Re: Linux 5.19-rc8

[Yury Norov <yury.norov@gmail.com>]

On Tue, Jul 26, 2022 at 10:12:21AM +0100, Russell King (Oracle) wrote:
> On Mon, Jul 25, 2022 at 10:55:18AM -0700, Linus Torvalds wrote:
> > On Mon, Jul 25, 2022 at 9:11 AM Guenter Roeck <linux@roeck-us.net> wrote:
> > >
> > > BUG: KFENCE: out-of-bounds read in _find_next_bit_le+0x10/0x48
> > 
> > Ok, I was hoping somebody more ARMy would look at this, particularly
> > since there is no call trace beyond the actual fault.
> 
> First I'm aware of it. Was it reported to linux-arm-kernel? I'm guessing
> the report wasn't Cc'd to me - I can't find anything in my mailbox about
> it.
> 
> > I think the fix might be something like this:
> > 
> >   diff --git a/arch/arm/lib/findbit.S b/arch/arm/lib/findbit.S
> >   index b5e8b9ae4c7d..b36ca301892e 100644
> >   --- a/arch/arm/lib/findbit.S
> >   +++ b/arch/arm/lib/findbit.S
> >   @@ -83,6 +83,8 @@ ENDPROC(_find_first_bit_le)
> >    ENTRY(_find_next_bit_le)
> >                 teq     r1, #0
> >                 beq     3b
> >   +             cmp     r2, r1
> >   +             bhs     3b
> >                 ands    ip, r2, #7
> >                 beq     1b                      @ If new byte, goto old routine
> >     ARM(                ldrb    r3, [r0, r2, lsr #3]    )
> > 
> > but my ARM asm is so broken that the above is just really random noise
> > that may or may not build - much less work.
> > 
> > I'll leave it to Russell &co to have a tested and working patch.
> 
> I think it needs a bit more than that, but as you point out in later
> emails, the compiler may do a better job for this.
> 
> One of the reasons for using byte loads was to avoid problems in the
> early days of Linux where these took void pointers and thus could be
> misaligned - and using word accesses would have resulted in much
> pain. However, that was changed to unsigned long pointers back in
> 2017, so in theory that should no longer be a concern.
> 
> I don't remember why we used void pointers there originally - that's
> something which dates back to the 1990s.

OK, then I'm sending the patch that switches it to generic code.

Thanks,
Yury