Re: kmemcheck caught read from freed memory (cfq_free_io_context)

["Vegard Nossum" <vegard.nossum@gmail.com>]

On Wed, Apr 2, 2008 at 6:59 PM, Paul E. McKenney
<paulmck@linux.vnet.ibm.com> wrote:
> On Wed, Apr 02, 2008 at 06:15:52PM +0200, Vegard Nossum wrote:
>  > On Wed, Apr 2, 2008 at 6:08 PM, Paul E. McKenney
>  > <paulmck@linux.vnet.ibm.com> wrote:
>  > > On Wed, Apr 02, 2008 at 02:01:13PM +0300, Pekka Enberg wrote:
>  > >  > No, kmemcheck is work in progress and does not know about
>  > >  > SLAB_DESTROY_BY_RCU yet. The reason I asked Vegard to post the warning
>  > >  > was because Peter, Vegard, and myself identified this particular
>  > >  > warning as a real problem. But yeah, kmemcheck can cause false
>  > >  > positives for RCU for now.
>  > >
>  > >  Would the following be an appropriate fix?  It seems to me to be in
>  > >  the same spirit as the existing check for s->ctor.
>  >
>  > In my opinion, no.
>  >
>  > It would fix the false positives, but would in fact also hide cases
>  > such as this one with cfq, e.g. the real cases of mis-use.
>
>  Though this case apparently does not qualify as misuse until such
>  time as CLONE_IO is implemented.
>
>  And doesn't the current check for ->ctor also potentially hide misuse?

Hm, no. Objects with a ctor should retain its "initializedness"
between allocations of the same object. This is one of the features of
slab allocation.

>  > Peter Zijlstra suggested this:
>  > > It would have to register an call_rcu callback itself in order to mark
>  > > it freed - and handle the race with the object being handed out again.
>
>  Glad you liked them!
>
>  And Peter's suggested approach would indeed be more accurate.  But I
>  will still put my patch forward as a stopgap.  ;-)

Yes, I realize that it will be needed as a temporary fix. It _is_
better to hide some real warnings in favour of the showing false ones.
(But a TODO comment is in order, I believe.)

Thanks.


Vegard