[PATCH v2] arm: fix memset-related crashes caused by recent GCC (4.7.2) optimizations

[Ivan Djelic]

On Sun, Mar 10, 2013 at 05:28:54PM +0000, Russell King - ARM Linux wrote:
> On Sun, Mar 10, 2013 at 06:06:11PM +0100, Alexander Holler wrote:
> > Am 07.03.2013 16:17, schrieb Russell King - ARM Linux:
> >> On Wed, Mar 06, 2013 at 08:15:17PM +0100, Dirk Behme wrote:
> >>> Am 11.02.2013 13:57, schrieb Ivan Djelic:
> >>>> Recent GCC versions (e.g. GCC-4.7.2) perform optimizations based on
> >>>> assumptions about the implementation of memset and similar functions.
> >>>> The current ARM optimized memset code does not return the value of
> >>>> its first argument, as is usually expected from standard implementations.
> >
> > I've just tried this patch with kernel 4.8.2 on an armv5-system where I  
> > use gcc 4.7.2 since several months and where most parts of the system  
> > are compiled with gcc 4.7.2 too.
> >
> > And I had at least one problem which manifested itself with
> 
> Yes, the patch _is_ wrong.  Reverted.  I was trusting Nicolas' review
> of it, but the patch is definitely wrong.  Look carefully at this
> fragment of code:
> 
> 1:      subs    r2, r2, #4              @ 1 do we have enough
>         blt     5f                      @ 1 bytes to align with?
>         cmp     r3, #2                  @ 1
>         strltb  r1, [ip], #1            @ 1
>         strleb  r1, [ip], #1            @ 1
>         strb    r1, [ip], #1            @ 1
>         add     r2, r2, r3              @ 1 (r2 = r2 - (4 - r3))
> /*
>  * The pointer is now aligned and the length is adjusted.  Try doing the
>  * memset again.
>  */
> 
> ENTRY(memset)
> /*
>  * Preserve the contents of r0 for the return value.
>  */
>         mov     ip, r0
>         ands    r3, ip, #3              @ 1 unaligned?
>         bne     1b                      @ 1
> 
> and consider what happens when 'r0' is not aligned to a word... We end
> up aligning the pointer in "1:" and then fall through into memset again
> which reloads the old misaligned pointer.

Oops... Indeed. Thanks very much for catching that and sorry for the regression.
--
Ivan.