[PATCH] drm/i915: Force vblanks around evasion when i915.watermark_test is set.

[PATCH] drm/i915: Force vblanks around evasion when i915.watermark_test is set.

[Maarten Lankhorst]

This will make it more likely that intermediary watermarks cause fifo
underruns, which is useful when writing watermark specific tests,
to make it more likely to trigger FIFO underruns in intermediary watermarks.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_params.c   | 1 +
 drivers/gpu/drm/i915/i915_params.h   | 1 +
 drivers/gpu/drm/i915/intel_display.c | 6 ++++++
 3 files changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_params.c b/drivers/gpu/drm/i915/i915_params.c
index 5e18cf9f754d..297d1cd53b15 100644
--- a/drivers/gpu/drm/i915/i915_params.c
+++ b/drivers/gpu/drm/i915/i915_params.c
@@ -45,6 +45,7 @@ struct i915_params i915 __read_mostly = {
 	.fastboot = 0,
 	.prefault_disable = 0,
 	.load_detect_test = 0,
+	.watermark_test = 0,
 	.reset = true,
 	.invert_brightness = 0,
 	.disable_display = 0,
diff --git a/drivers/gpu/drm/i915/i915_params.h b/drivers/gpu/drm/i915/i915_params.h
index 1323261a0cdd..91f976c6bac6 100644
--- a/drivers/gpu/drm/i915/i915_params.h
+++ b/drivers/gpu/drm/i915/i915_params.h
@@ -57,6 +57,7 @@ struct i915_params {
 	bool fastboot;
 	bool prefault_disable;
 	bool load_detect_test;
+	bool watermark_test;
 	bool reset;
 	bool disable_display;
 	bool verbose_state_checks;
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 8bb8d36f5cb9..9dcf304c7b1f 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -14184,6 +14184,9 @@ static void intel_begin_crtc_commit(struct drm_crtc *crtc,
 		to_intel_crtc_state(old_crtc_state);
 	bool modeset = needs_modeset(crtc->state);
 
+	if (i915.watermark_test)
+		intel_wait_for_vblank(dev, intel_crtc->pipe);
+
 	/* Perform vblank evasion around commit operation */
 	intel_pipe_update_start(intel_crtc);
 
@@ -14207,6 +14210,9 @@ static void intel_finish_crtc_commit(struct drm_crtc *crtc,
 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 
 	intel_pipe_update_end(intel_crtc, NULL);
+
+	if (i915.watermark_test)
+		intel_wait_for_vblank(crtc->dev, intel_crtc->pipe);
 }
 
 /**
-- 
2.5.5


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

Re: [PATCH] drm/i915: Force vblanks around evasion when i915.watermark_test is set.

[Daniel Vetter]

On Thu, Jun 09, 2016 at 01:36:50PM +0200, Maarten Lankhorst wrote:
> This will make it more likely that intermediary watermarks cause fifo
> underruns, which is useful when writing watermark specific tests,
> to make it more likely to trigger FIFO underruns in intermediary watermarks.

That's surprising ... with the vblank_wait added you're pretty much
guaranteed to never hit the evasion case. Note that the vblank evasion
code also just does a vblank wait if it detects a possible collision.

Given that I'd have assumed that this simply slows down atomic flips,
which is probably not what we want for testing all. Does this really help?
Any ideas why?
-Daniel

> 
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
>  drivers/gpu/drm/i915/i915_params.c   | 1 +
>  drivers/gpu/drm/i915/i915_params.h   | 1 +
>  drivers/gpu/drm/i915/intel_display.c | 6 ++++++
>  3 files changed, 8 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_params.c b/drivers/gpu/drm/i915/i915_params.c
> index 5e18cf9f754d..297d1cd53b15 100644
> --- a/drivers/gpu/drm/i915/i915_params.c
> +++ b/drivers/gpu/drm/i915/i915_params.c
> @@ -45,6 +45,7 @@ struct i915_params i915 __read_mostly = {
>  	.fastboot = 0,
>  	.prefault_disable = 0,
>  	.load_detect_test = 0,
> +	.watermark_test = 0,
>  	.reset = true,
>  	.invert_brightness = 0,
>  	.disable_display = 0,
> diff --git a/drivers/gpu/drm/i915/i915_params.h b/drivers/gpu/drm/i915/i915_params.h
> index 1323261a0cdd..91f976c6bac6 100644
> --- a/drivers/gpu/drm/i915/i915_params.h
> +++ b/drivers/gpu/drm/i915/i915_params.h
> @@ -57,6 +57,7 @@ struct i915_params {
>  	bool fastboot;
>  	bool prefault_disable;
>  	bool load_detect_test;
> +	bool watermark_test;
>  	bool reset;
>  	bool disable_display;
>  	bool verbose_state_checks;
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 8bb8d36f5cb9..9dcf304c7b1f 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -14184,6 +14184,9 @@ static void intel_begin_crtc_commit(struct drm_crtc *crtc,
>  		to_intel_crtc_state(old_crtc_state);
>  	bool modeset = needs_modeset(crtc->state);
>  
> +	if (i915.watermark_test)
> +		intel_wait_for_vblank(dev, intel_crtc->pipe);
> +
>  	/* Perform vblank evasion around commit operation */
>  	intel_pipe_update_start(intel_crtc);
>  
> @@ -14207,6 +14210,9 @@ static void intel_finish_crtc_commit(struct drm_crtc *crtc,
>  	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
>  
>  	intel_pipe_update_end(intel_crtc, NULL);
> +
> +	if (i915.watermark_test)
> +		intel_wait_for_vblank(crtc->dev, intel_crtc->pipe);
>  }
>  
>  /**
> -- 
> 2.5.5
> 
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH] drm/i915: Force vblanks around evasion when i915.watermark_test is set.

[Maarten Lankhorst]

Op 13-06-16 om 16:45 schreef Daniel Vetter:
> On Thu, Jun 09, 2016 at 01:36:50PM +0200, Maarten Lankhorst wrote:
>> This will make it more likely that intermediary watermarks cause fifo
>> underruns, which is useful when writing watermark specific tests,
>> to make it more likely to trigger FIFO underruns in intermediary watermarks.
> That's surprising ... with the vblank_wait added you're pretty much
> guaranteed to never hit the evasion case. Note that the vblank evasion
> code also just does a vblank wait if it detects a possible collision.
>
> Given that I'd have assumed that this simply slows down atomic flips,
> which is probably not what we want for testing all. Does this really help?
> Any ideas why?
This is a sanity check on the intermediate watermarks. The vblank test will ensure
that the old plane configuration doesn't cause underruns with watermarks, and that
the intermediate watermarks will work correctly with the new plane configuration too.

I think the latter vblank_wait may be killed, but the former could be really useful,
for example to find correct watervblank values in CHV.

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

Re: [PATCH] drm/i915: Force vblanks around evasion when i915.watermark_test is set.

[Daniel Vetter]

On Tue, Jun 14, 2016 at 08:27:57AM +0200, Maarten Lankhorst wrote:
> Op 13-06-16 om 16:45 schreef Daniel Vetter:
> > On Thu, Jun 09, 2016 at 01:36:50PM +0200, Maarten Lankhorst wrote:
> >> This will make it more likely that intermediary watermarks cause fifo
> >> underruns, which is useful when writing watermark specific tests,
> >> to make it more likely to trigger FIFO underruns in intermediary watermarks.
> > That's surprising ... with the vblank_wait added you're pretty much
> > guaranteed to never hit the evasion case. Note that the vblank evasion
> > code also just does a vblank wait if it detects a possible collision.
> >
> > Given that I'd have assumed that this simply slows down atomic flips,
> > which is probably not what we want for testing all. Does this really help?
> > Any ideas why?
> This is a sanity check on the intermediate watermarks. The vblank test will ensure
> that the old plane configuration doesn't cause underruns with watermarks, and that
> the intermediate watermarks will work correctly with the new plane configuration too.
> 
> I think the latter vblank_wait may be killed, but the former could be really useful,
> for example to find correct watervblank values in CHV.

Why can't we do the vblank wait in userspace? If we wait 1 vblank (or just
for the drm event of the previous commit), and then immediately issue the
next one, we will also run with intermediate watermarks for a full frame.
At least I still don't see a super-short race window for which we need to
have code in the kernel to properly provoke it ...
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx