lkml.org 
[lkml]   [2019]   [Jun]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3 3/4] drm/vblank: estimate vblank while disabling vblank if interrupt disabled
On Tue, Jun 25, 2019 at 06:59:14PM +0100, Robert Beckett wrote:
> If interrupts are disabled (e.g. via vblank_disable_fn) and we come to
> disable vblank, update the vblank count to best guess as to what it
> would be had the interrupts remained enabled, and update the timesamp to
> now.
>
> This avoids a stale vblank event being sent while disabling crtcs during
> atomic modeset.
>
> Fixes: 68036b08b91bc ("drm/vblank: Do not update vblank count if interrupts
> are already disabled.")
>
> Signed-off-by: Robert Beckett <bob.beckett@collabora.com>
> ---
> drivers/gpu/drm/drm_vblank.c | 16 +++++++++++++++-
> 1 file changed, 15 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> index 7dabb2bdb733..db68b8cbf797 100644
> --- a/drivers/gpu/drm/drm_vblank.c
> +++ b/drivers/gpu/drm/drm_vblank.c
> @@ -375,9 +375,23 @@ void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe)
> * interrupts were enabled. This avoids calling the ->disable_vblank()
> * operation in atomic context with the hardware potentially runtime
> * suspended.
> + * If interrupts are disabled (e.g. via blank_disable_fn) then make
> + * best guess as to what it would be now and make sure we have an up
> + * to date timestamp.
> */
> - if (!vblank->enabled)
> + if (!vblank->enabled) {
> + ktime_t now = ktime_get();

Would be nice to fake this a bit more accurately and round the timestamp
here to a multiple of the frame duration, i.e. ...
> + u32 diff = 0;
> + if (vblank->framedur_ns) {
> + u64 diff_ns = ktime_to_ns(ktime_sub(now, vblank->time));
> + diff = DIV_ROUND_CLOSEST_ULL(diff_ns,
> + vblank->framedur_ns);
> + }

now = vblank->time + diff * vblank_framedur_ns;

Picking the right macro for doing 64bit multiplies left to you :-)

> +
> + store_vblank(dev, pipe, diff, now, vblank->count);
> +
> goto out;
> + }
>
> /*
> * Update the count and timestamp to maintain the

Somewhat unhappy that we're duplicating this logic with
drm_update_vblank_count, but it's just 2 lines of math.
-Daniel

> --
> 2.18.0
>

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

\
 
 \ /
  Last update: 2019-06-25 22:12    [W:0.077 / U:6.360 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site