Re: [RFC][PATCH] rcu: Hotplug and PROVE_RCU_DELAY not playing welltogether

From: Paul E. McKenney
Date: Sun Jun 02 2013 - 06:25:07 EST

Next message: anish singh: "Re: [PATCH] [RFC]Watchdog:core: constant pinging until userspacetimesout when delay very less"
Previous message: Borislav Petkov: "Re: [PATCH] x86, microcode, AMD: Cocci spatch "memdup.spatch""
Next in thread: Paul E. McKenney: "Re: [RFC][PATCH] rcu: Hotplug and PROVE_RCU_DELAY not playing welltogether"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Fri, May 31, 2013 at 05:27:49PM -0400, Steven Rostedt wrote:
> Paul,
>
> I've been debugging the last couple of days why my tests have been
> locking up. One of my tracing tests, runs all available tracers. The
> lockup always happened with the mmiotrace, which is used to trace
> interactions between priority drivers and the kernel. But to do this
> easily, when the tracer gets registered, it disables all but the boot
> CPUs. The lockup always happened after it got done disabling the CPUs.
>
> Then I decided to try this:
>
> while :; do
> for i in 1 2 3; do
> echo 0 > /sys/devices/system/cpu/cpu$i/online
> done
> for i in 1 2 3; do
> echo 1 > /sys/devices/system/cpu/cpu$i/online
> done
> done
>
> Well, sure enough, that locked up too, with the same users. Doing a
> sysrq-w (showing all blocked tasks):

Impressive debugging!!! And that is what I call one gnarly deadlock!

Your patch looks like it should fix the problem, but my immediate
reaction was that it would be simpler to have rcu_gp_init()
do either cpu_maps_update_begin(), get_online_cpus(), or
cpu_hotplug_begin() if CONFIG_PROVE_RCU_DELAY instead of the
current mutex_lock(&rsp->onoff_mutex). (My first choice would be
get_online_cpus(), but I am not sure that I fully understand the
deadlock.)

Or am I missing something about the nature of this deadlock?

One concern is that if I made that change, and if any hotplug notifier
waited for a grace period, there would be another deadlock. Which
might well be why this acquires ->onoff_lock. Hmmm...

OK, another possible simplification would be to use udelay() or something
similar to do the waiting, and maybe dial down the delay from the current
two jiffies to (say) 200 microseconds. I could adjust the "if" condition
to make the delay more probable to get roughly the same testing intensity
as the current code has.

Other thoughts?

Thanx, Paul

> [ 2991.344562] task PC stack pid father
> [ 2991.344562] rcu_preempt D ffff88007986fdf8 0 10 2 0x00000000
> [ 2991.344562] ffff88007986fc98 0000000000000002 ffff88007986fc48 0000000000000908
> [ 2991.344562] ffff88007986c280 ffff88007986ffd8 ffff88007986ffd8 00000000001d3c80
> [ 2991.344562] ffff880079248a40 ffff88007986c280 0000000000000000 00000000fffd4295
> [ 2991.344562] Call Trace:
> [ 2991.344562] [<ffffffff815437ba>] schedule+0x64/0x66
> [ 2991.344562] [<ffffffff81541750>] schedule_timeout+0xbc/0xf9
> [ 2991.344562] [<ffffffff8154bec0>] ? ftrace_call+0x5/0x2f
> [ 2991.344562] [<ffffffff81049513>] ? cascade+0xa8/0xa8
> [ 2991.344562] [<ffffffff815417ab>] schedule_timeout_uninterruptible+0x1e/0x20
> [ 2991.344562] [<ffffffff810c980c>] rcu_gp_kthread+0x502/0x94b
> [ 2991.344562] [<ffffffff81062791>] ? __init_waitqueue_head+0x50/0x50
> [ 2991.344562] [<ffffffff810c930a>] ? rcu_gp_fqs+0x64/0x64
> [ 2991.344562] [<ffffffff81061cdb>] kthread+0xb1/0xb9
> [ 2991.344562] [<ffffffff81091e31>] ? lock_release_holdtime.part.23+0x4e/0x55
> [ 2991.344562] [<ffffffff81061c2a>] ? __init_kthread_worker+0x58/0x58
> [ 2991.344562] [<ffffffff8154c1dc>] ret_from_fork+0x7c/0xb0
> [ 2991.344562] [<ffffffff81061c2a>] ? __init_kthread_worker+0x58/0x58
> [ 2991.344562] kworker/0:1 D ffffffff81a30680 0 47 2 0x00000000
> [ 2991.344562] Workqueue: events cpuset_hotplug_workfn
> [ 2991.344562] ffff880078dbbb58 0000000000000002 0000000000000006 00000000000000d8
> [ 2991.344562] ffff880078db8100 ffff880078dbbfd8 ffff880078dbbfd8 00000000001d3c80
> [ 2991.344562] ffff8800779ca5c0 ffff880078db8100 ffffffff81541fcf 0000000000000000
> [ 2991.344562] Call Trace:
> [ 2991.344562] [<ffffffff81541fcf>] ? __mutex_lock_common+0x3d4/0x609
> [ 2991.344562] [<ffffffff815437ba>] schedule+0x64/0x66
> [ 2991.344562] [<ffffffff81543a39>] schedule_preempt_disabled+0x18/0x24
> [ 2991.344562] [<ffffffff81541fcf>] __mutex_lock_common+0x3d4/0x609
> [ 2991.344562] [<ffffffff8103d11b>] ? get_online_cpus+0x3c/0x50
> [ 2991.344562] [<ffffffff8103d11b>] ? get_online_cpus+0x3c/0x50
> [ 2991.344562] [<ffffffff815422ff>] mutex_lock_nested+0x3b/0x40
> [ 2991.344562] [<ffffffff8103d11b>] get_online_cpus+0x3c/0x50
> [ 2991.344562] [<ffffffff810af7e6>] rebuild_sched_domains_locked+0x6e/0x3a8
> [ 2991.344562] [<ffffffff810b0ec6>] rebuild_sched_domains+0x1c/0x2a
> [ 2991.344562] [<ffffffff810b109b>] cpuset_hotplug_workfn+0x1c7/0x1d3
> [ 2991.344562] [<ffffffff810b0ed9>] ? cpuset_hotplug_workfn+0x5/0x1d3
> [ 2991.344562] [<ffffffff81058e07>] process_one_work+0x2d4/0x4d1
> [ 2991.344562] [<ffffffff81058d3a>] ? process_one_work+0x207/0x4d1
> [ 2991.344562] [<ffffffff8105964c>] worker_thread+0x2e7/0x3b5
> [ 2991.344562] [<ffffffff81059365>] ? rescuer_thread+0x332/0x332
> [ 2991.344562] [<ffffffff81061cdb>] kthread+0xb1/0xb9
> [ 2991.344562] [<ffffffff81061c2a>] ? __init_kthread_worker+0x58/0x58
> [ 2991.344562] [<ffffffff8154c1dc>] ret_from_fork+0x7c/0xb0
> [ 2991.344562] [<ffffffff81061c2a>] ? __init_kthread_worker+0x58/0x58
> [ 2991.344562] bash D ffffffff81a4aa80 0 2618 2612 0x10000000
> [ 2991.344562] ffff8800379abb58 0000000000000002 0000000000000006 0000000000000c2c
> [ 2991.344562] ffff880077fea140 ffff8800379abfd8 ffff8800379abfd8 00000000001d3c80
> [ 2991.344562] ffff8800779ca5c0 ffff880077fea140 ffffffff81541fcf 0000000000000000
> [ 2991.344562] Call Trace:
> [ 2991.344562] [<ffffffff81541fcf>] ? __mutex_lock_common+0x3d4/0x609
