Attachment 874080 Details for Bug 1069028 – ixgbevf vlan stripping fix when not in netpoll path (v4)

[patch] ixgbevf vlan stripping fix when not in netpoll path (v4)

0001-ixgbevf-fix-vlan-acceleration.patch (text/plain), 2.82 KB, created by Nikolay Aleksandrov on 2014-03-13 17:07:59 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Nikolay Aleksandrov

Created: 2014-03-13 17:07:59 UTC

Size: 2.82 KB

patch

obsolete

>From 03624c2c217b25e7fd61eda43beb532ab44083e6 Mon Sep 17 00:00:00 2001
>From: Nikolay Aleksandrov <nikolay@redhat.com>
>Date: Tue, 11 Mar 2014 12:24:25 +0100
>Subject: [RHEL6.6 PATCH BZ 1069028 v2] ixgbevf: fix vlan acceleration
>
>BZ - https://bugzilla.redhat.com/show_bug.cgi?id=1069028
>
>Description:
>Currently if we're not in netpoll the stripped VLAN information will get
>lost resulting in connectivity problems if there're VLANs configured on
>top. Fix this to match upstream and add the VLAN information to the skb
>no matter which path will get chosen.
>Also make it closer to upstream by checking for the existence of an exact
>VLAN, another reason for checking for the exact VLAN is that if the VF
>is in a hypervisor VLAN and a different VLAN gets created on top of it,
>the original (hypervisor) VLAN will stop working.
>
>Note: Currently the IN_NETPOLL flag is always set, so that path is
>always chosen, but this patch also adds exact VLAN check on the other path
>in order to be consistent.
>
>v2: Use rcu_read_lock and check if the vlan is present inside of course it
>may disappear afterwards but that's not a problem, the packet will be
>handled correctly. Also mask out the bits of tag (vlan_tci really).
>
>Upstream:
> The patch is RHEL-specific, it makes the driver act as in upstream.
>
>Brew - https://brewweb.devel.redhat.com/taskinfo?taskID=7192289
>
>Tests:
>I have tested it locally by creating the same VLAN in the VM and in the
>host and checking for connectivity. The reporter tested it as can be seen
>here - https://bugzilla.redhat.com/show_bug.cgi?id=1069028#c26 and has also
>passed it to the client for testing.
>
>Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
>---
> drivers/net/ixgbevf/ixgbevf_main.c | 19 +++++++++++++++----
> 1 file changed, 15 insertions(+), 4 deletions(-)
>
>diff --git a/drivers/net/ixgbevf/ixgbevf_main.c b/drivers/net/ixgbevf/ixgbevf_main.c
>index 56b394621814..e87f8cd9a17d 100644
>--- a/drivers/net/ixgbevf/ixgbevf_main.c
>+++ b/drivers/net/ixgbevf/ixgbevf_main.c
>@@ -285,18 +285,29 @@ static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
> 				union ixgbe_adv_rx_desc *rx_desc)
> {
> 	struct ixgbevf_adapter *adapter = q_vector->adapter;
>-	bool is_vlan = (status & IXGBE_RXD_STAT_VP);
>+	bool is_vlan = (status & IXGBE_RXD_STAT_VP), used_vlan = false;
> 	u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
> 
>+	if (is_vlan) {
>+		rcu_read_lock();
>+		if (vlan_group_get_device(adapter->vlgrp, tag & VLAN_VID_MASK))
>+			used_vlan = true;
>+		rcu_read_unlock();
>+	}
>+
> 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
>-		if (adapter->vlgrp && is_vlan)
>+		if (used_vlan)
> 			vlan_gro_receive(&q_vector->napi,
> 					 adapter->vlgrp,
> 					 tag, skb);
> 		else
> 			napi_gro_receive(&q_vector->napi, skb);
>-	} else
>-		netif_rx(skb);
>+	} else {
>+		if (used_vlan)
>+			vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
>+		else
>+			netif_rx(skb);
>+	}
> }
> 
> /**
>-- 
>1.8.5.3
>

Actions: View | Diff

Attachments on bug 1069028: 869940 | 870927 | 873096 | 873098 | 873936 | 873949 | 874080