[PATCH] ARM: Don't use complete() during __cpu_die

[Krzysztof Kozlowski]

On śro, 2015-02-04 at 14:42 -0800, Stephen Boyd wrote:
> On 02/04/15 08:53, Krzysztof Kozlowski wrote:
> > The complete() should not be used on offlined CPU. Rewrite the
> > wait-complete mechanism with wait_on_bit_timeout().
> >
> > The CPU triggering hot unplug (e.g. CPU0) will loop until some bit is
> > cleared. In each iteration schedule_timeout() is used with initial sleep
> > time of 1 ms.  Later it is increased to 10 ms.
> >
> > The dying CPU will clear the bit which is safe in that context.
> >
> > This fixes following RCU warning on ARMv8 (Exynos 4412, Trats2) during
> > suspend to RAM:
> >
> > [   31.113925] ===============================
> > [   31.113928] [ INFO: suspicious RCU usage. ]
> > [   31.113935] 3.19.0-rc7-next-20150203 #1914 Not tainted
> > [   31.113938] -------------------------------
> > [   31.113943] kernel/sched/fair.c:4740 suspicious rcu_dereference_check() usage!
> > [   31.113946]
> > [   31.113946] other info that might help us debug this:
> > [   31.113946]
> > [   31.113952]
> > [   31.113952] RCU used illegally from offline CPU!
> > [   31.113952] rcu_scheduler_active = 1, debug_locks = 0
> > [   31.113957] 3 locks held by swapper/1/0:
> > [   31.113988]  #0:  ((cpu_died).wait.lock){......}, at: [<c005a114>] complete+0x14/0x44
> > [   31.114012]  #1:  (&p->pi_lock){-.-.-.}, at: [<c004a790>] try_to_wake_up+0x28/0x300
> > [   31.114035]  #2:  (rcu_read_lock){......}, at: [<c004f1b8>] select_task_rq_fair+0x5c/0xa04
> > [   31.114038]
> > [   31.114038] stack backtrace:
> > [   31.114046] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 3.19.0-rc7-next-20150203 #1914
> > [   31.114050] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
> > [   31.114076] [<c0014ce4>] (unwind_backtrace) from [<c0011c30>] (show_stack+0x10/0x14)
> > [   31.114091] [<c0011c30>] (show_stack) from [<c04dc048>] (dump_stack+0x70/0xbc)
> > [   31.114105] [<c04dc048>] (dump_stack) from [<c004f83c>] (select_task_rq_fair+0x6e0/0xa04)
> > [   31.114118] [<c004f83c>] (select_task_rq_fair) from [<c004a83c>] (try_to_wake_up+0xd4/0x300)
> > [   31.114129] [<c004a83c>] (try_to_wake_up) from [<c00598a0>] (__wake_up_common+0x4c/0x80)
> > [   31.114140] [<c00598a0>] (__wake_up_common) from [<c00598e8>] (__wake_up_locked+0x14/0x1c)
> > [   31.114150] [<c00598e8>] (__wake_up_locked) from [<c005a134>] (complete+0x34/0x44)
> > [   31.114167] [<c005a134>] (complete) from [<c04d6ca4>] (cpu_die+0x24/0x84)
> > [   31.114179] [<c04d6ca4>] (cpu_die) from [<c005a508>] (cpu_startup_entry+0x328/0x358)
> > [   31.114189] [<c005a508>] (cpu_startup_entry) from [<40008784>] (0x40008784)
> > [   31.114226] CPU1: shutdown
> >
> > Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com>
> > ---
> 
> Would it be better to use IPIs instead? The IPI handler could even call
> complete() as long as we IPI off the dying CPU to the killing CPU. I
> suppose this could be an extension of the current IPI_COMPLETION that we
> already have.

That sounds good. I wonder where register_ipi_completion() should be
put. This must execute before cpu_die() of dying CPU (if I get it
right). So it cannot be __cpu_die(). One candidate is __cpu_disable()
but it may not be called always. Any ideas?


> 
> >  arch/arm/kernel/smp.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++++---
> >  1 file changed, 49 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
> > index 86ef244c5a24..bb8ff465975f 100644
> > --- a/arch/arm/kernel/smp.c
> > +++ b/arch/arm/kernel/smp.c
> > @@ -213,7 +219,40 @@ int __cpu_disable(void)
> >  	return 0;
> >  }
> >  
> > -static DECLARE_COMPLETION(cpu_died);
> > +/*
> > + * Wait for 5000*1 ms for 'wait_cpu_die' bit to be cleared.
> > + * Actually the real wait time will be longer because of schedule()
> > + * called bit_wait_timeout.
> > + *
> > + * Returns 0 if bit was cleared (CPU died) or non-zero
> > + * otherwise (1 or negative ERRNO).
> > + */
> > +static int wait_for_cpu_die(void)
> > +{
> > +	int retries = 5000, sleep_ms = 1, ret = 0;
> > +
> > +	might_sleep();
> > +
> > +	smp_mb__before_atomic();
> > +	while (test_bit(CPU_DIE_WAIT_BIT, &wait_cpu_die)) {
> > +		ret = out_of_line_wait_on_bit_timeout(&wait_cpu_die,
> > +				CPU_DIE_WAIT_BIT, bit_wait_timeout,
> > +				TASK_UNINTERRUPTIBLE,
> > +				msecs_to_jiffies(sleep_ms));
> > +		if (!ret || (--retries <= 0))
> > +			break;
> > +
> > +		if (retries < 4000) {
> > +			/* After ~1000 ms increase sleeping time to 10 ms */
> > +			retries = 400;
> > +			sleep_ms = 10;
> > +		}
> > +
> > +		smp_mb__before_atomic(); /* For next test_bit() in loop */
> > +	}
> > +
> > +	return ret;
> > +}
> >  
> 
> Is there any reason we test the bit before testing it again in
> out_of_line_wait_on_bit_timeout()? Why can't we just call that function
> in a loop and let it handle checking the bit?

I think there is no strong reason. Maybe except that if bit was cleared
already then we skip the wait preparation steps. I followed the
convention of wait_on_bit in include/linux/wait.h.

I'll remove the test.

> 
> > @@ -267,7 +312,8 @@ void __ref cpu_die(void)
> >  	 * this returns, power and/or clocks can be removed at any point
> >  	 * from this CPU and its cache by platform_cpu_kill().
> >  	 */
> > -	complete(&cpu_died);
> > +	clear_bit(CPU_DIE_WAIT_BIT, &wait_cpu_die);
> > +	smp_mb__after_atomic();
> >  
> >  	/*
> >  	 * Ensure that the cache lines associated with that completion are
> 
> This comment here should be updated because the completion is gone.

Yes, indeed. I'll update it.

Thanks for feedback,
Krzysztof