Date: Mon, 28 Apr 2003 17:39:25 -0600 (MDT) From: "M. Warner Losh" <imp@bsdimp.com> To: hsu@FreeBSD.org Cc: jhb@FreeBSD.org Subject: Re: panic: sleeping thread owns a mutex Message-ID: <20030428.173925.21929852.imp@bsdimp.com> In-Reply-To: <200304282255.h3SMtuPi008021@mta4.rcsntx.swbell.net> References: <200304282255.h3SMtuPi008021@mta4.rcsntx.swbell.net>
next in thread | previous in thread | raw e-mail | index | archive | help
In message: <200304282255.h3SMtuPi008021@mta4.rcsntx.swbell.net>
Jeffrey Hsu <hsu@FreeBSD.org> writes:
: > Set a flag in your driver before you drop the wi lock that the
: > wiintr() function can check and bail out immediately if it is set.
: > For example:
:
: > foo_detach()
: > {
: > ...
: > sc->sc_dead = 1;
: > FOO_UNLOCK(sc); <--- Race 1
: > bus_teardown_intr(...)
We can't get an interrupt after this returns, and we're guaranteed
that the interrupt has terminated.
: > ...
: mtx_destroy(&sc->sc_mtx); <--- note this
: > }
:
: > foo_intr()
: > {
: > FOO_LOCK(sc);
: > if (sc->sc_dead) {
: > FOO_UNLOCK(sc);
: > return;
: > }
: > ...
: > }
:
: The sc_dead flag doesn't protect against foo_intr() attempting to
: lock a mutex that has been destroyed. fxp has the same problem
: and is one of the reasons, among others, I wasn't too happy with
: the fxp softc locks introduced there.
Since this race isn't possible, the dead solution is sufficient to
guard against the Race 1 above.
: The solution I have in mind involves using the DEAD flag in the interrupt
: handler to defer destroying the mutex if the interrupt handler is active.
bus_teardown_inter already assures that.
Warner
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030428.173925.21929852.imp>