rhashtable: Remove RCU marking from rhash_lock_head
authorHerbert Xu <herbert@gondor.apana.org.au>
Thu, 16 May 2019 07:19:46 +0000 (15:19 +0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 16 May 2019 16:45:20 +0000 (09:45 -0700)
The opaque type rhash_lock_head should not be marked with __rcu
because it can never be dereferenced.  We should apply the RCU
marking when we turn it into a pointer which can be dereferenced.

This patch does exactly that.  This fixes a number of sparse
warnings as well as getting rid of some unnecessary RCU checking.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/rhashtable.h
lib/rhashtable.c

index f7714d3b46bd03139e9072653a22b8750d5ad6e5..9f8bc06d413686dc48b2a90de19d326da157b465 100644 (file)
@@ -84,7 +84,7 @@ struct bucket_table {
 
        struct lockdep_map      dep_map;
 
-       struct rhash_lock_head __rcu *buckets[] ____cacheline_aligned_in_smp;
+       struct rhash_lock_head *buckets[] ____cacheline_aligned_in_smp;
 };
 
 /*
@@ -261,13 +261,13 @@ void rhashtable_free_and_destroy(struct rhashtable *ht,
                                 void *arg);
 void rhashtable_destroy(struct rhashtable *ht);
 
-struct rhash_lock_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
-                                                unsigned int hash);
-struct rhash_lock_head __rcu **__rht_bucket_nested(const struct bucket_table *tbl,
-                                                  unsigned int hash);
-struct rhash_lock_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
-                                                       struct bucket_table *tbl,
-                                                       unsigned int hash);
+struct rhash_lock_head **rht_bucket_nested(const struct bucket_table *tbl,
+                                          unsigned int hash);
+struct rhash_lock_head **__rht_bucket_nested(const struct bucket_table *tbl,
+                                            unsigned int hash);
+struct rhash_lock_head **rht_bucket_nested_insert(struct rhashtable *ht,
+                                                 struct bucket_table *tbl,
+                                                 unsigned int hash);
 
 #define rht_dereference(p, ht) \
        rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
@@ -284,21 +284,21 @@ struct rhash_lock_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
 #define rht_entry(tpos, pos, member) \
        ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
 
-static inline struct rhash_lock_head __rcu *const *rht_bucket(
+static inline struct rhash_lock_head *const *rht_bucket(
        const struct bucket_table *tbl, unsigned int hash)
 {
        return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
                                     &tbl->buckets[hash];
 }
 
-static inline struct rhash_lock_head __rcu **rht_bucket_var(
+static inline struct rhash_lock_head **rht_bucket_var(
        struct bucket_table *tbl, unsigned int hash)
 {
        return unlikely(tbl->nest) ? __rht_bucket_nested(tbl, hash) :
                                     &tbl->buckets[hash];
 }
 
-static inline struct rhash_lock_head __rcu **rht_bucket_insert(
+static inline struct rhash_lock_head **rht_bucket_insert(
        struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
 {
        return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
@@ -349,6 +349,12 @@ static inline void rht_unlock(struct bucket_table *tbl,
        local_bh_enable();
 }
 
+static inline struct rhash_head __rcu *__rht_ptr(
+       struct rhash_lock_head *const *bkt)
+{
+       return (struct rhash_head __rcu *)((unsigned long)*bkt & ~BIT(0));
+}
+
 /*
  * Where 'bkt' is a bucket and might be locked:
  *   rht_ptr() dereferences that pointer and clears the lock bit.
@@ -356,30 +362,30 @@ static inline void rht_unlock(struct bucket_table *tbl,
  *            access is guaranteed, such as when destroying the table.
  */
 static inline struct rhash_head *rht_ptr(
-       struct rhash_lock_head __rcu * const *bkt,
+       struct rhash_lock_head *const *bkt,
        struct bucket_table *tbl,
        unsigned int hash)
 {
-       const struct rhash_lock_head *p =
-               rht_dereference_bucket_rcu(*bkt, tbl, hash);
+       struct rhash_head __rcu *p = __rht_ptr(bkt);
 
-       if ((((unsigned long)p) & ~BIT(0)) == 0)
+       if (!p)
                return RHT_NULLS_MARKER(bkt);
-       return (void *)(((unsigned long)p) & ~BIT(0));
+
+       return rht_dereference_bucket_rcu(p, tbl, hash);
 }
 
 static inline struct rhash_head *rht_ptr_exclusive(
-       struct rhash_lock_head __rcu * const *bkt)
+       struct rhash_lock_head *const *bkt)
 {
-       const struct rhash_lock_head *p =
-               rcu_dereference_protected(*bkt, 1);
+       struct rhash_head __rcu *p = __rht_ptr(bkt);
 
        if (!p)
                return RHT_NULLS_MARKER(bkt);
-       return (void *)(((unsigned long)p) & ~BIT(0));
+
+       return rcu_dereference_protected(p, 1);
 }
 
-static inline void rht_assign_locked(struct rhash_lock_head __rcu **bkt,
+static inline void rht_assign_locked(struct rhash_lock_head **bkt,
                                     struct rhash_head *obj)
 {
        struct rhash_head __rcu **p = (struct rhash_head __rcu **)bkt;
@@ -390,7 +396,7 @@ static inline void rht_assign_locked(struct rhash_lock_head __rcu **bkt,
 }
 
 static inline void rht_assign_unlock(struct bucket_table *tbl,
-                                    struct rhash_lock_head __rcu **bkt,
+                                    struct rhash_lock_head **bkt,
                                     struct rhash_head *obj)
 {
        struct rhash_head __rcu **p = (struct rhash_head __rcu **)bkt;
@@ -587,7 +593,7 @@ static inline struct rhash_head *__rhashtable_lookup(
                .ht = ht,
                .key = key,
        };
-       struct rhash_lock_head __rcu * const *bkt;
+       struct rhash_lock_head *const *bkt;
        struct bucket_table *tbl;
        struct rhash_head *he;
        unsigned int hash;
@@ -703,7 +709,7 @@ static inline void *__rhashtable_insert_fast(
                .ht = ht,
                .key = key,
        };
-       struct rhash_lock_head __rcu **bkt;
+       struct rhash_lock_head **bkt;
        struct rhash_head __rcu **pprev;
        struct bucket_table *tbl;
        struct rhash_head *head;
@@ -989,7 +995,7 @@ static inline int __rhashtable_remove_fast_one(
        struct rhash_head *obj, const struct rhashtable_params params,
        bool rhlist)
 {
-       struct rhash_lock_head __rcu **bkt;
+       struct rhash_lock_head **bkt;
        struct rhash_head __rcu **pprev;
        struct rhash_head *he;
        unsigned int hash;
@@ -1141,7 +1147,7 @@ static inline int __rhashtable_replace_fast(
        struct rhash_head *obj_old, struct rhash_head *obj_new,
        const struct rhashtable_params params)
 {
-       struct rhash_lock_head __rcu **bkt;
+       struct rhash_lock_head **bkt;
        struct rhash_head __rcu **pprev;
        struct rhash_head *he;
        unsigned int hash;
index 6529fe1b45c13421d69665326ba620e7c39d01e0..7708699a5b96d0e09a611d9f2419e97a84a07e64 100644 (file)
@@ -34,7 +34,7 @@
 
 union nested_table {
        union nested_table __rcu *table;
-       struct rhash_lock_head __rcu *bucket;
+       struct rhash_lock_head *bucket;
 };
 
 static u32 head_hashfn(struct rhashtable *ht,
@@ -216,7 +216,7 @@ static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 }
 
 static int rhashtable_rehash_one(struct rhashtable *ht,
-                                struct rhash_lock_head __rcu **bkt,
+                                struct rhash_lock_head **bkt,
                                 unsigned int old_hash)
 {
        struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
@@ -269,7 +269,7 @@ static int rhashtable_rehash_chain(struct rhashtable *ht,
                                    unsigned int old_hash)
 {
        struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
-       struct rhash_lock_head __rcu **bkt = rht_bucket_var(old_tbl, old_hash);
+       struct rhash_lock_head **bkt = rht_bucket_var(old_tbl, old_hash);
        int err;
 
        if (!bkt)
@@ -478,7 +478,7 @@ fail:
 }
 
 static void *rhashtable_lookup_one(struct rhashtable *ht,
-                                  struct rhash_lock_head __rcu **bkt,
+                                  struct rhash_lock_head **bkt,
                                   struct bucket_table *tbl, unsigned int hash,
                                   const void *key, struct rhash_head *obj)
 {
@@ -529,7 +529,7 @@ static void *rhashtable_lookup_one(struct rhashtable *ht,
 }
 
 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
-                                                 struct rhash_lock_head __rcu **bkt,
+                                                 struct rhash_lock_head **bkt,
                                                  struct bucket_table *tbl,
                                                  unsigned int hash,
                                                  struct rhash_head *obj,
@@ -584,7 +584,7 @@ static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
 {
        struct bucket_table *new_tbl;
        struct bucket_table *tbl;
-       struct rhash_lock_head __rcu **bkt;
+       struct rhash_lock_head **bkt;
        unsigned int hash;
        void *data;
 
@@ -1166,8 +1166,8 @@ void rhashtable_destroy(struct rhashtable *ht)
 }
 EXPORT_SYMBOL_GPL(rhashtable_destroy);
 
-struct rhash_lock_head __rcu **__rht_bucket_nested(const struct bucket_table *tbl,
-                                                  unsigned int hash)
+struct rhash_lock_head **__rht_bucket_nested(const struct bucket_table *tbl,
+                                            unsigned int hash)
 {
        const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
        unsigned int index = hash & ((1 << tbl->nest) - 1);
@@ -1195,10 +1195,10 @@ struct rhash_lock_head __rcu **__rht_bucket_nested(const struct bucket_table *tb
 }
 EXPORT_SYMBOL_GPL(__rht_bucket_nested);
 
-struct rhash_lock_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
-                                                unsigned int hash)
+struct rhash_lock_head **rht_bucket_nested(const struct bucket_table *tbl,
+                                          unsigned int hash)
 {
-       static struct rhash_lock_head __rcu *rhnull;
+       static struct rhash_lock_head *rhnull;
 
        if (!rhnull)
                INIT_RHT_NULLS_HEAD(rhnull);
@@ -1206,9 +1206,9 @@ struct rhash_lock_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
 }
 EXPORT_SYMBOL_GPL(rht_bucket_nested);
 
-struct rhash_lock_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
-                                                       struct bucket_table *tbl,
-                                                       unsigned int hash)
+struct rhash_lock_head **rht_bucket_nested_insert(struct rhashtable *ht,
+                                                 struct bucket_table *tbl,
+                                                 unsigned int hash)
 {
        const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
        unsigned int index = hash & ((1 << tbl->nest) - 1);