Skip to contentSkip to conversation

drm/i915: Restore GT performance in headless mode with DMC loaded

PATCH Superseded Tvrtko Ursulin 6 participants 9 messages May 5, 2017 – May 8, 2017

Subsystems (2)

v1 of 6 Superseded Latest is v6 →

lore
Share & tools
lore

Tvrtko Ursulin

author maintainer
#0 lore raw

PATCH 1/1 drm/i915: Restore GT performance in headless mode with DMC loaded

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

This frantic activity on DC states has a terrible impact on the performance of the overall chip with huge latencies observed in the interrupt handlers and elsewhere. Simple tests like igt/gem_latency -n 0 are slowed down by a factor of eight.

Work around it by grabbing a modeset display power domain whilst there is any GT activity. This seems to be effective in making the DMC keep its paws off the chip.

On the other hand this may have a negative impact on the overall power budget of the chip and so could still affect performance.

This version limits the workaround got SKL GT3 and GT4 parts but this is just due the absence of testing on other platforms. It is possible we will have to apply it wider.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100572
Testcase: igt/gem_exec_nop/headless
Cc: Imre Deak <[email protected]>

---
 drivers/gpu/drm/i915/i915_drv.h         | 5 +++++
 drivers/gpu/drm/i915/i915_gem.c         | 4 ++++
 drivers/gpu/drm/i915/i915_gem_request.c | 3 +++
 3 files changed, 12 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +5 -0
index 320c16df1c9c..4d58e2e28c2f 100644
@@ -2990,6 +2990,11 @@ intel_info(const struct drm_i915_private *dev_priv)
 
 #define HAS_DECOUPLED_MMIO(dev_priv) (INTEL_INFO(dev_priv)->has_decoupled_mmio)
 
+#define NEEDS_CSR_GT_PERF_WA(dev_priv) \
+	HAS_CSR(dev_priv) && \
+	(IS_SKL_GT3(dev_priv) || IS_SKL_GT4(dev_priv)) && \
+	(dev_priv)->csr.dmc_payload
+
 #include "i915_trace.h"
 
 static inline bool intel_scanout_needs_vtd_wa(struct drm_i915_private *dev_priv)
drivers/gpu/drm/i915/i915_gem.c +4 -0
index b2727905ef2b..c52d863f409c 100644
@@ -3200,7 +3200,11 @@ i915_gem_idle_work_handler(struct work_struct *work)
 
 	if (INTEL_GEN(dev_priv) >= 6)
 		gen6_rps_idle(dev_priv);
+
 	intel_runtime_pm_put(dev_priv);
+
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET);
 out_unlock:
 	mutex_unlock(&dev->struct_mutex);
 
drivers/gpu/drm/i915/i915_gem_request.c +3 -0
index 10361c7e3b37..10a3b51f6362 100644
@@ -873,6 +873,9 @@ static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
 
 	GEM_BUG_ON(!dev_priv->gt.active_requests);
 
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
+
 	intel_runtime_pm_get_noresume(dev_priv);
 	dev_priv->gt.awake = true;
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Acked-by on v1 1/1
#1 lore raw
Quoted 11 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

This frantic activity on DC states has a terrible impact on the performance of the overall chip with huge latencies observed in the interrupt handlers and elsewhere. Simple tests like igt/gem_latency -n 0 are slowed down by a factor of eight.

Quoted 6 lines

Work around it by grabbing a modeset display power domain whilst there is any GT activity. This seems to be effective in making the DMC keep its paws off the chip.

On the other hand this may have a negative impact on the overall power budget of the chip and so could still affect performance.

Please add this as a comment to the code, I think in mark_busy(). I don't think this w/a will remain applicable forever and so merits a continual reminder and being discussed again in future.

Quoted 19 lines

This version limits the workaround got SKL GT3 and GT4 parts but this is just due the absence of testing on other platforms. It is possible we will have to apply it wider.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100572
Testcase: igt/gem_exec_nop/headless
Cc: Imre Deak <[email protected]>

---
 drivers/gpu/drm/i915/i915_drv.h         | 5 +++++
 drivers/gpu/drm/i915/i915_gem.c         | 4 ++++
 drivers/gpu/drm/i915/i915_gem_request.c | 3 +++
 3 files changed, 12 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +0 -0
index 320c16df1c9c..4d58e2e28c2f 100644

drivers/gpu/drm/i915/i915_drv.h intel_info(const struct drm_i915_private *dev_priv) · line 2996

Quoted 7 lines
drivers/gpu/drm/i915/i915_drv.h +4 -0
@@ -2990,6 +2990,11 @@ intel_info(const struct drm_i915_private *dev_priv)
 
 #define HAS_DECOUPLED_MMIO(dev_priv) (INTEL_INFO(dev_priv)->has_decoupled_mmio)
 
+#define NEEDS_CSR_GT_PERF_WA(dev_priv) \
+	HAS_CSR(dev_priv) && \
+	(IS_SKL_GT3(dev_priv) || IS_SKL_GT4(dev_priv)) && \
+	(dev_priv)->csr.dmc_payload

csr.dmc_payload is a bit of a surprise, but looks correct.

-Chris

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

✓ Fi.CI.BAT: success for drm/i915: Restore GT performance in headless mode with DMC loaded

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915: Restore GT performance in headless mode with DMC loaded
URL : https://patchwork.freedesktop.org/series/24017/
State : success

== Summary ==

Series 24017v1 drm/i915: Restore GT performance in headless mode with DMC loaded
https://patchwork.freedesktop.org/api/1.0/series/24017/revisions/1/mbox/

fi-bdw-5557u     total:278  pass:267  dwarn:0   dfail:0   fail:0   skip:11  time:429s
fi-bdw-gvtdvm    total:278  pass:256  dwarn:8   dfail:0   fail:0   skip:14  time:433s
fi-bsw-n3050     total:278  pass:242  dwarn:0   dfail:0   fail:0   skip:36  time:577s
fi-bxt-j4205     total:278  pass:259  dwarn:0   dfail:0   fail:0   skip:19  time:509s
fi-bxt-t5700     total:278  pass:258  dwarn:0   dfail:0   fail:0   skip:20  time:548s
fi-byt-j1900     total:278  pass:254  dwarn:0   dfail:0   fail:0   skip:24  time:490s
fi-byt-n2820     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:485s
fi-hsw-4770      total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:409s
fi-hsw-4770r     total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:408s
fi-ilk-650       total:278  pass:228  dwarn:0   dfail:0   fail:0   skip:50  time:416s
fi-ivb-3520m     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:495s
fi-ivb-3770      total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:471s
fi-kbl-7500u     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:458s
fi-kbl-7560u     total:278  pass:267  dwarn:1   dfail:0   fail:0   skip:10  time:566s
fi-skl-6260u     total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:458s
fi-skl-6700hq    total:278  pass:261  dwarn:0   dfail:0   fail:0   skip:17  time:569s
fi-skl-6700k     total:278  pass:256  dwarn:4   dfail:0   fail:0   skip:18  time:460s
fi-skl-6770hq    total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:489s
fi-skl-gvtdvm    total:278  pass:265  dwarn:0   dfail:0   fail:0   skip:13  time:431s
fi-snb-2520m     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:531s
fi-snb-2600      total:278  pass:249  dwarn:0   dfail:0   fail:0   skip:29  time:408s

369880c1680bf9bde467a40d2a03d3ad32341281 drm-tip: 2017y-05m-04d-15h-00m-33s UTC integration manifest e0b135e drm/i915: Restore GT performance in headless mode with DMC loaded

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_4630/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw
In reply to Tvrtko Ursulin
Quoted 5 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

Is it trapping on some interrupt register accesses or what? And if so which registers are affected?

Quoted 80 lines

This frantic activity on DC states has a terrible impact on the performance of the overall chip with huge latencies observed in the interrupt handlers and elsewhere. Simple tests like igt/gem_latency -n 0 are slowed down by a factor of eight.

Work around it by grabbing a modeset display power domain whilst there is any GT activity. This seems to be effective in making the DMC keep its paws off the chip.

On the other hand this may have a negative impact on the overall power budget of the chip and so could still affect performance.

This version limits the workaround got SKL GT3 and GT4 parts but this is just due the absence of testing on other platforms. It is possible we will have to apply it wider.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100572
Testcase: igt/gem_exec_nop/headless
Cc: Imre Deak <[email protected]>

---
 drivers/gpu/drm/i915/i915_drv.h         | 5 +++++
 drivers/gpu/drm/i915/i915_gem.c         | 4 ++++
 drivers/gpu/drm/i915/i915_gem_request.c | 3 +++
 3 files changed, 12 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +5 -0
index 320c16df1c9c..4d58e2e28c2f 100644
@@ -2990,6 +2990,11 @@ intel_info(const struct drm_i915_private *dev_priv)
 
 #define HAS_DECOUPLED_MMIO(dev_priv) (INTEL_INFO(dev_priv)->has_decoupled_mmio)
 
+#define NEEDS_CSR_GT_PERF_WA(dev_priv) \
+	HAS_CSR(dev_priv) && \
+	(IS_SKL_GT3(dev_priv) || IS_SKL_GT4(dev_priv)) && \
+	(dev_priv)->csr.dmc_payload
+
 #include "i915_trace.h"
 
 static inline bool intel_scanout_needs_vtd_wa(struct drm_i915_private *dev_priv)
drivers/gpu/drm/i915/i915_gem.c +4 -0
index b2727905ef2b..c52d863f409c 100644
@@ -3200,7 +3200,11 @@ i915_gem_idle_work_handler(struct work_struct *work)
 
 	if (INTEL_GEN(dev_priv) >= 6)
 		gen6_rps_idle(dev_priv);
+
 	intel_runtime_pm_put(dev_priv);
+
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET);
 out_unlock:
 	mutex_unlock(&dev->struct_mutex);
 
drivers/gpu/drm/i915/i915_gem_request.c +3 -0
index 10361c7e3b37..10a3b51f6362 100644
@@ -873,6 +873,9 @@ static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
 
 	GEM_BUG_ON(!dev_priv->gt.active_requests);
 
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
+
 	intel_runtime_pm_get_noresume(dev_priv);
 	dev_priv->gt.awake = true;
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#4 lore raw
Quoted 9 lines · Ville Syrjälä
Quoted 5 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

Is it trapping on some interrupt register accesses or what? And if so which registers are affected?

It looks like GT IIR or something along those lines it but I couldn't say with total confidence. It is just a guess. Firmware binary definitely "mentions" those registers as can be seen by inspecting it with a hex editor.

The data I collected at least seems to present a correlation between the batch frequency and DC state transition frequency:

tgt DC irqs irqs/s irq batch/s DC/s DC/batch
submit transitions /
freq batch
========================================================================
10000 20000 78300 7830.00 1.96 4000.00 2000.00 0.50
9901 14000 52855 7550.71 1.32 5714.29 2000.00 0.35
9524 13500 49100 7328.36 1.23 5970.15 2014.93 0.34
9091 13500 49200 7235.29 1.23 5882.35 1985.29 0.34
5000 16900 33290 3916.47 0.83 4705.88 1988.24 0.42
3333 27800 69550 4932.62 1.74 2836.88 1971.63 0.70
1667 57200 80200 2655.63 2.01 1324.50 1894.04 1.43
909 80000 80034 1482.11 2.00 740.74 1481.48 2.00
476 87000 80039 820.91 2.00 410.26 892.31 2.18
196 160000 80055 334.40 2.00 167.08 668.34 4.00

Submitted batches were ~100us long in all cases. So with low batch frequency it looks pretty believable. For example when we have 167.08 batches/s, we have 334.40 irq/s - which is double - as expected for execlists. And we get again double that in terms of DC transitions per second. Each irq is one GT IIR write from the GPU side, and another from the CPU side.

