Re: [PATCH] asm-generic: architecture independent readq/writeq for32bit environment

[Geert Uytterhoeven]

On Wed, Feb 8, 2012 at 17:28, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, Feb 8, 2012 at 8:14 AM, Hitoshi Mitake <h.mitake@xxxxxxxxx> wrote:
>> On Tue, Feb 7, 2012 at 12:56, Linus Torvalds
>> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>>>
>>> On Feb 6, 2012 6:47 PM, "Hitoshi Mitake" <h.mitake@xxxxxxxxx> wrote:
>>>>
>>>>Â All of them are endian aware
>>>
>>> I think that part is wrong.
>>>
>>> Pci is little-endian, so readl and writel are always already little-endian.
>>> Trying to make readq be endian-aware is wrong.
>>
>> Is every memory area which can be read/written by read[wl]/write[wl]
>> always little-endian?
>>
>> If so, of course I have to eliminate the big-endian version.
>> But I'm not sure about this point because the new readq/writeq is
>> in asm-generic which is used by every archs, so I'd like to ask you about it.
>
> Yes, PCI is always little-endian everywhere.
>
> Now, some architectures may have some unified IO space where parts of
> it is PCI, and other parts are programmed the same way but is for some
> other bus, but the PCI part will always be little-endian because
> otherwise no common driver would ever work. But those non-PCI things
> would never be relevant for an emulated readq/writeq anyway, and would
> need some bus-specific accessor functions.

IIRC, some crazy hardware manufacturers did swizzle PCI busses in hardware.
So doing a PCI "little endian" 32-bit read must be performed using a native
(big endian) 32-bit read. And it causes more headaches for 16-bit accesses.

Of course this must all be hidden in your readX() implementations, to make the
common drivers work.

Gr{oetje,eeting}s,

            Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
             Â Â -- Linus Torvalds
--
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/