct intel_wait *wait,
> -			  const struct i915_request *rq)
> -{
> -	return intel_wait_update_seqno(wait, i915_request_global_seqno(rq));
> -}
> -
> -static inline bool
> -intel_wait_check_seqno(const struct intel_wait *wait, u32 seqno)
> -{
> -	return wait->seqno == seqno;
> -}
> -
> -static inline bool
> -intel_wait_check_request(const struct intel_wait *wait,
> -			 const struct i915_request *rq)
> -{
> -	return intel_wait_check_seqno(wait, i915_request_global_seqno(rq));
> -}
> +void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
> +void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
>   
> -static inline bool intel_wait_complete(const struct intel_wait *wait)
> -{
> -	return RB_EMPTY_NODE(&wait->node);
> -}
> +void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine);
> +void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine);
>   
> -bool intel_engine_add_wait(struct intel_engine_cs *engine,
> -			   struct intel_wait *wait);
> -void intel_engine_remove_wait(struct intel_engine_cs *engine,
> -			      struct intel_wait *wait);
> -bool intel_engine_enable_signaling(struct i915_request *request, bool wakeup);
> -void intel_engine_cancel_signaling(struct i915_request *request);
> +bool intel_engine_signal_breadcrumbs(struct intel_engine_cs *engine);
> +void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
>   
> -static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
> +static inline void
> +intel_engine_queue_breadcrumbs(struct intel_engine_cs *engine)
>   {
> -	return READ_ONCE(engine->breadcrumbs.irq_wait);
> +	irq_work_queue(&engine->breadcrumbs.irq_work);
>   }
>   
> -unsigned int intel_engine_wakeup(struct intel_engine_cs *engine);
> -#define ENGINE_WAKEUP_WAITER BIT(0)
> -#define ENGINE_WAKEUP_ASLEEP BIT(1)
> -
> -void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine);
> -void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine);
> -
> -void __intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
> -void intel_engine_disarm_breadcrumbs(struct intel_engine_cs *engine);
> +bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine);
>   
>   void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine);
>   void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
>   
> +void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
> +				    struct drm_printer *p);
> +
>   static inline u32 *gen8_emit_pipe_control(u32 *batch, u32 flags, u32 offset)
>   {
>   	memset(batch, 0, 6 * sizeof(u32));
> diff --git a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> index 4a83a1c6c406..88e5ab586337 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> +++ b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> @@ -15,7 +15,6 @@ selftest(scatterlist, scatterlist_mock_selftests)
>   selftest(syncmap, i915_syncmap_mock_selftests)
>   selftest(uncore, intel_uncore_mock_selftests)
>   selftest(engine, intel_engine_cs_mock_selftests)
> -selftest(breadcrumbs, intel_breadcrumbs_mock_selftests)
>   selftest(timelines, i915_timeline_mock_selftests)
>   selftest(requests, i915_request_mock_selftests)
>   selftest(objects, i915_gem_object_mock_selftests)
> diff --git a/drivers/gpu/drm/i915/selftests/i915_request.c b/drivers/gpu/drm/i915/selftests/i915_request.c
> index 4d4b86b5fa11..5e4702a55688 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_request.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_request.c
> @@ -25,9 +25,12 @@
>   #include <linux/prime_numbers.h>
>   
>   #include "../i915_selftest.h"
> +#include "i915_random.h"
>   #include "igt_live_test.h"
> +#include "lib_sw_fence.h"
>   
>   #include "mock_context.h"
> +#include "mock_drm.h"
>   #include "mock_gem_device.h"
>   
>   static int igt_add_request(void *arg)
> @@ -247,6 +250,253 @@ static int igt_request_rewind(void *arg)
>   	return err;
>   }
>   
> +struct smoketest {
> +	struct intel_engine_cs *engine;
> +	struct i915_gem_context **contexts;
> +	unsigned int ncontexts, max_batch;
> +	atomic_long_t num_waits, num_fences;
> +	struct i915_request *(*request_alloc)(struct i915_gem_context *,
> +					      struct intel_engine_cs *);
> +};
> +
> +static struct i915_request *
> +__mock_request_alloc(struct i915_gem_context *ctx,
> +		     struct intel_engine_cs *engine)
> +{
> +	return mock_request(engine, ctx, 0);
> +}
> +
> +static struct i915_request *
> +__live_request_alloc(struct i915_gem_context *ctx,
> +		     struct intel_engine_cs *engine)
> +{
> +	return i915_request_alloc(engine, ctx);
> +}
> +
> +static int __igt_breadcrumbs_smoketest(void *arg)
> +{
> +	struct smoketest *t = arg;
> +	struct mutex * const BKL = &t->engine->i915->drm.struct_mutex;
> +	const unsigned int max_batch = min(t->ncontexts, t->max_batch) - 1;
> +	const unsigned int total = 4 * t->ncontexts + 1;
> +	unsigned int num_waits = 0, num_fences = 0;
> +	struct i915_request **requests;
> +	I915_RND_STATE(prng);
> +	unsigned int *order;
> +	int err = 0;
> +
> +	/*
> +	 * A very simple test to catch the most egregious of list handling bugs.
> +	 *
> +	 * At its heart, we simple create oodles of requests running across
> +	 * multiple kthreads and enable signaling on them, for the sole purpose
> +	 * of stressing our breadcrumb handling. The only inspection we do is
> +	 * that the fences were marked as signaled.
> +	 */
> +
> +	requests = kmalloc_array(total, sizeof(*requests), GFP_KERNEL);
> +	if (!requests)
> +		return -ENOMEM;
> +
> +	order = i915_random_order(total, &prng);
> +	if (!order) {
> +		err = -ENOMEM;
> +		goto out_requests;
> +	}
> +
> +	while (!kthread_should_stop()) {
> +		struct i915_sw_fence *submit, *wait;
> +		unsigned int n, count;
> +
> +		submit = heap_fence_create(GFP_KERNEL);
> +		if (!submit) {
> +			err = -ENOMEM;
> +			break;
> +		}
> +
> +		wait = heap_fence_create(GFP_KERNEL);
> +		if (!wait) {
> +			i915_sw_fence_commit(submit);
> +			heap_fence_put(submit);
> +			err = ENOMEM;
> +			break;
> +		}
> +
> +		i915_random_reorder(order, total, &prng);
> +		count = 1 + i915_prandom_u32_max_state(max_batch, &prng);
> +
> +		for (n = 0; n < count; n++) {
> +			struct i915_gem_context *ctx =
> +				t->contexts[order[n] % t->ncontexts];
> +			struct i915_request *rq;
> +
> +			mutex_lock(BKL);
> +
> +			rq = t->request_alloc(ctx, t->engine);
> +			if (IS_ERR(rq)) {
> +				mutex_unlock(BKL);
> +				err = PTR_ERR(rq);
> +				count = n;
> +				break;
> +			}
> +
> +			err = i915_sw_fence_await_sw_fence_gfp(&rq->submit,
> +							       submit,
> +							       GFP_KERNEL);

Submit fence is just to ensure request doesn't finish before you can 
enable signaling on it, via the wait fence?

> +
> +			requests[n] = i915_request_get(rq);
> +			i915_request_add(rq);
> +
> +			mutex_unlock(BKL);
> +
> +			if (err >= 0)
> +				err = i915_sw_fence_await_dma_fence(wait,
> +								    &rq->fence,
> +								    0,
> +								    GFP_KERNEL);

If above is true why this can't simply be i915_request_enable_breadcrumbs?

> +			if (err < 0) {

else if?

> +				i915_request_put(rq);
> +				count = n;
> +				break;
> +			}
> +		}
> +
> +		i915_sw_fence_commit(submit);
> +		i915_sw_fence_commit(wait);
> +
> +		if (!wait_event_timeout(wait->wait,
> +					i915_sw_fence_done(wait),
> +					HZ / 2)) {
> +			struct i915_request *rq = requests[count - 1];
> +
> +			pr_err("waiting for %d fences (last %llx:%lld) on %s timed out!\n",
> +			       count,
> +			       rq->fence.context, rq->fence.seqno,
> +			       t->engine->name);
> +			i915_gem_set_wedged(t->engine->i915);
> +			GEM_BUG_ON(!i915_request_completed(rq));
> +			i915_sw_fence_wait(wait);
> +			err = -EIO;
> +		}
> +
> +		for (n = 0; n < count; n++) {
> +			struct i915_request *rq = requests[n];
> +
> +			if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> +				      &rq->fence.flags)) {
> +				pr_err("%llu:%llu was not signaled!\n",
> +				       rq->fence.context, rq->fence.seqno);
> +				err = -EINVAL;
> +			}
> +
> +			i915_request_put(rq);
> +		}
> +
> +		heap_fence_put(wait);
> +		heap_fence_put(submit);
> +
> +		if (err < 0)
> +			break;
> +
> +		num_fences += count;

Isn't num_fences actually also growing by one, while num_requests would 
be growing by count?

> +		num_waits++;
> +
> +		cond_resched();

There is wait_event_timeout in the loop so this shouldn't be needed. 
AFAICS is something fails loop bails out.

> +	}
> +
> +	atomic_long_add(num_fences, &t->num_fences);
> +	atomic_long_add(num_waits, &t->num_waits);
> +
> +	kfree(order);
> +out_requests:
> +	kfree(requests);
> +	return err;
> +}
> +
> +static int mock_breadcrumbs_smoketest(void *arg)
> +{
> +	struct drm_i915_private *i915 = arg;
> +	struct smoketest t = {
> +		.engine = i915->engine[RCS],
> +		.ncontexts = 1024,
> +		.max_batch = 1024,
> +		.request_alloc = __mock_request_alloc
> +	};
> +	unsigned int ncpus = num_online_cpus();
> +	struct task_struct **threads;
> +	unsigned int n;
> +	int ret = 0;
> +
> +	/*
> +	 * Smoketest our breadcrumb/signal handling for requests across multiple
> +	 * threads. A very simple test to only catch the most egregious of bugs.
> +	 * See __igt_breadcrumbs_smoketest();
> +	 */
> +
> +	threads = kmalloc_array(ncpus, sizeof(*threads), GFP_KERNEL);
> +	if (!threads)
> +		return -ENOMEM;
> +
> +	t.contexts =
> +		kmalloc_array(t.ncontexts, sizeof(*t.contexts), GFP_KERNEL);
> +	if (!t.contexts) {
> +		ret = -ENOMEM;
> +		goto out_threads;
> +	}
> +
> +	mutex_lock(&t.engine->i915->drm.struct_mutex);
> +	for (n = 0; n < t.ncontexts; n++) {
> +		t.contexts[n] = mock_context(t.engine->i915, "mock");
> +		if (!t.contexts[n]) {
> +			ret = -ENOMEM;
> +			goto out_contexts;
> +		}
> +	}
> +
> +	for (n = 0; n < ncpus; n++) {
> +		threads[n] = kthread_run(__igt_breadcrumbs_smoketest,
> +					 &t, "igt/%d", n);
> +		if (IS_ERR(threads[n])) {
> +			ret = PTR_ERR(threads[n]);
> +			ncpus = n;

I would be tempted to just fail the test rather than carrying on. 
Especially since it seems selftest will anyway report failure since ret 
is preserved.

> +			break;
> +		}
> +
> +		get_task_struct(threads[n]);
> +	}

Thread creation looks like can be outside struct mutex.

> +	mutex_unlock(&t.engine->i915->drm.struct_mutex);
> +
> +	msleep(jiffies_to_msecs(i915_selftest.timeout_jiffies));
> +
> +	for (n = 0; n < ncpus; n++) {
> +		int err;
> +
> +		err = kthread_stop(threads[n]);
> +		if (err < 0 && !ret)
> +			ret = err;
> +
> +		put_task_struct(threads[n]);
> +	}
> +	pr_info("Completed %lu waits for %lu fence across %d cpus\n",
> +		atomic_long_read(&t.num_waits),
> +		atomic_long_read(&t.num_fences),
> +		ncpus);
> +
> +	mutex_lock(&t.engine->i915->drm.struct_mutex);
> +out_contexts:
> +	for (n = 0; n < t.ncontexts; n++) {
> +		if (!t.contexts[n])
> +			break;
> +		mock_context_close(t.contexts[n]);
> +	}
> +	mutex_unlock(&t.engine->i915->drm.struct_mutex);
> +	kfree(t.contexts);
> +out_threads:
> +	kfree(threads);
> +
> +	return ret;
> +}
> +
>   int i915_request_mock_selftests(void)
>   {
>   	static const struct i915_subtest tests[] = {
> @@ -254,6 +504,7 @@ int i915_request_mock_selftests(void)
>   		SUBTEST(igt_wait_request),
>   		SUBTEST(igt_fence_wait),
>   		SUBTEST(igt_request_rewind),
> +		SUBTEST(mock_breadcrumbs_smoketest),
>   	};
>   	struct drm_i915_private *i915;
>   	intel_wakeref_t wakeref;
> @@ -812,6 +1063,174 @@ static int live_sequential_engines(void *arg)
>   	return err;
>   }
>   
> +static int
> +max_batches(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
> +{
> +	struct i915_request *rq;
> +	int ret;
> +
> +	/*
> +	 * Before execlists, all contexts share the same ringbuffer. With
> +	 * execlists, each context/engine has a separate ringbuffer and
> +	 * for the purposes of this test, inexhaustible.
> +	 *
> +	 * For the global ringbuffer though, we have to be very careful
> +	 * that we do not wrap while preventing the execution of requests
> +	 * with a unsignaled fence.
> +	 */
> +	if (HAS_EXECLISTS(ctx->i915))
> +		return INT_MAX;
> +
> +	rq = i915_request_alloc(engine, ctx);
> +	if (IS_ERR(rq)) {
> +		ret = PTR_ERR(rq);
> +	} else {
> +		int sz;
> +
> +		ret = rq->ring->size - rq->reserved_space;
> +		i915_request_add(rq);
> +
> +		sz = rq->ring->emit - rq->head;
> +		if (sz < 0)
> +			sz += rq->ring->size;
> +		ret /= sz;
> +		ret /= 2; /* leave half spare, in case of emergency! */
> +
> +		/* One ring interleaved between requests from all cpus */
> +		ret /= num_online_cpus() + 1;

It would probably be more robust for any potential future users to leave 
out the num_cpus division to the caller.

> +	}
> +
> +	return ret;
> +}
> +
> +static int live_breadcrumbs_smoketest(void *arg)
> +{
> +	struct drm_i915_private *i915 = arg;
> +	struct smoketest t[I915_NUM_ENGINES];
> +	unsigned int ncpus = num_online_cpus();
> +	unsigned long num_waits, num_fences;
> +	struct intel_engine_cs *engine;
> +	struct task_struct **threads;
> +	struct igt_live_test live;
> +	enum intel_engine_id id;
> +	intel_wakeref_t wakeref;
> +	struct drm_file *file;
> +	unsigned int n;
> +	int ret = 0;
> +
> +	/*
> +	 * Smoketest our breadcrumb/signal handling for requests across multiple
> +	 * threads. A very simple test to only catch the most egregious of bugs.
> +	 * See __igt_breadcrumbs_smoketest();
> +	 *
> +	 * On real hardware this time.
> +	 */
> +
> +	wakeref = intel_runtime_pm_get(i915);
> +
> +	file = mock_file(i915);
> +	if (IS_ERR(file)) {
> +		ret = PTR_ERR(file);
> +		goto out_rpm;
> +	}
> +
> +	threads = kcalloc(ncpus * I915_NUM_ENGINES,
> +			  sizeof(*threads),
> +			  GFP_KERNEL);
> +	if (!threads)
> +		return -ENOMEM;

goto out_mock_file; to avoid two leaks.

> +
> +	memset(&t[0], 0, sizeof(t[0]));
> +	t[0].request_alloc = __live_request_alloc;
> +	t[0].ncontexts = 64;
> +	t[0].contexts = kmalloc_array(t[0].ncontexts,
> +				      sizeof(*t[0].contexts),
> +				      GFP_KERNEL);
> +	if (!t[0].contexts) {
> +		ret = -ENOMEM;
> +		goto out_threads;
> +	}
> +
> +	mutex_lock(&i915->drm.struct_mutex);
> +	for (n = 0; n < t[0].ncontexts; n++) {
> +		t[0].contexts[n] = live_context(i915, file);
> +		if (!t[0].contexts[n]) {
> +			ret = -ENOMEM;
> +			goto out_contexts;

Context list unwind is missing.

> +		}
> +	}
> +
> +	ret = igt_live_test_begin(&live, i915, __func__, "");
> +	if (ret)
> +		goto out_contexts;
> +
> +	for_each_engine(engine, i915, id) {
> +		t[id] = t[0];
> +		t[id].engine = engine;
> +		t[id].max_batch = max_batches(t[0].contexts[0], engine);
> +		if (t[id].max_batch < 0) {
> +			ret = t[id].max_batch;
> +			goto out_flush;
> +		}
> +		pr_debug("Limiting batches to %d requests on %s\n",
> +			 t[id].max_batch, engine->name);
> +
> +		for (n = 0; n < ncpus; n++) {
> +			struct task_struct *tsk;
> +
> +			tsk = kthread_run(__igt_breadcrumbs_smoketest,
> +					  &t[id], "igt/%d.%d", id, n);
> +			if (IS_ERR(tsk)) {
> +				ret = PTR_ERR(tsk);
> +				goto out_flush;
> +			}
> +
> +			get_task_struct(tsk);
> +			threads[id * ncpus + n] = tsk;
> +		}
> +	}
> +	mutex_unlock(&i915->drm.struct_mutex);

I think thread setup could be done outside struct mutext in two-pass, if 
only max_batches calculation is done first under it.

