Re: [PATCH] m68k: Use generic strncpy_from_user(), strlen_user(), and strnlen_user()

[Philippe De Muyter <phdm@macqel.be>]

On Wed, May 30, 2012 at 01:20:02PM +0200, Geert Uytterhoeven wrote:
> On Wed, May 30, 2012 at 12:22 PM, Philippe De Muyter <phdm@macqel.be> wrote:
> > On Tue, May 29, 2012 at 11:33:36PM +0200, Geert Uytterhoeven wrote:
> >> Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
> >> ---
> >> Do we also want
> >>
> >>     select HAVE_EFFICIENT_UNALIGNED_ACCESS if (!COLDFIRE && !M68000)
> >
> > Sorry, I did not follow what happened to unaligned accesses, but
> > CPU32 family (at least 68340) crashes on unaligned accesses.
> 
> We don't seem to have CONFIG_M68340 in arch/m68k/Kconfig.cpu?

I have a local port here (but based on an ancient linux kernel, 2.6.2 IIRC)

> But Freescale's website confirms both 68340 and 68360 are CPU32.
> 
> arch/m68k/include/asm/unaligned.h assumes CPU32 (CONFIG_MCPU32)
> can do unaligned accesses:

That's not true.  Accessing a 16- or 32-bit word at an odd address
with a 68340 generates an Address Error Exception.  I remember
discovering a bug in the ppp kernel code because of that.

> 
> #if defined(CONFIG_COLDFIRE) || defined(CONFIG_M68000)
> #include <linux/unaligned/be_struct.h>
> #include <linux/unaligned/le_byteshift.h>
> #include <linux/unaligned/generic.h>
> 
> #define get_unaligned   __get_unaligned_be
> #define put_unaligned   __put_unaligned_be
> 
> #else
> /*
>  * The m68k can do unaligned accesses itself.
>  */
> #include <linux/unaligned/access_ok.h>
> #include <linux/unaligned/generic.h>
> 
> #define get_unaligned   __get_unaligned_be
> #define put_unaligned   __put_unaligned_be
> 
> #endif
> 
> Is this wrong?

I can't tell from reading just the lines above, but I think one should add
"|| defined(CONFIG_MCPU32)" at the end of the if condition.

I also think that the Coldfire 5272 can do unaligned accesses, but I
cannot test that at the moment.

> 
> However, for CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS,
> the question is not whether unaligned accesses are supported, but
> whether they are more efficient than byte copies when copying larger blocks.

OK, thanks

Philippe

-- 
Philippe De Muyter +32 2 6101532 Macq SA rue de l'Aeronef 2 B-1140 Bruxelles