[PATCH] drm: Schedule the output_poll_work with 1s delay if we have delayed event

[PATCH] drm: Schedule the output_poll_work with 1s delay if we have delayed event

[Peter Ujfalusi]

Instead of scheduling the work to handle the initial delayed event, use 1s
delay.

When the delayed event is handled w/o delay - in a similar matter when the
poll had been initialized before drm_helper_probe_single_connector_modes()
is called - it triggers a race in Optimus setups.

Fixes: 339fd36238dd ("drm: drm_probe_helper: Fix output_poll_work scheduling")
Cc: stable@vger.kernel.org   # v4.9
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
---
Hi,

related bug report: https://bugs.freedesktop.org/show_bug.cgi?id=98690

Regards,
Peter

 drivers/gpu/drm/drm_probe_helper.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
index 98ed110e28ed..f30c14b0a72f 100644
--- a/drivers/gpu/drm/drm_probe_helper.c
+++ b/drivers/gpu/drm/drm_probe_helper.c
@@ -146,8 +146,9 @@ void drm_kms_helper_poll_enable_locked(struct drm_device *dev)
 	drm_connector_list_iter_put(&conn_iter);
 
 	if (dev->mode_config.delayed_event) {
+		/* Use short (1s) delay to handle the initial delayed event */
 		poll = true;
-		delay = 0;
+		delay = HZ;
 	}
 
 	if (poll)
-- 
2.11.0

Re: [PATCH] drm: Schedule the output_poll_work with 1s delay if we have delayed event

[Daniel Vetter]

On Wed, Jan 04, 2017 at 02:00:53PM +0200, Peter Ujfalusi wrote:
> Instead of scheduling the work to handle the initial delayed event, use 1s
> delay.
> 
> When the delayed event is handled w/o delay - in a similar matter when the
> poll had been initialized before drm_helper_probe_single_connector_modes()
> is called - it triggers a race in Optimus setups.
> 
> Fixes: 339fd36238dd ("drm: drm_probe_helper: Fix output_poll_work scheduling")
> Cc: stable@vger.kernel.org   # v4.9
> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
> ---
> Hi,
> 
> related bug report: https://bugs.freedesktop.org/show_bug.cgi?id=98690
> 
> Regards,
> Peter
> 
>  drivers/gpu/drm/drm_probe_helper.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
> index 98ed110e28ed..f30c14b0a72f 100644
> --- a/drivers/gpu/drm/drm_probe_helper.c
> +++ b/drivers/gpu/drm/drm_probe_helper.c
> @@ -146,8 +146,9 @@ void drm_kms_helper_poll_enable_locked(struct drm_device *dev)
>  	drm_connector_list_iter_put(&conn_iter);
>  
>  	if (dev->mode_config.delayed_event) {
> +		/* Use short (1s) delay to handle the initial delayed event */
>  		poll = true;
> -		delay = 0;
> +		delay = HZ;

This smells a bit like duct-tape papering over some other race. Lack of
locking or something else. Most likely some drivers enable polling a bit
too early in their load sequence. And if we can't figure this out, then we
need some really big FIXME: here that this papers over driver races.
-Daniel

>  	}
>  
>  	if (poll)
> -- 
> 2.11.0
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Re: [PATCH] drm: Schedule the output_poll_work with 1s delay if we have delayed event

[Peter Ujfalusi]

On 01/05/2017 10:43 AM, Daniel Vetter wrote:
> On Wed, Jan 04, 2017 at 02:00:53PM +0200, Peter Ujfalusi wrote:
>> Instead of scheduling the work to handle the initial delayed event, use 1s
>> delay.
>>
>> When the delayed event is handled w/o delay - in a similar matter when the
>> poll had been initialized before drm_helper_probe_single_connector_modes()
>> is called - it triggers a race in Optimus setups.
>>
>> Fixes: 339fd36238dd ("drm: drm_probe_helper: Fix output_poll_work scheduling")
>> Cc: stable@vger.kernel.org   # v4.9
>> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
>> ---
>> Hi,
>>
>> related bug report: https://bugs.freedesktop.org/show_bug.cgi?id=98690
>>
>> Regards,
>> Peter
>>
>>  drivers/gpu/drm/drm_probe_helper.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
>> index 98ed110e28ed..f30c14b0a72f 100644
>> --- a/drivers/gpu/drm/drm_probe_helper.c
>> +++ b/drivers/gpu/drm/drm_probe_helper.c
>> @@ -146,8 +146,9 @@ void drm_kms_helper_poll_enable_locked(struct drm_device *dev)
>>  	drm_connector_list_iter_put(&conn_iter);
>>  
>>  	if (dev->mode_config.delayed_event) {
>> +		/* Use short (1s) delay to handle the initial delayed event */
>>  		poll = true;
>> -		delay = 0;
>> +		delay = HZ;
> 
> This smells a bit like duct-tape papering over some other race.

Yes, I agree. We could revert 339fd36238dd also to put back the duct-tape.

> Lack of
> locking or something else. Most likely some drivers enable polling a bit
> too early in their load sequence.

All is pointing to Optimus. Intel and nouveau alone works, but we have
failure on Optimus laptops.
I have tried to narrow it down with another patch attached to the
bugzilla: https://bugs.freedesktop.org/attachment.cgi?id=128742

I expected it to fix the problem. But it did not. So I'm puzzled.

>And if we can't figure this out, then we
> need some really big FIXME: here that this papers over driver races.

Yeah, now we know that there is something wrong in some driver(s). We
can hide it with delaying the poll_work. It would be great if
nouveau/optimus guys would be able to take a look at this.

> -Daniel
> 
>>  	}
>>  
>>  	if (poll)
>> -- 
>> 2.11.0
>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 

-- 
Péter

Re: [PATCH] drm: Schedule the output_poll_work with 1s delay if we have delayed event

[Daniel Vetter]

On Thu, Jan 05, 2017 at 12:40:29PM +0200, Peter Ujfalusi wrote:
> On 01/05/2017 10:43 AM, Daniel Vetter wrote:
> > On Wed, Jan 04, 2017 at 02:00:53PM +0200, Peter Ujfalusi wrote:
> >> Instead of scheduling the work to handle the initial delayed event, use 1s
> >> delay.
> >>
> >> When the delayed event is handled w/o delay - in a similar matter when the
> >> poll had been initialized before drm_helper_probe_single_connector_modes()
> >> is called - it triggers a race in Optimus setups.
> >>
> >> Fixes: 339fd36238dd ("drm: drm_probe_helper: Fix output_poll_work scheduling")
> >> Cc: stable@vger.kernel.org   # v4.9
> >> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
> >> ---
> >> Hi,
> >>
> >> related bug report: https://bugs.freedesktop.org/show_bug.cgi?id=98690
> >>
> >> Regards,
> >> Peter
> >>
> >>  drivers/gpu/drm/drm_probe_helper.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
> >> index 98ed110e28ed..f30c14b0a72f 100644
> >> --- a/drivers/gpu/drm/drm_probe_helper.c
> >> +++ b/drivers/gpu/drm/drm_probe_helper.c
> >> @@ -146,8 +146,9 @@ void drm_kms_helper_poll_enable_locked(struct drm_device *dev)
> >>  	drm_connector_list_iter_put(&conn_iter);
> >>  
> >>  	if (dev->mode_config.delayed_event) {
> >> +		/* Use short (1s) delay to handle the initial delayed event */
> >>  		poll = true;
> >> -		delay = 0;
> >> +		delay = HZ;
> > 
> > This smells a bit like duct-tape papering over some other race.
> 
> Yes, I agree. We could revert 339fd36238dd also to put back the duct-tape.

Since the revert and this patch are the same kind of duct-tape (massive
delay) I prefer something like this patch ...

> 
> > Lack of
> > locking or something else. Most likely some drivers enable polling a bit
> > too early in their load sequence.
> 
> All is pointing to Optimus. Intel and nouveau alone works, but we have
> failure on Optimus laptops.
> I have tried to narrow it down with another patch attached to the
> bugzilla: https://bugs.freedesktop.org/attachment.cgi?id=128742
> 
> I expected it to fix the problem. But it did not. So I'm puzzled.
> 
> >And if we can't figure this out, then we
> > need some really big FIXME: here that this papers over driver races.
> 
> Yeah, now we know that there is something wrong in some driver(s). We
> can hide it with delaying the poll_work. It would be great if
> nouveau/optimus guys would be able to take a look at this.

... but with all the above information (optimus only, what blows up)
condensed into both the commit message and a big WARNING comment in the
code. Then I think this is reasonable to merge (but still a decent wtf),
since it addresses a rather serious regression.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch