Re: [PATCH 1/2] drm/atomic: Allow vblank-enabled + self-refresh "disable"

[Daniel Vetter]

On Fri, Jan 06, 2023 at 10:08:53AM -0800, Brian Norris wrote:
> Hi Daniel,
> 
> On Fri, Jan 06, 2023 at 06:53:49PM +0100, Daniel Vetter wrote:
> > On Thu, Jan 05, 2023 at 05:40:17PM -0800, Brian Norris wrote:
> > > The self-refresh helper framework overloads "disable" to sometimes mean
> > > "go into self-refresh mode," and this mode activates automatically
> > > (e.g., after some period of unchanging display output). In such cases,
> > > the display pipe is still considered "on", and user-space is not aware
> > > that we went into self-refresh mode. Thus, users may expect that
> > > vblank-related features (such as DRM_IOCTL_WAIT_VBLANK) still work
> > > properly.
> > > 
> > > However, we trigger the WARN_ONCE() here if a CRTC driver tries to leave
> > > vblank enabled here.
> > > 
> > > Add a new exception, such that we allow CRTCs to be "disabled" (with
> > > self-refresh active) with vblank interrupts still enabled.
> > > 
> > > Cc: <stable@xxxxxxxxxxxxxxx> # dependency for subsequent patch
> > > Signed-off-by: Brian Norris <briannorris@xxxxxxxxxxxx>
> > > ---
> > > 
> > >  drivers/gpu/drm/drm_atomic_helper.c | 6 ++++++
> > >  1 file changed, 6 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
> > > index d579fd8f7cb8..7b5eddadebd5 100644
> > > --- a/drivers/gpu/drm/drm_atomic_helper.c
> > > +++ b/drivers/gpu/drm/drm_atomic_helper.c
> > > @@ -1207,6 +1207,12 @@ disable_outputs(struct drm_device *dev, struct drm_atomic_state *old_state)
> > >  
> > >  		if (!drm_dev_has_vblank(dev))
> > >  			continue;
> > > +		/*
> > > +		 * Self-refresh is not a true "disable"; let vblank remain
> > > +		 * enabled.
> > > +		 */
> > > +		if (new_crtc_state->self_refresh_active)
> > > +			continue;
> > 
> > This very fishy, because we check in crtc_needs_disable whether this
> > output should stay on due to self-refresh. Which means you should never
> > end up in here.
> 
> That's not what crtc_needs_disable() does w.r.t. self-refresh. In fact,
> it's the opposite; see, for example, the
> |new_state->self_refresh_active| clause. That clause means that if we're
> entering self-refresh, we *intend* to disable (i.e., we return 'true').
> That's because like I mention above, the self-refresh helpers overload
> what "disable" means.
> 
> I'll also add my caveat again that I'm a bit new to DRM, so feel free to
> continue to correct me if I'm wrong :) Or perhaps Sean Paul could
> provide second opinions, as I believe he wrote this stuff.

I already replied in another thread with hopefully less nonsense from my
side :-)

> > And yes vblank better work in self refresh :-) If it doesn't, then you
> > need to fake it with a timer, that's at least what i915 has done for
> > transparent self-refresh.
> 
> OK! Then that sounds like it at least ACKs my general idea for this
> series. (Michel and I poked at a few ideas in the thread at [1] and
> landed on approx. this solution, or else a fake/timer like you suggest.)

Yeah once I stopped looking at this the wrong way round it does make sense
what you're doing. See my other reply, I think with just this series here
the vblanks still stall out? Or does your hw actually keep generating
vblank irq with the display off?

> > We might need a few more helpers. Also, probably more igt, or is this
> > something igt testing has uncovered? If so, please cite the igt testcase
> > which hits this.
> 
> The current patch only fixes a warning that comes when I try to do the
> second patch. The second patch is a direct product of an IGT test
> failure (a few of kms_vblank's subtests), and I linked [1] the KernelCI
> report there.

Ah yeah that makes sense. Would be good to cite that in this patch too,
because I guess the same kms_vblank tests can also hit this warning here?
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch