Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1163585 > unrolled thread

Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the Xen guest code

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the  Xen guest code Linus Torvalds <torvalds@linux-foundation.org> - 2015-06-11 22:40 +0200
    Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the  Xen guest code Linus Torvalds <torvalds@linux-foundation.org> - 2015-06-11 22:50 +0200
      Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from  the Xen guest code Oleg Nesterov <oleg@redhat.com> - 2015-06-13 01:40 +0200
      Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from  the Xen guest code Ingo Molnar <mingo@kernel.org> - 2015-06-13 09:30 +0200
        Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from  the Xen guest code Oleg Nesterov <oleg@redhat.com> - 2015-06-13 20:10 +0200

#1163585 — Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the Xen guest code

On Thu, Jun 11, 2015 at 7:07 AM, Ingo Molnar <mingo@kernel.org> wrote:
>
> But we want to remove the pgd_list, so convert the code over to walk
> all tasks in the system. This is an equivalent method.

This is not at all equivalent, and it looks stupid.

Don't use "for_each_process_thread(g, p)". You only care about each
mm, and threads all share the same mm, so just do

    for_each_process(p)

instead of iterating over all threads too.

No?

               Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [next] | [standalone]

#1163606

On Thu, Jun 11, 2015 at 1:37 PM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> Don't use "for_each_process_thread(g, p)". You only care about each
> mm, and threads all share the same mm, so just do
>
>     for_each_process(p)
>
> instead of iterating over all threads too.

Hmm. I may be wrong. It strikes me that one of the group leaders might
have exited but the subthreads live on. We'd see p->mm being NULL,
even though the mm was originally in use.

Ugh. So maybe the code really does need to iterate over all threads.

                  Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [next] | [standalone]

#1164435 — Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the Xen guest code

On 06/12, Boris Ostrovsky wrote:
>
> On 06/12/2015 04:53 PM, Oleg Nesterov wrote:
>>>
>>> 	for_each_process(p) {
>>>
>>> 		for_each_thread(p, t) {
>>> 			if (t->mm) {
>>> 				do_something(t->mm);
>>> 				break;
>>> 			}
>>> 		}
>>> 	}
>>>
>>> But either way I don't understand what protects this ->mm. Perhaps this needs
>>> find_lock_task_mm().
>>
>> And, I don't understand this code, probably this doesn't matter, but.
>>
>> unpin_all() is probably fine, but xen_mm_pin_all() can race with fork()
>> and miss the new child. Is it OK?
>
>
> Currently xen_mm_pin_all() is only called in the suspend path, out of
> stop_machine(), so presumably at that time fork is not possible.

OK, thanks, this also means that this code shouldn't worry about ->mm,
it should be stable.

But for_each_process() in sync_global_pgds() should, afaics.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [next] | [standalone]

#1164582 — Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the Xen guest code

* Oleg Nesterov <oleg@redhat.com> wrote:

> > So we could add tsk->mm_leader or so,
> 
> No, no, please do not. Just do something like
> 
> 	for_each_process(p) {
> 
> 		for_each_thread(p, t) {

So far that's what the for_each_process_thread() iterations I added do, right?

> 			if (t->mm) {
> 				do_something(t->mm);
> 				break;
> 			}
> 		}
> 	}
> 
> But either way I don't understand what protects this ->mm. Perhaps this needs 
> find_lock_task_mm().

That's indeed a bug: I'll add task_lock()/unlock() before looking at ->mm.

find_lock_task_mm() is not needed IMHO: we have a stable reference to 't', as a 
task can only go away via RCU expiry, and all the iterations I added are (supposed 
to) be RCU safe.

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [next] | [standalone]

#1164667 — Re: [PATCH 07/12] x86/virt/guest/xen: Remove use of pgd_list from the Xen guest code

On 06/13, Ingo Molnar wrote:
>
> * Oleg Nesterov <oleg@redhat.com> wrote:
>
> > > So we could add tsk->mm_leader or so,
> >
> > No, no, please do not. Just do something like
> >
> > 	for_each_process(p) {
> >
> > 		for_each_thread(p, t) {
>
> So far that's what the for_each_process_thread() iterations I added do, right?

Not really,

> > 			if (t->mm) {
> > 				do_something(t->mm);
> > 				break;
                                ^^^^^

Note this "break". We stop the inner loop right after we find a thread "t"
with ->mm != NULL. In the likely case t == p (group leader) so the inner
loop stops on the 1st iteration.


> > But either way I don't understand what protects this ->mm. Perhaps this needs
> > find_lock_task_mm().
>
> That's indeed a bug: I'll add task_lock()/unlock() before looking at ->mm.

Well, in this particular case we are safe. As Boris explained this is called
from stop_machine(). But sync_global_pgds() is not safe.

> find_lock_task_mm() is not needed IMHO: we have a stable reference to 't', as
> task can only go away via RCU expiry, and all the iterations I added are (supposed
> to) be RCU safe.

Sure, you can do lock/unlock by hand. But find_lock_task_mm() can simplify
the code because it checks subthreads if group_leader->mm == NULL. You can
simply do

	rcu_read_lock();
	for_each_process(p) {
		t = find_lock_task_mm(p);
		if (!t)
			continue;

		do_something(t->mm);
		task_unlock(t);
	}
	rcu_read_unlock();

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web