On Thu Apr 22 2021, Maciej Fijalkowski wrote:

On Thu, Apr 22, 2021 at 07:26:17AM +0200, Kurt Kanzenbach wrote:

quoted

+		/* pull rx packet timestamp if available and valid */
+		if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
+			timestamp = igb_ptp_rx_pktstamp(rx_ring->q_vector,
+							pktbuf);
+
+			if (timestamp) {
+				pkt_offset += IGB_TS_HDR_LEN;
+				size -= IGB_TS_HDR_LEN;
+			}
+		}

Small nit: since this is a hot path, maybe we could omit the additional
branch that you're introducing above and make igb_ptp_rx_pktstamp() to
return either 0 for error cases and IGB_TS_HDR_LEN if timestamp was fine?
timestamp itself would be passed as an arg.

So:
		if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
			ts_offset = igb_ptp_rx_pktstamp(rx_ring->q_vector,
							pktbuf, &timestamp);
			pkt_offset += ts_offset;
			size -= ts_offset;
		}

Thoughts? I feel like if we see that desc has timestamp enabled then let's
optimize it for successful case.

Yes, this should work as well. Actually I didn't like the if statement
either. Only one comment: It's not an offset but rather the timestamp
header length. I'd call it 'ts_len'.

quoted

 
 		/* retrieve a buffer from the ring */
 		if (!skb) {
-			unsigned int offset = igb_rx_offset(rx_ring);
-			unsigned char *hard_start;
+			unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
+			unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);

Probably we could do something similar in flavour of:
https://lore.kernel.org/bpf/20210118151318.12324-10-maciej.fijalkowski@intel.com/ (local)

which broke XDP_REDIRECT and got fixed in:
https://lore.kernel.org/bpf/20210303153928.11764-2-maciej.fijalkowski@intel.com/ (local)

You get the idea.

Yes, I do. However, I think such a change doesn't belong in this patch,
which is a bugfix for XDP. It looks like an optimization. Should I split
it into two patches and rather target net-next instead of net?

Thanks for your review.

Thanks,
Kurt