> > Another thing I notice was that fragments are handed 1 frame at a time, but > > making the device > > kick the TX-ring after each frame. Shouldn't for fragmented frames the ring > > be kicked after the > > last fragment has been given to the device? > > Good question. It should be guaranteed that no other frame is sent to > the driver in between two fragments. So you can kick the DMA only if > WLAN_FC_MOREFRAG is not set in the FC. I'm not sure if this should be > automated somehow by the stack.
Perhaps it is an idea to use a similar method as used in the ipw stack, where the xmit function is passing a list of sk_buffers. This way all fragments are passed to the device, and device can place all entries onto the buffer and kick the TX ring. IvD
pgpO6r70S98o3.pgp
Description: PGP signature
