From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755014AbdDLSrq (ORCPT ); Wed, 12 Apr 2017 14:47:46 -0400 Received: from a2nlsmtp01-03.prod.iad2.secureserver.net ([198.71.225.37]:36790 "EHLO a2nlsmtp01-03.prod.iad2.secureserver.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754721AbdDLSrn (ORCPT ); Wed, 12 Apr 2017 14:47:43 -0400 x-originating-ip: 107.180.71.197 From: Haiyang Zhang To: davem@davemloft.net, netdev@vger.kernel.org Cc: haiyangz@microsoft.com, kys@microsoft.com, olaf@aepfle.de, vkuznets@redhat.com, linux-kernel@vger.kernel.org Subject: [PATCH net-next,2/2] hv_netvsc: Exclude non-TCP port numbers from vRSS hashing Date: Wed, 12 Apr 2017 11:35:06 -0700 Message-Id: <1492022106-4070-2-git-send-email-haiyangz@exchange.microsoft.com> X-Mailer: git-send-email 1.7.1 In-Reply-To: <1492022106-4070-1-git-send-email-haiyangz@exchange.microsoft.com> References: <1492022106-4070-1-git-send-email-haiyangz@exchange.microsoft.com> Reply-To: haiyangz@microsoft.com X-CMAE-Envelope: MS4wfFtDUzQmLA8NNpwU0wCJP/GuoGmC/gDU1PKLVfBeklXUQeAgVCv45c6a0J7Xs6Pof4druvpDSmN9Atk+X/Mp9FcdAtOIZnrKAbBkrDcI0MDPKGkFoyGx wClmJnjNC4vZgBuR/ORyPzWSre+aLwE1c4sXuh8Ch2dTUEe/XI/xEgW3uS9bgd+m7WJyBNr2KXyUysc9lY3ifsA/mqOAwGgX9sXhq4tJ9BDns2/LGghe2ZuA 9k97jNhs+xogyRTTo6MuEQbvOsUlf+PVk8Jn4J+dWKgecj831d0Vovdy5vsNdMiq1IPHx8hSa8o0YHTT9asRontDCvJHTliezD1siYkIhEF7D9c1weVNzUMK NUAfVpQUqR16hxUoo74z7lcQQqV3lA== Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org From: Haiyang Zhang Azure hosts are not supporting non-TCP port numbers in vRSS hashing for now. For example, UDP packet loss rate will be high if port numbers are also included in vRSS hash. So, we created this patch to use only IP numbers for hashing in non-TCP traffic. Signed-off-by: Haiyang Zhang Reviewed-by: Stephen Hemminger --- drivers/net/hyperv/netvsc_drv.c | 32 +++++++++++++++++++++++++++++++- 1 files changed, 31 insertions(+), 1 deletions(-) diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c index e813eaf..51fa903 100644 --- a/drivers/net/hyperv/netvsc_drv.c +++ b/drivers/net/hyperv/netvsc_drv.c @@ -191,6 +191,36 @@ static int netvsc_close(struct net_device *net) return ppi; } +/* Azure hosts don't support non-TCP port numbers in hashing yet. We compute + * hash for non-TCP traffic with only IP numbers. + */ +static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk) +{ + struct flow_keys flow; + u32 hash; + static u32 hashrnd __read_mostly; + + net_get_random_once(&hashrnd, sizeof(hashrnd)); + + if (!skb_flow_dissect_flow_keys(skb, &flow, 0)) + return 0; + + if (flow.basic.ip_proto == IPPROTO_TCP) { + return skb_get_hash(skb); + } else { + if (flow.basic.n_proto == htons(ETH_P_IP)) + hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd); + else if (flow.basic.n_proto == htons(ETH_P_IPV6)) + hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd); + else + hash = 0; + + skb_set_hash(skb, hash, PKT_HASH_TYPE_L3); + } + + return hash; +} + static inline int netvsc_get_tx_queue(struct net_device *ndev, struct sk_buff *skb, int old_idx) { @@ -198,7 +228,7 @@ static inline int netvsc_get_tx_queue(struct net_device *ndev, struct sock *sk = skb->sk; int q_idx; - q_idx = ndc->tx_send_table[skb_get_hash(skb) & + q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) & (VRSS_SEND_TAB_SIZE - 1)]; /* If queue index changed record the new value */ -- 1.7.1