Skip to contentSkip to conversation

drm/i915/bxt: work around HW coherency issue

PATCH Superseded Imre Deak 4 participants 16 messages Aug 14, 2015 – Aug 26, 2015

Subsystems (2)
lore
Share & tools
lore

Imre Deak

author
#0 lore raw

v2 cover drm/i915/bxt: work around HW coherency issue

This is a v2 of [1]. Since v1 the HW team confirmed that there is an HW issue in A steppings with the GPU/CPU snoop logic, which explains why we need this workaround.

In v2 I fixed a typo and limited the workaround to A steppings, since in later steppings this problem is fixed.

[1]
http://lists.freedesktop.org/archives/intel-gfx/2015-June/068218.html

Imre Deak (2):
drm/i915/bxt: work around HW coherency issue when accessing GPU seqno drm/i915/bxt: work around HW coherency issue for cached GEM mappings

 drivers/gpu/drm/i915/i915_gem.c         | 11 ++++++++++-
 drivers/gpu/drm/i915/intel_lrc.c        | 20 ++++++++++++++++++++
 drivers/gpu/drm/i915/intel_ringbuffer.h |  7 +++++++
 3 files changed, 37 insertions(+), 1 deletion(-)
Signature · 6 lines
2.1.4

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#1 lore raw

PATCH v2 1/2 drm/i915/bxt: work around HW coherency issue when accessing GPU seqno

By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one.

Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view.

v2:
- fix using the proper logical && instead of a bitwise & (Jani, Mika)
- limit the workaround to A stepping, on later steppings this HW issue is fixed

Testcase: igt/store_dword_loop_render

---
 drivers/gpu/drm/i915/intel_lrc.c        | 20 ++++++++++++++++++++
 drivers/gpu/drm/i915/intel_ringbuffer.h |  7 +++++++
 2 files changed, 27 insertions(+)
drivers/gpu/drm/i915/intel_lrc.c +20 -0
index 138964a..46f2be0 100644
@@ -1691,12 +1691,32 @@ static int gen8_emit_flush_render(struct drm_i915_gem_request *request,
 
 static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency)
 {
+
+	/*
+	 * On BXT A steppings there is a HW coherency issue whereby the
+	 * MI_STORE_DATA_IMM storing the completed request's seqno
+	 * occasionally doesn't invalidate the CPU cache. Work around this by
+	 * clflushing the corresponding cacheline whenever the caller wants
+	 * the coherency to be guaranteed. Note that this cacheline is known
+	 * to be clean at this point, since we only write it in
+	 * gen8_set_seqno(), where we also do a clflush after the write. So
+	 * this clflush in practice becomes an invalidate operation.
+	 */
+
+	if (IS_BROXTON(ring->dev) && INTEL_REVID(ring->dev) < BXT_REVID_B0 &&
+	    !lazy_coherency)
+		intel_flush_status_page(ring, I915_GEM_HWS_INDEX);
+
 	return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
 }
 
 static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno)
 {
 	intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno);
+
+	/* See gen8_get_seqno() explaining the reason for the clflush. */
+	if (IS_BROXTON(ring->dev) && INTEL_REVID(ring->dev) < BXT_REVID_B0)
+		intel_flush_status_page(ring, I915_GEM_HWS_INDEX);
 }
 
 static int gen8_emit_request(struct drm_i915_gem_request *request)
drivers/gpu/drm/i915/intel_ringbuffer.h +7 -0
index 2e85fda..95b0b4b 100644
@@ -377,6 +377,13 @@ intel_ring_sync_index(struct intel_engine_cs *ring,
 	return idx;
 }
 
+static inline void
+intel_flush_status_page(struct intel_engine_cs *ring, int reg)
+{
+	drm_clflush_virt_range(&ring->status_page.page_addr[reg],
+			       sizeof(uint32_t));
+}
+
 static inline u32
 intel_read_status_page(struct intel_engine_cs *ring,
 		       int reg)
Signature · 6 lines
2.1.4

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [PATCH v2 1/2] drm/i915/bxt: work around HW coherency issue when accessing GPU seqno

Quoted 10 lines · Imre Deak

By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one.

