Fine-grained add / delete locking

Add a bucket-level lock bit. Use a spinlock only when actually
allocating, freeing, or splitting a bucket. Should improve
multi-thread add/del performance.

Change-Id: I3e40e2a8371685457f340d6584dea14e3207f2b0
Signed-off-by: Dave Barach <dave@barachs.net>

9 files changed, 234 insertions, 354 deletions

diff --git a/src/vppinfra/bihash_16_8.h b/src/vppinfra/bihash_16_8.h
index 6076487934d..5949e2b3549 100644
--- a/src/vppinfra/bihash_16_8.h
+++ b/src/vppinfra/bihash_16_8.h
@@ -13,12 +13,10 @@
  * limitations under the License.
  */
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _16_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_16_8_h__
 #define __included_bihash_16_8_h__
diff --git a/src/vppinfra/bihash_24_8.h b/src/vppinfra/bihash_24_8.h
index dcce2a21259..c028352d0e9 100644
--- a/src/vppinfra/bihash_24_8.h
+++ b/src/vppinfra/bihash_24_8.h
@@ -13,12 +13,10 @@
  * limitations under the License.
  */
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _24_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_24_8_h__
 #define __included_bihash_24_8_h__
diff --git a/src/vppinfra/bihash_40_8.h b/src/vppinfra/bihash_40_8.h
index 90adb8f5b10..fe57ba03966 100644
--- a/src/vppinfra/bihash_40_8.h
+++ b/src/vppinfra/bihash_40_8.h
@@ -14,12 +14,10 @@
  */
 
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _40_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_40_8_h__
 #define __included_bihash_40_8_h__
diff --git a/src/vppinfra/bihash_48_8.h b/src/vppinfra/bihash_48_8.h
index e981950b1bd..9aebe66d264 100644
--- a/src/vppinfra/bihash_48_8.h
+++ b/src/vppinfra/bihash_48_8.h
@@ -14,12 +14,10 @@
  */
 
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _48_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_48_8_h__
 #define __included_bihash_48_8_h__
diff --git a/src/vppinfra/bihash_8_8.h b/src/vppinfra/bihash_8_8.h
index 2deb64ef438..604cf30808f 100644
--- a/src/vppinfra/bihash_8_8.h
+++ b/src/vppinfra/bihash_8_8.h
@@ -13,12 +13,10 @@
  * limitations under the License.
  */
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _8_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_8_8_h__
 #define __included_bihash_8_8_h__
diff --git a/src/vppinfra/bihash_template.c b/src/vppinfra/bihash_template.c
index 53ffbf156e6..af624fee070 100644
--- a/src/vppinfra/bihash_template.c
+++ b/src/vppinfra/bihash_template.c
@@ -37,7 +37,6 @@ void BV (clib_bihash_init)
   (BVT (clib_bihash) * h, char *name, u32 nbuckets, uword memory_size)
 {
   uword bucket_size;
-  int i;
 
   nbuckets = 1 << (max_log2 (nbuckets));
 
@@ -47,6 +46,14 @@ void BV (clib_bihash_init)
   h->cache_hits = 0;
   h->cache_misses = 0;
 
+  /*
+   * Make sure the requested size is rational. The max table
+   * size without playing the alignment card is 64 Gbytes.
+   * If someone starts complaining that's not enough, we can shift
+   * the offset by CLIB_LOG2_CACHE_LINE_BYTES...
+   */
+  ASSERT (memory_size < (1ULL << BIHASH_BUCKET_OFFSET_BITS));
+
   h->alloc_arena = (uword) clib_mem_vm_alloc (memory_size);
   h->alloc_arena_next = h->alloc_arena;
   h->alloc_arena_size = memory_size;
@@ -54,11 +61,8 @@ void BV (clib_bihash_init)
   bucket_size = nbuckets * sizeof (h->buckets[0]);
   h->buckets = BV (alloc_aligned) (h, bucket_size);
 
-  h->writer_lock = BV (alloc_aligned) (h, CLIB_CACHE_LINE_BYTES);
-  h->writer_lock[0] = 0;
-
-  for (i = 0; i < nbuckets; i++)
-    BV (clib_bihash_reset_cache) (h->buckets + i);
+  h->alloc_lock = BV (alloc_aligned) (h, CLIB_CACHE_LINE_BYTES);
+  h->alloc_lock[0] = 0;
 
   h->fmt_fn = NULL;
 }
@@ -83,7 +87,7 @@ BV (value_alloc) (BVT (clib_bihash) * h, u32 log2_pages)
 {
   BVT (clib_bihash_value) * rv = 0;
 
-  ASSERT (h->writer_lock[0]);
+  ASSERT (h->alloc_lock[0]);
   if (log2_pages >= vec_len (h->freelists) || h->freelists[log2_pages] == 0)
     {
       vec_validate_init_empty (h->freelists, log2_pages, 0);
@@ -108,10 +112,13 @@ static void
 BV (value_free) (BVT (clib_bihash) * h, BVT (clib_bihash_value) * v,
 		 u32 log2_pages)
 {
-  ASSERT (h->writer_lock[0]);
+  ASSERT (h->alloc_lock[0]);
 
   ASSERT (vec_len (h->freelists) > log2_pages);
 
+  if (CLIB_DEBUG > 0)
+    memset (v, 0xFE, sizeof (*v) * (1 << log2_pages));
+
   v->next_free = h->freelists[log2_pages];
   h->freelists[log2_pages] = v;
 }
@@ -125,6 +132,8 @@ BV (make_working_copy) (BVT (clib_bihash) * h, BVT (clib_bihash_bucket) * b)
   u32 thread_index = os_get_thread_index ();
   int log2_working_copy_length;
 
+  ASSERT (h->alloc_lock[0]);
+
   if (thread_index >= vec_len (h->working_copies))
     {
       vec_validate (h->working_copies, thread_index);
@@ -154,10 +163,6 @@ BV (make_working_copy) (BVT (clib_bihash) * h, BVT (clib_bihash_bucket) * b)
       h->working_copies[thread_index] = working_copy;
     }
 
-  /* Lock the bucket... */
-  while (BV (clib_bihash_lock_bucket) (b) == 0)
-    ;
-
   v = BV (clib_bihash_get_value) (h, b->offset);
 
   clib_memcpy (working_copy, v, sizeof (*v) * (1 << b->log2_pages));
@@ -178,6 +183,8 @@ BV (split_and_rehash)
   BVT (clib_bihash_value) * new_values, *new_v;
   int i, j, length_in_kvs;
 
+  ASSERT (h->alloc_lock[0]);
+
   new_values = BV (value_alloc) (h, new_log2_pages);
   length_in_kvs = (1 << old_log2_pages) * BIHASH_KVP_PER_PAGE;
 
@@ -225,6 +232,8 @@ BV (split_and_rehash_linear)
   BVT (clib_bihash_value) * new_values;
   int i, j, new_length, old_length;
 
+  ASSERT (h->alloc_lock[0]);
+
   new_values = BV (value_alloc) (h, new_log2_pages);
   new_length = (1 << new_log2_pages) * BIHASH_KVP_PER_PAGE;
   old_length = (1 << old_log2_pages) * BIHASH_KVP_PER_PAGE;
@@ -266,7 +275,6 @@ int BV (clib_bihash_add_del)
   u32 bucket_index;
   BVT (clib_bihash_bucket) * b, tmp_b;
   BVT (clib_bihash_value) * v, *new_v, *save_new_v, *working_copy;
-  int rv = 0;
   int i, limit;
   u64 hash, new_hash;
   u32 new_log2_pages, old_log2_pages;
@@ -281,37 +289,36 @@ int BV (clib_bihash_add_del)
 
   hash >>= h->log2_nbuckets;
 
-  tmp_b.linear_search = 0;
-
-  while (__sync_lock_test_and_set (h->writer_lock, 1))
-    ;
+  BV (clib_bihash_lock_bucket) (b);
 
   /* First elt in the bucket? */
   if (BV (clib_bihash_bucket_is_empty) (b))
     {
       if (is_add == 0)
 	{
-	  rv = -1;
-	  goto unlock;
+	  BV (clib_bihash_unlock_bucket) (b);
+	  return (-1);
 	}
 
+      BV (clib_bihash_alloc_lock) (h);
       v = BV (value_alloc) (h, 0);
+      BV (clib_bihash_alloc_unlock) (h);
 
       *v->kvp = *add_v;
-      tmp_b.as_u64 = 0;
+      tmp_b.as_u64 = 0;		/* clears bucket lock */
       tmp_b.offset = BV (clib_bihash_get_offset) (h, v);
       tmp_b.refcnt = 1;
+      CLIB_MEMORY_BARRIER ();
 
       b->as_u64 = tmp_b.as_u64;
-      goto unlock;
+      BV (clib_bihash_unlock_bucket) (b);
+      return (0);
     }
 
