lkml.org 
[lkml]   [2023]   [Feb]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH net] net: dsa: mt7530: don't change PVC_EG_TAG when CPU port becomes VLAN-aware
From
On 6.02.2023 10:35, Frank Wunderlich wrote:
> Am 6. Februar 2023 00:50:53 MEZ schrieb Vladimir Oltean <vladimir.oltean@nxp.com>:
>> On Mon, Feb 06, 2023 at 02:02:48AM +0300, Arınç ÜNAL wrote:
>>> # ethtool -S eth1 | grep -v ': 0'
>>> NIC statistics:
>>> tx_bytes: 6272
>>> tx_packets: 81
>>> rx_bytes: 9089
>>> rx_packets: 136
>>> p05_TxUnicast: 52
>>> p05_TxMulticast: 3
>>> p05_TxBroadcast: 81
>>> p05_TxPktSz65To127: 136
>>> p05_TxBytes: 9633
>>> p05_RxFiltering: 11
>>> p05_RxUnicast: 11
>>> p05_RxMulticast: 26
>>> p05_RxBroadcast: 44
>>> p05_RxPktSz64: 47
>>> p05_RxPktSz65To127: 34
>>> p05_RxBytes: 6272
>>> # ethtool -S eth1 | grep -v ': 0'
>>> NIC statistics:
>>> tx_bytes: 6784
>>> tx_packets: 89
>>> rx_bytes: 9601
>>> rx_packets: 144
>>> p05_TxUnicast: 60
>>> p05_TxMulticast: 3
>>> p05_TxBroadcast: 81
>>> p05_TxPktSz65To127: 144
>>> p05_TxBytes: 10177
>>> p05_RxFiltering: 11
>>> p05_RxUnicast: 11
>>> p05_RxMulticast: 26
>>> p05_RxBroadcast: 52
>>> p05_RxPktSz64: 55
>>> p05_RxPktSz65To127: 34
>>> p05_RxBytes: 6784
>>> # ethtool -S eth1 | grep -v ': 0'
>>> NIC statistics:
>>> tx_bytes: 7424
>>> tx_packets: 99
>>> rx_bytes: 10241
>>> rx_packets: 154
>>> p05_TxUnicast: 70
>>> p05_TxMulticast: 3
>>> p05_TxBroadcast: 81
>>> p05_TxPktSz65To127: 154
>>> p05_TxBytes: 10857
>>> p05_RxFiltering: 11
>>> p05_RxUnicast: 11
>>> p05_RxMulticast: 26
>>> p05_RxBroadcast: 62
>>> p05_RxPktSz64: 65
>>> p05_RxPktSz65To127: 34
>>> p05_RxBytes: 7424
>>
>> I see no signs of packet loss on the DSA master or the CPU port.
>> However my analysis of the packets shows:
>>
>>> # tcpdump -i eth1 -e -n -Q in -XX
>>> tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
>>> listening on eth1, link-type NULL (BSD loopback), snapshot length 262144 bytes
>>> 03:50:38.645568 AF Unknown (2459068999), length 60:
>>> 0x0000: 9292 6a47 1ac0 e0d5 5ea4 edcc 0806 0001 ..jG....^.......
>> ^ ^ ^
>> | | |
>> | | ETH_P_ARP
>> | MAC SA:
>> | e0:d5:5e:a4:ed:cc
>> MAC DA:
>> 92:92:6a:47:1a:c0
>>
>>> 0x0010: 0800 0604 0002 e0d5 5ea4 edcc c0a8 0202 ........^.......
>>> 0x0020: 9292 6a47 1ac0 c0a8 0201 0000 0000 0000 ..jG............
>>> 0x0030: 0000 0000 0000 0000 0000 0000 ............
>>
>> So you have no tag_mtk header in the EtherType position where it's
>> supposed to be. This means you must be making use of the hardware DSA
>> untagging feature that Felix Fietkau added.
>>
>> Let's do some debugging. I'd like to know 2 things, in this order.
>> First, whether DSA sees the accelerated header (stripped by hardware, as
>> opposed to being present in the packet):
>>
>> diff --git a/net/dsa/tag.c b/net/dsa/tag.c
>> index b2fba1a003ce..e64628cf7fc1 100644
>> --- a/net/dsa/tag.c
>> +++ b/net/dsa/tag.c
>> @@ -75,12 +75,17 @@ static int dsa_switch_rcv(struct sk_buff *skb, struct net_device *dev,
>> if (!skb_has_extensions(skb))
>> skb->slow_gro = 0;
>>
>> + netdev_err(dev, "%s: skb %px metadata dst contains port id %d attached\n",
>> + __func__, skb, port);
>> +
>> skb->dev = dsa_master_find_slave(dev, 0, port);
>> if (likely(skb->dev)) {
>> dsa_default_offload_fwd_mark(skb);
>> nskb = skb;
>> }
>> } else {
>> + netdev_err(dev, "%s: there is no metadata dst attached to skb 0x%px\n",
>> + __func__, skb);
>> nskb = cpu_dp->rcv(skb, dev);
>> }
>>
>>
>> And second, which is what does the DSA master actually see, and put in
>> the skb metadata dst field:
>>
>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
>> index f1cb1efc94cf..e7ff569959b4 100644
>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
>> @@ -2077,11 +2077,23 @@ static int mtk_poll_rx(struct napi_struct *napi, int budget,
>> if (skb_vlan_tag_present(skb) && netdev_uses_dsa(netdev)) {
>> unsigned int port = ntohs(skb->vlan_proto) & GENMASK(2, 0);
>>
>> + netdev_err(netdev, "%s: skb->vlan_proto 0x%x port %d\n", __func__,
>> + ntohs(skb->vlan_proto), port);
>> +
>> if (port < ARRAY_SIZE(eth->dsa_meta) &&
>> - eth->dsa_meta[port])
>> + eth->dsa_meta[port]) {
>> + netdev_err(netdev, "%s: attaching metadata dst with port %d to skb 0x%px\n",
>> + __func__, port, skb);
>> skb_dst_set_noref(skb, &eth->dsa_meta[port]->dst);
>> + } else {
>> + netdev_err(netdev, "%s: not attaching any metadata dst to skb 0x%px\n",
>> + __func__, skb);
>> + }
>>
>> __vlan_hwaccel_clear_tag(skb);
>> + } else if (netdev_uses_dsa(netdev)) {
>> + netdev_err(netdev, "%s: received skb 0x%px without VLAN/DSA tag present\n",
>> + __func__, skb);
>> }
>>
>> skb_record_rx_queue(skb, 0);
>>
>> Be warned that there may be a considerable amount of output to the console,
>> so it would be best if you used a single switch port with small amounts
>> of traffic.
>
> Arınç have you tested with or without this series?
>
> https://patchwork.kernel.org/project/linux-mediatek/list/?series=707666

I'm trying this without it, this is the tree I'm testing.

https://github.com/arinc9/linux/commits/test-for-richard

Arınç

\
 
 \ /
  Last update: 2023-03-27 00:10    [W:0.161 / U:0.316 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site