Re: [PATCH 01/27] ARM: pxa: remove unused board files

[Dmitry Baryshkov]

вс, 8 янв. 2023 г. в 23:05, Arnd Bergmann <arnd@xxxxxxxx>:
>
> On Fri, Jan 6, 2023, at 11:06, Dmitry Baryshkov wrote:
> > пт, 6 янв. 2023 г. в 11:47, Arnd Bergmann <arnd@xxxxxxxx>:
> >> > Is there any conversion to DT you can easily point at as an example of
> >> > the kinds of changes needed?
> >>
> >> Robert Jarzmik and Daniel Mack worked on the conversion of the
> >> PXA platform to DT. Daniel contributed the port for Raumfeld,
> >> which should be complete, while Robert worked on more driver
> >> conversions and mentioned[1] that he had converted additional
> >> boards in the past but did not merge it upstream. They
> >> can probably point you to whatever is missing. I would expect
> >> the generic PXA drivers (spi, mmc, nand, i2c, audio, fb, gpio,
> >> keypad) to  basically work work a correct DT description,
> >> while the machine specific drivers (scoop and pcmcia mainly)
> >> will need DT support in the driver.
> >>
> >> In addition, Linus Walleij and Marc Zyngier have both expressed
> >> interest in keeping sa1100 (h3600, collie, assabet, jornada720)
> >> alive, but those don't have any DT support yet and require
> >> much more work. Also note that while you can now build a kernel
> >> that includes support for all little-endian ARMv4T and ARMv5
> >> machines, StrongARM machine still require a separate kernel
> >> build.
> >
> > I looked into converting collie to use DT several years ago. The major
> > problem was not in the StrongARM itself , but rather in the locomo
> > (platform-specific ASIC) and PCMCIA. Unfortunately I abandoned that
> > work ages ago. RMK didn't seem to be very interested, if I remember
> > correclty.
>
> At least locomo and sa1111 are both private to mach-sa1100
> now and no longer shared with pxa, so this should get
> a little easier. When I had last looking into cleaning up
> sa1100, my impression was that the main work would be converting
> most of the drivers to use dynamic resources instead of
> hardcoded addresses and interrupts. Looking at locomo again,
> my feeling is that this could remain largely unchanged,
> as the locomo downstream drivers (led, keyboard, lcd)
> are already abstracted enough and locomo itself can
> just be an mfd or soc driver.

locomo has been used on Sharp Poodle, if I'm not mistaken. And sa1111
is used on the lubbock, one of the crazy devkits.

>
> > I suspect that the platforms might need to be rebootstrapped from the
> > ground up. This sounds like a fun project for the next Connect demo :D
> >
> > BTW: collie is also supported by the qemu (in fact at some point I
> > mostly used qemu for debugging collie). I don't think that the LCD
> > emulation works, but the rest should be mostly good.
>
> Good to know about the LCD. The qemu support was clearly a
> strong reason for keeping this machine vs the others.

sa1100 LCD controller was significantly different and it used some
special format for the buffers, so I just skipped it at that point.

>
>      Arnd



-- 
With best wishes
Dmitry