[4/6] drm: Fix treatment of drm_vblank_offdelay in drm_vblank_on()

Commit Message

Mario Kleiner Feb. 8, 2016, 1:13 a.m. UTC

drm_vblank_offdelay can have three different types of values:

< 0 is to be always treated the same as dev->vblank_disable_immediate
= 0 is to be treated as "never disable vblanks"
> 0 is to be treated as disable immediate if kms driver wants it
    that way via dev->vblank_disable_immediate. Otherwise it is
    a disable timeout in msecs.

This got broken in Linux 3.18+ for the implementation of
drm_vblank_on. If the user specified a value of zero which should
always reenable vblank irqs in this function, a kms driver could
override the users choice by setting vblank_disable_immediate
to true. This patch fixes the regression and keeps the user in
control.

Signed-off-by: Mario Kleiner <[e-mail]>
Cc: <[e-mail]> # 3.18+
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
---
 drivers/gpu/drm/drm_irq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Daniel Vetter Feb. 9, 2016, 10:06 a.m. UTC | #1

On Mon, Feb 08, 2016 at 02:13:27AM +0100, Mario Kleiner wrote:
> drm_vblank_offdelay can have three different types of values:
> 
> < 0 is to be always treated the same as dev->vblank_disable_immediate
> = 0 is to be treated as "never disable vblanks"
> > 0 is to be treated as disable immediate if kms driver wants it
>     that way via dev->vblank_disable_immediate. Otherwise it is
>     a disable timeout in msecs.
> 
> This got broken in Linux 3.18+ for the implementation of
> drm_vblank_on. If the user specified a value of zero which should
> always reenable vblank irqs in this function, a kms driver could
> override the users choice by setting vblank_disable_immediate
> to true. This patch fixes the regression and keeps the user in
> control.
> 
> Signed-off-by: Mario Kleiner <[e-mail]>
> Cc: <[e-mail]> # 3.18+
> Cc: [e-mail]
> Cc: [e-mail]
> Cc: [e-mail]
> Cc: [e-mail]
> Cc: [e-mail]
> Cc: [e-mail]
> Cc: [e-mail]
> ---
>  drivers/gpu/drm/drm_irq.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
> index 5c27ad3..fb17c45 100644
> --- a/drivers/gpu/drm/drm_irq.c
> +++ b/drivers/gpu/drm/drm_irq.c
> @@ -1492,8 +1492,8 @@ void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
>  	 * re-enable interrupts if there are users left, or the
>  	 * user wishes vblank interrupts to be enabled all the time.
>  	 */
> -	if (atomic_read(&vblank->refcount) != 0 ||
> -	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
> +	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
> +	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))

Hm, shouldn't we change this to only enable the vblank irq if we need it,
i.e. offdelay == 0? For delayed disabling there's kinda no need to enable
it superflously after a modeset, if userspace didn't yet ask for vblank
timestamps. But then is was specifically added by Ville in cd19e52aee922,
so I guess someone really wants this.

Reviewed-by: Daniel Vetter <[e-mail]>

>  		WARN_ON(drm_vblank_enable(dev, pipe));
>  	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
>  }
> -- 
> 1.9.1
>

Ville Syrjala Feb. 9, 2016, 11:10 a.m. UTC | #2

On Tue, Feb 09, 2016 at 11:06:18AM +0100, Daniel Vetter wrote:
> On Mon, Feb 08, 2016 at 02:13:27AM +0100, Mario Kleiner wrote:
> > drm_vblank_offdelay can have three different types of values:
> > 
> > < 0 is to be always treated the same as dev->vblank_disable_immediate
> > = 0 is to be treated as "never disable vblanks"
> > > 0 is to be treated as disable immediate if kms driver wants it
> >     that way via dev->vblank_disable_immediate. Otherwise it is
> >     a disable timeout in msecs.
> > 
> > This got broken in Linux 3.18+ for the implementation of
> > drm_vblank_on. If the user specified a value of zero which should
> > always reenable vblank irqs in this function, a kms driver could
> > override the users choice by setting vblank_disable_immediate
> > to true. This patch fixes the regression and keeps the user in
> > control.
> > 
> > Signed-off-by: Mario Kleiner <[e-mail]>
> > Cc: <[e-mail]> # 3.18+
> > Cc: [e-mail]
> > Cc: [e-mail]
> > Cc: [e-mail]
> > Cc: [e-mail]
> > Cc: [e-mail]
> > Cc: [e-mail]
> > Cc: [e-mail]
> > ---
> >  drivers/gpu/drm/drm_irq.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
> > index 5c27ad3..fb17c45 100644
> > --- a/drivers/gpu/drm/drm_irq.c
> > +++ b/drivers/gpu/drm/drm_irq.c
> > @@ -1492,8 +1492,8 @@ void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
> >  	 * re-enable interrupts if there are users left, or the
> >  	 * user wishes vblank interrupts to be enabled all the time.
> >  	 */
> > -	if (atomic_read(&vblank->refcount) != 0 ||
> > -	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
> > +	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
> > +	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))
> 
> Hm, shouldn't we change this to only enable the vblank irq if we need it,
> i.e. offdelay == 0? For delayed disabling there's kinda no need to enable
> it superflously after a modeset, if userspace didn't yet ask for vblank
> timestamps. But then is was specifically added by Ville in cd19e52aee922,
> so I guess someone really wants this.

IIRC what I wanted was to just re-enable the interrupt for the offdelay==0
case. I think it just ended up as a mess due to changing some of the
semantics of offdelay<0 vs. offdelay==0 vs. disable_immediate during the 
review of the series. So yeah, given how drm_vblank_put() works now, I'd
just make this check for offdelay==0.

> 
> Reviewed-by: Daniel Vetter <[e-mail]>
> 
> >  		WARN_ON(drm_vblank_enable(dev, pipe));
> >  	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
> >  }
> > -- 
> > 1.9.1
> > 
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

Mario Kleiner Feb. 9, 2016, 1:29 p.m. UTC | #3

On 02/09/2016 12:10 PM, Ville Syrjälä wrote:
> On Tue, Feb 09, 2016 at 11:06:18AM +0100, Daniel Vetter wrote:
>> On Mon, Feb 08, 2016 at 02:13:27AM +0100, Mario Kleiner wrote:
>>> drm_vblank_offdelay can have three different types of values:
>>>
>>> < 0 is to be always treated the same as dev->vblank_disable_immediate
>>> = 0 is to be treated as "never disable vblanks"
>>>> 0 is to be treated as disable immediate if kms driver wants it
>>>      that way via dev->vblank_disable_immediate. Otherwise it is
>>>      a disable timeout in msecs.
>>>
>>> This got broken in Linux 3.18+ for the implementation of
>>> drm_vblank_on. If the user specified a value of zero which should
>>> always reenable vblank irqs in this function, a kms driver could
>>> override the users choice by setting vblank_disable_immediate
>>> to true. This patch fixes the regression and keeps the user in
>>> control.
>>>
>>> Signed-off-by: Mario Kleiner <[e-mail]>
>>> Cc: <[e-mail]> # 3.18+
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> Cc: [e-mail]
>>> ---
>>>   drivers/gpu/drm/drm_irq.c | 4 ++--
>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
>>> index 5c27ad3..fb17c45 100644
>>> --- a/drivers/gpu/drm/drm_irq.c
>>> +++ b/drivers/gpu/drm/drm_irq.c
>>> @@ -1492,8 +1492,8 @@ void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
>>>   	 * re-enable interrupts if there are users left, or the
>>>   	 * user wishes vblank interrupts to be enabled all the time.
>>>   	 */
>>> -	if (atomic_read(&vblank->refcount) != 0 ||
>>> -	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
>>> +	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
>>> +	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))
>>
>> Hm, shouldn't we change this to only enable the vblank irq if we need it,
>> i.e. offdelay == 0? For delayed disabling there's kinda no need to enable
>> it superflously after a modeset, if userspace didn't yet ask for vblank
>> timestamps. But then is was specifically added by Ville in cd19e52aee922,
>> so I guess someone really wants this.
>
> IIRC what I wanted was to just re-enable the interrupt for the offdelay==0
> case. I think it just ended up as a mess due to changing some of the
> semantics of offdelay<0 vs. offdelay==0 vs. disable_immediate during the
> review of the series. So yeah, given how drm_vblank_put() works now, I'd
> just make this check for offdelay==0.
>
>>
>> Reviewed-by: Daniel Vetter <[e-mail]>
>>

I can change that to offdelay==0 only, if you want. It was mostly about 
preserving what's there while at the same time fixing the important 
offdelay==0 user override.

