Re: [PATCH 00/14] replace call_rcu by kfree_rcu for simple kmem_cache_free callback

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Jun 18, 2024 at 07:21:42PM +0200, Vlastimil Babka wrote:
> On 6/18/24 6:48 PM, Paul E. McKenney wrote:
> > On Tue, Jun 18, 2024 at 11:31:00AM +0200, Uladzislau Rezki wrote:
> >> > On 6/17/24 8:42 PM, Uladzislau Rezki wrote:
> >> > >> +
> >> > >> +	s = container_of(work, struct kmem_cache, async_destroy_work);
> >> > >> +
> >> > >> +	// XXX use the real kmem_cache_free_barrier() or similar thing here
> >> > > It implies that we need to introduce kfree_rcu_barrier(), a new API, which i
> >> > > wanted to avoid initially.
> >> > 
> >> > I wanted to avoid new API or flags for kfree_rcu() users and this would
> >> > be achieved. The barrier is used internally so I don't consider that an
> >> > API to avoid. How difficult is the implementation is another question,
> >> > depending on how the current batching works. Once (if) we have sheaves
> >> > proven to work and move kfree_rcu() fully into SLUB, the barrier might
> >> > also look different and hopefully easier. So maybe it's not worth to
> >> > invest too much into that barrier and just go for the potentially
> >> > longer, but easier to implement?
> >> > 
> >> Right. I agree here. If the cache is not empty, OK, we just defer the
> >> work, even we can use a big 21 seconds delay, after that we just "warn"
> >> if it is still not empty and leave it as it is, i.e. emit a warning and
> >> we are done.
> >> 
> >> Destroying the cache is not something that must happen right away. 
> > 
> > OK, I have to ask...
> > 
> > Suppose that the cache is created and destroyed by a module and
> > init/cleanup time, respectively.  Suppose that this module is rmmod'ed
> > then very quickly insmod'ed.
> > 
> > Do we need to fail the insmod if the kmem_cache has not yet been fully
> > cleaned up?
> 
> We don't have any such link between kmem_cache and module to detect that, so
> we would have to start tracking that. Probably not worth the trouble.

Fair enough!

> >  If not, do we have two versions of the same kmem_cache in
> > /proc during the overlap time?
> 
> Hm could happen in /proc/slabinfo but without being harmful other than
> perhaps confusing someone. We could filter out the caches being destroyed
> trivially.

Or mark them in /proc/slabinfo?  Yet another column, yay!!!  Or script
breakage from flagging the name somehow, for example, trailing "/"
character.

> Sysfs and debugfs might be more problematic as I suppose directory names
> would clash. I'll have to check... might be even happening now when we do
> detect leaked objects and just leave the cache around... thanks for the
> question.

"It is a service that I provide."  ;-)

But yes, we might be living with it already and there might already
be ways people deal with it.

							Thanx, Paul

> >> > > Since you do it asynchronous can we just repeat
> >> > > and wait until it a cache is furry freed?
> >> > 
> >> > The problem is we want to detect the cases when it's not fully freed
> >> > because there was an actual read. So at some point we'd need to stop the
> >> > repeats because we know there can no longer be any kfree_rcu()'s in
> >> > flight since the kmem_cache_destroy() was called.
> >> > 
> >> Agree. As noted above, we can go with 21 seconds(as an example) interval
> >> and just perform destroy(without repeating).
> >> 
> >> --
> >> Uladzislau Rezki
> 




[Index of Archives]     [Kernel Development]     [Kernel Announce]     [Kernel Newbies]     [Linux Networking Development]     [Share Photos]     [IDE]     [Security]     [Git]     [Netfilter]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Device Mapper]

  Powered by Linux