[PATCH v1] hash: add tsx support for cuckoo hash

[PATCH v1] hash: add tsx support for cuckoo hash

[Shen Wei]

Introduced Intel TSX-enabled scalable multi-writer Cuckoo hash
insertion.

This patch introduced scalable multi-writer Cuckoo Hash insertion
based on a split Cuckoo Search and Move operation using Intel
TSX. It can do scalable hash insertion with 22 cores with little
performance loss and negligible TSX abortion rate.

* Added an extra rte_hash flag definition to switch default
  single writer Cuckoo Hash behavior to multiwriter.

* Added a make_space_insert_bfs_mw() function to do split Cuckoo
  search in BFS order.

* Added tsx_cuckoo_move_insert() to do Cuckoo move in Intel TSX
  protected manner.

* Added test_hash_multiwriter() as test case for multi-writer
  Cuckoo Hash.

Signed-off-by: Shen Wei <wei1.shen@intel.com>
Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
---
 app/test/Makefile                 |   1 +
 app/test/test_hash_multiwriter.c  | 272 ++++++++++++++++++++++++++++++++++++++
 lib/librte_hash/rte_cuckoo_hash.c | 228 ++++++++++++++++++++++++++++----
 lib/librte_hash/rte_hash.h        |   3 +
 4 files changed, 480 insertions(+), 24 deletions(-)
 create mode 100644 app/test/test_hash_multiwriter.c

diff --git a/app/test/Makefile b/app/test/Makefile
index a4907d5..697837a 100644
--- a/app/test/Makefile
+++ b/app/test/Makefile
@@ -87,6 +87,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_thash.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_perf.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_functions.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_scaling.c
+SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_multiwriter.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm.c
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm6.c
diff --git a/app/test/test_hash_multiwriter.c b/app/test/test_hash_multiwriter.c
new file mode 100644
index 0000000..bdb9840
--- /dev/null
+++ b/app/test/test_hash_multiwriter.c
@@ -0,0 +1,272 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *	 notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *	 notice, this list of conditions and the following disclaimer in
+ *	 the documentation and/or other materials provided with the
+ *	 distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *	 contributors may be used to endorse or promote products derived
+ *	 from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <inttypes.h>
+#include <locale.h>
+
+#include <rte_cycles.h>
+#include <rte_hash.h>
+#include <rte_hash_crc.h>
+#include <rte_launch.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+/*
+ * Check condition and return an error if true. Assumes that "handle" is the
+ * name of the hash structure pointer to be freed.
+ */
+#define RETURN_IF_ERROR(cond, str, ...) do {                            \
+	if (cond) {                                                     \
+		printf("ERROR line %d: " str "\n", __LINE__,            \
+							##__VA_ARGS__);	\
+		if (handle)                                             \
+			rte_hash_free(handle);                          \
+		return -1;                                              \
+	}                                                               \
+} while (0)
+
+#define RTE_APP_TEST_HASH_MULTIWRITER_FAILED 0
+
+struct {
+	uint32_t *keys;
+	uint32_t *found;
+	uint32_t nb_tsx_insertion;
+	struct rte_hash *h;
+} tbl_multiwriter_test_params;
+
+const uint32_t nb_entries = 16*1024*1024;
+const uint32_t nb_total_tsx_insertion = 15*1024*1024;
+uint32_t rounded_nb_total_tsx_insertion;
+
+static rte_atomic64_t gcycles;
+static rte_atomic64_t ginsertions;
+
+static int
+test_hash_multiwriter_worker(__attribute__((unused)) void *arg)
+{
+	uint64_t i, offset;
+	uint32_t lcore_id = rte_lcore_id();
+	uint64_t begin, cycles;
+
+	offset = (lcore_id - rte_get_master_lcore())
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	printf("Core #%d inserting %d: %'"PRId64" - %'"PRId64"\n",
+	       lcore_id, tbl_multiwriter_test_params.nb_tsx_insertion,
+	       offset, offset + tbl_multiwriter_test_params.nb_tsx_insertion);
+
+	begin = rte_rdtsc_precise();
+
+	for (i = offset;
+	     i < offset + tbl_multiwriter_test_params.nb_tsx_insertion;
+	     i++) {
+		if (rte_hash_add_key(tbl_multiwriter_test_params.h,
+				     tbl_multiwriter_test_params.keys + i) < 0)
+			break;
+	}
+
+	cycles = rte_rdtsc_precise() - begin;
+	rte_atomic64_add(&gcycles, cycles);
+	rte_atomic64_add(&ginsertions, i - offset);
+
+	for (; i < offset + tbl_multiwriter_test_params.nb_tsx_insertion; i++)
+		tbl_multiwriter_test_params.keys[i]
+			= RTE_APP_TEST_HASH_MULTIWRITER_FAILED;
+
+	return 0;
+}
+
+
+static int
+test_hash_multiwriter(void)
+{
+	unsigned int i, rounded_nb_total_tsx_insertion;
+	static unsigned calledCount = 1;
+
+	uint32_t *keys;
+	uint32_t *found;
+
+	struct rte_hash_parameters hash_params = {
+		.entries = nb_entries,
+		.key_len = sizeof(uint32_t),
+		.hash_func = rte_hash_crc,
+		.hash_func_init_val = 0,
+		.socket_id = rte_socket_id(),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT
+		| RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD,
+	};
+
+	struct rte_hash *handle;
+	char name[RTE_HASH_NAMESIZE];
+
+	const void *next_key;
+	void *next_data;
+	uint32_t iter = 0;
+
+	uint32_t duplicated_keys = 0;
+	uint32_t lost_keys = 0;
+
+	snprintf(name, 32, "test%u", calledCount++);
+	hash_params.name = name;
+
+	handle = rte_hash_create(&hash_params);
+	RETURN_IF_ERROR(handle == NULL, "hash creation failed");
+
+	tbl_multiwriter_test_params.h = handle;
+	tbl_multiwriter_test_params.nb_tsx_insertion =
+		nb_total_tsx_insertion / rte_lcore_count();
+
+	rounded_nb_total_tsx_insertion = (nb_total_tsx_insertion /
+		tbl_multiwriter_test_params.nb_tsx_insertion)
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	rte_srand(rte_rdtsc());
+
+	keys = rte_malloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+
+	if (keys == NULL) {
+		printf("RTE_MALLOC failed\n");
+		goto err1;
+	}
+
+	found = rte_zmalloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+	if (found == NULL) {
+		printf("RTE_ZMALLOC failed\n");
+		goto err2;
+	}
+
+	for (i = 0; i < nb_entries; i++)
+		keys[i] = i;
+
+	tbl_multiwriter_test_params.keys = keys;
+	tbl_multiwriter_test_params.found = found;
+
+	rte_atomic64_init(&gcycles);
+	rte_atomic64_clear(&gcycles);
+
+	rte_atomic64_init(&ginsertions);
+	rte_atomic64_clear(&ginsertions);
+
+	/* Fire all threads. */
+	rte_eal_mp_remote_launch(test_hash_multiwriter_worker,
+				 NULL, CALL_MASTER);
+	rte_eal_mp_wait_lcore();
+
+	while (rte_hash_iterate(handle, &next_key, &next_data, &iter) >= 0) {
+		/* Search for the key in the list of keys added .*/
+		i = *(const uint32_t *)next_key;
+		tbl_multiwriter_test_params.found[i]++;
+	}
+
+	for (i = 0; i < rounded_nb_total_tsx_insertion; i++) {
+		if (tbl_multiwriter_test_params.keys[i]
+		    != RTE_APP_TEST_HASH_MULTIWRITER_FAILED) {
+			if (tbl_multiwriter_test_params.found[i] > 1) {
+				duplicated_keys++;
+				break;
+			}
+			if (tbl_multiwriter_test_params.found[i] == 0) {
+				lost_keys++;
+				printf("key %d is lost\n", i);
+				break;
+			}
+		}
+	}
+
+	if (duplicated_keys > 0) {
+		printf("%d key duplicated\n", duplicated_keys);
+		goto err3;
+	}
+
+	if (lost_keys > 0) {
+		printf("%d key lost\n", lost_keys);
+		goto err3;
+	}
+
+	printf("No key corruped during multiwriter insertion.\n");
+
+	unsigned long long int cycles_per_insertion =
+		rte_atomic64_read(&gcycles)/
+		rte_atomic64_read(&ginsertions);
+
+	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
+
+	rte_free(tbl_multiwriter_test_params.found);
+	rte_free(tbl_multiwriter_test_params.keys);
+	rte_hash_free(handle);
+	return 0;
+
+err3:
+	rte_free(tbl_multiwriter_test_params.found);
+err2:
+	rte_free(tbl_multiwriter_test_params.keys);
+err1:
+	rte_hash_free(handle);
+	return -1;
+}
+
+static int
+test_hash_multiwriter_main(void)
+{
+	int r = -1;
+
+	if (rte_lcore_count() == 1) {
+		printf(
+			"More than one lcores are required to do multiwriter test\n");
+		return r;
+	}
+
+	if (!rte_tm_supported()) {
+		printf(
+			"Hardware transactional memory (lock elision) is NOT supported\n");
+		return r;
+	}
+
+	printf("Hardware transactional memory (lock elision) is supported\n");
+
+	setlocale(LC_NUMERIC, "");
+
+	r = test_hash_multiwriter();
+
+	return r;
+}
+
+
+static struct test_command hash_scaling_cmd = {
+	.command = "hash_multiwriter_autotest",
+	.callback = test_hash_multiwriter_main,
+};
+
+REGISTER_TEST_COMMAND(hash_scaling_cmd);
diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 7b7d1f8..5a01f51 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -100,7 +100,9 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
 
 #define KEY_ALIGNMENT			16
 
-#define LCORE_CACHE_SIZE		8
+#define LCORE_CACHE_SIZE		64
+
+#define RTE_HASH_BFS_QUEUE_MAX_LEN  5000
 
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 /*
@@ -190,6 +192,7 @@ struct rte_hash {
 							memory support */
 	struct lcore_cache *local_free_slots;
 	/**< Local cache per lcore, storing some indexes of the free slots */
+	uint8_t multiwrite_add; /**< Multi-write safe hash add behavior */
 } __rte_cache_aligned;
 
 /* Structure storing both primary and secondary hashes */
@@ -213,6 +216,9 @@ struct rte_hash_key {
 	char key[0];
 } __attribute__((aligned(KEY_ALIGNMENT)));
 
+#define RTE_HASH_KEY_FLAG_UNMOVED 0
+#define RTE_HASH_KEY_FLAG_MOVED 1
+
 /** Bucket structure */
 struct rte_hash_bucket {
 	struct rte_hash_signatures signatures[RTE_HASH_BUCKET_ENTRIES];
@@ -372,7 +378,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
 
 /*
  * If x86 architecture is used, select appropriate compare function,
- * which may use x86 instrinsics, otherwise use memcmp
+ * which may use x86 intrinsics, otherwise use memcmp
  */
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 	/* Select function to compare keys */
@@ -431,7 +437,16 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	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 */
+	/* Turn on multi-writer only with explicit flat from user and TM
+	 * support.
+	 */
+	if (params->extra_flag & RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD
+	    && h->hw_trans_mem_support)
+		h->multiwrite_add = 1;
+	else
+		h->multiwrite_add = 0;
+
+	/* Populate free slots ring. Entry zero is reserved for key misses. */
 	for (i = 1; i < params->entries + 1; i++)
 		rte_ring_sp_enqueue(r, (void *)((uintptr_t) i));
 
@@ -599,6 +614,131 @@ make_space_bucket(const struct rte_hash *h, struct rte_hash_bucket *bkt)
 
 }
 
+struct queue_node {
+	struct rte_hash_bucket *bkt;
+
+	struct queue_node *prev;
+	int prev_slot;
+};
+
+/* Shift buckets along cuckoo_path and fill the path head with new entry */
+static inline int
+tsx_cuckoo_move_insert(const struct rte_hash *h, struct queue_node *leaf,
+		       uint32_t leaf_slot, hash_sig_t sig,
+		       hash_sig_t alt_hash, uint32_t new_idx, uint8_t flag)
+{
+	#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4
+	int32_t try = 0;
+	uint32_t prev_alt_bkt_idx;
+	unsigned status;
+
+	struct queue_node *prev_node, *curr_node = leaf;
+	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
+	uint32_t prev_slot, curr_slot = leaf_slot;
+
+	int cuckoo_path_len = 0;
+
+	while (try < 10) {
+		status = rte_xbegin();
+		if (likely(status == RTE_XBEGIN_STARTED)) {
+			while (likely(curr_node->prev != NULL)) {
+				prev_node = curr_node->prev;
+				prev_bkt = prev_node->bkt;
+				prev_slot = curr_node->prev_slot;
+
+				prev_alt_bkt_idx
+					= prev_bkt->signatures[prev_slot].alt
+					& h->bucket_bitmask;
+
+				if (unlikely(&h->buckets[prev_alt_bkt_idx]
+					     != curr_bkt)) {
+					rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
+				}
+
+				curr_bkt->signatures[curr_slot]
+					= prev_bkt->signatures[prev_slot];
+				curr_bkt->key_idx[curr_slot]
+					= prev_bkt->key_idx[prev_slot];
+				curr_bkt->flag[curr_slot]
+					= RTE_HASH_KEY_FLAG_MOVED;
+
+				curr_slot = prev_slot;
+				curr_node = prev_node;
+				curr_bkt = curr_node->bkt;
+
+				cuckoo_path_len++;
+			}
+
+			curr_bkt->signatures[curr_slot].current = sig;
+			curr_bkt->signatures[curr_slot].alt = alt_hash;
+			curr_bkt->key_idx[curr_slot] = new_idx;
+			curr_bkt->flag[curr_slot] = flag;
+
+			rte_xend();
+
+			return 0;
+		}
+
+		/* If we abort we give up this cuckoo path. */
+		try++;
+		rte_pause();
+		continue;
+	}
+
+	return -1;
+}
+
+/* Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
+ * Cuckoo
+ */
+static inline int
+make_space_insert_bfs_mw(const struct rte_hash *h, struct rte_hash_bucket *bkt,
+			 hash_sig_t sig, hash_sig_t alt_hash,
+			 uint32_t new_idx, uint8_t flag)
+{
+	int i;
+	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
+	struct queue_node *tail, *head;
+	struct rte_hash_bucket *curr_bkt, *alt_bkt;
+
+	tail = queue;
+	head = queue + 1;
+	tail->bkt = bkt;
+	tail->prev = NULL;
+	tail->prev_slot = -1;
+
+	/* Cuckoo Search */
+	while (likely(tail != head && head <
+		      queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
+
+		curr_bkt = tail->bkt;
+		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
+				if (likely(tsx_cuckoo_move_insert(h, tail, i,
+					sig, alt_hash, new_idx, flag) == 0))
+					return 0;
+			}
+
+			/* Make sure current key's not already in its
+			 * secondary bucket.
+			 */
+			if (curr_bkt->flag[i] == RTE_HASH_KEY_FLAG_UNMOVED) {
+				/* Enqueue new node and keep prev node info */
+				alt_bkt =
+					&(h->buckets[curr_bkt->signatures[i].alt
+						     & h->bucket_bitmask]);
+				head->bkt = alt_bkt;
+				head->prev = tail;
+				head->prev_slot = i;
+				head++;
+			}
+		}
+		tail++;
+	}
+
+	return -ENOSPC;
+}
+
 /*
  * Function called to enqueue back an index in the cache/ring,
  * as slot has not being used and it can be used in the
@@ -712,30 +852,70 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	rte_memcpy(new_k->key, key, h->key_len);
 	new_k->pdata = data;
 
-	/* Insert new entry is there is room in the primary bucket */
-	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
-		/* Check if slot is available */
-		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
-			prim_bkt->signatures[i].current = sig;
-			prim_bkt->signatures[i].alt = alt_hash;
-			prim_bkt->key_idx[i] = new_idx;
-			return new_idx - 1;
+	unsigned status;
+	int32_t try = 0;
+
+	while (try < 10) {
+		status = rte_xbegin();
+		if (likely(status == RTE_XBEGIN_STARTED)) {
+			/* Insert new entry is there is room in the primary
+			 * bucket.
+			 */
+			for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+				/* Check if slot is available */
+				if (likely(prim_bkt->signatures[i].sig
+					   == NULL_SIGNATURE)) {
+					prim_bkt->signatures[i].current = sig;
+					prim_bkt->signatures[i].alt = alt_hash;
+					prim_bkt->key_idx[i] = new_idx;
+					prim_bkt->flag[i] =
+						RTE_HASH_KEY_FLAG_UNMOVED;
+					break;
+				}
+			}
+			rte_xend();
+
+			if (i != RTE_HASH_BUCKET_ENTRIES)
+				return new_idx - 1;
+
+			break;
+
+		} else {
+			/* If we abort we give up this cuckoo path. */
+			try++;
+			rte_pause();
+			continue;
 		}
 	}
 
 	/* Primary bucket is full, so we need to make space for new entry */
-	ret = make_space_bucket(h, prim_bkt);
-	/*
-	 * After recursive function.
-	 * Insert the new entry in the position of the pushed entry
-	 * if successful or return error and
-	 * store the new slot back in the ring
-	 */
-	if (ret >= 0) {
-		prim_bkt->signatures[ret].current = sig;
-		prim_bkt->signatures[ret].alt = alt_hash;
-		prim_bkt->key_idx[ret] = new_idx;
-		return new_idx - 1;
+	if (h->multiwrite_add) {
+		ret = make_space_insert_bfs_mw(h, prim_bkt, sig,
+			alt_hash, new_idx, RTE_HASH_KEY_FLAG_UNMOVED);
+
+		if (ret >= 0)
+			return new_idx - 1;
+
+		ret = make_space_insert_bfs_mw(h, sec_bkt, sig,
+			alt_hash, new_idx, RTE_HASH_KEY_FLAG_MOVED);
+
+		if (ret >= 0)
+			return new_idx - 1;
+
+	} else {
+		/*
+		 * After recursive function.
+		 * Insert the new entry in the position of the pushed entry
+		 * if successful or return error and
+		 * store the new slot back in the ring
+		 */
+		ret = make_space_bucket(h, prim_bkt);
+		if (ret >= 0) {
+			prim_bkt->signatures[ret].current = sig;
+			prim_bkt->signatures[ret].alt = alt_hash;
+			prim_bkt->key_idx[ret] = new_idx;
+			return new_idx - 1;
+		}
 	}
 
 	/* Error in addition, store new slot back in the ring and return error */
diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
index 724315a..c9612fb 100644
--- a/lib/librte_hash/rte_hash.h
+++ b/lib/librte_hash/rte_hash.h
@@ -60,6 +60,9 @@ extern "C" {
 /** Enable Hardware transactional memory support. */
 #define RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT	0x01
 
+/** Default behavior of insertion, single writer/multi writer */
+#define RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x02
+
 /** Signature of key that is stored internally. */
 typedef uint32_t hash_sig_t;
 
-- 
2.5.0

Re: [PATCH v1] hash: add tsx support for cuckoo hash

[Stephen Hemminger]

On Fri,  6 May 2016 21:05:02 +0100
Shen Wei <wei1.shen@intel.com> wrote:

> --- a/lib/librte_hash/rte_cuckoo_hash.c
> +++ b/lib/librte_hash/rte_cuckoo_hash.c
> @@ -1,7 +1,7 @@
>  /*-
>   *   BSD LICENSE
>   *
> - *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
> + *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
>   *   All rights reserved.
>   *
>   *   Redistribution and use in source and binary forms, with or without
> @@ -100,7 +100,9 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
>  
>  #define KEY_ALIGNMENT			16
>  
> -#define LCORE_CACHE_SIZE		8
> +#define LCORE_CACHE_SIZE		64
> +
> +#define RTE_HASH_BFS_QUEUE_MAX_LEN  5000
>  
>  #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
>  /*
> @@ -190,6 +192,7 @@ struct rte_hash {
>  							memory support */
>  	struct lcore_cache *local_free_slots;
>  	/**< Local cache per lcore, storing some indexes of the free slots */
> +	uint8_t multiwrite_add; /**< Multi-write safe hash add behavior */
>  } __rte_cache_aligned;
>  

I like the idea of using TSX to allow multi-writer safety, but there are
several problems with this patch.

1) It changes ABI, so it breaks old programs
2) What about older processors, need to detect and handle them at runtime.
3) Why can't this just be the default behavior with correct
   fallback to locking on older processors.

Actually lock ellision in DPDK is an interesting topic in general that
needs to be addressed.

Re: [PATCH v1] hash: add tsx support for cuckoo hash

[Shen, Wei1]

Hi Stephen,

Greetings. Thanks for your great feedback. Let’s me address your concern here.

1) It changes ABI, so it breaks old programs
The patch uses the extra_flag field in the rte_hash_parameters struct to set the default insertion behavior. Today there is only one bit used by this flag (RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT 0x1) and we used the next unused bit (RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x2) in this patch. So ABI are maintained.

2) What about older processors, need to detect and handle them at runtime.
Correct. This patch is based on the previous Transactional Memory patch. Since these previous patches already assume the user to check the presence of TSX so we build on top this assumption. But I personally agree with you that handling TSX check should be made easier.
http://dpdk.org/ml/archives/dev/2015-June/018571.html
http://dpdk.org/ml/archives/dev/2015-June/018566.html 

3) Why can't this just be the default behavior with correct fallback to locking on older processors.
This is an excellent point. We discussed this before. Our thought at that time is, since TSX insertion is a bit slower than without anything (TSX or other locks), it would benefit apps that is designed to have a single writer to the hash table (for instance in some master-slave model). We might need more feedback from user about whether making it default is more desirable if most the app is designed with multi-writer manner.

Thanks,


-- 
Best,

Wei Shen.






On 5/6/16, 9:56 PM, "Stephen Hemminger" <stephen@networkplumber.org> wrote:

>On Fri,  6 May 2016 21:05:02 +0100
>Shen Wei <wei1.shen@intel.com> wrote:
>
>> --- a/lib/librte_hash/rte_cuckoo_hash.c
>> +++ b/lib/librte_hash/rte_cuckoo_hash.c
>> @@ -1,7 +1,7 @@
>>  /*-
>>   *   BSD LICENSE
>>   *
>> - *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
>> + *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
>>   *   All rights reserved.
>>   *
>>   *   Redistribution and use in source and binary forms, with or without
>> @@ -100,7 +100,9 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
>>  
>>  #define KEY_ALIGNMENT			16
>>  
>> -#define LCORE_CACHE_SIZE		8
>> +#define LCORE_CACHE_SIZE		64
>> +
>> +#define RTE_HASH_BFS_QUEUEs_MAX_LEN  5000
>>  
>>  #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
>>  /*
>> @@ -190,6 +192,7 @@ struct rte_hash {
>>  							memory support */
>>  	struct lcore_cache *local_free_slots;
>>  	/**< Local cache per lcore, storing some indexes of the free slots */
>> +	uint8_t multiwrite_add; /**< Multi-write safe hash add behavior */
>>  } __rte_cache_aligned;
>>  
>
>I like the idea of using TSX to allow multi-writer safety, but there are
>several problems with this patch.
>
>1) It changes ABI, so it breaks old programs
>2) What about older processors, need to detect and handle them at runtime.
>3) Why can't this just be the default behavior with correct
>   fallback to locking on older processors.
>
>Actually lock ellision in DPDK is an interesting topic in general that
>needs to be addressed.

Re: [PATCH v1] hash: add tsx support for cuckoo hash

[De Lara Guarch, Pablo]

> -----Original Message-----
> From: Shen, Wei1
> Sent: Monday, May 09, 2016 5:52 PM
> To: Stephen Hemminger
> Cc: dev@dpdk.org; De Lara Guarch, Pablo; Maciocco, Christian; Gobriel,
> Sameh
> Subject: Re: [dpdk-dev] [PATCH v1] hash: add tsx support for cuckoo hash
> 
> Hi Stephen,
> 
> Greetings. Thanks for your great feedback. Let’s me address your concern
> here.
> 
> 1) It changes ABI, so it breaks old programs
> The patch uses the extra_flag field in the rte_hash_parameters struct to set
> the default insertion behavior. Today there is only one bit used by this flag
> (RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT 0x1) and we used the
> next unused bit (RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x2) in this
> patch. So ABI are maintained.

Agree on this. Also, if the problem is on the rte_hash (because of the change of
the cache size or the addition of a new field at the end), that should not be a problem,
as far as I understand, since it is modifying an internal structure
(rte_hash was made internal back in v2.1).

> 
> 2) What about older processors, need to detect and handle them at runtime.
> Correct. This patch is based on the previous Transactional Memory patch.
> Since these previous patches already assume the user to check the presence
> of TSX so we build on top this assumption. But I personally agree with you
> that handling TSX check should be made easier.
> http://dpdk.org/ml/archives/dev/2015-June/018571.html
> http://dpdk.org/ml/archives/dev/2015-June/018566.html
> 
> 3) Why can't this just be the default behavior with correct fallback to locking
> on older processors.
> This is an excellent point. We discussed this before. Our thought at that time
> is, since TSX insertion is a bit slower than without anything (TSX or other
> locks), it would benefit apps that is designed to have a single writer to the
> hash table (for instance in some master-slave model). We might need more
> feedback from user about whether making it default is more desirable if
> most the app is designed with multi-writer manner.
> 
> Thanks,
> 
> 
> --
> Best,
> 
> Wei Shen.
> 
> 
> 
> 
> 
> 
> On 5/6/16, 9:56 PM, "Stephen Hemminger"
> <stephen@networkplumber.org> wrote:
> 
> >On Fri,  6 May 2016 21:05:02 +0100
> >Shen Wei <wei1.shen@intel.com> wrote:
> >
> >> --- a/lib/librte_hash/rte_cuckoo_hash.c
> >> +++ b/lib/librte_hash/rte_cuckoo_hash.c
> >> @@ -1,7 +1,7 @@
> >>  /*-
> >>   *   BSD LICENSE
> >>   *
> >> - *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
> >> + *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
> >>   *   All rights reserved.
> >>   *
> >>   *   Redistribution and use in source and binary forms, with or without
> >> @@ -100,7 +100,9 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
> >>
> >>  #define KEY_ALIGNMENT			16
> >>
> >> -#define LCORE_CACHE_SIZE		8
> >> +#define LCORE_CACHE_SIZE		64
> >> +
> >> +#define RTE_HASH_BFS_QUEUEs_MAX_LEN  5000
> >>
> >>  #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
> >>  /*
> >> @@ -190,6 +192,7 @@ struct rte_hash {
> >>  							memory support */
> >>  	struct lcore_cache *local_free_slots;
> >>  	/**< Local cache per lcore, storing some indexes of the free slots */
> >> +	uint8_t multiwrite_add; /**< Multi-write safe hash add behavior */
> >>  } __rte_cache_aligned;
> >>
> >
> >I like the idea of using TSX to allow multi-writer safety, but there are
> >several problems with this patch.
> >
> >1) It changes ABI, so it breaks old programs
> >2) What about older processors, need to detect and handle them at
> runtime.
> >3) Why can't this just be the default behavior with correct
> >   fallback to locking on older processors.
> >
> >Actually lock ellision in DPDK is an interesting topic in general that
> >needs to be addressed.

Re: [PATCH v1] hash: add tsx support for cuckoo hash

[De Lara Guarch, Pablo]

Hi Wei,

> -----Original Message-----
> From: Shen, Wei1
> Sent: Friday, May 06, 2016 9:05 PM
> To: dev@dpdk.org
> Cc: De Lara Guarch, Pablo; Maciocco, Christian; Shen, Wei1; Gobriel, Sameh
> Subject: [PATCH v1] hash: add tsx support for cuckoo hash
> 
> Introduced Intel TSX-enabled scalable multi-writer Cuckoo hash
> insertion.
> 
> This patch introduced scalable multi-writer Cuckoo Hash insertion
> based on a split Cuckoo Search and Move operation using Intel
> TSX. It can do scalable hash insertion with 22 cores with little
> performance loss and negligible TSX abortion rate.
> 
> * Added an extra rte_hash flag definition to switch default
>   single writer Cuckoo Hash behavior to multiwriter.
> 
> * Added a make_space_insert_bfs_mw() function to do split Cuckoo
>   search in BFS order.
> 
> * Added tsx_cuckoo_move_insert() to do Cuckoo move in Intel TSX
>   protected manner.
> 
> * Added test_hash_multiwriter() as test case for multi-writer
>   Cuckoo Hash.
> 
> Signed-off-by: Shen Wei <wei1.shen@intel.com>
> Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
> ---
>  app/test/Makefile                 |   1 +
>  app/test/test_hash_multiwriter.c  | 272
> ++++++++++++++++++++++++++++++++++++++
>  lib/librte_hash/rte_cuckoo_hash.c | 228 ++++++++++++++++++++++++++++--
> --
>  lib/librte_hash/rte_hash.h        |   3 +
>  4 files changed, 480 insertions(+), 24 deletions(-)
>  create mode 100644 app/test/test_hash_multiwriter.c
> 

[...]

> diff --git a/app/test/test_hash_multiwriter.c
> b/app/test/test_hash_multiwriter.c
> new file mode 100644
> index 0000000..bdb9840
> --- /dev/null
> +++ b/app/test/test_hash_multiwriter.c
> @@ -0,0 +1,272 @@

[...]

> +
> +	if (duplicated_keys > 0) {
> +		printf("%d key duplicated\n", duplicated_keys);
> +		goto err3;
> +	}
> +
> +	if (lost_keys > 0) {
> +		printf("%d key lost\n", lost_keys);
> +		goto err3;
> +	}
> +
> +	printf("No key corruped during multiwriter insertion.\n");

Typo: corrupted

> +
> +	unsigned long long int cycles_per_insertion =
> +		rte_atomic64_read(&gcycles)/
> +		rte_atomic64_read(&ginsertions);
> +
> +	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
> +
> +	rte_free(tbl_multiwriter_test_params.found);
> +	rte_free(tbl_multiwriter_test_params.keys);
> +	rte_hash_free(handle);
> +	return 0;
> +
> +err3:
> +	rte_free(tbl_multiwriter_test_params.found);
> +err2:
> +	rte_free(tbl_multiwriter_test_params.keys);
> +err1:
> +	rte_hash_free(handle);
> +	return -1;
> +}
> +
> +static int
> +test_hash_multiwriter_main(void)
> +{
> +	int r = -1;
> +
> +	if (rte_lcore_count() == 1) {
> +		printf(
> +			"More than one lcores are required to do multiwriter
> test\n");

Typo: More than one lcore is required to do multiwriter test.

> +		return r;
> +	}
> +
> +	if (!rte_tm_supported()) {
> +		printf(
> +			"Hardware transactional memory (lock elision) is NOT
> supported\n");
> +		return r;
> +	}
> +
> +	printf("Hardware transactional memory (lock elision) is
> supported\n");
> +
> +	setlocale(LC_NUMERIC, "");
> +
> +	r = test_hash_multiwriter();
> +
> +	return r;
> +}
> +
> +
> +static struct test_command hash_scaling_cmd = {
> +	.command = "hash_multiwriter_autotest",
> +	.callback = test_hash_multiwriter_main,
> +};
> +
> +REGISTER_TEST_COMMAND(hash_scaling_cmd);
> diff --git a/lib/librte_hash/rte_cuckoo_hash.c
> b/lib/librte_hash/rte_cuckoo_hash.c
> index 7b7d1f8..5a01f51 100644
> --- a/lib/librte_hash/rte_cuckoo_hash.c
> +++ b/lib/librte_hash/rte_cuckoo_hash.c

[...]

> +/* Shift buckets along cuckoo_path and fill the path head with new entry */
> +static inline int
> +tsx_cuckoo_move_insert(const struct rte_hash *h, struct queue_node *leaf,
> +		       uint32_t leaf_slot, hash_sig_t sig,
> +		       hash_sig_t alt_hash, uint32_t new_idx, uint8_t flag)
> +{
> +	#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4

Could you move this up and indent it to the left?

> +	int32_t try = 0;

This variable (try) should not have any negative number, so you can change it to unsigned.

> +	uint32_t prev_alt_bkt_idx;
> +	unsigned status;
> +
> +	struct queue_node *prev_node, *curr_node = leaf;
> +	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
> +	uint32_t prev_slot, curr_slot = leaf_slot;
> +
> +	int cuckoo_path_len = 0;

This variable is not being used. It gets incremented below,
but is not used in anything useful, as far as I can see.
If you are not using it, then remove it.

> +
> +	while (try < 10) {

Magic number. Define a macro with this number instead.

> +		status = rte_xbegin();
> +		if (likely(status == RTE_XBEGIN_STARTED)) {
> +			while (likely(curr_node->prev != NULL)) {
> +				prev_node = curr_node->prev;
> +				prev_bkt = prev_node->bkt;
> +				prev_slot = curr_node->prev_slot;
> +
> +				prev_alt_bkt_idx
> +					= prev_bkt->signatures[prev_slot].alt
> +					& h->bucket_bitmask;
> +
> +				if (unlikely(&h->buckets[prev_alt_bkt_idx]
> +					     != curr_bkt)) {
> +
> 	rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
> +				}
> +
> +				curr_bkt->signatures[curr_slot]
> +					= prev_bkt->signatures[prev_slot];
> +				curr_bkt->key_idx[curr_slot]
> +					= prev_bkt->key_idx[prev_slot];
> +				curr_bkt->flag[curr_slot]
> +					= RTE_HASH_KEY_FLAG_MOVED;
> +
> +				curr_slot = prev_slot;
> +				curr_node = prev_node;
> +				curr_bkt = curr_node->bkt;
> +
> +				cuckoo_path_len++;
> +			}
> +
> +			curr_bkt->signatures[curr_slot].current = sig;
> +			curr_bkt->signatures[curr_slot].alt = alt_hash;
> +			curr_bkt->key_idx[curr_slot] = new_idx;
> +			curr_bkt->flag[curr_slot] = flag;
> +
> +			rte_xend();
> +
> +			return 0;
> +		}
> +
> +		/* If we abort we give up this cuckoo path. */
> +		try++;
> +		rte_pause();
> +		continue;

Is this continue necessary? It is at the end of a while loop, so it does not look it is.

> +	}
> +
> +	return -1;
> +}
> +
> +/* Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
> + * Cuckoo
> + */
> +static inline int
> +make_space_insert_bfs_mw(const struct rte_hash *h, struct
> rte_hash_bucket *bkt,
> +			 hash_sig_t sig, hash_sig_t alt_hash,
> +			 uint32_t new_idx, uint8_t flag)
> +{
> +	int i;

Unsigned i;

> +	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
> +	struct queue_node *tail, *head;
> +	struct rte_hash_bucket *curr_bkt, *alt_bkt;
> +
> +	tail = queue;
> +	head = queue + 1;
> +	tail->bkt = bkt;
> +	tail->prev = NULL;
> +	tail->prev_slot = -1;
> +
> +	/* Cuckoo Search */
> +	while (likely(tail != head && head <
> +		      queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
> +
> +		curr_bkt = tail->bkt;
> +		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
> +				if (likely(tsx_cuckoo_move_insert(h, tail, i,
> +					sig, alt_hash, new_idx, flag) == 0))

Add extra indentation here, to distinguish between the conditional
and the body (extra indentation to the right in "sig, alt_hash...").

> +					return 0;
> +			}
> +
> +			/* Make sure current key's not already in its
> +			 * secondary bucket.
> +			 */
> +			if (curr_bkt->flag[i] ==
> RTE_HASH_KEY_FLAG_UNMOVED) {
> +				/* Enqueue new node and keep prev node
> info */
> +				alt_bkt =
> +					&(h->buckets[curr_bkt-
> >signatures[i].alt
> +						     & h->bucket_bitmask]);
> +				head->bkt = alt_bkt;
> +				head->prev = tail;
> +				head->prev_slot = i;
> +				head++;
> +			}
> +		}
> +		tail++;
> +	}
> +
> +	return -ENOSPC;
> +}
> +
>  /*
>   * Function called to enqueue back an index in the cache/ring,
>   * as slot has not being used and it can be used in the
> @@ -712,30 +852,70 @@ __rte_hash_add_key_with_hash(const struct
> rte_hash *h, const void *key,
>  	rte_memcpy(new_k->key, key, h->key_len);
>  	new_k->pdata = data;
> 
> -	/* Insert new entry is there is room in the primary bucket */
> -	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> -		/* Check if slot is available */
> -		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
> -			prim_bkt->signatures[i].current = sig;
> -			prim_bkt->signatures[i].alt = alt_hash;
> -			prim_bkt->key_idx[i] = new_idx;
> -			return new_idx - 1;
> +	unsigned status;
> +	int32_t try = 0;
> +
> +	while (try < 10) {


Same comments about "unsigned try" and magic numbers.

> +		status = rte_xbegin();

The following code should only be executed if RTM is supported,
otherwise, there will be an illegal instruction.

> +		if (likely(status == RTE_XBEGIN_STARTED)) {
> +			/* Insert new entry is there is room in the primary
> +			 * bucket.

Typo: Insert new entry if there is room...

> +			 */
> +			for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +				/* Check if slot is available */
> +				if (likely(prim_bkt->signatures[i].sig
> +					   == NULL_SIGNATURE)) {

Extra indentation to the right, as said above.

> +					prim_bkt->signatures[i].current = sig;
> +					prim_bkt->signatures[i].alt =
> alt_hash;
> +					prim_bkt->key_idx[i] = new_idx;
> +					prim_bkt->flag[i] =
> +
> 	RTE_HASH_KEY_FLAG_UNMOVED;
> +					break;
> +				}
> +			}
> +			rte_xend();
> +
> +			if (i != RTE_HASH_BUCKET_ENTRIES)
> +				return new_idx - 1;
> +
> +			break;
> +
> +		} else {
> +			/* If we abort we give up this cuckoo path. */
> +			try++;
> +			rte_pause();
> +			continue;

Same as above, unnecessary continue?

>  		}
>  	}
> 
>  	/* Primary bucket is full, so we need to make space for new entry */

[PATCH v2] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[Wei Shen]

This patch introduced scalable multi-writer Cuckoo Hash insertion
based on a split Cuckoo Search and Move operation using Intel
TSX. It can do scalable hash insertion with 22 cores with little
performance loss and negligible TSX abortion rate.

* Added an extra rte_hash flag definition to switch default
  single writer Cuckoo Hash behavior to multiwriter.

* Added a make_space_insert_bfs_mw() function to do split Cuckoo
  search in BFS order.

* Added tsx_cuckoo_move_insert() to do Cuckoo move in Intel TSX
  protected manner.

* Added test_hash_multiwriter() as test case for multi-writer
  Cuckoo Hash.

Signed-off-by: Shen Wei <wei1.shen@intel.com>
Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
---
 app/test/Makefile                      |   1 +
 app/test/test_hash_multiwriter.c       | 272 +++++++++++++++++++++++++++++++++
 doc/guides/rel_notes/release_16_07.rst |  12 ++
 lib/librte_hash/rte_cuckoo_hash.c      | 231 +++++++++++++++++++++++++---
 lib/librte_hash/rte_hash.h             |   3 +
 5 files changed, 494 insertions(+), 25 deletions(-)
 create mode 100644 app/test/test_hash_multiwriter.c

diff --git a/app/test/Makefile b/app/test/Makefile
index 053f3a2..5476300 100644
--- a/app/test/Makefile
+++ b/app/test/Makefile
@@ -120,6 +120,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_thash.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_perf.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_functions.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_scaling.c
+SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_multiwriter.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm.c
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm_perf.c
diff --git a/app/test/test_hash_multiwriter.c b/app/test/test_hash_multiwriter.c
new file mode 100644
index 0000000..54a0d2c
--- /dev/null
+++ b/app/test/test_hash_multiwriter.c
@@ -0,0 +1,272 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *	 notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *	 notice, this list of conditions and the following disclaimer in
+ *	 the documentation and/or other materials provided with the
+ *	 distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *	 contributors may be used to endorse or promote products derived
+ *	 from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <inttypes.h>
+#include <locale.h>
+
+#include <rte_cycles.h>
+#include <rte_hash.h>
+#include <rte_hash_crc.h>
+#include <rte_launch.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+/*
+ * Check condition and return an error if true. Assumes that "handle" is the
+ * name of the hash structure pointer to be freed.
+ */
+#define RETURN_IF_ERROR(cond, str, ...) do {                            \
+	if (cond) {                                                     \
+		printf("ERROR line %d: " str "\n", __LINE__,            \
+							##__VA_ARGS__);	\
+		if (handle)                                             \
+			rte_hash_free(handle);                          \
+		return -1;                                              \
+	}                                                               \
+} while (0)
+
+#define RTE_APP_TEST_HASH_MULTIWRITER_FAILED 0
+
+struct {
+	uint32_t *keys;
+	uint32_t *found;
+	uint32_t nb_tsx_insertion;
+	struct rte_hash *h;
+} tbl_multiwriter_test_params;
+
+const uint32_t nb_entries = 16*1024*1024;
+const uint32_t nb_total_tsx_insertion = 15*1024*1024;
+uint32_t rounded_nb_total_tsx_insertion;
+
+static rte_atomic64_t gcycles;
+static rte_atomic64_t ginsertions;
+
+static int
+test_hash_multiwriter_worker(__attribute__((unused)) void *arg)
+{
+	uint64_t i, offset;
+	uint32_t lcore_id = rte_lcore_id();
+	uint64_t begin, cycles;
+
+	offset = (lcore_id - rte_get_master_lcore())
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	printf("Core #%d inserting %d: %'"PRId64" - %'"PRId64"\n",
+	       lcore_id, tbl_multiwriter_test_params.nb_tsx_insertion,
+	       offset, offset + tbl_multiwriter_test_params.nb_tsx_insertion);
+
+	begin = rte_rdtsc_precise();
+
+	for (i = offset;
+	     i < offset + tbl_multiwriter_test_params.nb_tsx_insertion;
+	     i++) {
+		if (rte_hash_add_key(tbl_multiwriter_test_params.h,
+				     tbl_multiwriter_test_params.keys + i) < 0)
+			break;
+	}
+
+	cycles = rte_rdtsc_precise() - begin;
+	rte_atomic64_add(&gcycles, cycles);
+	rte_atomic64_add(&ginsertions, i - offset);
+
+	for (; i < offset + tbl_multiwriter_test_params.nb_tsx_insertion; i++)
+		tbl_multiwriter_test_params.keys[i]
+			= RTE_APP_TEST_HASH_MULTIWRITER_FAILED;
+
+	return 0;
+}
+
+
+static int
+test_hash_multiwriter(void)
+{
+	unsigned int i, rounded_nb_total_tsx_insertion;
+	static unsigned calledCount = 1;
+
+	uint32_t *keys;
+	uint32_t *found;
+
+	struct rte_hash_parameters hash_params = {
+		.entries = nb_entries,
+		.key_len = sizeof(uint32_t),
+		.hash_func = rte_hash_crc,
+		.hash_func_init_val = 0,
+		.socket_id = rte_socket_id(),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT
+		| RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD,
+	};
+
+	struct rte_hash *handle;
+	char name[RTE_HASH_NAMESIZE];
+
+	const void *next_key;
+	void *next_data;
+	uint32_t iter = 0;
+
+	uint32_t duplicated_keys = 0;
+	uint32_t lost_keys = 0;
+
+	snprintf(name, 32, "test%u", calledCount++);
+	hash_params.name = name;
+
+	handle = rte_hash_create(&hash_params);
+	RETURN_IF_ERROR(handle == NULL, "hash creation failed");
+
+	tbl_multiwriter_test_params.h = handle;
+	tbl_multiwriter_test_params.nb_tsx_insertion =
+		nb_total_tsx_insertion / rte_lcore_count();
+
+	rounded_nb_total_tsx_insertion = (nb_total_tsx_insertion /
+		tbl_multiwriter_test_params.nb_tsx_insertion)
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	rte_srand(rte_rdtsc());
+
+	keys = rte_malloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+
+	if (keys == NULL) {
+		printf("RTE_MALLOC failed\n");
+		goto err1;
+	}
+
+	found = rte_zmalloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+	if (found == NULL) {
+		printf("RTE_ZMALLOC failed\n");
+		goto err2;
+	}
+
+	for (i = 0; i < nb_entries; i++)
+		keys[i] = i;
+
+	tbl_multiwriter_test_params.keys = keys;
+	tbl_multiwriter_test_params.found = found;
+
+	rte_atomic64_init(&gcycles);
+	rte_atomic64_clear(&gcycles);
+
+	rte_atomic64_init(&ginsertions);
+	rte_atomic64_clear(&ginsertions);
+
+	/* Fire all threads. */
+	rte_eal_mp_remote_launch(test_hash_multiwriter_worker,
+				 NULL, CALL_MASTER);
+	rte_eal_mp_wait_lcore();
+
+	while (rte_hash_iterate(handle, &next_key, &next_data, &iter) >= 0) {
+		/* Search for the key in the list of keys added .*/
+		i = *(const uint32_t *)next_key;
+		tbl_multiwriter_test_params.found[i]++;
+	}
+
+	for (i = 0; i < rounded_nb_total_tsx_insertion; i++) {
+		if (tbl_multiwriter_test_params.keys[i]
+		    != RTE_APP_TEST_HASH_MULTIWRITER_FAILED) {
+			if (tbl_multiwriter_test_params.found[i] > 1) {
+				duplicated_keys++;
+				break;
+			}
+			if (tbl_multiwriter_test_params.found[i] == 0) {
+				lost_keys++;
+				printf("key %d is lost\n", i);
+				break;
+			}
+		}
+	}
+
+	if (duplicated_keys > 0) {
+		printf("%d key duplicated\n", duplicated_keys);
+		goto err3;
+	}
+
+	if (lost_keys > 0) {
+		printf("%d key lost\n", lost_keys);
+		goto err3;
+	}
+
+	printf("No key corrupted during multiwriter insertion.\n");
+
+	unsigned long long int cycles_per_insertion =
+		rte_atomic64_read(&gcycles)/
+		rte_atomic64_read(&ginsertions);
+
+	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
+
+	rte_free(tbl_multiwriter_test_params.found);
+	rte_free(tbl_multiwriter_test_params.keys);
+	rte_hash_free(handle);
+	return 0;
+
+err3:
+	rte_free(tbl_multiwriter_test_params.found);
+err2:
+	rte_free(tbl_multiwriter_test_params.keys);
+err1:
+	rte_hash_free(handle);
+	return -1;
+}
+
+static int
+test_hash_multiwriter_main(void)
+{
+	int r = -1;
+
+	if (rte_lcore_count() == 1) {
+		printf(
+			"More than one lcore is required to do multiwriter test\n");
+		return 0;
+	}
+
+	if (!rte_tm_supported()) {
+		printf(
+			"Hardware transactional memory (lock elision) is NOT supported\n");
+		return 0;
+	}
+
+	printf("Hardware transactional memory (lock elision) is supported\n");
+
+	setlocale(LC_NUMERIC, "");
+
+	r = test_hash_multiwriter();
+
+	return r;
+}
+
+
+static struct test_command hash_scaling_cmd = {
+	.command = "hash_multiwriter_autotest",
+	.callback = test_hash_multiwriter_main,
+};
+
+REGISTER_TEST_COMMAND(hash_scaling_cmd);
diff --git a/doc/guides/rel_notes/release_16_07.rst b/doc/guides/rel_notes/release_16_07.rst
index 131723c..f8264fb 100644
--- a/doc/guides/rel_notes/release_16_07.rst
+++ b/doc/guides/rel_notes/release_16_07.rst
@@ -70,6 +70,18 @@ New Features
   * Enable RSS per network interface through the configuration file.
   * Streamline the CLI code.
 
+* **Added multi-writer support for RTE Hash with Intel TSX.**
+
+  The following features/modifications have been added to rte_hash library:
+
+  * Enabled application developers to use an extra flag for rte_hash creation
+    to specify default behavior (multi-thread safe/unsafe) with rte_hash_add_key
+    function.
+  * Changed Cuckoo search algorithm to breadth first search for multi-writer
+    routine and split Cuckoo Search and Move operations in order to reduce
+    transactional code region and improve TSX performance.
+  * Added a hash multi-writer test case for test app.
+
 
 Resolved Issues
 ---------------
diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 7b7d1f8..3cb6770 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -100,7 +100,13 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
 
 #define KEY_ALIGNMENT			16
 
-#define LCORE_CACHE_SIZE		8
+#define LCORE_CACHE_SIZE		64
+
+#define RTE_HASH_BFS_QUEUE_MAX_LEN  1000
+
+#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4
+
+#define RTE_HASH_TSX_MAX_RETRY 10
 
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 /*
@@ -190,6 +196,7 @@ struct rte_hash {
 							memory support */
 	struct lcore_cache *local_free_slots;
 	/**< Local cache per lcore, storing some indexes of the free slots */
+	uint8_t multiwriter_add; /**< Multi-write safe hash add behavior */
 } __rte_cache_aligned;
 
 /* Structure storing both primary and secondary hashes */
@@ -372,7 +379,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
 
 /*
  * If x86 architecture is used, select appropriate compare function,
- * which may use x86 instrinsics, otherwise use memcmp
+ * which may use x86 intrinsics, otherwise use memcmp
  */
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 	/* Select function to compare keys */
@@ -431,7 +438,16 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	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 */
+	/* Turn on multi-writer only with explicit flat from user and TM
+	 * support.
+	 */
+	if (params->extra_flag & RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD
+	    && h->hw_trans_mem_support)
+		h->multiwriter_add = 1;
+	else
+		h->multiwriter_add = 0;
+
+	/* Populate free slots ring. Entry zero is reserved for key misses. */
 	for (i = 1; i < params->entries + 1; i++)
 		rte_ring_sp_enqueue(r, (void *)((uintptr_t) i));
 
@@ -599,6 +615,123 @@ make_space_bucket(const struct rte_hash *h, struct rte_hash_bucket *bkt)
 
 }
 
+struct queue_node {
+	struct rte_hash_bucket *bkt; /* Current bucket on the bfs search */
+
+	struct queue_node *prev;     /* Parent(bucket) in search path */
+	int prev_slot;               /* Parent(slot) in search path */
+};
+
+/* Shift buckets along cuckoo_path and fill the path head with new entry */
+static inline int
+tsx_cuckoo_move_insert(const struct rte_hash *h, struct queue_node *leaf,
+			uint32_t leaf_slot, hash_sig_t sig,
+			hash_sig_t alt_hash, uint32_t new_idx)
+{
+	unsigned try = 0;
+	unsigned status;
+	uint32_t prev_alt_bkt_idx;
+
+	struct queue_node *prev_node, *curr_node = leaf;
+	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
+	uint32_t prev_slot, curr_slot = leaf_slot;
+
+	while (try < RTE_HASH_TSX_MAX_RETRY) {
+		status = rte_xbegin();
+		if (likely(status == RTE_XBEGIN_STARTED)) {
+			while (likely(curr_node->prev != NULL)) {
+				prev_node = curr_node->prev;
+				prev_bkt = prev_node->bkt;
+				prev_slot = curr_node->prev_slot;
+
+				prev_alt_bkt_idx
+					= prev_bkt->signatures[prev_slot].alt
+					    & h->bucket_bitmask;
+
+				if (unlikely(&h->buckets[prev_alt_bkt_idx]
+					     != curr_bkt)) {
+					rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
+				}
+
+				/* Need to swap current/alt sig to allow later Cuckoo insert to
+				 * move elements back to its primary bucket if available
+				 */
+				curr_bkt->signatures[curr_slot].alt =
+				    prev_bkt->signatures[prev_slot].current;
+				curr_bkt->signatures[curr_slot].current =
+				    prev_bkt->signatures[prev_slot].alt;
+				curr_bkt->key_idx[curr_slot]
+				    = prev_bkt->key_idx[prev_slot];
+
+				curr_slot = prev_slot;
+				curr_node = prev_node;
+				curr_bkt = curr_node->bkt;
+			}
+
+			curr_bkt->signatures[curr_slot].current = sig;
+			curr_bkt->signatures[curr_slot].alt = alt_hash;
+			curr_bkt->key_idx[curr_slot] = new_idx;
+
+			rte_xend();
+
+			return 0;
+		}
+
+		/* If we abort we give up this cuckoo path, since most likely it's
+		 * no longer valid as TSX detected data conflict
+		 */
+		try++;
+		rte_pause();
+	}
+
+	return -1;
+}
+
+/*
+ * Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
+ * Cuckoo
+ */
+static inline int
+make_space_insert_bfs_mw(const struct rte_hash *h, struct rte_hash_bucket *bkt,
+			hash_sig_t sig, hash_sig_t alt_hash,
+			uint32_t new_idx)
+{
+	unsigned i;
+	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
+	struct queue_node *tail, *head;
+	struct rte_hash_bucket *curr_bkt, *alt_bkt;
+
+	tail = queue;
+	head = queue + 1;
+	tail->bkt = bkt;
+	tail->prev = NULL;
+	tail->prev_slot = -1;
+
+	/* Cuckoo bfs Search */
+	while (likely(tail != head && head <
+		            queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
+		curr_bkt = tail->bkt;
+		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
+				if (likely(tsx_cuckoo_move_insert(h, tail, i,
+					         sig, alt_hash, new_idx) == 0))
+					return 0;
+			}
+
+			/* Enqueue new node and keep prev node info */
+			alt_bkt = &(h->buckets[curr_bkt->signatures[i].alt
+						    & h->bucket_bitmask]);
+			head->bkt = alt_bkt;
+			head->prev = tail;
+			head->prev_slot = i;
+			head++;
+		}
+		tail++;
+	}
+
+	return -ENOSPC;
+}
+
 /*
  * Function called to enqueue back an index in the cache/ring,
  * as slot has not being used and it can be used in the
@@ -712,30 +845,78 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	rte_memcpy(new_k->key, key, h->key_len);
 	new_k->pdata = data;
 
-	/* Insert new entry is there is room in the primary bucket */
-	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
-		/* Check if slot is available */
-		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
-			prim_bkt->signatures[i].current = sig;
-			prim_bkt->signatures[i].alt = alt_hash;
-			prim_bkt->key_idx[i] = new_idx;
+	if (h->multiwriter_add) {
+		unsigned status;
+		unsigned try = 0;
+
+		while (try < RTE_HASH_TSX_MAX_RETRY) {
+			status = rte_xbegin();
+			if (likely(status == RTE_XBEGIN_STARTED)) {
+				/* Insert new entry if there is room in the primary
+				* bucket.
+				*/
+				for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+					/* Check if slot is available */
+					if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
+						prim_bkt->signatures[i].current = sig;
+						prim_bkt->signatures[i].alt = alt_hash;
+						prim_bkt->key_idx[i] = new_idx;
+						break;
+					}
+				}
+				rte_xend();
+
+				if (i != RTE_HASH_BUCKET_ENTRIES)
+					return new_idx - 1;
+
+				break; /* break off try loop if transaction commits */
+			} else {
+				/* If we abort we give up this cuckoo path. */
+				try++;
+				rte_pause();
+			}
+		}
+
+		/* Primary bucket full, need to make space for new entry */
+		ret = make_space_insert_bfs_mw(h, prim_bkt, sig, alt_hash,
+		                               new_idx);
+
+		if (ret >= 0)
 			return new_idx - 1;
+
+		/* Also search secondary bucket to get better occupancy */
+		ret = make_space_insert_bfs_mw(h, sec_bkt, sig, alt_hash,
+		                               new_idx);
+
+		if (ret >= 0)
+			return new_idx - 1;
+	} else {
+		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+			/* Check if slot is available */
+			if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
+				prim_bkt->signatures[i].current = sig;
+				prim_bkt->signatures[i].alt = alt_hash;
+				prim_bkt->key_idx[i] = new_idx;
+				break;
+			}
 		}
-	}
 
-	/* Primary bucket is full, so we need to make space for new entry */
-	ret = make_space_bucket(h, prim_bkt);
-	/*
-	 * After recursive function.
-	 * Insert the new entry in the position of the pushed entry
-	 * if successful or return error and
-	 * store the new slot back in the ring
-	 */
-	if (ret >= 0) {
-		prim_bkt->signatures[ret].current = sig;
-		prim_bkt->signatures[ret].alt = alt_hash;
-		prim_bkt->key_idx[ret] = new_idx;
-		return new_idx - 1;
+		if (i != RTE_HASH_BUCKET_ENTRIES)
+			return new_idx - 1;
+
+		/* Primary bucket full, need to make space for new entry
+		 * After recursive function.
+		 * Insert the new entry in the position of the pushed entry
+		 * if successful or return error and
+		 * store the new slot back in the ring
+		 */
+		ret = make_space_bucket(h, prim_bkt);
+		if (ret >= 0) {
+			prim_bkt->signatures[ret].current = sig;
+			prim_bkt->signatures[ret].alt = alt_hash;
+			prim_bkt->key_idx[ret] = new_idx;
+			return new_idx - 1;
+		}
 	}
 
 	/* Error in addition, store new slot back in the ring and return error */
diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
index 724315a..c9612fb 100644
--- a/lib/librte_hash/rte_hash.h
+++ b/lib/librte_hash/rte_hash.h
@@ -60,6 +60,9 @@ extern "C" {
 /** Enable Hardware transactional memory support. */
 #define RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT	0x01
 
+/** Default behavior of insertion, single writer/multi writer */
+#define RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x02
+
 /** Signature of key that is stored internally. */
 typedef uint32_t hash_sig_t;
 
-- 
2.5.5

Re: [PATCH v1] hash: add tsx support for cuckoo hash

[Shen, Wei1]

Thanks Pablo for reviewing the patch. I just sent out v2 patch. It fixed those problems you found and also removed the RTE_HASH_KEY_FLAG_MOVED flag used in v1, which would cause problem when key deletion happens (current key deletion doesn’t restore keys back to its primary bucket). Now it just swaps current/alt sig when a push happens, as make_space_bucket does.

