[PATCH] platform: support non-pci platform devices

[Rob Clark]

On Mon, Jun 16, 2014 at 7:46 AM, Thierry Reding
<thierry.reding at gmail.com> wrote:
> On Sat, Jun 14, 2014 at 03:58:53PM -0400, Rob Clark wrote:
>> This makes things not completely fail if DDX implements platformProbe()
>> but the device is not actually a PCI device.  Also, the platform device
>> name does not always match the DDX name, so deal with that.  I'm sure
>> there are more cases that find_non_pci_driver() needs to handle for
>> that, but this is a start.
>>
>> Signed-off-by: Rob Clark <robdclark at gmail.com>
>> ---
>> NOTE: I still need to figure out a sane way to workaround the existing
>> bug with non-pci platform devices.  Currently, if DDX implements the
>> platformProbe() hook, then in xf86platformProbeDev() it will get claimed
>> in the autoAddGPU loop, resulting that the old-school Probe() fallback
>> (if you have .conf file) fails too because the device is already claimed.
>> We kind of need a way that the DDX can detect that the xserver does not
>> have this fix, so that it can work around the bug by failing
>> platformProbe().
>>
>>  hw/xfree86/common/xf86platformBus.c | 65 ++++++++++++++++++++++++++++++++++---
>>  1 file changed, 61 insertions(+), 4 deletions(-)
>
> What ever happened to the series I sent a while ago to make this kind of
> setup work with an xorg.conf snippet? The idea at the time was to add an
> OutputClass section that allows matching by kernel driver, so that a DDX
> module could simply provide a corresponding xorg.conf snippet in
> /usr/share/X11/xorg.conf.d.

mostly just that I couldn't find that series, so in anger I re-implemented ;-)

tbh, I don't really mind which approach is taken.  What I did, and I
guess what you did initially, seems like a natural extension of how
things work for PCI devices.  But getting rid of the tables in code
and getting this all from .conf files would be a nice cleanup for both
platform and pci devices.  But I think we can treat that as a
cleanup/enhancement rather than bug fix.

I'd recommend we merge something closer to my current patchset (plus
something like Hans's SERVER_SUPPORTS_PLATFORM_DEVS idea) now (and on
any stable branch that has AutoAddGPU), since that is fixing an actual
bug, and then do .conf file / OutputClass stuff after on master for
next xserver release.

BR,
-R

> Also see here:
>
>         http://lists.x.org/archives/xorg-devel/2014-February/040671.html
>
> I did originally propose something like you did, but that was shot down
> by Dave (I think). Generally the OutputClass series seemed like an
> accepted approach (and had potential users even for PCI devices). Keith
> said on IRC that he had some comments but never got around to posting
> them it seems.
>
> Thierry