Re: [PATCH] drm: sun8i-ui/vi: Fix layer zpos change/atomic modesetting

[OndÅej Jirman]

On Wed, Sep 18, 2019 at 10:16:17PM +0200, Maxime Ripard wrote:
> On Wed, Sep 18, 2019 at 05:23:09PM +0200, OndÅej Jirman wrote:
> > Hi,
> >
> > On Wed, Sep 18, 2019 at 04:17:34PM +0200, Maxime Ripard wrote:
> > > Hi,
> > >
> > > On Sun, Sep 15, 2019 at 12:03:37AM +0200, megous@xxxxxxxxxx wrote:
> > > > From: Ondrej Jirman <megous@xxxxxxxxxx>
> > > >
> > > > There are various issues that this re-work of sun8i_[uv]i_layer_enable
> > > > function fixes:
> > > >
> > > > - Make sure that we re-initialize zpos on reset
> > > > - Minimize register updates by doing them only when state changes
> > > > - Fix issue where DE pipe might get disabled even if it is no longer
> > > >   used by the layer that's currently calling sun8i_ui_layer_enable
> > > > - .atomic_disable callback is not really needed because .atomic_update
> > > >   can do the disable too, so drop the duplicate code
> > > >
> > > > Signed-off-by: Ondrej Jirman <megous@xxxxxxxxxx>
> > >
> > > It looks like these fixes should be in separate patches. Is there any
> > > reason it's not the case?
> >
> > Bullet points just describe the resulting effect/benefits of the change to fix
> > the pipe control register update issue (see the referenced e-mail).
> 
> It's definitely ok to have multiple patches needed to address a single
> perceived issue.

Yes, but I can't simply split the patch. In order for each change to work on its
own, they'd have to be done differently than the final result.

I wouldn't mind at all if it was just a simple splitting, but you're asking
for too much work, this time, for no benefit that I can see.

> A commit is not about what you're fixing but what you're changing. And
> the fact that you have tha bullet list in the first place proves that
> you have multiple logical changes in your patch.
> 
> And even then, your commit log mentions that you're fixing multiple
> issues (without explaining them).

I can reword the commit message if that helps, and skip the bullet list if it
is confusing. There's a single core issue and that is that the driver doesn't
update the pipe/channel configuration correctly leading to disabling of
arbitrary layers (not even those being updated - update of UI layer may disable
VI layer as a side effect for example) at wrong times. And only changes
necessary to debug/fix this are included.

I may try generating a nicer patch with a different diff options, if it makes it
more readable for review.

> > I can maybe split off the first bullet point into a separate patch. But
> > I can't guarantee it will not make the original issue worse, because it might
> > have been hiding the other issue with register updates.
> >
> > The rest is just a result of the single logical change. It doesn't work
> > individually, it all has the goal of fixing the issue as a whole.
> >
> > If I were to split it I would have to actually re-implement .atomic_disable
> > callback only to remove it in the next patch. I don't see the benefit.
> 
> Your commit log says that you remove atomic_disable. Why would you
> remove it, to add it back, to remove it again?

Because if I remove it I need to re-implement the functionality in the update
callback. The core will change what is called based on presence of callbacks.
It's not a simple removal.

If I first implement the new sun8i_[uv]i_layer_enable and update callback,
keeping a disable callback would not work, because the new update callback
will only work if disable callback is not defined (because it it is, then
the drm core will not call the update callback in all cases that I need).

regards,
	o.

> Maxime



> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel