Re: [PATCH v10 2/5] rtc: goldfish: introduce goldfish_ioread32()/goldfish_iowrite32()

[Laurent Vivier]

Le 19/01/2022 à 09:49, Arnd Bergmann a écrit :
> On Wed, Jan 19, 2022 at 9:21 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > On Wed, Jan 19, 2022 at 1:05 AM Laurent Vivier <laurent@xxxxxxxxx> wrote:
> >
> > I've just discovered include/linux/goldfish.h, which already has gf_*()
> > accessors for 64-bit, so it'd make sense to move the above there,
> > and adjust the names.
>
> Yes, good idea.

So the idea is to put goldfish accessors inside a "#ifdef CONFIG_M68K ... #else ... #endif" in 
include/linux/goldfish.h and not in generic-asm/io.h for the generic version and 
m68k/include/ams/io.h for the m68k version?

> > Arnd: note that the existing ones do use __raw_writel().
>
> It looks like Laurent introduced that bug in da31de35cd2f ("tty: goldfish: use
> __raw_writel()/__raw_readl()") and could fix it up here. Laurent, was the intent

The idea was to use the native endianness of the CPU, I missed it can differ from the one of the 
architecture.

> of this earlier patch also to make the driver usabel for m68k, or are there
> any other targets you looked at that had mixed up endianness?

Yes, the intent was to make it usable for m68k.
I think all the targets that use goldfish are little-endian, it's why there was no problem until now.

Let me know which solution you prefer, I will update the series accordingly.

Thanks,
Laurent