> [ 2991.344562] [<ffffffff815437ba>] schedule+0x64/0x66
> [ 2991.344562] [<ffffffff81543a39>] schedule_preempt_disabled+0x18/0x24
> [ 2991.344562] [<ffffffff81541fcf>] __mutex_lock_common+0x3d4/0x609
> [ 2991.344562] [<ffffffff81530078>] ? rcu_cpu_notify+0x2f5/0x86e
> [ 2991.344562] [<ffffffff81530078>] ? rcu_cpu_notify+0x2f5/0x86e
> [ 2991.344562] [<ffffffff815422ff>] mutex_lock_nested+0x3b/0x40
> [ 2991.344562] [<ffffffff81530078>] rcu_cpu_notify+0x2f5/0x86e
> [ 2991.344562] [<ffffffff81091c99>] ? __lock_is_held+0x32/0x53
> [ 2991.344562] [<ffffffff81548912>] notifier_call_chain+0x6b/0x98
> [ 2991.344562] [<ffffffff810671fd>] __raw_notifier_call_chain+0xe/0x10
> [ 2991.344562] [<ffffffff8103cf64>] __cpu_notify+0x20/0x32
> [ 2991.344562] [<ffffffff8103cf8d>] cpu_notify_nofail+0x17/0x36
> [ 2991.344562] [<ffffffff815225de>] _cpu_down+0x154/0x259
> [ 2991.344562] [<ffffffff81522710>] cpu_down+0x2d/0x3a
> [ 2991.344562] [<ffffffff81526351>] store_online+0x4e/0xe7
> [ 2991.344562] [<ffffffff8134d764>] dev_attr_store+0x20/0x22
> [ 2991.344562] [<ffffffff811b3c5f>] sysfs_write_file+0x108/0x144
> [ 2991.344562] [<ffffffff8114c5ef>] vfs_write+0xfd/0x158
> [ 2991.344562] [<ffffffff8114c928>] SyS_write+0x5c/0x83
> [ 2991.344562] [<ffffffff8154c494>] tracesys+0xdd/0xe2
>
> As well as held locks:
>
> [ 3034.728033] Showing all locks held in the system:
> [ 3034.728033] 1 lock held by rcu_preempt/10:
> [ 3034.728033] #0: (rcu_preempt_state.onoff_mutex){+.+...}, at: [<ffffffff810c9471>] rcu_gp_kthread+0x167/0x94b
> [ 3034.728033] 4 locks held by kworker/0:1/47:
> [ 3034.728033] #0: (events){.+.+.+}, at: [<ffffffff81058d3a>] process_one_work+0x207/0x4d1
> [ 3034.728033] #1: (cpuset_hotplug_work){+.+.+.}, at: [<ffffffff81058d3a>] process_one_work+0x207/0x4d1
> [ 3034.728033] #2: (cpuset_mutex){+.+.+.}, at: [<ffffffff810b0ec1>] rebuild_sched_domains+0x17/0x2a
> [ 3034.728033] #3: (cpu_hotplug.lock){+.+.+.}, at: [<ffffffff8103d11b>] get_online_cpus+0x3c/0x50
> [ 3034.728033] 1 lock held by mingetty/2563:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
> [ 3034.728033] 1 lock held by mingetty/2565:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
> [ 3034.728033] 1 lock held by mingetty/2569:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
> [ 3034.728033] 1 lock held by mingetty/2572:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
> [ 3034.728033] 1 lock held by mingetty/2575:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
> [ 3034.728033] 7 locks held by bash/2618:
> [ 3034.728033] #0: (sb_writers#5){.+.+.+}, at: [<ffffffff8114bc3f>] file_start_write+0x2a/0x2c
> [ 3034.728033] #1: (&buffer->mutex#2){+.+.+.}, at: [<ffffffff811b3b93>] sysfs_write_file+0x3c/0x144
> [ 3034.728033] #2: (s_active#54){.+.+.+}, at: [<ffffffff811b3c3e>] sysfs_write_file+0xe7/0x144
> [ 3034.728033] #3: (x86_cpu_hotplug_driver_mutex){+.+.+.}, at: [<ffffffff810217c2>] cpu_hotplug_driver_lock+0x17/0x19
> [ 3034.728033] #4: (cpu_add_remove_lock){+.+.+.}, at: [<ffffffff8103d196>] cpu_maps_update_begin+0x17/0x19
> [ 3034.728033] #5: (cpu_hotplug.lock){+.+.+.}, at: [<ffffffff8103cfd8>] cpu_hotplug_begin+0x2c/0x6d
> [ 3034.728033] #6: (rcu_preempt_state.onoff_mutex){+.+...}, at: [<ffffffff81530078>] rcu_cpu_notify+0x2f5/0x86e
> [ 3034.728033] 1 lock held by bash/2980:
> [ 3034.728033] #0: (&ldata->atomic_read_lock){+.+...}, at: [<ffffffff8131e28a>] n_tty_read+0x252/0x7e8
>
> Things looked a little weird. Also, this is a deadlock that lockdep did
> not catch. But what we have here does not look like a circular lock
> issue:
>
> Bash is blocked in rcu_cpu_notify():
>
> 1961 /* Exclude any attempts to start a new grace period. */
> 1962 mutex_lock(&rsp->onoff_mutex);
>
>
> kworker is blocked in get_online_cpus(), which makes sense as we are
> currently taking down a CPU.
>
> But rcu_preempt is not blocked on anything. It is simply sleeping in
> rcu_gp_kthread (really rcu_gp_init) here:
>
> 1453 #ifdef CONFIG_PROVE_RCU_DELAY
> 1454 if ((prandom_u32() % (rcu_num_nodes * 8)) == 0 &&
> 1455 system_state == SYSTEM_RUNNING)
> 1456 schedule_timeout_uninterruptible(2);
> 1457 #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
>
> And it does this while holding the onoff_mutex that bash is waiting for.
>
> Doing a function trace, it showed me where it happened:
>
> [ 125.940066] rcu_pree-10 3.... 28384115273: schedule_timeout_uninterruptible <-rcu_gp_kthread
> [...]
> [ 125.940066] rcu_pree-10 3d..3 28384202439: sched_switch: prev_comm=rcu_preempt prev_pid=10 prev_prio=120 prev_state=D ==> next_comm=watchdog/3 next_pid=38 next_prio=120
>
> The watchdog ran, and then:
>
> [ 125.940066] watchdog-38 3d..3 28384692863: sched_switch: prev_comm=watchdog/3 prev_pid=38 prev_prio=120 prev_state=P ==> next_comm=modprobe next_pid=2848 next_prio=118
>
> Not sure what modprobe was doing, but shortly after that:
>
> [ 125.940066] modprobe-2848 3d..3 28385041749: sched_switch: prev_comm=modprobe prev_pid=2848 prev_prio=118 prev_state=R+ ==> next_comm=migration/3 next_pid=40 next_prio=0
>
> Where the migration thread took down the CPU:
>
> [ 125.940066] migratio-40 3d..3 28389148276: sched_switch: prev_comm=migration/3 prev_pid=40 prev_prio=0 prev_state=P ==> next_comm=swapper/3 next_pid=0 next_prio=120
>
> which finally did:
>
> [ 125.940066] <idle>-0 3...1 28389282142: arch_cpu_idle_dead <-cpu_startup_entry
> [ 125.940066] <idle>-0 3...1 28389282548: native_play_dead <-arch_cpu_idle_dead
> [ 125.940066] <idle>-0 3...1 28389282924: play_dead_common <-native_play_dead
> [ 125.940066] <idle>-0 3...1 28389283468: idle_task_exit <-play_dead_common
> [ 125.940066] <idle>-0 3...1 28389284644: amd_e400_remove_cpu <-play_dead_common
>
>
> CPU 3 is now offline, the rcu_preempt thread that ran on CPU 3 is still
> doing a schedule_timeout_uninterruptible() and it registered it's
> timeout to the timer base for CPU 3. You would think that it would get
> migrated right? The issue here is that the timer migration happens at
> the CPU notifier for CPU_DEAD. The problem is that the rcu notifier for
> CPU_DOWN is blocked waiting for the onoff_mutex to be released, which is
> held by the thread that just put itself into a uninterruptible sleep,
> that wont wake up until the CPU_DEAD notifier of the timer
> infrastructure is called, which wont happen until the rcu notifier
> finishes. Here's our deadlock!
>
> Not sure how to solve this. I added this hack. Although, even though
> it's a hack, it's a fix that's for a hack that is used for debugging
> purposes only.
>
> I added a waitqueue and a flag. The flag gets set when a cpu is going
> down and cleared after it is down or dead. The rcu_preempt thread will
> add itself to the waitqueue and then check if any CPU is going down. If
> one is, it wont do the schedule, if one is not, then it does the
> schedule. When a cpu goes down, the rcu cpu notifier will wake up the