Quoted 8 lines

Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view.

v2:
- fix using the proper logical && instead of a bitwise & (Jani, Mika)
- limit the workaround to A stepping, on later steppings this HW issue is fixed

We have vfuncs in order to avoid the pointer dance (and boy is it a pretty and quite convoluted dance).
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#3 lore raw

Re: [PATCH v2 1/2] drm/i915/bxt: work around HW coherency issue when accessing GPU seqno

Quoted 15 lines · Chris Wilson
Quoted 14 lines · Imre Deak

By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one.

Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view.

Quoted 7 lines
Quoted 4 lines

v2:
- fix using the proper logical && instead of a bitwise & (Jani, Mika)
- limit the workaround to A stepping, on later steppings this HW issue is fixed

We have vfuncs in order to avoid the pointer dance (and boy is it a pretty and quite convoluted dance).

Ok, I'll add new get_seqno/set_seqno vfuncs.

Quoted 2 lines

-Chris

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

author
#4 lore raw

PATCH v3 1/2 drm/i915/bxt: work around HW coherency issue when accessing GPU seqno

In reply to Imre Deak

By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one.

Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view.

v2:
- fix using the proper logical && instead of a bitwise & (Jani, Mika)
- limit the workaround to A stepping, on later steppings this HW issue is fixed
v3:
- use a separate get_seqno/set_seqno vfunc (Chris)

Testcase: igt/store_dword_loop_render

---
 drivers/gpu/drm/i915/intel_lrc.c        | 64 ++++++++++++++++++++++++++++-----
 drivers/gpu/drm/i915/intel_ringbuffer.h |  7 ++++
 2 files changed, 63 insertions(+), 8 deletions(-)
drivers/gpu/drm/i915/intel_lrc.c +56 -8
index 138964a..1269d57 100644
@@ -1699,6 +1699,34 @@ static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno)
 	intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno);
 }
 
+static u32 bxt_a_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency)
+{
+
+	/*
+	 * On BXT A steppings there is a HW coherency issue whereby the
+	 * MI_STORE_DATA_IMM storing the completed request's seqno
+	 * occasionally doesn't invalidate the CPU cache. Work around this by
+	 * clflushing the corresponding cacheline whenever the caller wants
+	 * the coherency to be guaranteed. Note that this cacheline is known
+	 * to be clean at this point, since we only write it in
+	 * bxt_a_set_seqno(), where we also do a clflush after the write. So
+	 * this clflush in practice becomes an invalidate operation.
+	 */
+
+	if (!lazy_coherency)
+		intel_flush_status_page(ring, I915_GEM_HWS_INDEX);
+
+	return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
+}
+
+static void bxt_a_set_seqno(struct intel_engine_cs *ring, u32 seqno)
+{
+	intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno);
+
+	/* See bxt_a_get_seqno() explaining the reason for the clflush. */
+	intel_flush_status_page(ring, I915_GEM_HWS_INDEX);
+}
+
 static int gen8_emit_request(struct drm_i915_gem_request *request)
 {
 	struct intel_ringbuffer *ringbuf = request->ringbuf;
@@ -1868,8 +1896,13 @@ static int logical_render_ring_init(struct drm_device *dev)
 		ring->init_hw = gen8_init_render_ring;
 	ring->init_context = gen8_init_rcs_context;
 	ring->cleanup = intel_fini_pipe_control;
-	ring->get_seqno = gen8_get_seqno;
-	ring->set_seqno = gen8_set_seqno;
+	if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) {
+		ring->get_seqno = bxt_a_get_seqno;
+		ring->set_seqno = bxt_a_set_seqno;
+	} else {
+		ring->get_seqno = gen8_get_seqno;
+		ring->set_seqno = gen8_set_seqno;
+	}
 	ring->emit_request = gen8_emit_request;
 	ring->emit_flush = gen8_emit_flush_render;
 	ring->irq_get = gen8_logical_ring_get_irq;
@@ -1915,8 +1948,13 @@ static int logical_bsd_ring_init(struct drm_device *dev)
 		GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT;
 
