Re: [PATCH 0/4] exit: Make unlikely case in mm_update_next_owner() more scalable

[Eric W. Biederman]

Michal Hocko <mhocko@xxxxxxxxxx> writes:

> On Fri 01-06-18 10:25:59, Eric W. Biederman wrote:
>> Michal Hocko <mhocko@xxxxxxxxxx> writes:
>> 
>> > On Fri 01-06-18 09:32:42, Eric W. Biederman wrote:
>> >> Michal Hocko <mhocko@xxxxxxxxxx> writes:
>> > [...]
>> >> > Group leader exiting early without tearing down the whole thread
>> >> > group should be quite rare as well. No question that somebody might do
>> >> > that on purpose though...
>> >> 
>> >> The group leader exiting early is a completely legitimate and reasonable
>> >> thing to do, even if it is rare.
>> >
>> > I am not saying it isn't legitimate. But the most common case is the
>> > main thread waiting for its threads or calling exit which would tear the
>> > whole group down. Is there any easy way to achieve this other than tkill
>> > to group leader? Calling exit(3) from the leader performs group exit
>> > IIRC.
>> 
>> pthread_exit from the group leader.
>
> Right, forgot to mention this one but this would be quite exotic,
> right?

Not exotic.  It is easy to do and well defined.

It would be easy to do if you are running a thread pool and closing
unnecessary threads.  Or frankly anything where the application is not
assigning a special role to the initial thread.

It does seem rare enough that no one has noticed how attrocious
mm_update_next_owner is until now.

My key point is that it is easy to trigger which makes the current
mm_update_next_owner a fundamentally flawed design, and something that
needs to be fixed.

Eric