> +
> +	msleep(jiffies_to_msecs(i915_selftest.timeout_jiffies));
> +
> +out_flush:
> +	num_waits = 0;
> +	num_fences = 0;
> +	for_each_engine(engine, i915, id) {
> +		for (n = 0; n < ncpus; n++) {
> +			struct task_struct *tsk = threads[id * ncpus + n];
> +			int err;
> +
> +			if (!tsk)
> +				continue;
> +
> +			err = kthread_stop(tsk);
> +			if (err < 0 && !ret)
> +				ret = err;
> +
> +			put_task_struct(tsk);
> +		}
> +
> +		num_waits += atomic_long_read(&t[id].num_waits);
> +		num_fences += atomic_long_read(&t[id].num_fences);
> +	}
> +	pr_info("Completed %lu waits for %lu fence across %d engines and %d cpus\n",

fences plural

> +		num_waits, num_fences, RUNTIME_INFO(i915)->num_rings, ncpus);
> +
> +	mutex_lock(&i915->drm.struct_mutex);
> +	ret = igt_live_test_end(&live) ?: ret;
> +out_contexts:
> +	mutex_unlock(&i915->drm.struct_mutex);
> +	kfree(t[0].contexts);
> +out_threads:
> +	kfree(threads);
> +	mock_file_free(i915, file);
> +out_rpm:
> +	intel_runtime_pm_put(i915, wakeref);
> +
> +	return ret;
> +}
> +
>   int i915_request_live_selftests(struct drm_i915_private *i915)
>   {
>   	static const struct i915_subtest tests[] = {
> @@ -819,6 +1238,7 @@ int i915_request_live_selftests(struct drm_i915_private *i915)
>   		SUBTEST(live_all_engines),
>   		SUBTEST(live_sequential_engines),
>   		SUBTEST(live_empty_request),
> +		SUBTEST(live_breadcrumbs_smoketest),
>   	};
>   
>   	if (i915_terminally_wedged(&i915->gpu_error))
> diff --git a/drivers/gpu/drm/i915/selftests/igt_spinner.c b/drivers/gpu/drm/i915/selftests/igt_spinner.c
> index 0e70df0230b8..9ebd9225684e 100644
> --- a/drivers/gpu/drm/i915/selftests/igt_spinner.c
> +++ b/drivers/gpu/drm/i915/selftests/igt_spinner.c
> @@ -185,11 +185,6 @@ void igt_spinner_fini(struct igt_spinner *spin)
>   
>   bool igt_wait_for_spinner(struct igt_spinner *spin, struct i915_request *rq)
>   {
> -	if (!wait_event_timeout(rq->execute,
> -				READ_ONCE(rq->global_seqno),
> -				msecs_to_jiffies(10)))
> -		return false;
> -
>   	return !(wait_for_us(i915_seqno_passed(hws_seqno(spin, rq),
>   					       rq->fence.seqno),
>   			     10) &&
> diff --git a/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c b/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
> deleted file mode 100644
> index f03b407fdbe2..000000000000
> --- a/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
> +++ /dev/null
> @@ -1,470 +0,0 @@
> -/*
> - * Copyright © 2016 Intel Corporation
> - *
> - * Permission is hereby granted, free of charge, to any person obtaining a
> - * copy of this software and associated documentation files (the "Software"),
> - * to deal in the Software without restriction, including without limitation
> - * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> - * and/or sell copies of the Software, and to permit persons to whom the
> - * Software is furnished to do so, subject to the following conditions:
> - *
> - * The above copyright notice and this permission notice (including the next
> - * paragraph) shall be included in all copies or substantial portions of the
> - * Software.
> - *
> - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> - * IN THE SOFTWARE.
> - *
> - */
> -
> -#include "../i915_selftest.h"
> -#include "i915_random.h"
> -
> -#include "mock_gem_device.h"
> -#include "mock_engine.h"
> -
> -static int check_rbtree(struct intel_engine_cs *engine,
> -			const unsigned long *bitmap,
> -			const struct intel_wait *waiters,
> -			const int count)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct rb_node *rb;
> -	int n;
> -
> -	if (&b->irq_wait->node != rb_first(&b->waiters)) {
> -		pr_err("First waiter does not match first element of wait-tree\n");
> -		return -EINVAL;
> -	}
> -
> -	n = find_first_bit(bitmap, count);
> -	for (rb = rb_first(&b->waiters); rb; rb = rb_next(rb)) {
> -		struct intel_wait *w = container_of(rb, typeof(*w), node);
> -		int idx = w - waiters;
> -
> -		if (!test_bit(idx, bitmap)) {
> -			pr_err("waiter[%d, seqno=%d] removed but still in wait-tree\n",
> -			       idx, w->seqno);
> -			return -EINVAL;
> -		}
> -
> -		if (n != idx) {
> -			pr_err("waiter[%d, seqno=%d] does not match expected next element in tree [%d]\n",
> -			       idx, w->seqno, n);
> -			return -EINVAL;
> -		}
> -
> -		n = find_next_bit(bitmap, count, n + 1);
> -	}
> -
> -	return 0;
> -}
> -
> -static int check_completion(struct intel_engine_cs *engine,
> -			    const unsigned long *bitmap,
> -			    const struct intel_wait *waiters,
> -			    const int count)
> -{
> -	int n;
> -
> -	for (n = 0; n < count; n++) {
> -		if (intel_wait_complete(&waiters[n]) != !!test_bit(n, bitmap))
> -			continue;
> -
> -		pr_err("waiter[%d, seqno=%d] is %s, but expected %s\n",
> -		       n, waiters[n].seqno,
> -		       intel_wait_complete(&waiters[n]) ? "complete" : "active",
> -		       test_bit(n, bitmap) ? "active" : "complete");
> -		return -EINVAL;
> -	}
> -
> -	return 0;
> -}
> -
> -static int check_rbtree_empty(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	if (b->irq_wait) {
> -		pr_err("Empty breadcrumbs still has a waiter\n");
> -		return -EINVAL;
> -	}
> -
> -	if (!RB_EMPTY_ROOT(&b->waiters)) {
> -		pr_err("Empty breadcrumbs, but wait-tree not empty\n");
> -		return -EINVAL;
> -	}
> -
> -	return 0;
> -}
> -
> -static int igt_random_insert_remove(void *arg)
> -{
> -	const u32 seqno_bias = 0x1000;
> -	I915_RND_STATE(prng);
> -	struct intel_engine_cs *engine = arg;
> -	struct intel_wait *waiters;
> -	const int count = 4096;
> -	unsigned int *order;
> -	unsigned long *bitmap;
> -	int err = -ENOMEM;
> -	int n;
> -
> -	mock_engine_reset(engine);
> -
> -	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
> -	if (!waiters)
> -		goto out_engines;
> -
> -	bitmap = kcalloc(DIV_ROUND_UP(count, BITS_PER_LONG), sizeof(*bitmap),
> -			 GFP_KERNEL);
> -	if (!bitmap)
> -		goto out_waiters;
> -
> -	order = i915_random_order(count, &prng);
> -	if (!order)
> -		goto out_bitmap;
> -
> -	for (n = 0; n < count; n++)
> -		intel_wait_init_for_seqno(&waiters[n], seqno_bias + n);
> -
> -	err = check_rbtree(engine, bitmap, waiters, count);
> -	if (err)
> -		goto out_order;
> -
> -	/* Add and remove waiters into the rbtree in random order. At each
> -	 * step, we verify that the rbtree is correctly ordered.
> -	 */
> -	for (n = 0; n < count; n++) {
> -		int i = order[n];
> -
> -		intel_engine_add_wait(engine, &waiters[i]);
> -		__set_bit(i, bitmap);
> -
> -		err = check_rbtree(engine, bitmap, waiters, count);
> -		if (err)
> -			goto out_order;
> -	}
> -
> -	i915_random_reorder(order, count, &prng);
> -	for (n = 0; n < count; n++) {
> -		int i = order[n];
> -
> -		intel_engine_remove_wait(engine, &waiters[i]);
> -		__clear_bit(i, bitmap);
> -
> -		err = check_rbtree(engine, bitmap, waiters, count);
> -		if (err)
> -			goto out_order;
> -	}
> -
> -	err = check_rbtree_empty(engine);
> -out_order:
> -	kfree(order);
> -out_bitmap:
> -	kfree(bitmap);
> -out_waiters:
> -	kvfree(waiters);
> -out_engines:
> -	mock_engine_flush(engine);
> -	return err;
> -}
> -
> -static int igt_insert_complete(void *arg)
> -{
> -	const u32 seqno_bias = 0x1000;
> -	struct intel_engine_cs *engine = arg;
> -	struct intel_wait *waiters;
> -	const int count = 4096;
> -	unsigned long *bitmap;
> -	int err = -ENOMEM;
> -	int n, m;
> -
> -	mock_engine_reset(engine);
> -
> -	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
> -	if (!waiters)
> -		goto out_engines;
> -
> -	bitmap = kcalloc(DIV_ROUND_UP(count, BITS_PER_LONG), sizeof(*bitmap),
> -			 GFP_KERNEL);
> -	if (!bitmap)
> -		goto out_waiters;
> -
> -	for (n = 0; n < count; n++) {
> -		intel_wait_init_for_seqno(&waiters[n], n + seqno_bias);
> -		intel_engine_add_wait(engine, &waiters[n]);
> -		__set_bit(n, bitmap);
> -	}
> -	err = check_rbtree(engine, bitmap, waiters, count);
> -	if (err)
> -		goto out_bitmap;
> -
> -	/* On each step, we advance the seqno so that several waiters are then
> -	 * complete (we increase the seqno by increasingly larger values to
> -	 * retire more and more waiters at once). All retired waiters should
> -	 * be woken and removed from the rbtree, and so that we check.
> -	 */
> -	for (n = 0; n < count; n = m) {
> -		int seqno = 2 * n;
> -
> -		GEM_BUG_ON(find_first_bit(bitmap, count) != n);
> -
> -		if (intel_wait_complete(&waiters[n])) {
> -			pr_err("waiter[%d, seqno=%d] completed too early\n",
> -			       n, waiters[n].seqno);
> -			err = -EINVAL;
> -			goto out_bitmap;
> -		}
> -
> -		/* complete the following waiters */
> -		mock_seqno_advance(engine, seqno + seqno_bias);
> -		for (m = n; m <= seqno; m++) {
> -			if (m == count)
> -				break;
> -
> -			GEM_BUG_ON(!test_bit(m, bitmap));
> -			__clear_bit(m, bitmap);
> -		}
> -
> -		intel_engine_remove_wait(engine, &waiters[n]);
> -		RB_CLEAR_NODE(&waiters[n].node);
> -
> -		err = check_rbtree(engine, bitmap, waiters, count);
> -		if (err) {
> -			pr_err("rbtree corrupt after seqno advance to %d\n",
> -			       seqno + seqno_bias);
> -			goto out_bitmap;
> -		}
> -
> -		err = check_completion(engine, bitmap, waiters, count);
> -		if (err) {
> -			pr_err("completions after seqno advance to %d failed\n",
> -			       seqno + seqno_bias);
> -			goto out_bitmap;
> -		}
> -	}
> -
> -	err = check_rbtree_empty(engine);
> -out_bitmap:
> -	kfree(bitmap);
> -out_waiters:
> -	kvfree(waiters);
> -out_engines:
> -	mock_engine_flush(engine);
> -	return err;
> -}
> -
> -struct igt_wakeup {
> -	struct task_struct *tsk;
> -	atomic_t *ready, *set, *done;
> -	struct intel_engine_cs *engine;
> -	unsigned long flags;
> -#define STOP 0
> -#define IDLE 1
> -	wait_queue_head_t *wq;
> -	u32 seqno;
> -};
> -
> -static bool wait_for_ready(struct igt_wakeup *w)
> -{
> -	DEFINE_WAIT(ready);
> -
> -	set_bit(IDLE, &w->flags);
> -	if (atomic_dec_and_test(w->done))
> -		wake_up_var(w->done);
> -
> -	if (test_bit(STOP, &w->flags))
> -		goto out;
> -
> -	for (;;) {
> -		prepare_to_wait(w->wq, &ready, TASK_INTERRUPTIBLE);
> -		if (atomic_read(w->ready) == 0)
> -			break;
> -
> -		schedule();
> -	}
> -	finish_wait(w->wq, &ready);
> -
> -out:
> -	clear_bit(IDLE, &w->flags);
> -	if (atomic_dec_and_test(w->set))
> -		wake_up_var(w->set);
> -
> -	return !test_bit(STOP, &w->flags);
> -}
> -
> -static int igt_wakeup_thread(void *arg)
> -{
> -	struct igt_wakeup *w = arg;
> -	struct intel_wait wait;
> -
> -	while (wait_for_ready(w)) {
> -		GEM_BUG_ON(kthread_should_stop());
> -
> -		intel_wait_init_for_seqno(&wait, w->seqno);
> -		intel_engine_add_wait(w->engine, &wait);
> -		for (;;) {
> -			set_current_state(TASK_UNINTERRUPTIBLE);
> -			if (i915_seqno_passed(intel_engine_get_seqno(w->engine),
> -					      w->seqno))
> -				break;
> -
> -			if (test_bit(STOP, &w->flags)) /* emergency escape */
> -				break;
> -
> -			schedule();
> -		}
> -		intel_engine_remove_wait(w->engine, &wait);
> -		__set_current_state(TASK_RUNNING);
> -	}
> -
> -	return 0;
> -}
> -
> -static void igt_wake_all_sync(atomic_t *ready,
> -			      atomic_t *set,
> -			      atomic_t *done,
> -			      wait_queue_head_t *wq,
> -			      int count)
> -{
> -	atomic_set(set, count);
> -	atomic_set(ready, 0);
> -	wake_up_all(wq);
> -
> -	wait_var_event(set, !atomic_read(set));
> -	atomic_set(ready, count);
> -	atomic_set(done, count);
> -}
> -
> -static int igt_wakeup(void *arg)
> -{
> -	I915_RND_STATE(prng);
> -	struct intel_engine_cs *engine = arg;
> -	struct igt_wakeup *waiters;
> -	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
> -	const int count = 4096;
> -	const u32 max_seqno = count / 4;
> -	atomic_t ready, set, done;
> -	int err = -ENOMEM;
> -	int n, step;
> -
> -	mock_engine_reset(engine);
> -
> -	waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
> -	if (!waiters)
> -		goto out_engines;
> -
> -	/* Create a large number of threads, each waiting on a random seqno.
> -	 * Multiple waiters will be waiting for the same seqno.
> -	 */
> -	atomic_set(&ready, count);
> -	for (n = 0; n < count; n++) {
> -		waiters[n].wq = &wq;
> -		waiters[n].ready = &ready;
> -		waiters[n].set = &set;
> -		waiters[n].done = &done;
> -		waiters[n].engine = engine;
> -		waiters[n].flags = BIT(IDLE);
> -
> -		waiters[n].tsk = kthread_run(igt_wakeup_thread, &waiters[n],
> -					     "i915/igt:%d", n);
> -		if (IS_ERR(waiters[n].tsk))
> -			goto out_waiters;
> -
> -		get_task_struct(waiters[n].tsk);
> -	}
> -
> -	for (step = 1; step <= max_seqno; step <<= 1) {
> -		u32 seqno;
> -
> -		/* The waiter threads start paused as we assign them a random
> -		 * seqno and reset the engine. Once the engine is reset,
> -		 * we signal that the threads may begin their wait upon their
> -		 * seqno.
> -		 */
> -		for (n = 0; n < count; n++) {
> -			GEM_BUG_ON(!test_bit(IDLE, &waiters[n].flags));
> -			waiters[n].seqno =
> -				1 + prandom_u32_state(&prng) % max_seqno;
> -		}
> -		mock_seqno_advance(engine, 0);
> -		igt_wake_all_sync(&ready, &set, &done, &wq, count);
> -
> -		/* Simulate the GPU doing chunks of work, with one or more
> -		 * seqno appearing to finish at the same time. A random number
> -		 * of threads will be waiting upon the update and hopefully be
> -		 * woken.
> -		 */
> -		for (seqno = 1; seqno <= max_seqno + step; seqno += step) {
> -			usleep_range(50, 500);
> -			mock_seqno_advance(engine, seqno);
> -		}
> -		GEM_BUG_ON(intel_engine_get_seqno(engine) < 1 + max_seqno);
> -
> -		/* With the seqno now beyond any of the waiting threads, they
> -		 * should all be woken, see that they are complete and signal
> -		 * that they are ready for the next test. We wait until all
> -		 * threads are complete and waiting for us (i.e. not a seqno).
> -		 */
> -		if (!wait_var_event_timeout(&done,
> -					    !atomic_read(&done), 10 * HZ)) {
> -			pr_err("Timed out waiting for %d remaining waiters\n",
> -			       atomic_read(&done));
> -			err = -ETIMEDOUT;
> -			break;
> -		}
> -
> -		err = check_rbtree_empty(engine);
> -		if (err)
> -			break;
> -	}
> -
> -out_waiters:
> -	for (n = 0; n < count; n++) {
> -		if (IS_ERR(waiters[n].tsk))
> -			break;
> -
> -		set_bit(STOP, &waiters[n].flags);
> -	}
> -	mock_seqno_advance(engine, INT_MAX); /* wakeup any broken waiters */
> -	igt_wake_all_sync(&ready, &set, &done, &wq, n);
> -
> -	for (n = 0; n < count; n++) {
> -		if (IS_ERR(waiters[n].tsk))
> -			break;
> -
> -		kthread_stop(waiters[n].tsk);
> -		put_task_struct(waiters[n].tsk);
> -	}
> -
> -	kvfree(waiters);
> -out_engines:
> -	mock_engine_flush(engine);
> -	return err;
> -}
> -
> -int intel_breadcrumbs_mock_selftests(void)
> -{
> -	static const struct i915_subtest tests[] = {
> -		SUBTEST(igt_random_insert_remove),
> -		SUBTEST(igt_insert_complete),
> -		SUBTEST(igt_wakeup),
> -	};
> -	struct drm_i915_private *i915;
> -	int err;
> -
> -	i915 = mock_gem_device();
> -	if (!i915)
> -		return -ENOMEM;
> -
> -	err = i915_subtests(tests, i915->engine[RCS]);
> -	drm_dev_put(&i915->drm);
> -
> -	return err;
> -}
> diff --git a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> index 2c38ea5892d9..7b6f3bea9ef8 100644
> --- a/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> +++ b/drivers/gpu/drm/i915/selftests/intel_hangcheck.c
> @@ -1127,7 +1127,7 @@ static int __igt_reset_evict_vma(struct drm_i915_private *i915,
>   
>   	wait_for_completion(&arg.completion);
>   
> -	if (wait_for(waitqueue_active(&rq->execute), 10)) {
> +	if (wait_for(!list_empty(&rq->fence.cb_list), 10)) {

What is actually supposed to happen here? Callback will be added since 
thread is waiting for eviction, which waits on hang and so reset to 
clear up the hang, and test says 10ms should be enough for the kthread 
to set that up. But if it doesn't happen in 10ms it declares thread did 
"not wait". But maybe it did not evict, or try to evict either for some 
other reason? Or it tried and succeeded without waiting.

Maybe I am overcomplicating things since this doesn't belong to this 
patch anyway. I was just intrigued with what is this supposed to test. I 
guess I expected to see some more checks to verify vma is still where we 
expect it etc.

>   		struct drm_printer p = drm_info_printer(i915->drm.dev);
>   
>   		pr_err("igt/evict_vma kthread did not wait\n");
> diff --git a/drivers/gpu/drm/i915/selftests/lib_sw_fence.c b/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
> index b26f07b55d86..2bfa72c1654b 100644
> --- a/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
> +++ b/drivers/gpu/drm/i915/selftests/lib_sw_fence.c
> @@ -76,3 +76,57 @@ void timed_fence_fini(struct timed_fence *tf)
>   	destroy_timer_on_stack(&tf->timer);
>   	i915_sw_fence_fini(&tf->fence);
>   }
> +
> +struct heap_fence {
> +	struct i915_sw_fence fence;
> +	union {
> +		struct kref ref;
> +		struct rcu_head rcu;
> +	};

Pure magic... I was confused how they can share space and rcu be usable 
when not initialized, but then after nosing around realized it is just 
offset it needs.

