Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Mar 10, 2008 at 03:12:44PM -0400, John Baldwin wrote:
> On Saturday 08 March 2008 06:21:29 am Kostik Belousov wrote:
> > On Sat, Mar 08, 2008 at 02:01:51AM +0300, Oleg Sidorkin wrote:
> > > Hello,
> > >=20
> > > I'm running 7.0-Stable on the ASUS P5K-VM + Intel Q6600 box.
> > > If Palm device is disconnected after synchronization, system crashes =

> > > following stacktrace:
> > >=20
> > > Does anyone have any ideas?

> >=20
> > I posted the patch several days ago, see
> > http://lists.freebsd.org/pipermail/f...rch/084076.ht=

> So this axes the Giant trick stuff, is this because there are race with t=

> cdevsw pointer itself?

There is a separate race with cdevsw being uninitialized while device is
created that is fixed in the patch.

The panic reported is caused by the dereference of the cdev->si_devsw
in the giant_trick wrappers without holding the dev_mtx. In fact, dropping
dev_mtx in dev(vn)_refthreads and then dereferencing the si_devsw.
The destroy_devl() routine zeroes out si_devsw before waiting for
si_threadcount to settle to zero. This works as a barrier for
dev_refthread. Dereferencing si_devsw afterward requires either
the same check or returning giant_trick from dev_refthread or axing
gianttrick (dev_refthread returns cdevsw). I choose the later.

Content-Type: application/pgp-signature
Content-Disposition: inline

Version: GnuPG v1.4.8 (FreeBSD)