Re: 2.6.16-rc1: 28ms latency when process with lots of swapped memory exits

[Ingo Molnar <mingo@elte.hu>]

* Hugh Dickins <hugh@veritas.com> wrote:

> > It seems to trigger when a process with a large amount of memory swapped
> > out exits.
> > 
> > Can this be solved with a cond_resched?
> 
> Not that easily, I think.
> 
> Are you testing with CONFIG_PREEMPT=y, as I'd expect?  I thought 
> cond_resched() adds nothing to that case (and we keep on intending but 
> forgetting to make it compile away to nothing in that case). Or am I 
> confused?

it still has an effect: if PREEMPT_BKL is disabled it drops the BKL. But 
if PREEMPT_BKL is enabled too then cond_resched() has no effect. 
(cond_resched_locked() does though)

back to the problem reported by Lee:

> >     Xorg-18254 0dn.2    3us < (2110048)
> >     Xorg-18254 0dn.2    4us : preempt_schedule (free_swap_and_cache)
> >     Xorg-18254 0dn.2    5us : free_swap_and_cache (unmap_vmas)
> >     Xorg-18254 0dn.2    6us : swap_info_get (free_swap_and_cache)
> >     Xorg-18254 0dn.3    6us : swap_entry_free (free_swap_and_cache)
> >     Xorg-18254 0dn.3    7us : find_trylock_page (free_swap_and_cache)

hm, where does the latency come from? We do have a lockbreaker in 
unmap_vmas():

                        if (need_resched() ||
                                (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
                                if (i_mmap_lock) {
                                        *tlbp = NULL;
                                        goto out;
                                }
                                cond_resched();
                        }


why doesnt this break up the 28ms latency?

	Ingo