> thread before it does more work.
>
> This patch does prevent the issue from occurring.
>
> Signed-off-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
>
> Index: linux-trace.git/kernel/rcutree.c
> ===================================================================
> --- linux-trace.git.orig/kernel/rcutree.c
> +++ linux-trace.git/kernel/rcutree.c
> @@ -1396,6 +1396,34 @@ rcu_start_gp_per_cpu(struct rcu_state *r
> __note_new_gpnum(rsp, rnp, rdp);
> }
>
> +#ifdef CONFIG_PROVE_RCU_DELAY
> +/* Need to kick rcu_preempt kthread if cpu is going down */
> +static DECLARE_WAIT_QUEUE_HEAD(rcu_delay_wait);
> +static bool rcu_delay_cpu_going_down;
> +static void rcu_delay_down_done(void)
> +{
> + rcu_delay_cpu_going_down = false;
> +}
> +static void rcu_delay_down(void)
> +{
> + rcu_delay_cpu_going_down = true;
> + /* Make sure the rcu_preempt thread see this set */
> + smp_wmb();
> + /*
> + * We may wake up rcu_preempt threads for other CPUs, but
> + * that's OK. It's sleeping for debug purposes only.
> + */
> + wake_up_interruptible(&rcu_delay_wait);
> +}
> +#else
> +static inline void rcu_delay_down_done(void)
> +{
> +}
> +static inline void rcu_delay_down(void)
> +{
> +}
> +#endif
> +
> /*
> * Initialize a new grace period.
> */
> @@ -1452,8 +1480,26 @@ static int rcu_gp_init(struct rcu_state
> raw_spin_unlock_irq(&rnp->lock);
> #ifdef CONFIG_PROVE_RCU_DELAY
> if ((prandom_u32() % (rcu_num_nodes * 8)) == 0 &&
> - system_state == SYSTEM_RUNNING)
> - schedule_timeout_uninterruptible(2);
> + system_state == SYSTEM_RUNNING) {
> + DECLARE_WAITQUEUE(wait, current);
> + /*
> + * If the current CPU goes offline, the rcu_preempt may
> + * never wake up from the schedule_timeout(). This is
> + * because it holds the onoff_mutex which gets taken
> + * by the RCU CPU down notifier. This will prevent the timer
> + * notifier from migrating the rcu_preempt and letting
> + * it wake up, causing a deadlock.
> + * Thus we have this hacky waitqueue and flag to make sure
> + * that if the CPU goes down, we either skip the
> + * schedule_timeout or we wake up the rcu_preempt thread.
> + */
> + set_current_state(TASK_UNINTERRUPTIBLE);
> + add_wait_queue(&rcu_delay_wait, &wait);
> + if (!rcu_delay_cpu_going_down)
> + schedule_timeout(2);
> + __set_current_state(TASK_RUNNING);
> + remove_wait_queue(&rcu_delay_wait, &wait);
> + }
> #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
> cond_resched();
> }
> @@ -3074,8 +3120,10 @@ static int __cpuinit rcu_cpu_notify(stru
> case CPU_ONLINE:
> case CPU_DOWN_FAILED:
> rcu_boost_kthread_setaffinity(rnp, -1);
> + rcu_delay_down_done();
> break;
> case CPU_DOWN_PREPARE:
> + rcu_delay_down();
> rcu_boost_kthread_setaffinity(rnp, cpu);
> break;
> case CPU_DYING:
> @@ -3087,6 +3135,7 @@ static int __cpuinit rcu_cpu_notify(stru
> case CPU_DEAD_FROZEN:
> case CPU_UP_CANCELED:
> case CPU_UP_CANCELED_FROZEN:
> + rcu_delay_down_done();
> for_each_rcu_flavor(rsp)
> rcu_cleanup_dead_cpu(cpu, rsp);
> break;
>
>

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: anish singh: "Re: [PATCH] [RFC]Watchdog:core: constant pinging until userspacetimesout when delay very less"
Previous message: Borislav Petkov: "Re: [PATCH] x86, microcode, AMD: Cocci spatch "memdup.spatch""
Next in thread: Paul E. McKenney: "Re: [RFC][PATCH] rcu: Hotplug and PROVE_RCU_DELAY not playing welltogether"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]