MT7621/MT7530: fdb learning bug also in 5.15

[dl3it]

device: Xiaomi Router 4A Gigabit

The fdb learning bug is also present in kernel 5.15.
see also #9420

Devices roaming between wifi access points don't get network access until the fdb entry in mt7530 is cleared by timeout.

[Vladdrako]

@quarkysg Can you port your patch to 5.15?

[neheb]

ping @LGA1150

[LGA1150]

I'm busy with work recently and may not have time for this, sorry.

[quarkysg]

So because for WiFi clients DSA will add an ARL entry in the mt7530 for CPU port (as source frame originates from mt76 to mt7530), and because the mt7530 auto learn is disabled, the ARL is not updated when packet comes in via another switch port, as when CPU send data to the mt7530, it will shoot it back to the CPU port (since that's what the ARL table says), and the Linux stack will likely discard it.

When the ARL entry times out, and new packet comes in from the LAN port for the WiFi client associated to the AP, the Linux kernel knows that it comes from the switch, so sends replies to the mt7530. But now the ARL table in mt7530 doesn't have any record of the WiFi client MAC address, so goes thru ARP protocol again. This likely will cause a thruput degradation.

My conclusion is that the DSA framework is 'broken' when it comes to kernel assisted SA learning. It is not creating the proper ARL entries in the mt7530 when the mt7530's auto SA learning is disabled.

@Vladdrako the mt7530 driver in 5.15.33 is slightly different from the one in 5.10.x, but it should be fairly straigh forward to amend my patch to work with it. I'll try to create a patch file for it for you to try, if you are brave enough to test my patch.

[Vladdrako]

the mt7530 driver in 5.15.33 is slightly different from the one in 5.10.x, but it should be fairly straigh forward to amend my patch to work with it. I'll try to create a patch file for it for you to try, if you are brave enough to test my patch.

No problem :)

[dl3it]

@quarkysg I have an extra device here in place for testing purposes... If you like, I'll give your patch a try...

[quarkysg]

@Vladdrako, @dl3it If you have the time, do help test out this patch. Do note that I have not compile or run tested this patch, as I have yet to switch to the 5.15 kernel and my E8450 is still running my test builds based on 5.10 kernel.

[dl3it]

I gave it a try... something fails.

CC drivers/net/dsa/mt7530.o
drivers/net/dsa/mt7530.c: In function 'mt7530_port_vlan_add':
drivers/net/dsa/mt7530.c:1585:18: error: 'const struct switchdev_obj_port_vlan' has no member named 'vid_end'
1585 | if (!vlan->vid_end || dsa_is_cpu_port(ds, port))
| ^~
In file included from ./include/linux/kernel.h:19,
from ./include/linux/skbuff.h:13,
from ./include/linux/if_ether.h:19,
from ./include/linux/etherdevice.h:20,
from drivers/net/dsa/mt7530.c:6:
drivers/net/dsa/mt7530.c:1592:83: error: 'const struct switchdev_obj_port_vlan' has no member named 'vid_end'
1592 | func, port, priv->ports[port].pvid, pvid, vlan->vid_end);
| ^~
./include/linux/printk.h:422:33: note: in definition of macro 'printk_index_wrap'
422 | _p_func(_fmt, ##VA_ARGS);
| ^~~~~~~~~~~
./include/linux/printk.h:503:9: note: in expansion of macro 'printk'
503 | printk(KERN_WARNING pr_fmt(fmt), ##VA_ARGS)
| ^~~~~~
drivers/net/dsa/mt7530.c:1590:25: note: in expansion of macro 'pr_warn'
1590 | pr_warn("%s: port[%d] already configured with vid[%d] - "
| ^~~~~~~
drivers/net/dsa/mt7530.c:1594:46: error: 'const struct switchdev_obj_port_vlan' has no member named 'vid_end'
1594 | priv->ports[port].pvid = vlan->vid_end;
| ^~
In file included from drivers/net/dsa/mt7530.c:25:
drivers/net/dsa/mt7530.c:1598:44: error: 'const struct switchdev_obj_port_vlan' has no member named 'vid_end'
1598 | G0_PORT_VID(vlan->vid_end));
| ^~
drivers/net/dsa/mt7530.h:281:44: note: in definition of macro 'G0_PORT_VID'
281 | #define G0_PORT_VID(x) (((x) & 0xfff) << 0)
| ^
make[8]: *** [scripts/Makefile.build:277: drivers/net/dsa/mt7530.o] Error 1
make[7]: *** [scripts/Makefile.build:540: drivers/net/dsa] Error 2
make[6]: *** [scripts/Makefile.build:540: drivers/net] Error 2
make[5]: *** [Makefile:1868: drivers] Error 2
make[5]: Leaving directory '/mnt/Build/openwrt.mi3g/build_dir/target-mipsel_24kc_musl/linux-ramips_mt7621/linux-5.15.33'

[Vladdrako]

something fails.

Same errors.
EDIT: if I replaced vid_end with vid it compiled, but I'm not sure this is correct))

[quarkysg]

@dl3it @Vladdrako Really sorry. I didn't detect that the struct switchdev_obj_port_vlan structure has changed. I have amended the patch. You can try the patch again. Same link.

@Vladdrako it's correct.

[quarkysg]

A quick read on the mt7530.c driver code shows that in 5.10, the driver turns on the flag to make DSA configure the switch with VLAN when vlan_filtering is not explicitly configure, while in 5.15, this flag was removed.

I would suggest to test out configuration with and without vlan_filtering with bridges that's configured.

My recommendation is to assign all switch ports (lan & wan) to one bridge and assign them separate VLAN numbers. Otherwise DSA will assign all switch ports to VLAN 0, which is reserved for priority tagged frames.

[dl3it]

first quick test: device is unreachable on ethernet. I`ll connect the serial interface to check further details...

[Vladdrako]

first quick test: device is unreachable on ethernet. I`ll connect the serial interface to check further details...

No such issue for me.

[dl3it]

everything comes up as expected...

root@keller:/# bridge v
port vlan-id
wan 100
117
118
119
120
128
br-gast 1 PVID Egress Untagged
switch0 100
117
118
119
120
128
switch0.119 1 PVID Egress Untagged
br-iot 1 PVID Egress Untagged
switch0.120 1 PVID Egress Untagged
br-lan 1 PVID Egress Untagged
switch0.100 1 PVID Egress Untagged
br-media 1 PVID Egress Untagged
switch0.118 1 PVID Egress Untagged
br-rapw 1 PVID Egress Untagged
switch0.128 1 PVID Egress Untagged
br-work 1 PVID Egress Untagged
switch0.117 1 PVID Egress Untagged
wlan0 1 PVID Egress Untagged
wlan1 1 PVID Egress Untagged
wlan1-1 1 PVID Egress Untagged
wlan1-2 1 PVID Egress Untagged
wlan1-3 1 PVID Egress Untagged
root@keller:/# brctl show
bridge name bridge id STP enabled interfaces
br-rapw 7fff.50d2f526b39e no switch0.128
wlan1-3
br-work 7fff.50d2f526b39e no switch0.117
wlan1
br-media 7fff.50d2f526b39e no wlan1-2
switch0.118
br-iot 7fff.50d2f526b39e no switch0.120
wlan0
switch0 7fff.50d2f526b39e no lan2
wan
lan1
br-gast 7fff.50d2f526b39e no switch0.119
wlan1-1
br-lan 7fff.50d2f526b39e no switch0.100

but the device has no access to the WAN. tcpdump shows packets, but they don't go through the switch. E.g. packet tagged with VLAN 100 seen on WAN are not seen on switch0.100