Re: [PATCH rcu 5/8] slab: Explain why SLAB_DESTROY_BY_RCU reference before locking

[Paul E. McKenney]

On Thu, Oct 20, 2022 at 09:10:49AM +0200, Vlastimil Babka wrote:
> On 10/20/22 00:46, Paul E. McKenney wrote:
> > It is not obvious to the casual user why it is absolutely necessary to
> > acquire a reference to a SLAB_DESTROY_BY_RCU structure before acquiring
> > a lock in that structure.  Therefore, add a comment explaining this point.
> 
> s/SLAB_DESTROY_BY_RCU/SLAB_TYPESAFE_BY_RCU/ in subject, commit log and the
> added comment? :)

Boy, I was certainly living in the past when I did this patch, wasn't I?

Thank you, will fix on next rebase.

> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > Cc: Christoph Lameter <cl@xxxxxxxxx>
> > Cc: Pekka Enberg <penberg@xxxxxxxxxx>
> > Cc: David Rientjes <rientjes@xxxxxxxxxx>
> > Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Vlastimil Babka <vbabka@xxxxxxx>
> > Cc: Roman Gushchin <roman.gushchin@xxxxxxxxx>
> > Cc: Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx>
> > Cc: <linux-mm@xxxxxxxxx>
> > ---
> >  include/linux/slab.h | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/include/linux/slab.h b/include/linux/slab.h
> > index 90877fcde70bd..446303e385265 100644
> > --- a/include/linux/slab.h
> > +++ b/include/linux/slab.h
> > @@ -76,6 +76,12 @@
> >   * rcu_read_lock before reading the address, then rcu_read_unlock after
> >   * taking the spinlock within the structure expected at that address.
> >   *
> > + * Note that it is not possible to acquire a lock within a structure
> > + * allocated with SLAB_DESTROY_BY_RCU without first acquiring a reference
> > + * as described above.  The reason is that SLAB_DESTROY_BY_RCU pages are
> > + * not zeroed before being given to the slab, which means that any locks
> > + * must be initialized after each and every kmem_struct_alloc().
> > + *
> 
> Wonder if slab caches with a constructor should be OK here as AFAIK it
> should mean the object has to be in the initialized state both when
> allocated and freed?

It does look that way, thank you!

And __i915_request_ctor(), sighand_ctor(), and anon_vma_ctor() actually
do this, initializing a lock in the process.

The ctor function could just initialize the locks, and all would be well.
In addition, this makes sequence-lock-like approaches a bit easier, as in
"just use a sequence lock".

I will update with attribution.

							Thanx, Paul

> >   * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
> >   */
> >  /* Defer freeing slabs to RCU */
>