What is the rule of thumb with regard to spl*() (in my case splbio()) in
conjunction with simple_lock()?  Must they be nested, and if so in what
order?

For example, I have a pending change to vrele which (I think) should look
like this:

        s = splbio();
        simple_lock(&vp->v_interlock);
        vp->v_usecount--;
        if (vp->v_usecount > 0) {
                simple_unlock(&vp->v_interlock);
                splx(s);
                return;
        }

        simple_lock(&vnode_free_list_slock);
        if (vp->v_holdcnt > 0)
                TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
        else
                TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
        simple_unlock(&vnode_free_list_slock);
        splx(s);

The purpose of the change is that "v_usecount--" and the tailq insertion
need to be atomic vis-a-vis disk interrupts.  Given that the section under
splbio() has to interact with the simple_lock()s, is this the best way to
do this, or might

        simple_lock(&vp->v_interlock);
        s = splbio();
        vp->v_usecount--;
        if (vp->v_usecount > 0) {
                simple_unlock(&vp->v_interlock);
                splx(s);
                return;
        }

        simple_lock(&vnode_free_list_slock);
        if (vp->v_holdcnt > 0)
                TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
        else
                TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
        splx(s);
        simple_unlock(&vnode_free_list_slock);

be as good?

Thanks,
						Konrad Schroder
						perseant@hhhh.org