diff mbox

drm/vblank: Fix delta_ns to an absolute value

Message ID 1504249636-11681-1-git-send-email-hoegeun.kwon@samsung.com (mailing list archive)
State New, archived
Headers show

Commit Message

Hoegeun Kwon Sept. 1, 2017, 7:07 a.m. UTC
If scanout started, we should reduce etime by delta_ns. But delta_ns
is negative if scanout has not started. If delta_ns is negative,
subtraction of delta_ns from etime increases etime. This is wrong, the
etime should not be increased, so you have to make delta_ns an
absolute value.

Signed-off-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
---

Hello all,

I think that the etime should not be increased.
In cases where delta_ns is negative, if you get time again after an
interrupt call, there is a problem that the time obtained from the
interrupt becomes the future time instead of the past time.

Please let me know if this patch is wrong.

Best regards,
Hoegeun

 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Ville Syrjälä Sept. 1, 2017, 1:36 p.m. UTC | #1
On Fri, Sep 01, 2017 at 04:07:16PM +0900, Hoegeun Kwon wrote:
> If scanout started, we should reduce etime by delta_ns. But delta_ns
> is negative if scanout has not started. If delta_ns is negative,
> subtraction of delta_ns from etime increases etime. This is wrong, the
> etime should not be increased, so you have to make delta_ns an
> absolute value.
> 
> Signed-off-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
> ---
> 
> Hello all,
> 
> I think that the etime should not be increased.
> In cases where delta_ns is negative, if you get time again after an
> interrupt call, there is a problem that the time obtained from the
> interrupt becomes the future time instead of the past time.
> 
> Please let me know if this patch is wrong.

It is wrong. The timestamp corresponds to the first active pixel of the
frame/field. So while between vblank start and active start we expect
to get a timestamp that is in the future.

> 
> Best regards,
> Hoegeun
> 
>  drivers/gpu/drm/drm_vblank.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> index 70f2b95..a3e0176 100644
> --- a/drivers/gpu/drm/drm_vblank.c
> +++ b/drivers/gpu/drm/drm_vblank.c
> @@ -684,7 +684,7 @@ bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
>  	/* Subtract time delta from raw timestamp to get final
>  	 * vblank_time timestamp for end of vblank.
>  	 */
> -	etime = ktime_sub_ns(etime, delta_ns);
> +	etime = ktime_sub_ns(etime, abs(delta_ns));
>  	*vblank_time = ktime_to_timeval(etime);
>  
>  	DRM_DEBUG_VBL("crtc %u : v p(%d,%d)@ %ld.%ld -> %ld.%ld [e %d us, %d rep]\n",
> -- 
> 1.9.1
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Daniel Vetter Sept. 4, 2017, 7:36 a.m. UTC | #2
On Fri, Sep 01, 2017 at 04:36:25PM +0300, Ville Syrjälä wrote:
> On Fri, Sep 01, 2017 at 04:07:16PM +0900, Hoegeun Kwon wrote:
> > If scanout started, we should reduce etime by delta_ns. But delta_ns
> > is negative if scanout has not started. If delta_ns is negative,
> > subtraction of delta_ns from etime increases etime. This is wrong, the
> > etime should not be increased, so you have to make delta_ns an
> > absolute value.
> > 
> > Signed-off-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
> > ---
> > 
> > Hello all,
> > 
> > I think that the etime should not be increased.
> > In cases where delta_ns is negative, if you get time again after an
> > interrupt call, there is a problem that the time obtained from the
> > interrupt becomes the future time instead of the past time.
> > 
> > Please let me know if this patch is wrong.
> 
> It is wrong. The timestamp corresponds to the first active pixel of the
> frame/field. So while between vblank start and active start we expect
> to get a timestamp that is in the future.

A bit more strict, it's up to drivers when exactly the vblank interrupt
fires. Some fire at vblank start, some at the end, some somewhen around
that. The only rule is that the vblank must not fire before the point of
no return for another page_flip call, i.e. after the vblank event goes
out, the next page_flip (or atomic ioctl) must go to the next vblank.

So yeah both positive and negative deltas make perfect sense.
-Daniel

> 
> > 
> > Best regards,
> > Hoegeun
> > 
> >  drivers/gpu/drm/drm_vblank.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> > index 70f2b95..a3e0176 100644
> > --- a/drivers/gpu/drm/drm_vblank.c
> > +++ b/drivers/gpu/drm/drm_vblank.c
> > @@ -684,7 +684,7 @@ bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
> >  	/* Subtract time delta from raw timestamp to get final
> >  	 * vblank_time timestamp for end of vblank.
> >  	 */
> > -	etime = ktime_sub_ns(etime, delta_ns);
> > +	etime = ktime_sub_ns(etime, abs(delta_ns));
> >  	*vblank_time = ktime_to_timeval(etime);
> >  
> >  	DRM_DEBUG_VBL("crtc %u : v p(%d,%d)@ %ld.%ld -> %ld.%ld [e %d us, %d rep]\n",
> > -- 
> > 1.9.1
> > 
> > _______________________________________________
> > dri-devel mailing list
> > dri-devel@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> -- 
> Ville Syrjälä
> Intel OTC
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Hoegeun Kwon Sept. 5, 2017, 6:36 a.m. UTC | #3
On 09/04/2017 04:36 PM, Daniel Vetter wrote:
> On Fri, Sep 01, 2017 at 04:36:25PM +0300, Ville Syrjälä wrote:
>> On Fri, Sep 01, 2017 at 04:07:16PM +0900, Hoegeun Kwon wrote:
>>> If scanout started, we should reduce etime by delta_ns. But delta_ns
>>> is negative if scanout has not started. If delta_ns is negative,
>>> subtraction of delta_ns from etime increases etime. This is wrong, the
>>> etime should not be increased, so you have to make delta_ns an
>>> absolute value.
>>>
>>> Signed-off-by: Hoegeun Kwon <hoegeun.kwon@samsung.com>
>>> ---
>>>
>>> Hello all,
>>>
>>> I think that the etime should not be increased.
>>> In cases where delta_ns is negative, if you get time again after an
>>> interrupt call, there is a problem that the time obtained from the
>>> interrupt becomes the future time instead of the past time.
>>>
>>> Please let me know if this patch is wrong.
>> It is wrong. The timestamp corresponds to the first active pixel of the
>> frame/field. So while between vblank start and active start we expect
>> to get a timestamp that is in the future.
> A bit more strict, it's up to drivers when exactly the vblank interrupt
> fires. Some fire at vblank start, some at the end, some somewhen around
> that. The only rule is that the vblank must not fire before the point of
> no return for another page_flip call, i.e. after the vblank event goes
> out, the next page_flip (or atomic ioctl) must go to the next vblank.
>
> So yeah both positive and negative deltas make perfect sense.
> -Daniel

Thanks Ville and Daniel,

Your response has been really helpful.
Thanks so much.

Best regards,
Hoegeun

>>> Best regards,
>>> Hoegeun
>>>
>>>   drivers/gpu/drm/drm_vblank.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
>>> index 70f2b95..a3e0176 100644
>>> --- a/drivers/gpu/drm/drm_vblank.c
>>> +++ b/drivers/gpu/drm/drm_vblank.c
>>> @@ -684,7 +684,7 @@ bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
>>>   	/* Subtract time delta from raw timestamp to get final
>>>   	 * vblank_time timestamp for end of vblank.
>>>   	 */
>>> -	etime = ktime_sub_ns(etime, delta_ns);
>>> +	etime = ktime_sub_ns(etime, abs(delta_ns));
>>>   	*vblank_time = ktime_to_timeval(etime);
>>>   
>>>   	DRM_DEBUG_VBL("crtc %u : v p(%d,%d)@ %ld.%ld -> %ld.%ld [e %d us, %d rep]\n",
>>> -- 
>>> 1.9.1
>>>
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>> -- 
>> Ville Syrjälä
>> Intel OTC
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
diff mbox

Patch

diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
index 70f2b95..a3e0176 100644
--- a/drivers/gpu/drm/drm_vblank.c
+++ b/drivers/gpu/drm/drm_vblank.c
@@ -684,7 +684,7 @@  bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 	/* Subtract time delta from raw timestamp to get final
 	 * vblank_time timestamp for end of vblank.
 	 */
-	etime = ktime_sub_ns(etime, delta_ns);
+	etime = ktime_sub_ns(etime, abs(delta_ns));
 	*vblank_time = ktime_to_timeval(etime);
 
 	DRM_DEBUG_VBL("crtc %u : v p(%d,%d)@ %ld.%ld -> %ld.%ld [e %d us, %d rep]\n",