[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On Mon, 24 Oct 2016, Paul Bolle wrote:
> [Detailed post, but please give it a quick scan.]
Please file the information in the bug you filed. Please attach dmesg
(again, on the bug) with drm.debug=14 and running your patch.
BR,
Jani.
>
> On Wed, 2016-10-12 at 14:06 +0200, Paul Bolle wrote:
>> On Wed, 2016-10-12 at 14:08 +0300, Joonas Lahtinen wrote:
>> > Bisecting the offending commit between v4.8 and v4.8.1 would be a good
>> > start.
>>
>> That would be between v4.7 and v4.8. (I guess my report was
>> ambiguous.)
>>
>> That might take some time. Because bisecting always takes a long time
>> and especially since hitting this WARNING sometimes takes over an hour.
>> Anyhow, please prod me if I stay silent for too long.
>
> 0) So I've lost my courage to do a bisect when my first attempt landed
> me in v4.6-rc3. This is about for issue popping up between v4.7 and
> v4.8-rc1.
>
> 1) So I used the most reliable debugging tool that I actually
> understand: printk():
>
> diff --git a/drivers/gpu/drm/i915/intel_display.c
> b/drivers/gpu/drm/i915/intel_display.c
> index fbcfed63a76e..791de414cf1e 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -14771,10 +14771,16 @@ skl_max_scale(struct intel_crtc *intel_crtc, struct
> intel_crtc_state *crtc_state
> return DRM_PLANE_HELPER_NO_SCALING;
>
> crtc_clock = crtc_state->base.adjusted_mode.crtc_clock;
> - cdclk = to_intel_atomic_state(crtc_state->base.state)->cdclk;
> + if (WARN_ON_ONCE(!crtc_clock))
> + return DRM_PLANE_HELPER_NO_SCALING;
>
> - if (WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock))
> + cdclk = to_intel_atomic_state(crtc_state->base.state)->cdclk;
> + if (WARN_ON_ONCE(cdclk < crtc_clock)) {
> + printk(KERN_DEBUG "i915: cdclk < crtc_clock: %d < %d\n",
> cdclk, crtc_clock);
> return DRM_PLANE_HELPER_NO_SCALING;
> + }
> +
> + printk_ratelimited(KERN_DEBUG "i915: cdclk >= crtc_clock: %d >=
> %d\n", cdclk, crtc_clock);
>
> /*
> * skl max scale is lower of:
>
> 2) This taught me that crtc_clock always is 373250 on my machine. cdclk
> mostly is 45, but every now and then it briefly is 337500.
>
> 3) Now the interesting pattern is that cdclk drops to 337500 only after
> two quick calls of skl_max_scale() with cdclk set to 45, and a
> roughly 300ms pause before the third call of that function. Example:
>
> <7>[23758.501933] i915: cdclk >= crtc_clock: 45 >= 373250
> <7>[23758.515211] i915: cdclk >= crtc_clock: 45 >= 373250
> <7>[23758.869057] i915: cdclk < crtc_clock: 337500 < 373250
>
> This pattern of cdclk being 337500 after roughly 300ms is surprisingly
> stable.
>
> 4) So _perhaps_ there's some roughly 300ms timeout, somehow, somewhere,
> that sets cdclk to 337500 and triggers this issue. Ideas?
>
> To be continued,
>
>
> Paul Bolle
--
Jani Nikula, Intel Open Source Technology Center
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
[Detailed post, but please give it a quick scan.]
On Wed, 2016-10-12 at 14:06 +0200, Paul Bolle wrote:
> On Wed, 2016-10-12 at 14:08 +0300, Joonas Lahtinen wrote:
> > Bisecting the offending commit between v4.8 and v4.8.1 would be a good
> > start.
>
> That would be between v4.7 and v4.8. (I guess my report was
> ambiguous.)
>
> That might take some time. Because bisecting always takes a long time
> and especially since hitting this WARNING sometimes takes over an hour.
> Anyhow, please prod me if I stay silent for too long.
0) So I've lost my courage to do a bisect when my first attempt landed
me in v4.6-rc3. This is about for issue popping up between v4.7 and
v4.8-rc1.
1) So I used the most reliable debugging tool that I actually
understand: printk():
diff --git a/drivers/gpu/drm/i915/intel_display.c
b/drivers/gpu/drm/i915/intel_display.c
index fbcfed63a76e..791de414cf1e 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -14771,10 +14771,16 @@ skl_max_scale(struct intel_crtc *intel_crtc, struct
intel_crtc_state *crtc_state
return DRM_PLANE_HELPER_NO_SCALING;
crtc_clock = crtc_state->base.adjusted_mode.crtc_clock;
- cdclk = to_intel_atomic_state(crtc_state->base.state)->cdclk;
+ if (WARN_ON_ONCE(!crtc_clock))
+ return DRM_PLANE_HELPER_NO_SCALING;
- if (WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock))
+ cdclk = to_intel_atomic_state(crtc_state->base.state)->cdclk;
+ if (WARN_ON_ONCE(cdclk < crtc_clock)) {
+ printk(KERN_DEBUG "i915: cdclk < crtc_clock: %d < %d\n", cdclk,
crtc_clock);
return DRM_PLANE_HELPER_NO_SCALING;
+ }
+
+ printk_ratelimited(KERN_DEBUG "i915: cdclk >= crtc_clock: %d >= %d\n",
cdclk, crtc_clock);
/*
* skl max scale is lower of:
2) This taught me that crtc_clock always is 373250 on my machine. cdclk
mostly is 45, but every now and then it briefly is 337500.
