Re: [BUG] bnx2 + vlan + TSO : doesnt work

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mardi 18 janvier 2011 à 01:32 -0500, Jesse Gross a écrit :
> On Tue, Jan 18, 2011 at 1:21 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > Le lundi 17 janvier 2011 à 22:09 -0800, David Miller a écrit :
> >> From: Jesse Gross <jesse@nicira.com>
> >> Date: Mon, 17 Jan 2011 16:13:18 -0800
> >>
> >> > I think it is better for netif_skb_features() to actually return the
> >> > correct features rather than bypass it here.  NETIF_F_HW_VLAN_TX
> >> > doesn't depend on any other offloads, so we can always include it if
> >> > it is in dev->features.
> >> >
> >> > Separately, this means there is a problem with bnx2 because it allows
> >> > vlan insertion to be turned off, which would have the same effect.
> >> > Maybe it is looking directly at skb->protocol or similar for TSO.
> >>
> >> Please, someone cons up an acceptable fix fast.
> >>
> >
> > I just woke up, and honestly dont understand why only bnx2 is affected
> > by this problem of masking NETIF_F_HW_VLAN_TX
> 
> If NETIF_F_HW_VLAN_TX is masked then the tag will be inserted in
> software, which is generally OK.  The problem is that some drivers
> assume that if they can do vlan tagging in hardware it will always be
> used and therefore don't expect vlan tags when setting up TSO, etc.
> 

OK, but then this driver gave us the hint core network was actually
always doing software vlan, tagging ;)

> >
> > And I dont understand all this netif_skb_features() stuff : if we want
> > to actually test dev->features & NETIF_F_HW_VLAN_TX, and this flag
> > doesnt depend on other offloads, why are we doing features &
> > vlan_features.
> 
> The idea is to put all of the logic in one place rather than having
> pieces that are really interdependent scattered around in the
> different offloads.  So we could test dev->features directly for vlans
> but I would rather just have netif_skb_features() return the right
> values to start off with.
> 
> >
> > Jesse, I dont understand why you say "bnx2 allows vlan insertion to be
> > turned off". Really.
> 
> You can disable it using Ethtool, which will turn off
> NETIF_F_HW_VLAN_TX the same as this bug.
> 
> I'm running a quick test on a patch to always allow NETIF_F_HW_VLAN_TX
> to be returned from netif_skb_features().

Thanks