Re: [PATCH] sched/isolation: Make NO_HZ_FULL select CPU_ISOLATION

From: Paul E. McKenney
Date: Mon Dec 04 2017 - 12:17:18 EST


On Mon, Dec 04, 2017 at 04:53:15PM +0100, Frederic Weisbecker wrote:
> 2017-12-02 20:24 UTC+01:00, Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>:
> > On Sat, Dec 02, 2017 at 02:59:12PM +0100, Frederic Weisbecker wrote:
> >> 2017-11-30 21:20 UTC+01:00, Paul E. McKenney
> >> <paulmck@xxxxxxxxxxxxxxxxxx>:
> >> > Commit 5c4991e24c69 ("sched/isolation: Split out new
> >> > CONFIG_CPU_ISOLATION=y config from CONFIG_NO_HZ_FULL") can result in
> >> > RCU
> >> > CPU stall warnings when running rcutorture with CONFIG_NO_HZ_FULL_ALL=y
> >> > and CONFIG_CPU_ISOLATION=n. These warnings are caused by RCU's
> >> > grace-period kthreads sleeping for a few jiffies, but never being
> >> > awakened:
> >> >
> >> > [ 116.353432] rcu_preempt kthread starved for 9974 jiffies!
> >> > g4294967208
> >> > +c4294967207 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x402 ->cpu=0
> >> > [ 116.355517] rcu_preempt I 7464 8 2 0x80000000
> >> > [ 116.356543] Call Trace:
> >> > [ 116.357008] __schedule+0x493/0x620
> >> > [ 116.357682] schedule+0x24/0x40
> >> > [ 116.358291] schedule_timeout+0x330/0x3b0
> >> > [ 116.359024] ? preempt_count_sub+0xea/0x140
> >> > [ 116.359806] ? collect_expired_timers+0xb0/0xb0
> >> > [ 116.360660] rcu_gp_kthread+0x6bf/0xef0
> >> >
> >> > This commit therefore makes NO_HZ_FULL select CPU_ISOLATION, which
> >> > prevents this behavior and seems like it was the original intention in
> >> > any case.
> >>
> >> Although CONFIG_NO_HZ should indeed select CONFIG_CPU_ISOLATION, I'm
> >> surprised about this stall. I'm even more surprised that setting
> >> CONFIG_CPU_ISOLATION=y is enough to fix the issue because
> >> CONFIG_NO_HZ_FULL_ALL shortcuts CONFIG_CPU_ISOLATION entirely (which
> >> is not good, but work in progress...).
> >
> > Yes, and after applying this patch, I get failures a few commits
> > later, which appears to be due to other changes that break
> > CONFIG_NO_HZ_FULL_ALL=y. So I have another patch staged that removes
> > CONFIG_NO_HZ_FULL_ALL, on the grounds that no one else has complained,
> > so rcutorture is likely to be the only user, and I don't see the point
> > of having a Kconfig option for only one user.
>
> I would prefer to keep it. It's useful for automated boot testing
> based on configs such as 0-day or -tip test machines. But I'm likely
> to migrate it to isolcpus implementation. Maybe something along the
> lines of CONFIG_CPU_ISOLATION_ALL.

How about instead allowing something like "nohz_full=1-" specify that
all CPUs other than CPU 0 should be nohz_full CPUs? That would shrink
the code by eliminating CONFIG_NO_HZ_FULL_ALL while still allowing
easy automation of that particular scenario.

(Right now, the boot code complains about "nohz_full=1-", which means
that whatever is generating the boot parameters needs to know how many
CPUs there really are, which as you say can be a pain.)

> >> Did you have any nohz_full= or isolcpus= boot options?
> >
> > Replacing CONFIG_NO_HZ_FULL_ALL=y with nohz_full=1-7 works, that
> > is CONFIG_NO_HZ_FULL=y, CONFIG_NO_HZ_FULL_ALL=n, and nohz_full=1-7
> > on an eight-CPU test.
> >
> > But it is relatively easy to test. Running the rcutorture TREE04
> > scenario on a four-socket x86 gets me RCU CPU stall warnings within
> > a few minutes more than half the time. ;-)
>
> Indeed I managed to trigger something. If it's the same thing I should
> be able to track down the root cause.
>
> [ 123.907557] ??? Writer stall state RTWS_STUTTER(8) g160 c160 f0x0
> ->state 0x1 cpu 2
> [ 123.915184] rcu_torture_wri S 0 111 2 0x80080000
> [ 123.920673] Call Trace:
> [ 123.923096] ? __schedule+0x2bf/0xbb0
> [ 123.926715] ? _raw_spin_unlock_irqrestore+0x59/0x70
> [ 123.931657] schedule+0x3c/0x90
> [ 123.934777] schedule_timeout+0x1e1/0x560

It might well be the same thing, as this schedule_timeout() does look
familiar. I have some diagnostic patches in -rcu, please see below
for the overall effect.

Thanx, Paul

> [ 123.938785] ? __next_timer_interrupt+0xd0/0xd0
> [ 123.943276] stutter_wait+0xc5/0xe0
> [ 123.946738] rcu_torture_writer+0x1ae/0x730
> [ 123.950912] ? rcu_torture_pipe_update+0xf0/0xf0
> [ 123.955491] kthread+0x15f/0x1a0
> [ 123.958702] ? kthread_unpark+0x60/0x60
> [ 123.962523] ret_from_fork+0x24/0x30
> [ 123.966091] rcu_preempt: wait state: 1 ->state: 0x402
> [ 123.971112] rcu_sched: wait state: 1 ->state: 0x402
> [ 123.975953] rcu_bh: wait state: 1 ->state: 0x402

------------------------------------------------------------------------

diff --git a/kernel/time/timer.c b/kernel/time/timer.c
index ffebcf878fba..23af27461d8c 100644
--- a/kernel/time/timer.c
+++ b/kernel/time/timer.c
@@ -1755,8 +1755,13 @@ static void process_timeout(struct timer_list *t)
*/
signed long __sched schedule_timeout(signed long timeout)
{
+ struct timer_base *base;
struct process_timer timer;
unsigned long expire;
+ unsigned long flags;
+ unsigned long i;
+ unsigned int idx, idx_now;
+ unsigned long j;

switch (timeout)
{
@@ -1793,6 +1798,17 @@ signed long __sched schedule_timeout(signed long timeout)
timer_setup_on_stack(&timer.timer, process_timeout, 0);
__mod_timer(&timer.timer, expire, 0);
schedule();
+ j = jiffies;
+ if (timeout < 5 && time_after(j, expire + 8 * HZ) && timer_pending(&timer.timer)) {
+ base = lock_timer_base(&timer.timer, &flags);
+ idx = timer_get_idx(&timer.timer);
+ idx_now = calc_wheel_index(j, base->clk);
+ raw_spin_unlock_irqrestore(&base->lock, flags);
+ pr_info("%s: Waylayed timer base->clk: %#lx jiffies: %#lx base->next_expiry: %#lx timer->flags: %#x timer->expires %#lx idx: %x idx_now: %x base->pending_map ", __func__, base->clk, j, base->next_expiry, timer.timer.flags, timer.timer.expires, idx, idx_now);
+ for (i = 0; i < WHEEL_SIZE / sizeof(base->pending_map[0]) / 8; i++)
+ pr_cont("%016lx", base->pending_map[i]);
+ pr_cont("\n");
+ }
del_singleshot_timer_sync(&timer.timer);

/* Remove the timer from the object tracker */