Re: [PATCH 3/3] offloading: Force software GSO for multiple vlan tags.

[Jesse Gross <jesse@nicira.com>]

2011/1/7 Michał Mirosław <mirqus@gmail.com>:
> Hi,
>
> Sorry for late reply, I noticed this patch only after it went to Linus' tree.
>
> 2010/10/30 Jesse Gross <jesse@nicira.com>:
>> We currently use vlan_features to check for TSO support if there is
>> a vlan tag.  However, it's quite likely that the NIC is not able to
>> do TSO when there is an arbitrary number of tags.  Therefore if there
>> is more than one tag (in-band or out-of-band), fall back to software
>> emulation.
>>
>> Signed-off-by: Jesse Gross <jesse@nicira.com>
>> CC: Ben Hutchings <bhutchings@solarflare.com>
>> ---
>>  include/linux/netdevice.h |    7 +++----
>>  net/core/dev.c            |   16 ++++++++++++++++
>>  2 files changed, 19 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
>> index 072652d..980c752 100644
>> --- a/include/linux/netdevice.h
>> +++ b/include/linux/netdevice.h
>> @@ -2234,6 +2234,8 @@ unsigned long netdev_fix_features(unsigned long features, const char *name);
>>  void netif_stacked_transfer_operstate(const struct net_device *rootdev,
>>                                        struct net_device *dev);
>>
>> +int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev);
>> +
>>  static inline int net_gso_ok(int features, int gso_type)
>>  {
>>        int feature = gso_type << NETIF_F_GSO_SHIFT;
>> @@ -2249,10 +2251,7 @@ static inline int skb_gso_ok(struct sk_buff *skb, int features)
>>  static inline int netif_needs_gso(struct net_device *dev, struct sk_buff *skb)
>>  {
>>        if (skb_is_gso(skb)) {
>> -               int features = dev->features;
>> -
>> -               if (skb->protocol == htons(ETH_P_8021Q) || skb->vlan_tci)
>> -                       features &= dev->vlan_features;
>> +               int features = netif_get_vlan_features(skb, dev);
>>
>>                return (!skb_gso_ok(skb, features) ||
>>                        unlikely(skb->ip_summed != CHECKSUM_PARTIAL));
>> diff --git a/net/core/dev.c b/net/core/dev.c
>> index 8bdda70..8d74988 100644
>> --- a/net/core/dev.c
>> +++ b/net/core/dev.c
>> @@ -1969,6 +1969,22 @@ static inline void skb_orphan_try(struct sk_buff *skb)
>>        }
>>  }
>>
>> +int netif_get_vlan_features(struct sk_buff *skb, struct net_device *dev)
>> +{
>> +       __be16 protocol = skb->protocol;
>> +
>> +       if (protocol == htons(ETH_P_8021Q)) {
>> +               struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
>> +               protocol = veh->h_vlan_encapsulated_proto;
>> +       } else if (!skb->vlan_tci)
>> +               return dev->features;
>> +
>> +       if (protocol != htons(ETH_P_8021Q))
>> +               return dev->features & dev->vlan_features;
>> +       else
>> +               return 0;
>> +}
>
> This clears all features for multiply-tagged frames. At least SG,
> FRAGLIST and HW_CSUM are perfectly valid for those frames.
>
> This doesn't really matter if this function stays used only in
> netif_needs_gso(). It's name and placement suggests otherwise, though.

You're right and in fact I have some upcoming changes that expands the
use of this function to places that do care about these offloads.
I've generalized it as you suggest.

Thanks.