This was actually Imre's suggestion btw. Before I was only looking at the irq/s to DC/s correlation which was confusing me a lot since there are more of the latter, so I thought it can't be the trigger. But once Imre mentioned the possibility that things are triggered by IIR register writes numbers started making more sense.

Then with higher batch frequencies the ratio starts falling, is it because DC transitions are too slow to keep up, or something else I am not sure.

Regards,

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

#5 lore raw
Quoted 9 lines · Tvrtko Ursulin
Quoted 7 lines · Ville Syrjälä
Quoted 5 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

Quoted 5 lines
Quoted 2 lines

Is it trapping on some interrupt register accesses or what? And if so which registers are affected?

It looks like GT IIR or something along those lines it but I couldn't say with total confidence.

<read DC counters>
for i in seq 1 100 ; do IGT_NO_FORCEWAKE=1 intel_reg read <whatever> ; done <read DC counters>

Should be a pretty trivial to run that against the suspect registers.

Quoted 22 lines

It is just a guess. Firmware binary
definitely "mentions" those registers as can be seen by inspecting it with a hex editor.

The data I collected at least seems to present a correlation between the batch frequency and DC state transition frequency:

tgt DC irqs irqs/s irq batch/s DC/s DC/batch
submit transitions /
freq batch
========================================================================
10000 20000 78300 7830.00 1.96 4000.00 2000.00 0.50
9901 14000 52855 7550.71 1.32 5714.29 2000.00 0.35
9524 13500 49100 7328.36 1.23 5970.15 2014.93 0.34
9091 13500 49200 7235.29 1.23 5882.35 1985.29 0.34
5000 16900 33290 3916.47 0.83 4705.88 1988.24 0.42
3333 27800 69550 4932.62 1.74 2836.88 1971.63 0.70
1667 57200 80200 2655.63 2.01 1324.50 1894.04 1.43
909 80000 80034 1482.11 2.00 740.74 1481.48 2.00
476 87000 80039 820.91 2.00 410.26 892.31 2.18
196 160000 80055 334.40 2.00 167.08 668.34 4.00

Quoted 6 lines

Submitted batches were ~100us long in all cases. So with low batch frequency it looks pretty believable. For example when we have 167.08 batches/s, we have 334.40 irq/s - which is double - as expected for execlists. And we get again double that in terms of DC transitions per second. Each irq is one GT IIR write from the GPU side, and another from the CPU side.

GPU doesn't actually write the IIRs. It's just latching stuff from the ISR. Whether the ISR edge or some higher level interrupt event actually causes the DMC to kick into action isn't clear at all. My original impressions was that it just traps the register accesses.

Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#6 lore raw
In reply to Tvrtko Ursulin
Quoted 37 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

This frantic activity on DC states has a terrible impact on the performance of the overall chip with huge latencies observed in the interrupt handlers and elsewhere. Simple tests like igt/gem_latency -n 0 are slowed down by a factor of eight.

Work around it by grabbing a modeset display power domain whilst there is any GT activity. This seems to be effective in making the DMC keep its paws off the chip.

On the other hand this may have a negative impact on the overall power budget of the chip and so could still affect performance.

This version limits the workaround got SKL GT3 and GT4 parts but this is just due the absence of testing on other platforms. It is possible we will have to apply it wider.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100572
Testcase: igt/gem_exec_nop/headless
Cc: Imre Deak <[email protected]>

---
 drivers/gpu/drm/i915/i915_drv.h         | 5 +++++
 drivers/gpu/drm/i915/i915_gem.c         | 4 ++++
 drivers/gpu/drm/i915/i915_gem_request.c | 3 +++
 3 files changed, 12 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +0 -0
index 320c16df1c9c..4d58e2e28c2f 100644

drivers/gpu/drm/i915/i915_drv.h intel_info(const struct drm_i915_private *dev_priv) · line 2996

