On Fri, 11 Nov 2011, Shaohua Li wrote: > Looks this could be a real dead lock. we hold a lock to free a object, > but the free need allocate a new object. if the new object and the freed > object are from the same slab, there is a deadlock. unfreeze partials is never called when going through get_partial_node() so there is no deadlock AFAICT. > discard_slab() doesn't need hold the lock if the slab is already removed > from partial list. how about below patch, only compile tested. In general I think it is good to move the call to discard_slab() out from under the list_lock in unfreeze_partials(). Could you fold discard_page_list into unfreeze_partials()? __flush_cpu_slab still calls discard_page_list with disabled interrupts even after your patch. -- 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>