vxge: improve tx performance by using mmiowb() instead of wmb()
authorBenjamin LaHaise <ben.lahaise@neterion.com>
Tue, 4 Aug 2009 10:21:21 +0000 (10:21 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 5 Aug 2009 20:10:44 +0000 (13:10 -0700)
wmb() is extremely heavy on x86.  The semantics required in the driver are
provided by mmiowb(), so use that and improve tx performance on P4 Xeons by
5-10%.

Signed-off-by: Benjamin LaHaise <ben.lahaise@neterion.com>
Signed-off-by: Sreenivasa Honnur <sreenivasa.honnur@neterion.com>
Signed-off-by: Ramkrishna Vepa <ram.vepa@neterion.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vxge/vxge-traffic.c

index c249cd25a5d0415957dffbb606bd218cda76c5b9..0efb769894ef42b01a667b589e7182b3f09a65f8 100644 (file)
@@ -1070,11 +1070,11 @@ static void __vxge_hw_non_offload_db_post(struct __vxge_hw_fifo *fifo,
                VXGE_HW_NODBW_GET_NO_SNOOP(no_snoop),
                &fifo->nofl_db->control_0);
 
-       wmb();
+       mmiowb();
 
        writeq(txdl_ptr, &fifo->nofl_db->txdl_ptr);
-       wmb();
 
+       mmiowb();
 }
 
 /**