Messages in this thread | | | Date | Fri, 04 Sep 2009 00:08:42 +0200 | From | Eric Dumazet <> | Subject | Re: [PATCH] slub: fix slab_pad_check() |
| |
Christoph Lameter a écrit : > On Thu, 3 Sep 2009, Paul E. McKenney wrote: > >> 2. CPU 0 discovers that the slab cache can now be destroyed. >> >> It determines that there are no users, and has guaranteed >> that there will be no future users. So it knows that it >> can safely do kmem_cache_destroy(). >> >> 3. In absence of rcu_barrier(), kmem_cache_destroy() would >> immediately tear down the slab data structures. > > Of course. This has been discussed before. > > You need to ensure that no objects are in use before destroying a slab. In > case of DESTROY_BY_RCU you must ensure that there are no potential > readers. So use a suitable rcu barrier or something else like a > synchronize_rcu... > >>> But going through the RCU period is pointless since no user of the cache >>> remains. >> Which is irrelevant. The outstanding RCU callback was posted by the >> slab cache itself, -not- by the user of the slab cache. > > There will be no rcu callbacks generated at kmem_cache_destroy with the > patch I posted. > >>> The dismantling does not need RCU since there are no operations on the >>> objects in progress. So simply switch DESTROY_BY_RCU off for close. >> Unless I am missing something, this patch re-introduces the bug that >> the rcu_barrier() was added to prevent. So, in absence of a better >> explanation of what I am missing: > > The "fix" was ill advised. Slab users must ensure that no objects are in > use before destroying a slab. Only the slab users know how the objects > are being used. The slab allocator itself cannot know how to ensure that > there are no pending references. Putting a rcu_barrier in there creates an > inconsistency in the operation of kmem_cache_destroy() and an expectation > of functionality that the function cannot provide. >
Problem is not _objects_ Christoph, but _slabs_, and your patch is not working.
Its true that when User calls kmem_cache_destroy(), all _objects_ were previously freed. This is mandatory, with or withou SLAB_DESTROY_BY_RCU thing
Problem is that slub has some internal state, including some to-be-freed _slabs_, that User have no control at all on it.
User cannot "know" slabs are freed, inuse, or whatever state in cache or call_rcu queues.
Face it, SLAB_DESTROY_BY_RCU is internal affair (to slub/slab/... allocators)
We absolutely need a rcu_barrier() somewhere, believe it or not. You can argue that it should be done *before*, but it gives no speedup, only potential bugs.
Only case User should do its rcu_barrier() itself is if it knows some call_rcu() are pending and are delaying _objects_ freeing (typical !SLAB_DESTROY_RCU usage in RCU algos).
I dont even understand why you care so much about kmem_cache_destroy(SLAB_DESTROY_BY_RCU), given that almost nobody use it. We took almost one month to find out what the bug was in first place... -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |