Re: [Intel-gfx] [PATCH v8 1/3] gpu: drm: separate panel orientation property creating and value setting

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, 15 Feb 2022 at 16:37, Simon Ser <contact@xxxxxxxxxxx> wrote:
>
> On Tuesday, February 15th, 2022 at 15:38, Emil Velikov <emil.l.velikov@xxxxxxxxx> wrote:
>
> > On Tue, 15 Feb 2022 at 13:55, Simon Ser <contact@xxxxxxxxxxx> wrote:
> > >
> > > On Tuesday, February 15th, 2022 at 13:04, Emil Velikov <emil.l.velikov@xxxxxxxxx> wrote:
> > >
> > > > Greetings everyone,
> > > >
> > > > Padron for joining in so late o/
> > > >
> > > > On Tue, 8 Feb 2022 at 08:42, Hsin-Yi Wang <hsinyi@xxxxxxxxxxxx> wrote:
> > > > >
> > > > > drm_dev_register() sets connector->registration_state to
> > > > > DRM_CONNECTOR_REGISTERED and dev->registered to true. If
> > > > > drm_connector_set_panel_orientation() is first called after
> > > > > drm_dev_register(), it will fail several checks and results in following
> > > > > warning.
> > > > >
> > > > > Add a function to create panel orientation property and set default value
> > > > > to UNKNOWN, so drivers can call this function to init the property earlier
> > > > > , and let the panel set the real value later.
> > > > >
> > > >
> > > > The warning illustrates a genuine race condition, where userspace will
> > > > read the old/invalid property value/state. So this patch masks away
> > > > the WARNING without addressing the actual issue.
> > > > Instead can we fix the respective drivers, so that no properties are
> > > > created after drm_dev_register()?
> > > >
> > > > Longer version:
> > > > As we look into drm_dev_register() it's in charge of creating the
> > > > dev/sysfs nodes (et al). Note that connectors cannot disappear at
> > > > runtime.
> > > > For panel orientation, we are creating an immutable connector
> > > > properly, meaning that as soon as drm_dev_register() is called we must
> > > > ensure that the property is available (if applicable) and set to the
> > > > correct value.
> > >
> > > Unfortunately we can't quite do this. To apply the panel orientation quirks we
> > > need to grab the EDID of the eDP connector, and this happened too late in my
> > > testing.
> > >
> > > What we can do is create the prop early during module load, and update it when
> > > we read the EDID (at the place where we create it right now). User-space will
> > > receive a hotplug event after the EDID is read, so will be able to pick up the
> > > new value if any.
> >
> > Didn't quite get that, are you saying that a GETPROPERTY for the EDID,
> > the ioctl blocks or that we get an empty EDID?
>
> I'm not referring to GETPROPERTY, I'm referring to the driver getting the EDID
> from the sink (here, the eDP panel). In my experimentations with amdgpu I
> noticed that the driver module load finished before the EDID was available to
> the driver. Maybe other drivers behave differently and probe connectors when
> loaded, not sure.
>
I see thanks.

> > The EDID hotplug even thing is neat - sounds like it also signals on
> > panel orientation, correct?
> > On such an event, which properties userspace should be re-fetching -
> > everything or guess randomly?
> >
> > Looking through the documentation, I cannot see a clear answer :-\
>
> User-space should re-fetch *all* properties. In practice some user-space may
> only be fetching some properties, but that should get fixed in user-space.
>
> Also the kernel can indicate that only a single connector changed via the
> "CONNECTOR" uevent prop, or even a single connector property via "PROPERTY".
> See [1] for a user-space implementation. But all of this is purely an optional
> optimization. Re-fetching all properties is a bit slower (especially if some
> drmModeGetConnector calls force-probe connectors) but works perfectly fine.
>
Looking at KDE/kwin (the one I'm running) - doesn't seem like it
handles any of the three "HOTPLUG", "PROPERTY" or "CONNECTOR" uevents
:-\
Skimming through GNOME/mutter - it handles "HOTPLUG", but not the optional ones.

Guess we're in the clear wrt potential races, even though the
documentation on the topic is lacklustre.

> It would be nice to document, if you have the time feel free to send a patch
> and CC danvet, pq and me.
>

Doubt I will have the time in the upcoming weeks, but I'll add it to
my ever-growing TODO list :-P

Thanks
Emil



[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux