Re: [RFC][RFT] memcg fix cgroup_mutex deadlock when cpusetreclaims memory

[Balbir Singh <balbir@linux.vnet.ibm.com>]

* KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2008-12-10 20:32:03]:

> Balbir Singh said:
> > * KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2008-12-10
> > 17:49:06]:
> >
> >> On Wed, 10 Dec 2008 10:49:47 +0530
> >> Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> >>
> >> > Hi,
> >> >
> >> > Here is a proposed fix for the memory controller cgroup_mutex deadlock
> >> > reported. It is lightly tested and reviewed. I need help with review
> >> > and test. Is the reported deadlock reproducible after this patch? A
> >> > careful review of the cpuset impact will also be highly appreciated.
> >> >
> >> > From: Balbir Singh <balbir@linux.vnet.ibm.com>
> >> >
> >> > cpuset_migrate_mm() holds cgroup_mutex throughout the duration of
> >> > do_migrate_pages(). The issue with that is that
> >> >
> >> > 1. It can lead to deadlock with memcg, as do_migrate_pages()
> >> >    enters reclaim
> >> > 2. It can lead to long latencies, preventing users from creating/
> >> >    destroying other cgroups anywhere else
> >> >
> >> > The patch holds callback_mutex through the duration of
> >> cpuset_migrate_mm() and
> >> > gives up cgroup_mutex while doing so.
> >> >
> >> > Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
> >> > ---
> >> >
> >> >  include/linux/cpuset.h |   13 ++++++++++++-
> >> >  kernel/cpuset.c        |   23 ++++++++++++-----------
> >> >  2 files changed, 24 insertions(+), 12 deletions(-)
> >> >
> >> > diff -puN kernel/cgroup.c~cpuset-remove-cgroup-mutex-from-update-path
> >> kernel/cgroup.c
> >> > diff -puN kernel/cpuset.c~cpuset-remove-cgroup-mutex-from-update-path
> >> kernel/cpuset.c
> >> > --- a/kernel/cpuset.c~cpuset-remove-cgroup-mutex-from-update-path
> >> > +++ a/kernel/cpuset.c
> >> > @@ -369,7 +369,7 @@ static void guarantee_online_mems(const
> >> >   * task has been modifying its cpuset.
> >> >   */
> >> >
> >> > -void cpuset_update_task_memory_state(void)
> >> > +void __cpuset_update_task_memory_state(bool held)
> >> >  {
> >> >  	int my_cpusets_mem_gen;
> >> >  	struct task_struct *tsk = current;
> >> > @@ -380,7 +380,8 @@ void cpuset_update_task_memory_state(voi
> >> >  	rcu_read_unlock();
> >> >
> >> >  	if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
> >> > -		mutex_lock(&callback_mutex);
> >> > +		if (!held)
> >> > +			mutex_lock(&callback_mutex);
> >> >  		task_lock(tsk);
> >> >  		cs = task_cs(tsk); /* Maybe changed when task not locked */
> >> >  		guarantee_online_mems(cs, &tsk->mems_allowed);
> >> > @@ -394,7 +395,8 @@ void cpuset_update_task_memory_state(voi
> >> >  		else
> >> >  			tsk->flags &= ~PF_SPREAD_SLAB;
> >> >  		task_unlock(tsk);
> >> > -		mutex_unlock(&callback_mutex);
> >> > +		if (!held)
> >> > +			mutex_unlock(&callback_mutex);
> >> >  		mpol_rebind_task(tsk, &tsk->mems_allowed);
> >> >  	}
> >> >  }
> >> > @@ -949,13 +951,15 @@ static int update_cpumask(struct cpuset
> >> >   *    so that the migration code can allocate pages on these nodes.
> >> >   *
> >> >   *    Call holding cgroup_mutex, so current's cpuset won't change
> >> > - *    during this call, as manage_mutex holds off any cpuset_attach()
> >> > + *    during this call, as callback_mutex holds off any
> >> cpuset_attach()
> >> >   *    calls.  Therefore we don't need to take task_lock around the
> >> >   *    call to guarantee_online_mems(), as we know no one is changing
> >> >   *    our task's cpuset.
> >> >   *
> >> >   *    Hold callback_mutex around the two modifications of our tasks
> >> > - *    mems_allowed to synchronize with cpuset_mems_allowed().
> >> > + *    mems_allowed to synchronize with cpuset_mems_allowed(). Give
> >> > + *    up cgroup_mutex to avoid deadlocking with other subsystems
> >> > + *    as we enter reclaim from do_migrate_pages().
> >> >   *
> >> >   *    While the mm_struct we are migrating is typically from some
> >> >   *    other task, the task_struct mems_allowed that we are hacking
> >> > @@ -976,17 +980,14 @@ static void cpuset_migrate_mm(struct mm_
> >> >  {
> >> >  	struct task_struct *tsk = current;
> >> >
> >> > -	cpuset_update_task_memory_state();
> >> > -
> >> > +	cgroup_unlock();
> >> >  	mutex_lock(&callback_mutex);
> >> > +	cpuset_update_task_memory_state_locked();
> >> >  	tsk->mems_allowed = *to;
> >> > -	mutex_unlock(&callback_mutex);
> >> > -
> >> >  	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
> >> > -
> >> > -	mutex_lock(&callback_mutex);
> >> >  	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
> >> >  	mutex_unlock(&callback_mutex);
> >> > +	cgroup_lock();
> >> >  }
> >> >
> >>
> >> Hmm...can't this happen ?
> >>
> >> Assume there is a task X and cgroup Z1 and Z2. Z1 and Z2 doesn't need to
> >> be in
> >> the same hierarchy.
> >> ==
> >> 	CPU A attach task X to cgroup Z1
> >> 		cgroup_lock()
> >> 			for_each_subsys_state()
> >
> > You mean for_each_subsys() right?
> >
> >> 				=> attach(X,Z)
> >> 					=> migrate_mm()
> >> 						=> cgroup_unlock()
> >> 							migration
> >>
> >> 	CPU B attach task X to cgroup Z2 at the same time
> >> 		cgroup_lock()
> >> 			replace css_set.
> >> ==
> >>
> >> Works on CPU B can't break for_each_subsys_state() in CPU A ?
> >>
> >
> > for_each_subsys is hierarchy aware, so if we try to add the same task
> > to different hierachies, it should not be a problem right?
> >
> Ah, maybe. But what happens when Z1 and Z2 is the same hierarchy ?
> Are there some locks ?
>

If they are in the same hierarchy, tsk->cgroups and tsk->cg_list is
updated atomically and for_each_subsys should not be affected.

Needs more thought and coffee though


-- 
	Balbir

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>