On Mon 21-01-13 20:12:17, Glauber Costa wrote: > On 01/21/2013 08:07 PM, Michal Hocko wrote: > >> > And the reason why kmemcg holds the set_limit mutex > >> > is just to protect from itself, then there is no *need* to hold any > >> > extra lock (and we'll never be able to stop holding the creation lock, > >> > whatever it is). So my main point here is not memcg_mutex vs > >> > set_limit_mutex, but rather, memcg_mutex is needed anyway, and once it > >> > is taken, the set_limit_mutex *can* be held, but doesn't need to. > > So you can update kmem specific usage of set_limit_mutex. > Meaning ? I thought you've said it is not needed and the code says that: - memcg_propagate_kmem is called with memcg_mutex held in css_alloc - memcg_update_kmem_limit takes both of them - kmem_cache_destroy_memcg_children _doesn't_ take both So one obvious way to go would be changing kmem_cache_destroy_memcg_children to memcg_mutex and removing set_limit_mutex from other two paths. This would leave set_limit_mutex lock for its original intention. -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>