Messages in this thread | | | Date | Fri, 20 Apr 2012 22:41:29 +0200 | From | Oleg Nesterov <> | Subject | Re: [PATCH 1/2] mm: set task exit code before complete_vfork_done() |
| |
On 04/20, Konstantin Khlebnikov wrote: > > Oleg Nesterov wrote: >> >> /* sync mm's RSS info before statistics gathering */ >> if (tsk->mm) >> sync_mm_rss(tsk->mm); >> >> Which "statistics gathering" ? Probably I missed something, but >> after the quick grep it seems to me that this is only needed for >> taskstats_exit()->xacct_add_tsk(). >> >> So why we can't simply add sync_mm_rss() into xacct_add_tsk() ? > >> Yes, this way we do not "account" put_user(clear_child_tid) but >> I think we do not care. > > Why we don't care? Each thread can corrupt these counters by one. > I do not think that we are satisfied with nearly accurate rss accounting. > +/- one page for each clone()-exit().
Not actually "for each" in practice. Each exit does sync_ (with this patch from xacct_add_tsk), the net effect should be small.
And. This is what we do now, nobody ever complained.
>> IOW, what do you think about the trivial patch below? Uncompiled, >> untested, probably incomplete. acct_update_integrals() looks >> suspicious too. > > what a mess! =)
Thanks ;)
But it is much, much simpler than your patches, don't you agree?
Oleg.
| |