Re: Do we really need curr_target in signal_struct ?

[Oleg Nesterov <oleg@redhat.com>]

On 01/29, Rakib Mullick wrote:
>
> On Wed, Jan 29, 2014 at 8:55 PM, Oleg Nesterov <oleg@redhat.com> wrote:
> > On 01/29, Rakib Mullick wrote:
> >>
> >> On Tue, Jan 28, 2014 at 10:43 PM, Oleg Nesterov <oleg@redhat.com> wrote:
> >>
> >
> >> AFAIU, ->current_target is only a loop breaker to avoid infinite loop,
> >
> > No. It caches the last result of "find a thread which can handle this
> > group-wide signal".
> >
> The reason behind of my understanding is the following comments:
>
>                                 /*
>                                  * No thread needs to be woken.
>                                  * Any eligible threads will see
>                                  * the signal in the queue soon.
>                                  */
>
> What if, there's no thread in a group wants_signal()?

then complete_signal() returns without signal_wake_up().

> Or it can't
> practically happen?

It can. Say, all threads has blocked this signal. And other reasons.

> >> but - by using while_each_thread() we can remove it completely, thus
> >> helps to get rid from maintaining it too.
> >
> > ... and remove the optimization above.
> >
> >> I'll prepare a proper patch with you suggestions for reviewing.
> >
> > I am not sure we want this patch. Once again, I do not know how much
> > ->curr_target helps, and certainaly it can't help always. But you
> > should not blindly remove it just because yes, sure, it is not strictly
> > needed to find a wants_signal() thread.
> >
> Are you thinking that , since things are not broken, then we shouldn't
> try to do anything?

Hmm. No.

I am thinking that, since you misunderstood the purpose of ->curr_target,
I should probably try to argue with your patch which blindly removes this
optimization ?

I also think that this logic doesn't look perfect, but this is another
story.

Oleg.