Re: [PATCH 0/3] static keys: fix test/set races

[Jason Baron]

On 06/29/2013 03:20 AM, Ingo Molnar wrote:
> * jbaron@xxxxxxxxxx <jbaron@xxxxxxxxxx> wrote:
>
> > Hi,
> >
> > As pointed out by Andi Kleen, some static key users can be racy because they
> > check the value of the key->enabled, and then subsequently update the branch
> > direction. A number of call sites have 'higher' level locking that avoids this
> > race, but the usage in the scheduler features does not. See:
> > http://lkml.indiana.edu/hypermail/linux/kernel/1304.2/01655.html
> But that's not an issue at all - switching the scheduler features is for
> development and debugging only, and in some cases higher level locking
> would be needed to solve it 'properly', beyond what the keys API could
> give ...
>
> So this is pretty pointless, sorry, please don't complicate this facility.
>
> Thanks,
>
> 	Ingo

Hi Ingo,

Yes, I agree that 'higher' level locking may be required for some 
callers of the newly proposed interface. However, I do think that the 
static_key_slow_set_true()/false() provides a nice abstraction for some 
callers, while addressing test/set() races, by making that sequence atomic.

I view the proposed inteface of set_true()/set_false() as somewhat 
analogous to an atomic_set() call. In the same way, the current 
static_key_slow_inc()/dec() are analogous to atomic_inc()/dec().

It arguably makes the code code a bit more readable, transforming 
sequences such as:

if (!static_key_enabled(&control_var))
        static_key_slow_inc(&control_var);

into:

        static_key_slow_set_true(&control_var);


I see at least 3 users of static_keys in the tree which I think would 
benefit  from this transformation. The 2 attached with this series, and 
the usage in kernel/tracepoint.c.

Thanks,

-Jason

--
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/