Re: [9fans] i/o error: wrenwrite

[Charles Forsyth <forsyth@terzarima.net>]

>>Plan9 has a bug that puts it into endless loop on my hardware
>>after reboot without fshalt. I am asking where to look at to
>>fix the bug.

quite right.

if you can recompile a kernel while the system is up, and run that,
you could try adding code to /sys/src/9/pc/sdata.c to
atainterrupt (say) to help diagnose the i/o error more precisely.
"i/o error" isn't much help as you've discovered.
there might already be debugging code that can be switched on.
(i did have a very quick look at it myself but i didn't see
anything just right for this case, but i might easily have missed it.)

round about here, say:
	if(status & Err)
		drive->error = inb(cmdport+Error);

i suspect the error might happen much earlier, before it gets to
generate an interrupt, but that should become apparent if you
see no error output from that point.  just a print() of the status
might do.  i suspect the controller ends up in a state where its
registers don't look right to the driver, or indeed where the
controller needs a bit of a whack from the driver after the power cycling,
in which case it probably can't generate IO at all and it's (perhaps)
timing out the command.  it's interesting that it's apparently only
wrenwrite not wrenread that generates errors (or i missed something).
that in itself might help pinpoint the problem.