Re: [PATCH RFC v2] m68k: remove get_fs()/set_fs()

[Linus Torvalds <torvalds@linux-foundation.org>]

On Wed, Jul 7, 2021 at 6:48 PM Michael Schmitz <schmitzmic@gmail.com> wrote:
>
> Followup to my patch "Fixes to Linus' 'remove set_fs patch'", with
> kernel mode copy macros used in __get/put_kernel_nofault().
> (As soon as I use the same __get/put_user_asm() macros with
> parameters, like in Christoph's patch, everything falls to pieces.)

Ok, strange. If this works for you, then the _concept_ is fine, and
there's something odd going on with the "macros with 'moves' vs 'move'
as a parameter" thing.

> The fact that I had to touch include/linux/uaccess.h at all means
> I got something horribly wrong in my use of MAKE_MM_SEG() to
> set the segments.

Oh, that's straightforward enough: the m68k segment code uses the same
'mm_segment_t' type that the geberic kernel just wants to make empty
for the non-CONFIG_SET_FS case.

So the m68k segment type should just use a different name.

In fact, the m68k manuals call them "alternate function code
registers", so maybe it shouldn't be about "mm_segment_t" at all, but
the type should really be named as such.

The "segment" naming is just legacy x86 nomenclature, so I suspect it
would be a good thing to really walk away from that and make it very
explicit about what this is on m68k.

So maybe it could be

    typedef struct {
        unsigned char val:3;
    } function_code_reg_t;

or something like that.

I suspect some real m68k person should make up the name, not me. The
alternate function code registers are all from after my time ;)

              Linus