> +};
> +
> +static int __i915_sw_fence_call
> +heap_fence_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
> +{
> +	struct heap_fence *h = container_of(fence, typeof(*h), fence);
> +
> +	switch (state) {
> +	case FENCE_COMPLETE:
> +		break;
> +
> +	case FENCE_FREE:
> +		heap_fence_put(&h->fence);
> +	}
> +
> +	return NOTIFY_DONE;
> +}
> +
> +struct i915_sw_fence *heap_fence_create(gfp_t gfp)
> +{
> +	struct heap_fence *h;
> +
> +	h = kmalloc(sizeof(*h), gfp);
> +	if (!h)
> +		return NULL;
> +
> +	i915_sw_fence_init(&h->fence, heap_fence_notify);
> +	refcount_set(&h->ref.refcount, 2);
> +
> +	return &h->fence;
> +}
> +
> +static void heap_fence_release(struct kref *ref)
> +{
> +	struct heap_fence *h = container_of(ref, typeof(*h), ref);
> +
> +	i915_sw_fence_fini(&h->fence);
> +
> +	kfree_rcu(h, rcu);
> +}
> +
> +void heap_fence_put(struct i915_sw_fence *fence)
> +{
> +	struct heap_fence *h = container_of(fence, typeof(*h), fence);
> +
> +	kref_put(&h->ref, heap_fence_release);
> +}
> diff --git a/drivers/gpu/drm/i915/selftests/lib_sw_fence.h b/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
> index 474aafb92ae1..1f9927e10f3a 100644
> --- a/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
> +++ b/drivers/gpu/drm/i915/selftests/lib_sw_fence.h
> @@ -39,4 +39,7 @@ struct timed_fence {
>   void timed_fence_init(struct timed_fence *tf, unsigned long expires);
>   void timed_fence_fini(struct timed_fence *tf);
>   
> +struct i915_sw_fence *heap_fence_create(gfp_t gfp);
> +void heap_fence_put(struct i915_sw_fence *fence);
> +
>   #endif /* _LIB_SW_FENCE_H_ */
> diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.c b/drivers/gpu/drm/i915/selftests/mock_engine.c
> index 3b226ebc6bc4..08f0cab02e0f 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_engine.c
> +++ b/drivers/gpu/drm/i915/selftests/mock_engine.c
> @@ -86,17 +86,21 @@ static struct mock_request *first_request(struct mock_engine *engine)
>   static void advance(struct mock_request *request)
>   {
>   	list_del_init(&request->link);
> -	mock_seqno_advance(request->base.engine, request->base.global_seqno);
> +	intel_engine_write_global_seqno(request->base.engine,
> +					request->base.global_seqno);
>   	i915_request_mark_complete(&request->base);
>   	GEM_BUG_ON(!i915_request_completed(&request->base));
> +
> +	intel_engine_queue_breadcrumbs(request->base.engine);
>   }
>   
>   static void hw_delay_complete(struct timer_list *t)
>   {
>   	struct mock_engine *engine = from_timer(engine, t, hw_delay);
>   	struct mock_request *request;
> +	unsigned long flags;
>   
> -	spin_lock(&engine->hw_lock);
> +	spin_lock_irqsave(&engine->hw_lock, flags);

I can't see that it is called outside the timer context?

>   
>   	/* Timer fired, first request is complete */
>   	request = first_request(engine);
> @@ -116,7 +120,7 @@ static void hw_delay_complete(struct timer_list *t)
>   		advance(request);
>   	}
>   
> -	spin_unlock(&engine->hw_lock);
> +	spin_unlock_irqrestore(&engine->hw_lock, flags);
>   }
>   
>   static void mock_context_unpin(struct intel_context *ce)
> @@ -191,11 +195,12 @@ static void mock_submit_request(struct i915_request *request)
>   	struct mock_request *mock = container_of(request, typeof(*mock), base);
>   	struct mock_engine *engine =
>   		container_of(request->engine, typeof(*engine), base);
> +	unsigned long flags;
>   
>   	i915_request_submit(request);
>   	GEM_BUG_ON(!request->global_seqno);
>   
> -	spin_lock_irq(&engine->hw_lock);
> +	spin_lock_irqsave(&engine->hw_lock, flags);
>   	list_add_tail(&mock->link, &engine->hw_queue);
>   	if (mock->link.prev == &engine->hw_queue) {
>   		if (mock->delay)
> @@ -203,7 +208,7 @@ static void mock_submit_request(struct i915_request *request)
>   		else
>   			advance(mock);
>   	}
> -	spin_unlock_irq(&engine->hw_lock);
> +	spin_unlock_irqrestore(&engine->hw_lock, flags);
>   }
>   
>   struct intel_engine_cs *mock_engine(struct drm_i915_private *i915,
> @@ -273,7 +278,7 @@ void mock_engine_flush(struct intel_engine_cs *engine)
>   
>   void mock_engine_reset(struct intel_engine_cs *engine)
>   {
> -	intel_write_status_page(engine, I915_GEM_HWS_INDEX, 0);
> +	intel_engine_write_global_seqno(engine, 0);
>   }
>   
>   void mock_engine_free(struct intel_engine_cs *engine)
> diff --git a/drivers/gpu/drm/i915/selftests/mock_engine.h b/drivers/gpu/drm/i915/selftests/mock_engine.h
> index 133d0c21790d..b9cc3a245f16 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_engine.h
> +++ b/drivers/gpu/drm/i915/selftests/mock_engine.h
> @@ -46,10 +46,4 @@ void mock_engine_flush(struct intel_engine_cs *engine);
>   void mock_engine_reset(struct intel_engine_cs *engine);
>   void mock_engine_free(struct intel_engine_cs *engine);
>   
> -static inline void mock_seqno_advance(struct intel_engine_cs *engine, u32 seqno)
> -{
> -	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
> -	intel_engine_wakeup(engine);
> -}
> -
>   #endif /* !__MOCK_ENGINE_H__ */
> 

Phew. Approx 3hrs with a short snack break and minimal interruptions. :)

Regards,

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

Re: [PATCH 27/33] drm/i915: Replace global breadcrumbs with per-context interrupt tracking

[Chris Wilson]

Quoting Tvrtko Ursulin (2019-01-25 13:54:05)
> 
> On 25/01/2019 02:29, Chris Wilson wrote:
> > diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> > index 2171df2d3019..c09a6644a2ab 100644
> > --- a/drivers/gpu/drm/i915/i915_request.c
> > +++ b/drivers/gpu/drm/i915/i915_request.c
> > @@ -60,7 +60,7 @@ static bool i915_fence_signaled(struct dma_fence *fence)
> >   
> >   static bool i915_fence_enable_signaling(struct dma_fence *fence)
> >   {
> > -     return intel_engine_enable_signaling(to_request(fence), true);
> > +     return i915_request_enable_breadcrumb(to_request(fence));
> 
> enable_signaling would be better IMO, enable_breadcrumb sounds like it 
> could do with breadcrumb emission. And it would align with 
> i915_fence_enable_signaling.

But isn't it to entirely do with our breadcrumbs...

I'm going the other way, while the request/interrupt logic is called
breadcrumbs, we want to link the dma-fence signaling back to
breadcrumbs.

I liked the new name :)

> >   /**
> >    * i915_request_started - check if the request has begun being executed
> >    * @rq: the request
> > @@ -345,7 +369,23 @@ static inline bool i915_request_started(const struct i915_request *rq)
> >               return true;
> >   
> >       /* Remember: started but may have since been preempted! */
> > -     return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
> > +     return __i915_request_has_started(rq);
> > +}
> 
> Is it worth having both i915_request_started and 
> __i915_request_has_started. AFAIU fence cannot be signaled unless the 
> seqno has advanced, so if __i915_request_has_started would be dropped 
> and the caller just use the former, I don't think there would be any 
> issues. Callers always check for completed before hand. I guess it saves 
> a redundant fence signaled check.

I was trying to consolidate down on the fence->flags logic, and decided
half the checks were redundant.

I was also dipping my toe into the water to see how everyone reacted to
a possible
s/i915_request_completed/i915_request_has_completed/
s/i915_request_started/i915_request_has_started/
i915_request_is_running

> > +static inline bool i915_request_is_running(const struct i915_request *rq)
> > +{
> > +     if (!i915_request_is_active(rq))
> > +             return false;
> > +
> > +     return __i915_request_has_started(rq);
> 
> return i915_request_is_active(rq) && __i915_request_has_started(rq); 
> probably doesn't fit in 80chars to be that attractive?

It keeps the pattern with the other status checks (started, completed,
is_running) though!

> > -static unsigned int __intel_breadcrumbs_wakeup(struct intel_breadcrumbs *b)
> > +static void irq_enable(struct intel_engine_cs *engine)
> > +{
> > +     if (!engine->irq_enable)
> > +             return;
> > +
> > +     /* Caller disables interrupts */
> 
> GEM_BUG_ON(!irqs_disabled); ?

Nah, I thought I had removed it before, but apparently there was more
than one check here.

It doesn't matter if we set the IMR while interrupts are disabled, we
get a hang report. We can't turn off interrupts without parking, and we
can't park without interrupts. That's probably the best point to say
"oi!"

> > +                     /*
> > +                      * We may race with direct invocation of
> > +                      * dma_fence_signal(), e.g. i915_request_retire(),
> > +                      * in which case we can skip processing it ourselves.
> > +                      */
> > +                     if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> > +                                  &rq->fence.flags))
> > +                             continue;
> 
> Doesn't look this is worth it. We could then race a line below and still 
> end up with the already signaled rq on the local list. So would be 
> tempted to drop this for simplicity.

It's virtually free since we already have the exclusive cacheline for
rq->fence.flags. ~o~
(Except when it races and then we skip the atomic and list manipulation)

There's some benefit in having little reminders that many paths lead to
dma_fence_signal.

> > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > index 3f9f0a5a0f44..d7c31ad1d0ea 100644
> > --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > @@ -486,8 +486,8 @@ static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
> >   
> >       for (i = 0; i < GEN7_XCS_WA; i++) {
> >               *cs++ = MI_STORE_DWORD_INDEX;
> > -             *cs++ = I915_GEM_HWS_INDEX_ADDR;
> > -             *cs++ = rq->global_seqno;
> > +             *cs++ = I915_GEM_HWS_SEQNO_ADDR;
> > +             *cs++ = rq->fence.seqno;
> 
> Why this, and in this patch?

A slip of the fingers. Spotted that last night as well and forgot to
move the chunk.

> > +             for (n = 0; n < count; n++) {
> > +                     struct i915_gem_context *ctx =
> > +                             t->contexts[order[n] % t->ncontexts];
> > +                     struct i915_request *rq;
> > +
> > +                     mutex_lock(BKL);
> > +
> > +                     rq = t->request_alloc(ctx, t->engine);
> > +                     if (IS_ERR(rq)) {
> > +                             mutex_unlock(BKL);
> > +                             err = PTR_ERR(rq);
> > +                             count = n;
> > +                             break;
> > +                     }
> > +
> > +                     err = i915_sw_fence_await_sw_fence_gfp(&rq->submit,
> > +                                                            submit,
> > +                                                            GFP_KERNEL);
> 
> Submit fence is just to ensure request doesn't finish before you can 
> enable signaling on it, via the wait fence?

Yes. We put everything into a background queue, complete our setup, then
let the scheduler do its thing.

> > +
> > +                     requests[n] = i915_request_get(rq);
> > +                     i915_request_add(rq);
> > +
> > +                     mutex_unlock(BKL);
> > +
> > +                     if (err >= 0)
> > +                             err = i915_sw_fence_await_dma_fence(wait,
> > +                                                                 &rq->fence,
> > +                                                                 0,
> > +                                                                 GFP_KERNEL);
> 
> If above is true why this can't simply be i915_request_enable_breadcrumbs?

I was trying to test breadcrumbs at a high level without 'cheating'.

For i915_request_enable_breadcrumbs, I keep thinking the test should be
asking if the engine then has it on its signaling list, but that feels
like a circular argument, directly following the implementation itself

> > +                     if (err < 0) {
> 
> else if?

Now where would I be able to copy'n'paste that from? ;)

> > +                             i915_request_put(rq);
> > +                             count = n;
> > +                             break;
> > +                     }
> > +             }
> > +
> > +             i915_sw_fence_commit(submit);
> > +             i915_sw_fence_commit(wait);
> > +
> > +             if (!wait_event_timeout(wait->wait,
> > +                                     i915_sw_fence_done(wait),
> > +                                     HZ / 2)) {
> > +                     struct i915_request *rq = requests[count - 1];
> > +
> > +                     pr_err("waiting for %d fences (last %llx:%lld) on %s timed out!\n",
> > +                            count,
> > +                            rq->fence.context, rq->fence.seqno,
> > +                            t->engine->name);
> > +                     i915_gem_set_wedged(t->engine->i915);
> > +                     GEM_BUG_ON(!i915_request_completed(rq));
> > +                     i915_sw_fence_wait(wait);
> > +                     err = -EIO;
> > +             }
> > +
> > +             for (n = 0; n < count; n++) {
> > +                     struct i915_request *rq = requests[n];
> > +
> > +                     if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
> > +                                   &rq->fence.flags)) {
> > +                             pr_err("%llu:%llu was not signaled!\n",
> > +                                    rq->fence.context, rq->fence.seqno);
> > +                             err = -EINVAL;
> > +                     }
> > +
> > +                     i915_request_put(rq);
> > +             }
> > +
> > +             heap_fence_put(wait);
> > +             heap_fence_put(submit);
> > +
> > +             if (err < 0)
> > +                     break;
> > +
> > +             num_fences += count;
> 
> Isn't num_fences actually also growing by one, while num_requests would 
> be growing by count?

Each request is the rq->fence. We're inspecting requests/fences and
trusting i915_sw_fence.

> > +             num_waits++;
> > +
> > +             cond_resched();
> 
> There is wait_event_timeout in the loop so this shouldn't be needed. 
> AFAICS is something fails loop bails out.

Force of habit. Rather be overkill than have somebody moan about
RCU/khungtaskd stalls.

> > +     for (n = 0; n < ncpus; n++) {
> > +             threads[n] = kthread_run(__igt_breadcrumbs_smoketest,
> > +                                      &t, "igt/%d", n);
> > +             if (IS_ERR(threads[n])) {
> > +                     ret = PTR_ERR(threads[n]);
> > +                     ncpus = n;
> 
> I would be tempted to just fail the test rather than carrying on. 
> Especially since it seems selftest will anyway report failure since ret 
> is preserved.

We have to tidy up, that's why we carry on, keeping the error state.

> > +                     break;
> > +             }
> > +
> > +             get_task_struct(threads[n]);
> > +     }
> 
> Thread creation looks like can be outside struct mutex.

Could be.

> > +static int
> > +max_batches(struct i915_gem_context *ctx, struct intel_engine_cs *engine)
> > +{
> > +     struct i915_request *rq;
> > +     int ret;
> > +
> > +     /*
> > +      * Before execlists, all contexts share the same ringbuffer. With
> > +      * execlists, each context/engine has a separate ringbuffer and
> > +      * for the purposes of this test, inexhaustible.
> > +      *
> > +      * For the global ringbuffer though, we have to be very careful
> > +      * that we do not wrap while preventing the execution of requests
> > +      * with a unsignaled fence.
> > +      */
> > +     if (HAS_EXECLISTS(ctx->i915))
> > +             return INT_MAX;
> > +
> > +     rq = i915_request_alloc(engine, ctx);
> > +     if (IS_ERR(rq)) {
> > +             ret = PTR_ERR(rq);
> > +     } else {
> > +             int sz;
> > +
> > +             ret = rq->ring->size - rq->reserved_space;
> > +             i915_request_add(rq);
> > +
> > +             sz = rq->ring->emit - rq->head;
> > +             if (sz < 0)
> > +                     sz += rq->ring->size;
> > +             ret /= sz;
> > +             ret /= 2; /* leave half spare, in case of emergency! */
> > +
> > +             /* One ring interleaved between requests from all cpus */
> > +             ret /= num_online_cpus() + 1;
> 
> It would probably be more robust for any potential future users to leave 
> out the num_cpus division to the caller.

But then I would only have INT_MAX/ncpus for execlists!!! :)

> > +static int live_breadcrumbs_smoketest(void *arg)
> > +{
> > +     struct drm_i915_private *i915 = arg;
> > +     struct smoketest t[I915_NUM_ENGINES];
> > +     unsigned int ncpus = num_online_cpus();
> > +     unsigned long num_waits, num_fences;
> > +     struct intel_engine_cs *engine;
> > +     struct task_struct **threads;
> > +     struct igt_live_test live;
> > +     enum intel_engine_id id;
> > +     intel_wakeref_t wakeref;
> > +     struct drm_file *file;
> > +     unsigned int n;
> > +     int ret = 0;
> > +
> > +     /*
> > +      * Smoketest our breadcrumb/signal handling for requests across multiple
> > +      * threads. A very simple test to only catch the most egregious of bugs.
> > +      * See __igt_breadcrumbs_smoketest();
> > +      *
> > +      * On real hardware this time.
> > +      */
> > +
> > +     wakeref = intel_runtime_pm_get(i915);
> > +
> > +     file = mock_file(i915);
> > +     if (IS_ERR(file)) {
> > +             ret = PTR_ERR(file);
> > +             goto out_rpm;
> > +     }
> > +
> > +     threads = kcalloc(ncpus * I915_NUM_ENGINES,
> > +                       sizeof(*threads),
> > +                       GFP_KERNEL);
> > +     if (!threads)
> > +             return -ENOMEM;
> 
> goto out_mock_file; to avoid two leaks.
> 
> > +
> > +     memset(&t[0], 0, sizeof(t[0]));
> > +     t[0].request_alloc = __live_request_alloc;
> > +     t[0].ncontexts = 64;
> > +     t[0].contexts = kmalloc_array(t[0].ncontexts,
> > +                                   sizeof(*t[0].contexts),
> > +                                   GFP_KERNEL);
> > +     if (!t[0].contexts) {
> > +             ret = -ENOMEM;
> > +             goto out_threads;
> > +     }
> > +
> > +     mutex_lock(&i915->drm.struct_mutex);
> > +     for (n = 0; n < t[0].ncontexts; n++) {
> > +             t[0].contexts[n] = live_context(i915, file);
> > +             if (!t[0].contexts[n]) {
> > +                     ret = -ENOMEM;
> > +                     goto out_contexts;
> 
> Context list unwind is missing.

The magic of live_context() is that are added to the file and released
when the file is freed if not earlier.

> > +     ret = igt_live_test_begin(&live, i915, __func__, "");
> > +     if (ret)
> > +             goto out_contexts;
> > +
> > +     for_each_engine(engine, i915, id) {
> > +             t[id] = t[0];
> > +             t[id].engine = engine;
> > +             t[id].max_batch = max_batches(t[0].contexts[0], engine);
> > +             if (t[id].max_batch < 0) {
> > +                     ret = t[id].max_batch;
> > +                     goto out_flush;
> > +             }
> > +             pr_debug("Limiting batches to %d requests on %s\n",
> > +                      t[id].max_batch, engine->name);
> > +
> > +             for (n = 0; n < ncpus; n++) {
> > +                     struct task_struct *tsk;
> > +
> > +                     tsk = kthread_run(__igt_breadcrumbs_smoketest,
> > +                                       &t[id], "igt/%d.%d", id, n);
> > +                     if (IS_ERR(tsk)) {
> > +                             ret = PTR_ERR(tsk);
> > +                             goto out_flush;
> > +                     }
> > +
> > +                     get_task_struct(tsk);
> > +                     threads[id * ncpus + n] = tsk;
> > +             }
> > +     }
> > +     mutex_unlock(&i915->drm.struct_mutex);
> 
> I think thread setup could be done outside struct mutext in two-pass, if 
> only max_batches calculation is done first under it.

Then do it as one pass, since having a thread start a few microseconds
earlier is immaterial in the grand scheme of things.

> > +     msleep(jiffies_to_msecs(i915_selftest.timeout_jiffies));
> > +
> > +out_flush:
> > +     num_waits = 0;
> > +     num_fences = 0;
> > +     for_each_engine(engine, i915, id) {
> > +             for (n = 0; n < ncpus; n++) {
> > +                     struct task_struct *tsk = threads[id * ncpus + n];
> > +                     int err;
> > +
> > +                     if (!tsk)
> > +                             continue;
> > +
> > +                     err = kthread_stop(tsk);
> > +                     if (err < 0 && !ret)
> > +                             ret = err;
> > +
> > +                     put_task_struct(tsk);
> > +             }
> > +
> > +             num_waits += atomic_long_read(&t[id].num_waits);
> > +             num_fences += atomic_long_read(&t[id].num_fences);
> > +     }
> > +     pr_info("Completed %lu waits for %lu fence across %d engines and %d cpus\n",
> 
> fences plural

But it might only be 1! ;)
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 27/33] drm/i915: Replace global breadcrumbs with per-context interrupt tracking

[Chris Wilson]

Quoting Chris Wilson (2019-01-25 14:26:49)
> Quoting Tvrtko Ursulin (2019-01-25 13:54:05)
> > 
> > On 25/01/2019 02:29, Chris Wilson wrote:
> > > +
> > > +                     requests[n] = i915_request_get(rq);
> > > +                     i915_request_add(rq);
> > > +
> > > +                     mutex_unlock(BKL);
> > > +
> > > +                     if (err >= 0)
> > > +                             err = i915_sw_fence_await_dma_fence(wait,
> > > +                                                                 &rq->fence,
> > > +                                                                 0,
> > > +                                                                 GFP_KERNEL);
> > 
> > If above is true why this can't simply be i915_request_enable_breadcrumbs?
> 
> I was trying to test breadcrumbs at a high level without 'cheating'.
> 
> For i915_request_enable_breadcrumbs, I keep thinking the test should be
> asking if the engine then has it on its signaling list, but that feels
> like a circular argument, directly following the implementation itself
> 
> > > +                     if (err < 0) {
> > 
> > else if?
> 
> Now where would I be able to copy'n'paste that from? ;)

Actually, no. If it is separate, I'll give you a newline instead.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 28/33] drm/i915: Drop fake breadcrumb irq

[Chris Wilson]

Missed breadcrumb detection is defunct due to the tight coupling with
dma_fence signaling and the myriad ways we may signal fences from
everywhere but from an interrupt, i.e. we frequently signal a fence
before we even see its interrupt. This means that even if we miss an
interrupt for a fence, it still is signaled before our breadcrumb
hangcheck fires, so simplify the breadcrumb hangchecking by moving it
into the GPU hangcheck and forgo fake interrupts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c           |  93 -----------
 drivers/gpu/drm/i915/i915_gpu_error.c         |   2 -
 drivers/gpu/drm/i915/i915_gpu_error.h         |   5 -
 drivers/gpu/drm/i915/intel_breadcrumbs.c      | 147 +-----------------
 drivers/gpu/drm/i915/intel_hangcheck.c        |   2 +
 drivers/gpu/drm/i915/intel_ringbuffer.h       |   5 -
 .../gpu/drm/i915/selftests/igt_live_test.c    |   7 -
 7 files changed, 5 insertions(+), 256 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index e0ca3987eed6..fb216ce8dd60 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1322,9 +1322,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 			   intel_engine_last_submit(engine),
 			   jiffies_to_msecs(jiffies -
 					    engine->hangcheck.action_timestamp));
-		seq_printf(m, "\tfake irq active? %s\n",
-			   yesno(test_bit(engine->id,
-					  &dev_priv->gpu_error.missed_irq_rings)));
 
 		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
 			   (long long)engine->hangcheck.acthd,
@@ -3900,94 +3897,6 @@ DEFINE_SIMPLE_ATTRIBUTE(i915_wedged_fops,
 			i915_wedged_get, i915_wedged_set,
 			"%llu\n");
 
-static int
-fault_irq_set(struct drm_i915_private *i915,
-	      unsigned long *irq,
-	      unsigned long val)
-{
-	int err;
-
-	err = mutex_lock_interruptible(&i915->drm.struct_mutex);
-	if (err)
-		return err;
-
-	err = i915_gem_wait_for_idle(i915,
-				     I915_WAIT_LOCKED |
-				     I915_WAIT_INTERRUPTIBLE,
-				     MAX_SCHEDULE_TIMEOUT);
-	if (err)
-		goto err_unlock;
-
-	*irq = val;
-	mutex_unlock(&i915->drm.struct_mutex);
-
-	/* Flush idle worker to disarm irq */
-	drain_delayed_work(&i915->gt.idle_work);
-
-	return 0;
-
-err_unlock:
-	mutex_unlock(&i915->drm.struct_mutex);
-	return err;
-}
-
-static int
-i915_ring_missed_irq_get(void *data, u64 *val)
-{
-	struct drm_i915_private *dev_priv = data;
-
-	*val = dev_priv->gpu_error.missed_irq_rings;
-	return 0;
-}
-
-static int
-i915_ring_missed_irq_set(void *data, u64 val)
-{
-	struct drm_i915_private *i915 = data;
-
-	return fault_irq_set(i915, &i915->gpu_error.missed_irq_rings, val);
-}
-
-DEFINE_SIMPLE_ATTRIBUTE(i915_ring_missed_irq_fops,
-			i915_ring_missed_irq_get, i915_ring_missed_irq_set,
-			"0x%08llx\n");
-
-static int
-i915_ring_test_irq_get(void *data, u64 *val)
-{
-	struct drm_i915_private *dev_priv = data;
-
-	*val = dev_priv->gpu_error.test_irq_rings;
-
-	return 0;
-}
-
-static int
-i915_ring_test_irq_set(void *data, u64 val)
-{
-	struct drm_i915_private *i915 = data;
-
-	/* GuC keeps the user interrupt permanently enabled for submission */
-	if (USES_GUC_SUBMISSION(i915))
-		return -ENODEV;
-
-	/*
-	 * From icl, we can no longer individually mask interrupt generation
-	 * from each engine.
-	 */
-	if (INTEL_GEN(i915) >= 11)
-		return -ENODEV;
-
-	val &= INTEL_INFO(i915)->ring_mask;
-	DRM_DEBUG_DRIVER("Masking interrupts on rings 0x%08llx\n", val);
-
-	return fault_irq_set(i915, &i915->gpu_error.test_irq_rings, val);
-}
-
-DEFINE_SIMPLE_ATTRIBUTE(i915_ring_test_irq_fops,
-			i915_ring_test_irq_get, i915_ring_test_irq_set,
-			"0x%08llx\n");
-
 #define DROP_UNBOUND	BIT(0)
 #define DROP_BOUND	BIT(1)
 #define DROP_RETIRE	BIT(2)
@@ -4750,8 +4659,6 @@ static const struct i915_debugfs_files {
 } i915_debugfs_files[] = {
 	{"i915_wedged", &i915_wedged_fops},
 	{"i915_cache_sharing", &i915_cache_sharing_fops},
-	{"i915_ring_missed_irq", &i915_ring_missed_irq_fops},
-	{"i915_ring_test_irq", &i915_ring_test_irq_fops},
 	{"i915_gem_drop_caches", &i915_drop_caches_fops},
 #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR)
 	{"i915_error_state", &i915_error_state_fops},
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 420b94341433..965356c460ce 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -723,8 +723,6 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
 	err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
 	err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
 	err_printf(m, "CCID: 0x%08x\n", error->ccid);
-	err_printf(m, "Missed interrupts: 0x%08lx\n",
-		   m->i915->gpu_error.missed_irq_rings);
 
 	for (i = 0; i < error->nfence; i++)
 		err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
index 74757c424aab..53b1f22dd365 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.h
+++ b/drivers/gpu/drm/i915/i915_gpu_error.h
@@ -204,8 +204,6 @@ struct i915_gpu_error {
 
 	atomic_t pending_fb_pin;
 
-	unsigned long missed_irq_rings;
-
 	/**
 	 * State variable controlling the reset flow and count
 	 *
@@ -274,9 +272,6 @@ struct i915_gpu_error {
 	 */
 	wait_queue_head_t reset_queue;
 
-	/* For missed irq/seqno simulation. */
-	unsigned long test_irq_rings;
-
 	struct i915_gpu_restart *restart;
 };
 
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index b0795b0ad227..cacaa1d04d17 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -91,7 +91,6 @@ bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
 
 	spin_lock(&b->irq_lock);
 
-	b->irq_fired = true;
 	if (b->irq_armed && list_empty(&b->signalers))
 		__intel_breadcrumbs_disarm_irq(b);
 
@@ -172,86 +171,6 @@ static void signal_irq_work(struct irq_work *work)
 	intel_engine_breadcrumbs_irq(engine);
 }
 
-static unsigned long wait_timeout(void)
-{
-	return round_jiffies_up(jiffies + DRM_I915_HANGCHECK_JIFFIES);
-}
-
-static noinline void missed_breadcrumb(struct intel_engine_cs *engine)
-{
-	if (GEM_SHOW_DEBUG()) {
-		struct drm_printer p = drm_debug_printer(__func__);
-
-		intel_engine_dump(engine, &p,
-				  "%s missed breadcrumb at %pS\n",
-				  engine->name, __builtin_return_address(0));
-	}
-
-	set_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
-}
-
-static void intel_breadcrumbs_hangcheck(struct timer_list *t)
-{
-	struct intel_engine_cs *engine =
-		from_timer(engine, t, breadcrumbs.hangcheck);
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	if (!b->irq_armed)
-		return;
-
-	if (b->irq_fired)
-		goto rearm;
-
-	/*
-	 * We keep the hangcheck timer alive until we disarm the irq, even
-	 * if there are no waiters at present.
-	 *
-	 * If the waiter was currently running, assume it hasn't had a chance
-	 * to process the pending interrupt (e.g, low priority task on a loaded
-	 * system) and wait until it sleeps before declaring a missed interrupt.
-	 *
-	 * If the waiter was asleep (and not even pending a wakeup), then we
-	 * must have missed an interrupt as the GPU has stopped advancing
-	 * but we still have a waiter. Assuming all batches complete within
-	 * DRM_I915_HANGCHECK_JIFFIES [1.5s]!
-	 */
-	synchronize_hardirq(engine->i915->drm.irq);
-	if (intel_engine_signal_breadcrumbs(engine)) {
-		missed_breadcrumb(engine);
-		mod_timer(&b->fake_irq, jiffies + 1);
-	} else {
-rearm:
-		b->irq_fired = false;
-		mod_timer(&b->hangcheck, wait_timeout());
-	}
-}
-
-static void intel_breadcrumbs_fake_irq(struct timer_list *t)
-{
-	struct intel_engine_cs *engine =
-		from_timer(engine, t, breadcrumbs.fake_irq);
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	/*
-	 * The timer persists in case we cannot enable interrupts,
-	 * or if we have previously seen seqno/interrupt incoherency
-	 * ("missed interrupt" syndrome, better known as a "missed breadcrumb").
-	 * Here the worker will wake up every jiffie in order to kick the
-	 * oldest waiter to do the coherent seqno check.
-	 */
-
-	if (!intel_engine_signal_breadcrumbs(engine) && !b->irq_armed)
-		return;
-
-	/* If the user has disabled the fake-irq, restore the hangchecking */
-	if (!test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings)) {
-		mod_timer(&b->hangcheck, wait_timeout());
-		return;
-	}
-
-	mod_timer(&b->fake_irq, jiffies + 1);
-}
-
 void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
@@ -274,43 +193,14 @@ void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine)
 	spin_unlock_irq(&b->irq_lock);
 }
 
-static bool use_fake_irq(const struct intel_breadcrumbs *b)
-{
-	const struct intel_engine_cs *engine =
-		container_of(b, struct intel_engine_cs, breadcrumbs);
-
-	if (!test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
-		return false;
-
-	/*
-	 * Only start with the heavy weight fake irq timer if we have not
-	 * seen any interrupts since enabling it the first time. If the
-	 * interrupts are still arriving, it means we made a mistake in our
-	 * engine->seqno_barrier(), a timing error that should be transient
-	 * and unlikely to reoccur.
-	 */
-	return !b->irq_fired;
-}
-
-static void enable_fake_irq(struct intel_breadcrumbs *b)
-{
-	/* Ensure we never sleep indefinitely */
-	if (!b->irq_enabled || use_fake_irq(b))
-		mod_timer(&b->fake_irq, jiffies + 1);
-	else
-		mod_timer(&b->hangcheck, wait_timeout());
-}
-
-static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
+static void __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
 {
 	struct intel_engine_cs *engine =
 		container_of(b, struct intel_engine_cs, breadcrumbs);
-	struct drm_i915_private *i915 = engine->i915;
-	bool enabled;
 
 	lockdep_assert_held(&b->irq_lock);
 	if (b->irq_armed)
-		return false;
+		return;
 
 	/*
 	 * The breadcrumb irq will be disarmed on the interrupt after the
@@ -328,16 +218,8 @@ static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
 	 * the driver is idle) we disarm the breadcrumbs.
 	 */
 
-	/* No interrupts? Kick the waiter every jiffie! */
-	enabled = false;
-	if (!b->irq_enabled++ &&
-	    !test_bit(engine->id, &i915->gpu_error.test_irq_rings)) {
+	if (!b->irq_enabled++)
 		irq_enable(engine);
-		enabled = true;
-	}
-
-	enable_fake_irq(b);
-	return enabled;
 }
 
 void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
@@ -348,18 +230,6 @@ void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
 	INIT_LIST_HEAD(&b->signalers);
 
 	init_irq_work(&b->irq_work, signal_irq_work);
-
-	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
-	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
-}
-
-static void cancel_fake_irq(struct intel_engine_cs *engine)
-{
-	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-
-	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
-	del_timer_sync(&b->hangcheck);
-	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
 }
 
 void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
@@ -369,13 +239,6 @@ void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
 
 	spin_lock_irqsave(&b->irq_lock, flags);
 
-	/*
-	 * Leave the fake_irq timer enabled (if it is running), but clear the
-	 * bit so that it turns itself off on its next wake up and goes back
-	 * to the long hangcheck interval if still required.
-	 */
-	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
-
 	if (b->irq_enabled)
 		irq_enable(engine);
 	else
@@ -386,7 +249,6 @@ void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
 
 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
 {
-	cancel_fake_irq(engine);
 }
 
 bool i915_request_enable_breadcrumb(struct i915_request *rq)
@@ -482,7 +344,4 @@ void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
 		}
 	}
 	spin_unlock_irq(&b->irq_lock);
-
-	if (test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
-		drm_printf(p, "Fake irq active\n");
 }
diff --git a/drivers/gpu/drm/i915/intel_hangcheck.c b/drivers/gpu/drm/i915/intel_hangcheck.c
index 5662d6fed523..a219c796e56d 100644
--- a/drivers/gpu/drm/i915/intel_hangcheck.c
+++ b/drivers/gpu/drm/i915/intel_hangcheck.c
@@ -275,6 +275,8 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 	for_each_engine(engine, dev_priv, id) {
 		struct hangcheck hc;
 
+		intel_engine_signal_breadcrumbs(engine);
+
 		hangcheck_load_sample(engine, &hc);
 		hangcheck_accumulate_sample(engine, &hc);
 		hangcheck_store_sample(engine, &hc);
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index eb0f8667a246..8038397239ee 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -382,14 +382,9 @@ struct intel_engine_cs {
 
 		struct irq_work irq_work; /* for use from inside irq_lock */
 
-		struct timer_list fake_irq; /* used after a missed interrupt */
-		struct timer_list hangcheck; /* detect missed interrupts */
-
-		unsigned int hangcheck_interrupts;
 		unsigned int irq_enabled;
 
 		bool irq_armed;
-		bool irq_fired;
 	} breadcrumbs;
 
 	struct {
diff --git a/drivers/gpu/drm/i915/selftests/igt_live_test.c b/drivers/gpu/drm/i915/selftests/igt_live_test.c
index 5deb485fb942..3e902761cd16 100644
--- a/drivers/gpu/drm/i915/selftests/igt_live_test.c
+++ b/drivers/gpu/drm/i915/selftests/igt_live_test.c
@@ -35,7 +35,6 @@ int igt_live_test_begin(struct igt_live_test *t,
 		return err;
 	}
 
-	i915->gpu_error.missed_irq_rings = 0;
 	t->reset_global = i915_reset_count(&i915->gpu_error);
 
 	for_each_engine(engine, i915, id)
@@ -75,11 +74,5 @@ int igt_live_test_end(struct igt_live_test *t)
 		return -EIO;
 	}
 
-	if (i915->gpu_error.missed_irq_rings) {
-		pr_err("%s(%s): Missed interrupts on engines %lx\n",
-		       t->func, t->name, i915->gpu_error.missed_irq_rings);
-		return -EIO;
-	}
-
 	return 0;
 }
-- 
2.20.1

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

Re: [PATCH 28/33] drm/i915: Drop fake breadcrumb irq

[Tvrtko Ursulin]

On 25/01/2019 02:30, Chris Wilson wrote:
> Missed breadcrumb detection is defunct due to the tight coupling with
> dma_fence signaling and the myriad ways we may signal fences from
> everywhere but from an interrupt, i.e. we frequently signal a fence
> before we even see its interrupt. This means that even if we miss an
> interrupt for a fence, it still is signaled before our breadcrumb
> hangcheck fires, so simplify the breadcrumb hangchecking by moving it
> into the GPU hangcheck and forgo fake interrupts.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Regards,

Tvrtko

> ---
>   drivers/gpu/drm/i915/i915_debugfs.c           |  93 -----------
>   drivers/gpu/drm/i915/i915_gpu_error.c         |   2 -
>   drivers/gpu/drm/i915/i915_gpu_error.h         |   5 -
>   drivers/gpu/drm/i915/intel_breadcrumbs.c      | 147 +-----------------
>   drivers/gpu/drm/i915/intel_hangcheck.c        |   2 +
>   drivers/gpu/drm/i915/intel_ringbuffer.h       |   5 -
>   .../gpu/drm/i915/selftests/igt_live_test.c    |   7 -
>   7 files changed, 5 insertions(+), 256 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index e0ca3987eed6..fb216ce8dd60 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -1322,9 +1322,6 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
>   			   intel_engine_last_submit(engine),
>   			   jiffies_to_msecs(jiffies -
>   					    engine->hangcheck.action_timestamp));
> -		seq_printf(m, "\tfake irq active? %s\n",
> -			   yesno(test_bit(engine->id,
> -					  &dev_priv->gpu_error.missed_irq_rings)));
>   
>   		seq_printf(m, "\tACTHD = 0x%08llx [current 0x%08llx]\n",
>   			   (long long)engine->hangcheck.acthd,
> @@ -3900,94 +3897,6 @@ DEFINE_SIMPLE_ATTRIBUTE(i915_wedged_fops,
>   			i915_wedged_get, i915_wedged_set,
>   			"%llu\n");
>   
> -static int
> -fault_irq_set(struct drm_i915_private *i915,
> -	      unsigned long *irq,
> -	      unsigned long val)
> -{
> -	int err;
> -
> -	err = mutex_lock_interruptible(&i915->drm.struct_mutex);
> -	if (err)
> -		return err;
> -
> -	err = i915_gem_wait_for_idle(i915,
> -				     I915_WAIT_LOCKED |
> -				     I915_WAIT_INTERRUPTIBLE,
> -				     MAX_SCHEDULE_TIMEOUT);
> -	if (err)
> -		goto err_unlock;
> -
> -	*irq = val;
> -	mutex_unlock(&i915->drm.struct_mutex);
> -
> -	/* Flush idle worker to disarm irq */
> -	drain_delayed_work(&i915->gt.idle_work);
> -
> -	return 0;
> -
> -err_unlock:
> -	mutex_unlock(&i915->drm.struct_mutex);
> -	return err;
> -}
> -
> -static int
> -i915_ring_missed_irq_get(void *data, u64 *val)
> -{
> -	struct drm_i915_private *dev_priv = data;
> -
> -	*val = dev_priv->gpu_error.missed_irq_rings;
> -	return 0;
> -}
> -
> -static int
> -i915_ring_missed_irq_set(void *data, u64 val)
> -{
> -	struct drm_i915_private *i915 = data;
> -
> -	return fault_irq_set(i915, &i915->gpu_error.missed_irq_rings, val);
> -}
> -
> -DEFINE_SIMPLE_ATTRIBUTE(i915_ring_missed_irq_fops,
> -			i915_ring_missed_irq_get, i915_ring_missed_irq_set,
> -			"0x%08llx\n");
> -
> -static int
> -i915_ring_test_irq_get(void *data, u64 *val)
> -{
> -	struct drm_i915_private *dev_priv = data;
> -
> -	*val = dev_priv->gpu_error.test_irq_rings;
> -
> -	return 0;
> -}
> -
> -static int
> -i915_ring_test_irq_set(void *data, u64 val)
> -{
> -	struct drm_i915_private *i915 = data;
> -
> -	/* GuC keeps the user interrupt permanently enabled for submission */
> -	if (USES_GUC_SUBMISSION(i915))
> -		return -ENODEV;
> -
> -	/*
> -	 * From icl, we can no longer individually mask interrupt generation
> -	 * from each engine.
> -	 */
> -	if (INTEL_GEN(i915) >= 11)
> -		return -ENODEV;
> -
> -	val &= INTEL_INFO(i915)->ring_mask;
> -	DRM_DEBUG_DRIVER("Masking interrupts on rings 0x%08llx\n", val);
> -
> -	return fault_irq_set(i915, &i915->gpu_error.test_irq_rings, val);
> -}
> -
> -DEFINE_SIMPLE_ATTRIBUTE(i915_ring_test_irq_fops,
> -			i915_ring_test_irq_get, i915_ring_test_irq_set,
> -			"0x%08llx\n");
> -
>   #define DROP_UNBOUND	BIT(0)
>   #define DROP_BOUND	BIT(1)
>   #define DROP_RETIRE	BIT(2)
> @@ -4750,8 +4659,6 @@ static const struct i915_debugfs_files {
>   } i915_debugfs_files[] = {
>   	{"i915_wedged", &i915_wedged_fops},
>   	{"i915_cache_sharing", &i915_cache_sharing_fops},
> -	{"i915_ring_missed_irq", &i915_ring_missed_irq_fops},
> -	{"i915_ring_test_irq", &i915_ring_test_irq_fops},
>   	{"i915_gem_drop_caches", &i915_drop_caches_fops},
>   #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR)
>   	{"i915_error_state", &i915_error_state_fops},
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
> index 420b94341433..965356c460ce 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
> @@ -723,8 +723,6 @@ static void __err_print_to_sgl(struct drm_i915_error_state_buf *m,
>   	err_printf(m, "FORCEWAKE: 0x%08x\n", error->forcewake);
>   	err_printf(m, "DERRMR: 0x%08x\n", error->derrmr);
>   	err_printf(m, "CCID: 0x%08x\n", error->ccid);
> -	err_printf(m, "Missed interrupts: 0x%08lx\n",
> -		   m->i915->gpu_error.missed_irq_rings);
>   
>   	for (i = 0; i < error->nfence; i++)
>   		err_printf(m, "  fence[%d] = %08llx\n", i, error->fence[i]);
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
> index 74757c424aab..53b1f22dd365 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.h
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.h
> @@ -204,8 +204,6 @@ struct i915_gpu_error {
>   
>   	atomic_t pending_fb_pin;
>   
> -	unsigned long missed_irq_rings;
> -
>   	/**
>   	 * State variable controlling the reset flow and count
>   	 *
> @@ -274,9 +272,6 @@ struct i915_gpu_error {
>   	 */
>   	wait_queue_head_t reset_queue;
>   
> -	/* For missed irq/seqno simulation. */
> -	unsigned long test_irq_rings;
> -
>   	struct i915_gpu_restart *restart;
>   };
>   
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index b0795b0ad227..cacaa1d04d17 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -91,7 +91,6 @@ bool intel_engine_breadcrumbs_irq(struct intel_engine_cs *engine)
>   
>   	spin_lock(&b->irq_lock);
>   
> -	b->irq_fired = true;
>   	if (b->irq_armed && list_empty(&b->signalers))
>   		__intel_breadcrumbs_disarm_irq(b);
>   
> @@ -172,86 +171,6 @@ static void signal_irq_work(struct irq_work *work)
>   	intel_engine_breadcrumbs_irq(engine);
>   }
>   
> -static unsigned long wait_timeout(void)
> -{
> -	return round_jiffies_up(jiffies + DRM_I915_HANGCHECK_JIFFIES);
> -}
> -
> -static noinline void missed_breadcrumb(struct intel_engine_cs *engine)
> -{
> -	if (GEM_SHOW_DEBUG()) {
> -		struct drm_printer p = drm_debug_printer(__func__);
> -
> -		intel_engine_dump(engine, &p,
> -				  "%s missed breadcrumb at %pS\n",
> -				  engine->name, __builtin_return_address(0));
> -	}
> -
> -	set_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
> -}
> -
> -static void intel_breadcrumbs_hangcheck(struct timer_list *t)
> -{
> -	struct intel_engine_cs *engine =
> -		from_timer(engine, t, breadcrumbs.hangcheck);
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	if (!b->irq_armed)
> -		return;
> -
> -	if (b->irq_fired)
> -		goto rearm;
> -
> -	/*
> -	 * We keep the hangcheck timer alive until we disarm the irq, even
> -	 * if there are no waiters at present.
> -	 *
> -	 * If the waiter was currently running, assume it hasn't had a chance
> -	 * to process the pending interrupt (e.g, low priority task on a loaded
> -	 * system) and wait until it sleeps before declaring a missed interrupt.
> -	 *
> -	 * If the waiter was asleep (and not even pending a wakeup), then we
> -	 * must have missed an interrupt as the GPU has stopped advancing
> -	 * but we still have a waiter. Assuming all batches complete within
> -	 * DRM_I915_HANGCHECK_JIFFIES [1.5s]!
> -	 */
> -	synchronize_hardirq(engine->i915->drm.irq);
> -	if (intel_engine_signal_breadcrumbs(engine)) {
> -		missed_breadcrumb(engine);
> -		mod_timer(&b->fake_irq, jiffies + 1);
> -	} else {
> -rearm:
> -		b->irq_fired = false;
> -		mod_timer(&b->hangcheck, wait_timeout());
> -	}
> -}
> -
> -static void intel_breadcrumbs_fake_irq(struct timer_list *t)
> -{
> -	struct intel_engine_cs *engine =
> -		from_timer(engine, t, breadcrumbs.fake_irq);
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	/*
> -	 * The timer persists in case we cannot enable interrupts,
> -	 * or if we have previously seen seqno/interrupt incoherency
> -	 * ("missed interrupt" syndrome, better known as a "missed breadcrumb").
> -	 * Here the worker will wake up every jiffie in order to kick the
> -	 * oldest waiter to do the coherent seqno check.
> -	 */
> -
> -	if (!intel_engine_signal_breadcrumbs(engine) && !b->irq_armed)
> -		return;
> -
> -	/* If the user has disabled the fake-irq, restore the hangchecking */
> -	if (!test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings)) {
> -		mod_timer(&b->hangcheck, wait_timeout());
> -		return;
> -	}
> -
> -	mod_timer(&b->fake_irq, jiffies + 1);
> -}
> -
>   void intel_engine_pin_breadcrumbs_irq(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> @@ -274,43 +193,14 @@ void intel_engine_unpin_breadcrumbs_irq(struct intel_engine_cs *engine)
>   	spin_unlock_irq(&b->irq_lock);
>   }
>   
> -static bool use_fake_irq(const struct intel_breadcrumbs *b)
> -{
> -	const struct intel_engine_cs *engine =
> -		container_of(b, struct intel_engine_cs, breadcrumbs);
> -
> -	if (!test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
> -		return false;
> -
> -	/*
> -	 * Only start with the heavy weight fake irq timer if we have not
> -	 * seen any interrupts since enabling it the first time. If the
> -	 * interrupts are still arriving, it means we made a mistake in our
> -	 * engine->seqno_barrier(), a timing error that should be transient
> -	 * and unlikely to reoccur.
> -	 */
> -	return !b->irq_fired;
> -}
> -
> -static void enable_fake_irq(struct intel_breadcrumbs *b)
> -{
> -	/* Ensure we never sleep indefinitely */
> -	if (!b->irq_enabled || use_fake_irq(b))
> -		mod_timer(&b->fake_irq, jiffies + 1);
> -	else
> -		mod_timer(&b->hangcheck, wait_timeout());
> -}
> -
> -static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
> +static void __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
>   {
>   	struct intel_engine_cs *engine =
>   		container_of(b, struct intel_engine_cs, breadcrumbs);
> -	struct drm_i915_private *i915 = engine->i915;
> -	bool enabled;
>   
>   	lockdep_assert_held(&b->irq_lock);
>   	if (b->irq_armed)
> -		return false;
> +		return;
>   
>   	/*
>   	 * The breadcrumb irq will be disarmed on the interrupt after the
> @@ -328,16 +218,8 @@ static bool __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
>   	 * the driver is idle) we disarm the breadcrumbs.
>   	 */
>   
> -	/* No interrupts? Kick the waiter every jiffie! */
> -	enabled = false;
> -	if (!b->irq_enabled++ &&
> -	    !test_bit(engine->id, &i915->gpu_error.test_irq_rings)) {
> +	if (!b->irq_enabled++)
>   		irq_enable(engine);
> -		enabled = true;
> -	}
> -
> -	enable_fake_irq(b);
> -	return enabled;
>   }
>   
>   void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
> @@ -348,18 +230,6 @@ void intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
>   	INIT_LIST_HEAD(&b->signalers);
>   
>   	init_irq_work(&b->irq_work, signal_irq_work);
> -
> -	timer_setup(&b->fake_irq, intel_breadcrumbs_fake_irq, 0);
> -	timer_setup(&b->hangcheck, intel_breadcrumbs_hangcheck, 0);
> -}
> -
> -static void cancel_fake_irq(struct intel_engine_cs *engine)
> -{
> -	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -
> -	del_timer_sync(&b->fake_irq); /* may queue b->hangcheck */
> -	del_timer_sync(&b->hangcheck);
> -	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
>   }
>   
>   void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
> @@ -369,13 +239,6 @@ void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
>   
>   	spin_lock_irqsave(&b->irq_lock, flags);
>   
> -	/*
> -	 * Leave the fake_irq timer enabled (if it is running), but clear the
> -	 * bit so that it turns itself off on its next wake up and goes back
> -	 * to the long hangcheck interval if still required.
> -	 */
> -	clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
> -
>   	if (b->irq_enabled)
>   		irq_enable(engine);
>   	else
> @@ -386,7 +249,6 @@ void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
>   
>   void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
>   {
> -	cancel_fake_irq(engine);
>   }
>   
>   bool i915_request_enable_breadcrumb(struct i915_request *rq)
> @@ -482,7 +344,4 @@ void intel_engine_print_breadcrumbs(struct intel_engine_cs *engine,
>   		}
>   	}
>   	spin_unlock_irq(&b->irq_lock);
> -
> -	if (test_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings))
> -		drm_printf(p, "Fake irq active\n");
>   }
> diff --git a/drivers/gpu/drm/i915/intel_hangcheck.c b/drivers/gpu/drm/i915/intel_hangcheck.c
> index 5662d6fed523..a219c796e56d 100644
> --- a/drivers/gpu/drm/i915/intel_hangcheck.c
> +++ b/drivers/gpu/drm/i915/intel_hangcheck.c
> @@ -275,6 +275,8 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
>   	for_each_engine(engine, dev_priv, id) {
>   		struct hangcheck hc;
>   
> +		intel_engine_signal_breadcrumbs(engine);
> +
>   		hangcheck_load_sample(engine, &hc);
>   		hangcheck_accumulate_sample(engine, &hc);
>   		hangcheck_store_sample(engine, &hc);
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index eb0f8667a246..8038397239ee 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -382,14 +382,9 @@ struct intel_engine_cs {
>   
>   		struct irq_work irq_work; /* for use from inside irq_lock */
>   
> -		struct timer_list fake_irq; /* used after a missed interrupt */
> -		struct timer_list hangcheck; /* detect missed interrupts */
> -
> -		unsigned int hangcheck_interrupts;
>   		unsigned int irq_enabled;
>   
>   		bool irq_armed;
> -		bool irq_fired;
>   	} breadcrumbs;
>   
>   	struct {
> diff --git a/drivers/gpu/drm/i915/selftests/igt_live_test.c b/drivers/gpu/drm/i915/selftests/igt_live_test.c
> index 5deb485fb942..3e902761cd16 100644
> --- a/drivers/gpu/drm/i915/selftests/igt_live_test.c
> +++ b/drivers/gpu/drm/i915/selftests/igt_live_test.c
> @@ -35,7 +35,6 @@ int igt_live_test_begin(struct igt_live_test *t,
>   		return err;
>   	}
>   
> -	i915->gpu_error.missed_irq_rings = 0;
>   	t->reset_global = i915_reset_count(&i915->gpu_error);
>   
>   	for_each_engine(engine, i915, id)
> @@ -75,11 +74,5 @@ int igt_live_test_end(struct igt_live_test *t)
>   		return -EIO;
>   	}
>   
> -	if (i915->gpu_error.missed_irq_rings) {
> -		pr_err("%s(%s): Missed interrupts on engines %lx\n",
> -		       t->func, t->name, i915->gpu_error.missed_irq_rings);
> -		return -EIO;
> -	}
> -
>   	return 0;
>   }
> 
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 29/33] drm/i915: Implement an "idle" barrier

[Chris Wilson]

We have a number of tasks that we like to run when idle and parking the
GPU into a powersaving mode. A few of those tasks are using the global
idle point as a convenient moment when all previous execution has been
required (and so we know that the GPU is not still touching random
user memory). However, on a busy system we are unlikely to see global
idle points, and would prefer a much more incremental system of being
able to retire after all current execution has completed.

Enter the idle barrier and idle tasks.

To determine a point in the future when all current tasks are complete,
we schedule a new low priority request that will be executed after all
current work is complete, and by imposing a barrier before all future
work. We therefore know we retire that barrier, the GPU is no longer
touching any memory released before the barrier was submitting allowing
us to run a set of idle tasks clear of any dangling GPU references.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h      |  5 ++
 drivers/gpu/drm/i915/i915_gem.c      | 90 ++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_request.c  |  9 +++
 drivers/gpu/drm/i915/i915_timeline.c |  3 +
 4 files changed, 107 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d072f3369ee1..5ca77e2e53fb 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2021,6 +2021,9 @@ struct drm_i915_private {
 		 */
 		struct delayed_work idle_work;
 
+		struct i915_gem_active idle_barrier;
+		struct list_head idle_tasks;
+
 		ktime_t last_init_time;
 
 		struct i915_vma *scratch;
@@ -3040,6 +3043,8 @@ void i915_gem_fini(struct drm_i915_private *dev_priv);
 void i915_gem_cleanup_engines(struct drm_i915_private *dev_priv);
 int i915_gem_wait_for_idle(struct drm_i915_private *dev_priv,
 			   unsigned int flags, long timeout);
+void i915_gem_add_idle_task(struct drm_i915_private *i915,
+			    struct i915_gem_active *idle);
 int __must_check i915_gem_suspend(struct drm_i915_private *dev_priv);
 void i915_gem_suspend_late(struct drm_i915_private *dev_priv);
 void i915_gem_resume(struct drm_i915_private *dev_priv);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 101a0f644787..0a8bcf6e7098 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -141,6 +141,15 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)
 	return 0;
 }
 
+static void call_idle_tasks(struct list_head *tasks)
+{
+	struct i915_gem_active *tsk, *tn;
+
+	list_for_each_entry_safe(tsk, tn, tasks, link)
+		tsk->retire(tsk, NULL);
+	INIT_LIST_HEAD(tasks);
+}
+
 static u32 __i915_gem_park(struct drm_i915_private *i915)
 {
 	intel_wakeref_t wakeref;
@@ -169,6 +178,8 @@ static u32 __i915_gem_park(struct drm_i915_private *i915)
 	 */
 	synchronize_irq(i915->drm.irq);
 
+	call_idle_tasks(&i915->gt.idle_tasks);
+
 	intel_engines_park(i915);
 	i915_timelines_park(i915);
 
@@ -2906,6 +2917,81 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
 	return active;
 }
 
