Re: [PATCH net-next v2 11/20] net: packetengines: slight optimization of addr

[Joe Perches <joe@perches.com>]

On Sat, 2013-12-28 at 23:18 +0800, Ding Tianhong wrote:
> 于 2013/12/28 21:58, Sergei Shtylyov 写道:
> > Hello.
> > 
> > On 28-12-2013 10:17, Ding Tianhong wrote:
> > 
> >> Use possibly more efficient ether_addr_equal
> >> to instead of memcmp.
> > 
> >> Cc: "David S. Miller" <davem@davemloft.net>
> >> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
> >> ---
> >>   drivers/net/ethernet/packetengines/yellowfin.c |   12 ++++++------
> >>   1 files changed, 6 insertions(+), 6 deletions(-)
> > 
> >> diff --git a/drivers/net/ethernet/packetengines/yellowfin.c b/drivers/net/ethernet/packetengines/yellowfin.c
> >> index d28593b..b83ac0e 100644
> >> --- a/drivers/net/ethernet/packetengines/yellowfin.c
> >> +++ b/drivers/net/ethernet/packetengines/yellowfin.c
> >> @@ -1097,12 +1097,12 @@ static int yellowfin_rx(struct net_device *dev)
> >>               if (status2 & 0x80) dev->stats.rx_dropped++;
> >>   #ifdef YF_PROTOTYPE        /* Support for prototype hardware errata. */
> >>           } else if ((yp->flags & HasMACAddrBug)  &&
> >> -            memcmp(le32_to_cpu(yp->rx_ring_dma +
> >> -                entry*sizeof(struct yellowfin_desc)),
> >> -                dev->dev_addr, 6) != 0 &&
> >> -            memcmp(le32_to_cpu(yp->rx_ring_dma +
> >> -                entry*sizeof(struct yellowfin_desc)),
> >> -                "\377\377\377\377\377\377", 6) != 0) {
> >> +            !ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
> >> +                              entry * sizeof(struct yellowfin_desc)),
> >> +                              dev->dev_addr) &&
> > 
> >    Previous line was aligned correctly, the above line should start under le32_to_cpu.
> > 
> >> +            !ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
> >> +                              entry * sizeof(struct yellowfin_desc)),
> > 
> >    Start the continuation lines under 'yp', please.
> > 
> >> +                              "\377\377\377\377\377\377")) {
> > 
> >    This line should start under le32_to_cpu.
> > 
> > WBR, Sergei
> > 
> 
> Hi sergei:
> you mean this way?
> 	!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
> 				      entry * sizeof(struct yellowfin_desc)),
> 			  dev->dev_addr) &&
> 
> 	!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
> 				      entry * sizeof(struct yellowfin_desc)),
> 			  "\377\377\377\377\377\377")) {

Does this really matter?
Does anyone have a packetengine NIC anymore?

\377 octal is 0xff, so this is matching a broadcast address.
is_broadcast_ether_addr(addr) would be appropriate.

So would using a temporary address.

u8 *addr = (u8 *)(unsigned long)le32_to_cpu(etc)

but the whole thing looks very suspect as an le32
value could not be added to correctly on a
big-endian arch anyway.

My guess is this was tested only on an x86 and
it should be:

	u8 *addr = (u8 *)(unsigned long)(le32_to_cpu(yp->rx_ring_dma) +
					 entry * sizeof(struct yellowfin_desc));

It maybe better just to leave these two alone.