Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 21 Oct 2022 06:43:09 -0700
From:   "Paul E. McKenney" <paulmck@kernel.org>
To:     Christoph Lameter <cl@gentwo.de>
Cc:     rcu@vger.kernel.org, linux-kernel@vger.kernel.org,
        kernel-team@fb.com, rostedt@goodmis.org,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Vlastimil Babka <vbabka@suse.cz>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Hyeonggon Yoo <42.hyeyoo@gmail.com>, linux-mm@kvack.org
Subject: Re: [PATCH rcu 5/8] slab: Explain why SLAB_DESTROY_BY_RCU reference
 before locking
Message-ID: <20221021134309.GG5600@paulmck-ThinkPad-P17-Gen-1>
Reply-To: paulmck@kernel.org
References: <20221019224652.GA2499358@paulmck-ThinkPad-P17-Gen-1>
 <20221019224659.2499511-5-paulmck@kernel.org>
 <03d5730-9241-542d-76c6-728be4487c4@gentwo.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <03d5730-9241-542d-76c6-728be4487c4@gentwo.de>
Precedence: bulk

On Fri, Oct 21, 2022 at 09:44:23AM +0200, Christoph Lameter wrote:
> On Wed, 19 Oct 2022, 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.
> 
> Sorry but this is not correct and difficult to comprehend.
> 
> 1. You do not need a reference to a slab object after it was allocated.
>    Objects must be properly protected by rcu_locks.
> 
> 2. Locks are initialized once on slab allocation via a constructor (*not* on object allocation via kmem_cache_alloc)
> 
> 3. Modifying locks at allocation/free is not possible since references to
>    these objects may still persist after free and before alloc.
> 
> 4. The old term SLAB_DESTROY_BY_RCU is used here.

Thank you for looking this over, but Vlastimil beat you to it.  How does
the update below look?

							Thanx, Paul

------------------------------------------------------------------------

commit ff4c536e6b44e2e185e38c3653851f92e07139da
Author: Paul E. McKenney <paulmck@kernel.org>
Date:   Mon Sep 26 08:57:56 2022 -0700

    slab: Explain why SLAB_TYPESAFE_BY_RCU reference before locking
    
    It is not obvious to the casual user why it is absolutely necessary to
    acquire a reference to a SLAB_TYPESAFE_BY_RCU structure before acquiring
    a lock in that structure.  Therefore, add a comment explaining this point.
    
    [ paulmck: Apply Vlastimil Babka feedback. ]
    
    Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
    Cc: Christoph Lameter <cl@linux.com>
    Cc: Pekka Enberg <penberg@kernel.org>
    Cc: David Rientjes <rientjes@google.com>
    Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
    Cc: Andrew Morton <akpm@linux-foundation.org>
    Cc: Vlastimil Babka <vbabka@suse.cz>
    Cc: Roman Gushchin <roman.gushchin@linux.dev>
    Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com>
    Cc: <linux-mm@kvack.org>

diff --git a/include/linux/slab.h b/include/linux/slab.h
index 90877fcde70bd..487418c7ea8cd 100644
--- a/include/linux/slab.h
+++ b/include/linux/slab.h
@@ -76,6 +76,17 @@
  * 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_TYPESAFE_BY_RCU without first acquiring a reference
+ * as described above.  The reason is that SLAB_TYPESAFE_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().
+ * Alternatively, make the ctor passed to kmem_cache_create() initialize
+ * the locks at page-allocation time, as is done in __i915_request_ctor(),
+ * sighand_ctor(), and anon_vma_ctor().  Such a ctor permits readers
+ * to safely acquire those ctor-initialized locks under rcu_read_lock()
+ * protection.
+ *
  * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
  */
 /* Defer freeing slabs to RCU */