 	ring->init_hw = gen8_init_common_ring;
-	ring->get_seqno = gen8_get_seqno;
-	ring->set_seqno = gen8_set_seqno;
+	if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) {
+		ring->get_seqno = bxt_a_get_seqno;
+		ring->set_seqno = bxt_a_set_seqno;
+	} else {
+		ring->get_seqno = gen8_get_seqno;
+		ring->set_seqno = gen8_set_seqno;
+	}
 	ring->emit_request = gen8_emit_request;
 	ring->emit_flush = gen8_emit_flush;
 	ring->irq_get = gen8_logical_ring_get_irq;
@@ -1965,8 +2003,13 @@ static int logical_blt_ring_init(struct drm_device *dev)
 		GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT;
 
 	ring->init_hw = gen8_init_common_ring;
-	ring->get_seqno = gen8_get_seqno;
-	ring->set_seqno = gen8_set_seqno;
+	if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) {
+		ring->get_seqno = bxt_a_get_seqno;
+		ring->set_seqno = bxt_a_set_seqno;
+	} else {
+		ring->get_seqno = gen8_get_seqno;
+		ring->set_seqno = gen8_set_seqno;
+	}
 	ring->emit_request = gen8_emit_request;
 	ring->emit_flush = gen8_emit_flush;
 	ring->irq_get = gen8_logical_ring_get_irq;
@@ -1990,8 +2033,13 @@ static int logical_vebox_ring_init(struct drm_device *dev)
 		GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT;
 
 	ring->init_hw = gen8_init_common_ring;
-	ring->get_seqno = gen8_get_seqno;
-	ring->set_seqno = gen8_set_seqno;
+	if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) {
+		ring->get_seqno = bxt_a_get_seqno;
+		ring->set_seqno = bxt_a_set_seqno;
+	} else {
+		ring->get_seqno = gen8_get_seqno;
+		ring->set_seqno = gen8_set_seqno;
+	}
 	ring->emit_request = gen8_emit_request;
 	ring->emit_flush = gen8_emit_flush;
 	ring->irq_get = gen8_logical_ring_get_irq;
drivers/gpu/drm/i915/intel_ringbuffer.h +7 -0
index 2e85fda..95b0b4b 100644
@@ -377,6 +377,13 @@ intel_ring_sync_index(struct intel_engine_cs *ring,
 	return idx;
 }
 
+static inline void
+intel_flush_status_page(struct intel_engine_cs *ring, int reg)
+{
+	drm_clflush_virt_range(&ring->status_page.page_addr[reg],
+			       sizeof(uint32_t));
+}
+
 static inline u32
 intel_read_status_page(struct intel_engine_cs *ring,
 		       int reg)
Signature · 6 lines
2.1.4

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Reviewed-by on v3 1/2
#5 lore raw

Re: [PATCH v3 1/2] drm/i915/bxt: work around HW coherency issue when accessing GPU seqno

Quoted 21 lines · Imre Deak

By running igt/store_dword_loop_render on BXT we can hit a coherency problem where the seqno written at GPU command completion time is not seen by the CPU. This results in __i915_wait_request seeing the stale seqno and not completing the request (not considering the lost interrupt/GPU reset mechanism). I also verified that this isn't a case of a lost interrupt, or that the command didn't complete somehow: when the coherency issue occured I read the seqno via an uncached GTT mapping too. While the cached version of the seqno still showed the stale value the one read via the uncached mapping was the correct one.

Work around this issue by clflushing the corresponding CPU cacheline following any store of the seqno and preceding any reading of it. When reading it do this only when the caller expects a coherent view.

v2:
- fix using the proper logical && instead of a bitwise & (Jani, Mika)
- limit the workaround to A stepping, on later steppings this HW issue is fixed
v3:
- use a separate get_seqno/set_seqno vfunc (Chris)

Quoted 2 lines

Testcase: igt/store_dword_loop_render

I'm not going to quibble about the whitespace or the code duplication that isn't your fault...

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#6 lore raw

PATCH v2 2/2 drm/i915/bxt: work around HW coherency issue for cached GEM mappings

In reply to Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached GPU mappings, so fall back to uncached mappings. Note that this still won't fix cases where userspace expects a coherent view without synchronizing (via a set domain call). It still makes sense to limit the kernel's notion of the mapping to be uncached, for example for relocations to work properly during execbuffer time. Also in case user space does synchronize the buffer, this will still guarantee that we'll do the proper clflushing for the buffer.

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed

Testcast: igt/gem_store_dword_batches_loop/cached-mapping

---
 drivers/gpu/drm/i915/i915_gem.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_gem.c +10 -1
index 407b6b3..987ffa8 100644
@@ -3742,7 +3742,16 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,
 		level = I915_CACHE_NONE;
 		break;
 	case I915_CACHING_CACHED:
-		level = I915_CACHE_LLC;
+		/*
+		 * Due to a HW issue on BXT A stepping, GPU stores via a
+		 * snooped mapping may leave stale data in a corresponding CPU
+		 * cacheline, whereas normally such cachelines would get
+		 * invalidated. As a workaround assume that these stores are
+		 * not coherent, which means we'll flush the CPU cache manually
+		 * whenever doing a CPU/GPU sync operation.
+		 */
+		level = IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0 ?
+			I915_CACHE_NONE : I915_CACHE_LLC;
 		break;
 	case I915_CACHING_DISPLAY:
 		level = HAS_WT(dev) ? I915_CACHE_WT : I915_CACHE_NONE;
Signature · 6 lines
2.1.4

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#7 lore raw

Re: [PATCH v2 2/2] drm/i915/bxt: work around HW coherency issue for cached GEM mappings

Quoted 11 lines · Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached GPU mappings, so fall back to uncached mappings. Note that this still won't fix cases where userspace expects a coherent view without synchronizing (via a set domain call). It still makes sense to limit the kernel's notion of the mapping to be uncached, for example for relocations to work properly during execbuffer time. Also in case user space does synchronize the buffer, this will still guarantee that we'll do the proper clflushing for the buffer.

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed

This has to report the failure, ENODEV otherwise userspace will be terribly confused (it will try to CPU coherent access assuming it will be fast, when it is better to use alternative paths). -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#8 lore raw

Re: [PATCH v2 2/2] drm/i915/bxt: work around HW coherency issue for cached GEM mappings

Quoted 10 lines · Chris Wilson
Quoted 9 lines · Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached GPU mappings, so fall back to uncached mappings. Note that this still won't fix cases where userspace expects a coherent view without synchronizing (via a set domain call). It still makes sense to limit the kernel's notion of the mapping to be uncached, for example for relocations to work properly during execbuffer time. Also in case user space does synchronize the buffer, this will still guarantee that we'll do the proper clflushing for the buffer.

Quoted 6 lines
Quoted 2 lines

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed

This has to report the failure, ENODEV otherwise userspace will be terribly confused (it will try to CPU coherent access assuming it will be fast, when it is better to use alternative paths).

Ok, I was not sure how existing user space would handle the failure, but if it has the fall-back logic then ENODEV is the better solution. Will change this.

Quoted 2 lines

-Chris

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

author
#9 lore raw

PATCH v3 2/2 drm/i915/bxt: don't allow cached GEM mappings on A stepping

In reply to Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached (CPU snooped) GPU mappings, so fail such requests. User space is supposed to fall back to uncached mappings in this case.

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed v3:
- return error instead of trying to work around the issue in kernel, since that could confuse user space (Chris)

Testcast: igt/gem_store_dword_batches_loop/cached-mapping

---
 drivers/gpu/drm/i915/i915_gem.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +9 -0
index 407b6b3..95fd4ff 100644
@@ -3742,6 +3742,15 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,
 		level = I915_CACHE_NONE;
 		break;
 	case I915_CACHING_CACHED:
+		/*
+		 * Due to a HW issue on BXT A stepping, GPU stores via a
+		 * snooped mapping may leave stale data in a corresponding CPU
+		 * cacheline, whereas normally such cachelines would get
+		 * invalidated.
+		 */
+		if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0)
+			return -ENODEV;
+
 		level = I915_CACHE_LLC;
 		break;
 	case I915_CACHING_DISPLAY:
Signature · 6 lines
2.1.4

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Reviewed-by on v3 2/2
#10 lore raw

Re: [PATCH v3 2/2] drm/i915/bxt: don't allow cached GEM mappings on A stepping

Quoted 5 lines · Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached (CPU snooped) GPU mappings, so fail such requests. User space is supposed to fall back to uncached mappings in this case.

