In message <20001011164939.N20533@dr-evil.z.zembu.com>Jason R Thorpe writes
>On Wed, Oct 11, 2000 at 04:31:29PM -0700, Jonathan Stone wrote:
>
> > I'm guessing that the splimp() around all of if_up() is the culprit,
> > since it's OK if I bring the interface up a second time (or ifconfig
> > it down, when its already down) and that does all the same SIOC
> > ioctls, right up to where the kernel checkis if the interface is already up
> > (down).
>
>No, the culprit is the code that frobs the power for the PCMCIA slot.

Oh. Thanks, Jason.  So the problem is pcic_chip_socket_enable() (or
whatever) victimizing the sound-playback process with a tlseep(), not
something at high spl blocking the 50Hz interrupts for isochronous USB
audio data?  Yuck.  That means it'll still happen even if I fixed the
native ymf-7xx driver.

Any chance of using a kernel thread to do the power dance, the way
usb (iirc) now uses a thread for long-latency operations?