Hi,
did anyone have a look at the sgiseeq.c driver ? It seems it is
very suboptimal to what the hardware is capable to do. From just a quick
glance we are copying the packets payload from the skb to KSEG1
preallocated tx-rings and then run the dma instead of just dmaing from
the skbuff we just got handed (After a dma_cache_wback_inv for the
skbuf).
On receive we basically do the same the other way round. We copy the
packets from KSEG1 to the "final" skb.
From my perspective this is ok when you only have "cache_flush_all"
but with the finer granularity of dma_cache_wback_inv or dma_cache_inv
we would be more happy to just dma from the skbuff shouldnt we ?
Flo
--
Florian Lohoff flo@rfc822.org +49-5201-669912
Nine nineth on september the 9th Welcome to the new billenium
PGP signature