Quoted 8 lines

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed v3:
- return error instead of trying to work around the issue in kernel, since that could confuse user space (Chris)

Testcast: igt/gem_store_dword_batches_loop/cached-mapping

I checked in the two userspaces where I have used snooping that both handle an error correctly, so

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#11 lore raw

Re: [PATCH v3 2/2] drm/i915/bxt: don't allow cached GEM mappings on A stepping

Quoted 12 lines · Chris Wilson
Quoted 11 lines · Imre Deak

Due to a coherency issue on BXT A steppings we can't guarantee a coherent view of cached (CPU snooped) GPU mappings, so fail such requests. User space is supposed to fall back to uncached mappings in this case.

v2:
- limit the WA to A steppings, on later stepping this HW issue is fixed v3:
- return error instead of trying to work around the issue in kernel, since that could confuse user space (Chris)

Quoted 7 lines
Quoted 2 lines

Testcast: igt/gem_store_dword_batches_loop/cached-mapping

I checked in the two userspaces where I have used snooping that both handle an error correctly, so

Both patches applied, thanks.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Tested-by on v2 2/2
#12 lore raw

Re: [PATCH v2 2/2] drm/i915/bxt: work around HW coherency issue for cached GEM mappings

In reply to Imre Deak

Task id: 7201 -------------------------------------Summary-------------------------------------
Platform Delta drm-intel-nightly Series Applied
ILK -1 302/302 301/302

SNB                                  315/315              315/315
IVB                                  336/336              336/336
BYT                                  283/283              283/283
HSW                                  378/378              378/378

-------------------------------------Detailed-------------------------------------
Platform Test drm-intel-nightly Series Applied
*ILK igt@kms_flip@flip-vs-dpms-interruptible PASS(1) DMESG_WARN(1)
Note: You need to pay more attention to line start with '*'
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#13 lore raw
In reply to Imre Deak
Quoted 3 lines · Imre Deak

This is a v2 of [1]. Since v1 the HW team confirmed that there is an HW issue in A steppings with the GPU/CPU snoop logic, which explains why we need this workaround.

I've been testing this extensively, and I do believe we can use clflush for all gen - it is a measurable improvement over using the heavyweight/locked read of ACTHD, and none of the coherency tests have thrown any warnings (over a period of a couple of months now). -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#14 lore raw
Quoted 9 lines · Chris Wilson
Quoted 3 lines · Imre Deak

This is a v2 of [1]. Since v1 the HW team confirmed that there is an HW issue in A steppings with the GPU/CPU snoop logic, which explains why we need this workaround.

I've been testing this extensively, and I do believe we can use clflush for all gen - it is a measurable improvement over using the heavyweight/locked read of ACTHD, and none of the coherency tests have thrown any warnings (over a period of a couple of months now).

Ok. Do you mean only places where there is already the ACTHD w/a, or to extend it also to other platforms?

Imo doing this as a follow-up to this patchset would make sense, to keep platform specific changes separate.

--Imre

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#15 lore raw
Quoted 6 lines · Imre Deak
Quoted 5 lines · Chris Wilson
Quoted 3 lines · Imre Deak

This is a v2 of [1]. Since v1 the HW team confirmed that there is an HW issue in A steppings with the GPU/CPU snoop logic, which explains why we need this workaround.

Quoted 7 lines
Quoted 4 lines

I've been testing this extensively, and I do believe we can use clflush for all gen - it is a measurable improvement over using the heavyweight/locked read of ACTHD, and none of the coherency tests have thrown any warnings (over a period of a couple of months now).

Ok. Do you mean only places where there is already the ACTHD w/a, or to extend it also to other platforms?

I mean replace the current ACTHD w/a.

Quoted 2 lines

Imo doing this as a follow-up to this patchset would make sense, to keep platform specific changes separate.

Agreed. Get the w/a in place, then do the conversion so that we have a safe fallback.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx