6.8-stable review patch. If anyone has any objections, please let me know.
------------------
From: Eric Dumazet edumazet@google.com
[ Upstream commit 581073f626e387d3e7eed55c48c8495584ead7ba ]
trafgen performance considerably sank on hosts with many cores after the blamed commit.
packet_read_pending() is very expensive, and calling it in af_packet fast path defeats Daniel intent in commit b013840810c2 ("packet: use percpu mmap tx frame pending refcount")
tpacket_destruct_skb() makes room for one packet, we can immediately wakeup a producer, no need to completely drain the tx ring.
Fixes: 89ed5b519004 ("af_packet: Block execution of tasks waiting for transmit to complete in AF_PACKET") Signed-off-by: Eric Dumazet edumazet@google.com Cc: Neil Horman nhorman@tuxdriver.com Cc: Daniel Borkmann daniel@iogearbox.net Reviewed-by: Willem de Bruijn willemb@google.com Link: https://lore.kernel.org/r/20240515163358.4105915-1-edumazet@google.com Signed-off-by: Jakub Kicinski kuba@kernel.org Signed-off-by: Sasha Levin sashal@kernel.org --- net/packet/af_packet.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c index e6a8701a38dbe..91c9dc0108a2d 100644 --- a/net/packet/af_packet.c +++ b/net/packet/af_packet.c @@ -2522,8 +2522,7 @@ static void tpacket_destruct_skb(struct sk_buff *skb) ts = __packet_set_timestamp(po, ph, skb); __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
- if (!packet_read_pending(&po->tx_ring)) - complete(&po->skb_completion); + complete(&po->skb_completion); }
sock_wfree(skb);