Re: [PATCH] exit: Use read lock for do_notify_parent() instead of write lock

[Kirill Tkhai <ktkhai@parallels.com>]

Hi, Oleg,

thanks for your review.

В Пт, 10/04/2015 в 19:50 +0200, Oleg Nesterov пишет:
> Kirill,
> 
> I'll try to read this patch tomorrow, currently I am hopelessly buried
> in user-space programming :/
> 
> But I have to admit that so far I dislike this patch very much. It adds
> a lot of complications and for what?
> 
> Yes, yes, yes. tasklist_lock is another BKL and must die. We need the
> per-process lock. Until then I do not think the hacks like this make
> any sense, unless you have the "real" workload with before/after
> performance numbers.

I don't think the complication is very huge. We add one rule about
exit_state. Yes, the state becomes unstable under read_lock(), but
only wait_consider_task() is affected by this.

Ok, what do you mean when you're speaking about killing tasklist_lock?
Can't we leave it for fork() and __unhash_process() only, but change
other places which lock it for write? Every of the places will get rid
of it by its own way. EXIT_NOTIFY is for do_exit().

Or you want to kill it completelly?

I didn't test the patch on special workload or large SMP systems.
The results for 4 CPU box (kernel compilation):

[origin]
1)534.37user 32.15system 2:29.32elapsed 379%CPU (0avgtext+0avgdata 142488maxresident)k
0inputs+724264outputs (0major+23852891minor)pagefaults 0swaps
2)534.85user 32.81system 2:28.67elapsed 381%CPU (0avgtext+0avgdata 142476maxresident)k
0inputs+724264outputs (0major+23853531minor)pagefaults 0swaps

[patched]
1)531.65user 32.69system 2:27.41elapsed 382%CPU (0avgtext+0avgdata 142580maxresident)k
0inputs+724256outputs (0major+23854620minor)pagefaults 0swaps
2)530.92user 32.51system 2:28.18elapsed 380%CPU (0avgtext+0avgdata 142544maxresident)k
0inputs+724256outputs (0major+23852925minor)pagefaults 0swaps

My test machine has HDD, so it's not the best test for the patch. I'll try something
else later. But I don't expect exciting results on workloads like this.

> 
> On 04/09, Kirill Tkhai wrote:
> >
> > I suggest to execute do_notify_parent() under read_lock(). It allows more tasks
> > to use it in parallel. Read lock gives enough guarantees for us: child's parent
> > won't change during the notification.
> 
> Well, write_unlock() + read_lock() is not nice too...
> 
> > include/asm-generic/qrwlock.h:
> > static inline void queue_reduce_locked_write_to_read(struct qrwlock *lock)
> > {
> > 	smp_mb__before_atomic();
> > 	atomic_add(_QR_BIAS - _QW_LOCKED, &lock->cnts);
> > }
> 
> Yes, downgrade() will be better.
> 
> Still, this only removes do_notify_parent() from the write_lock'ed section.

Yeah, but the plan is to go successively to removing write lock from every
place it's used, except of hashing in fork() and unhashing in __unhash_process().

> (lets ignore kill_orphaned_pgrp(), we want to make will_become_orphaned_pgrp
>  lockless. Look at get_signal).
> 
> And this changes the rules: currently ->exit_state is stable under read_lock,
> except -> EXIT_DEAD transition. OK, this is probably fine, but we need to
> recheck. At least this was certainly wrong some time before iirc.
> 
> > @@ -594,7 +597,10 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
> >
> >  	write_lock_irq(&tasklist_lock);
> >  	forget_original_parent(tsk, &dead);
> > +	tsk->exit_state = EXIT_NOTIFY;
> > +	write_unlock_irq(&tasklist_lock);
> 
> And unless I missed something this EXIT_NOTIFY turns the concurrent
> do_wait() into the busy-wait loop.
> 
> Now suppose that CONFIG_SMP=n and the rt parent preempts the exiting
> child right after it drops tasklist: deadlock?

You sure, thank. We need to disable preemption there.

> > +	read_lock(&tasklist_lock);
> >  	if (group_dead)
> >  		kill_orphaned_pgrp(tsk->group_leader, NULL);
> >
> > @@ -612,13 +618,14 @@ static void exit_notify(struct task_struct *tsk, int group_dead)
> >  	}
> >
> >  	tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
> 
> This needs WRITE_ONCE(). Otherwise gcc can do, say,
> 
> 	tsk->exit_state = EXIT_ZOMBIE;
> 	if (autoreap)
> 		tsk->exit_state = EXIT_DEAD;
> 
> which will lead to kernel crash (both parent and child can release this
> task).

Ah, thanks.
> 
> 
> > -	if (tsk->exit_state == EXIT_DEAD)
> > +	smp_wmb(); /* Pairs with read_lock() in do_wait() */
> 
> Why? this barries looks unnecessary.

Sure, it's unnecessary for do_wait().

> OTOH. We need to set EXIT_XXX before __wake_up_parent(). OK, OK, we do not
> because of the busy-wait loop, but busy-wait is not an option.
> 
> > @@ -1317,6 +1324,13 @@ static int wait_consider_task(struct wait_opts *wo, int ptrace,
> >  		return 0;
> >  	}
> >
> > +	if (unlikely(exit_state == EXIT_NOTIFY)) {
> > +		if (wo->wo_flags & WNOHANG)
> > +			return 0;
> > +		read_unlock(&tasklist_lock);
> > +		return -REPEAT_DOWAIT;
> > +	}
> 
> No, no, no. If you do something like this, please (ab)use wo->notask_error.
> And wait_consider_task() should continue after that,

Kirill