2018-09-26 14:27:27

by Eugeniy Paltsev

Subject: [PATCH] DRM: VBLANK: provide valid timestamp for EVENT_FLIP

If driver/HW doesn't support vblank functionality (for example
UDL driver, ARCPGU driver, ...) we always have vblank->time == 0.
In result we always provide zero timestamp for
DRM_EVENT_FLIP_COMPLETE. This breaks userspace apps (for example
weston) which relies on timestamp value.

Setup time to provide valid timestamp for DRM_EVENT_FLIP_COMPLETE
event.

Cc: [email protected]
Signed-off-by: Eugeniy Paltsev <[email protected]>
---
drivers/gpu/drm/drm_vblank.c | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
index 28cdcf76b6f9..0d19aca48782 100644
--- a/drivers/gpu/drm/drm_vblank.c
+++ b/drivers/gpu/drm/drm_vblank.c
@@ -911,6 +911,15 @@ void drm_crtc_send_vblank_event(struct drm_crtc *crtc,

if (dev->num_crtcs > 0) {
seq = drm_vblank_count_and_time(dev, pipe, &now);
+
+ /*
+ * If driver/HW doesn't support vblank functionality we
+ * always have vblank->time == 0. Setup time to provide valid
+ * timestamp for DRM_EVENT_FLIP_COMPLETE event.
+ */
+ if (!now && e->event.base.type == DRM_EVENT_FLIP_COMPLETE)
+ now = ktime_get();
+
} else {
seq = 0;

--
2.14.4

2018-09-26 16:37:48

by Ville Syrjälä

Subject: Re: [PATCH] DRM: VBLANK: provide valid timestamp for EVENT_FLIP

On Wed, Sep 26, 2018 at 05:25:35PM +0300, Eugeniy Paltsev wrote:
> If driver/HW doesn't support vblank functionality (for example
> UDL driver, ARCPGU driver, ...) we always have vblank->time == 0.
> In result we always provide zero timestamp for
> DRM_EVENT_FLIP_COMPLETE. This breaks userspace apps (for example
> weston) which relies on timestamp value.
>
> Setup time to provide valid timestamp for DRM_EVENT_FLIP_COMPLETE
> event.
>
> Cc: [email protected]
> Signed-off-by: Eugeniy Paltsev <[email protected]>
> ---
> drivers/gpu/drm/drm_vblank.c | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> index 28cdcf76b6f9..0d19aca48782 100644
> --- a/drivers/gpu/drm/drm_vblank.c
> +++ b/drivers/gpu/drm/drm_vblank.c
> @@ -911,6 +911,15 @@ void drm_crtc_send_vblank_event(struct drm_crtc *crtc,
>
> if (dev->num_crtcs > 0) {

If those drivers don't support vblank stuff why are they calling
drm_vblank_init() w/ num_crtcs > 0? If they didn't they'd take the
else branch which already does what you want.

> seq = drm_vblank_count_and_time(dev, pipe, &now);
> +
> + /*
> + * If driver/HW doesn't support vblank functionality we
> + * always have vblank->time == 0. Setup time to provide valid
> + * timestamp for DRM_EVENT_FLIP_COMPLETE event.
> + */
> + if (!now && e->event.base.type == DRM_EVENT_FLIP_COMPLETE)
> + now = ktime_get();
> +
> } else {
> seq = 0;
>
> --
> 2.14.4
>
> _______________________________________________
> dri-devel mailing list
> [email protected]
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

--
Ville Syrj?l?
Intel

2018-09-28 14:32:48

by Eugeniy Paltsev

Subject: Re: [PATCH] DRM: VBLANK: provide valid timestamp for EVENT_FLIP

On Wed, 2018-09-26 at 19:37 +0300, Ville Syrjälä wrote:
> On Wed, Sep 26, 2018 at 05:25:35PM +0300, Eugeniy Paltsev wrote:
> > If driver/HW doesn't support vblank functionality (for example
> > UDL driver, ARCPGU driver, ...) we always have vblank->time == 0.
> > In result we always provide zero timestamp for
> > DRM_EVENT_FLIP_COMPLETE. This breaks userspace apps (for example
> > weston) which relies on timestamp value.
> >
> > Setup time to provide valid timestamp for DRM_EVENT_FLIP_COMPLETE
> > event.
> >
> > Signed-off-by: Eugeniy Paltsev <[email protected]>
> > ---
> > drivers/gpu/drm/drm_vblank.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> > index 28cdcf76b6f9..0d19aca48782 100644
> > --- a/drivers/gpu/drm/drm_vblank.c
> > +++ b/drivers/gpu/drm/drm_vblank.c
> > @@ -911,6 +911,15 @@ void drm_crtc_send_vblank_event(struct drm_crtc *crtc,
> >
> > if (dev->num_crtcs > 0) {
>
> If those drivers don't support vblank stuff why are they calling
> bldrm_vank_init() w/ num_crtcs > 0? If they didn't they'd take the
> else branch which already does what you want.

Hmm...
As I can see UDL call drm_vblank_init() with num_crtcs = 1 but UDL HW
doesn't support vblank at all.
I guess I need to fix it.

> > seq = drm_vblank_count_and_time(dev, pipe, &now);
> > +
> > + /*
> > + * If driver/HW doesn't support vblank functionality we
> > + * always have vblank->time == 0. Setup time to provide valid
> > + * timestamp for DRM_EVENT_FLIP_COMPLETE event.
> > + */
> > + if (!now && e->event.base.type == DRM_EVENT_FLIP_COMPLETE)
> > + now = ktime_get();
> > +
> > } else {
> > seq = 0;
> >
> > --
> > 2.14.4
> >
> > _______________________________________________
> > dri-devel mailing list
> > [email protected]
> > https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_dri-2Ddevel&d=DwIDAw&c=DPL6_X_6JkXFx7AXWqB0tg&r=ZlJN1M
> > riPUTkBKCrPSx67GmaplEUGcAEk9yPtCLdUXI&m=5bbb9SeVXTMPDa1bXQxqlBIOnSdKaKY5W0BkMCgwijE&s=50EhApCZAeMMDUEd6FVpXf5owd1Bi3vn5GaBt0KE2Cw&e=
>
>
--
Eugeniy Paltsev