Re: [PATCH] cgroup: replace global percpu_rwsem with signal_struct->group_rwsem when writing cgroup.procs/threads

[escape <escape@linux.alibaba.com>]

在 2025/9/4 09:40, Chen Ridong 写道:
>
> On 2025/9/4 4:45, Tejun Heo wrote:
>> Hello, Michal.
>>
>> On Wed, Sep 03, 2025 at 10:03:39PM +0200, Michal Koutný wrote:
>>> On Wed, Sep 03, 2025 at 06:53:36AM -1000, Tejun Heo <tj@kernel.org> wrote:
>>>> If you use CLONE_INTO_CGROUP, cgroup migration doesn't just become cold. It
>>>> disappears completely and CLONE_INTO_CGROUP doesn't need any global locks
>>>> from cgroup side.
>>> CLONE_INTO_CGROUP uses cgroup_mutex and threadgroup rwsem like regular
>>> migration, no? Its effect is atomicity wrt clone.
>>> Or, Tejum, what do you mean that it disappears? (I think we cannot give
>>> up cgroup_mutex as it ensures synchronization of possible parent's
>>> migration.)
>> Sorry, I was confused. We no longer need to write lock threadgroup rwsem
>> when CLONE_INTO_CGROUP'ing into an empty cgroup. We do still need
>> cgroup_mutex.
>>
>>    671c11f0619e ("cgroup: Elide write-locking threadgroup_rwsem when updating csses on an empty subtree")
>>
>> Thanks.
>>
> I'm still a bit confused. Commit 671c11f0619e ("cgroup: Elide write-locking threadgroup_rwsem when
> updating csses on an empty subtree") only applies to CSS updates. However, cloning with
> CLONE_INTO_CGROUP still requires acquiring the threadgroup_rwsem.
>
> cgroup_can_fork
>    cgroup_css_set_fork
>      	if (kargs->flags & CLONE_INTO_CGROUP)
> 		cgroup_lock();
> 	cgroup_threadgroup_change_begin(current);
>
When using CLONE_INTO_CGROUP, there is no need to write cgroup.procs. So 
although
a read lock is required here, there is no contention with the write lock.

Thanks.