On Thu, 2013-08-01 at 21:58 +0100, Russell King - ARM Linux wrote:

On Thu, Aug 01, 2013 at 01:55:28PM -0700, Joe Perches wrote:

quoted

On Thu, 2013-08-01 at 21:33 +0100, Russell King - ARM Linux wrote:

quoted

On Thu, Aug 01, 2013 at 01:31:09PM -0700, Olof Johansson wrote:

quoted

On Thu, Aug 1, 2013 at 1:14 PM, Joe Perches [off-list ref] wrote:

quoted

diff --git a/drivers/net/ethernet/pasemi/pasemi_mac.h b/drivers/net/ethernet/pasemi/pasemi_mac.h

[]

quoted

quoted

quoted

@@ -79,11 +79,11 @@ struct pasemi_mac {
        int             last_cs;
        int             num_cs;
        u32             dma_if;
-       u8              type;
 #define MAC_TYPE_GMAC  1
 #define MAC_TYPE_XAUI  2

-       u8              mac_addr[6];
+       u8              mac_addr[ETH_ALEN];
+       u8              type;

Just promote 'type' to u32 instead, saves you from moving the #defines
down, etc, etc.

type is already u8, why change it?
That would also change struct size.

quoted

There's a more fundamental question which has to be asked: what is the
point of moving that in the first place?

Some is_<foo>_ether_addr tests assume __aligned(2)
by a casting char * to u16/be16 * and dereferencing.

see patch 0/3 and include/linux/etherdevice.h

This seems rather obscure - I mean, it's not obvious to driver authors
that should be the case.  Would it not be better to make this a little
more obvious somehow?  Maybe __aligned(2) against mac_addr?  Or
maybe have a debugging check for it?

That'd be for David Miller (cc'd).

I believe he's argued in the past that any alignment check
for mac addresses was unnecessary.

For all I know it really might not matter because pasemi
can successfully dereference a ushort against an odd char
pointer.

I just noticed it and thought it'd be better moved.