On Tue, Apr 26, 2011 at 01:45:31PM +0200, Sedat Dilek wrote:
> On Tue, Apr 26, 2011 at 7:06 AM, Paul E. McKenney
> <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
> > On Sun, Apr 24, 2011 at 09:43:31AM -0700, Paul E. McKenney wrote:
> >> On Sun, Apr 24, 2011 at 11:36:44AM +0200, Sedat Dilek wrote:
> >> > On Sun, Apr 24, 2011 at 8:27 AM, Paul E. McKenney
> >> > <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
> >>
> >> [ . . . ]
> >>
> >> > > OK, this looks unrelated, but just in case, could you please try it
> >> > > again with the following patch? (Not mainlinable, debug only.)
> >> > >
> >> > > Also, it does look like you are still seeing a grace-period hang.
> >> > > Could you please send the output of the script? Same one as last time.
> >> > >
> >> > > Thanx, Paul
> >> > >
> >> > > ------------------------------------------------------------------------
> >> > >
> >> > > debugobjects.c | 8 +++++---
> >> > > 1 file changed, 5 insertions(+), 3 deletions(-)
> >> > >
> >> > > diff --git a/lib/debugobjects.c b/lib/debugobjects.c
> >> > > index 9d86e45..10a7c7a 100644
> >> > > --- a/lib/debugobjects.c
> >> > > +++ b/lib/debugobjects.c
> >> > > @@ -289,10 +289,12 @@ static void debug_object_is_on_stack(void *addr, int onstack)
> >> > > return;
> >> > >
> >> > > limit++;
> >> > > - if (is_on_stack)
> >> > > + if (is_on_stack) {
> >> > > + struct rcu_head *p = (struct rcu_head *)addr;
> >> > > printk(KERN_WARNING
> >> > > - "ODEBUG: object is on stack, but not annotated\n");
> >> > > - else
> >> > > + "ODEBUG: object is on stack, but not annotated: %p\n",
> >> > > + p->func);
> >> > > + } else
> >> > > printk(KERN_WARNING
> >> > > "ODEBUG: object is not on stack, but annotated\n");
> >> > > WARN_ON(1);
> >> > >
> >> >
> >> > Somehow your attached patch was not applicable.
> >> > As the changes were a few lines I applied it by myself.
> >> > Attached are log, dmesg and patches (orig + mine)
> >>
> >> Hmmm... Does 0xc10231a1 correspond to a function in your build? If so,
> >> could you please let me know which one?
> >>
> >> OK, so according to "ps" the per-CPU kthread is runnable, but it appears
> >> to never run. You only have one CPU, so it cannot be waiting due to
> >> running on the wrong CPU. The only other loop is in wait_event(), and
> >> that code looks good -- besides, if wait_event() was broken, we would
> >> be seeing breakage everywhere.
> >>
> >> Peter, any thoughts on what I might have done wrong to get the scheduler
> >> into a state where it was ignoring a runnable realtime task?
> >
> > Hello, Sedat,
> >
> > Here is a diagnostic patch to apply on top of sedat.2011.04.23a from
> > the -rcu git tree. Could you please try it out, let me know what
> > happens, and run the last collectdebugfs.sh during the test?
> >
> > Thanx, Paul
> >
> > ------------------------------------------------------------------------
> >
> > diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> > index 6cf6e47..65ae701 100644
> > --- a/kernel/rcutree.c
> > +++ b/kernel/rcutree.c
> > @@ -1524,9 +1524,9 @@ static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
> > return;
> > if (to_rt) {
> > policy = SCHED_NORMAL;
> > - sp.sched_priority = RCU_KTHREAD_PRIO;
> > + sp.sched_priority = 0;
> > } else {
> > - policy = SCHED_FIFO;
> > + policy = SCHED_NORMAL;
> > sp.sched_priority = 0;
> > }
> > sched_setscheduler_nocheck(t, policy, &sp);
> > @@ -1566,8 +1566,8 @@ static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
> > sp.sched_priority = 0;
> > sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
> > schedule();
> > - sp.sched_priority = RCU_KTHREAD_PRIO;
> > - sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
> > + sp.sched_priority = 0;
> > + sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
> > del_timer(&yield_timer);
> > }
> >
> > @@ -1671,8 +1671,8 @@ static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
> > WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
> > per_cpu(rcu_cpu_kthread_task, cpu) = t;
> > wake_up_process(t);
> > - sp.sched_priority = RCU_KTHREAD_PRIO;
> > - sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
> > + sp.sched_priority = 0;
> > + sched_setscheduler_nocheck(t, SCHED_NORMAL, &sp);
> > return 0;
> > }
> >
> > @@ -1713,8 +1713,8 @@ static int rcu_node_kthread(void *arg)
> > continue;
> > }
> > per_cpu(rcu_cpu_has_work, cpu) = 1;
> > - sp.sched_priority = RCU_KTHREAD_PRIO;
> > - sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
> > + sp.sched_priority = 0;
> > + sched_setscheduler_nocheck(t, SCHED_NORMAL, &sp);
> > preempt_enable();
> > }
> > }
> > diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
> > index a21413d..baee185 100644
> > --- a/kernel/rcutree_plugin.h
> > +++ b/kernel/rcutree_plugin.h
> > @@ -1307,8 +1307,8 @@ static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
> > rnp->boost_kthread_task = t;
> > raw_spin_unlock_irqrestore(&rnp->lock, flags);
> > wake_up_process(t);
> > - sp.sched_priority = RCU_KTHREAD_PRIO;
> > - sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
> > + sp.sched_priority = 0;
> > + sched_setscheduler_nocheck(t, SCHED_NORMAL, &sp);
> > return 0;
> > }
> >
> >
>
> Hi Paul,
>
> I have tested with your patch and kept the kernel-config file from
> previous tests (don't get confused by the new name).
> Hope this helps you.
>
> I have some questions to k-c options espcially X86_UP and
> CONFIG_RCU_FANOUT=32 options.
> To what extent can they influence our RCU issue?
> The below options were not set for this round of testing, but I would
> like to have a feedback.
> Thanks in advance.
>
> Would these settings be more optimal for a UP-machine?
>
> # CONFIG_SMP is not set
> # CONFIG_M486 is not set
> CONFIG_M686=y
> CONFIG_NR_CPUS=1
These should be fine.
> CONFIG_X86_UP_APIC=y
> CONFIG_X86_UP_IOAPIC=y
These I don't know about.
> CONFIG_HIGHMEM4G=y
This one seems good for allowing the system to go as long as possible.
> Is CONFIG_RCU_FANOUT=32 OK?
On a UP system, this one doesn't matter.
> With reverting commit 687d7a960aea46e016182c7ce346d62c4dbd0366 ("rcu:
> restrict TREE_RCU to SMP builds with !PREEMPT").
Thank you for trying this one out!
I don't see any sign of a grace-period hang. Did your test complete
correctly?
Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
