On Wed, 30 Sep 2015 08:41:57 +0200
Sedat Dilek [off-list ref] wrote:

quoted

What shall I do... play with lockdep (print_irqtrace_events) in
del_timer_sync()?

I recalled that when Jiri was thinking towards a "compiler bug" that
the part in del_timer_sync() emebedded in the "ifdef CONFIG_LOCKDEP"
is somehow "mis-compiled".
Furthermore, I see that try_to_del_timer_sync() is invoked within
del_timer_sync() which does a spin_unlock_irqrestore().

Which is fine.

[ kernel/time/timer.c ]

int del_timer_sync(struct timer_list *timer)
{
#ifdef CONFIG_LOCKDEP
unsigned long flags;

/*
* If lockdep gives a backtrace here, please reference
* the synchronization rules above.
*/
local_irq_save(flags);
lock_map_acquire(&timer->lockdep_map);
lock_map_release(&timer->lockdep_map);
local_irq_restore(flags);
#endif
/*
* don't use it in hardirq context, because it
* could lead to deadlock.
*/
WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
for (;;) {
int ret = try_to_del_timer_sync(timer);
if (ret >= 0)
return ret;
cpu_relax();
}
}
EXPORT_SYMBOL(del_timer_sync);
#endif
...
int try_to_del_timer_sync(struct timer_list *timer)
{
struct tvec_base *base;
unsigned long flags;
int ret = -1;

debug_assert_init(timer);

base = lock_timer_base(timer, &flags);

lock_timer_base() will grab the base->lock and save the current state
of interrupts in flags, and then disable interrupts.

if (base->running_timer != timer) {
timer_stats_timer_clear_start_info(timer);
ret = detach_if_pending(timer, base, true);
}
spin_unlock_irqrestore(&base->lock, flags);

this will release the base->lock and put the interrupt state back to
what it was before it took the lock.

I still don't see anything wrong with the code.

-- Steve

return ret;
}
EXPORT_SYMBOL(try_to_del_timer_sync);
...

Is the attached patch suitable for a test?

I remember I tried to turn lockdep kernel-config for amd64 but that
was somehow tricky.

Shall I ask timer folks?

Thoughts?

Thanks!

- Sedat -