Messages in this thread | | | From | Andrea Righi <> | Subject | Re: [RFC PATCH 5/5] refresh VM committed space after a task migration | Date | Wed, 11 Jun 2008 12:37:57 +0200 (MEST) |
| |
Dave Hansen wrote: > On Tue, 2008-06-10 at 01:33 +0200, Andrea Righi wrote: >> + preempt_disable(); >> + committed = atomic_long_read(&p->vm_committed_space); >> + atomic_long_sub(committed, &old_mem->vmacct.vm_committed_space); >> + atomic_long_add(committed, &mem->vmacct.vm_committed_space); >> + preempt_enable(); >> out: >> mmput(mm); >> } > > Why bother with the preempt stuff here? What does the actually protect > against? I assume that you're trying to keep other tasks that might run > on this CPU from seeing weird, inconsistent numbers in here. Is there > some other looks that keeps *other* cpus from seeing this? > > In any case, I think it needs a big, fat comment.
Yes, true, mem_cgroup_move_task() is called after the task->cgroups pointer has been changed. So, even if task changes its committed space between the atomic_long_sub() and atomic_long_add() it will be correctly accounted in the new cgroup.
-Andrea
| |