lkml.org 
[lkml]   [2011]   [Jun]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] [BUGFIX] update mm->owner even if no next owner.
On Fri, 10 Jun 2011, KAMEZAWA Hiroyuki wrote:
>
> I think this can be a fix.

Sorry, I think not: I've not digested your rationale,
but three things stand out:

1. Why has this only just started happening? I may not have run that
test on 3.0-rc1, but surely I ran it for hours with 2.6.39;
maybe not with khugepaged, but certainly with ksmd.

2. Your hunk below:
> - if (!mm_need_new_owner(mm, p))
> + if (!mm_need_new_owner(mm, p)) {
> + rcu_assign_pointer(mm->owner, NULL);
is now setting mm->owner to NULL at times when we were sure it did not
need updating before (task is not the owner): you're damaging mm->owner.

3. There's a patch from Andrea in 3.0-rc1 which looks very likely to be
relevant, 692e0b35427a "mm: thp: optimize memcg charge in khugepaged".
I'll try reproducing without that tonight (I crashed in 20 minutes
this morning, so it's not too hard).

Hugh

> maybe good to CC Oleg.
> ==
> From dff52fb35af0cf36486965d19ee79e04b59f1dc4 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Fri, 10 Jun 2011 13:15:14 +0900
> Subject: [PATCH] [BUGFIX] update mm->owner even if no next owner.
>
> A panic is reported.
>
> > Call Trace:
> >  [<ffffffff81139792>] mem_cgroup_from_task+0x15/0x17
> >  [<ffffffff8113a75a>] __mem_cgroup_try_charge+0x148/0x4b4
> >  [<ffffffff810493f3>] ? need_resched+0x23/0x2d
> >  [<ffffffff814cbf43>] ? preempt_schedule+0x46/0x4f
> >  [<ffffffff8113afe8>] mem_cgroup_charge_common+0x9a/0xce
> >  [<ffffffff8113b6d1>] mem_cgroup_newpage_charge+0x5d/0x5f
> >  [<ffffffff81134024>] khugepaged+0x5da/0xfaf
> >  [<ffffffff81078ea0>] ? __init_waitqueue_head+0x4b/0x4b
> >  [<ffffffff81133a4a>] ? add_mm_counter.constprop.5+0x13/0x13
> >  [<ffffffff81078625>] kthread+0xa8/0xb0
> >  [<ffffffff814d13e8>] ? sub_preempt_count+0xa1/0xb4
> >  [<ffffffff814d5664>] kernel_thread_helper+0x4/0x10
> >  [<ffffffff814ce858>] ? retint_restore_args+0x13/0x13
> >  [<ffffffff8107857d>] ? __init_kthread_worker+0x5a/0x5a
>
> The code is.
> > return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
> > struct mem_cgroup, css);
>
>
> What happens here is accssing a freed task struct "p" from mm->owner.
> So, it's doubtful that mm->owner points to freed task struct.
>
> At thread exit, we need to handle mm->owner. If exitting-thread == mm->owner,
> we modify mm->owner to points to other exisiting task. But, we do not update
> mm->owner when there are no more threads. But if a kernel thread, like khugepaged,
> picks up a mm_struct without updating mm->users, there is a trouble.
>
> When mm_users shows that the task is the last task belongs to mm.
> mm->owner is not updated and remained to point to the task. So, in this case,
> mm->owner points to a not exisiting task. This was good because if there
> are no thread, no charge happens in old days. But now, we have ksm and
> khugepaged.
>
> rcu_read_lock() used in memcg is of no use because mm->owner can be
> freed before we take rcu_read_lock.
> Then, mm->owner should be cleared if there are no next owner.
>
> Reported-by: Hugh Dickins <hughd@google.com>
> Reported-by: Dave Jones <davej@redhat.com>
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> ---
> kernel/exit.c | 6 ++++--
> 1 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/exit.c b/kernel/exit.c
> index 20a4064..dbc3736 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -582,8 +582,10 @@ void mm_update_next_owner(struct mm_struct *mm)
> struct task_struct *c, *g, *p = current;
>
> retry:
> - if (!mm_need_new_owner(mm, p))
> + if (!mm_need_new_owner(mm, p)) {
> + rcu_assign_pointer(mm->owner, NULL);
> return;
> + }
>
> read_lock(&tasklist_lock);
> /*
> @@ -617,7 +619,7 @@ retry:
> * most likely racing with swapoff (try_to_unuse()) or /proc or
> * ptrace or page migration (get_task_mm()). Mark owner as NULL.
> */
> - mm->owner = NULL;
> + rcu_assign_pointer(mm->owner, NULL);
> return;
>
> assign_new_owner:
> --
> 1.7.4.1
\
 
 \ /
  Last update: 2011-06-10 23:51    [W:0.105 / U:0.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site