Quoted 7 lines
drivers/gpu/drm/i915/i915_drv.h +4 -0
@@ -2990,6 +2990,11 @@ intel_info(const struct drm_i915_private *dev_priv)
 
 #define HAS_DECOUPLED_MMIO(dev_priv) (INTEL_INFO(dev_priv)->has_decoupled_mmio)
 
+#define NEEDS_CSR_GT_PERF_WA(dev_priv) \
+	HAS_CSR(dev_priv) && \
+	(IS_SKL_GT3(dev_priv) || IS_SKL_GT4(dev_priv)) && \
+	(dev_priv)->csr.dmc_payload

Nitpick, the whole thing could use braces around it for consistency, although I don't see any sane use of the macro causing precedence surprises.

BR,
Jani.

Quoted 33 lines
drivers/gpu/drm/i915/i915_drv.h +1 -0
+
 #include "i915_trace.h"
 
 static inline bool intel_scanout_needs_vtd_wa(struct drm_i915_private *dev_priv)
drivers/gpu/drm/i915/i915_gem.c +4 -0
index b2727905ef2b..c52d863f409c 100644
@@ -3200,7 +3200,11 @@ i915_gem_idle_work_handler(struct work_struct *work)
 
 	if (INTEL_GEN(dev_priv) >= 6)
 		gen6_rps_idle(dev_priv);
+
 	intel_runtime_pm_put(dev_priv);
+
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET);
 out_unlock:
 	mutex_unlock(&dev->struct_mutex);
 
drivers/gpu/drm/i915/i915_gem_request.c +3 -0
index 10361c7e3b37..10a3b51f6362 100644
@@ -873,6 +873,9 @@ static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
 
 	GEM_BUG_ON(!dev_priv->gt.active_requests);
 
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
+
 	intel_runtime_pm_get_noresume(dev_priv);
 	dev_priv->gt.awake = true;
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author maintainer
#7 lore raw

PATCH v2 1/1 drm/i915: Restore GT performance in headless mode with DMC loaded

In reply to Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

It seems that the DMC likes to transition between the DC states a lot when there are no connected displays (no active power domains) during simple command submission.

This frantic activity on DC states has a terrible impact on the performance of the overall chip with huge latencies observed in the interrupt handlers and elsewhere. Simple tests like igt/gem_latency -n 0 are slowed down by a factor of eight.

Work around it by grabbing a modeset display power domain whilst there is any GT activity. This seems to be effective in making the DMC keep its paws off the chip.

On the other hand this may have a negative impact on the overall power budget of the chip and so could still affect performance.

This version limits the workaround got SKL GT3 and GT4 parts but this is just due the absence of testing on other platforms. It is possible we will have to apply it wider.

v2:
* Add commit text as comment in i915_gem_mark_busy. (Chris Wilson)
* Protect macro body with braces. (Jani Nikula)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100572
Testcase: igt/gem_exec_nop/headless

---
 drivers/gpu/drm/i915/i915_drv.h         |  5 +++++
 drivers/gpu/drm/i915/i915_gem.c         |  4 ++++
 drivers/gpu/drm/i915/i915_gem_request.c | 17 +++++++++++++++++
 3 files changed, 26 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +5 -0
index 320c16df1c9c..509b398d054d 100644
@@ -2990,6 +2990,11 @@ intel_info(const struct drm_i915_private *dev_priv)
 
 #define HAS_DECOUPLED_MMIO(dev_priv) (INTEL_INFO(dev_priv)->has_decoupled_mmio)
 
+#define NEEDS_CSR_GT_PERF_WA(dev_priv) \
+	(HAS_CSR(dev_priv) && \
+	(IS_SKL_GT3(dev_priv) || IS_SKL_GT4(dev_priv)) && \
+	(dev_priv)->csr.dmc_payload)
+
 #include "i915_trace.h"
 
 static inline bool intel_scanout_needs_vtd_wa(struct drm_i915_private *dev_priv)
drivers/gpu/drm/i915/i915_gem.c +4 -0
index b2727905ef2b..c52d863f409c 100644
@@ -3200,7 +3200,11 @@ i915_gem_idle_work_handler(struct work_struct *work)
 
 	if (INTEL_GEN(dev_priv) >= 6)
 		gen6_rps_idle(dev_priv);
+
 	intel_runtime_pm_put(dev_priv);
+
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET);
 out_unlock:
 	mutex_unlock(&dev->struct_mutex);
 
drivers/gpu/drm/i915/i915_gem_request.c +17 -0
index 10361c7e3b37..c47c983b874e 100644
@@ -873,6 +873,23 @@ static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
 
 	GEM_BUG_ON(!dev_priv->gt.active_requests);
 
+	/*
+	 * It seems that the DMC likes to transition between the DC states
+	 * a lot when there are no connected displays (no active power
+	 * domains) during simple command submission.
+	 *
+	 * This frantic activity on DC states has a terrible impact on the
+	 * performance of the overall chip with huge latencies observed in
+	 * the interrupt handlers and elsewhere. Simple tests like
+	 * igt/gem_latency -n 0 are slowed down by a factor of eight.
+	 *
+	 * Work around it by grabbing a modeset display power domain whilst
+	 * there is any GT activity. This seems to be effective in making
+	 * the DMC keep its paws off the chip.
+	 */
+	if (NEEDS_CSR_GT_PERF_WA(dev_priv))
+		intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
+
 	intel_runtime_pm_get_noresume(dev_priv);
 	dev_priv->gt.awake = true;
 
Signature · 6 lines
2.9.3

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

✓ Fi.CI.BAT: success for drm/i915: Restore GT performance in headless mode with DMC loaded (rev2)

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915: Restore GT performance in headless mode with DMC loaded (rev2)
URL : https://patchwork.freedesktop.org/series/24017/
State : success

== Summary ==

Series 24017v2 drm/i915: Restore GT performance in headless mode with DMC loaded
https://patchwork.freedesktop.org/api/1.0/series/24017/revisions/2/mbox/

fi-bdw-5557u     total:278  pass:267  dwarn:0   dfail:0   fail:0   skip:11  time:432s
fi-bdw-gvtdvm    total:278  pass:256  dwarn:8   dfail:0   fail:0   skip:14  time:424s
fi-bsw-n3050     total:278  pass:242  dwarn:0   dfail:0   fail:0   skip:36  time:586s
fi-bxt-j4205     total:278  pass:259  dwarn:0   dfail:0   fail:0   skip:19  time:506s
fi-bxt-t5700     total:278  pass:258  dwarn:0   dfail:0   fail:0   skip:20  time:564s
fi-byt-j1900     total:278  pass:254  dwarn:0   dfail:0   fail:0   skip:24  time:487s
fi-byt-n2820     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:479s
fi-hsw-4770      total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:416s
fi-hsw-4770r     total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:406s
fi-ilk-650       total:278  pass:228  dwarn:0   dfail:0   fail:0   skip:50  time:416s
fi-ivb-3520m     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:498s
fi-ivb-3770      total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:466s
fi-kbl-7500u     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:456s
fi-kbl-7560u     total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:571s
fi-skl-6260u     total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:451s
fi-skl-6700hq    total:278  pass:261  dwarn:0   dfail:0   fail:0   skip:17  time:565s
fi-skl-6700k     total:278  pass:256  dwarn:4   dfail:0   fail:0   skip:18  time:455s
fi-skl-6770hq    total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:492s
fi-snb-2520m     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:528s
fi-snb-2600      total:278  pass:249  dwarn:0   dfail:0   fail:0   skip:29  time:409s

fi-skl-gvtdvm failed to collect. IGT log at Patchwork_4639/fi-skl-gvtdvm/igt.log

f3c6147e2a7879ee7face0b7b634dd26704fc3d4 drm-tip: 2017y-05m-08d-11h-23m-22s UTC integration manifest 6a61edd drm/i915: Restore GT performance in headless mode with DMC loaded

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_4639/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx