| Message ID | 20210825083832.2425886-3-dqfext@gmail.com (mailing list archive) |
|---|---|
| State | RFC |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | DSA slave with customise netdev features | expand |
| Context | Check | Description |
|---|---|---|
| netdev/cover_letter | success | Link |
| netdev/fixes_present | success | Link |
| netdev/patch_count | success | Link |
| netdev/tree_selection | success | Clearly marked for net-next |
| netdev/subject_prefix | success | Link |
| netdev/cc_maintainers | success | CCed 13 of 13 maintainers |
| netdev/source_inline | success | Was 0 now: 0 |
| netdev/verify_signedoff | success | Link |
| netdev/module_param | success | Was 0 now: 0 |
| netdev/build_32bit | success | Errors and warnings before: 0 this patch: 0 |
| netdev/kdoc | success | Errors and warnings before: 0 this patch: 0 |
| netdev/verify_fixes | success | Link |
| netdev/checkpatch | success | total: 0 errors, 0 warnings, 0 checks, 68 lines checked |
| netdev/build_allmodconfig_warn | success | Errors and warnings before: 0 this patch: 0 |
| netdev/header_inline | success | Link |
On Wed, Aug 25, 2021 at 04:38:31PM +0800, DENG Qingfang wrote: > Advertise TX VLAN offload features, and handle VLAN tag insertion in > the tag_xmit function. > > Signed-off-by: DENG Qingfang <dqfext@gmail.com> > --- > net/dsa/tag_mtk.c | 46 ++++++++++++++++++++++------------------------ > 1 file changed, 22 insertions(+), 24 deletions(-) > > diff --git a/net/dsa/tag_mtk.c b/net/dsa/tag_mtk.c > index 415d8ece242a..e407abefa06c 100644 > --- a/net/dsa/tag_mtk.c > +++ b/net/dsa/tag_mtk.c > @@ -22,7 +22,6 @@ static struct sk_buff *mtk_tag_xmit(struct sk_buff *skb, > struct net_device *dev) > { > struct dsa_port *dp = dsa_slave_to_port(dev); > - u8 xmit_tpid; > u8 *mtk_tag; > > /* Build the special tag after the MAC Source Address. If VLAN header > @@ -31,33 +30,31 @@ static struct sk_buff *mtk_tag_xmit(struct sk_buff *skb, > * the both special and VLAN tag at the same time and then look up VLAN > * table with VID. > */ > - switch (skb->protocol) { > - case htons(ETH_P_8021Q): > - xmit_tpid = MTK_HDR_XMIT_TAGGED_TPID_8100; > - break; > - case htons(ETH_P_8021AD): > - xmit_tpid = MTK_HDR_XMIT_TAGGED_TPID_88A8; > - break; > - default: > - xmit_tpid = MTK_HDR_XMIT_UNTAGGED; > - skb_push(skb, MTK_HDR_LEN); > - dsa_alloc_etype_header(skb, MTK_HDR_LEN); > - } > - You cannot just remove the old code. Only things like 8021q uppers will send packets with the VLAN in the hwaccel area. If you have an application that puts the VLAN in the actual AF_PACKET payload, like: https://github.com/vladimiroltean/tsn-scripts/blob/master/isochron/send.c then you need to handle the VLAN being in the skb payload.
On Thu, Aug 26, 2021 at 03:03:49AM +0300, Vladimir Oltean wrote: > > You cannot just remove the old code. Only things like 8021q uppers will > send packets with the VLAN in the hwaccel area. > > If you have an application that puts the VLAN in the actual AF_PACKET > payload, like: > > https://github.com/vladimiroltean/tsn-scripts/blob/master/isochron/send.c > > then you need to handle the VLAN being in the skb payload. I've actually tested this (only apply patch 2 without .features) and it still worked. The comment says the VLAN tag need to be combined with the special tag in order to perform VLAN table lookup, so we can set its destination port vector to all zeroes and the switch will forward it like a data frame (TX forward offload), but as we allow multiple bridges which are either VLAN-unaware or VLAN-aware with the same VID, there is no way to determine the destination bridge unless we maintain some VLAN translation mapping.
On 8/26/2021 7:29 AM, DENG Qingfang wrote: > On Thu, Aug 26, 2021 at 03:03:49AM +0300, Vladimir Oltean wrote: >> >> You cannot just remove the old code. Only things like 8021q uppers will >> send packets with the VLAN in the hwaccel area. >> >> If you have an application that puts the VLAN in the actual AF_PACKET >> payload, like: >> >> https://github.com/vladimiroltean/tsn-scripts/blob/master/isochron/send.c >> >> then you need to handle the VLAN being in the skb payload. > > I've actually tested this (only apply patch 2 without .features) and it > still worked. OK and this works because now you always push a MTK_HDR_LEN tag which you were not doing before for VLAN-tagged frames, right? You don't seem to be clearing mtk_tag[2] and mtk_tag[3] anymore, is that intended? > > The comment says the VLAN tag need to be combined with the special tag in > order to perform VLAN table lookup, so we can set its destination port > vector to all zeroes and the switch will forward it like a data frame > (TX forward offload), but as we allow multiple bridges which are either > VLAN-unaware or VLAN-aware with the same VID, there is no way to determine > the destination bridge unless we maintain some VLAN translation mapping. There is no "bridge" on the other side from the CPU port's perspective on egress, only physical ports, so how does that comment apply?