-  /* Note: this leaves the cache disabled */
-  BV (make_working_copy) (h, b);
-
-  v = BV (clib_bihash_get_value) (h, h->saved_bucket.offset);
-
+  /* WARNING: we're still looking at the live copy... */
   limit = BIHASH_KVP_PER_PAGE;
+  v = BV (clib_bihash_get_value) (h, b->offset);
+
   v += (b->linear_search == 0) ? hash & ((1 << b->log2_pages) - 1) : 0;
   if (b->linear_search)
     limit <<= b->log2_pages;
@@ -319,6 +326,13 @@ int BV (clib_bihash_add_del)
   if (is_add)
     {
       /*
+       * Because reader threads are looking at live data,
+       * we have to be extra careful. Readers do NOT hold the
+       * bucket lock. We need to be SLOWER than a search, past the
+       * point where readers CHECK the bucket lock.
+       */
+
+      /*
        * For obvious (in hindsight) reasons, see if we're supposed to
        * replace an existing key, then look for an empty slot.
        */
@@ -326,52 +340,79 @@ int BV (clib_bihash_add_del)
 	{
 	  if (!memcmp (&(v->kvp[i]), &add_v->key, sizeof (add_v->key)))
 	    {
+	      CLIB_MEMORY_BARRIER ();	/* Add a delay */
 	      clib_memcpy (&(v->kvp[i]), add_v, sizeof (*add_v));
-	      CLIB_MEMORY_BARRIER ();
-	      /* Restore the previous (k,v) pairs */
-	      b->as_u64 = h->saved_bucket.as_u64;
-	      goto unlock;
+	      BV (clib_bihash_unlock_bucket) (b);
+	      return (0);
 	    }
 	}
+      /*
+       * Look for an empty slot. If found, use it
+       */
       for (i = 0; i < limit; i++)
 	{
 	  if (BV (clib_bihash_is_free) (&(v->kvp[i])))
 	    {
-	      clib_memcpy (&(v->kvp[i]), add_v, sizeof (*add_v));
-	      CLIB_MEMORY_BARRIER ();
-	      b->as_u64 = h->saved_bucket.as_u64;
+	      /*
+	       * Copy the value first, so that if a reader manages
+	       * to match the new key, the value will be right...
+	       */
+	      clib_memcpy (&(v->kvp[i].value),
+			   &add_v->value, sizeof (add_v->value));
+	      CLIB_MEMORY_BARRIER ();	/* Make sure the value has settled */
+	      clib_memcpy (&(v->kvp[i]), &add_v->key, sizeof (add_v->key));
 	      b->refcnt++;
-	      goto unlock;
+	      BV (clib_bihash_unlock_bucket) (b);
+	      return (0);
 	    }
 	}
-      /* no room at the inn... split case... */
+      /* Out of space in this bucket, split the bucket... */
     }
-  else
+  else				/* delete case */
     {
       for (i = 0; i < limit; i++)
 	{
+	  /* Found the key? Kill it... */
 	  if (!memcmp (&(v->kvp[i]), &add_v->key, sizeof (add_v->key)))
 	    {
 	      memset (&(v->kvp[i]), 0xff, sizeof (*(add_v)));
-	      CLIB_MEMORY_BARRIER ();
-	      if (PREDICT_TRUE (h->saved_bucket.refcnt > 1))
+	      /* Is the bucket empty? */
+	      if (PREDICT_TRUE (b->refcnt > 1))
 		{
-		  h->saved_bucket.refcnt -= 1;
-		  b->as_u64 = h->saved_bucket.as_u64;
-		  goto unlock;
+		  b->refcnt--;
+		  BV (clib_bihash_unlock_bucket) (b);
+		  return (0);
 		}
-	      else
+	      else		/* yes, free it */
 		{
-		  tmp_b.as_u64 = 0;
-		  goto free_old_bucket;
+		  /* Save old bucket value, need log2_pages to free it */
+		  tmp_b.as_u64 = b->as_u64;
+		  CLIB_MEMORY_BARRIER ();
+
+		  /* Kill and unlock the bucket */
+		  b->as_u64 = 0;
+
+		  /* And free the backing storage */
+		  BV (clib_bihash_alloc_lock) (h);
+		  /* Note: v currently points into the middle of the bucket */
+		  v = BV (clib_bihash_get_value) (h, tmp_b.offset);
+		  BV (value_free) (h, v, tmp_b.log2_pages);
+		  BV (clib_bihash_alloc_unlock) (h);
+		  return (0);
 		}
 	    }
 	}
-      rv = -3;
-      b->as_u64 = h->saved_bucket.as_u64;
-      goto unlock;
+      /* Not found... */
+      BV (clib_bihash_unlock_bucket) (b);
+      return (-3);
     }
 
+  /* Move readers to a (locked) temp copy of the bucket */
+  BV (clib_bihash_alloc_lock) (h);
+  BV (make_working_copy) (h, b);
+
+  v = BV (clib_bihash_get_value) (h, h->saved_bucket.offset);
+
   old_log2_pages = h->saved_bucket.log2_pages;
   new_log2_pages = old_log2_pages + 1;
   mark_bucket_linear = 0;
@@ -436,21 +477,11 @@ expand_ok:
   tmp_b.offset = BV (clib_bihash_get_offset) (h, save_new_v);
   tmp_b.linear_search = mark_bucket_linear;
   tmp_b.refcnt = h->saved_bucket.refcnt + 1;
-
-free_old_bucket:
-
+  tmp_b.lock = 0;
   CLIB_MEMORY_BARRIER ();
   b->as_u64 = tmp_b.as_u64;
-  v = BV (clib_bihash_get_value) (h, h->saved_bucket.offset);
-
-  BV (value_free) (h, v, h->saved_bucket.log2_pages);
-
-unlock:
-  BV (clib_bihash_reset_cache) (b);
-  BV (clib_bihash_unlock_bucket) (b);
-  CLIB_MEMORY_BARRIER ();
-  h->writer_lock[0] = 0;
-  return rv;
+  BV (clib_bihash_alloc_unlock) (h);
+  return (0);
 }
 
