sys/usb/usbus/cdc/ecm: inerface thread blocked when connected host interface is down

[fabian18]

Description

USBUS cdcecm thread stalls when sending ping and the receiving interface on Linux is down.

I expected an interrupt to happen after a timeout if a transaction fails. In the MCU datasheet I found the TRFAIL interrupt which was not enabled yet. But with this patch when I try to enable it, it is fireing all the time and does not solve the problem.

cdcecm.patch.txt

Steps to reproduce the issue

Flash examples/gnrc_networking to same54-xpro withe USEMODULE+="usbus_cdc_ecm auto_init_usbus"

2024-12-20 14:53:01,433 # ifconfig
2024-12-20 14:53:01,437 # Iface  6  HWaddr: FC:C2:3D:2F:8E:90  Link: down 
2024-12-20 14:53:01,442 #           L2-PDU:1500  MTU:1500  HL:64  RTR  
2024-12-20 14:53:01,445 #           Source address length: 6
2024-12-20 14:53:01,447 #           Link type: wired
2024-12-20 14:53:01,448 #           
2024-12-20 14:53:01,451 #           Statistics for Layer 2
2024-12-20 14:53:01,454 #             RX packets 0  bytes 0
2024-12-20 14:53:01,458 #             TX packets 0 (Multicast: 0)  bytes 0
2024-12-20 14:53:01,462 #             TX succeeded 0 errors 0
2024-12-20 14:53:01,464 #           Statistics for IPv6
2024-12-20 14:53:01,467 #             RX packets 0  bytes 0
2024-12-20 14:53:01,471 #             TX packets 0 (Multicast: 0)  bytes 0
2024-12-20 14:53:01,475 #             TX succeeded 0 errors 0
2024-12-20 14:53:01,475 # 
2024-12-20 14:53:01,478 # Iface  7  HWaddr: FE:76:BD:F3:19:D8 
2024-12-20 14:53:01,483 #           L2-PDU:1500  MTU:1500  HL:64  RTR  
2024-12-20 14:53:01,486 #           Source address length: 6
2024-12-20 14:53:01,488 #           Link type: wired
2024-12-20 14:53:01,494 #           inet6 addr: fe80::fc76:bdff:fef3:19d8  scope: link  VAL
2024-12-20 14:53:01,497 #           inet6 group: ff02::2
2024-12-20 14:53:01,499 #           inet6 group: ff02::1
2024-12-20 14:53:01,503 #           inet6 group: ff02::1:fff3:19d8
2024-12-20 14:53:01,504 #           
2024-12-20 14:53:01,507 #           Statistics for Layer 2
2024-12-20 14:53:01,510 #             RX packets 14  bytes 2360
2024-12-20 14:53:01,514 #             TX packets 4 (Multicast: 4)  bytes 202
2024-12-20 14:53:01,518 #             TX succeeded 0 errors 0
2024-12-20 14:53:01,520 #           Statistics for IPv6
2024-12-20 14:53:01,524 #             RX packets 11  bytes 1216
2024-12-20 14:53:01,528 #             TX packets 4 (Multicast: 4)  bytes 224
2024-12-20 14:53:01,531 #             TX succeeded 4 errors 0

Interface 7 is cdcecm.

Set interface on Linux side down:

$ sudo ip link set enxfe76bdf305d8 down
9: enxfe76bdf305d8: <BROADCAST,MULTICAST> mtu 1500 qdisc fq_codel state DOWN group default qlen 1000
    link/ether fe:76:bd:f3:05:d8 brd ff:ff:ff:ff:ff:ff

Send ping over cdcecm interface.

> ping ff02::1%7
2024-12-20 14:55:50,706 # ping ff02::1%7
2024-12-20 14:55:53,706 # 
2024-12-20 14:55:53,709 # --- ff02::1%7 PING statistics ---
2024-12-20 14:55:53,714 # 3 packets transmitted, 0 packets received, 100% packet loss

Another ifconfig stalls and interface 7 doe not show up.

2024-12-20 14:56:56,129 # ifconfig
2024-12-20 14:56:56,133 # Iface  6  HWaddr: FC:C2:3D:2F:8E:90  Link: down 
2024-12-20 14:56:56,138 #           L2-PDU:1500  MTU:1500  HL:64  RTR  
2024-12-20 14:56:56,141 #           Source address length: 6
2024-12-20 14:56:56,143 #           Link type: wired
2024-12-20 14:56:56,144 #           
2024-12-20 14:56:56,147 #           Statistics for Layer 2
2024-12-20 14:56:56,150 #             RX packets 0  bytes 0
2024-12-20 14:56:56,154 #             TX packets 0 (Multicast: 0)  bytes 0
2024-12-20 14:56:56,157 #             TX succeeded 0 errors 0
2024-12-20 14:56:56,160 #           Statistics for IPv6
2024-12-20 14:56:56,163 #             RX packets 0  bytes 0
2024-12-20 14:56:56,167 #             TX packets 0 (Multicast: 0)  bytes 0
2024-12-20 14:56:56,170 #             TX succeeded 0 errors 0
2024-12-20 14:56:56,171 # 

The second send blocks the interface thread because there

RIOT/sys/usb/usbus/cdc/ecm/cdc_ecm_netdev.c

Line 78 in 4f55461

mutex_lock(&cdcecm->out_lock);

a mutex is locked twice without being unlock after a successful transfer, because there is no successful transfer.

Expected results

Even thought the Linux interface is down, RIOT should not stall after sending a ping.

Actual results

RIOT stalls because a mutex is not unlocked, when sending over a cdcecm interface when the Linux interface is down.

Versions

[fabian18]

When the Linux interface is up again the mutex is unlocked.