On Mon, Apr 03, 2000 at 12:50:48AM -0400, Thor Lancelot Simon wrote:
> Bill and I dug through disksort() (including lots of instrumentation),
> the IDE drivers, and a bunch of the buffer cache code on Friday.  Though
> I don't think my x86 assembly is really up to it, I was planning to look
> at the low-level interrupt handling code next.  I'm starting to wonder
> if maybe we're not always restoring the correct mask or something like
> that -- why would we lose with *some* interrupts but not others?

I just added some code to wdcwait() to detect busy-wait at this level.
Just set wdcdebug_mask to 0x40 to enable it.

I tried it on my system, but it doesn't looks like it's looping here on my
system.
I also have short hangs with my audio board (endsonic audiopci97, see the
mail I just posted to port-i386), which seems worse when I run a 'cvs update'
at the same time (realplayer is almost unusable while doing this, where
it works quite well when I don't stress the disk too much).

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--