Re: gcc 4.5.1 / as 2.20.51.0.11 miscompiling drivers/char/i8k.c ?

[Jakub Jelinek]

On Mon, Nov 15, 2010 at 07:30:35PM +0100, Jim Bos wrote:
> On 11/15/2010 07:08 PM, Linus Torvalds wrote:
> > On Mon, Nov 15, 2010 at 9:40 AM, Jim Bos <jim876@xxxxxxxxx> wrote:
> >>
> >> Hmm, that doesn't work.
> >>
> >> [ Not sure if you read to whole thread but initial workaround was to
> >> change the asm(..) to asm volatile(..) which did work. ]
> > 
> > Since I have a different gcc than yours (and I'm not going to compile
> > my own), have you posted your broken .s file anywhere? In fact, with
> > the noinline (and the removal of the "+m" thing - iow just the patch
> > you tried), what does just the "i8k_smm" function assembly look like
> > for you after you've done a "make drivers/char/i8k.s"?
> > 
> > If the asm just doesn't exist AT ALL, that's just odd. Because every
> > single call-site of i8k_smm() clearly looks at the return value. So
> > the volatile really shouldn't make any difference from that
> > standpoint. Odd.
> > 
> >                        Linus
> > 
> 
> Attached version with plain 2.6.36 source and version with the committed
> patch, i.e with the '"+m" (*regs)'

Thanks, this actually helped to see the problem.
The problem is not inside of i8k_smm, which is not inlined, but in the
callers.
ipa-pure-const.c pass thinks i8k_smm is a pure function, thus
regs = {};
regs.eax = 166;
x = i8k_smm (&regs);
if (!x) x = regs.eax;
in the callers is optimized into
regs = {}
regs.eax = 166;
x = i8k_smm (&regs);
if (!x) x = 166;
Now, not sure why this happens, as there is
    case GIMPLE_ASM:
      for (i = 0; i < gimple_asm_nclobbers (stmt); i++)
        {
          tree op = gimple_asm_clobber_op (stmt, i);
          if (simple_cst_equal(TREE_VALUE (op), memory_identifier_string) == 1)
            {
              if (dump_file)
                fprintf (dump_file, "    memory asm clobber is not const/pure");
              /* Abandon all hope, ye who enter here. */
              local->pure_const_state = IPA_NEITHER;
            }
        }
Debugging...

	Jakub
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/