| Message ID | 20250124085744.434869-1-cratiu@nvidia.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [net,v2] bonding: Correctly support GSO ESP offload | expand |
On Fri, Jan 24, 2025 at 10:57:44AM +0200, Cosmin Ratiu wrote: > --- > drivers/net/bonding/bond_main.c | 19 ++++++++++--------- > 1 file changed, 10 insertions(+), 9 deletions(-) > > diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c > index 7b78c2bada81..e45bba240cbc 100644 > --- a/drivers/net/bonding/bond_main.c > +++ b/drivers/net/bonding/bond_main.c > @@ -1538,17 +1538,20 @@ static netdev_features_t bond_fix_features(struct net_device *dev, > NETIF_F_HIGHDMA | NETIF_F_LRO) > > #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > - NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) > + NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE | \ > + NETIF_F_GSO_PARTIAL) > > #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > NETIF_F_GSO_SOFTWARE) > > +#define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP) > + > > static void bond_compute_features(struct bonding *bond) > { > + netdev_features_t gso_partial_features = BOND_GSO_PARTIAL_FEATURES; > unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | > IFF_XMIT_DST_RELEASE_PERM; > - netdev_features_t gso_partial_features = NETIF_F_GSO_ESP; > netdev_features_t vlan_features = BOND_VLAN_FEATURES; > netdev_features_t enc_features = BOND_ENC_FEATURES; > #ifdef CONFIG_XFRM_OFFLOAD > @@ -1582,8 +1585,9 @@ static void bond_compute_features(struct bonding *bond) > BOND_XFRM_FEATURES); > #endif /* CONFIG_XFRM_OFFLOAD */ > > - if (slave->dev->hw_enc_features & NETIF_F_GSO_PARTIAL) > - gso_partial_features &= slave->dev->gso_partial_features; > + gso_partial_features = netdev_increment_features(gso_partial_features, > + slave->dev->gso_partial_features, > + BOND_GSO_PARTIAL_FEATURES); > > mpls_features = netdev_increment_features(mpls_features, > slave->dev->mpls_features, > @@ -1598,12 +1602,8 @@ static void bond_compute_features(struct bonding *bond) > } > bond_dev->hard_header_len = max_hard_header_len; > > - if (gso_partial_features & NETIF_F_GSO_ESP) > - bond_dev->gso_partial_features |= NETIF_F_GSO_ESP; > - else > - bond_dev->gso_partial_features &= ~NETIF_F_GSO_ESP; > - > done: > + bond_dev->gso_partial_features = gso_partial_features; > bond_dev->vlan_features = vlan_features; > bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | > NETIF_F_HW_VLAN_CTAG_TX | if (!bond_has_slaves(bond)) goto done; If there is no slaves, should we add the gso_partial_features? Thanks Hangbin > @@ -6046,6 +6046,7 @@ void bond_setup(struct net_device *bond_dev) > bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; > bond_dev->features |= bond_dev->hw_features; > bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; > + bond_dev->features |= NETIF_F_GSO_PARTIAL; > #ifdef CONFIG_XFRM_OFFLOAD > bond_dev->hw_features |= BOND_XFRM_FEATURES; > /* Only enable XFRM features if this is an active-backup config */ > -- > 2.45.0 >
On Fri, 2025-01-24 at 09:54 +0000, Hangbin Liu wrote: > On Fri, Jan 24, 2025 at 10:57:44AM +0200, Cosmin Ratiu wrote: > > --- > > drivers/net/bonding/bond_main.c | 19 ++++++++++--------- > > 1 file changed, 10 insertions(+), 9 deletions(-) > > > > diff --git a/drivers/net/bonding/bond_main.c > > b/drivers/net/bonding/bond_main.c > > index 7b78c2bada81..e45bba240cbc 100644 > > --- a/drivers/net/bonding/bond_main.c > > +++ b/drivers/net/bonding/bond_main.c > > @@ -1538,17 +1538,20 @@ static netdev_features_t > > bond_fix_features(struct net_device *dev, > > NETIF_F_HIGHDMA | NETIF_F_LRO) > > > > #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > > - NETIF_F_RXCSUM | > > NETIF_F_GSO_SOFTWARE) > > + NETIF_F_RXCSUM | > > NETIF_F_GSO_SOFTWARE | \ > > + NETIF_F_GSO_PARTIAL) > > > > #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > > NETIF_F_GSO_SOFTWARE) > > > > +#define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP) > > + > > > > static void bond_compute_features(struct bonding *bond) > > { > > + netdev_features_t gso_partial_features = > > BOND_GSO_PARTIAL_FEATURES; > > unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | > > IFF_XMIT_DST_RELEASE_PERM; > > - netdev_features_t gso_partial_features = NETIF_F_GSO_ESP; > > netdev_features_t vlan_features = BOND_VLAN_FEATURES; > > netdev_features_t enc_features = BOND_ENC_FEATURES; > > #ifdef CONFIG_XFRM_OFFLOAD > > @@ -1582,8 +1585,9 @@ static void bond_compute_features(struct > > bonding *bond) > > > > BOND_XFRM_FEATURES); > > #endif /* CONFIG_XFRM_OFFLOAD */ > > > > - if (slave->dev->hw_enc_features & > > NETIF_F_GSO_PARTIAL) > > - gso_partial_features &= slave->dev- > > >gso_partial_features; > > + gso_partial_features = > > netdev_increment_features(gso_partial_features, > > + > > slave->dev->gso_partial_features, > > + > > BOND_GSO_PARTIAL_FEATURES); > > > > mpls_features = > > netdev_increment_features(mpls_features, > > slave- > > >dev->mpls_features, > > @@ -1598,12 +1602,8 @@ static void bond_compute_features(struct > > bonding *bond) > > } > > bond_dev->hard_header_len = max_hard_header_len; > > > > - if (gso_partial_features & NETIF_F_GSO_ESP) > > - bond_dev->gso_partial_features |= NETIF_F_GSO_ESP; > > - else > > - bond_dev->gso_partial_features &= > > ~NETIF_F_GSO_ESP; > > - > > done: > > + bond_dev->gso_partial_features = gso_partial_features; > > bond_dev->vlan_features = vlan_features; > > bond_dev->hw_enc_features = enc_features | > > NETIF_F_GSO_ENCAP_ALL | > > NETIF_F_HW_VLAN_CTAG_TX | > > if (!bond_has_slaves(bond)) > goto done; > > If there is no slaves, should we add the gso_partial_features? The other partial feature sets are added after 'done:', why not do the same for gso_partial_features for consistency? 'gso_partial_features' is otherwise not set anywhere else and relies on it being set to zero when allocated in alloc_netdev_mqs. I think it's better for it to be explicitly initialized in all cases here, like the other feature sets. Cosmin.
On Fri, Jan 24, 2025 at 10:35:40AM +0000, Cosmin Ratiu wrote: > On Fri, 2025-01-24 at 09:54 +0000, Hangbin Liu wrote: > > On Fri, Jan 24, 2025 at 10:57:44AM +0200, Cosmin Ratiu wrote: > > > --- > > > drivers/net/bonding/bond_main.c | 19 ++++++++++--------- > > > 1 file changed, 10 insertions(+), 9 deletions(-) > > > > > > diff --git a/drivers/net/bonding/bond_main.c > > > b/drivers/net/bonding/bond_main.c > > > index 7b78c2bada81..e45bba240cbc 100644 > > > --- a/drivers/net/bonding/bond_main.c > > > +++ b/drivers/net/bonding/bond_main.c > > > @@ -1538,17 +1538,20 @@ static netdev_features_t > > > bond_fix_features(struct net_device *dev, > > > NETIF_F_HIGHDMA | NETIF_F_LRO) > > > > > > #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > > > - NETIF_F_RXCSUM | > > > NETIF_F_GSO_SOFTWARE) > > > + NETIF_F_RXCSUM | > > > NETIF_F_GSO_SOFTWARE | \ > > > + NETIF_F_GSO_PARTIAL) > > > > > > #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ > > > NETIF_F_GSO_SOFTWARE) > > > > > > +#define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP) > > > + > > > > > > static void bond_compute_features(struct bonding *bond) > > > { > > > + netdev_features_t gso_partial_features = > > > BOND_GSO_PARTIAL_FEATURES; > > > unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | > > > IFF_XMIT_DST_RELEASE_PERM; > > > - netdev_features_t gso_partial_features = NETIF_F_GSO_ESP; > > > netdev_features_t vlan_features = BOND_VLAN_FEATURES; > > > netdev_features_t enc_features = BOND_ENC_FEATURES; > > > #ifdef CONFIG_XFRM_OFFLOAD > > > @@ -1582,8 +1585,9 @@ static void bond_compute_features(struct > > > bonding *bond) > > > > > > BOND_XFRM_FEATURES); > > > #endif /* CONFIG_XFRM_OFFLOAD */ > > > > > > - if (slave->dev->hw_enc_features & > > > NETIF_F_GSO_PARTIAL) > > > - gso_partial_features &= slave->dev- > > > >gso_partial_features; > > > + gso_partial_features = > > > netdev_increment_features(gso_partial_features, > > > + > > > slave->dev->gso_partial_features, > > > + > > > BOND_GSO_PARTIAL_FEATURES); > > > > > > mpls_features = > > > netdev_increment_features(mpls_features, > > > slave- > > > >dev->mpls_features, > > > @@ -1598,12 +1602,8 @@ static void bond_compute_features(struct > > > bonding *bond) > > > } > > > bond_dev->hard_header_len = max_hard_header_len; > > > > > > - if (gso_partial_features & NETIF_F_GSO_ESP) > > > - bond_dev->gso_partial_features |= NETIF_F_GSO_ESP; > > > - else > > > - bond_dev->gso_partial_features &= > > > ~NETIF_F_GSO_ESP; > > > - > > > done: > > > + bond_dev->gso_partial_features = gso_partial_features; > > > bond_dev->vlan_features = vlan_features; > > > bond_dev->hw_enc_features = enc_features | > > > NETIF_F_GSO_ENCAP_ALL | > > > NETIF_F_HW_VLAN_CTAG_TX | > > > > if (!bond_has_slaves(bond)) > > goto done; > > > > If there is no slaves, should we add the gso_partial_features? > > The other partial feature sets are added after 'done:', why not do the > same for gso_partial_features for consistency? 