3) Now the interesting pattern is that cdclk drops to 337500 only after
two quick calls of skl_max_scale() with cdclk set to 45, and a
roughly 300ms pause before the third call of that function. Example:
<7>[23758.501933] i915: cdclk >= crtc_clock: 45 >= 373250
<7>[23758.515211] i915: cdclk >= crtc_clock: 45 >= 373250
<7>[23758.869057] i915: cdclk < crtc_clock: 337500 < 373250
This pattern of cdclk being 337500 after roughly 300ms is surprisingly
stable.
4) So _perhaps_ there's some roughly 300ms timeout, somehow, somewhere,
that sets cdclk to 337500 and triggers this issue. Ideas?
To be continued,
Paul Bolle
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
[Adding Matt]
On Wed, 2016-10-12 at 14:08 +0300, Joonas Lahtinen wrote:
> Bisecting the offending commit between v4.8 and v4.8.1 would be a good
> start.
0) Why use a personal notebook when one can just post any half baked
idea to lkml?
1) I stumbled on https://bugzilla.redhat.com/show_bug.cgi?id=1361357 (s
ame hardware, rather similar trace).
2) That report was about the first Fedora (rawhide) kernel release
after this commit http://pkgs.fedoraproject.org/cgit/rpms/kernel.git/co
mmit/?h=f25=7d2c2f2d91793b5da452bee9bea4fa32051c8608Â ("Bring in
patch-series from drm-next to fix skl_update_other_pipe_wm issues").
3) That seventeen part series ended up in v4.8. The last commit of that
series is commit 5b483747a925 ("drm/i915: Remove wm_config from
dev_priv/intel_atomic_state").
4) So, with a little bit of luck, my bisect might only need to look at
those seventeen commits. Still, it will probably be next week before I
have a day or two to actually perform that bisect.
To be continued,
Paul Bolle
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On Wed, 2016-10-12 at 14:06 +0200, Paul Bolle wrote: > That might take some time. Because bisecting always takes a long time > and especially since hitting this WARNING sometimes takes over an hour. > Anyhow, please prod me if I stay silent for too long. For the record: I just had to power cycle this laptop because it got into that lovely state where it just locks without accepting any input (no, I don't have netconsole enabled). Assuming this lockup is related: this could be more urgent than I thought. Paul Bolle
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On Wed, 12 Oct 2016, Paul Bolle wrote: > On Wed, 2016-10-12 at 14:08 +0300, Joonas Lahtinen wrote: >> Bisecting the offending commit between v4.8 and v4.8.1 would be a good >> start. > > That would be between v4.7 and v4.8. (I guess my report was ambiguous.) > > That might take some time. Because bisecting always takes a long time > and especially since hitting this WARNING sometimes takes over an hour. > Anyhow, please prod me if I stay silent for too long. In the mean time, please file a bug over at [1] so we don't lose track. BR, Jani. [1] https://bugs.freedesktop.org/enter_bug.cgi?product=DRI=DRM/Intel -- Jani Nikula, Intel Open Source Technology Center
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On Wed, 2016-10-12 at 17:34 +0300, Jani Nikula wrote: > In the mean time, please file a bug over at [1] so we don't lose > track. Done: Â https://bugs.freedesktop.org/show_bug.cgi?id=98214 Paul Bolle
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On ke, 2016-10-12 at 11:56 +0200, Paul Bolle wrote: > On a laptop that tracks the latest stable release (Ie, it now runs > v4.8.1) I see this WARNING > Â Â Â WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock) > > Full trace pasted below. I never saw this WARNING before v4.8. Since > v4.8 I've had it in all (four, actually) boots. > > What am I expected to do about this WARNING? > Bisecting the offending commit between v4.8 and v4.8.1 would be a good start. Regards, Joonas -- Joonas Lahtinen Open Source Technology Center Intel Corporation
[Intel-gfx] drm/i915: WARN_ON_ONCE(!crtc_clock || cdclk < crtc_clock)
On Wed, 2016-10-12 at 14:08 +0300, Joonas Lahtinen wrote: > Bisecting the offending commit between v4.8 and v4.8.1 would be a good > start. That would be between v4.7 and v4.8. (I guess my report was ambiguous.) That might take some time. Because bisecting always takes a long time and especially since hitting this WARNING sometimes takes over an hour. Anyhow, please prod me if I stay silent for too long. Thanks, Paul Bolle
