Re: [-mm] Add an owner to the mm_struct (v2)

[Paul Menage]

On Fri, Mar 28, 2008 at 7:52 AM, Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote:
>
>  mm->owner_lock is there to protect mm->owner field from changing simultaneously
>  as tasks fork/exit.
>

But the *hardware* already does that for you - individual writes to
pointers are already atomic operations and so will be serialized.
Using a lock to guard something only does anything useful if at least
one of the critical regions that takes the lock consists of more than
a single atomic operation, or if you have a mixture of read sections
and write sections. Now it's true that your critical region in
mm_fork_init_owner() is more than a single atomic op, but I'm arguing
below that it's a no-op. So that just leaves the single region

spin_lock(&mm->owner_lock);
mm->owner = new_owner;
spin_unlock(&mm->owner_lock);

which isn't observably different if you remove the spinlock.

>
>  Oh! yes.. my bad again. The check should have been p == p->thread_group, but
>  that is not required either. The check should now ideally be
>
>  if (!(clone_flags & CLONE_VM))
>

OK, so if the new thread has its own mm (and hence will already have
mm->owner set up to point to p in mm_init()) then we do:

>  +       if (mm->owner != p)
>  +               rcu_assign_pointer(mm->owner, p->group_leader);

which is a no-op since we know mm->owner == p.

>
>  Yes.. I think we need to call it earlier.
>

No, I think we need to call it later - after we've cleared current->mm
(from within task_lock(current)) - so we can't rely on p->mm in this
function, we have to pass it in. If we call it before while
current->mm == mm, then we risk a race where the (new or existing)
owner exits and passes it back to us *after* we've done a check to see
if we need to find a new owner. If we ensure that current->mm != mm
before we call mm_update_next_owner(), then we know we're not a
candidate for receiving the ownership if we don't have it already.

>
>  But there is no way to guarantee that, what is the new_owner exec's after we've
>  done the check and assigned. Won't we end up breaking the invariant? How about
>  we have mm_update_new_owner() call in exec_mmap() as well? That way, we can
>  still use owner_lock and keep the invariant.
>

Oops, I thought that exit_mm() already got called in the execve()
path, but you're right, it doesn't.

Yes, exit_mmap() should call mm_update_next_owner() after the call to
task_unlock(), i.e. after it's set its new mm.

So I need to express the invariant more carefully.

What we need to preserve is that, for every mm at all times, mm->owner
points to a valid task. So either:

1) mm->owner->mm == mm AND mm->owner will check to see whether it
needs to pass ownership before it exits or execs.

OR

2) mm->owner is the last user of mm and is about to free mm.

OR

3) mm->owner is currently searching for another user of mm to pass the
ownership to.

In order to get from state 3 to state 1 safely we have to hold
task_lock(new_owner). Otherwise we can race with an exit or exec in
new_owner, resulting in a process that has already passed the point of
checking current->mm->owner.

I don't see why we need mm->owner_lock to maintain this invariant.
(But am quite prepared to be proven wrong).

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