Re: [Xen-devel] [PATCH v3 1/4] drm: Add drm_crtc_has_vblank()

[Daniel Vetter <daniel@xxxxxxxx>]

On Wed, Jan 22, 2020 at 09:53:42AM +0100, Thomas Zimmermann wrote:
> Hi
> 
> Am 22.01.20 um 09:31 schrieb Daniel Vetter:
> > On Mon, Jan 20, 2020 at 01:20:48PM +0100, Thomas Zimmermann wrote:
> >> The new interface drm_crtc_has_vblank() return true if vblanking has
> >> been initialized for a certain CRTC, or false otherwise. This function
> >> will be useful for initializing CRTC state.
> >>
> >> Signed-off-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
> >> ---
> >>  drivers/gpu/drm/drm_vblank.c | 21 +++++++++++++++++++++
> >>  include/drm/drm_vblank.h     |  1 +
> >>  2 files changed, 22 insertions(+)
> >>
> >> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
> >> index 1659b13b178c..c20102899411 100644
> >> --- a/drivers/gpu/drm/drm_vblank.c
> >> +++ b/drivers/gpu/drm/drm_vblank.c
> >> @@ -501,6 +501,27 @@ int drm_vblank_init(struct drm_device *dev, unsigned 
> >> int num_crtcs)
> >>  }
> >>  EXPORT_SYMBOL(drm_vblank_init);
> >>  
> >> +/**
> >> + * drm_crtc_has_vblank - test if vblanking has been initialized for
> >> + *                       a CRTC
> >> + * @crtc: the CRTC
> >> + *
> >> + * Drivers may call this function to test if vblank support is
> >> + * initialized for a CRTC. For most hardware this means that vblanking
> >> + * can also be enabled on the CRTC.
> >> + *
> >> + * Returns:
> >> + * True if vblanking has been initialized for the given CRTC, false
> >> + * otherwise.
> >> + */
> >> +bool drm_crtc_has_vblank(const struct drm_crtc *crtc)
> > 
> > So making this specific to a CRTC sounds like a good idea. But it's not
> > the reality, drm_vblank.c assumes that either everything or nothing
> > supports vblanks.
> > 
> > The reason for dev->num_crtcs is historical baggage, it predates kms by a
> > few years. For kms drivers the only two valid values are either 0 or
> > dev->mode_config.num_crtcs. Yes that's an entire different can of worms
> > that's been irking me since forever (ideally drm_vblank_init would somehow
> > loose the num_crtcs argument for kms drivers, but some drivers call this
> > before they've done all the drm_crtc_init calls so it's complicated).
> 
> Maybe as a first step, drm_vblank_init() could use
> dev->mode_config.num_crtcs if the supplied number of CRTCs is zero.
> 
> > 
> > Hence drm_dev_has_vblank as I suggested. That would also allow you to
> > replace a bunch of if (dev->num_crtcs) checks in drm_vblank.c, which
> > should help quite a bit in code readability.
> 
> OK, but I still don't understand why this interface is better overall.
> We don't loose anything by passing in the crtc instead of the device
> structure. And if there's ever a per-crtc vblank initialization, we'd
> have the interface in place already. The tests with "if
> (dev->num_crtcs)" could probably be removed in most places in any case.

You can't use it in drm_vblank.c code, because we only have the
drm_device, not the drm_crtc (in most places at least). Your other patch
series to deprecate the drm_device callbacks for vblanks is a huge step
into the direction to fix that, but still more work needed: We'd
essentially need to copypaste drm_vblank.c into drm_crtc_vblank.c for kms
drivers, and in that copy switch from (dev, pipe) to crtc everywhere. Plus
then move the drm_vblank structure into struct drm_crtc.

Wrt removing the check: In a pile of cases it changes the return value,
which matters both for vblank usage in helper code and the ioctl itself.
From a quick look most of the checks that don't matter are already wrapped
in a WARN.

> We should also consider forking the vblank code for non-KMS drivers.
> While working in this, I found the support for legacy drivers is getting
> in the way at times. With such a fork, legacy drivers could continue
> using struct drm_vblank_crtc, while modern drivers could maybe store
> vblank state directly in struct drm_crtc.

Hm if you want to do all that then the drm_crtc_has_vblank makes sense.
But only after we've done the full split. So maybe make the public
function drm_crtc_has_vblank, which calls the internal-only
drm_has_vblank, and use that internally in drm_vblank.c?

btw I still think a sub-struct for vblank stuff in drm_crtc makes sense,
and drm_vblank_crtc seems to mostly fit the bill.

That way we're at least not adding the the conversion pain of switching
the vblank code over to drm_crtc fully.

Thoughts?
-Daniel

> Anyway, all this is for another patch. Unless you change your mind, I'll
> replace drm_crtc_has_vblank() with drm_dev_has_vblank() for the
> patchset's next iteration.
> 
> Best regards
> Thomas
> 
> > 
> > Cheers, Daniel
> > 
> >> +{
> >> +  struct drm_device *dev = crtc->dev;
> >> +
> >> +  return crtc->index < dev->num_crtcs;
> >> +}
> >> +EXPORT_SYMBOL(drm_crtc_has_vblank);
> >> +
> >>  /**
> >>   * drm_crtc_vblank_waitqueue - get vblank waitqueue for the CRTC
> >>   * @crtc: which CRTC's vblank waitqueue to retrieve
> >> diff --git a/include/drm/drm_vblank.h b/include/drm/drm_vblank.h
> >> index c16c44052b3d..531a6bc12b7e 100644
> >> --- a/include/drm/drm_vblank.h
> >> +++ b/include/drm/drm_vblank.h
> >> @@ -206,6 +206,7 @@ struct drm_vblank_crtc {
> >>  };
> >>  
> >>  int drm_vblank_init(struct drm_device *dev, unsigned int num_crtcs);
> >> +bool drm_crtc_has_vblank(const struct drm_crtc *crtc);
> >>  u64 drm_crtc_vblank_count(struct drm_crtc *crtc);
> >>  u64 drm_crtc_vblank_count_and_time(struct drm_crtc *crtc,
> >>                               ktime_t *vblanktime);
> >> -- 
> >> 2.24.1
> >>
> > 
> 
> -- 
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Maxfeldstr. 5, 90409 Nürnberg, Germany
> (HRB 36809, AG Nürnberg)
> Geschäftsführer: Felix Imendörffer
> 




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

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel