Hi,
I came across this thread/patchset from around June last year:
http://lists.ozlabs.org/pipermail/linuxppc-dev/2009-June/073086.html
where Wolfgang proposed a generic OF-driven UIO driver. The
discussion seemed to stall after Grant Likely indicated he didn't like
the use of a linux-specific compatible binding in the device tree
(compatible="generic-uio").
I guess I have a couple of questions:
* did this patchset go anywhere? I've been using it here the last
few days and it works great.
and more generally:
* Is there a better way to handle the OF bindings for this sort of thing?
Grant's complaint seems to come up often - when you have generic
controllers in a system (SPI/I2C also spring to mind), we need a way
of signalling somehow to the kernel that each instance has a
particular usage intended.
However, the device-tree guys complain whenever anyone tries to encode
anything non-hardware related into the DTS itself.
I guess I'd like to just open up a discussion, see if there's been any
progress towards a general solution.
Thanks,
John
--
John Williams
PetaLogix - Linux Solutions for a Reconfigurable World
w: http://www.petalogix.com p: +61-7-30090663 f: +61-7-30090663
John,
> I came across this thread/patchset from around June last year:
>
> http://lists.ozlabs.org/pipermail/linuxppc-dev/2009-June/073086.html
>
> where Wolfgang proposed a generic OF-driven UIO driver. The
Wolfram, please ;)
> discussion seemed to stall after Grant Likely indicated he didn't like
> the use of a linux-specific compatible binding in the device tree
> (compatible="generic-uio").
I agree with him on that.
> I guess I have a couple of questions:
>
> * did this patchset go anywhere? I've been using it here the last
> few days and it works great.
The idea was to create a mechanism to instantiate bindings at runtime, similar
to new_id for PCI/PCMCIA, e.g.:
$ echo "commodore,c64" > /sys/bus/of_platform/drivers/of_uio_genirq/new_compatible
so we don't have to maintain an ever growing list of hardcoded
compatible-properties for those UIO-devices.
> * Is there a better way to handle the OF bindings for this sort of thing?
Run-time instantiation might help in a couple of other cases; still, in the
progress of unifying/extending the OF-support, it was discussed if it was
possible to get rid of of_platform entirely. It looks like a very challenging
task, but seems to be favoured designwise (at least I do).
> However, the device-tree guys complain whenever anyone tries to encode
> anything non-hardware related into the DTS itself.
Well, if I get a device tree including special properties for Linux and BSD and
whatever may follow, that could get quite confusing :)
> I guess I'd like to just open up a discussion, see if there's been any
> progress towards a general solution.
I decided to wait for the outcome of the of_platform-removal-idea. Though, I
have to admit that in the last weeks I haven't followed of-related things due
to other commitments.
Regards,
Wolfram
--
Pengutronix e.K. | Wolfram Sang |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Wolfram,
On Thursday 28 January 2010 11:45:45 Wolfram Sang wrote:
> > I guess I'd like to just open up a discussion, see if there's been any
> > progress towards a general solution.
>
> I decided to wait for the outcome of the of_platform-removal-idea. Though,
> I have to admit that in the last weeks I haven't followed of-related
> things due to other commitments.
Is there any update on this in the meantime? Do you know of any plans to re-
work/-push this patchset? Just checking...
Thanks,
Stefan
--
DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-0 Fax: (+49)-8142-66989-80 Email: [email protected]
> Is there any update on this in the meantime? Do you know of any plans to re-
> work/-push this patchset? Just checking...
I had another look recently; I wouldn't like to just copy/paste the
"new_id"-mechanism from PCI because the code would then exist three
times in the kernel (PCMCIA being the third user I am aware of). I
wondered if it was feasible to make the "new-id"-code generic and make
those subsystems just use it. I haven't started working on it, though.
Kind regards,
Wolfram
--
Pengutronix e.K. | Wolfram Sang |
Industrial Linux Solutions | http://www.pengutronix.de/ |