-mario

>>>   		WARN_ON(drm_vblank_enable(dev, pipe));
>>>   	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
>>>   }
>>> --
>>> 1.9.1
>>>
>>
>> --
>> Daniel Vetter
>> Software Engineer, Intel Corporation
>> http://blog.ffwll.ch
>

Ville Syrjala Feb. 9, 2016, 1:41 p.m. UTC | #4

On Tue, Feb 09, 2016 at 02:29:49PM +0100, Mario Kleiner wrote:
> On 02/09/2016 12:10 PM, Ville Syrjälä wrote:
> > On Tue, Feb 09, 2016 at 11:06:18AM +0100, Daniel Vetter wrote:
> >> On Mon, Feb 08, 2016 at 02:13:27AM +0100, Mario Kleiner wrote:
> >>> drm_vblank_offdelay can have three different types of values:
> >>>
> >>> < 0 is to be always treated the same as dev->vblank_disable_immediate
> >>> = 0 is to be treated as "never disable vblanks"
> >>>> 0 is to be treated as disable immediate if kms driver wants it
> >>>      that way via dev->vblank_disable_immediate. Otherwise it is
> >>>      a disable timeout in msecs.
> >>>
> >>> This got broken in Linux 3.18+ for the implementation of
> >>> drm_vblank_on. If the user specified a value of zero which should
> >>> always reenable vblank irqs in this function, a kms driver could
> >>> override the users choice by setting vblank_disable_immediate
> >>> to true. This patch fixes the regression and keeps the user in
> >>> control.
> >>>
> >>> Signed-off-by: Mario Kleiner <[e-mail]>
> >>> Cc: <[e-mail]> # 3.18+
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> Cc: [e-mail]
> >>> ---
> >>>   drivers/gpu/drm/drm_irq.c | 4 ++--
> >>>   1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
> >>> index 5c27ad3..fb17c45 100644
> >>> --- a/drivers/gpu/drm/drm_irq.c
> >>> +++ b/drivers/gpu/drm/drm_irq.c
> >>> @@ -1492,8 +1492,8 @@ void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
> >>>   	 * re-enable interrupts if there are users left, or the
> >>>   	 * user wishes vblank interrupts to be enabled all the time.
> >>>   	 */
> >>> -	if (atomic_read(&vblank->refcount) != 0 ||
> >>> -	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
> >>> +	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
> >>> +	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))
> >>
> >> Hm, shouldn't we change this to only enable the vblank irq if we need it,
> >> i.e. offdelay == 0? For delayed disabling there's kinda no need to enable
> >> it superflously after a modeset, if userspace didn't yet ask for vblank
> >> timestamps. But then is was specifically added by Ville in cd19e52aee922,
> >> so I guess someone really wants this.
> >
> > IIRC what I wanted was to just re-enable the interrupt for the offdelay==0
> > case. I think it just ended up as a mess due to changing some of the
> > semantics of offdelay<0 vs. offdelay==0 vs. disable_immediate during the
> > review of the series. So yeah, given how drm_vblank_put() works now, I'd
> > just make this check for offdelay==0.
> >
> >>
> >> Reviewed-by: Daniel Vetter <[e-mail]>
> >>
> 
> I can change that to offdelay==0 only, if you want. It was mostly about 
> preserving what's there while at the same time fixing the important 
> offdelay==0 user override.

Yeah, just offdelay==0 seems best. Otherwise I think we could actually
leave the interrupt enabled indefinitely w/ offdelay>0 since there's not
going to be a drm_vblank_put() to arm the disable timer.

> 
> -mario
> 
> >>>   		WARN_ON(drm_vblank_enable(dev, pipe));
> >>>   	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
> >>>   }
> >>> --
> >>> 1.9.1
> >>>
> >>
> >> --
> >> Daniel Vetter
> >> Software Engineer, Intel Corporation
> >> http://blog.ffwll.ch
> >

Daniel Vetter Feb. 9, 2016, 2:31 p.m. UTC | #5

On Tue, Feb 09, 2016 at 03:41:28PM +0200, Ville Syrjälä wrote:
> On Tue, Feb 09, 2016 at 02:29:49PM +0100, Mario Kleiner wrote:
> > On 02/09/2016 12:10 PM, Ville Syrjälä wrote:
> > > On Tue, Feb 09, 2016 at 11:06:18AM +0100, Daniel Vetter wrote:
> > >> On Mon, Feb 08, 2016 at 02:13:27AM +0100, Mario Kleiner wrote:
> > >>> drm_vblank_offdelay can have three different types of values:
> > >>>
> > >>> < 0 is to be always treated the same as dev->vblank_disable_immediate
> > >>> = 0 is to be treated as "never disable vblanks"
> > >>>> 0 is to be treated as disable immediate if kms driver wants it
> > >>>      that way via dev->vblank_disable_immediate. Otherwise it is
> > >>>      a disable timeout in msecs.
> > >>>
> > >>> This got broken in Linux 3.18+ for the implementation of
> > >>> drm_vblank_on. If the user specified a value of zero which should
> > >>> always reenable vblank irqs in this function, a kms driver could
> > >>> override the users choice by setting vblank_disable_immediate
> > >>> to true. This patch fixes the regression and keeps the user in
> > >>> control.
> > >>>
> > >>> Signed-off-by: Mario Kleiner <[e-mail]>
> > >>> Cc: <[e-mail]> # 3.18+
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> Cc: [e-mail]
> > >>> ---
> > >>>   drivers/gpu/drm/drm_irq.c | 4 ++--
> > >>>   1 file changed, 2 insertions(+), 2 deletions(-)
> > >>>
> > >>> diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
> > >>> index 5c27ad3..fb17c45 100644
> > >>> --- a/drivers/gpu/drm/drm_irq.c
> > >>> +++ b/drivers/gpu/drm/drm_irq.c
> > >>> @@ -1492,8 +1492,8 @@ void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
> > >>>   	 * re-enable interrupts if there are users left, or the
> > >>>   	 * user wishes vblank interrupts to be enabled all the time.
> > >>>   	 */
> > >>> -	if (atomic_read(&vblank->refcount) != 0 ||
> > >>> -	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
> > >>> +	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
> > >>> +	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))
> > >>
> > >> Hm, shouldn't we change this to only enable the vblank irq if we need it,
> > >> i.e. offdelay == 0? For delayed disabling there's kinda no need to enable
> > >> it superflously after a modeset, if userspace didn't yet ask for vblank
> > >> timestamps. But then is was specifically added by Ville in cd19e52aee922,
> > >> so I guess someone really wants this.
> > >
> > > IIRC what I wanted was to just re-enable the interrupt for the offdelay==0
> > > case. I think it just ended up as a mess due to changing some of the
> > > semantics of offdelay<0 vs. offdelay==0 vs. disable_immediate during the
> > > review of the series. So yeah, given how drm_vblank_put() works now, I'd
> > > just make this check for offdelay==0.
> > >
> > >>
> > >> Reviewed-by: Daniel Vetter <[e-mail]>
> > >>
> > 
> > I can change that to offdelay==0 only, if you want. It was mostly about 
> > preserving what's there while at the same time fixing the important 
> > offdelay==0 user override.
> 
> Yeah, just offdelay==0 seems best. Otherwise I think we could actually
> leave the interrupt enabled indefinitely w/ offdelay>0 since there's not
> going to be a drm_vblank_put() to arm the disable timer.

Yeah, r-b on that, after Ville clarified the intention of his commit. When
you respin pls cite that commit, and explain why intention/implementation
diverged quickly.

Thanks, Daniel

> 
> > 
> > -mario
> > 
> > >>>   		WARN_ON(drm_vblank_enable(dev, pipe));
> > >>>   	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
> > >>>   }
> > >>> --
> > >>> 1.9.1
> > >>>
> > >>
> > >> --
> > >> Daniel Vetter
> > >> Software Engineer, Intel Corporation
> > >> http://blog.ffwll.ch
> > >
> 
> -- 
> Ville Syrjälä
> Intel OTC

Patch

diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
index 5c27ad3..fb17c45 100644
--- a/drivers/gpu/drm/drm_irq.c
+++ b/drivers/gpu/drm/drm_irq.c
@@ -1492,8 +1492,8 @@  void drm_vblank_on(struct drm_device *dev, unsigned int pipe)
 	 * re-enable interrupts if there are users left, or the
 	 * user wishes vblank interrupts to be enabled all the time.
 	 */
-	if (atomic_read(&vblank->refcount) != 0 ||
-	    (!dev->vblank_disable_immediate && drm_vblank_offdelay == 0))
+	if (atomic_read(&vblank->refcount) != 0 || drm_vblank_offdelay == 0 ||
+	    (!dev->vblank_disable_immediate && drm_vblank_offdelay > 0))
 		WARN_ON(drm_vblank_enable(dev, pipe));
 	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
 }