[dpdk-dev] [PATCH] hash: free internal ring when freeing hash

[dpdk-dev] [PATCH] hash: free internal ring when freeing hash

[Pablo de Lara]

Since freeing a ring is now possible, then when freeing
a hash table, its internal ring can be freed as well.
Therefore when a new table, with the same name as a previously
freed table, is created, there is no need to look up
the already allocated ring.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
---

This patch depends on patch "ring: add function to free a ring"
(http://dpdk.org/dev/patchwork/patch/7376/)

 lib/librte_hash/rte_cuckoo_hash.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 7019763..409fc2e 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -180,7 +180,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	struct rte_hash_list *hash_list;
 	struct rte_ring *r = NULL;
 	char hash_name[RTE_HASH_NAMESIZE];
-	void *ptr, *k = NULL;
+	void *k = NULL;
 	void *buckets = NULL;
 	char ring_name[RTE_RING_NAMESIZE];
 	unsigned i;
@@ -288,13 +288,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
 #endif
 
 	snprintf(ring_name, sizeof(ring_name), "HT_%s", params->name);
-	r = rte_ring_lookup(ring_name);
-	if (r != NULL) {
-		/* clear the free ring */
-		while (rte_ring_dequeue(r, &ptr) == 0)
-			rte_pause();
-	} else
-		r = rte_ring_create(ring_name, rte_align32pow2(params->entries + 1),
+	r = rte_ring_create(ring_name, rte_align32pow2(params->entries + 1),
 				params->socket_id, 0);
 	if (r == NULL) {
 		RTE_LOG(ERR, HASH, "memory allocation failed\n");
@@ -363,6 +357,7 @@ rte_hash_free(struct rte_hash *h)
 
 	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 
+	rte_ring_free(h->free_slots);
 	rte_free(h->key_store);
 	rte_free(h->buckets);
 	rte_free(h);
-- 
2.4.3

Re: [dpdk-dev] [PATCH] hash: free internal ring when freeing hash

[Bruce Richardson]

On Fri, Oct 02, 2015 at 05:07:13PM +0100, Pablo de Lara wrote:
> Since freeing a ring is now possible, then when freeing
> a hash table, its internal ring can be freed as well.
> Therefore when a new table, with the same name as a previously
> freed table, is created, there is no need to look up
> the already allocated ring.
> 
> Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>

Acked-by: Bruce Richardson <bruce.richardson@intel.com>

Re: [dpdk-dev] [PATCH] hash: free internal ring when freeing hash

[Thomas Monjalon]

2015-10-27 11:38, Bruce Richardson:
> On Fri, Oct 02, 2015 at 05:07:13PM +0100, Pablo de Lara wrote:
> > Since freeing a ring is now possible, then when freeing
> > a hash table, its internal ring can be freed as well.
> > Therefore when a new table, with the same name as a previously
> > freed table, is created, there is no need to look up
> > the already allocated ring.
> > 
> > Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
> 
> Acked-by: Bruce Richardson <bruce.richardson@intel.com>

Applied, thanks

[dpdk-dev] [PATCH v2] hash: fix scaling by reducing contention

[Pablo de Lara]

From: "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com>

If using multiple cores on a system with hardware transactional
memory support, thread scaling does not work, as there was a single
point in the hash library which is a bottleneck for all threads,
which is the "free_slots" ring, which stores all the indices of
the free slots in the table.

This patch fixes the problem, by creating a local cache per logical core,
which stores locally indices of free slots,
so most times, writer threads will not interfere each other.

Fixes: 48a399119619 ("hash: replace with cuckoo hash implementation")

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
---

Changes in v2:
 - Include patch dependency below

This patch depends on patch "hash: free internal ring when freeing hash"
(http://www.dpdk.org/dev/patchwork/patch/7377/)

 app/test/test_hash_scaling.c         |   1 +
 doc/guides/rel_notes/release_2_2.rst |   5 ++
 lib/librte_hash/rte_cuckoo_hash.c    | 144 ++++++++++++++++++++++++++++++-----
 lib/librte_hash/rte_hash.h           |   3 +
 4 files changed, 133 insertions(+), 20 deletions(-)

diff --git a/app/test/test_hash_scaling.c b/app/test/test_hash_scaling.c
index 39602cb..e7cb7e2 100644
--- a/app/test/test_hash_scaling.c
+++ b/app/test/test_hash_scaling.c
@@ -133,6 +133,7 @@ test_hash_scaling(int locking_mode)
 		.hash_func = rte_hash_crc,
 		.hash_func_init_val = 0,
 		.socket_id = rte_socket_id(),
+		.extra_flag = 1 << RTE_HASH_TRANS_MEM_SUPPORT_FLAG,
 	};
 	struct rte_hash *handle;
 	char name[RTE_HASH_NAMESIZE];
diff --git a/doc/guides/rel_notes/release_2_2.rst b/doc/guides/rel_notes/release_2_2.rst
index 89e4d58..14d2ed9 100644
--- a/doc/guides/rel_notes/release_2_2.rst
+++ b/doc/guides/rel_notes/release_2_2.rst
@@ -79,6 +79,11 @@ Drivers
 
   Fixed issue when releasing null control queue.
 
+* **hash: Fixed thread scaling by reducing contention.**
+
+  Fixed issue in hash library where, using multiple cores with
+  hardware transactional memory support, thread scaling did not work,
+  due to the global ring that is shared by all cores.
 
 Libraries
 ~~~~~~~~~
diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 409fc2e..43788a4 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -96,8 +96,15 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
 
 #define KEY_ALIGNMENT			16
 
+#define LCORE_CACHE_SIZE		8
+
 typedef int (*rte_hash_cmp_eq_t)(const void *key1, const void *key2, size_t key_len);
 
+struct lcore_cache {
+	unsigned len; /**< Cache len */
+	void *objs[LCORE_CACHE_SIZE]; /**< Cache objects */
+} __rte_cache_aligned;
+
 /** A hash table structure. */
 struct rte_hash {
 	char name[RTE_HASH_NAMESIZE];   /**< Name of the hash. */
@@ -117,6 +124,10 @@ struct rte_hash {
 	struct rte_hash_bucket *buckets;	/**< Table with buckets storing all the
 							hash values and key indexes
 							to the key table*/
+	uint8_t hw_trans_mem_support;	/**< Hardware transactional
+							memory support */
+	struct lcore_cache *local_free_slots;
+	/**< Local cache per lcore, storing some indexes of the free slots */
 } __rte_cache_aligned;
 
 /* Structure storing both primary and secondary hashes */
@@ -183,6 +194,8 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	void *k = NULL;
 	void *buckets = NULL;
 	char ring_name[RTE_RING_NAMESIZE];
+	unsigned num_key_slots;
+	unsigned hw_trans_mem_support = 0;
 	unsigned i;
 
 	hash_list = RTE_TAILQ_CAST(rte_hash_tailq.head, rte_hash_list);
@@ -202,6 +215,10 @@ rte_hash_create(const struct rte_hash_parameters *params)
 		return NULL;
 	}
 
+	/* Check extra flags bit to check extra options. */
+	if (params->extra_flag & (1 << RTE_HASH_TRANS_MEM_SUPPORT_FLAG))
+		hw_trans_mem_support = 1;
+
 	snprintf(hash_name, sizeof(hash_name), "HT_%s", params->name);
 
 	/* Guarantee there's no existing */
@@ -238,7 +255,18 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	const uint32_t key_entry_size = sizeof(struct rte_hash_key) + params->key_len;
 
 	/* Store all keys and leave the first entry as a dummy entry for lookup_bulk */
-	const uint64_t key_tbl_size = (uint64_t) key_entry_size * (params->entries + 1);
+	if (hw_trans_mem_support)
+		/*
+		 * Increase number of slots by total number of indices
+		 * that can be stored in the lcore caches
+		 * except for the first cache
+		 */
+		num_key_slots = params->entries + (RTE_MAX_LCORE - 1) *
+					LCORE_CACHE_SIZE + 1;
+	else
+		num_key_slots = params->entries + 1;
+
+	const uint64_t key_tbl_size = (uint64_t) key_entry_size * num_key_slots;
 
 	k = rte_zmalloc_socket(NULL, key_tbl_size,
 			RTE_CACHE_LINE_SIZE, params->socket_id);
@@ -288,13 +316,19 @@ rte_hash_create(const struct rte_hash_parameters *params)
 #endif
 
 	snprintf(ring_name, sizeof(ring_name), "HT_%s", params->name);
-	r = rte_ring_create(ring_name, rte_align32pow2(params->entries + 1),
-				params->socket_id, 0);
+	r = rte_ring_create(ring_name, rte_align32pow2(num_key_slots),
+			params->socket_id, 0);
 	if (r == NULL) {
 		RTE_LOG(ERR, HASH, "memory allocation failed\n");
 		goto err;
 	}
 
+	if (hw_trans_mem_support) {
+		h->local_free_slots = rte_zmalloc_socket(NULL,
+				sizeof(struct lcore_cache) * RTE_MAX_LCORE,
+				RTE_CACHE_LINE_SIZE, params->socket_id);
+	}
+
 	/* Setup hash context */
 	snprintf(h->name, sizeof(h->name), "%s", params->name);
 	h->entries = params->entries;
@@ -307,9 +341,9 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	h->buckets = buckets;
 	h->hash_func = (params->hash_func == NULL) ?
 		DEFAULT_HASH_FUNC : params->hash_func;
-
 	h->key_store = k;
 	h->free_slots = r;
+	h->hw_trans_mem_support = hw_trans_mem_support;
 
 	/* populate the free slots ring. Entry zero is reserved for key misses */
 	for (i = 1; i < params->entries + 1; i++)
@@ -357,6 +391,9 @@ rte_hash_free(struct rte_hash *h)
 