 int BV (clib_bihash_search)
@@ -460,9 +491,6 @@ int BV (clib_bihash_search)
   u64 hash;
   u32 bucket_index;
   BVT (clib_bihash_value) * v;
-#if BIHASH_KVP_CACHE_SIZE > 0
-  BVT (clib_bihash_kv) * kvp;
-#endif
   BVT (clib_bihash_bucket) * b;
   int i, limit;
 
@@ -476,23 +504,12 @@ int BV (clib_bihash_search)
   if (BV (clib_bihash_bucket_is_empty) (b))
     return -1;
 
-#if BIHASH_KVP_CACHE_SIZE > 0
-  /* Check the cache, if currently enabled */
-  if (PREDICT_TRUE ((b->cache_lru & (1 << 15)) == 0))
+  if (PREDICT_FALSE (b->lock))
     {
-      limit = BIHASH_KVP_CACHE_SIZE;
-      kvp = b->cache;
-      for (i = 0; i < limit; i++)
-	{
-	  if (BV (clib_bihash_key_compare) (kvp[i].key, search_key->key))
-	    {
-	      *valuep = kvp[i];
-	      h->cache_hits++;
-	      return 0;
-	    }
-	}
+      volatile BVT (clib_bihash_bucket) * bv = b;
+      while (bv->lock)
+	;
     }
-#endif
 
   hash >>= h->log2_nbuckets;
 
@@ -507,53 +524,12 @@ int BV (clib_bihash_search)
       if (BV (clib_bihash_key_compare) (v->kvp[i].key, search_key->key))
 	{
 	  *valuep = v->kvp[i];
-
-#if BIHASH_KVP_CACHE_SIZE > 0
-	  u8 cache_slot;
-	  /* Shut off the cache */
-	  if (BV (clib_bihash_lock_bucket) (b))
-	    {
-	      cache_slot = BV (clib_bihash_get_lru) (b);
-	      b->cache[cache_slot] = v->kvp[i];
-	      BV (clib_bihash_update_lru) (b, cache_slot);
-
-	      /* Reenable the cache */
-	      BV (clib_bihash_unlock_bucket) (b);
-	      h->cache_misses++;
-	    }
-#endif
 	  return 0;
 	}
     }
   return -1;
 }
 
-u8 *BV (format_bihash_lru) (u8 * s, va_list * args)
-{
-#if BIHASH_KVP_SIZE > 0
-  int i;
-  BVT (clib_bihash_bucket) * b = va_arg (*args, BVT (clib_bihash_bucket) *);
-  u16 cache_lru = b->cache_lru;
-
-  s = format (s, "cache %s, order ", cache_lru & (1 << 15) ? "on" : "off");
-
-  for (i = 0; i < BIHASH_KVP_CACHE_SIZE; i++)
-    s = format (s, "[%d] ", ((cache_lru >> (3 * i)) & 7));
-
-  return (s);
-#else
-  return format (s, "cache not configured");
-#endif
-}
-
-void
-BV (clib_bihash_update_lru_not_inline) (BVT (clib_bihash_bucket) * b, u8 slot)
-{
-#if BIHASH_KVP_SIZE > 0
-  BV (clib_bihash_update_lru) (b, slot);
-#endif
-}
-
 u8 *BV (format_bihash) (u8 * s, va_list * args)
 {
   BVT (clib_bihash) * h = va_arg (*args, BVT (clib_bihash) *);
diff --git a/src/vppinfra/bihash_template.h b/src/vppinfra/bihash_template.h
index eb50d32d521..6303992bc57 100644
--- a/src/vppinfra/bihash_template.h
+++ b/src/vppinfra/bihash_template.h
@@ -49,9 +49,7 @@ typedef struct BV (clib_bihash_value)
   };
 } BVT (clib_bihash_value);
 
-#if BIHASH_KVP_CACHE_SIZE > 5
-#error Requested KVP cache LRU data exceeds 16 bits
-#endif
+#define BIHASH_BUCKET_OFFSET_BITS 36
 
 typedef struct
 {
@@ -59,28 +57,23 @@ typedef struct
   {
     struct
     {
-      u64 offset:37;
+      u64 offset:BIHASH_BUCKET_OFFSET_BITS;
+      u64 lock:1;
       u64 linear_search:1;
       u64 log2_pages:8;
       i64 refcnt:16;
     };
     u64 as_u64;
   };
-#if BIHASH_KVP_CACHE_SIZE > 0
-  u16 cache_lru;
-    BVT (clib_bihash_kv) cache[BIHASH_KVP_CACHE_SIZE];
-#endif
 } BVT (clib_bihash_bucket);
 
-#if BIHASH_KVP_CACHE_SIZE == 0
 STATIC_ASSERT_SIZEOF (BVT (clib_bihash_bucket), sizeof (u64));
-#endif
 
 typedef struct
 {
   BVT (clib_bihash_value) * values;
   BVT (clib_bihash_bucket) * buckets;
-  volatile u32 *writer_lock;
+  volatile u32 *alloc_lock;
 
     BVT (clib_bihash_value) ** working_copies;
   int *working_copy_lengths;
@@ -110,127 +103,38 @@ typedef struct
 
 } BVT (clib_bihash);
 
-
-static inline void
-BV (clib_bihash_update_lru) (BVT (clib_bihash_bucket) * b, u8 slot)
+static inline void BV (clib_bihash_alloc_lock) (BVT (clib_bihash) * h)
 {
-#if BIHASH_KVP_CACHE_SIZE > 1
-  u16 value, tmp, mask;
-  u8 found_lru_pos;
-  u16 save_hi;
-
-  ASSERT (slot < BIHASH_KVP_CACHE_SIZE);
-
-  /* First, find the slot in cache_lru */
-  mask = slot;
-  if (BIHASH_KVP_CACHE_SIZE > 1)
-    mask |= slot << 3;
-  if (BIHASH_KVP_CACHE_SIZE > 2)
-    mask |= slot << 6;
-  if (BIHASH_KVP_CACHE_SIZE > 3)
-    mask |= slot << 9;
-  if (BIHASH_KVP_CACHE_SIZE > 4)
-    mask |= slot << 12;
-
-  value = b->cache_lru;
-  tmp = value ^ mask;
-
-  /* Already the most-recently used? */
-  if ((tmp & 7) == 0)
-    return;
-
-  found_lru_pos = ((tmp & (7 << 3)) == 0) ? 1 : 0;
-  if (BIHASH_KVP_CACHE_SIZE > 2)
-    found_lru_pos = ((tmp & (7 << 6)) == 0) ? 2 : found_lru_pos;
-  if (BIHASH_KVP_CACHE_SIZE > 3)
-    found_lru_pos = ((tmp & (7 << 9)) == 0) ? 3 : found_lru_pos;
-  if (BIHASH_KVP_CACHE_SIZE > 4)
-    found_lru_pos = ((tmp & (7 << 12)) == 0) ? 4 : found_lru_pos;
-
-  ASSERT (found_lru_pos);
-
-  /* create a mask to kill bits in or above slot */
-  mask = 0xFFFF << found_lru_pos;
-  mask <<= found_lru_pos;
-  mask <<= found_lru_pos;
-  mask ^= 0xFFFF;
-  tmp = value & mask;
-
-  /* Save bits above slot */
-  mask ^= 0xFFFF;
-  mask <<= 3;
-  save_hi = value & mask;
-
-  value = save_hi | (tmp << 3) | slot;
-
-  b->cache_lru = value;
-#endif
-}
-
-void
-BV (clib_bihash_update_lru_not_inline) (BVT (clib_bihash_bucket) * b,
-					u8 slot);
-
-static inline u8 BV (clib_bihash_get_lru) (BVT (clib_bihash_bucket) * b)
-{
-#if BIHASH_KVP_CACHE_SIZE > 0
-  return (b->cache_lru >> (3 * (BIHASH_KVP_CACHE_SIZE - 1))) & 7;
-#else
-  return 0;
-#endif
+  while (__atomic_test_and_set (h->alloc_lock, __ATOMIC_ACQUIRE))
+    ;
 }
 
-static inline void BV (clib_bihash_reset_cache) (BVT (clib_bihash_bucket) * b)
+static inline void BV (clib_bihash_alloc_unlock) (BVT (clib_bihash) * h)
 {
-#if BIHASH_KVP_CACHE_SIZE > 0
-  u16 initial_lru_value;
-
-  memset (b->cache, 0xff, sizeof (b->cache));
-
-  /*
-   * We'll want the cache to be loaded from slot 0 -> slot N, so
-   * the initial LRU order is reverse index order.
-   */
-  if (BIHASH_KVP_CACHE_SIZE == 1)
-    initial_lru_value = 0;
-  else if (BIHASH_KVP_CACHE_SIZE == 2)
-    initial_lru_value = (0 << 3) | (1 << 0);
-  else if (BIHASH_KVP_CACHE_SIZE == 3)
-    initial_lru_value = (0 << 6) | (1 << 3) | (2 << 0);
-  else if (BIHASH_KVP_CACHE_SIZE == 4)
-    initial_lru_value = (0 << 9) | (1 << 6) | (2 << 3) | (3 << 0);
-  else if (BIHASH_KVP_CACHE_SIZE == 5)
-    initial_lru_value = (0 << 12) | (1 << 9) | (2 << 6) | (3 << 3) | (4 << 0);
-
-  b->cache_lru = initial_lru_value;
-#endif
+  __atomic_clear (h->alloc_lock, __ATOMIC_RELEASE);
 }
 
