lkml.org 
[lkml]   [2010]   [Dec]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/2] ksm: annotate ksm_thread_mutex is no deadlock source
On Wed, 1 Dec 2010, KOSAKI Motohiro wrote:
>
> commit 62b61f611e(ksm: memory hotremove migration only) made following
> new lockdep warning.
>
> =======================================================
> [ INFO: possible circular locking dependency detected ]
> -------------------------------------------------------
> bash/1621 is trying to acquire lock:
> ((memory_chain).rwsem){.+.+.+}, at: [<ffffffff81079339>]
> __blocking_notifier_call_chain+0x69/0xc0
>
> but task is already holding lock:
> (ksm_thread_mutex){+.+.+.}, at: [<ffffffff8113a3aa>]
> ksm_memory_callback+0x3a/0xc0
>
> which lock already depends on the new lock.
>
> the existing dependency chain (in reverse order) is:
>
> -> #1 (ksm_thread_mutex){+.+.+.}:
> [<ffffffff8108b70a>] lock_acquire+0xaa/0x140
> [<ffffffff81505d74>] __mutex_lock_common+0x44/0x3f0
> [<ffffffff81506228>] mutex_lock_nested+0x48/0x60
> [<ffffffff8113a3aa>] ksm_memory_callback+0x3a/0xc0
> [<ffffffff8150c21c>] notifier_call_chain+0x8c/0xe0
> [<ffffffff8107934e>] __blocking_notifier_call_chain+0x7e/0xc0
> [<ffffffff810793a6>] blocking_notifier_call_chain+0x16/0x20
> [<ffffffff813afbfb>] memory_notify+0x1b/0x20
> [<ffffffff81141b7c>] remove_memory+0x1cc/0x5f0
> [<ffffffff813af53d>] memory_block_change_state+0xfd/0x1a0
> [<ffffffff813afd62>] store_mem_state+0xe2/0xf0
> [<ffffffff813a0bb0>] sysdev_store+0x20/0x30
> [<ffffffff811bc116>] sysfs_write_file+0xe6/0x170
> [<ffffffff8114f398>] vfs_write+0xc8/0x190
> [<ffffffff8114fc14>] sys_write+0x54/0x90
> [<ffffffff810028b2>] system_call_fastpath+0x16/0x1b
>
> -> #0 ((memory_chain).rwsem){.+.+.+}:
> [<ffffffff8108b5ba>] __lock_acquire+0x155a/0x1600
> [<ffffffff8108b70a>] lock_acquire+0xaa/0x140
> [<ffffffff81506601>] down_read+0x51/0xa0
> [<ffffffff81079339>] __blocking_notifier_call_chain+0x69/0xc0
> [<ffffffff810793a6>] blocking_notifier_call_chain+0x16/0x20
> [<ffffffff813afbfb>] memory_notify+0x1b/0x20
> [<ffffffff81141f1e>] remove_memory+0x56e/0x5f0
> [<ffffffff813af53d>] memory_block_change_state+0xfd/0x1a0
> [<ffffffff813afd62>] store_mem_state+0xe2/0xf0
> [<ffffffff813a0bb0>] sysdev_store+0x20/0x30
> [<ffffffff811bc116>] sysfs_write_file+0xe6/0x170
> [<ffffffff8114f398>] vfs_write+0xc8/0x190
> [<ffffffff8114fc14>] sys_write+0x54/0x90
> [<ffffffff810028b2>] system_call_fastpath+0x16/0x1b
>
> But it's false positive. Both memory_chain.rwsem and ksm_thread_mutex
> have outer lock (mem_hotplug_mutex). then, they can't make deadlock.
>
> Thus, This patch annotate ksm_thread_mutex is not deadlock source.
>
> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>

Thank you (I assume it does the job, I've not yet checked). My only
issue with this is that the comment you added below tells a different
story from the fuller comment you give above. Maybe change it to:

* mutex_lock_nested() is necessary because lockdep was alarmed that
* here we take ksm_thread_mutex inside notifier chain mutex, and
* later take notifier chain mutex inside ksm_thread_mutex to unlock
* it: but that's safe because both are inside mem_hotplug_mutex.

Acked-by: Hugh Dickins <hughd@google.com>

> ---
> mm/ksm.c | 4 +++-
> 1 files changed, 3 insertions(+), 1 deletions(-)
>
> diff --git a/mm/ksm.c b/mm/ksm.c
> index 65ab5c7..5aa4900 100644
> --- a/mm/ksm.c
> +++ b/mm/ksm.c
> @@ -1724,8 +1724,10 @@ static int ksm_memory_callback(struct notifier_block *self,
> /*
> * Keep it very simple for now: just lock out ksmd and
> * MADV_UNMERGEABLE while any memory is going offline.
> + * Mutex_lock_nested() is necessary to tell that
> + * ksm_thread_mutex is not unlocked here intentionally.
> */
> - mutex_lock(&ksm_thread_mutex);
> + mutex_lock_nested(&ksm_thread_mutex, SINGLE_DEPTH_NESTING);
> break;
>
> case MEM_OFFLINE:
> --
> 1.6.5.2


\
 
 \ /
  Last update: 2010-12-01 21:15    [W:0.065 / U:13.616 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site