On Saturday 29 April 2006 01:38, SODA Noriyuki wrote:
> >     * we have no splhard<something>() so splhardusb/splusb don't
> >     follow our naming convention. Most likely we would need to make
> >     splsoftusb == current splusb and make splhardusb == splusb.
>
> One of confusing part is that interrupt handlers which are protected
> by splhardusb() are called with IPL_USB (== IPL_BIO) level, and
> interrupt handlers which are protected by splusb() level are called
> with IPL_SOFTNET (in the __HAVE_GENERIC_SOFT_INTERRUPTS case) or
> IPL_SOFTCLOCK (in !defined(__HAVE_GENERIC_SOFT_INTERRUPTS) level ;)
>
> >     * we should probably define splusb under splvm and spltty (I'm
> >       ignoring the current problem of spltty being lower priority than
> >       splnet -- maybe it's time to fix this ?)
>
> Probably this is not so simple change, because we want non-isochronous
> devices to keep using lower priority, but want isochronous device
> interrupt handlers to be called without lowering interrupt priority,
> and also have to protect some usb controller resources from both priority
> levels.

How about using mutexes, like on FreeBSD, instead of splxxx?

--HPS