Date: Wed, 20 Dec 2006 12:14:52 +0100 (CET) From: Martin Blapp <mb@imp.ch> To: Attilio Rao <attilio@FreeBSD.org> Cc: cvs-src@FreeBSD.org, Martin Blapp <mbr@FreeBSD.org>, cvs-all@FreeBSD.org, src-committers@FreeBSD.org Subject: Re: cvs commit: src/sys/kern tty.c Message-ID: <20061220120419.M53548@godot.imp.ch> In-Reply-To: <3bbf2fe10612191850y3a546fcew486575385b1400df@mail.gmail.com> References: <200612192234.kBJMYYo0055529@repoman.freebsd.org> <3bbf2fe10612191850y3a546fcew486575385b1400df@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi,
> Since proctree_lock is a sx lock which uses 2 condition variables,
> they alredy drop Giant (DROP_GIANT()) before sleeping. Are you secure
> it is the right thing to do here?
Yes I am.
sx_slock() can call cv_wait() which later can call cv_wait_unlock ...
if there is already a lock we need to share ... After a possible
sleep cv_wait_unlock looks like:
cv_wait_unlock(struct cv *cvp, struct mtx *mp)
[...]
sleepq_lock(cvp);
cvp->cv_waiters++;
DROP_GIANT();
mtx_unlock(mp);
[...]
It can call DROP_GIANT for a short amount of time. This was the race.
I already fixed it (but haven't seen the real cause for the crashes) once in
STABLE and HEAD, but backed it out again because I (and jhb) were sure that the
tty driver was protected fully by the GIANT lock after I fixed severall
possible race places in kern_proc.c. I had no idea that sx_slock() and sx_slock
both can drop GIANT for a short time and that we need carefully recheck
variables.
Martin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20061220120419.M53548>