 	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 
+	if (h->hw_trans_mem_support)
+		rte_free(h->local_free_slots);
+
 	rte_ring_free(h->free_slots);
 	rte_free(h->key_store);
 	rte_free(h->buckets);
@@ -402,6 +439,12 @@ rte_hash_reset(struct rte_hash *h)
 	/* Repopulate the free slots ring. Entry zero is reserved for key misses */
 	for (i = 1; i < h->entries + 1; i++)
 		rte_ring_sp_enqueue(h->free_slots, (void *)((uintptr_t) i));
+
+	if (h->hw_trans_mem_support) {
+		/* Reset local caches per lcore */
+		for (i = 0; i < RTE_MAX_LCORE; i++)
+			h->local_free_slots[i].len = 0;
+	}
 }
 
 /* Search for an entry that can be pushed to its alternative location */
@@ -468,6 +511,17 @@ make_space_bucket(const struct rte_hash *h, struct rte_hash_bucket *bkt)
 
 }
 
+static inline void
+enqueue_slot(const struct rte_hash *h, struct lcore_cache *cached_free_slots,
+		void *slot_id)
+{
+	if (h->hw_trans_mem_support) {
+		cached_free_slots->objs[cached_free_slots->len] = slot_id;
+		cached_free_slots->len++;
+	} else
+		rte_ring_sp_enqueue(h->free_slots, slot_id);
+}
+
 static inline int32_t
 __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 						hash_sig_t sig, void *data)
@@ -477,9 +531,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	unsigned i;
 	struct rte_hash_bucket *prim_bkt, *sec_bkt;
 	struct rte_hash_key *new_k, *k, *keys = h->key_store;
-	void *slot_id;
+	void *slot_id = NULL;
 	uint32_t new_idx;
 	int ret;
+	unsigned n_slots;
+	unsigned lcore_id;
+	struct lcore_cache *cached_free_slots = NULL;
 
 	prim_bucket_idx = sig & h->bucket_bitmask;
 	prim_bkt = &h->buckets[prim_bucket_idx];
@@ -491,8 +548,28 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	rte_prefetch0(sec_bkt);
 
 	/* Get a new slot for storing the new key */
-	if (rte_ring_sc_dequeue(h->free_slots, &slot_id) != 0)
-		return -ENOSPC;
+	if (h->hw_trans_mem_support) {
+		lcore_id = rte_lcore_id();
+		cached_free_slots = &h->local_free_slots[lcore_id];
+		/* Try to get a free slot from the local cache */
+		if (cached_free_slots->len == 0) {
+			/* Need to get another burst of free slots from global ring */
+			n_slots = rte_ring_mc_dequeue_burst(h->free_slots,
+					cached_free_slots->objs, LCORE_CACHE_SIZE);
+			if (n_slots == 0)
+				return -ENOSPC;
+
+			cached_free_slots->len += n_slots;
+		}
+
+		/* Get a free slot from the local cache */
+		cached_free_slots->len--;
+		slot_id = cached_free_slots->objs[cached_free_slots->len];
+	} else {
+		if (rte_ring_sc_dequeue(h->free_slots, &slot_id) != 0)
+			return -ENOSPC;
+	}
+
 	new_k = RTE_PTR_ADD(keys, (uintptr_t)slot_id * h->key_entry_size);
 	rte_prefetch0(new_k);
 	new_idx = (uint32_t)((uintptr_t) slot_id);
@@ -500,11 +577,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	/* Check if key is already inserted in primary location */
 	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
 		if (prim_bkt->signatures[i].current == sig &&
-				prim_bkt->signatures[i].alt == alt_hash)  {
+				prim_bkt->signatures[i].alt == alt_hash) {
 			k = (struct rte_hash_key *) ((char *)keys +
 					prim_bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				rte_ring_sp_enqueue(h->free_slots, slot_id);
+				/* Enqueue index of free slot back in the ring. */
+				enqueue_slot(h, cached_free_slots, slot_id);
 				/* Update data */
 				k->pdata = data;
 				/*
@@ -519,11 +597,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	/* Check if key is already inserted in secondary location */
 	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
 		if (sec_bkt->signatures[i].alt == sig &&
-				sec_bkt->signatures[i].current == alt_hash)  {
+				sec_bkt->signatures[i].current == alt_hash) {
 			k = (struct rte_hash_key *) ((char *)keys +
 					sec_bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				rte_ring_sp_enqueue(h->free_slots, slot_id);
+				/* Enqueue index of free slot back in the ring. */
+				enqueue_slot(h, cached_free_slots, slot_id);
 				/* Update data */
 				k->pdata = data;
 				/*
@@ -566,10 +645,9 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	}
 
 	/* Error in addition, store new slot back in the ring and return error */
-	rte_ring_sp_enqueue(h->free_slots,
-		(void *)((uintptr_t) new_idx));
-	return ret;
+	enqueue_slot(h, cached_free_slots, (void *)((uintptr_t) new_idx));
 
+	return ret;
 }
 
 int32_t
@@ -701,6 +779,34 @@ rte_hash_lookup_data(const struct rte_hash *h, const void *key, void **data)
 	return __rte_hash_lookup_with_hash(h, key, rte_hash_hash(h, key), data);
 }
 
+static inline void
+remove_entry(const struct rte_hash *h, struct rte_hash_bucket *bkt, unsigned i)
+{
+	unsigned lcore_id, n_slots;
+	struct lcore_cache *cached_free_slots;
+
+	bkt->signatures[i].sig = NULL_SIGNATURE;
+	if (h->hw_trans_mem_support) {
+		lcore_id = rte_lcore_id();
+		cached_free_slots = &h->local_free_slots[lcore_id];
+		/* Cache full, need to free it. */
+		if (cached_free_slots->len == LCORE_CACHE_SIZE) {
+			/* Need to enqueue the free slots in global ring. */
+			n_slots = rte_ring_mp_enqueue_burst(h->free_slots,
+						cached_free_slots->objs,
+						LCORE_CACHE_SIZE);
+			cached_free_slots->len -= n_slots;
+		}
+		/* Put index of new free slot in cache. */
+		cached_free_slots->objs[cached_free_slots->len] =
+				(void *)((uintptr_t)bkt->key_idx[i]);
+		cached_free_slots->len++;
+	} else {
+		rte_ring_sp_enqueue(h->free_slots,
+				(void *)((uintptr_t)bkt->key_idx[i]));
+	}
+}
+
 static inline int32_t
 __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 						hash_sig_t sig)
@@ -721,9 +827,8 @@ __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 			k = (struct rte_hash_key *) ((char *)keys +
 					bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				bkt->signatures[i].sig = NULL_SIGNATURE;
-				rte_ring_sp_enqueue(h->free_slots,
-						(void *)((uintptr_t)bkt->key_idx[i]));
+				remove_entry(h, bkt, i);
+
 				/*
 				 * Return index where key is stored,
 				 * substracting the first dummy index
@@ -745,9 +850,8 @@ __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 			k = (struct rte_hash_key *) ((char *)keys +
 					bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				bkt->signatures[i].sig = NULL_SIGNATURE;
-				rte_ring_sp_enqueue(h->free_slots,
-						(void *)((uintptr_t)bkt->key_idx[i]));
+				remove_entry(h, bkt, i);
+
 				/*
 				 * Return index where key is stored,
 				 * substracting the first dummy index
diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
index 175c0bb..24d4c60 100644
--- a/lib/librte_hash/rte_hash.h
+++ b/lib/librte_hash/rte_hash.h
@@ -56,6 +56,9 @@ extern "C" {
 #define RTE_HASH_LOOKUP_BULK_MAX		64
 #define RTE_HASH_LOOKUP_MULTI_MAX		RTE_HASH_LOOKUP_BULK_MAX
 
+/** Enable Hardware transactional memory support. */
+#define RTE_HASH_TRANS_MEM_SUPPORT_FLAG		0
+
 /** Signature of key that is stored internally. */
 typedef uint32_t hash_sig_t;
 
-- 
2.4.3

Re: [dpdk-dev] [PATCH v2] hash: fix scaling by reducing contention

[Bruce Richardson]

On Fri, Oct 30, 2015 at 12:22:52PM +0000, Pablo de Lara wrote:
> From: "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com>
> 
> If using multiple cores on a system with hardware transactional
> memory support, thread scaling does not work, as there was a single
> point in the hash library which is a bottleneck for all threads,
> which is the "free_slots" ring, which stores all the indices of
> the free slots in the table.
> 
> This patch fixes the problem, by creating a local cache per logical core,
> which stores locally indices of free slots,
> so most times, writer threads will not interfere each other.
> 
> Fixes: 48a399119619 ("hash: replace with cuckoo hash implementation")
> 
> Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
> ---
> 
> Changes in v2:
>  - Include patch dependency below
> 
> This patch depends on patch "hash: free internal ring when freeing hash"
> (http://www.dpdk.org/dev/patchwork/patch/7377/)
> 
>  app/test/test_hash_scaling.c         |   1 +
