Re: remove set_fs for UML

[Al Viro]

On Tue, Dec 21, 2021 at 10:05:03PM +0100, Richard Weinberger wrote:
> Christoph,
> 
> ----- Ursprüngliche Mail -----
> > Von: "hch" <hch@xxxxxx>
> > An: "richard" <richard@xxxxxx>, "anton ivanov" <anton.ivanov@xxxxxxxxxxxxxxxxxx>
> > CC: "x86" <x86@xxxxxxxxxx>, "linux-um" <linux-um@xxxxxxxxxxxxxxxxxxx>, "linux-kernel" <linux-kernel@xxxxxxxxxxxxxxx>
> > Gesendet: Mittwoch, 15. Dezember 2021 17:56:10
> > Betreff: remove set_fs for UML
> 
> > Hi Richard and Anton,
> > 
> > this series removes address space overrides using set_fs for UML.
> > 
> > Diffstat:
> > um/Kconfig                   |    1 -
> > um/include/asm/thread_info.h |    4 ----
> > um/include/asm/uaccess.h     |   21 +++++++++++++++++++--
> > um/kernel/skas/uaccess.c     |   25 -------------------------
> > x86/include/asm/mtrr.h       |    8 +-------
> > x86/kernel/setup.c           |    7 ++++++-
> > x86/um/asm/segment.h         |    8 --------
> >  7 files changed, 26 insertions(+), 48 deletions(-)
> 
> So far UML seems to work with these changes applied. :-)
> I have applied both patches to my UML tree for now, I assume x86 maintainers are fine with
> patch 1/2?

Hmmm...  AFAICS, the right thing to do would be to have __get_kernel_nofault
and __put_kernel_nofault in arch/x86/um/asm/something.  The question is how
to avoid duplicating the x86 implementation of those (along with the asm-goto
fun, etc.)...

But Christoph is right, it's not a new problem.  As far as I'm concerned,
that series looks fine.