On 01/27/2017 01:33 PM, John Fastabend wrote:
This is an experimental implementation of rx zero copy for af_packet.
Its a bit rough and likely has errors but the plan is to clean it up
over the next few months.
And seeing I said I would post it in another thread a few days back
here it is.
This sounds good (believe me I have been thinking along the lines :)
From driver Rx side, we always premap RX buffers so best to map them to
shmem for PF_PACKET sockets.
Also, I like the idea that user can put selected queue (may be queues in
future?) to PF_PACKET mode keeping rest of the queues as it is.
Zero copy and removing skb setup & processing overhead on RX certainly
makes things faster and help latency. Zero copy is good on Tx however
without skb should we figure out how to use segmentation and checksum
offloading features of HW. Can this be considered in tpacket V4 hdr!
-Tushar