>  doc/guides/rel_notes/release_2_2.rst |   5 ++
>  lib/librte_hash/rte_cuckoo_hash.c    | 144 ++++++++++++++++++++++++++++++-----
>  lib/librte_hash/rte_hash.h           |   3 +
>  4 files changed, 133 insertions(+), 20 deletions(-)
> 
> diff --git a/app/test/test_hash_scaling.c b/app/test/test_hash_scaling.c
> index 39602cb..e7cb7e2 100644
> --- a/app/test/test_hash_scaling.c
> +++ b/app/test/test_hash_scaling.c
> @@ -133,6 +133,7 @@ test_hash_scaling(int locking_mode)
>  		.hash_func = rte_hash_crc,
>  		.hash_func_init_val = 0,
>  		.socket_id = rte_socket_id(),
> +		.extra_flag = 1 << RTE_HASH_TRANS_MEM_SUPPORT_FLAG,

This flag could do with having it's name adjusted, because as used here it's
not really a flag, but a shift value.

<snip>
>  
> +static inline void
> +enqueue_slot(const struct rte_hash *h, struct lcore_cache *cached_free_slots,
> +		void *slot_id)
> +{
> +	if (h->hw_trans_mem_support) {
> +		cached_free_slots->objs[cached_free_slots->len] = slot_id;
> +		cached_free_slots->len++;
> +	} else
> +		rte_ring_sp_enqueue(h->free_slots, slot_id);
> +}
> +

This function could do with a comment explaining that it's just used inside the
add function and is not a generic enqueue function as it assumes that we are just
putting back an element we have just dequeued, so there is guaranteed to be space
in the local cache for it. Otherwise the lack of support for a full cache is
a glaring omission. :-)

/Bruce

[dpdk-dev] [PATCH v3] hash: fix scaling by reducing contention

[Pablo de Lara]

From: "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com>

If using multiple cores on a system with hardware transactional
memory support, thread scaling does not work, as there was a single
point in the hash library which is a bottleneck for all threads,
which is the "free_slots" ring, which stores all the indices of
the free slots in the table.

This patch fixes the problem, by creating a local cache per logical core,
which stores locally indices of free slots,
so most times, writer threads will not interfere each other.

Fixes: 48a399119619 ("hash: replace with cuckoo hash implementation")

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
---

Changes in v3:
 - Renamed new extra option in hash parameters
 - Comment and renamed enqueue_slot function for better understanding

Changes in v2:
 - Included patch dependency below

This patch depends on patch "hash: free internal ring when freeing hash"
(http://www.dpdk.org/dev/patchwork/patch/7377/)

 app/test/test_hash_scaling.c         |   1 +
 doc/guides/rel_notes/release_2_2.rst |   5 ++
 lib/librte_hash/rte_cuckoo_hash.c    | 150 ++++++++++++++++++++++++++++++-----
 lib/librte_hash/rte_hash.h           |   3 +
 4 files changed, 139 insertions(+), 20 deletions(-)

diff --git a/app/test/test_hash_scaling.c b/app/test/test_hash_scaling.c
index 39602cb..744e5e3 100644
--- a/app/test/test_hash_scaling.c
+++ b/app/test/test_hash_scaling.c
@@ -133,6 +133,7 @@ test_hash_scaling(int locking_mode)
 		.hash_func = rte_hash_crc,
 		.hash_func_init_val = 0,
 		.socket_id = rte_socket_id(),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT
 	};
 	struct rte_hash *handle;
 	char name[RTE_HASH_NAMESIZE];
diff --git a/doc/guides/rel_notes/release_2_2.rst b/doc/guides/rel_notes/release_2_2.rst
index 89e4d58..14d2ed9 100644
--- a/doc/guides/rel_notes/release_2_2.rst
+++ b/doc/guides/rel_notes/release_2_2.rst
@@ -79,6 +79,11 @@ Drivers
 
   Fixed issue when releasing null control queue.
 
+* **hash: Fixed thread scaling by reducing contention.**
+
+  Fixed issue in hash library where, using multiple cores with
+  hardware transactional memory support, thread scaling did not work,
+  due to the global ring that is shared by all cores.
 
 Libraries
 ~~~~~~~~~
diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 409fc2e..f797aeb 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -96,8 +96,15 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
 
 #define KEY_ALIGNMENT			16
 
+#define LCORE_CACHE_SIZE		8
+
 typedef int (*rte_hash_cmp_eq_t)(const void *key1, const void *key2, size_t key_len);
 
+struct lcore_cache {
+	unsigned len; /**< Cache len */
+	void *objs[LCORE_CACHE_SIZE]; /**< Cache objects */
+} __rte_cache_aligned;
+
 /** A hash table structure. */
 struct rte_hash {
 	char name[RTE_HASH_NAMESIZE];   /**< Name of the hash. */
@@ -117,6 +124,10 @@ struct rte_hash {
 	struct rte_hash_bucket *buckets;	/**< Table with buckets storing all the
 							hash values and key indexes
 							to the key table*/
+	uint8_t hw_trans_mem_support;	/**< Hardware transactional
+							memory support */
+	struct lcore_cache *local_free_slots;
+	/**< Local cache per lcore, storing some indexes of the free slots */
 } __rte_cache_aligned;
 
 /* Structure storing both primary and secondary hashes */
@@ -183,6 +194,8 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	void *k = NULL;
 	void *buckets = NULL;
 	char ring_name[RTE_RING_NAMESIZE];
+	unsigned num_key_slots;
+	unsigned hw_trans_mem_support = 0;
 	unsigned i;
 
 	hash_list = RTE_TAILQ_CAST(rte_hash_tailq.head, rte_hash_list);
@@ -202,6 +215,10 @@ rte_hash_create(const struct rte_hash_parameters *params)
 		return NULL;
 	}
 
+	/* Check extra flags field to check extra options. */
+	if (params->extra_flag & RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT)
+		hw_trans_mem_support = 1;
+
 	snprintf(hash_name, sizeof(hash_name), "HT_%s", params->name);
 
 	/* Guarantee there's no existing */
@@ -238,7 +255,18 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	const uint32_t key_entry_size = sizeof(struct rte_hash_key) + params->key_len;
 
 	/* Store all keys and leave the first entry as a dummy entry for lookup_bulk */
-	const uint64_t key_tbl_size = (uint64_t) key_entry_size * (params->entries + 1);
+	if (hw_trans_mem_support)
+		/*
+		 * Increase number of slots by total number of indices
+		 * that can be stored in the lcore caches
+		 * except for the first cache
+		 */
+		num_key_slots = params->entries + (RTE_MAX_LCORE - 1) *
+					LCORE_CACHE_SIZE + 1;
+	else
+		num_key_slots = params->entries + 1;
+
+	const uint64_t key_tbl_size = (uint64_t) key_entry_size * num_key_slots;
 
 	k = rte_zmalloc_socket(NULL, key_tbl_size,
 			RTE_CACHE_LINE_SIZE, params->socket_id);
@@ -288,13 +316,19 @@ rte_hash_create(const struct rte_hash_parameters *params)
 #endif
 
 	snprintf(ring_name, sizeof(ring_name), "HT_%s", params->name);
-	r = rte_ring_create(ring_name, rte_align32pow2(params->entries + 1),
-				params->socket_id, 0);
+	r = rte_ring_create(ring_name, rte_align32pow2(num_key_slots),
+			params->socket_id, 0);
 	if (r == NULL) {
 		RTE_LOG(ERR, HASH, "memory allocation failed\n");
 		goto err;
 	}
 
+	if (hw_trans_mem_support) {
+		h->local_free_slots = rte_zmalloc_socket(NULL,
+				sizeof(struct lcore_cache) * RTE_MAX_LCORE,
+				RTE_CACHE_LINE_SIZE, params->socket_id);
+	}
+
 	/* Setup hash context */
 	snprintf(h->name, sizeof(h->name), "%s", params->name);
 	h->entries = params->entries;
@@ -307,9 +341,9 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	h->buckets = buckets;
 	h->hash_func = (params->hash_func == NULL) ?
 		DEFAULT_HASH_FUNC : params->hash_func;
-
 	h->key_store = k;
 	h->free_slots = r;
+	h->hw_trans_mem_support = hw_trans_mem_support;
 
 	/* populate the free slots ring. Entry zero is reserved for key misses */
 	for (i = 1; i < params->entries + 1; i++)
@@ -357,6 +391,9 @@ rte_hash_free(struct rte_hash *h)
 
 	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 
+	if (h->hw_trans_mem_support)
+		rte_free(h->local_free_slots);
+
 	rte_ring_free(h->free_slots);
 	rte_free(h->key_store);
 	rte_free(h->buckets);
@@ -402,6 +439,12 @@ rte_hash_reset(struct rte_hash *h)
 	/* Repopulate the free slots ring. Entry zero is reserved for key misses */
 	for (i = 1; i < h->entries + 1; i++)
 		rte_ring_sp_enqueue(h->free_slots, (void *)((uintptr_t) i));
+
+	if (h->hw_trans_mem_support) {
+		/* Reset local caches per lcore */
+		for (i = 0; i < RTE_MAX_LCORE; i++)
+			h->local_free_slots[i].len = 0;
+	}
 }
 
 /* Search for an entry that can be pushed to its alternative location */
@@ -468,6 +511,23 @@ make_space_bucket(const struct rte_hash *h, struct rte_hash_bucket *bkt)
 
 }
 
+/*
+ * Function called to enqueue back an index in the cache/ring,
+ * as slot has not being used and it can be used in the
+ * next addition attempt.
+ */
+static inline void
+enqueue_slot_back(const struct rte_hash *h,
+		struct lcore_cache *cached_free_slots,
+		void *slot_id)
+{
+	if (h->hw_trans_mem_support) {
+		cached_free_slots->objs[cached_free_slots->len] = slot_id;
+		cached_free_slots->len++;
+	} else
+		rte_ring_sp_enqueue(h->free_slots, slot_id);
+}
+
 static inline int32_t
 __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 						hash_sig_t sig, void *data)
@@ -477,9 +537,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	unsigned i;
 	struct rte_hash_bucket *prim_bkt, *sec_bkt;
 	struct rte_hash_key *new_k, *k, *keys = h->key_store;
-	void *slot_id;
+	void *slot_id = NULL;
 	uint32_t new_idx;
 	int ret;
+	unsigned n_slots;
+	unsigned lcore_id;
+	struct lcore_cache *cached_free_slots = NULL;
 
 	prim_bucket_idx = sig & h->bucket_bitmask;
 	prim_bkt = &h->buckets[prim_bucket_idx];
@@ -491,8 +554,28 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	rte_prefetch0(sec_bkt);
 
 	/* Get a new slot for storing the new key */
-	if (rte_ring_sc_dequeue(h->free_slots, &slot_id) != 0)
-		return -ENOSPC;
+	if (h->hw_trans_mem_support) {
+		lcore_id = rte_lcore_id();
+		cached_free_slots = &h->local_free_slots[lcore_id];
+		/* Try to get a free slot from the local cache */
+		if (cached_free_slots->len == 0) {
+			/* Need to get another burst of free slots from global ring */
+			n_slots = rte_ring_mc_dequeue_burst(h->free_slots,
+					cached_free_slots->objs, LCORE_CACHE_SIZE);
+			if (n_slots == 0)
+				return -ENOSPC;
+
+			cached_free_slots->len += n_slots;
+		}
+
+		/* Get a free slot from the local cache */
+		cached_free_slots->len--;
+		slot_id = cached_free_slots->objs[cached_free_slots->len];
+	} else {
+		if (rte_ring_sc_dequeue(h->free_slots, &slot_id) != 0)
+			return -ENOSPC;
+	}
+
 	new_k = RTE_PTR_ADD(keys, (uintptr_t)slot_id * h->key_entry_size);
 	rte_prefetch0(new_k);
 	new_idx = (uint32_t)((uintptr_t) slot_id);
@@ -500,11 +583,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	/* Check if key is already inserted in primary location */
 	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
 		if (prim_bkt->signatures[i].current == sig &&
-				prim_bkt->signatures[i].alt == alt_hash)  {
+				prim_bkt->signatures[i].alt == alt_hash) {
 			k = (struct rte_hash_key *) ((char *)keys +
 					prim_bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				rte_ring_sp_enqueue(h->free_slots, slot_id);
+				/* Enqueue index of free slot back in the ring. */
+				enqueue_slot_back(h, cached_free_slots, slot_id);
 				/* Update data */
 				k->pdata = data;
 				/*
@@ -519,11 +603,12 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	/* Check if key is already inserted in secondary location */
 	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
 		if (sec_bkt->signatures[i].alt == sig &&
-				sec_bkt->signatures[i].current == alt_hash)  {
+				sec_bkt->signatures[i].current == alt_hash) {
 			k = (struct rte_hash_key *) ((char *)keys +
 					sec_bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				rte_ring_sp_enqueue(h->free_slots, slot_id);
+				/* Enqueue index of free slot back in the ring. */
+				enqueue_slot_back(h, cached_free_slots, slot_id);
 				/* Update data */
 				k->pdata = data;
 				/*
@@ -566,10 +651,9 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	}
 
 	/* Error in addition, store new slot back in the ring and return error */
-	rte_ring_sp_enqueue(h->free_slots,
-		(void *)((uintptr_t) new_idx));
-	return ret;
+	enqueue_slot_back(h, cached_free_slots, (void *)((uintptr_t) new_idx));
 
+	return ret;
 }
 
 int32_t
@@ -701,6 +785,34 @@ rte_hash_lookup_data(const struct rte_hash *h, const void *key, void **data)
 	return __rte_hash_lookup_with_hash(h, key, rte_hash_hash(h, key), data);
 }
 
+static inline void
+remove_entry(const struct rte_hash *h, struct rte_hash_bucket *bkt, unsigned i)
+{
+	unsigned lcore_id, n_slots;
+	struct lcore_cache *cached_free_slots;
+
+	bkt->signatures[i].sig = NULL_SIGNATURE;
+	if (h->hw_trans_mem_support) {
+		lcore_id = rte_lcore_id();
+		cached_free_slots = &h->local_free_slots[lcore_id];
+		/* Cache full, need to free it. */
+		if (cached_free_slots->len == LCORE_CACHE_SIZE) {
+			/* Need to enqueue the free slots in global ring. */
+			n_slots = rte_ring_mp_enqueue_burst(h->free_slots,
+						cached_free_slots->objs,
+						LCORE_CACHE_SIZE);
+			cached_free_slots->len -= n_slots;
+		}
+		/* Put index of new free slot in cache. */
+		cached_free_slots->objs[cached_free_slots->len] =
+				(void *)((uintptr_t)bkt->key_idx[i]);
+		cached_free_slots->len++;
+	} else {
+		rte_ring_sp_enqueue(h->free_slots,
+				(void *)((uintptr_t)bkt->key_idx[i]));
+	}
+}
+
 static inline int32_t
 __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 						hash_sig_t sig)
@@ -721,9 +833,8 @@ __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 			k = (struct rte_hash_key *) ((char *)keys +
 					bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				bkt->signatures[i].sig = NULL_SIGNATURE;
-				rte_ring_sp_enqueue(h->free_slots,
-						(void *)((uintptr_t)bkt->key_idx[i]));
+				remove_entry(h, bkt, i);
+
 				/*
 				 * Return index where key is stored,
 				 * substracting the first dummy index
@@ -745,9 +856,8 @@ __rte_hash_del_key_with_hash(const struct rte_hash *h, const void *key,
 			k = (struct rte_hash_key *) ((char *)keys +
 					bkt->key_idx[i] * h->key_entry_size);
 			if (h->rte_hash_cmp_eq(key, k->key, h->key_len) == 0) {
-				bkt->signatures[i].sig = NULL_SIGNATURE;
-				rte_ring_sp_enqueue(h->free_slots,
-						(void *)((uintptr_t)bkt->key_idx[i]));
+				remove_entry(h, bkt, i);
+
 				/*
 				 * Return index where key is stored,
 				 * substracting the first dummy index
diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
index 175c0bb..b678766 100644
--- a/lib/librte_hash/rte_hash.h
+++ b/lib/librte_hash/rte_hash.h
@@ -56,6 +56,9 @@ extern "C" {
 #define RTE_HASH_LOOKUP_BULK_MAX		64
 #define RTE_HASH_LOOKUP_MULTI_MAX		RTE_HASH_LOOKUP_BULK_MAX
 
+/** Enable Hardware transactional memory support. */
+#define RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT	0x01
+
 /** Signature of key that is stored internally. */
 typedef uint32_t hash_sig_t;
 
-- 
2.4.3

Re: [dpdk-dev] [PATCH v3] hash: fix scaling by reducing contention

[Richardson, Bruce]

> -----Original Message-----
> From: De Lara Guarch, Pablo
> Sent: Friday, October 30, 2015 2:37 PM
> To: dev@dpdk.org
> Cc: Richardson, Bruce <bruce.richardson@intel.com>; De Lara Guarch, Pablo
> <pablo.de.lara.guarch@intel.com>
> Subject: [PATCH v3] hash: fix scaling by reducing contention
> 
> From: "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com>
> 
> If using multiple cores on a system with hardware transactional memory
> support, thread scaling does not work, as there was a single point in the
> hash library which is a bottleneck for all threads, which is the
> "free_slots" ring, which stores all the indices of the free slots in the
> table.
> 
> This patch fixes the problem, by creating a local cache per logical core,
> which stores locally indices of free slots, so most times, writer threads
> will not interfere each other.
> 
> Fixes: 48a399119619 ("hash: replace with cuckoo hash implementation")
> 
> Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>

Acked-by: Bruce Richardson <bruce.richardson@intel.com>

Re: [dpdk-dev] [PATCH v3] hash: fix scaling by reducing contention

[Thomas Monjalon]

> > If using multiple cores on a system with hardware transactional memory
> > support, thread scaling does not work, as there was a single point in the
> > hash library which is a bottleneck for all threads, which is the
> > "free_slots" ring, which stores all the indices of the free slots in the
> > table.
> > 
> > This patch fixes the problem, by creating a local cache per logical core,
> > which stores locally indices of free slots, so most times, writer threads
> > will not interfere each other.
> > 
> > Fixes: 48a399119619 ("hash: replace with cuckoo hash implementation")
> > 
> > Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
> 
> Acked-by: Bruce Richardson <bruce.richardson@intel.com>

Applied, thanks