On Tue, 18 Aug 2009 15:35:22 +0100
Andrew Haley <aph@xxxxxxxxxx> wrote:
> > void orion_nand_read_buf(uint8_t *buf, int len)
> > {
> > void *io_base = vobb;
> > uint64_t *buf64;
> > int i = 0;
> >
> > buf64 = (uint64_t *)buf;
> > while (i < len/8) {
> > uint64_t x;
> > asm ("ldrd\t%0, [%1]" : "=r" (x) : "r" (io_base));
>
> This is wrong. It tells gcc that an address is input to the asm and
> a uint64_t is returned. But it doesn't tell gcc that some memory
> at io_base is being used. So, it's legitimate to hoist the asm out
> of the loop.
>
> Add a memory clobber and an earlyclobber:
>
> asm ("ldrd\t%0, [%1]" : "=&r" (x) : "r" (io_base));
Thanks, changing it to
asm ("ldrd\t%0, [%1]" : "=&r" (x) : "r" (io_base) : "memory");
fixes the issue. I'll re-submit the patch to the kernel mailing list.
> > Error: destination register must be even -- `ldrd r1,[r2]'
> >
> > when compiling this code.
>
> I think that is a bug.
>
> > So is this a GCC bug or should I head back to the Linux people with it?
>
> No-one here is going to fix bugs in gcc 3.4.4. I think you'll need to
> use an explicit register variable to work round it.
Sorry, I was referring to the Linux code above, not the GCC 3.4.4 issue
I got.
// Simon
