Nouveau - Nov 2018 - [PATCH] drm/nouveau: tegra: Initialize mode configuration

On Tue, Nov 06, 2018 at 06:41:22PM +0200, Ville Syrjälä
wrote:
> On Tue, Nov 06, 2018 at 05:24:15PM +0100, Thierry Reding wrote:
> > From: Thierry Reding <treding at nvidia.com>
> > 
> > Irrespective of whether or not the device has any usable outputs, the
> > modesetting helpers will try to register all the resources such as
CRTCs
> > and planes. Unfortunately, the helpers rely on drm_mode_config_init()
to
> > properly set up internal data structures. Since the Tegra GPU does not
> > have a display engine, the Nouveau driver doesn't set this up for
Tegra,
> > which results in the following oops on driver probe:
> 
> Remove DRIVER_MODESET ?
Yeah, that works as well. Technically I suppose somebody could be
instantiating the IP in an SoC and include a display engine, in which
case we'd need the DRIVER_MODESET feature again, but I'm not aware of
anyone doing that, so it may not be worth planning for that at this
point.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL:
<https://lists.freedesktop.org/archives/nouveau/attachments/20181107/905eb2dc/attachment.sig>

On Wed, Nov 7, 2018 at 10:34 AM Thierry Reding <thierry.reding at
gmail.com> wrote:
>
> On Tue, Nov 06, 2018 at 06:41:22PM +0200, Ville Syrjälä wrote:
> > On Tue, Nov 06, 2018 at 05:24:15PM +0100, Thierry Reding wrote:
> > > From: Thierry Reding <treding at nvidia.com>
> > >
> > > Irrespective of whether or not the device has any usable outputs,
the
> > > modesetting helpers will try to register all the resources such
as CRTCs
> > > and planes. Unfortunately, the helpers rely on
drm_mode_config_init() to
> > > properly set up internal data structures. Since the Tegra GPU
does not
> > > have a display engine, the Nouveau driver doesn't set this up
for Tegra,
> > > which results in the following oops on driver probe:
> >
> > Remove DRIVER_MODESET ?
>
> Yeah, that works as well. Technically I suppose somebody could be
> instantiating the IP in an SoC and include a display engine, in which
> case we'd need the DRIVER_MODESET feature again, but I'm not aware
of
> anyone doing that, so it may not be worth planning for that at this
> point.
I thought DRIVER_MODESET was required even for rendernodes. Is that no
longer the case?

Note that e.g. GF117 GPUs also don't have a display engine hooked up
in nouveau (or in hardware).

  -ilia

On Wed, Nov 07, 2018 at 10:52:11AM -0500, Ilia Mirkin
wrote:
> On Wed, Nov 7, 2018 at 10:34 AM Thierry Reding <thierry.reding at
gmail.com> wrote:
> >
> > On Tue, Nov 06, 2018 at 06:41:22PM +0200, Ville Syrjälä wrote:
> > > On Tue, Nov 06, 2018 at 05:24:15PM +0100, Thierry Reding wrote:
> > > > From: Thierry Reding <treding at nvidia.com>
> > > >
> > > > Irrespective of whether or not the device has any usable
outputs, the
> > > > modesetting helpers will try to register all the resources
such as CRTCs
> > > > and planes. Unfortunately, the helpers rely on
drm_mode_config_init() to
> > > > properly set up internal data structures. Since the Tegra
GPU does not
> > > > have a display engine, the Nouveau driver doesn't set
this up for Tegra,
> > > > which results in the following oops on driver probe:
> > >
> > > Remove DRIVER_MODESET ?
> >
> > Yeah, that works as well. Technically I suppose somebody could be
> > instantiating the IP in an SoC and include a display engine, in which
> > case we'd need the DRIVER_MODESET feature again, but I'm not
aware of
> > anyone doing that, so it may not be worth planning for that at this
> > point.
> 
> I thought DRIVER_MODESET was required even for rendernodes. Is that no
> longer the case?
Indeed. Removing the DRIVER_MODESET feature gets the driver to load
properly, but trying to actually use Nouveau (e.g. kmscube) crashes
in other places.

Now that you mention it, I vaguely remember trying to do something
similar a long time ago and I had to split up DRIVER_MODESET into more
fine-grained features (basically DRIVER_MODESET and DRIVER_LEGACY, if
I remember correctly) to avoid crashes. But this must have been three
or four years ago and I can't recall the exact details.
> Note that e.g. GF117 GPUs also don't have a display engine hooked up
> in nouveau (or in hardware).
That's interesting. I think there'd be some symmetric elegance to it if
the PCI driver supported DRIVER_MODESET irrespective of the feature set
of the card and if the platform driver did the same. In the end, this
isn't really a driver-specific feature, but a device-specific feature.
Or to put it another way: the driver inherently supports modesetting
features, irrespective of a device's specific capabilities.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL:
<https://lists.freedesktop.org/archives/nouveau/attachments/20181107/6959f76e/attachment.sig>