+static void idle_barrier(struct drm_i915_private *i915)
+{
+	struct i915_gt_timelines *gt = &i915->gt.timelines;
+	struct i915_timeline *tl;
+	struct i915_request *rq;
+	int err = 0;
+
+	if (list_empty(&i915->gt.idle_tasks))
+		return;
+
+	if (!i915->gt.active_requests) {
+		call_idle_tasks(&i915->gt.idle_tasks);
+		return;
+	}
+
+	/* Keep just one idle barrier in flight, amalgamating tasks instead */
+	if (i915_gem_active_isset(&i915->gt.idle_barrier))
+		return;
+
+	GEM_TRACE("adding idle barrier\n");
+
+	rq = i915_request_alloc(i915->engine[RCS], i915->kernel_context);
+	if (IS_ERR(rq))
+		return;
+
+	/* run after all current requests have executed, but before any new */
+	mutex_lock(&gt->mutex);
+	list_for_each_entry(tl, &gt->active_list, link) {
+		struct i915_request *last;
+
+		if (tl == rq->timeline)
+			continue;
+
+		err = i915_timeline_set_barrier(tl, rq);
+		if (err == -EEXIST)
+			continue;
+		if (err)
+			break;
+
+		last = i915_gem_active_raw(&tl->last_request,
+					   &i915->drm.struct_mutex);
+		if (!last)
+			continue;
+
+		mutex_unlock(&gt->mutex); /* allocation ahead! */
+		err = i915_request_await_dma_fence(rq, &last->fence);
+		mutex_lock(&gt->mutex);
+		if (err)
+			break;
+
+		/* restart after reacquiring the lock */
+		tl = list_entry(&gt->active_list, typeof(*tl), link);
+	}
+	mutex_unlock(&gt->mutex);
+
+	if (err == 0) {
+		list_splice_init(&i915->gt.idle_tasks, &rq->active_list);
+		i915_gem_active_set(&i915->gt.idle_barrier, rq);
+	}
+
+	i915_request_add(rq);
+}
+
+void i915_gem_add_idle_task(struct drm_i915_private *i915,
+			    struct i915_gem_active *task)
+{
+	lockdep_assert_held(&i915->drm.struct_mutex);
+	GEM_TRACE("adding idle task hint:%pS\n", task->retire);
+
+	if (i915->gt.active_requests)
+		list_add(&task->link, &i915->gt.idle_tasks);
+	else
+		task->retire(task, NULL);
+}
+
 static void
 i915_gem_retire_work_handler(struct work_struct *work)
 {
@@ -2916,6 +3002,7 @@ i915_gem_retire_work_handler(struct work_struct *work)
 	/* Come back later if the device is busy... */
 	if (mutex_trylock(&dev->struct_mutex)) {
 		i915_retire_requests(dev_priv);
+		idle_barrier(dev_priv);
 		mutex_unlock(&dev->struct_mutex);
 	}
 
@@ -5182,6 +5269,7 @@ void i915_gem_fini(struct drm_i915_private *dev_priv)
 
 	/* Flush any outstanding unpin_work. */
 	i915_gem_drain_workqueue(dev_priv);
+	GEM_BUG_ON(!list_empty(&dev_priv->gt.idle_tasks));
 
 	mutex_lock(&dev_priv->drm.struct_mutex);
 	intel_uc_fini_hw(dev_priv);
@@ -5302,6 +5390,8 @@ int i915_gem_init_early(struct drm_i915_private *dev_priv)
 
 	INIT_LIST_HEAD(&dev_priv->gt.active_rings);
 	INIT_LIST_HEAD(&dev_priv->gt.closed_vma);
+	INIT_LIST_HEAD(&dev_priv->gt.idle_tasks);
+	init_request_active(&dev_priv->gt.idle_barrier, NULL);
 
 	i915_gem_init__mm(dev_priv);
 
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index c09a6644a2ab..b397155fe8a7 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -532,6 +532,11 @@ static int add_barrier(struct i915_request *rq, struct i915_gem_active *active)
 	return barrier ? i915_request_await_dma_fence(rq, &barrier->fence) : 0;
 }
 
+static int add_idle_barrier(struct i915_request *rq)
+{
+	return add_barrier(rq, &rq->i915->gt.idle_barrier);
+}
+
 static int add_timeline_barrier(struct i915_request *rq)
 {
 	return add_barrier(rq, &rq->timeline->barrier);
@@ -679,6 +684,10 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	 */
 	rq->head = rq->ring->emit;
 
+	ret = add_idle_barrier(rq);
+	if (ret)
+		goto err_unwind;
+
 	ret = add_timeline_barrier(rq);
 	if (ret)
 		goto err_unwind;
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 8f5c57304064..60b2e1c3abf4 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -270,6 +270,9 @@ int i915_timeline_set_barrier(struct i915_timeline *tl, struct i915_request *rq)
 	/* Must maintain ordering wrt existing barriers */
 	old = i915_gem_active_raw(&tl->barrier, &rq->i915->drm.struct_mutex);
 	if (old) {
+		if (old == rq)
+			return -EEXIST;
+
 		err = i915_request_await_dma_fence(rq, &old->fence);
 		if (err)
 			return err;
-- 
2.20.1

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

Re: [PATCH 29/33] drm/i915: Implement an "idle" barrier

[Chris Wilson]

Quoting Chris Wilson (2019-01-25 02:30:01)
> We have a number of tasks that we like to run when idle and parking the
> GPU into a powersaving mode. A few of those tasks are using the global
> idle point as a convenient moment when all previous execution has been
> required (and so we know that the GPU is not still touching random
> user memory). However, on a busy system we are unlikely to see global
> idle points, and would prefer a much more incremental system of being
> able to retire after all current execution has completed.
> 
> Enter the idle barrier and idle tasks.
> 
> To determine a point in the future when all current tasks are complete,
> we schedule a new low priority request that will be executed after all
> current work is complete, and by imposing a barrier before all future
> work. We therefore know we retire that barrier, the GPU is no longer
> touching any memory released before the barrier was submitting allowing
> us to run a set of idle tasks clear of any dangling GPU references.

The biggest downside is that this does impose a global barrier onto the
system, a point which PI cannot pass.

After sleeping on it, I think I prefer not to take this approach, as
interesting as it is, just yet (for HW semaphores, we can just keep track
of the read references).
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 30/33] drm/i915: Keep timeline HWSP allocated until the system is idle

[Chris Wilson]

In preparation for enabling HW semaphores, we need to keep in flight
timeline HWSP alive until the entire system is idle, as any other
timeline active on the GPU may still refer back to the already retired
timeline. We both have to delay recycling available cachelines and
unpinning old HWSP until the next idle point (i.e. on parking).

That we have to keep the HWSP alive for external references on HW raises
an interesting conundrum. On a busy system, we may never see a global
idle point, essentially meaning the resource will be leaking until we
are forced to sleep. What we need is a set of RCU primitives for the GPU!
This should also help mitigate the resource starvation issues
promulgating from keeping all logical state pinned until idle (instead
of as currently handled until the next context switch).

v2: Use idle barriers to free stale HWSP as soon as all current requests
are idle, rather than rely on the system reaching a global idle point.
(Tvrtko)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_request.c  |  33 ++++----
 drivers/gpu/drm/i915/i915_timeline.c | 118 +++++++++++++++++++++++----
 drivers/gpu/drm/i915/i915_timeline.h |   1 +
 3 files changed, 121 insertions(+), 31 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index b397155fe8a7..690fb4eb6d15 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -331,11 +331,6 @@ void i915_request_retire_upto(struct i915_request *rq)
 	} while (tmp != rq);
 }
 
-static u32 timeline_get_seqno(struct i915_timeline *tl)
-{
-	return tl->seqno += 1 + tl->has_initial_breadcrumb;
-}
-
 static void move_to_timeline(struct i915_request *request,
 			     struct i915_timeline *timeline)
 {
@@ -555,8 +550,10 @@ struct i915_request *
 i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 {
 	struct drm_i915_private *i915 = engine->i915;
-	struct i915_request *rq;
 	struct intel_context *ce;
+	struct i915_timeline *tl;
+	struct i915_request *rq;
+	u32 seqno;
 	int ret;
 
 	lockdep_assert_held(&i915->drm.struct_mutex);
@@ -631,7 +628,15 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 		}
 	}
 
-	rq->rcustate = get_state_synchronize_rcu();
+	tl = ce->ring->timeline;
+	GEM_BUG_ON(tl == &engine->timeline);
+	ret = i915_timeline_get_seqno(tl, &seqno);
+	if (ret)
+		goto err_free;
+
+	spin_lock_init(&rq->lock);
+	dma_fence_init(&rq->fence, &i915_fence_ops, &rq->lock,
+		       tl->fence_context, seqno);
 
 	INIT_LIST_HEAD(&rq->active_list);
 	rq->i915 = i915;
@@ -639,16 +644,9 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	rq->gem_context = ctx;
 	rq->hw_context = ce;
 	rq->ring = ce->ring;
-	rq->timeline = ce->ring->timeline;
-	GEM_BUG_ON(rq->timeline == &engine->timeline);
-	rq->hwsp_seqno = rq->timeline->hwsp_seqno;
-
-	spin_lock_init(&rq->lock);
-	dma_fence_init(&rq->fence,
-		       &i915_fence_ops,
-		       &rq->lock,
-		       rq->timeline->fence_context,
-		       timeline_get_seqno(rq->timeline));
+	rq->timeline = tl;
+	rq->hwsp_seqno = tl->hwsp_seqno;
+	rq->rcustate = get_state_synchronize_rcu();
 
 	/* We bump the ref for the fence chain */
 	i915_sw_fence_init(&i915_request_get(rq)->submit, submit_notify);
@@ -713,6 +711,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	GEM_BUG_ON(!list_empty(&rq->sched.signalers_list));
 	GEM_BUG_ON(!list_empty(&rq->sched.waiters_list));
 
+err_free:
 	kmem_cache_free(i915->requests, rq);
 err_unreserve:
 	unreserve_gt(i915);
diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
index 60b2e1c3abf4..ef5484931ad5 100644
--- a/drivers/gpu/drm/i915/i915_timeline.c
+++ b/drivers/gpu/drm/i915/i915_timeline.c
@@ -10,8 +10,9 @@
 #include "i915_syncmap.h"
 
 struct i915_timeline_hwsp {
-	struct i915_vma *vma;
+	struct i915_gt_timelines *gt;
 	struct list_head free_link;
+	struct i915_vma *vma;
 	u64 free_bitmap;
 };
 
@@ -65,6 +66,7 @@ static struct i915_vma *hwsp_alloc(struct i915_timeline *timeline, int *offset)
 		vma->private = hwsp;
 		hwsp->vma = vma;
 		hwsp->free_bitmap = ~0ull;
+		hwsp->gt = gt;
 
 		spin_lock(&gt->hwsp_lock);
 		list_add(&hwsp->free_link, &gt->hwsp_free_list);
@@ -84,14 +86,9 @@ static struct i915_vma *hwsp_alloc(struct i915_timeline *timeline, int *offset)
 	return hwsp->vma;
 }
 
-static void hwsp_free(struct i915_timeline *timeline)
+static void __idle_hwsp_free(struct i915_timeline_hwsp *hwsp, int offset)
 {
-	struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
-	struct i915_timeline_hwsp *hwsp;
-
-	hwsp = i915_timeline_hwsp(timeline);
-	if (!hwsp) /* leave global HWSP alone! */
-		return;
+	struct i915_gt_timelines *gt = hwsp->gt;
 
 	spin_lock(&gt->hwsp_lock);
 
@@ -99,7 +96,7 @@ static void hwsp_free(struct i915_timeline *timeline)
 	if (!hwsp->free_bitmap)
 		list_add_tail(&hwsp->free_link, &gt->hwsp_free_list);
 
-	hwsp->free_bitmap |= BIT_ULL(timeline->hwsp_offset / CACHELINE_BYTES);
+	hwsp->free_bitmap |= BIT_ULL(offset / CACHELINE_BYTES);
 
 	/* And if no one is left using it, give the page back to the system */
 	if (hwsp->free_bitmap == ~0ull) {
@@ -143,7 +140,9 @@ int i915_timeline_init(struct drm_i915_private *i915,
 
 	vaddr = i915_gem_object_pin_map(hwsp->obj, I915_MAP_WB);
 	if (IS_ERR(vaddr)) {
-		hwsp_free(timeline);
+		if (!i915_timeline_is_global(timeline))
+			__idle_hwsp_free(i915_timeline_hwsp(timeline),
+					 timeline->hwsp_offset);
 		i915_vma_put(hwsp);
 		return PTR_ERR(vaddr);
 	}
@@ -228,10 +227,8 @@ void i915_timeline_fini(struct i915_timeline *timeline)
 {
 	GEM_BUG_ON(timeline->pin_count);
 	GEM_BUG_ON(!list_empty(&timeline->requests));
-	GEM_BUG_ON(i915_gem_active_isset(&timeline->barrier));
 
 	i915_syncmap_free(&timeline->sync);
-	hwsp_free(timeline);
 
 	i915_gem_object_unpin_map(timeline->hwsp_ggtt->obj);
 	i915_vma_put(timeline->hwsp_ggtt);
@@ -303,6 +300,85 @@ int i915_timeline_pin(struct i915_timeline *tl)
 	return err;
 }
 
+static u32 timeline_advance(struct i915_timeline *tl)
+{
+	GEM_BUG_ON(!tl->pin_count);
+	GEM_BUG_ON(tl->seqno & tl->has_initial_breadcrumb);
+
+	return tl->seqno += 1 + tl->has_initial_breadcrumb;
+}
+
+static void timeline_rollback(struct i915_timeline *tl)
+{
+	tl->seqno -= 1 + tl->has_initial_breadcrumb;
+}
+
+struct idle_hwsp_free {
+	struct i915_gem_active active;
+	struct i915_timeline_hwsp *hwsp;
+	int offset;
+};
+
+static void idle_hwsp_free(struct i915_gem_active *active,
+			   struct i915_request *rq)
+{
+	struct idle_hwsp_free *idle =
+		container_of(active, typeof(*idle), active);
+
+	i915_vma_unpin(idle->hwsp->vma);
+	__idle_hwsp_free(idle->hwsp, idle->offset);
+
+	kfree(idle);
+}
+
+static noinline int
+__i915_timeline_get_seqno(struct i915_timeline *tl, u32 *seqno)
+{
+	struct idle_hwsp_free *idle;
+	struct i915_vma *vma;
+	int offset;
+
+	idle = kmalloc(sizeof(*idle), GFP_KERNEL);
+	if (!idle)
+		return -ENOMEM;
+
+	init_request_active(&idle->active, idle_hwsp_free);
+	idle->hwsp = i915_timeline_hwsp(tl);
+	idle->offset = tl->hwsp_offset;
+
+	vma = hwsp_alloc(tl, &offset);
+	if (IS_ERR(vma)) {
+		timeline_rollback(tl);
+		kfree(idle);
+		return PTR_ERR(vma);
+	}
+
+	/*
+	 * We need to keep the previous HWSP cacheline alive until the GPU
+	 * is idle so that we can be sure that any inflight references to it
+	 * have been retired.
+	 */
+	i915_gem_add_idle_task(tl->i915, &idle->active);
+
+	tl->hwsp_ggtt = i915_vma_get(vma);
+	tl->hwsp_offset = offset;
+	__i915_vma_pin(tl->hwsp_ggtt);
+
+	*seqno = timeline_advance(tl);
+	return 0;
+}
+
+int i915_timeline_get_seqno(struct i915_timeline *tl, u32 *seqno)
+{
+	*seqno = timeline_advance(tl);
+
+	/* Replace the HWSP on wraparound for HW semaphores */
+	if (unlikely(!*seqno && !i915_timeline_is_global(tl)))
+		return __i915_timeline_get_seqno(tl, seqno);
+
+	return 0;
+}
+
 void i915_timeline_unpin(struct i915_timeline *tl)
 {
 	GEM_BUG_ON(!tl->pin_count);
@@ -321,15 +397,29 @@ void i915_timeline_unpin(struct i915_timeline *tl)
 	__i915_vma_unpin(tl->hwsp_ggtt);
 }
 
-void __i915_timeline_free(struct kref *kref)
+static void idle_timeline_free(struct i915_gem_active *active,
+			       struct i915_request *rq)
 {
 	struct i915_timeline *timeline =
-		container_of(kref, typeof(*timeline), kref);
+		container_of(active, typeof(*timeline), barrier);
+
+	if (!i915_timeline_is_global(timeline))
+		__idle_hwsp_free(i915_timeline_hwsp(timeline),
+				 timeline->hwsp_offset);
 
 	i915_timeline_fini(timeline);
 	kfree(timeline);
 }
 
+void __i915_timeline_free(struct kref *kref)
+{
+	struct i915_timeline *timeline =
+		container_of(kref, typeof(*timeline), kref);
+
+	init_request_active(&timeline->barrier, idle_timeline_free);
+	i915_gem_add_idle_task(timeline->i915, &timeline->barrier);
+}
+
 void i915_timelines_fini(struct drm_i915_private *i915)
 {
 	struct i915_gt_timelines *gt = &i915->gt.timelines;
diff --git a/drivers/gpu/drm/i915/i915_timeline.h b/drivers/gpu/drm/i915/i915_timeline.h
index d167e04073c5..37529981d909 100644
--- a/drivers/gpu/drm/i915/i915_timeline.h
+++ b/drivers/gpu/drm/i915/i915_timeline.h
@@ -159,6 +159,7 @@ static inline bool i915_timeline_sync_is_later(struct i915_timeline *tl,
 }
 
 int i915_timeline_pin(struct i915_timeline *tl);
+int i915_timeline_get_seqno(struct i915_timeline *tl, u32 *seqno);
 void i915_timeline_unpin(struct i915_timeline *tl);
 
 void i915_timelines_init(struct drm_i915_private *i915);
-- 
2.20.1

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

[PATCH 31/33] drm/i915/execlists: Refactor out can_merge_rq()

[Chris Wilson]

In the next patch, we add another user that wants to check whether
requests can be merge into a single HW execution, and in the future we
want to add more conditions under which requests from the same context
cannot be merge. In preparation, extract out can_merge_rq().

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_lrc.c | 27 ++++++++++++++++++---------
 1 file changed, 18 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index cbf20273e9f8..151e4ae48ea0 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -284,12 +284,11 @@ static inline bool need_preempt(const struct intel_engine_cs *engine,
 }
 
 __maybe_unused static inline bool
-assert_priority_queue(const struct intel_engine_execlists *execlists,
-		      const struct i915_request *prev,
+assert_priority_queue(const struct i915_request *prev,
 		      const struct i915_request *next)
 {
-	if (!prev)
-		return true;
+	const struct intel_engine_execlists *execlists =
+		&prev->engine->execlists;
 
 	/*
 	 * Without preemption, the prev may refer to the still active element
@@ -594,6 +593,17 @@ static bool can_merge_ctx(const struct intel_context *prev,
 	return true;
 }
 
+static bool can_merge_rq(const struct i915_request *prev,
+			 const struct i915_request *next)
+{
+	GEM_BUG_ON(!assert_priority_queue(prev, next));
+
+	if (!can_merge_ctx(prev->hw_context, next->hw_context))
+		return false;
+
+	return true;
+}
+
 static void port_assign(struct execlist_port *port, struct i915_request *rq)
 {
 	GEM_BUG_ON(rq == port_request(port));
@@ -744,8 +754,6 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 		int i;
 
 		priolist_for_each_request_consume(rq, rn, p, i) {
-			GEM_BUG_ON(!assert_priority_queue(execlists, last, rq));
-
 			/*
 			 * Can we combine this request with the current port?
 			 * It has to be the same context/ringbuffer and not
@@ -757,8 +765,10 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 			 * second request, and so we never need to tell the
 			 * hardware about the first.
 			 */
-			if (last &&
-			    !can_merge_ctx(rq->hw_context, last->hw_context)) {
+			if (last && !can_merge_rq(last, rq)) {
+				if (last->hw_context == rq->hw_context)
+					goto done;
+
 				/*
 				 * If we are on the second port and cannot
 				 * combine this request with the last, then we
@@ -778,7 +788,6 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 				    ctx_single_port_submission(rq->hw_context))
 					goto done;
 
-				GEM_BUG_ON(last->hw_context == rq->hw_context);
 
 				if (submit)
 					port_assign(port, last);
-- 
2.20.1

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

[PATCH 32/33] drm/i915: Use HW semaphores for inter-engine synchronisation on gen8+

[Chris Wilson]

Having introduced per-context seqno, we know have a means to identity
progress across the system without feel of rollback as befell the
global_seqno. That is we can program a MI_SEMAPHORE_WAIT operation in
advance of submission safe in the knowledge that our target seqno and
address is stable.

However, since we are telling the GPU to busy-spin on the target address
until it matches the signaling seqno, we only want to do so when we are
sure that busy-spin will be completed quickly. To achieve this we only
submit the request to HW once the signaler is itself executing (modulo
preemption causing us to wait longer), and we only do so for default and
above priority requests (so that idle priority tasks never themselves
hog the GPU waiting for others).

But what AB-BA deadlocks? If you remove B, there can be no deadlock...
The issue is that with a deep ELSP queue, we can queue up a pair of
AB-BA on different engines, thus forming a classic mutual exclusion
deadlock. We side-step that issue by restricting the queue depth to
avoid having multiple semaphores in flight and so we only ever take one
set of locks at a time.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_request.c       | 149 +++++++++++++++++++++-
 drivers/gpu/drm/i915/i915_request.h       |   1 +
 drivers/gpu/drm/i915/i915_scheduler.c     |   1 +
 drivers/gpu/drm/i915/i915_scheduler.h     |   1 +
 drivers/gpu/drm/i915/i915_sw_fence.c      |   4 +-
 drivers/gpu/drm/i915/i915_sw_fence.h      |   3 +
 drivers/gpu/drm/i915/intel_gpu_commands.h |   5 +
 drivers/gpu/drm/i915/intel_lrc.c          |  14 +-
 8 files changed, 174 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 690fb4eb6d15..a4318d9a6f34 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -22,8 +22,9 @@
  *
  */
 
-#include <linux/prefetch.h>
 #include <linux/dma-fence-array.h>
+#include <linux/irq_work.h>
+#include <linux/prefetch.h>
 #include <linux/sched.h>
 #include <linux/sched/clock.h>
 #include <linux/sched/signal.h>
@@ -331,6 +332,76 @@ void i915_request_retire_upto(struct i915_request *rq)
 	} while (tmp != rq);
 }
 
+struct execute_cb {
+	struct list_head link;
+	struct irq_work work;
+	struct i915_sw_fence *fence;
+};
+
+static void irq_execute_cb(struct irq_work *wrk)
+{
+	struct execute_cb *cb = container_of(wrk, typeof(*cb), work);
+
+	i915_sw_fence_complete(cb->fence);
+	kfree(cb);
+}
+
+static void __notify_execute_cb(struct i915_request *rq)
+{
+	struct execute_cb *cb;
+
+	lockdep_assert_held(&rq->lock);
+
+	if (list_empty(&rq->execute_cb))
+		return;
+
+	list_for_each_entry(cb, &rq->execute_cb, link)
+		irq_work_queue(&cb->work);
+
+	/*
+	 * XXX Rollback on __i915_request_unsubmit()
+	 *
+	 * In the future, perhaps when we have an active time-slicing scheduler,
+	 * it will be interesting to unsubmit parallel execution and remove
+	 * busywaits from the GPU until their master is restarted. This is
+	 * quite hairy, we have to carefully rollback the fence and do a
+	 * preempt-to-idle cycle on the target engine, all the while the
+	 * master execute_cb may refire.
+	 */
+	INIT_LIST_HEAD(&rq->execute_cb);
+}
+
+static int
+i915_request_await_execution(struct i915_request *rq,
+			     struct i915_request *signal,
+			     gfp_t gfp)
+{
+	struct execute_cb *cb;
+	unsigned long flags;
+
+	if (test_bit(I915_FENCE_FLAG_ACTIVE, &signal->fence.flags))
+		return 0;
+
+	cb = kmalloc(sizeof(*cb), gfp);
+	if (!cb)
+		return -ENOMEM;
+
+	cb->fence = &rq->submit;
+	i915_sw_fence_await(cb->fence);
+	init_irq_work(&cb->work, irq_execute_cb);
+
+	spin_lock_irqsave(&signal->lock, flags);
+	if (test_bit(I915_FENCE_FLAG_ACTIVE, &signal->fence.flags)) {
+		i915_sw_fence_complete(cb->fence);
+		kfree(cb);
+	} else {
+		list_add_tail(&cb->link, &signal->execute_cb);
+	}
+	spin_unlock_irqrestore(&signal->lock, flags);
+
+	return 0;
+}
+
 static void move_to_timeline(struct i915_request *request,
 			     struct i915_timeline *timeline)
 {
@@ -377,6 +448,7 @@ void __i915_request_submit(struct i915_request *request)
 	if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &request->fence.flags) &&
 	    !i915_request_enable_breadcrumb(request))
 		intel_engine_queue_breadcrumbs(engine);
+	__notify_execute_cb(request);
 	spin_unlock(&request->lock);
 
 	engine->emit_fini_breadcrumb(request,
@@ -639,6 +711,7 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 		       tl->fence_context, seqno);
 
 	INIT_LIST_HEAD(&rq->active_list);
+	INIT_LIST_HEAD(&rq->execute_cb);
 	rq->i915 = i915;
 	rq->engine = engine;
 	rq->gem_context = ctx;
@@ -719,6 +792,77 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
 	return ERR_PTR(ret);
 }
 
+static int
+emit_semaphore_wait(struct i915_request *to,
+		    struct i915_request *from,
+		    gfp_t gfp)
+{
+	u32 *cs;
+
+	GEM_BUG_ON(i915_timeline_is_global(from->timeline));
+	GEM_BUG_ON(!from->timeline->has_initial_breadcrumb);
+
+	/*
+	 * If we know our signaling request has started, we know that it
+	 * must, at least, have passed its initial breadcrumb and that its
+	 * seqno can only increase, therefore any change in its breadcrumb
+	 * must indicate completion. By using a "not equal to start" compare
+	 * we avoid the murky issue of how to handle seqno wraparound in an
+	 * async environment (short answer, we must stop the world whenever
+	 * any context wraps!) as the likelihood of missing one request then
+	 * seeing the same start value for a new request is 1 in 2^31, and
+	 * even then we know that the new request has started and is in
+	 * progress, so we are sure it will complete soon enough (not to
+	 * worry about).
+	 */
+	if (i915_request_started(from)) {
+		cs = intel_ring_begin(to, 4);
+		if (IS_ERR(cs))
+			return PTR_ERR(cs);
+
+		*cs++ = MI_SEMAPHORE_WAIT |
+			MI_SEMAPHORE_GLOBAL_GTT |
+			MI_SEMAPHORE_POLL |
+			MI_SEMAPHORE_SAD_NEQ_SDD;
+		*cs++ = from->fence.seqno - 1;
+		*cs++ = i915_timeline_seqno_address(from->timeline);
+		*cs++ = 0;
+
+		intel_ring_advance(to, cs);
+	} else {
+		int err;
+
+		err = i915_request_await_execution(to, from, gfp);
+		if (err)
+			return err;
+
+		cs = intel_ring_begin(to, 4);
+		if (IS_ERR(cs))
+			return PTR_ERR(cs);
+
+		/*
+		 * Using greater-than-or-equal here means we have to worry
+		 * about seqno wraparound. To side step that issue, we swap
+		 * the timeline HWSP upon wrapping, so that everyone listening
+		 * for the old (pre-wrap) values do not see the much smaller
+		 * (post-wrap) values than they were expecting (and so wait
+		 * forever).
+		 */
+		*cs++ = MI_SEMAPHORE_WAIT |
+			MI_SEMAPHORE_GLOBAL_GTT |
+			MI_SEMAPHORE_POLL |
+			MI_SEMAPHORE_SAD_GTE_SDD;
+		*cs++ = from->fence.seqno;
+		*cs++ = i915_timeline_seqno_address(from->timeline);
+		*cs++ = 0;
+
+		intel_ring_advance(to, cs);
+	}
+
+	to->sched.semaphore = true;
+	return 0;
+}
+
 static int
 i915_request_await_request(struct i915_request *to, struct i915_request *from)
 {
@@ -742,6 +886,9 @@ i915_request_await_request(struct i915_request *to, struct i915_request *from)
 		ret = i915_sw_fence_await_sw_fence_gfp(&to->submit,
 						       &from->submit,
 						       I915_FENCE_GFP);
+	} else if (HAS_EXECLISTS(to->i915) &&
+		   to->gem_context->sched.priority >= I915_PRIORITY_NORMAL) {
+		ret = emit_semaphore_wait(to, from, I915_FENCE_GFP);
 	} else {
 		ret = i915_sw_fence_await_dma_fence(&to->submit,
 						    &from->fence, 0,
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 3cffb96203b9..b7c717cec15a 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -127,6 +127,7 @@ struct i915_request {
 	 */
 	struct i915_sw_fence submit;
 	wait_queue_entry_t submitq;
+	struct list_head execute_cb;
 
 	/*
 	 * A list of everyone we wait upon, and everyone who waits upon us.
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index fb5d953430e5..c41a36e48d12 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -29,6 +29,7 @@ void i915_sched_node_init(struct i915_sched_node *node)
 	INIT_LIST_HEAD(&node->waiters_list);
 	INIT_LIST_HEAD(&node->link);
 	node->attr.priority = I915_PRIORITY_INVALID;
+	node->semaphore = false;
 }
 
 static struct i915_dependency *
diff --git a/drivers/gpu/drm/i915/i915_scheduler.h b/drivers/gpu/drm/i915/i915_scheduler.h
index dbe9cb7ecd82..d764cf10536f 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.h
+++ b/drivers/gpu/drm/i915/i915_scheduler.h
@@ -72,6 +72,7 @@ struct i915_sched_node {
 	struct list_head waiters_list; /* those after us, they depend upon us */
 	struct list_head link;
 	struct i915_sched_attr attr;
+	bool semaphore;
 };
 
 struct i915_dependency {
diff --git a/drivers/gpu/drm/i915/i915_sw_fence.c b/drivers/gpu/drm/i915/i915_sw_fence.c
index 7c58b049ecb5..8d1400d378d7 100644
--- a/drivers/gpu/drm/i915/i915_sw_fence.c
+++ b/drivers/gpu/drm/i915/i915_sw_fence.c
@@ -192,7 +192,7 @@ static void __i915_sw_fence_complete(struct i915_sw_fence *fence,
 	__i915_sw_fence_notify(fence, FENCE_FREE);
 }
 
-static void i915_sw_fence_complete(struct i915_sw_fence *fence)
+void i915_sw_fence_complete(struct i915_sw_fence *fence)
 {
 	debug_fence_assert(fence);
 
@@ -202,7 +202,7 @@ static void i915_sw_fence_complete(struct i915_sw_fence *fence)
 	__i915_sw_fence_complete(fence, NULL);
 }
 
-static void i915_sw_fence_await(struct i915_sw_fence *fence)
+void i915_sw_fence_await(struct i915_sw_fence *fence)
 {
 	debug_fence_assert(fence);
 	WARN_ON(atomic_inc_return(&fence->pending) <= 1);
diff --git a/drivers/gpu/drm/i915/i915_sw_fence.h b/drivers/gpu/drm/i915/i915_sw_fence.h
index 0e055ea0179f..6dec9e1d1102 100644
--- a/drivers/gpu/drm/i915/i915_sw_fence.h
+++ b/drivers/gpu/drm/i915/i915_sw_fence.h
@@ -79,6 +79,9 @@ int i915_sw_fence_await_reservation(struct i915_sw_fence *fence,
 				    unsigned long timeout,
 				    gfp_t gfp);
 
+void i915_sw_fence_await(struct i915_sw_fence *fence);
+void i915_sw_fence_complete(struct i915_sw_fence *fence);
+
 static inline bool i915_sw_fence_signaled(const struct i915_sw_fence *fence)
 {
 	return atomic_read(&fence->pending) <= 0;
diff --git a/drivers/gpu/drm/i915/intel_gpu_commands.h b/drivers/gpu/drm/i915/intel_gpu_commands.h
index 105e2a9e874a..bafad94f751b 100644
--- a/drivers/gpu/drm/i915/intel_gpu_commands.h
+++ b/drivers/gpu/drm/i915/intel_gpu_commands.h
@@ -106,7 +106,12 @@
 #define   MI_SEMAPHORE_TARGET(engine)	((engine)<<15)
 #define MI_SEMAPHORE_WAIT	MI_INSTR(0x1c, 2) /* GEN8+ */
 #define   MI_SEMAPHORE_POLL		(1<<15)
+#define   MI_SEMAPHORE_SAD_GT_SDD	(0<<12)
 #define   MI_SEMAPHORE_SAD_GTE_SDD	(1<<12)
+#define   MI_SEMAPHORE_SAD_LT_SDD	(2<<12)
+#define   MI_SEMAPHORE_SAD_LTE_SDD	(3<<12)
+#define   MI_SEMAPHORE_SAD_EQ_SDD	(4<<12)
+#define   MI_SEMAPHORE_SAD_NEQ_SDD	(5<<12)
 #define MI_STORE_DWORD_IMM	MI_INSTR(0x20, 1)
 #define MI_STORE_DWORD_IMM_GEN4	MI_INSTR(0x20, 2)
 #define   MI_MEM_VIRTUAL	(1 << 22) /* 945,g33,965 */
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 151e4ae48ea0..113d6fa8b190 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -414,7 +414,8 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine)
 	 * stream, so give it the equivalent small priority bump to prevent
 	 * it being gazumped a second time by another peer.
 	 */
-	if ((prio & ACTIVE_PRIORITY) != ACTIVE_PRIORITY) {
+	if ((prio & ACTIVE_PRIORITY) != ACTIVE_PRIORITY &&
+	    i915_request_started(active)) {
 		prio |= ACTIVE_PRIORITY;
 		active->sched.attr.priority = prio;
 		list_move_tail(&active->sched.link,
@@ -598,6 +599,17 @@ static bool can_merge_rq(const struct i915_request *prev,
 {
 	GEM_BUG_ON(!assert_priority_queue(prev, next));
 
+	/*
+	 * To avoid AB-BA deadlocks, we simply restrict ourselves to only
+	 * submitting one semaphore (think HW spinlock) to HW at a time. This
+	 * prevents the execution callback on a later sempahore from being
+	 * queued on another engine, so no cycle can be formed. Preemption
+	 * rules should mean that if this semaphore is preempted, its
+	 * dependency chain is preserved and suitably promoted via PI.
+	 */
+	if (prev->sched.semaphore && !i915_request_started(prev))
+		return false;
+
 	if (!can_merge_ctx(prev->hw_context, next->hw_context))
 		return false;
 
-- 
2.20.1

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

[PATCH 33/33] drm/i915: Prioritise non-busywait semaphore workloads

[Chris Wilson]

We don't want to busywait on the GPU if we have other work to do. If we
give non-busywaiting workloads higher (initial) priority than workloads
that require a busywait, we will prioritise work that is ready to run
immediately.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_request.c   | 3 +++
 drivers/gpu/drm/i915/i915_scheduler.h | 7 ++++---
 drivers/gpu/drm/i915/intel_lrc.c      | 2 +-
 3 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index a4318d9a6f34..5bda7356773c 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1132,6 +1132,9 @@ void i915_request_add(struct i915_request *request)
 	if (engine->schedule) {
 		struct i915_sched_attr attr = request->gem_context->sched;
 
+		if (!request->sched.semaphore)
+			attr.priority |= I915_PRIORITY_NOSEMAPHORE;
+
 		/*
 		 * Boost priorities to new clients (new request flows).
 		 *
diff --git a/drivers/gpu/drm/i915/i915_scheduler.h b/drivers/gpu/drm/i915/i915_scheduler.h
index d764cf10536f..7f194a8db785 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.h
+++ b/drivers/gpu/drm/i915/i915_scheduler.h
@@ -24,14 +24,15 @@ enum {
 	I915_PRIORITY_INVALID = INT_MIN
 };
 
-#define I915_USER_PRIORITY_SHIFT 2
+#define I915_USER_PRIORITY_SHIFT 3
 #define I915_USER_PRIORITY(x) ((x) << I915_USER_PRIORITY_SHIFT)
 
 #define I915_PRIORITY_COUNT BIT(I915_USER_PRIORITY_SHIFT)
 #define I915_PRIORITY_MASK (I915_PRIORITY_COUNT - 1)
 
-#define I915_PRIORITY_WAIT	((u8)BIT(0))
-#define I915_PRIORITY_NEWCLIENT	((u8)BIT(1))
+#define I915_PRIORITY_WAIT		((u8)BIT(0))
+#define I915_PRIORITY_NEWCLIENT		((u8)BIT(1))
+#define I915_PRIORITY_NOSEMAPHORE	((u8)BIT(2))
 
 struct i915_sched_attr {
 	/**
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 113d6fa8b190..0be09ceb6c7c 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -164,7 +164,7 @@
 #define WA_TAIL_DWORDS 2
 #define WA_TAIL_BYTES (sizeof(u32) * WA_TAIL_DWORDS)
 
-#define ACTIVE_PRIORITY (I915_PRIORITY_NEWCLIENT)
+#define ACTIVE_PRIORITY (I915_PRIORITY_NEWCLIENT | I915_PRIORITY_NOSEMAPHORE)
 
 static int execlists_context_deferred_alloc(struct i915_gem_context *ctx,
 					    struct intel_engine_cs *engine,
-- 
2.20.1

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

✗ Fi.CI.CHECKPATCH: warning for series starting with [01/33] drm/i915/execlists: Move RPCS setup to context pin

[Patchwork]

== Series Details ==

Series: series starting with [01/33] drm/i915/execlists: Move RPCS setup to context pin
URL   : https://patchwork.freedesktop.org/series/55708/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
6abb3bdb8c37 drm/i915/execlists: Move RPCS setup to context pin
44e01a9b8ba3 drm/i915: Measure the required reserved size for request emission
a8be5f976f88 drm/i915: Remove manual breadcumb counting
b4de6f7897c8 drm/i915: Compute the HWS offsets explicitly
2ba1c33378f9 drm/i915/execlists: Suppress preempting self
-:22: WARNING:COMMIT_LOG_LONG_LINE: Possible unwrapped commit description (prefer a maximum 75 chars per line)
#22: 
References: a2bf92e8cc16 ("drm/i915/execlists: Avoid kicking priority on the current context")

-:22: ERROR:GIT_COMMIT_ID: Please use git commit description style 'commit <12+ chars of sha1> ("<title line>")' - ie: 'commit a2bf92e8cc16 ("drm/i915/execlists: Avoid kicking priority on the current context")'
#22: 
References: a2bf92e8cc16 ("drm/i915/execlists: Avoid kicking priority on the current context")

total: 1 errors, 1 warnings, 0 checks, 156 lines checked
d23882a69bf8 drm/i915/execlists: Suppress redundant preemption
f6a7e760a383 drm/i915/selftests: Apply a subtest filter
6a91ea6fd2c7 drm/i915: Make all GPU resets atomic
-:33: CHECK:USLEEP_RANGE: usleep_range is preferred over udelay; see Documentation/timers/timers-howto.txt
#33: FILE: drivers/gpu/drm/i915/i915_reset.c:149:
+	udelay(50);

-:39: CHECK:USLEEP_RANGE: usleep_range is preferred over udelay; see Documentation/timers/timers-howto.txt
#39: FILE: drivers/gpu/drm/i915/i915_reset.c:154:
+	udelay(50);

total: 0 errors, 0 warnings, 2 checks, 176 lines checked
6955a17b590b drm/i915/guc: Disable global reset
0e46060a69ff drm/i915: Remove GPU reset dependence on struct_mutex
-:1279: WARNING:BOOL_BITFIELD: Avoid using bool as bitfield.  Prefer bool bitfields as unsigned int or u<8|16|32>
#1279: FILE: drivers/gpu/drm/i915/intel_hangcheck.c:35:
+	bool wedged:1;

-:1280: WARNING:BOOL_BITFIELD: Avoid using bool as bitfield.  Prefer bool bitfields as unsigned int or u<8|16|32>
#1280: FILE: drivers/gpu/drm/i915/intel_hangcheck.c:36:
+	bool stalled:1;

total: 0 errors, 2 warnings, 0 checks, 1705 lines checked
4f8bac3aae86 drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest
54d87f290739 drm/i915: Issue engine resets onto idle engines
ceab76765075 drm/i915: Stop tracking MRU activity on VMA
0f21396816a4 drm/i915: Pull VM lists under the VM mutex.
fdb647847171 drm/i915: Move vma lookup to its own lock
-:161: WARNING:USE_SPINLOCK_T: struct spinlock should be spinlock_t
#161: FILE: drivers/gpu/drm/i915/i915_gem_object.h:94:
+		struct spinlock lock;

total: 0 errors, 1 warnings, 0 checks, 290 lines checked
56860703e5fd drm/i915: Always allocate an object/vma for the HWSP
a2ba7cb66608 drm/i915: Add timeline barrier support
ca873c95645e drm/i915: Move list of timelines under its own lock
5263a04ba167 drm/i915: Introduce concept of per-timeline (context) HWSP
2d8b70963a93 drm/i915: Enlarge vma->pin_count
83d209a5ffec drm/i915: Allocate a status page for each timeline
9b0e47a0b162 drm/i915: Share per-timeline HWSP using a slab suballocator
-:79: CHECK:SPACING: No space is necessary after a cast
#79: FILE: drivers/gpu/drm/i915/i915_timeline.c:43:
+	BUILD_BUG_ON(BITS_PER_TYPE(u64) * CACHELINE_BYTES > PAGE_SIZE);

total: 0 errors, 0 warnings, 1 checks, 415 lines checked
4bd03fef836a drm/i915: Track the context's seqno in its own timeline HWSP
083196e5154c drm/i915: Track active timelines
67501f1cc318 drm/i915: Identify active requests
cf16be448222 drm/i915: Remove the intel_engine_notify tracepoint
e94b3cd6cc75 drm/i915: Replace global breadcrumbs with per-context interrupt tracking
-:18: ERROR:GIT_COMMIT_ID: Please use git commit description style 'commit <12+ chars of sha1> ("<title line>")' - ie: 'commit 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")'
#18: 
Before commit 688e6c725816, the solution was simple. Every client waking

-:21: ERROR:GIT_COMMIT_ID: Please use git commit description style 'commit <12+ chars of sha1> ("<title line>")' - ie: 'commit 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")'
#21: 
688e6c725816 introduced an rbtree so that only the earliest waiter on

-:55: WARNING:COMMIT_LOG_LONG_LINE: Possible unwrapped commit description (prefer a maximum 75 chars per line)
#55: 
References: 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")

-:55: ERROR:GIT_COMMIT_ID: Please use git commit description style 'commit <12+ chars of sha1> ("<title line>")' - ie: 'commit 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")'
#55: 
References: 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")

-:2179: WARNING:FUNCTION_ARGUMENTS: function definition argument 'struct i915_gem_context *' should also have an identifier name
#2179: FILE: drivers/gpu/drm/i915/selftests/i915_request.c:258:
+	struct i915_request *(*request_alloc)(struct i915_gem_context *,

-:2179: WARNING:FUNCTION_ARGUMENTS: function definition argument 'struct intel_engine_cs *' should also have an identifier name
#2179: FILE: drivers/gpu/drm/i915/selftests/i915_request.c:258:
+	struct i915_request *(*request_alloc)(struct i915_gem_context *,

-:2205: WARNING:LINE_SPACING: Missing a blank line after declarations
#2205: FILE: drivers/gpu/drm/i915/selftests/i915_request.c:284:
+	struct i915_request **requests;
+	I915_RND_STATE(prng);

-:2632: WARNING:FILE_PATH_CHANGES: added, moved or deleted file(s), does MAINTAINERS need updating?
#2632: 
deleted file mode 100644

total: 3 errors, 5 warnings, 0 checks, 2572 lines checked
a265ac2617cc drm/i915: Drop fake breadcrumb irq
174f53eceb3f drm/i915: Implement an "idle" barrier
3b1ab1239de6 drm/i915: Keep timeline HWSP allocated until the system is idle
1ed416939ae4 drm/i915/execlists: Refactor out can_merge_rq()
a8ca508c4233 drm/i915: Use HW semaphores for inter-engine synchronisation on gen8+
-:306: CHECK:SPACING: spaces preferred around that '<<' (ctx:VxV)
#306: FILE: drivers/gpu/drm/i915/intel_gpu_commands.h:109:
+#define   MI_SEMAPHORE_SAD_GT_SDD	(0<<12)
                                  	  ^

-:308: CHECK:SPACING: spaces preferred around that '<<' (ctx:VxV)
#308: FILE: drivers/gpu/drm/i915/intel_gpu_commands.h:111:
+#define   MI_SEMAPHORE_SAD_LT_SDD	(2<<12)
                                  	  ^

-:309: CHECK:SPACING: spaces preferred around that '<<' (ctx:VxV)
#309: FILE: drivers/gpu/drm/i915/intel_gpu_commands.h:112:
+#define   MI_SEMAPHORE_SAD_LTE_SDD	(3<<12)
                                   	  ^

-:310: CHECK:SPACING: spaces preferred around that '<<' (ctx:VxV)
#310: FILE: drivers/gpu/drm/i915/intel_gpu_commands.h:113:
+#define   MI_SEMAPHORE_SAD_EQ_SDD	(4<<12)
                                  	  ^

-:311: CHECK:SPACING: spaces preferred around that '<<' (ctx:VxV)
#311: FILE: drivers/gpu/drm/i915/intel_gpu_commands.h:114:
+#define   MI_SEMAPHORE_SAD_NEQ_SDD	(5<<12)
                                   	  ^

total: 0 errors, 0 warnings, 5 checks, 270 lines checked
008bb5355846 drm/i915: Prioritise non-busywait semaphore workloads

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

✗ Fi.CI.SPARSE: warning for series starting with [01/33] drm/i915/execlists: Move RPCS setup to context pin

[Patchwork]

== Series Details ==

Series: series starting with [01/33] drm/i915/execlists: Move RPCS setup to context pin
URL   : https://patchwork.freedesktop.org/series/55708/
State : warning

== Summary ==

$ dim sparse origin/drm-tip
Sparse version: v0.5.2
Commit: drm/i915/execlists: Move RPCS setup to context pin
Okay!

Commit: drm/i915: Measure the required reserved size for request emission
Okay!

Commit: drm/i915: Remove manual breadcumb counting
Okay!

Commit: drm/i915: Compute the HWS offsets explicitly
Okay!

Commit: drm/i915/execlists: Suppress preempting self
-drivers/gpu/drm/i915/intel_ringbuffer.h:602:23: warning: expression using sizeof(void)

Commit: drm/i915/execlists: Suppress redundant preemption
Okay!

Commit: drm/i915/selftests: Apply a subtest filter
Okay!

Commit: drm/i915: Make all GPU resets atomic
Okay!

Commit: drm/i915/guc: Disable global reset
Okay!

Commit: drm/i915: Remove GPU reset dependence on struct_mutex
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3546:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3541:16: warning: expression using sizeof(void)

Commit: drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest
Okay!

Commit: drm/i915: Issue engine resets onto idle engines
Okay!

Commit: drm/i915: Stop tracking MRU activity on VMA
Okay!

Commit: drm/i915: Pull VM lists under the VM mutex.
Okay!

Commit: drm/i915: Move vma lookup to its own lock
Okay!

Commit: drm/i915: Always allocate an object/vma for the HWSP
Okay!

Commit: drm/i915: Add timeline barrier support
Okay!

Commit: drm/i915: Move list of timelines under its own lock
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3541:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3544:16: warning: expression using sizeof(void)

Commit: drm/i915: Introduce concept of per-timeline (context) HWSP
Okay!

Commit: drm/i915: Enlarge vma->pin_count
Okay!

Commit: drm/i915: Allocate a status page for each timeline
+./include/linux/mm.h:619:13: error: not a function <noident>
+./include/linux/mm.h:619:13: error: undefined identifier '__builtin_mul_overflow'
+./include/linux/mm.h:619:13: warning: call with no type!

Commit: drm/i915: Share per-timeline HWSP using a slab suballocator
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3544:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3548:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/i915_timeline.c:89:38: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/i915_timeline.c:89:38: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/i915_timeline.c:92:44: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/i915_timeline.c:92:44: warning: expression using sizeof(void)
+./include/linux/slab.h:664:13: error: undefined identifier '__builtin_mul_overflow'
+./include/linux/slab.h:664:13: warning: call with no type!

Commit: drm/i915: Track the context's seqno in its own timeline HWSP
Okay!

Commit: drm/i915: Track active timelines
Okay!

Commit: drm/i915: Identify active requests
Okay!

Commit: drm/i915: Remove the intel_engine_notify tracepoint
Okay!

Commit: drm/i915: Replace global breadcrumbs with per-context interrupt tracking
+drivers/gpu/drm/i915/selftests/i915_request.c:280:40: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/i915_request.c:280:40: warning: expression using sizeof(void)
-./include/linux/mm.h:619:13: error: not a function <noident>
-./include/linux/mm.h:619:13: error: not a function <noident>
-./include/linux/mm.h:619:13: error: undefined identifier '__builtin_mul_overflow'
-./include/linux/mm.h:619:13: warning: call with no type!
+./include/linux/slab.h:664:13: error: not a function <noident>
+./include/linux/slab.h:664:13: error: not a function <noident>

Commit: drm/i915: Drop fake breadcrumb irq
Okay!

Commit: drm/i915: Implement an "idle" barrier
-drivers/gpu/drm/i915/selftests/../i915_drv.h:3548:16: warning: expression using sizeof(void)
+drivers/gpu/drm/i915/selftests/../i915_drv.h:3553:16: warning: expression using sizeof(void)

Commit: drm/i915: Keep timeline HWSP allocated until the system is idle
Okay!

Commit: drm/i915/execlists: Refactor out can_merge_rq()
Okay!

Commit: drm/i915: Use HW semaphores for inter-engine synchronisation on gen8+
Okay!

Commit: drm/i915: Prioritise non-busywait semaphore workloads
Okay!

_______________________________________________
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 [01/33] drm/i915/execlists: Move RPCS setup to context pin

[Patchwork]

== Series Details ==

Series: series starting with [01/33] drm/i915/execlists: Move RPCS setup to context pin
URL   : https://patchwork.freedesktop.org/series/55708/
State : failure

== Summary ==

CI Bug Log - changes from CI_DRM_5478 -> Patchwork_12035
====================================================

Summary
-------

  **FAILURE**

  Serious unknown changes coming with Patchwork_12035 absolutely need to be
  verified manually.
  
  If you think the reported changes have nothing to do with the changes
  introduced in Patchwork_12035, please notify your bug team to allow them
  to document this new failure mode, which will reduce false positives in CI.

  External URL: https://patchwork.freedesktop.org/api/1.0/series/55708/revisions/1/mbox/

Possible new issues
-------------------

  Here are the unknown changes that may have been introduced in Patchwork_12035:

### IGT changes ###

#### Possible regressions ####

  * igt@gem_ctx_create@basic-files:
    - fi-cfl-guc:         PASS -> DMESG-WARN
    - fi-kbl-guc:         PASS -> DMESG-WARN

  
#### Suppressed ####

  The following results come from untrusted machines, tests, or statuses.
  They do not affect the overall result.

  * igt@gem_ringfill@basic-default-fd:
    - {fi-icl-u3}:        PASS -> INCOMPLETE
    - {fi-icl-u2}:        PASS -> INCOMPLETE

  
Known issues
------------

  Here are the changes found in Patchwork_12035 that come from known issues:

### IGT changes ###

#### Issues hit ####

  * igt@i915_selftest@live_execlists:
    - fi-apl-guc:         PASS -> INCOMPLETE [fdo#103927]

  * igt@kms_chamelium@hdmi-hpd-fast:
    - fi-kbl-7500u:       PASS -> FAIL [fdo#108767]

  * igt@pm_rpm@module-reload:
    - fi-hsw-4770:        NOTRUN -> INCOMPLETE [fdo#107807]

  * igt@prime_vgem@basic-fence-flip:
    - fi-ilk-650:         PASS -> FAIL [fdo#104008]

  
#### Possible fixes ####

  * igt@i915_module_load@reload:
    - fi-blb-e6850:       INCOMPLETE [fdo#107718] -> PASS

  * igt@i915_selftest@live_hangcheck:
    - fi-bwr-2160:        DMESG-FAIL [fdo#108735] -> PASS

  * igt@kms_busy@basic-flip-b:
    - fi-gdg-551:         FAIL [fdo#103182] -> PASS

  * igt@kms_pipe_crc_basic@nonblocking-crc-pipe-b:
    - fi-byt-clapper:     FAIL [fdo#107362] -> PASS

  * igt@kms_pipe_crc_basic@suspend-read-crc-pipe-a:
    - fi-byt-clapper:     FAIL [fdo#103191] / [fdo#107362] -> PASS +1

  
  {name}: This element is suppressed. This means it is ignored when computing
          the status of the difference (SUCCESS, WARNING, or FAILURE).

  [fdo#103182]: https://bugs.freedesktop.org/show_bug.cgi?id=103182
  [fdo#103191]: https://bugs.freedesktop.org/show_bug.cgi?id=103191
  [fdo#103927]: https://bugs.freedesktop.org/show_bug.cgi?id=103927
  [fdo#104008]: https://bugs.freedesktop.org/show_bug.cgi?id=104008
  [fdo#107362]: https://bugs.freedesktop.org/show_bug.cgi?id=107362
  [fdo#107718]: https://bugs.freedesktop.org/show_bug.cgi?id=107718
  [fdo#107807]: https://bugs.freedesktop.org/show_bug.cgi?id=107807
  [fdo#108622]: https://bugs.freedesktop.org/show_bug.cgi?id=108622
  [fdo#108735]: https://bugs.freedesktop.org/show_bug.cgi?id=108735
  [fdo#108767]: https://bugs.freedesktop.org/show_bug.cgi?id=108767
  [fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271


Participating hosts (44 -> 40)
------------------------------

  Additional (1): fi-hsw-4770 
  Missing    (5): fi-kbl-soraka fi-ilk-m540 fi-byt-squawks fi-bsw-cyan fi-pnv-d510 


Build changes
-------------

    * Linux: CI_DRM_5478 -> Patchwork_12035

  CI_DRM_5478: c6308f6a6b019fbea6c4179e2668f1e38b673eab @ git://anongit.freedesktop.org/gfx-ci/linux
  IGT_4790: dcdf4b04e16312f8f52ad389388d834f9d74b8f0 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
  Patchwork_12035: 008bb53558465c1b58c4be2368a7e206943a1dff @ git://anongit.freedesktop.org/gfx-ci/linux


== Linux commits ==

008bb5355846 drm/i915: Prioritise non-busywait semaphore workloads
a8ca508c4233 drm/i915: Use HW semaphores for inter-engine synchronisation on gen8+
1ed416939ae4 drm/i915/execlists: Refactor out can_merge_rq()
3b1ab1239de6 drm/i915: Keep timeline HWSP allocated until the system is idle
174f53eceb3f drm/i915: Implement an "idle" barrier
a265ac2617cc drm/i915: Drop fake breadcrumb irq
e94b3cd6cc75 drm/i915: Replace global breadcrumbs with per-context interrupt tracking
cf16be448222 drm/i915: Remove the intel_engine_notify tracepoint
67501f1cc318 drm/i915: Identify active requests
083196e5154c drm/i915: Track active timelines
4bd03fef836a drm/i915: Track the context's seqno in its own timeline HWSP
9b0e47a0b162 drm/i915: Share per-timeline HWSP using a slab suballocator
83d209a5ffec drm/i915: Allocate a status page for each timeline
2d8b70963a93 drm/i915: Enlarge vma->pin_count
5263a04ba167 drm/i915: Introduce concept of per-timeline (context) HWSP
ca873c95645e drm/i915: Move list of timelines under its own lock
a2ba7cb66608 drm/i915: Add timeline barrier support
56860703e5fd drm/i915: Always allocate an object/vma for the HWSP
fdb647847171 drm/i915: Move vma lookup to its own lock
0f21396816a4 drm/i915: Pull VM lists under the VM mutex.
ceab76765075 drm/i915: Stop tracking MRU activity on VMA
54d87f290739 drm/i915: Issue engine resets onto idle engines
4f8bac3aae86 drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest
0e46060a69ff drm/i915: Remove GPU reset dependence on struct_mutex
6955a17b590b drm/i915/guc: Disable global reset
6a91ea6fd2c7 drm/i915: Make all GPU resets atomic
f6a7e760a383 drm/i915/selftests: Apply a subtest filter
d23882a69bf8 drm/i915/execlists: Suppress redundant preemption
2ba1c33378f9 drm/i915/execlists: Suppress preempting self
b4de6f7897c8 drm/i915: Compute the HWS offsets explicitly
a8be5f976f88 drm/i915: Remove manual breadcumb counting
44e01a9b8ba3 drm/i915: Measure the required reserved size for request emission
6abb3bdb8c37 drm/i915/execlists: Move RPCS setup to context pin

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_12035/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx