Re: [PATCH] slub: fix slab_pad_check()

[Eric Dumazet <eric.dumazet@gmail.com>]

Christoph Lameter a écrit :
> On Thu, 3 Sep 2009, Eric Dumazet wrote:
> 
>> Christoph Lameter a ?crit :
>>> On Thu, 3 Sep 2009, Eric Dumazet wrote:
>>>
>>>> on a SLAB_DESTROY_BY_RCU cache, there is no need to try to optimize this
>>>> rcu_barrier() call, unless we want superfast reboot/halt sequences...
>>> I stilll think that the action to quiesce rcu is something that the caller
>>> of kmem_cache_destroy must take care of.
>> Do you mean :
>>
>> if (kmem_cache_shrink(s) == 0) {
>> 	rcu_barrier();
>> 	kmem_cache_destroy_no_rcu_barrier(s);
>> } else {
>> 	kmem_cache_destroy_with_rcu_barrier_because_SLAB_DESTROY_BY_RCU_cache(s);
>> }
>>
>> What would be the point ?
> 
> The above is port of slub?

No, I am trying to code what you suggest, and I could not find a clean way with
current API (SLAB/SLUB/SLOB)

> 
> I mean that (in this case) the net subsystem would have to deal with RCU quietness
> before disposing of the slab cache. There may be multiple ways of dealing
> with RCU. The RCU barrier may be unnecessary for future uses. Typically
> one would expect that all deferred handling of structures must be complete
> for correctness before disposing of the whole cache.

Point is we cannot deal with RCU quietness before disposing the slab cache,
(if SLAB_DESTROY_BY_RCU was set on the cache) since this disposing *will* 
make call_rcu() calls when a full slab is freed/purged.
And when RCU grace period is elapsed, the callback *will* need access to 
the cache we want to dismantle. Better to not have kfreed()/poisoned it...

I believe you mix two RCU uses here.

1) The one we all know, is use normal caches (!SLAB_DESTROY_BY_RCU) 
(or kmalloc()), and use call_rcu(... kfree_something)

   In this case, you are 100% right that the subsystem itself has
   to call rcu_barrier() (or respect whatever self-synchro) itself,
   before calling kmem_cache_destroy()

2) The SLAB_DESTROY_BY_RCU one.

   Part of cache dismantle needs to call rcu_barrier() itself.
   Caller doesnt have to use rcu_barrier(). It would be a waste of time,
   as kmem_cache_destroy() will refill rcu wait queues with its own stuff.