Re: [PATCH] release_task: kill the no longer needed get/put_pid(thread_pid)

[Mateusz Guzik <mjguzik@gmail.com>]

On Tue, Apr 15, 2025 at 9:35 AM Christian Brauner <brauner@kernel.org> wrote:
>
> On Mon, Apr 14, 2025 at 11:26:35PM +0200, Mateusz Guzik wrote:
> > On Mon, Apr 14, 2025 at 10:46 PM Oleg Nesterov <oleg@redhat.com> wrote:
> > > I'd prefer to move add_device_randomness() after release_thread(),
> > > but perhaps this needs another minor cleanup?
> > >
> >
> > I have no opinion on that front. As far as cosmetics go I would not
> > touch it past the nit I mentioned in my previous e-mail, but I'm not
> > going to protest any other changes.
> >
> > imo the real thing to do concerning the routine is to figure out if
> > the call is even of any benefit -- it very well may be this is call is
> > only there because nobody with random-fu bothered to remove it.
> > Personally I have epsilon knowledge of that area, so I'm not even
> > going to try to evaluate it. But it would be nice if someone(tm)
> > reached out to random folk concerning it. Worst case, even if it still
> > has to be there, maybe the sucker can trylock and bail. It does
> > contribute to contention, likely for no good reason.
>
> Afaict, add_device_randomness() cannot block. So why can't we just move
> this into put_task_struct_rcu_user() (or some variant that's local to
> kernel/exit.c)? I see no reason why this would need to be done
> synchronously?

It can move anywhere as long as ->se.sum_exec_runtime is not scrambled.

I'm not going to argue where it should land. I do note chances are the
call can be straight up removed and if I was looking to shuffle it
around some more, I would first try to find out if it can in fact just
go. I don't believe it warrants any effort at this stage though.

That said, apart from this remark I'm not going to argue about any placement.
-- 
Mateusz Guzik <mjguzik gmail.com>