Virtualization - May 2023 - [PATCH 3/3] fork, vhost: Use CLONE_THREAD to fix freezer/ps regression

On Wed, May 31, 2023 at 3:25?PM Oleg Nesterov <oleg at redhat.com>
wrote:
>
> On 05/31, Jason Wang wrote:
> >
> > ? 2023/5/23 20:15, Oleg Nesterov ??:
> > >
> > >             /* make sure flag is seen after deletion */
> > >             smp_wmb();
> > >             llist_for_each_entry_safe(work, work_next, node,
node) {
> > >                     clear_bit(VHOST_WORK_QUEUED,
&work->flags);
> > >
> > >I am not sure about smp_wmb + clear_bit. Once we clear
VHOST_WORK_QUEUED,
> > >vhost_work_queue() can add this work again and change
work->node->next.
> > >
> > >That is why we use _safe, but we need to ensure that
llist_for_each_safe()
> > >completes LOAD(work->node->next) before VHOST_WORK_QUEUED is
cleared.
> >
> > This should be fine since store is not speculated, so
work->node->next needs
> > to be loaded before VHOST_WORK_QUEUED is cleared to meet the loop
condition.
>
> I don't understand you. OK, to simplify, suppose we have 2 global vars
>
>         void *PTR = something_non_null;
>         unsigned long FLAGS = -1ul;
>
> Now I think this code
>
>         CPU_0                           CPU_1
>
>         void *ptr = PTR;                if (!test_and_set_bit(0, FLAGS))
>         clear_bit(0, FLAGS);                    PTR = NULL;
>         BUG_ON(!ptr);
>
> is racy and can hit the BUG_ON(!ptr).
This seems different to the above case? And you can hit BUG_ON with
the following execution sequence:

[cpu 0] clear_bit(0, FLAGS);
[cpu 1] if (!test_and_set_bit(0, FLAGS))
[cpu 1] PTR = NULL;
[cpu 0] BUG_ON(!ptr)

In vhost code, there's a condition before the clear_bit() which sits
inside llist_for_each_entry_safe():

#define llist_for_each_entry_safe(pos, n, node, member)                        \
        for (pos = llist_entry((node), typeof(*pos), member);                  \
             member_address_is_nonnull(pos, member) &&                  
\
                (n = llist_entry(pos->member.next, typeof(*n), member),
true); \
             pos = n)

The clear_bit() is a store which is not speculated, so there's a
control dependency, the store can't be executed until the condition
expression is evaluated which requires pos->member.next
(work->node.next) to be loaded.
>
> I guess it is fine on x86, but in general you need smp_mb__before_atomic()
> before clear_bit(), or clear_bit_unlock().
>
> > >                     __set_current_state(TASK_RUNNING);
> > >
> > >Why do we set TASK_RUNNING inside the loop? Does this mean that
work->fn()
> > >can return with current->state != RUNNING ?
> >
> > It is because the state were set to TASK_INTERRUPTIBLE in the
beginning of
> > the loop otherwise it might be side effect while executing
work->fn().
>
> Again, I don't understand you. So let me repeat: can work->fn()
return with
> current->_state != TASK_RUNNING ? If not (and I'd say it should
not), you can
> do __set_current_state(TASK_RUNNING) once, before
llist_for_each_entry_safe().
>
Ok, that should be fine.

Thanks

> > >Now the main question. Whatever we do, SIGKILL/SIGSTOP/etc can
come right
> > >before we call work->fn(). Is it "safe" to run this
callback with
> > >signal_pending() or fatal_signal_pending() ?
> >
> > It looks safe since:
> >
> > 1) vhost hold refcnt of the mm
> > 2) release will sync with the worker
>
> Well, that's not what I asked... nevermind, please forget.
>
> Thanks.
>
> Oleg.
>

On 05/31, Jason Wang wrote:
>
> On Wed, May 31, 2023 at 3:25?PM Oleg Nesterov <oleg at redhat.com>
wrote:
> >
> > On 05/31, Jason Wang wrote:
> > >
> > > ? 2023/5/23 20:15, Oleg Nesterov ??:
> > > >
> > > >             /* make sure flag is seen after deletion */
> > > >             smp_wmb();
> > > >             llist_for_each_entry_safe(work, work_next, node,
node) {
> > > >                     clear_bit(VHOST_WORK_QUEUED,
&work->flags);
> > > >
> > > >I am not sure about smp_wmb + clear_bit. Once we clear
VHOST_WORK_QUEUED,
> > > >vhost_work_queue() can add this work again and change
work->node->next.
> > > >
> > > >That is why we use _safe, but we need to ensure that
llist_for_each_safe()
> > > >completes LOAD(work->node->next) before
VHOST_WORK_QUEUED is cleared.
> > >
> > > This should be fine since store is not speculated, so
work->node->next needs
> > > to be loaded before VHOST_WORK_QUEUED is cleared to meet the loop
condition.
> >
> > I don't understand you. OK, to simplify, suppose we have 2 global
vars
> >
> >         void *PTR = something_non_null;
> >         unsigned long FLAGS = -1ul;
> >
> > Now I think this code
> >
> >         CPU_0                           CPU_1
> >
> >         void *ptr = PTR;                if (!test_and_set_bit(0,
FLAGS))
> >         clear_bit(0, FLAGS);                    PTR = NULL;
> >         BUG_ON(!ptr);
> >
> > is racy and can hit the BUG_ON(!ptr).
>
> This seems different to the above case?
not sure,
> And you can hit BUG_ON with
> the following execution sequence:
>
> [cpu 0] clear_bit(0, FLAGS);
> [cpu 1] if (!test_and_set_bit(0, FLAGS))
> [cpu 1] PTR = NULL;
> [cpu 0] BUG_ON(!ptr)
I don't understand this part... yes, we can hit this BUG_ON() without mb in
between, this is what I tried to say.
> In vhost code, there's a condition before the clear_bit() which sits
> inside llist_for_each_entry_safe():
>
> #define llist_for_each_entry_safe(pos, n, node, member)                    
\
>         for (pos = llist_entry((node), typeof(*pos), member);              
\
>              member_address_is_nonnull(pos, member) &&             
\
>                 (n = llist_entry(pos->member.next, typeof(*n), member),
true); \
>              pos = n)
>
> The clear_bit() is a store which is not speculated, so there's a
> control dependency, the store can't be executed until the condition
> expression is evaluated which requires pos->member.next
> (work->node.next) to be loaded.
But llist_for_each_entry_safe() doesn't check "n", I mean, it is
not that we have
something like

	n = llist_entry(...);
	if (n)
		clear_bit(...);

so I do not see how can we rely on the load-store control dependency.

Oleg.