On Tuesday 18 January 2011 19:11:01 Lars-Peter Clausen wrote:

Currently io{read,write}{16,32} expand to
        *addr = cpu_to_leXX(cpu_to_beXX(val))
and
        val = beXX_to_cpu(leXX_to_cpu(*addr))

While it should rather be:
        *addr = cpu_to_be{16,32}(val)
and
        val = be{16,32}_to_cpu(*addr)

The current implementation works on litte-endian targets(where cpu_to_leXX is a
noop), but breaks on on big-endian targets, this patch fixes it.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>

The existing code looks broken on big-endian indeed, thanks
for the report!

However, you cannot use __raw_readl in general, because that
function knows nothing about the various kinds of I/O accesses
(potentially) handled by readl. Think of architectures where
readl is not a pointer dereference but something else.

The right solution is probably to use swab16/swab32 for the
big-endian functions. This also corrects the iowrite functions
which really should be using cpu_to_be32 instead of be32_to_cpu
(although they are always defined to be the same afaict.

	Arnd