imx8mp router loopback, causing NETDEV WATCHDOG to reset and unable to restore the mac

[yrj]

 Hello, may I ask if our A board and B board are connected to the switch at the same time, and then A-B mutual ping, the switch has a ring network due to misoperation (both ends of a network cable are connected to the switch), there is a problem with the kernel (direct restart, or debug is stuck, etc.), the detailed log is in the attachment.

[ 314.850236] ------------[ cut here ]------------
[ 314.854871] NETDEV WATCHDOG: ens3 (fec): transmit queue 0 timed out 3384 ms
[ 314.861910] WARNING: CPU: 0 PID: 15 at net/sched/sch_generic.c:525 dev_watchdog+0x234/0x23c
[ 314.870271] Modules linked in:
[ 314.873329] CPU: 0 PID: 15 Comm: ksoftirqd/0 Not tainted 6.6.3-g9d3450dbcab9-dirty #71
[ 314.881246] Hardware name: NXP i.MX8MPlus EVK board (DT)
[ 314.886555] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 314.893516] pc : dev_watchdog+0x234/0x23c
[ 314.897528] lr : dev_watchdog+0x234/0x23c
[ 314.901538] sp : ffff8000823e3c20
[ 314.904850] x29: ffff8000823e3c20 x28: 0000000000000100 x27: ffff8000823e3cd0
[ 314.911989] x26: ffff8000820d69c0 x25: 0000000000000d38 x24: ffff8000820d6000
[ 314.919128] x23: 0000000000000000 x22: ffff0000055503dc x21: ffff000005550000
[ 314.926266] x20: ffff00000589c400 x19: ffff000005550488 x18: 0000000000000006
[ 314.933404] x17: ffff8000821b62e8 x16: 0000000073e4d9ea x15: ffff8000823e3640
[ 314.940541] x14: 0000000000000000 x13: ffff8000820f0da0 x12: 0000000000000672
[ 314.947679] x11: 0000000000000226 x10: ffff800082148da0 x9 : ffff8000820f0da0
[ 314.954818] x8 : 00000000ffffefff x7 : ffff800082148da0 x6 : 80000000fffff000
[ 314.961956] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
[ 314.969093] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff00000412d880

[yrj]

内核版本是：6.6.3

[Zhiming_Liu]

Hi @yrj 

这个问题更多得是和交换机有关，建议通过交换机的环路检测功能解决环网问题。

Best Regards,
Zhiming

[yrj]

Hi @Zhiming_Liu 

      不小心造成环路，imx8mp可以预防吗

[Zhiming_Liu]

Hi,

最佳的方法是交换机侧设置防御，一般都有这种功能，可以参考这个：

https://www.cisco.com/c/zh_cn/support/docs/smb/switches/cisco-250-series-smart-switches/smb5794-enab...

当然也可以在驱动层添加广播检测统计，但是目前fec驱动不支持这项功能。

你可以通过一个脚本在用户层监测广播，如果出现大量广播，就把网卡down掉。

Best Regards,
Zhiming

[yrj]

Hi @Zhiming_Liu

     网络风暴发生时，系统有概率直接重启，或者debug调试口卡死，或者系统跑飞（如b.log），需要重启才能恢复。