On 2014/2/21 23:40, Tejun Heo wrote:
> Hello,
>
> On Fri, Feb 21, 2014 at 10:16:22AM +0800, Fengguang Wu wrote:
>> commit fb47fea7a59cf3d6387c566084a6684b5005af83
>> Author: Tejun Heo <tj@xxxxxxxxxx>
>> AuthorDate: Thu Feb 13 15:16:35 2014 -0500
>> Commit: Tejun Heo <tj@xxxxxxxxxx>
>> CommitDate: Thu Feb 13 15:16:35 2014 -0500
>>
>> cgroup: drop task_lock() protection around task->cgroups
>>
>> For optimization, task_lock() is additionally used to protect
>> task->cgroups. The optimization is pretty dubious as either
>> css_set_rwsem is grabbed anyway or PF_EXITING already protects
>> task->cgroups. It adds only overhead and confusion at this point.
>> Let's drop task_[un]lock() and update comments accordingly.
>>
>> Signed-off-by: Tejun Heo <tj@xxxxxxxxxx>
>>
>> [main] Setsockopt(1 2b 80d1000 4) on fd 223 [17:2:768]
>> [ 27.030764]
>> [ 27.031119] ===============================
>> [ 27.031833] [ INFO: suspicious RCU usage. ]
>> [ 27.032536] 3.14.0-rc3-02458-g837caba #2 Not tainted
>> [ 27.033378] -------------------------------
>> [ 27.044237] include/linux/cgroup.h:697 suspicious rcu_dereference_check() usage!
>> [ 27.045795]
>> [ 27.045795] other info that might help us debug this:
>> [ 27.045795]
>> [ 27.047114]
>> [ 27.047114] rcu_scheduler_active = 1, debug_locks = 0
>> [main] Setsockopt(1 c 80d1000 4) on fd 225 [39:5:0]
>> [ 27.048751] 2 locks held by trinity-c0/4479:
>> [ 27.049478] #0: (callback_mutex){+.+...}, at: [<81118395>] cpuset_cpus_allowed+0x1e/0x123
>> [ 27.051132] #1: (&(&p->alloc_lock)->rlock){+.+...}, at: [<8111839c>] cpuset_cpus_allowed+0x25/0x123
>> [ 27.052788]
>> [ 27.052788] stack backtrace:
>> [ 27.053528] CPU: 0 PID: 4479 Comm: trinity-c0 Not tainted 3.14.0-rc3-02458-g837caba #2
>> [ 27.064971] 00000000 00000000
>> 919eff28 81877cc3[main] Setsockopt(1 7 80d1000 4) on fd 226 [1:5:1]
>
> So, this is from removing task_lock from task_css_set_check() and
> adding rcu_read_lock() in cpuset_cpus_allowed() should fix it.
Yeah, rcu_read_lock() should be sufficient.
> I'm
> not sure how much of task_lock() locking we currently have in cpuset
> is actually necessary tho. Shouldn't we be able to do most with just
> callback_mutex, if not cpuset_mutex? Li, any ideas?
>
task_lock() is also used to protect task->mems_allowed. I'll see if we
can get rid of most (if not all) task_lock() lockings in cpuset.
--
To unsubscribe from this list: send the line "unsubscribe cgroups" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
