RPI3 ethernet lockup

[Greg Troxel <gdt%lexort.com@localhost>]

I have a RPI3 that has been very solid, running recent but not
yesterday's netbsd-8.  uSD / and /boot, as normal, and then a 4T
spinning disk on USB.  It has an adafruit 2.5A power supply on a UPS.

I was doing an rsync reading contents from the spinning disk; the box
has a residential 50 Mb/s fiber connection.  It went unresponsive, and
after power cycling, I find in the logs:

Sep 25 13:09:54 foo /netbsd: usmsc0: warning: Failed to read register 0x114
Sep 25 13:09:54 foo /netbsd: usmsc0: warning: MII is busy
Sep 25 13:09:58 foo /netbsd: usmsc0: usb error on tx: TIMEOUT
Sep 25 13:09:59 foo /netbsd: usmsc0: warning: Failed to read register 0x114
Sep 25 13:09:59 foo /netbsd: usmsc0: warning: MII is busy
Sep 25 13:10:04 foo /netbsd: usmsc0: warning: Failed to read register 0x114
Sep 25 13:10:04 foo /netbsd: usmsc0: warning: MII is busy
^@(408 nulls)
first message from syslogd on boot


I wonder if others have seen this sort of trouble.  It seems the kernel
hung hard after this, and it might be good to convert this into
something that causes a reboot.

Has anyone tried to use the watchdog:
  bcmpm0 at obio0: Power management, Reset and Watchdog controller

and found that it works for such failures, in either kernel or user
tickle mode?  (I just enabled it with user tickle.)