> I think the 'disconnect' feature has to be explicitely asked by the driver
> (I don't have the draft under hands) for each command, so I don't think
> this is the problem.

Hmm, no, it doesn't sound very plausible, if it's per-command.

If it was "global", I'd be more suspicious that some drives might think they 
don't need to be asked, in violation of the (draft?  You mean it isn't even a) 
standard (yet?).

> I'm pretty sure now that there is a problem with interrupts in the IDE driver
> (or maybe a more general problem which only affects the IDE driver).
> It seems that sometimes the IDE driver gets an IRQ even when we're still at
> splbio() ...

Now that's _really_ interesting.

If it's a general problem, I'm not so sure it only affects the IDE driver.

I was already wondering if there isn't something screwy going on with the PS/2 
mouse driver and its interrupts.  It seems especially prone to trouble while 
I'm downloading data via my dial-up ISP, through my internal modem's serial 
port -- in other words, when the serial port driver and the PS/2 mouse driver 
are both getting interrupts.  The net result seems to be bogus mouse events, 
which can cause various problems for clients who don't/can't filter them out 
properly.  I don't know if/how it affects/involves the serial port driver.

I was able to reduce the frequency of this by tinkering with the PS/2 mouse 
driver, but still get the occasional blow-up.

> Still investigating ...

I'd like to say I'm "still investigating" but I've hardly started, because I'm 
not really sure _where_ to start....


--Jim Howard  <jiho@mail.c-zone.net>