Peter Zijlstra:
> On Mon, Sep 03, 2018 at 10:54:23AM +0200, Peter Zijlstra wrote:
>> On Mon, Sep 03, 2018 at 09:38:15AM +0200, Thomas Gleixner wrote:
>>> On Mon, 3 Sep 2018, Peter Zijlstra wrote:
>>>> On Sat, Sep 01, 2018 at 11:51:26AM +0930, Kevin Shanahan wrote:
>>>>> commit 01548f4d3e8e94caf323a4f664eb347fd34a34ab
>>>>> Author: Martin Schwidefsky <schwidef...@de.ibm.com>
>>>>> Date: Tue Aug 18 17:09:42 2009 +0200
>>>>>
>>>>> clocksource: Avoid clocksource watchdog circular locking dependency
>>>>>
>>>>> stop_machine from a multithreaded workqueue is not allowed because
>>>>> of a circular locking dependency between cpu_down and the workqueue
>>>>> execution. Use a kernel thread to do the clocksource downgrade.
>>>>
>>>> I cannot find stop_machine usage there; either it went away or I need to
>>>> like wake up.
>>>
>>> timekeeping_notify() which is involved in switching clock source uses stomp
>>> machine.
>>
>> ARGH... OK, lemme see if I can come up with something other than
>> endlessly spawning that kthread.
>>
>> A special purpose kthread_worker would make more sense than that.
>
> Can someone test this?
>
> ---
> kernel/time/clocksource.c | 28 ++++++++++++++++++++++------
> 1 file changed, 22 insertions(+), 6 deletions(-)
>
> diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
> index f74fb00d8064..898976d0082a 100644
> --- a/kernel/time/clocksource.c
> +++ b/kernel/time/clocksource.c
> @@ -112,13 +112,28 @@ static int finished_booting;
> static u64 suspend_start;
>
> #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
> -static void clocksource_watchdog_work(struct work_struct *work);
> +static void clocksource_watchdog_work(struct kthread_work *work);
> static void clocksource_select(void);
>
> static LIST_HEAD(watchdog_list);
> static struct clocksource *watchdog;
> static struct timer_list watchdog_timer;
> -static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
> +
> +/*
> + * We must use a kthread_worker here, because:
> + *
> + * clocksource_watchdog_work()
> + * clocksource_select()
> + * __clocksource_select()
> + * timekeeping_notify()
> + * stop_machine()
> + *
> + * cannot be called from a reqular workqueue, because of deadlocks between
> + * workqueue and stopmachine.
> + */
> +static struct kthread_worker *watchdog_worker;
> +static DEFINE_KTHREAD_WORK(watchdog_work, clocksource_watchdog_work);
> +
> static DEFINE_SPINLOCK(watchdog_lock);
> static int watchdog_running;
> static atomic_t watchdog_reset_pending;
> @@ -158,7 +173,7 @@ static void __clocksource_unstable(struct clocksource *cs)
>
> /* kick clocksource_watchdog_work() */
> if (finished_booting)
> - schedule_work(&watchdog_work);
> + kthread_queue_work(watchdog_worker, &watchdog_work);
> }
>
> /**
> @@ -199,7 +214,7 @@ static void clocksource_watchdog(struct timer_list
> *unused)
> /* Clocksource already marked unstable? */
> if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
> if (finished_booting)
> - schedule_work(&watchdog_work);
> + kthread_queue_work(watchdog_worker,
> &watchdog_work);
> continue;
> }
>
> @@ -269,7 +284,7 @@ static void clocksource_watchdog(struct timer_list
> *unused)
> */
> if (cs != curr_clocksource) {
> cs->flags |= CLOCK_SOURCE_RESELECT;
> - schedule_work(&watchdog_work);
> + kthread_queue_work(watchdog_worker,
> &watchdog_work);
> } else {
> tick_clock_notify();
> }
> @@ -418,7 +433,7 @@ static int __clocksource_watchdog_work(void)
> return select;
> }
>
> -static void clocksource_watchdog_work(struct work_struct *work)
> +static void clocksource_watchdog_work(struct kthread_work *work)
> {
> mutex_lock(&clocksource_mutex);
> if (__clocksource_watchdog_work())
> @@ -806,6 +821,7 @@ static int __init clocksource_done_booting(void)
> {
> mutex_lock(&clocksource_mutex);
> curr_clocksource = clocksource_default_clock();
> + watchdog_worker = kthread_create_worker(0, "cs-watchdog");
> finished_booting = 1;
> /*
> * Run the watchdog first to eliminate unstable clock sources
>
Applied on mainline tag v4.19-rc2. Tested without additional parameters,
with "quiet" and with "debug", my PC booted successfully in all three
cases, whereas it stalled almost always in these three cases before.
Thanks!