'gso_partial_features' > is otherwise not set anywhere else and relies on it being set to zero > when allocated in alloc_netdev_mqs. I think it's better for it to be > explicitly initialized in all cases here, like the other feature sets. > > Cosmin. OK. Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c index 7b78c2bada81..e45bba240cbc 100644 --- a/drivers/net/bonding/bond_main.c +++ b/drivers/net/bonding/bond_main.c @@ -1538,17 +1538,20 @@ static netdev_features_t bond_fix_features(struct net_device *dev, NETIF_F_HIGHDMA | NETIF_F_LRO) #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ - NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) + NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE | \ + NETIF_F_GSO_PARTIAL) #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ NETIF_F_GSO_SOFTWARE) +#define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP) + static void bond_compute_features(struct bonding *bond) { + netdev_features_t gso_partial_features = BOND_GSO_PARTIAL_FEATURES; unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; - netdev_features_t gso_partial_features = NETIF_F_GSO_ESP; netdev_features_t vlan_features = BOND_VLAN_FEATURES; netdev_features_t enc_features = BOND_ENC_FEATURES; #ifdef CONFIG_XFRM_OFFLOAD @@ -1582,8 +1585,9 @@ static void bond_compute_features(struct bonding *bond) BOND_XFRM_FEATURES); #endif /* CONFIG_XFRM_OFFLOAD */ - if (slave->dev->hw_enc_features & NETIF_F_GSO_PARTIAL) - gso_partial_features &= slave->dev->gso_partial_features; + gso_partial_features = netdev_increment_features(gso_partial_features, + slave->dev->gso_partial_features, + BOND_GSO_PARTIAL_FEATURES); mpls_features = netdev_increment_features(mpls_features, slave->dev->mpls_features, @@ -1598,12 +1602,8 @@ static void bond_compute_features(struct bonding *bond) } bond_dev->hard_header_len = max_hard_header_len; - if (gso_partial_features & NETIF_F_GSO_ESP) - bond_dev->gso_partial_features |= NETIF_F_GSO_ESP; - else - bond_dev->gso_partial_features &= ~NETIF_F_GSO_ESP; - done: + bond_dev->gso_partial_features = gso_partial_features; bond_dev->vlan_features = vlan_features; bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | NETIF_F_HW_VLAN_CTAG_TX | @@ -6046,6 +6046,7 @@ void bond_setup(struct net_device *bond_dev) bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; bond_dev->features |= bond_dev->hw_features; bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; + bond_dev->features |= NETIF_F_GSO_PARTIAL; #ifdef CONFIG_XFRM_OFFLOAD bond_dev->hw_features |= BOND_XFRM_FEATURES; /* Only enable XFRM features if this is an active-backup config */
The referenced fix is incomplete. It correctly computes bond_dev->gso_partial_features across slaves, but unfortunately netdev_fix_features discards gso_partial_features from the feature set if NETIF_F_GSO_PARTIAL isn't set in bond_dev->features. This is visible with ethtool -k bond0 | grep esp: tx-esp-segmentation: off [requested on] esp-hw-offload: on esp-tx-csum-hw-offload: on This patch reworks the bonding GSO offload support by: - making aggregating gso_partial_features across slaves similar to the other feature sets (this part is a no-op). - advertising the default partial gso features on empty bond devs, same as with other feature sets (also a no-op). - adding NETIF_F_GSO_PARTIAL to hw_enc_features filtered across slaves. - adding NETIF_F_GSO_PARTIAL to features in bond_setup() With all of these, 'ethtool -k bond0 | grep esp' now reports: tx-esp-segmentation: on esp-hw-offload: on esp-tx-csum-hw-offload: on Fixes: 4861333b4217 ("bonding: add ESP offload features when slaves support") Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com> --- drivers/net/bonding/bond_main.c | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)