-static inline int BV (clib_bihash_lock_bucket) (BVT (clib_bihash_bucket) * b)
+static inline void BV (clib_bihash_lock_bucket) (BVT (clib_bihash_bucket) * b)
 {
-#if BIHASH_KVP_CACHE_SIZE > 0
-  u16 cache_lru_bit;
-  u16 rv;
-
-  cache_lru_bit = 1 << 15;
+  BVT (clib_bihash_bucket) unlocked_bucket, locked_bucket;
 
-  rv = __sync_fetch_and_or (&b->cache_lru, cache_lru_bit);
-  /* Was already locked? */
-  if (rv & (1 << 15))
-    return 0;
-#endif
-  return 1;
+  do
+    {
+      locked_bucket.as_u64 = unlocked_bucket.as_u64 = b->as_u64;
+      unlocked_bucket.lock = 0;
+      locked_bucket.lock = 1;
+    }
+  while (__atomic_compare_exchange_n (&b->as_u64, &unlocked_bucket.as_u64,
+				      locked_bucket.as_u64, 1 /* weak */ ,
+				      __ATOMIC_ACQUIRE,
+				      __ATOMIC_ACQUIRE) == 0);
 }
 
 static inline void BV (clib_bihash_unlock_bucket)
   (BVT (clib_bihash_bucket) * b)
 {
-#if BIHASH_KVP_CACHE_SIZE > 0
-  u16 cache_lru;
-
-  cache_lru = b->cache_lru & ~(1 << 15);
-  b->cache_lru = cache_lru;
-#endif
+  CLIB_MEMORY_BARRIER ();
+  b->lock = 0;
 }
 
 static inline void *BV (clib_bihash_get_value) (BVT (clib_bihash) * h,
@@ -245,7 +149,8 @@ static inline void *BV (clib_bihash_get_value) (BVT (clib_bihash) * h,
 static inline int BV (clib_bihash_bucket_is_empty)
   (BVT (clib_bihash_bucket) * b)
 {
-  return b->as_u64 == 0;
+  /* Note: applied to locked buckets, test offset */
+  return b->offset == 0;
 }
 
 static inline uword BV (clib_bihash_get_offset) (BVT (clib_bihash) * h,
@@ -286,9 +191,6 @@ static inline int BV (clib_bihash_search_inline_with_hash)
   u32 bucket_index;
   BVT (clib_bihash_value) * v;
   BVT (clib_bihash_bucket) * b;
-#if BIHASH_KVP_CACHE_SIZE > 0
-  BVT (clib_bihash_kv) * kvp;
-#endif
   int i, limit;
 
   bucket_index = hash & (h->nbuckets - 1);
@@ -297,23 +199,12 @@ static inline int BV (clib_bihash_search_inline_with_hash)
   if (PREDICT_FALSE (BV (clib_bihash_bucket_is_empty) (b)))
     return -1;
 
-#if BIHASH_KVP_CACHE_SIZE > 0
-  /* Check the cache, if not currently locked */
-  if (PREDICT_TRUE ((b->cache_lru & (1 << 15)) == 0))
+  if (PREDICT_FALSE (b->lock))
     {
-      limit = BIHASH_KVP_CACHE_SIZE;
-      kvp = b->cache;
-      for (i = 0; i < limit; i++)
-	{
-	  if (BV (clib_bihash_key_compare) (kvp[i].key, key_result->key))
-	    {
-	      *key_result = kvp[i];
-	      h->cache_hits++;
-	      return 0;
-	    }
-	}
+      volatile BVT (clib_bihash_bucket) * bv = b;
+      while (bv->lock)
+	;
     }
-#endif
 
   hash >>= h->log2_nbuckets;
 
@@ -330,21 +221,6 @@ static inline int BV (clib_bihash_search_inline_with_hash)
       if (BV (clib_bihash_key_compare) (v->kvp[i].key, key_result->key))
 	{
 	  *key_result = v->kvp[i];
-
-#if BIHASH_KVP_CACHE_SIZE > 0
-	  u8 cache_slot;
-	  /* Try to lock the bucket */
-	  if (BV (clib_bihash_lock_bucket) (b))
-	    {
-	      cache_slot = BV (clib_bihash_get_lru) (b);
-	      b->cache[cache_slot] = v->kvp[i];
-	      BV (clib_bihash_update_lru) (b, cache_slot);
-
-	      /* Unlock the bucket */
-	      BV (clib_bihash_unlock_bucket) (b);
-	      h->cache_misses++;
-	    }
-#endif
 	  return 0;
 	}
     }
@@ -401,9 +277,6 @@ static inline int BV (clib_bihash_search_inline_2_with_hash)
   u32 bucket_index;
   BVT (clib_bihash_value) * v;
   BVT (clib_bihash_bucket) * b;
-#if BIHASH_KVP_CACHE_SIZE > 0
-  BVT (clib_bihash_kv) * kvp;
-#endif
   int i, limit;
 
   ASSERT (valuep);
@@ -414,23 +287,12 @@ static inline int BV (clib_bihash_search_inline_2_with_hash)
   if (PREDICT_FALSE (BV (clib_bihash_bucket_is_empty) (b)))
     return -1;
 
-  /* Check the cache, if currently unlocked */
-#if BIHASH_KVP_CACHE_SIZE > 0
-  if (PREDICT_TRUE ((b->cache_lru & (1 << 15)) == 0))
+  if (PREDICT_FALSE (b->lock))
     {
-      limit = BIHASH_KVP_CACHE_SIZE;
-      kvp = b->cache;
-      for (i = 0; i < limit; i++)
-	{
-	  if (BV (clib_bihash_key_compare) (kvp[i].key, search_key->key))
-	    {
-	      *valuep = kvp[i];
-	      h->cache_hits++;
-	      return 0;
-	    }
-	}
+      volatile BVT (clib_bihash_bucket) * bv = b;
+      while (bv->lock)
+	;
     }
-#endif
 
   hash >>= h->log2_nbuckets;
   v = BV (clib_bihash_get_value) (h, b->offset);
@@ -446,22 +308,6 @@ static inline int BV (clib_bihash_search_inline_2_with_hash)
       if (BV (clib_bihash_key_compare) (v->kvp[i].key, search_key->key))
 	{
 	  *valuep = v->kvp[i];
-
-#if BIHASH_KVP_CACHE_SIZE > 0
-	  u8 cache_slot;
-
-	  /* Try to lock the bucket */
-	  if (BV (clib_bihash_lock_bucket) (b))
-	    {
-	      cache_slot = BV (clib_bihash_get_lru) (b);
-	      b->cache[cache_slot] = v->kvp[i];
-	      BV (clib_bihash_update_lru) (b, cache_slot);
-
-	      /* Reenable the cache */
-	      BV (clib_bihash_unlock_bucket) (b);
-	      h->cache_misses++;
-	    }
-#endif
 	  return 0;
 	}
     }
diff --git a/src/vppinfra/bihash_vec8_8.h b/src/vppinfra/bihash_vec8_8.h
index cc78f1d51d7..2394600ba7c 100644
--- a/src/vppinfra/bihash_vec8_8.h
+++ b/src/vppinfra/bihash_vec8_8.h
@@ -13,12 +13,10 @@
  * limitations under the License.
  */
 #undef BIHASH_TYPE
-#undef BIHASH_KVP_CACHE_SIZE
 #undef BIHASH_KVP_PER_PAGE
 
 #define BIHASH_TYPE _vec8_8
 #define BIHASH_KVP_PER_PAGE 4
-#define BIHASH_KVP_CACHE_SIZE 0
 
 #ifndef __included_bihash_vec8_8_h__
 #define __included_bihash_vec8_8_h__
diff --git a/src/vppinfra/test_bihash_template.c b/src/vppinfra/test_bihash_template.c
index bdcf2cd6a81..80e11511f62 100644
--- a/src/vppinfra/test_bihash_template.c
+++ b/src/vppinfra/test_bihash_template.c
@@ -17,6 +17,7 @@
 #include <vppinfra/error.h>
 #include <sys/resource.h>
 #include <stdio.h>
+#include <pthread.h>
 
 #include <vppinfra/bihash_8_8.h>
 #include <vppinfra/bihash_template.h>
@@ -25,6 +26,9 @@
 
 typedef struct
 {
+  volatile u32 thread_barrier;
+  volatile u32 threads_running;
+  volatile u64 sequence_number;
   u64 seed;
   u32 nbuckets;
   u32 nitems;
@@ -34,11 +38,13 @@ typedef struct
   int careful_delete_tests;
   int verbose;
   int non_random_keys;
+  u32 nthreads;
   uword *key_hash;
   u64 *keys;
   uword hash_memory_size;
     BVT (clib_bihash) hash;
   clib_time_t clib_time;
+  void *global_heap;
 
   unformat_input_t *input;
 
@@ -89,6 +95,102 @@ test_bihash_vec64 (test_main_t * tm)
   return 0;
 }
 
+void *
+test_bihash_thread_fn (void *arg)
+{
+  BVT (clib_bihash) * h;
+  BVT (clib_bihash_kv) kv;
+  test_main_t *tm = &test_main;
+
+  int i, j;
+
+  u32 my_thread_index = (u32) (u64) arg;
+  __os_thread_index = my_thread_index;
+  clib_mem_set_per_cpu_heap (tm->global_heap);
+
+  while (tm->thread_barrier)
+    ;
+
+  h = &tm->hash;
+
+  for (i = 0; i < tm->ncycles; i++)
+    {
+      for (j = 0; j < tm->nitems; j++)
+	{
+	  kv.key = ((u64) my_thread_index << 32) | (u64) j;
+	  kv.value = ((u64) my_thread_index << 32) | (u64) j;
+	  (void) __atomic_add_fetch (&tm->sequence_number, 1,
+				     __ATOMIC_ACQUIRE);
+	  BV (clib_bihash_add_del) (h, &kv, 1 /* is_add */ );
+	}
+      for (j = 0; j < tm->nitems; j++)
+	{
+	  kv.key = ((u64) my_thread_index << 32) | (u64) j;
+	  kv.value = ((u64) my_thread_index << 32) | (u64) j;
+	  (void) __atomic_add_fetch (&tm->sequence_number, 1,
+				     __ATOMIC_ACQUIRE);
+	  BV (clib_bihash_add_del) (h, &kv, 0 /* is_add */ );
+	}
+    }
+
+  (void) __atomic_sub_fetch (&tm->threads_running, 1, __ATOMIC_ACQUIRE);
+  while (1)
+    {
+      struct timespec ts, tsrem;
+      ts.tv_sec = 1;
+      ts.tv_nsec = 0;
+
+      while (nanosleep (&ts, &tsrem) < 0)
+	ts = tsrem;
+    }
+  return (0);			/* not so much */
+}
+
+static clib_error_t *
+test_bihash_threads (test_main_t * tm)
+{
+  int i;
+  pthread_t handle;
+  BVT (clib_bihash) * h;
+  int rv;
+
+  h = &tm->hash;
+
+  BV (clib_bihash_init) (h, "test", tm->nbuckets, tm->hash_memory_size);
+
+  tm->thread_barrier = 1;
+
+  /* Start the worker threads */
+  for (i = 0; i < tm->nthreads; i++)
+    {
+      rv = pthread_create (&handle, NULL, test_bihash_thread_fn,
+			   (void *) (u64) i);
+      if (rv)
+	{
+	  clib_unix_warning ("pthread_create returned %d", rv);
+	}
+    }
+  tm->threads_running = i;
+  tm->sequence_number = 0;
+  CLIB_MEMORY_BARRIER ();
+
+  /* start the workers */
+  tm->thread_barrier = 0;
+
+  while (tm->threads_running)
+    {
+      struct timespec ts, tsrem;
+      ts.tv_sec = 0;
+      ts.tv_nsec = 20 * 1000 * 1000;	/* sleep for 20ms at a time */
+
+      while (nanosleep (&ts, &tsrem) < 0)
+	ts = tsrem;
+    }
+
+  return 0;
+}
+
+
 static clib_error_t *
 test_bihash (test_main_t * tm)
 {
@@ -283,39 +385,6 @@ test_bihash (test_main_t * tm)
 }
 
 clib_error_t *
-test_bihash_cache (test_main_t * tm)
-{
-  u32 lru;
-  BVT (clib_bihash_bucket) _b, *b = &_b;
-
-  BV (clib_bihash_reset_cache) (b);
-
-  fformat (stdout, "Initial LRU config: %U\n", BV (format_bihash_lru), b);
-
-  BV (clib_bihash_update_lru_not_inline) (b, 3);
-
-  fformat (stdout, "use slot 3, LRU config: %U\n", BV (format_bihash_lru), b);
-
-  BV (clib_bihash_update_lru) (b, 1);
-
-  fformat (stdout, "use slot 1 LRU config: %U\n", BV (format_bihash_lru), b);
-
-  lru = BV (clib_bihash_get_lru) (b);
-
-  fformat (stdout, "least-recently-used is %d\n", lru);
-
-  BV (clib_bihash_update_lru) (b, 4);
-
-  fformat (stdout, "use slot 4 LRU config: %U\n", BV (format_bihash_lru), b);
-
-  lru = BV (clib_bihash_get_lru) (b);
-
-  fformat (stdout, "least-recently-used is %d\n", lru);
-
-  return 0;
-}
-
-clib_error_t *
 test_bihash_main (test_main_t * tm)
 {
   unformat_input_t *i = tm->input;
@@ -351,9 +420,8 @@ test_bihash_main (test_main_t * tm)
 	;
       else if (unformat (i, "vec64"))
 	which = 1;
-      else if (unformat (i, "cache"))
+      else if (unformat (i, "threads %u", &tm->nthreads))
 	which = 2;
-
       else if (unformat (i, "verbose"))
 	tm->verbose = 1;
       else
@@ -378,7 +446,7 @@ test_bihash_main (test_main_t * tm)
       break;
 
     case 2:
-      error = test_bihash_cache (tm);
+      error = test_bihash_threads (tm);
       break;
 
     default:
@@ -398,6 +466,8 @@ main (int argc, char *argv[])
 
   clib_mem_init (0, 4095ULL << 20);
 
+  tm->global_heap = clib_mem_get_per_cpu_heap ();
+
   tm->input = &i;
   tm->seed = 0xdeaddabe;