Re: [PATCH v12 2/5] tty: goldfish: introduce gf_ioread32()/gf_iowrite32()

[Laurent Vivier]

Le 26/01/2022 à 14:41, Greg KH a écrit :
> On Fri, Jan 21, 2022 at 09:07:35PM +0100, Laurent Vivier wrote:
> > Revert
> > commit da31de35cd2f ("tty: goldfish: use __raw_writel()/__raw_readl()")
>
> Why?
>
> > and define gf_ioread32()/gf_iowrite32() to be able to use accessors
> > defined by the architecture.
>
> What does this do?
>
> > Cc: stable@xxxxxxxxxxxxxxx # v5.11+
> > Fixes: da31de35cd2f ("tty: goldfish: use __raw_writel()/__raw_readl()")
> > Signed-off-by: Laurent Vivier <laurent@xxxxxxxxx>
> > ---
> >   drivers/tty/goldfish.c   | 20 ++++++++++----------
> >   include/linux/goldfish.h | 15 +++++++++++----
> >   2 files changed, 21 insertions(+), 14 deletions(-)
...
> > --
> > 2.34.1
>
> This feels like a step backwards.  Why keep this level of indirection
> for no good reason?

It was proposed by Arnd on my previous iteration of the series when I wanted to update goldfish-rtc 
in the same way:

https://lore.kernel.org/all/CAK8P3a1H6-sd_+FqnOq0Zhj=L51EWuW5VCcYeTENcp3+PkTC4Q@xxxxxxxxxxxxxx/

Keeping __raw_XXX() functions works on most of the cases except if the current CPU endianness can 
differ from the architecture one (like a ppc64le kernel (little-endian) running a ppc64 machine 
(big-endian)).

The best solution would be to update QEMU to set the device as a little-endian one, but google 
didn't merge its implemention in upstream QEMU and this would break all other OSes using goldfish 
devices on big-endian architectures.

Thanks,
Laurent