On Thu, Aug 26, 2021 at 01:29:56PM +0800, DENG Qingfang wrote: > On Thu, Aug 26, 2021 at 03:03:49AM +0300, Vladimir Oltean wrote: > > > > You cannot just remove the old code. Only things like 8021q uppers will > > send packets with the VLAN in the hwaccel area. > > > > If you have an application that puts the VLAN in the actual AF_PACKET > > payload, like: > > > > https://github.com/vladimiroltean/tsn-scripts/blob/master/isochron/send.c > > > > then you need to handle the VLAN being in the skb payload. > > I've actually tested this (only apply patch 2 without .features) and it > still worked. > > The comment says the VLAN tag need to be combined with the special tag in > order to perform VLAN table lookup, It does say this. > so we can set its destination port vector to all zeroes and the switch > will forward it like a data frame (TX forward offload), And it does not say this. So this is supported after all with mt7530? Are you looking to add support for that? > but as we allow multiple bridges which are either VLAN-unaware or > VLAN-aware with the same VID, there is no way to determine the > destination bridge unless we maintain some VLAN translation mapping. What does "VLAN translation mapping" mean, practically? Other drivers which cannot remap VIDs to internal VLANs just restrict a single VLAN-aware bridge, and potentially multiple VLAN-unaware ones.
On Thu, Aug 26, 2021 at 05:13:26PM +0300, Vladimir Oltean wrote: > On Thu, Aug 26, 2021 at 01:29:56PM +0800, DENG Qingfang wrote: > > > > The comment says the VLAN tag need to be combined with the special tag in > > order to perform VLAN table lookup, > > It does say this. > > > so we can set its destination port vector to all zeroes and the switch > > will forward it like a data frame (TX forward offload), > > And it does not say this. So this is supported after all with mt7530? > Are you looking to add support for that? I already run-tested that, it works, as long as there is only one bridge. > > > but as we allow multiple bridges which are either VLAN-unaware or > > VLAN-aware with the same VID, there is no way to determine the > > destination bridge unless we maintain some VLAN translation mapping. > > What does "VLAN translation mapping" mean, practically? It's just VLAN remapping, as you stated below. > Other drivers which cannot remap VIDs to internal VLANs just restrict a > single VLAN-aware bridge, and potentially multiple VLAN-unaware ones.
diff --git a/net/dsa/tag_mtk.c b/net/dsa/tag_mtk.c index 415d8ece242a..e407abefa06c 100644 --- a/net/dsa/tag_mtk.c +++ b/net/dsa/tag_mtk.c @@ -22,7 +22,6 @@ static struct sk_buff *mtk_tag_xmit(struct sk_buff *skb, struct net_device *dev) { struct dsa_port *dp = dsa_slave_to_port(dev); - u8 xmit_tpid; u8 *mtk_tag; /* Build the special tag after the MAC Source Address. If VLAN header @@ -31,33 +30,31 @@ static struct sk_buff *mtk_tag_xmit(struct sk_buff *skb, * the both special and VLAN tag at the same time and then look up VLAN * table with VID. */ - switch (skb->protocol) { - case htons(ETH_P_8021Q): - xmit_tpid = MTK_HDR_XMIT_TAGGED_TPID_8100; - break; - case htons(ETH_P_8021AD): - xmit_tpid = MTK_HDR_XMIT_TAGGED_TPID_88A8; - break; - default: - xmit_tpid = MTK_HDR_XMIT_UNTAGGED; - skb_push(skb, MTK_HDR_LEN); - dsa_alloc_etype_header(skb, MTK_HDR_LEN); - } - + skb_push(skb, MTK_HDR_LEN); + dsa_alloc_etype_header(skb, MTK_HDR_LEN); mtk_tag = dsa_etype_header_pos_tx(skb); - /* Mark tag attribute on special tag insertion to notify hardware - * whether that's a combined special tag with 802.1Q header. - */ - mtk_tag[0] = xmit_tpid; - mtk_tag[1] = (1 << dp->index) & MTK_HDR_XMIT_DP_BIT_MASK; - - /* Tag control information is kept for 802.1Q */ - if (xmit_tpid == MTK_HDR_XMIT_UNTAGGED) { - mtk_tag[2] = 0; - mtk_tag[3] = 0; + if (skb_vlan_tag_present(skb)) { + switch (skb->vlan_proto) { + case htons(ETH_P_8021Q): + mtk_tag[0] = MTK_HDR_XMIT_TAGGED_TPID_8100; + break; + case htons(ETH_P_8021AD): + mtk_tag[0] = MTK_HDR_XMIT_TAGGED_TPID_88A8; + break; + default: + return NULL; + } + + ((__be16 *)mtk_tag)[1] = htons(skb_vlan_tag_get(skb)); + __vlan_hwaccel_clear_tag(skb); + } else { + mtk_tag[0] = MTK_HDR_XMIT_UNTAGGED; + ((__be16 *)mtk_tag)[1] = 0; } + mtk_tag[1] = (1 << dp->index) & MTK_HDR_XMIT_DP_BIT_MASK; + return skb; } @@ -96,6 +93,7 @@ static const struct dsa_device_ops mtk_netdev_ops = { .xmit = mtk_tag_xmit, .rcv = mtk_tag_rcv, .needed_headroom = MTK_HDR_LEN, + .features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX, }; MODULE_LICENSE("GPL");
Advertise TX VLAN offload features, and handle VLAN tag insertion in the tag_xmit function. Signed-off-by: DENG Qingfang <dqfext@gmail.com> --- net/dsa/tag_mtk.c | 46 ++++++++++++++++++++++------------------------ 1 file changed, 22 insertions(+), 24 deletions(-)