Panic on kern_event.c
Mark Johnston
markj at freebsd.org
Sun Nov 18 00:35:59 UTC 2018
On Fri, Nov 16, 2018 at 05:56:54PM +0100, Sylvain GALLIANO wrote:
> Le ven. 16 nov. 2018 à 16:42, Mark Johnston <markj at freebsd.org> a écrit :
>
> > On Fri, Nov 16, 2018 at 03:47:39PM +0100, Sylvain GALLIANO wrote:
> > > Le jeu. 15 nov. 2018 à 23:10, Mark Johnston <markj at freebsd.org> a écrit
> > :
> > >
> > > > On Thu, Nov 08, 2018 at 05:05:03PM +0100, Sylvain GALLIANO wrote:
> > > > > Hi,
> > > > >
> > > > > I replaced
> > > > > << printf("XXX knote %p already in tailq status:%x kq_count:%d [%p
> > %p]
> > > > >
> > > >
> > %u\n",kn,kn->kn_status,kq->kq_count,kn->kn_tqe.tqe_next,kn->kn_tqe.tqe_prev,__LINE__);
> > > > > by
> > > > > >> panic("XXX knote %p already in tailq status:%x kq_count:%d [%p
> > %p]
> > > > >
> > > >
> > %u\n",kn,kn->kn_status,kq->kq_count,kn->kn_tqe.tqe_next,kn->kn_tqe.tqe_prev,__LINE__);
> > > > >
> > > > > Here is the stack during panic:
> > > > > panic: XXX knote 0xfffff801e1c6ddc0 already in tailq status:1
> > kq_count:2
> > > > > [0 0xfffff8000957a978] 2671
> > > > >
> > > > Could you please give the following patch a try?
> > > >
> > > > If possible, could you also ktrace one of the active syslog-ng
> > processes
> > > > for some time, perhaps 15 seconds, and share the kdump? I have been
> > > > trying to reproduce the problem without any luck.
> > > >
> > > Unfortunately patched kernel is not stable:
> > > - some processes run at 100% CPU (STOP state) and cannot be killed
> > > - sometime the system completely freeze (need a hard reboot)
Could you please give this updated a patch a try?
diff --git a/sys/kern/kern_event.c b/sys/kern/kern_event.c
index d9c670e29d60..f989d612bac4 100644
--- a/sys/kern/kern_event.c
+++ b/sys/kern/kern_event.c
@@ -224,6 +224,7 @@ SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
#define KQ_LOCK(kq) do { \
mtx_lock(&(kq)->kq_lock); \
} while (0)
+#define KQ_LOCKPTR(kq) (&(kq)->kq_lock)
#define KQ_FLUX_WAKEUP(kq) do { \
if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
(kq)->kq_state &= ~KQ_FLUXWAIT; \
@@ -687,8 +688,11 @@ filt_timerexpire(void *knx)
struct kq_timer_cb_data *kc;
kn = knx;
+ KQ_OWNED(kn->kn_kq);
kn->kn_data++;
- KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
+
+ if (!kn_in_flux(kn) || (kn->kn_status & KN_SCAN) != 0)
+ KNOTE_ACTIVATE(kn, 1);
if ((kn->kn_flags & EV_ONESHOT) != 0)
return;
@@ -753,7 +757,7 @@ filt_timerattach(struct knote *kn)
kn->kn_flags |= EV_CLEAR; /* automatically set */
kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */
kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
- callout_init(&kc->c, 1);
+ callout_init_mtx(&kc->c, KQ_LOCKPTR(kn->kn_kq), 0);
filt_timerstart(kn, to);
return (0);
@@ -772,8 +776,10 @@ filt_timerstart(struct knote *kn, sbintime_t to)
kc->next = to + sbinuptime();
kc->to = to;
}
+ KQ_LOCK(kn->kn_kq);
callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn,
PCPU_GET(cpuid), C_ABSOLUTE);
+ KQ_UNLOCK(kn->kn_kq);
}
static void
@@ -826,7 +832,6 @@ filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
KQ_LOCK(kq);
if (kn->kn_status & KN_QUEUED)
knote_dequeue(kn);
-
kn->kn_status &= ~KN_ACTIVE;
kn->kn_data = 0;
KQ_UNLOCK(kq);
@@ -1843,12 +1848,13 @@ kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
}
TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
+ marker->kn_status |= KN_QUEUED;
influx = 0;
while (count) {
KQ_OWNED(kq);
kn = TAILQ_FIRST(&kq->kq_head);
- if ((kn->kn_status == KN_MARKER && kn != marker) ||
+ if (((kn->kn_status & KN_MARKER) != 0 && kn != marker) ||
kn_in_flux(kn)) {
if (influx) {
influx = 0;
@@ -1861,24 +1867,21 @@ kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
}
TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
- if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
- kn->kn_status &= ~KN_QUEUED;
- kq->kq_count--;
+ kn->kn_status &= ~KN_QUEUED;
+ if ((kn->kn_status & KN_DISABLED) == KN_DISABLED)
continue;
- }
if (kn == marker) {
KQ_FLUX_WAKEUP(kq);
if (count == maxevents)
goto retry;
goto done;
}
+ kq->kq_count--;
KASSERT(!kn_in_flux(kn),
("knote %p is unexpectedly in flux", kn));
if ((kn->kn_flags & EV_DROP) == EV_DROP) {
- kn->kn_status &= ~KN_QUEUED;
kn_enter_flux(kn);
- kq->kq_count--;
KQ_UNLOCK(kq);
/*
* We don't need to lock the list since we've
@@ -1888,9 +1891,7 @@ kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
KQ_LOCK(kq);
continue;
} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
- kn->kn_status &= ~KN_QUEUED;
kn_enter_flux(kn);
- kq->kq_count--;
KQ_UNLOCK(kq);
/*
* We don't need to lock the list since we've
@@ -1910,10 +1911,9 @@ kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
if (kn->kn_fop->f_event(kn, 0) == 0) {
KQ_LOCK(kq);
KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
- kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
- KN_SCAN);
+ /* XXXMJ races with EVFILT_TIMER activation. */
+ kn->kn_status &= ~(KN_ACTIVE | KN_SCAN);
kn_leave_flux(kn);
- kq->kq_count--;
kn_list_unlock(knl);
influx = 1;
continue;
@@ -1937,11 +1937,10 @@ kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
}
if (kn->kn_flags & EV_DISPATCH)
kn->kn_status |= KN_DISABLED;
- kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
- kq->kq_count--;
- } else
- TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
-
+ kn->kn_status &= ~KN_ACTIVE;
+ } else if ((kn->kn_status & KN_QUEUED) == 0)
+ knote_enqueue(kn);
+
kn->kn_status &= ~KN_SCAN;
kn_leave_flux(kn);
kn_list_unlock(knl);
More information about the freebsd-current
mailing list