1Using hlist_nulls to protect read-mostly linked lists and 2objects using SLAB_TYPESAFE_BY_RCU allocations. 3 4Please read the basics in Documentation/RCU/listRCU.txt 5 6Using special makers (called 'nulls') is a convenient way 7to solve following problem : 8 9A typical RCU linked list managing objects which are 10allocated with SLAB_TYPESAFE_BY_RCU kmem_cache can 11use following algos : 12 131) Lookup algo 14-------------- 15rcu_read_lock() 16begin: 17obj = lockless_lookup(key); 18if (obj) { 19 if (!try_get_ref(obj)) // might fail for free objects 20 goto begin; 21 /* 22 * Because a writer could delete object, and a writer could 23 * reuse these object before the RCU grace period, we 24 * must check key after getting the reference on object 25 */ 26 if (obj->key != key) { // not the object we expected 27 put_ref(obj); 28 goto begin; 29 } 30} 31rcu_read_unlock(); 32 33Beware that lockless_lookup(key) cannot use traditional hlist_for_each_entry_rcu() 34but a version with an additional memory barrier (smp_rmb()) 35 36lockless_lookup(key) 37{ 38 struct hlist_node *node, *next; 39 for (pos = rcu_dereference((head)->first); 40 pos && ({ next = pos->next; smp_rmb(); prefetch(next); 1; }) && 41 ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1; }); 42 pos = rcu_dereference(next)) 43 if (obj->key == key) 44 return obj; 45 return NULL; 46 47And note the traditional hlist_for_each_entry_rcu() misses this smp_rmb() : 48 49 struct hlist_node *node; 50 for (pos = rcu_dereference((head)->first); 51 pos && ({ prefetch(pos->next); 1; }) && 52 ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1; }); 53 pos = rcu_dereference(pos->next)) 54 if (obj->key == key) 55 return obj; 56 return NULL; 57} 58 59Quoting Corey Minyard : 60 61"If the object is moved from one list to another list in-between the 62 time the hash is calculated and the next field is accessed, and the 63 object has moved to the end of a new list, the traversal will not 64 complete properly on the list it should have, since the object will 65 be on the end of the new list and there's not a way to tell it's on a 66 new list and restart the list traversal. I think that this can be 67 solved by pre-fetching the "next" field (with proper barriers) before 68 checking the key." 69 702) Insert algo : 71---------------- 72 73We need to make sure a reader cannot read the new 'obj->obj_next' value 74and previous value of 'obj->key'. Or else, an item could be deleted 75from a chain, and inserted into another chain. If new chain was empty 76before the move, 'next' pointer is NULL, and lockless reader can 77not detect it missed following items in original chain. 78 79/* 80 * Please note that new inserts are done at the head of list, 81 * not in the middle or end. 82 */ 83obj = kmem_cache_alloc(...); 84lock_chain(); // typically a spin_lock() 85obj->key = key; 86/* 87 * we need to make sure obj->key is updated before obj->next 88 * or obj->refcnt 89 */ 90smp_wmb(); 91atomic_set(&obj->refcnt, 1); 92hlist_add_head_rcu(&obj->obj_node, list); 93unlock_chain(); // typically a spin_unlock() 94 95 963) Remove algo 97-------------- 98Nothing special here, we can use a standard RCU hlist deletion. 99But thanks to SLAB_TYPESAFE_BY_RCU, beware a deleted object can be reused 100very very fast (before the end of RCU grace period) 101 102if (put_last_reference_on(obj) { 103 lock_chain(); // typically a spin_lock() 104 hlist_del_init_rcu(&obj->obj_node); 105 unlock_chain(); // typically a spin_unlock() 106 kmem_cache_free(cachep, obj); 107} 108 109 110 111-------------------------------------------------------------------------- 112With hlist_nulls we can avoid extra smp_rmb() in lockless_lookup() 113and extra smp_wmb() in insert function. 114 115For example, if we choose to store the slot number as the 'nulls' 116end-of-list marker for each slot of the hash table, we can detect 117a race (some writer did a delete and/or a move of an object 118to another chain) checking the final 'nulls' value if 119the lookup met the end of chain. If final 'nulls' value 120is not the slot number, then we must restart the lookup at 121the beginning. If the object was moved to the same chain, 122then the reader doesn't care : It might eventually 123scan the list again without harm. 124 125 1261) lookup algo 127 128 head = &table[slot]; 129 rcu_read_lock(); 130begin: 131 hlist_nulls_for_each_entry_rcu(obj, node, head, member) { 132 if (obj->key == key) { 133 if (!try_get_ref(obj)) // might fail for free objects 134 goto begin; 135 if (obj->key != key) { // not the object we expected 136 put_ref(obj); 137 goto begin; 138 } 139 goto out; 140 } 141/* 142 * if the nulls value we got at the end of this lookup is 143 * not the expected one, we must restart lookup. 144 * We probably met an item that was moved to another chain. 145 */ 146 if (get_nulls_value(node) != slot) 147 goto begin; 148 obj = NULL; 149 150out: 151 rcu_read_unlock(); 152 1532) Insert function : 154-------------------- 155 156/* 157 * Please note that new inserts are done at the head of list, 158 * not in the middle or end. 159 */ 160obj = kmem_cache_alloc(cachep); 161lock_chain(); // typically a spin_lock() 162obj->key = key; 163/* 164 * changes to obj->key must be visible before refcnt one 165 */ 166smp_wmb(); 167atomic_set(&obj->refcnt, 1); 168/* 169 * insert obj in RCU way (readers might be traversing chain) 170 */ 171hlist_nulls_add_head_rcu(&obj->obj_node, list); 172unlock_chain(); // typically a spin_unlock() 173