5.4-p1 crash

[Doug White]

On Mon, 20 Jun 2005, Philippe PEGON wrote:

> Philippe PEGON wrote:
> > Mitch Parks wrote:
> >
> >> On Sun, 19 Jun 2005, Doug White wrote:
> >>
> >>> On Fri, 17 Jun 2005, Mitch Parks wrote:
> >>>
> >>>> Below are details regarding another crash on a Dell 2600 SMP (HTT
> >>>> and USB
> >>>> disabled). It has been 9 days since the last crash. I didn't have
> >>>> the serial
> >>>> console in place for this last crash, but it is now.
> >>>
> >>>
> >>>
> >>> As noted, the ttwakeup() panic is a known bug. The best thing we have
> >>> for
> >>> a fix is this patch:
> >>>
> >>> http://people.freebsd.org/~mlaier/tty.t_pgrp.diff
> >>>
> >>> Please give it a try and report back if you have any more panics (or
> >>> don't :-) ).
> >>
> >>
> >>
> >> Thanks! This patch appears to be for 5.3, but I manually applied the
> >> chunk of the patch that didn't apply cleanly and the countdown is on.
> >>
> >> I'll report back in 10 days unless something bad happens before then.
> >>
> >> Below is the patch chunk #10 that I actually applied rather than the
> >> one given. If I've done something bad here by removing the PGRP_LOCK
> >> please let me know.
> >
> >
> > I'm not a kernel developper, but if you remove
> >
> > PGRP_LOCK(tp->t_pgrp);
> >
> > and the PGRP_UNLOCK(tp->t_pgrp) in the if condition (removed by the
> > orginal patch)
> >
> > there is maybe another "PGRP_UNLOCK(tp->t_pgrp);" to remove if the if
> > condition doesn't match, line 2528 in the original 5.4-p1 tty.c ?
>
> after having applied the patch (with your modification), there is no
> "sx_sunlock(&proctree_lock)" in the ttyinfo function if the three
> conditions failed. Maybe we have just to replace
> "PGRP_UNLOCK(tp->t_pgrp);" line 2528 by "sx_sunlock(&proctree_lock)" ?
> I think that we need the helps of a kernel developper.

No, that would be a leaked lock, which would cause hangs.  More likely its
some other case that got missed that needs locks extended to it, or the
aliased pgrp isn't the underlying problem.

I've run out of time to debug this, unfortunately...

>
> >
> >>
> >> ....
> >> Hunk #6 succeeded at 1154 (offset -51 lines).
> >> Hunk #7 succeeded at 1215 (offset -6 lines).
> >> Hunk #8 succeeded at 1203 (offset -51 lines).
> >> Hunk #9 succeeded at 1946 (offset -5 lines).
> >> Hunk #10 failed at 2562.
> >> Hunk #11 succeeded at 2847 (offset -212 lines).
> >> 1 out of 11 hunks failed--saving rejects to tty.c.rej
> >>
> >>
> >> @@ -2495,19 +2511,21 @@
> >>          * On return following a ttyprintf(), we set tp->t_rocount to
> >> 0 so
> >>          * that pending input will be retyped on BS.
> >>          */
> >> +       sx_slock(&proctree_lock);
> >>         if (tp->t_session == NULL) {
> >> +               sx_sunlock(&proctree_lock);
> >>                 ttyprintf(tp, "not a controlling terminal\n");
> >>                 tp->t_rocount = 0;
> >>                 return;
> >>         }
> >>         if (tp->t_pgrp == NULL) {
> >> +               sx_sunlock(&proctree_lock);
> >>                 ttyprintf(tp, "no foreground process group\n");
> >>                 tp->t_rocount = 0;
> >>                 return;
> >>         }
> >> -       PGRP_LOCK(tp->t_pgrp);
> >> -       if ((p = LIST_FIRST(&tp->t_pgrp->pg_members)) == 0) {
> >> -               PGRP_UNLOCK(tp->t_pgrp);
> >> +       if ((p = LIST_FIRST(&tp->t_pgrp->pg_members)) == NULL) {
> >> +               sx_sunlock(&proctree_lock);
> >>                 ttyprintf(tp, "empty foreground process group\n");
> >>                 tp->t_rocount = 0;
> >>                 return;
> >>
> >> Or the complete patch:
> >> http://kuoi.asui.uidaho.edu/~mitch/crash/tty_5.4.patch
> >>
> >> Mitch Parks
> >> mitch at kuoi.asui.uidaho.edu
> >> _______________________________________________
> >> freebsd-stable at freebsd.org mailing list
> >> http://lists.freebsd.org/mailman/listinfo/freebsd-stable
> >> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at freebsd.org"
> >
> >
>
>
>

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite at gumbysoft.com          |  www.FreeBSD.org