There is still some line too long checkpatch warning, but fixing those would hurt code readability rather than improving them. So I left as it.

Thanks.
-- 
Best,

Wei Shen.

On 6/15/16, 4:45 PM, "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com> wrote:

Hi Wei,

> -----Original Message-----
> From: Shen, Wei1
> Sent: Friday, May 06, 2016 9:05 PM
> To: dev@dpdk.org
> Cc: De Lara Guarch, Pablo; Maciocco, Christian; Shen, Wei1; Gobriel, Sameh
> Subject: [PATCH v1] hash: add tsx support for cuckoo hash
> 
> Introduced Intel TSX-enabled scalable multi-writer Cuckoo hash
> insertion.
> 
> This patch introduced scalable multi-writer Cuckoo Hash insertion
> based on a split Cuckoo Search and Move operation using Intel
> TSX. It can do scalable hash insertion with 22 cores with little
> performance loss and negligible TSX abortion rate.
> 
> * Added an extra rte_hash flag definition to switch default
>   single writer Cuckoo Hash behavior to multiwriter.
> 
> * Added a make_space_insert_bfs_mw() function to do split Cuckoo
>   search in BFS order.
> 
> * Added tsx_cuckoo_move_insert() to do Cuckoo move in Intel TSX
>   protected manner.
> 
> * Added test_hash_multiwriter() as test case for multi-writer
>   Cuckoo Hash.
> 
> Signed-off-by: Shen Wei <wei1.shen@intel.com>
> Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
> ---
>  app/test/Makefile                 |   1 +
>  app/test/test_hash_multiwriter.c  | 272
> ++++++++++++++++++++++++++++++++++++++
>  lib/librte_hash/rte_cuckoo_hash.c | 228 ++++++++++++++++++++++++++++--
> --
>  lib/librte_hash/rte_hash.h        |   3 +
>  4 files changed, 480 insertions(+), 24 deletions(-)
>  create mode 100644 app/test/test_hash_multiwriter.c
> 

[...]

> diff --git a/app/test/test_hash_multiwriter.c
> b/app/test/test_hash_multiwriter.c
> new file mode 100644
> index 0000000..bdb9840
> --- /dev/null
> +++ b/app/test/test_hash_multiwriter.c
> @@ -0,0 +1,272 @@

[...]

> +
> +	if (duplicated_keys > 0) {
> +		printf("%d key duplicated\n", duplicated_keys);
> +		goto err3;
> +	}
> +
> +	if (lost_keys > 0) {
> +		printf("%d key lost\n", lost_keys);
> +		goto err3;
> +	}
> +
> +	printf("No key corruped during multiwriter insertion.\n");

Typo: corrupted

> +
> +	unsigned long long int cycles_per_insertion =
> +		rte_atomic64_read(&gcycles)/
> +		rte_atomic64_read(&ginsertions);
> +
> +	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
> +
> +	rte_free(tbl_multiwriter_test_params.found);
> +	rte_free(tbl_multiwriter_test_params.keys);
> +	rte_hash_free(handle);
> +	return 0;
> +
> +err3:
> +	rte_free(tbl_multiwriter_test_params.found);
> +err2:
> +	rte_free(tbl_multiwriter_test_params.keys);
> +err1:
> +	rte_hash_free(handle);
> +	return -1;
> +}
> +
> +static int
> +test_hash_multiwriter_main(void)
> +{
> +	int r = -1;
> +
> +	if (rte_lcore_count() == 1) {
> +		printf(
> +			"More than one lcores are required to do multiwriter
> test\n");

Typo: More than one lcore is required to do multiwriter test.

> +		return r;
> +	}
> +
> +	if (!rte_tm_supported()) {
> +		printf(
> +			"Hardware transactional memory (lock elision) is NOT
> supported\n");
> +		return r;
> +	}
> +
> +	printf("Hardware transactional memory (lock elision) is
> supported\n");
> +
> +	setlocale(LC_NUMERIC, "");
> +
> +	r = test_hash_multiwriter();
> +
> +	return r;
> +}
> +
> +
> +static struct test_command hash_scaling_cmd = {
> +	.command = "hash_multiwriter_autotest",
> +	.callback = test_hash_multiwriter_main,
> +};
> +
> +REGISTER_TEST_COMMAND(hash_scaling_cmd);
> diff --git a/lib/librte_hash/rte_cuckoo_hash.c
> b/lib/librte_hash/rte_cuckoo_hash.c
> index 7b7d1f8..5a01f51 100644
> --- a/lib/librte_hash/rte_cuckoo_hash.c
> +++ b/lib/librte_hash/rte_cuckoo_hash.c

[...]

> +/* Shift buckets along cuckoo_path and fill the path head with new entry */
> +static inline int
> +tsx_cuckoo_move_insert(const struct rte_hash *h, struct queue_node *leaf,
> +		       uint32_t leaf_slot, hash_sig_t sig,
> +		       hash_sig_t alt_hash, uint32_t new_idx, uint8_t flag)
> +{
> +	#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4

Could you move this up and indent it to the left?

> +	int32_t try = 0;

This variable (try) should not have any negative number, so you can change it to unsigned.

> +	uint32_t prev_alt_bkt_idx;
> +	unsigned status;
> +
> +	struct queue_node *prev_node, *curr_node = leaf;
> +	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
> +	uint32_t prev_slot, curr_slot = leaf_slot;
> +
> +	int cuckoo_path_len = 0;

This variable is not being used. It gets incremented below,
but is not used in anything useful, as far as I can see.
If you are not using it, then remove it.

> +
> +	while (try < 10) {

Magic number. Define a macro with this number instead.

> +		status = rte_xbegin();
> +		if (likely(status == RTE_XBEGIN_STARTED)) {
> +			while (likely(curr_node->prev != NULL)) {
> +				prev_node = curr_node->prev;
> +				prev_bkt = prev_node->bkt;
> +				prev_slot = curr_node->prev_slot;
> +
> +				prev_alt_bkt_idx
> +					= prev_bkt->signatures[prev_slot].alt
> +					& h->bucket_bitmask;
> +
> +				if (unlikely(&h->buckets[prev_alt_bkt_idx]
> +					     != curr_bkt)) {
> +
> 	rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
> +				}
> +
> +				curr_bkt->signatures[curr_slot]
> +					= prev_bkt->signatures[prev_slot];
> +				curr_bkt->key_idx[curr_slot]
> +					= prev_bkt->key_idx[prev_slot];
> +				curr_bkt->flag[curr_slot]
> +					= RTE_HASH_KEY_FLAG_MOVED;
> +
> +				curr_slot = prev_slot;
> +				curr_node = prev_node;
> +				curr_bkt = curr_node->bkt;
> +
> +				cuckoo_path_len++;
> +			}
> +
> +			curr_bkt->signatures[curr_slot].current = sig;
> +			curr_bkt->signatures[curr_slot].alt = alt_hash;
> +			curr_bkt->key_idx[curr_slot] = new_idx;
> +			curr_bkt->flag[curr_slot] = flag;
> +
> +			rte_xend();
> +
> +			return 0;
> +		}
> +
> +		/* If we abort we give up this cuckoo path. */
> +		try++;
> +		rte_pause();
> +		continue;

Is this continue necessary? It is at the end of a while loop, so it does not look it is.

> +	}
> +
> +	return -1;
> +}
> +
> +/* Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
> + * Cuckoo
> + */
> +static inline int
> +make_space_insert_bfs_mw(const struct rte_hash *h, struct
> rte_hash_bucket *bkt,
> +			 hash_sig_t sig, hash_sig_t alt_hash,
> +			 uint32_t new_idx, uint8_t flag)
> +{
> +	int i;

Unsigned i;

> +	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
> +	struct queue_node *tail, *head;
> +	struct rte_hash_bucket *curr_bkt, *alt_bkt;
> +
> +	tail = queue;
> +	head = queue + 1;
> +	tail->bkt = bkt;
> +	tail->prev = NULL;
> +	tail->prev_slot = -1;
> +
> +	/* Cuckoo Search */
> +	while (likely(tail != head && head <
> +		      queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
> +
> +		curr_bkt = tail->bkt;
> +		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
> +				if (likely(tsx_cuckoo_move_insert(h, tail, i,
> +					sig, alt_hash, new_idx, flag) == 0))

Add extra indentation here, to distinguish between the conditional
and the body (extra indentation to the right in "sig, alt_hash...").

> +					return 0;
> +			}
> +
> +			/* Make sure current key's not already in its
> +			 * secondary bucket.
> +			 */
> +			if (curr_bkt->flag[i] ==
> RTE_HASH_KEY_FLAG_UNMOVED) {
> +				/* Enqueue new node and keep prev node
> info */
> +				alt_bkt =
> +					&(h->buckets[curr_bkt-
> >signatures[i].alt
> +						     & h->bucket_bitmask]);
> +				head->bkt = alt_bkt;
> +				head->prev = tail;
> +				head->prev_slot = i;
> +				head++;
> +			}
> +		}
> +		tail++;
> +	}
> +
> +	return -ENOSPC;
> +}
> +
>  /*
>   * Function called to enqueue back an index in the cache/ring,
>   * as slot has not being used and it can be used in the
> @@ -712,30 +852,70 @@ __rte_hash_add_key_with_hash(const struct
> rte_hash *h, const void *key,
>  	rte_memcpy(new_k->key, key, h->key_len);
>  	new_k->pdata = data;
> 
> -	/* Insert new entry is there is room in the primary bucket */
> -	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> -		/* Check if slot is available */
> -		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
> -			prim_bkt->signatures[i].current = sig;
> -			prim_bkt->signatures[i].alt = alt_hash;
> -			prim_bkt->key_idx[i] = new_idx;
> -			return new_idx - 1;
> +	unsigned status;
> +	int32_t try = 0;
> +
> +	while (try < 10) {


Same comments about "unsigned try" and magic numbers.

> +		status = rte_xbegin();

The following code should only be executed if RTM is supported,
otherwise, there will be an illegal instruction.

> +		if (likely(status == RTE_XBEGIN_STARTED)) {
> +			/* Insert new entry is there is room in the primary
> +			 * bucket.

Typo: Insert new entry if there is room...

> +			 */
> +			for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +				/* Check if slot is available */
> +				if (likely(prim_bkt->signatures[i].sig
> +					   == NULL_SIGNATURE)) {

Extra indentation to the right, as said above.

> +					prim_bkt->signatures[i].current = sig;
> +					prim_bkt->signatures[i].alt =
> alt_hash;
> +					prim_bkt->key_idx[i] = new_idx;
> +					prim_bkt->flag[i] =
> +
> 	RTE_HASH_KEY_FLAG_UNMOVED;
> +					break;
> +				}
> +			}
> +			rte_xend();
> +
> +			if (i != RTE_HASH_BUCKET_ENTRIES)
> +				return new_idx - 1;
> +
> +			break;
> +
> +		} else {
> +			/* If we abort we give up this cuckoo path. */
> +			try++;
> +			rte_pause();
> +			continue;

Same as above, unnecessary continue?

>  		}
>  	}
> 
>  	/* Primary bucket is full, so we need to make space for new entry */

Re: [PATCH v2] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[Ananyev, Konstantin]

Hi Wei,

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wei Shen
> Sent: Thursday, June 16, 2016 5:53 AM
> To: dev@dpdk.org
> Cc: De Lara Guarch, Pablo; stephen@networkplumber.org; Tai, Charlie; Maciocco, Christian; Gobriel, Sameh; Shen, Wei1
> Subject: [dpdk-dev] [PATCH v2] rte_hash: add scalable multi-writer insertion w/ Intel TSX
> 
> This patch introduced scalable multi-writer Cuckoo Hash insertion
> based on a split Cuckoo Search and Move operation using Intel
> TSX. It can do scalable hash insertion with 22 cores with little
> performance loss and negligible TSX abortion rate.
> 
> * Added an extra rte_hash flag definition to switch default
>   single writer Cuckoo Hash behavior to multiwriter.
> 
> * Added a make_space_insert_bfs_mw() function to do split Cuckoo
>   search in BFS order.
> 
> * Added tsx_cuckoo_move_insert() to do Cuckoo move in Intel TSX
>   protected manner.
> 
> * Added test_hash_multiwriter() as test case for multi-writer
>   Cuckoo Hash.
> 
> Signed-off-by: Shen Wei <wei1.shen@intel.com>
> Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
> ---
>  app/test/Makefile                      |   1 +
>  app/test/test_hash_multiwriter.c       | 272 +++++++++++++++++++++++++++++++++
>  doc/guides/rel_notes/release_16_07.rst |  12 ++
>  lib/librte_hash/rte_cuckoo_hash.c      | 231 +++++++++++++++++++++++++---
>  lib/librte_hash/rte_hash.h             |   3 +
>  5 files changed, 494 insertions(+), 25 deletions(-)
>  create mode 100644 app/test/test_hash_multiwriter.c
> 
> diff --git a/app/test/Makefile b/app/test/Makefile
> index 053f3a2..5476300 100644
> --- a/app/test/Makefile
> +++ b/app/test/Makefile
> @@ -120,6 +120,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_thash.c
>  SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_perf.c
>  SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_functions.c
>  SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_scaling.c
> +SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_multiwriter.c
> 
>  SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm.c
>  SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm_perf.c
> diff --git a/app/test/test_hash_multiwriter.c b/app/test/test_hash_multiwriter.c
> new file mode 100644
> index 0000000..54a0d2c
> --- /dev/null
> +++ b/app/test/test_hash_multiwriter.c
> @@ -0,0 +1,272 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2016 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *	 notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *	 notice, this list of conditions and the following disclaimer in
> + *	 the documentation and/or other materials provided with the
> + *	 distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *	 contributors may be used to endorse or promote products derived
> + *	 from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +#include <inttypes.h>
> +#include <locale.h>
> +
> +#include <rte_cycles.h>
> +#include <rte_hash.h>
> +#include <rte_hash_crc.h>
> +#include <rte_launch.h>
> +#include <rte_malloc.h>
> +#include <rte_random.h>
> +#include <rte_spinlock.h>
> +
> +#include "test.h"
> +
> +/*
> + * Check condition and return an error if true. Assumes that "handle" is the
> + * name of the hash structure pointer to be freed.
> + */
> +#define RETURN_IF_ERROR(cond, str, ...) do {                            \
> +	if (cond) {                                                     \
> +		printf("ERROR line %d: " str "\n", __LINE__,            \
> +							##__VA_ARGS__);	\
> +		if (handle)                                             \
> +			rte_hash_free(handle);                          \
> +		return -1;                                              \
> +	}                                                               \
> +} while (0)
> +
> +#define RTE_APP_TEST_HASH_MULTIWRITER_FAILED 0
> +
> +struct {
> +	uint32_t *keys;
> +	uint32_t *found;
> +	uint32_t nb_tsx_insertion;
> +	struct rte_hash *h;
> +} tbl_multiwriter_test_params;
> +
> +const uint32_t nb_entries = 16*1024*1024;
> +const uint32_t nb_total_tsx_insertion = 15*1024*1024;
> +uint32_t rounded_nb_total_tsx_insertion;
> +
> +static rte_atomic64_t gcycles;
> +static rte_atomic64_t ginsertions;
> +
> +static int
> +test_hash_multiwriter_worker(__attribute__((unused)) void *arg)
> +{
> +	uint64_t i, offset;
> +	uint32_t lcore_id = rte_lcore_id();
> +	uint64_t begin, cycles;
> +
> +	offset = (lcore_id - rte_get_master_lcore())
> +		* tbl_multiwriter_test_params.nb_tsx_insertion;
> +
> +	printf("Core #%d inserting %d: %'"PRId64" - %'"PRId64"\n",
> +	       lcore_id, tbl_multiwriter_test_params.nb_tsx_insertion,
> +	       offset, offset + tbl_multiwriter_test_params.nb_tsx_insertion);
> +
> +	begin = rte_rdtsc_precise();
> +
> +	for (i = offset;
> +	     i < offset + tbl_multiwriter_test_params.nb_tsx_insertion;
> +	     i++) {
> +		if (rte_hash_add_key(tbl_multiwriter_test_params.h,
> +				     tbl_multiwriter_test_params.keys + i) < 0)
> +			break;
> +	}
> +
> +	cycles = rte_rdtsc_precise() - begin;
> +	rte_atomic64_add(&gcycles, cycles);
> +	rte_atomic64_add(&ginsertions, i - offset);
> +
> +	for (; i < offset + tbl_multiwriter_test_params.nb_tsx_insertion; i++)
> +		tbl_multiwriter_test_params.keys[i]
> +			= RTE_APP_TEST_HASH_MULTIWRITER_FAILED;
> +
> +	return 0;
> +}
> +
> +
> +static int
> +test_hash_multiwriter(void)
> +{
> +	unsigned int i, rounded_nb_total_tsx_insertion;
> +	static unsigned calledCount = 1;
> +
> +	uint32_t *keys;
> +	uint32_t *found;
> +
> +	struct rte_hash_parameters hash_params = {
> +		.entries = nb_entries,
> +		.key_len = sizeof(uint32_t),
> +		.hash_func = rte_hash_crc,
> +		.hash_func_init_val = 0,
> +		.socket_id = rte_socket_id(),
> +		.extra_flag = RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT
> +		| RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD,
> +	};
> +
> +	struct rte_hash *handle;
> +	char name[RTE_HASH_NAMESIZE];
> +
> +	const void *next_key;
> +	void *next_data;
> +	uint32_t iter = 0;
> +
> +	uint32_t duplicated_keys = 0;
> +	uint32_t lost_keys = 0;
> +
> +	snprintf(name, 32, "test%u", calledCount++);
> +	hash_params.name = name;
> +
> +	handle = rte_hash_create(&hash_params);
> +	RETURN_IF_ERROR(handle == NULL, "hash creation failed");
> +
> +	tbl_multiwriter_test_params.h = handle;
> +	tbl_multiwriter_test_params.nb_tsx_insertion =
> +		nb_total_tsx_insertion / rte_lcore_count();
> +
> +	rounded_nb_total_tsx_insertion = (nb_total_tsx_insertion /
> +		tbl_multiwriter_test_params.nb_tsx_insertion)
> +		* tbl_multiwriter_test_params.nb_tsx_insertion;
> +
> +	rte_srand(rte_rdtsc());
> +
> +	keys = rte_malloc(NULL, sizeof(uint32_t) * nb_entries, 0);
> +
> +	if (keys == NULL) {
> +		printf("RTE_MALLOC failed\n");
> +		goto err1;
> +	}
> +
> +	found = rte_zmalloc(NULL, sizeof(uint32_t) * nb_entries, 0);
> +	if (found == NULL) {
> +		printf("RTE_ZMALLOC failed\n");
> +		goto err2;
> +	}
> +
> +	for (i = 0; i < nb_entries; i++)
> +		keys[i] = i;
> +
> +	tbl_multiwriter_test_params.keys = keys;
> +	tbl_multiwriter_test_params.found = found;
> +
> +	rte_atomic64_init(&gcycles);
> +	rte_atomic64_clear(&gcycles);
> +
> +	rte_atomic64_init(&ginsertions);
> +	rte_atomic64_clear(&ginsertions);
> +
> +	/* Fire all threads. */
> +	rte_eal_mp_remote_launch(test_hash_multiwriter_worker,
> +				 NULL, CALL_MASTER);
> +	rte_eal_mp_wait_lcore();
> +
> +	while (rte_hash_iterate(handle, &next_key, &next_data, &iter) >= 0) {
> +		/* Search for the key in the list of keys added .*/
> +		i = *(const uint32_t *)next_key;
> +		tbl_multiwriter_test_params.found[i]++;
> +	}
> +
> +	for (i = 0; i < rounded_nb_total_tsx_insertion; i++) {
> +		if (tbl_multiwriter_test_params.keys[i]
> +		    != RTE_APP_TEST_HASH_MULTIWRITER_FAILED) {
> +			if (tbl_multiwriter_test_params.found[i] > 1) {
> +				duplicated_keys++;
> +				break;
> +			}
> +			if (tbl_multiwriter_test_params.found[i] == 0) {
> +				lost_keys++;
> +				printf("key %d is lost\n", i);
> +				break;
> +			}
> +		}
> +	}
> +
> +	if (duplicated_keys > 0) {
> +		printf("%d key duplicated\n", duplicated_keys);
> +		goto err3;
> +	}
> +
> +	if (lost_keys > 0) {
> +		printf("%d key lost\n", lost_keys);
> +		goto err3;
> +	}
> +
> +	printf("No key corrupted during multiwriter insertion.\n");
> +
> +	unsigned long long int cycles_per_insertion =
> +		rte_atomic64_read(&gcycles)/
> +		rte_atomic64_read(&ginsertions);
> +
> +	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
> +
> +	rte_free(tbl_multiwriter_test_params.found);
> +	rte_free(tbl_multiwriter_test_params.keys);
> +	rte_hash_free(handle);
> +	return 0;
> +
> +err3:
> +	rte_free(tbl_multiwriter_test_params.found);
> +err2:
> +	rte_free(tbl_multiwriter_test_params.keys);
> +err1:
> +	rte_hash_free(handle);
> +	return -1;
> +}
> +
> +static int
> +test_hash_multiwriter_main(void)
> +{
> +	int r = -1;
> +
> +	if (rte_lcore_count() == 1) {
> +		printf(
> +			"More than one lcore is required to do multiwriter test\n");
> +		return 0;
> +	}
> +
> +	if (!rte_tm_supported()) {
> +		printf(
> +			"Hardware transactional memory (lock elision) is NOT supported\n");
> +		return 0;
> +	}
> +
> +	printf("Hardware transactional memory (lock elision) is supported\n");
> +
> +	setlocale(LC_NUMERIC, "");
> +
> +	r = test_hash_multiwriter();
> +
> +	return r;
> +}
> +
> +
> +static struct test_command hash_scaling_cmd = {
> +	.command = "hash_multiwriter_autotest",
> +	.callback = test_hash_multiwriter_main,
> +};
> +
> +REGISTER_TEST_COMMAND(hash_scaling_cmd);
> diff --git a/doc/guides/rel_notes/release_16_07.rst b/doc/guides/rel_notes/release_16_07.rst
> index 131723c..f8264fb 100644
> --- a/doc/guides/rel_notes/release_16_07.rst
> +++ b/doc/guides/rel_notes/release_16_07.rst
> @@ -70,6 +70,18 @@ New Features
>    * Enable RSS per network interface through the configuration file.
>    * Streamline the CLI code.
> 
> +* **Added multi-writer support for RTE Hash with Intel TSX.**
> +
> +  The following features/modifications have been added to rte_hash library:
> +
> +  * Enabled application developers to use an extra flag for rte_hash creation
> +    to specify default behavior (multi-thread safe/unsafe) with rte_hash_add_key
> +    function.
> +  * Changed Cuckoo search algorithm to breadth first search for multi-writer
> +    routine and split Cuckoo Search and Move operations in order to reduce
> +    transactional code region and improve TSX performance.
> +  * Added a hash multi-writer test case for test app.
> +
> 
>  Resolved Issues
>  ---------------
> diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
> index 7b7d1f8..3cb6770 100644
> --- a/lib/librte_hash/rte_cuckoo_hash.c
> +++ b/lib/librte_hash/rte_cuckoo_hash.c
> @@ -1,7 +1,7 @@
>  /*-
>   *   BSD LICENSE
>   *
> - *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
> + *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
>   *   All rights reserved.
>   *
>   *   Redistribution and use in source and binary forms, with or without
> @@ -100,7 +100,13 @@ EAL_REGISTER_TAILQ(rte_hash_tailq)
> 
>  #define KEY_ALIGNMENT			16
> 
> -#define LCORE_CACHE_SIZE		8
> +#define LCORE_CACHE_SIZE		64
> +
> +#define RTE_HASH_BFS_QUEUE_MAX_LEN  1000
> +
> +#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4
> +
> +#define RTE_HASH_TSX_MAX_RETRY 10
> 
>  #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
>  /*
> @@ -190,6 +196,7 @@ struct rte_hash {
>  							memory support */
>  	struct lcore_cache *local_free_slots;
>  	/**< Local cache per lcore, storing some indexes of the free slots */
> +	uint8_t multiwriter_add; /**< Multi-write safe hash add behavior */
>  } __rte_cache_aligned;
> 
>  /* Structure storing both primary and secondary hashes */
> @@ -372,7 +379,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
> 
>  /*
>   * If x86 architecture is used, select appropriate compare function,
> - * which may use x86 instrinsics, otherwise use memcmp
> + * which may use x86 intrinsics, otherwise use memcmp
>   */
>  #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
>  	/* Select function to compare keys */
> @@ -431,7 +438,16 @@ rte_hash_create(const struct rte_hash_parameters *params)
>  	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 */
> +	/* Turn on multi-writer only with explicit flat from user and TM
> +	 * support.
> +	 */
> +	if (params->extra_flag & RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD
> +	    && h->hw_trans_mem_support)
> +		h->multiwriter_add = 1;
> +	else
> +		h->multiwriter_add = 0;


Wonder why MULTIPLE-WRITER support has to be implemented only for machines with TSX support?
>From initial discussion my understanding was that it would work on both arhitectures with and without TSX:
on non-TSX platforms approach with spin-lock will be used.
Do I miss something here?   

> +
> +	/* Populate free slots ring. Entry zero is reserved for key misses. */
>  	for (i = 1; i < params->entries + 1; i++)
>  		rte_ring_sp_enqueue(r, (void *)((uintptr_t) i));
> 
> @@ -599,6 +615,123 @@ make_space_bucket(const struct rte_hash *h, struct rte_hash_bucket *bkt)
> 
>  }
> 
> +struct queue_node {
> +	struct rte_hash_bucket *bkt; /* Current bucket on the bfs search */
> +
> +	struct queue_node *prev;     /* Parent(bucket) in search path */
> +	int prev_slot;               /* Parent(slot) in search path */
> +};
> +
> +/* Shift buckets along cuckoo_path and fill the path head with new entry */
> +static inline int
> +tsx_cuckoo_move_insert(const struct rte_hash *h, struct queue_node *leaf,
> +			uint32_t leaf_slot, hash_sig_t sig,
> +			hash_sig_t alt_hash, uint32_t new_idx)
> +{
> +	unsigned try = 0;
> +	unsigned status;
> +	uint32_t prev_alt_bkt_idx;
> +
> +	struct queue_node *prev_node, *curr_node = leaf;
> +	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
> +	uint32_t prev_slot, curr_slot = leaf_slot;
> +
> +	while (try < RTE_HASH_TSX_MAX_RETRY) {
> +		status = rte_xbegin();

Hmm, would it compile for non-IA platform?
As I remember, we have rte_xbegin/xend/... defined only for x86 arch,
and I don't see here any mechanism to exclude that code from compilation on non-x86 arch
(#ifdef RTE_ARCH_X86 or so).

Konstantin

> +		if (likely(status == RTE_XBEGIN_STARTED)) {
> +			while (likely(curr_node->prev != NULL)) {
> +				prev_node = curr_node->prev;
> +				prev_bkt = prev_node->bkt;
> +				prev_slot = curr_node->prev_slot;
> +
> +				prev_alt_bkt_idx
> +					= prev_bkt->signatures[prev_slot].alt
> +					    & h->bucket_bitmask;
> +
> +				if (unlikely(&h->buckets[prev_alt_bkt_idx]
> +					     != curr_bkt)) {
> +					rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
> +				}
> +
> +				/* Need to swap current/alt sig to allow later Cuckoo insert to
> +				 * move elements back to its primary bucket if available
> +				 */
> +				curr_bkt->signatures[curr_slot].alt =
> +				    prev_bkt->signatures[prev_slot].current;
> +				curr_bkt->signatures[curr_slot].current =
> +				    prev_bkt->signatures[prev_slot].alt;
> +				curr_bkt->key_idx[curr_slot]
> +				    = prev_bkt->key_idx[prev_slot];
> +
> +				curr_slot = prev_slot;
> +				curr_node = prev_node;
> +				curr_bkt = curr_node->bkt;
> +			}
> +
> +			curr_bkt->signatures[curr_slot].current = sig;
> +			curr_bkt->signatures[curr_slot].alt = alt_hash;
> +			curr_bkt->key_idx[curr_slot] = new_idx;
> +
> +			rte_xend();
> +
> +			return 0;
> +		}
> +
> +		/* If we abort we give up this cuckoo path, since most likely it's
> +		 * no longer valid as TSX detected data conflict
> +		 */
> +		try++;
> +		rte_pause();
> +	}
> +
> +	return -1;
> +}
> +
> +/*
> + * Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
> + * Cuckoo
> + */
> +static inline int
> +make_space_insert_bfs_mw(const struct rte_hash *h, struct rte_hash_bucket *bkt,
> +			hash_sig_t sig, hash_sig_t alt_hash,
> +			uint32_t new_idx)
> +{
> +	unsigned i;
> +	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
> +	struct queue_node *tail, *head;
> +	struct rte_hash_bucket *curr_bkt, *alt_bkt;
> +
> +	tail = queue;
> +	head = queue + 1;
> +	tail->bkt = bkt;
> +	tail->prev = NULL;
> +	tail->prev_slot = -1;
> +
> +	/* Cuckoo bfs Search */
> +	while (likely(tail != head && head <
> +		            queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
> +		curr_bkt = tail->bkt;
> +		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
> +				if (likely(tsx_cuckoo_move_insert(h, tail, i,
> +					         sig, alt_hash, new_idx) == 0))
> +					return 0;
> +			}
> +
> +			/* Enqueue new node and keep prev node info */
> +			alt_bkt = &(h->buckets[curr_bkt->signatures[i].alt
> +						    & h->bucket_bitmask]);
> +			head->bkt = alt_bkt;
> +			head->prev = tail;
> +			head->prev_slot = i;
> +			head++;
> +		}
> +		tail++;
> +	}
> +
> +	return -ENOSPC;
> +}
> +
>  /*
>   * Function called to enqueue back an index in the cache/ring,
>   * as slot has not being used and it can be used in the
> @@ -712,30 +845,78 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
>  	rte_memcpy(new_k->key, key, h->key_len);
>  	new_k->pdata = data;
> 
> -	/* Insert new entry is there is room in the primary bucket */
> -	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> -		/* Check if slot is available */
> -		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
> -			prim_bkt->signatures[i].current = sig;
> -			prim_bkt->signatures[i].alt = alt_hash;
> -			prim_bkt->key_idx[i] = new_idx;
> +	if (h->multiwriter_add) {
> +		unsigned status;
> +		unsigned try = 0;
> +
> +		while (try < RTE_HASH_TSX_MAX_RETRY) {
> +			status = rte_xbegin();
> +			if (likely(status == RTE_XBEGIN_STARTED)) {
> +				/* Insert new entry if there is room in the primary
> +				* bucket.
> +				*/
> +				for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +					/* Check if slot is available */
> +					if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
> +						prim_bkt->signatures[i].current = sig;
> +						prim_bkt->signatures[i].alt = alt_hash;
> +						prim_bkt->key_idx[i] = new_idx;
> +						break;
> +					}
> +				}
> +				rte_xend();
> +
> +				if (i != RTE_HASH_BUCKET_ENTRIES)
> +					return new_idx - 1;
> +
> +				break; /* break off try loop if transaction commits */
> +			} else {
> +				/* If we abort we give up this cuckoo path. */
> +				try++;
> +				rte_pause();
> +			}
> +		}
> +
> +		/* Primary bucket full, need to make space for new entry */
> +		ret = make_space_insert_bfs_mw(h, prim_bkt, sig, alt_hash,
> +		                               new_idx);
> +
> +		if (ret >= 0)
>  			return new_idx - 1;
> +
> +		/* Also search secondary bucket to get better occupancy */
> +		ret = make_space_insert_bfs_mw(h, sec_bkt, sig, alt_hash,
> +		                               new_idx);
> +
> +		if (ret >= 0)
> +			return new_idx - 1;
> +	} else {
> +		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
> +			/* Check if slot is available */
> +			if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
> +				prim_bkt->signatures[i].current = sig;
> +				prim_bkt->signatures[i].alt = alt_hash;
> +				prim_bkt->key_idx[i] = new_idx;
> +				break;
> +			}
>  		}
> -	}
> 
> -	/* Primary bucket is full, so we need to make space for new entry */
> -	ret = make_space_bucket(h, prim_bkt);
> -	/*
> -	 * After recursive function.
> -	 * Insert the new entry in the position of the pushed entry
> -	 * if successful or return error and
> -	 * store the new slot back in the ring
> -	 */
> -	if (ret >= 0) {
> -		prim_bkt->signatures[ret].current = sig;
> -		prim_bkt->signatures[ret].alt = alt_hash;
> -		prim_bkt->key_idx[ret] = new_idx;
> -		return new_idx - 1;
> +		if (i != RTE_HASH_BUCKET_ENTRIES)
> +			return new_idx - 1;
> +
> +		/* Primary bucket full, need to make space for new entry
> +		 * After recursive function.
> +		 * Insert the new entry in the position of the pushed entry
> +		 * if successful or return error and
> +		 * store the new slot back in the ring
> +		 */
> +		ret = make_space_bucket(h, prim_bkt);
> +		if (ret >= 0) {
> +			prim_bkt->signatures[ret].current = sig;
> +			prim_bkt->signatures[ret].alt = alt_hash;
> +			prim_bkt->key_idx[ret] = new_idx;
> +			return new_idx - 1;
> +		}
>  	}
> 
>  	/* Error in addition, store new slot back in the ring and return error */
> diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
> index 724315a..c9612fb 100644
> --- a/lib/librte_hash/rte_hash.h
> +++ b/lib/librte_hash/rte_hash.h
> @@ -60,6 +60,9 @@ extern "C" {
>  /** Enable Hardware transactional memory support. */
>  #define RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT	0x01
> 
> +/** Default behavior of insertion, single writer/multi writer */
> +#define RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x02
> +
>  /** Signature of key that is stored internally. */
>  typedef uint32_t hash_sig_t;
> 
> --
> 2.5.5

[PATCH v3] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[Wei Shen]

Here's the latest version of the rte_hash multi-writer patch.
It's re-based on top of the latest head as of Jun 16, 2016.

http://dpdk.org/dev/patchwork/patch/13886/
http://dpdk.org/dev/patchwork/patch/12589/

v3 changes:

* Made spinlock as fall back behavior when developer choose to use multi-writer
  behavior while HTM is not available.
* Created a rte_cuckoo_hash_x86.h file to hold all x86-specific related
  cuckoo_hash functions. And rte_cuckoo_hash.c uses compile time flag to
  select x86 file or other platform-specific implementations. While HTM check
  is still done at runtime (same with RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT)
* Moved rte_hash private struct definitions to rte_cuckoo_hash.h, to allow
  rte_cuckoo_hash_x86.h or future platform dependent functions to include.
* Following renaming for consistent names when new platform TM support are
  added.
  - rte_hash_cuckoo_insert_mw_tm for trying insertion without moving buckets
   around.
  - rte_hash_cuckoo_move_insert_mw_tm for trying insertion by moving buckets
   around.

v2 changes:

* Address issues pointed out by reviews on mailing list.
* Removed the RTE_HASH_KEY_FLAG_MOVED flag used in v1, which would cause
  problem when key deletion happens.

Wei Shen (1):
  rte_hash: add scalable multi-writer insertion w/ Intel TSX

 app/test/Makefile                      |   1 +
 app/test/test_hash_multiwriter.c       | 287 +++++++++++++++++++++++++++++++++
 doc/guides/rel_notes/release_16_07.rst |  12 ++
 lib/librte_hash/rte_cuckoo_hash.c      | 258 ++++++++++-------------------
 lib/librte_hash/rte_cuckoo_hash.h      | 219 +++++++++++++++++++++++++
 lib/librte_hash/rte_cuckoo_hash_x86.h  | 193 ++++++++++++++++++++++
 lib/librte_hash/rte_hash.h             |   3 +
 7 files changed, 796 insertions(+), 177 deletions(-)
 create mode 100644 app/test/test_hash_multiwriter.c
 create mode 100644 lib/librte_hash/rte_cuckoo_hash.h
 create mode 100644 lib/librte_hash/rte_cuckoo_hash_x86.h

-- 
2.5.5

[PATCH v3] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[Wei Shen]

This patch introduced scalable multi-writer Cuckoo Hash insertion
based on a split Cuckoo Search and Move operation using Intel
TSX. It can do scalable hash insertion with 22 cores with little
performance loss and negligible TSX abortion rate.

* Added an extra rte_hash flag definition to switch default single writer
  Cuckoo Hash behavior to multiwriter.
    - If HTM is available, it would use hardware feature for concurrency.
    - If HTM is not available, it would fall back to spinlock.

* Created a rte_cuckoo_hash_x86.h file to hold all x86-arch related
  cuckoo_hash functions. And rte_cuckoo_hash.c uses compile time flag to
  select x86 file or other platform-specific implementations. While HTM check
  is still done at runtime (same idea with
  RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT)

* Moved rte_hash private struct definitions to rte_cuckoo_hash.h, to allow
  rte_cuckoo_hash_x86.h or future platform dependent functions to include.

* Following new functions are created for consistent names when new platform
  TM support are added.
    - rte_hash_cuckoo_move_insert_mw_tm: do insertion with bucket movement.
    - rte_hash_cuckoo_insert_mw_tm: do insertion without bucket movement.

* One extra multi-writer test case is added.

Signed-off-by: Shen Wei <wei1.shen@intel.com>
Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
---
 app/test/Makefile                      |   1 +
 app/test/test_hash_multiwriter.c       | 287 +++++++++++++++++++++++++++++++++
 doc/guides/rel_notes/release_16_07.rst |  12 ++
 lib/librte_hash/rte_cuckoo_hash.c      | 258 ++++++++++-------------------
 lib/librte_hash/rte_cuckoo_hash.h      | 219 +++++++++++++++++++++++++
 lib/librte_hash/rte_cuckoo_hash_x86.h  | 193 ++++++++++++++++++++++
 lib/librte_hash/rte_hash.h             |   3 +
 7 files changed, 796 insertions(+), 177 deletions(-)
 create mode 100644 app/test/test_hash_multiwriter.c
 create mode 100644 lib/librte_hash/rte_cuckoo_hash.h
 create mode 100644 lib/librte_hash/rte_cuckoo_hash_x86.h

diff --git a/app/test/Makefile b/app/test/Makefile
index 053f3a2..5476300 100644
--- a/app/test/Makefile
+++ b/app/test/Makefile
@@ -120,6 +120,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_thash.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_perf.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_functions.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_scaling.c
+SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_multiwriter.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm.c
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm_perf.c
diff --git a/app/test/test_hash_multiwriter.c b/app/test/test_hash_multiwriter.c
new file mode 100644
index 0000000..b0f31b0
--- /dev/null
+++ b/app/test/test_hash_multiwriter.c
@@ -0,0 +1,287 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *	 notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *	 notice, this list of conditions and the following disclaimer in
+ *	 the documentation and/or other materials provided with the
+ *	 distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *	 contributors may be used to endorse or promote products derived
+ *	 from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <inttypes.h>
+#include <locale.h>
+
+#include <rte_cycles.h>
+#include <rte_hash.h>
+#include <rte_hash_crc.h>
+#include <rte_launch.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+/*
+ * Check condition and return an error if true. Assumes that "handle" is the
+ * name of the hash structure pointer to be freed.
+ */
+#define RETURN_IF_ERROR(cond, str, ...) do {                            \
+	if (cond) {                                                     \
+		printf("ERROR line %d: " str "\n", __LINE__,            \
+							##__VA_ARGS__);	\
+		if (handle)                                             \
+			rte_hash_free(handle);                          \
+		return -1;                                              \
+	}                                                               \
+} while (0)
+
+#define RTE_APP_TEST_HASH_MULTIWRITER_FAILED 0
+
+struct {
+	uint32_t *keys;
+	uint32_t *found;
+	uint32_t nb_tsx_insertion;
+	struct rte_hash *h;
+} tbl_multiwriter_test_params;
+
+const uint32_t nb_entries = 16*1024*1024;
+const uint32_t nb_total_tsx_insertion = 15*1024*1024;
+uint32_t rounded_nb_total_tsx_insertion;
+
+static rte_atomic64_t gcycles;
+static rte_atomic64_t ginsertions;
+
+static int use_htm;
+
+static int
+test_hash_multiwriter_worker(__attribute__((unused)) void *arg)
+{
+	uint64_t i, offset;
+	uint32_t lcore_id = rte_lcore_id();
+	uint64_t begin, cycles;
+
+	offset = (lcore_id - rte_get_master_lcore())
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	printf("Core #%d inserting %d: %'"PRId64" - %'"PRId64"\n",
+	       lcore_id, tbl_multiwriter_test_params.nb_tsx_insertion,
+	       offset, offset + tbl_multiwriter_test_params.nb_tsx_insertion);
+
+	begin = rte_rdtsc_precise();
+
+	for (i = offset;
+	     i < offset + tbl_multiwriter_test_params.nb_tsx_insertion;
+	     i++) {
+		if (rte_hash_add_key(tbl_multiwriter_test_params.h,
+				     tbl_multiwriter_test_params.keys + i) < 0)
+			break;
+	}
+
+	cycles = rte_rdtsc_precise() - begin;
+	rte_atomic64_add(&gcycles, cycles);
+	rte_atomic64_add(&ginsertions, i - offset);
+
+	for (; i < offset + tbl_multiwriter_test_params.nb_tsx_insertion; i++)
+		tbl_multiwriter_test_params.keys[i]
+			= RTE_APP_TEST_HASH_MULTIWRITER_FAILED;
+
+	return 0;
+}
+
+
+static int
+test_hash_multiwriter(void)
+{
+	unsigned int i, rounded_nb_total_tsx_insertion;
+	static unsigned calledCount = 1;
+
+	uint32_t *keys;
+	uint32_t *found;
+
+	struct rte_hash_parameters hash_params = {
+		.entries = nb_entries,
+		.key_len = sizeof(uint32_t),
+		.hash_func = rte_hash_crc,
+		.hash_func_init_val = 0,
+		.socket_id = rte_socket_id(),
+	};
+	if (use_htm)
+		hash_params.extra_flag =
+			RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT
+				| RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD;
+	else
+		hash_params.extra_flag =
+			RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD;
+
+	struct rte_hash *handle;
+	char name[RTE_HASH_NAMESIZE];
+
+	const void *next_key;
+	void *next_data;
+	uint32_t iter = 0;
+
+	uint32_t duplicated_keys = 0;
+	uint32_t lost_keys = 0;
+
+	snprintf(name, 32, "test%u", calledCount++);
+	hash_params.name = name;
+
+	handle = rte_hash_create(&hash_params);
+	RETURN_IF_ERROR(handle == NULL, "hash creation failed");
+
+	tbl_multiwriter_test_params.h = handle;
+	tbl_multiwriter_test_params.nb_tsx_insertion =
+		nb_total_tsx_insertion / rte_lcore_count();
+
+	rounded_nb_total_tsx_insertion = (nb_total_tsx_insertion /
+		tbl_multiwriter_test_params.nb_tsx_insertion)
+		* tbl_multiwriter_test_params.nb_tsx_insertion;
+
+	rte_srand(rte_rdtsc());
+
+	keys = rte_malloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+
+	if (keys == NULL) {
+		printf("RTE_MALLOC failed\n");
+		goto err1;
+	}
+
+	found = rte_zmalloc(NULL, sizeof(uint32_t) * nb_entries, 0);
+	if (found == NULL) {
+		printf("RTE_ZMALLOC failed\n");
+		goto err2;
+	}
+
+	for (i = 0; i < nb_entries; i++)
+		keys[i] = i;
+
+	tbl_multiwriter_test_params.keys = keys;
+	tbl_multiwriter_test_params.found = found;
+
+	rte_atomic64_init(&gcycles);
+	rte_atomic64_clear(&gcycles);
+
+	rte_atomic64_init(&ginsertions);
+	rte_atomic64_clear(&ginsertions);
+
+	/* Fire all threads. */
+	rte_eal_mp_remote_launch(test_hash_multiwriter_worker,
+				 NULL, CALL_MASTER);
+	rte_eal_mp_wait_lcore();
+
+	while (rte_hash_iterate(handle, &next_key, &next_data, &iter) >= 0) {
+		/* Search for the key in the list of keys added .*/
+		i = *(const uint32_t *)next_key;
+		tbl_multiwriter_test_params.found[i]++;
+	}
+
+	for (i = 0; i < rounded_nb_total_tsx_insertion; i++) {
+		if (tbl_multiwriter_test_params.keys[i]
+		    != RTE_APP_TEST_HASH_MULTIWRITER_FAILED) {
+			if (tbl_multiwriter_test_params.found[i] > 1) {
+				duplicated_keys++;
+				break;
+			}
+			if (tbl_multiwriter_test_params.found[i] == 0) {
+				lost_keys++;
+				printf("key %d is lost\n", i);
+				break;
+			}
+		}
+	}
+
+	if (duplicated_keys > 0) {
+		printf("%d key duplicated\n", duplicated_keys);
+		goto err3;
+	}
+
+	if (lost_keys > 0) {
+		printf("%d key lost\n", lost_keys);
+		goto err3;
+	}
+
+	printf("No key corrupted during multiwriter insertion.\n");
+
+	unsigned long long int cycles_per_insertion =
+		rte_atomic64_read(&gcycles)/
+		rte_atomic64_read(&ginsertions);
+
+	printf(" cycles per insertion: %llu\n", cycles_per_insertion);
+
+	rte_free(tbl_multiwriter_test_params.found);
+	rte_free(tbl_multiwriter_test_params.keys);
+	rte_hash_free(handle);
+	return 0;
+
+err3:
+	rte_free(tbl_multiwriter_test_params.found);
+err2:
+	rte_free(tbl_multiwriter_test_params.keys);
+err1:
+	rte_hash_free(handle);
+	return -1;
+}
+
+static int
+test_hash_multiwriter_main(void)
+{
+	int r = -1;
+
+	if (rte_lcore_count() == 1) {
+		printf("More than one lcore is required to do multiwriter test\n");
+		return 0;
+	}
+
+
+	setlocale(LC_NUMERIC, "");
+
+
+	if (!rte_tm_supported()) {
+		printf("Hardware transactional memory (lock elision) "
+			"is NOT supported\n");
+	} else {
+		printf("Hardware transactional memory (lock elision) "
+			"is supported\n");
+
+		printf("Test multi-writer with Hardware transactional memory\n");
+
+		use_htm = 1;
+		r = test_hash_multiwriter();
+	}
+
+	printf("Test multi-writer without Hardware transactional memory\n");
+	use_htm = 0;
+	r = test_hash_multiwriter();
+
+	return r;
+}
+
+
+static struct test_command hash_scaling_cmd = {
+	.command = "hash_multiwriter_autotest",
+	.callback = test_hash_multiwriter_main,
+};
+
+REGISTER_TEST_COMMAND(hash_scaling_cmd);
diff --git a/doc/guides/rel_notes/release_16_07.rst b/doc/guides/rel_notes/release_16_07.rst
index ff8cc0d..dd96286 100644
--- a/doc/guides/rel_notes/release_16_07.rst
+++ b/doc/guides/rel_notes/release_16_07.rst
@@ -76,6 +76,18 @@ New Features
   monitoring applications, enabling the support of broader liveness
   reporting to external processes.
 
+* **Added multi-writer support for RTE Hash with Intel TSX.**
+
+  The following features/modifications have been added to rte_hash library:
+
+  * Enabled application developers to use an extra flag for rte_hash creation
+    to specify default behavior (multi-thread safe/unsafe) with rte_hash_add_key
+    function.
+  * Changed Cuckoo search algorithm to breadth first search for multi-writer
+    routine and split Cuckoo Search and Move operations in order to reduce
+    transactional code region and improve TSX performance.
+  * Added a hash multi-writer test case for test app.
+
 
 Resolved Issues
 ---------------
diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
index 7b7d1f8..e3cc3a7 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -59,12 +59,10 @@
 #include <rte_compat.h>
 
 #include "rte_hash.h"
-#if defined(RTE_ARCH_X86)
-#include "rte_cmp_x86.h"
-#endif
+#include "rte_cuckoo_hash.h"
 
-#if defined(RTE_ARCH_ARM64)
-#include "rte_cmp_arm64.h"
+#if defined(RTE_ARCH_X86)
+#include "rte_cuckoo_hash_x86.h"
 #endif
 
 TAILQ_HEAD(rte_hash_list, rte_tailq_entry);
@@ -74,153 +72,6 @@ static struct rte_tailq_elem rte_hash_tailq = {
 };
 EAL_REGISTER_TAILQ(rte_hash_tailq)
 
-/* Macro to enable/disable run-time checking of function parameters */
-#if defined(RTE_LIBRTE_HASH_DEBUG)
-#define RETURN_IF_TRUE(cond, retval) do { \
-	if (cond) \
-		return retval; \
-} while (0)
-#else
-#define RETURN_IF_TRUE(cond, retval)
-#endif
-
-/* Hash function used if none is specified */
-#if defined(RTE_MACHINE_CPUFLAG_SSE4_2) || defined(RTE_MACHINE_CPUFLAG_CRC32)
-#include <rte_hash_crc.h>
-#define DEFAULT_HASH_FUNC       rte_hash_crc
-#else
-#include <rte_jhash.h>
-#define DEFAULT_HASH_FUNC       rte_jhash
-#endif
-
-/** Number of items per bucket. */
-#define RTE_HASH_BUCKET_ENTRIES		4
-
-#define NULL_SIGNATURE			0
-
-#define KEY_ALIGNMENT			16
-
-#define LCORE_CACHE_SIZE		8
-
-#if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
-/*
- * All different options to select a key compare function,
- * based on the key size and custom function.
- */
-enum cmp_jump_table_case {
-	KEY_CUSTOM = 0,
-	KEY_16_BYTES,
-	KEY_32_BYTES,
-	KEY_48_BYTES,
-	KEY_64_BYTES,
-	KEY_80_BYTES,
-	KEY_96_BYTES,
-	KEY_112_BYTES,
-	KEY_128_BYTES,
-	KEY_OTHER_BYTES,
-	NUM_KEY_CMP_CASES,
-};
-
-/*
- * Table storing all different key compare functions
- * (multi-process supported)
- */
-const rte_hash_cmp_eq_t cmp_jump_table[NUM_KEY_CMP_CASES] = {
-	NULL,
-	rte_hash_k16_cmp_eq,
-	rte_hash_k32_cmp_eq,
-	rte_hash_k48_cmp_eq,
-	rte_hash_k64_cmp_eq,
-	rte_hash_k80_cmp_eq,
-	rte_hash_k96_cmp_eq,
-	rte_hash_k112_cmp_eq,
-	rte_hash_k128_cmp_eq,
-	memcmp
-};
-#else
-/*
- * All different options to select a key compare function,
- * based on the key size and custom function.
- */
-enum cmp_jump_table_case {
-	KEY_CUSTOM = 0,
-	KEY_OTHER_BYTES,
-	NUM_KEY_CMP_CASES,
-};
-
-/*
- * Table storing all different key compare functions
- * (multi-process supported)
- */
-const rte_hash_cmp_eq_t cmp_jump_table[NUM_KEY_CMP_CASES] = {
-	NULL,
-	memcmp
-};
-
-#endif
-
-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. */
-	uint32_t entries;               /**< Total table entries. */
-	uint32_t num_buckets;           /**< Number of buckets in table. */
-	uint32_t key_len;               /**< Length of hash key. */
-	rte_hash_function hash_func;    /**< Function used to calculate hash. */
-	uint32_t hash_func_init_val;    /**< Init value used by hash_func. */
-	rte_hash_cmp_eq_t rte_hash_custom_cmp_eq;
-	/**< Custom function used to compare keys. */
-	enum cmp_jump_table_case cmp_jump_table_idx;
-	/**< Indicates which compare function to use. */
-	uint32_t bucket_bitmask;        /**< Bitmask for getting bucket index
-						from hash signature. */
-	uint32_t key_entry_size;         /**< Size of each key entry. */
-
-	struct rte_ring *free_slots;    /**< Ring that stores all indexes
-						of the free slots in the key table */
-	void *key_store;                /**< Table storing all keys and data */
-	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 */
-struct rte_hash_signatures {
-	union {
-		struct {
-			hash_sig_t current;
-			hash_sig_t alt;
-		};
-		uint64_t sig;
-	};
-};
-
-/* Structure that stores key-value pair */
-struct rte_hash_key {
-	union {
-		uintptr_t idata;
-		void *pdata;
-	};
-	/* Variable key size */
-	char key[0];
-} __attribute__((aligned(KEY_ALIGNMENT)));
-
-/** Bucket structure */
-struct rte_hash_bucket {
-	struct rte_hash_signatures signatures[RTE_HASH_BUCKET_ENTRIES];
-	/* Includes dummy key index that always contains index 0 */
-	uint32_t key_idx[RTE_HASH_BUCKET_ENTRIES + 1];
-	uint8_t flag[RTE_HASH_BUCKET_ENTRIES];
-} __rte_cache_aligned;
-
 struct rte_hash *
 rte_hash_find_existing(const char *name)
 {
@@ -372,7 +223,7 @@ rte_hash_create(const struct rte_hash_parameters *params)
 
 /*
  * If x86 architecture is used, select appropriate compare function,
- * which may use x86 instrinsics, otherwise use memcmp
+ * which may use x86 intrinsics, otherwise use memcmp
  */
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 	/* Select function to compare keys */
@@ -431,7 +282,23 @@ rte_hash_create(const struct rte_hash_parameters *params)
 	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 */
+	/* Turn on multi-writer only with explicit flat from user and TM
+	 * support.
+	 */
+	if (params->extra_flag & RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD) {
+		if (h->hw_trans_mem_support) {
+			h->add_key = ADD_KEY_MULTIWRITER_TM;
+		} else {
+			h->add_key = ADD_KEY_MULTIWRITER;
+			h->multiwriter_lock = rte_malloc(NULL,
+							sizeof(rte_spinlock_t),
+							LCORE_CACHE_SIZE);
+			rte_spinlock_init(h->multiwriter_lock);
+		}
+	} else
+		h->add_key = ADD_KEY_SINGLEWRITER;
+
+	/* Populate free slots ring. Entry zero is reserved for key misses. */
 	for (i = 1; i < params->entries + 1; i++)
 		rte_ring_sp_enqueue(r, (void *)((uintptr_t) i));
 
@@ -482,6 +349,8 @@ rte_hash_free(struct rte_hash *h)
 	if (h->hw_trans_mem_support)
 		rte_free(h->local_free_slots);
 
+	if (h->add_key == ADD_KEY_MULTIWRITER)
+		rte_free(h->multiwriter_lock);
 	rte_ring_free(h->free_slots);
 	rte_free(h->key_store);
 	rte_free(h->buckets);
@@ -632,6 +501,9 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	unsigned lcore_id;
 	struct lcore_cache *cached_free_slots = NULL;
 
+	if (h->add_key == ADD_KEY_MULTIWRITER)
+		rte_spinlock_lock(h->multiwriter_lock);
+
 	prim_bucket_idx = sig & h->bucket_bitmask;
 	prim_bkt = &h->buckets[prim_bucket_idx];
 	rte_prefetch0(prim_bkt);
@@ -712,35 +584,67 @@ __rte_hash_add_key_with_hash(const struct rte_hash *h, const void *key,
 	rte_memcpy(new_k->key, key, h->key_len);
 	new_k->pdata = data;
 
-	/* Insert new entry is there is room in the primary bucket */
-	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
-		/* Check if slot is available */
-		if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
-			prim_bkt->signatures[i].current = sig;
-			prim_bkt->signatures[i].alt = alt_hash;
-			prim_bkt->key_idx[i] = new_idx;
+#if defined(RTE_ARCH_X86) /* currently only x86 support HTM */
+	if (h->add_key == ADD_KEY_MULTIWRITER_TM) {
+		ret = rte_hash_cuckoo_insert_mw_tm(prim_bkt,
+				sig, alt_hash, new_idx);
+		if (ret >= 0)
+			return new_idx - 1;
+
+		/* Primary bucket full, need to make space for new entry */
+		ret = rte_hash_cuckoo_make_space_mw_tm(h, prim_bkt, sig,
+							alt_hash, new_idx);
+
+		if (ret >= 0)
+			return new_idx - 1;
+
+		/* Also search secondary bucket to get better occupancy */
+		ret = rte_hash_cuckoo_make_space_mw_tm(h, sec_bkt, sig,
+							alt_hash, new_idx);
+
+		if (ret >= 0)
+			return new_idx - 1;
+	} else {
+#endif
+		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+			/* Check if slot is available */
+			if (likely(prim_bkt->signatures[i].sig == NULL_SIGNATURE)) {
+				prim_bkt->signatures[i].current = sig;
+				prim_bkt->signatures[i].alt = alt_hash;
+				prim_bkt->key_idx[i] = new_idx;
+				break;
+			}
+		}
+
+		if (i != RTE_HASH_BUCKET_ENTRIES) {
+			if (h->add_key == ADD_KEY_MULTIWRITER)
+				rte_spinlock_unlock(h->multiwriter_lock);
 			return new_idx - 1;
 		}
-	}
 
-	/* Primary bucket is full, so we need to make space for new entry */
-	ret = make_space_bucket(h, prim_bkt);
-	/*
-	 * After recursive function.
-	 * Insert the new entry in the position of the pushed entry
-	 * if successful or return error and
-	 * store the new slot back in the ring
-	 */
-	if (ret >= 0) {
-		prim_bkt->signatures[ret].current = sig;
-		prim_bkt->signatures[ret].alt = alt_hash;
-		prim_bkt->key_idx[ret] = new_idx;
-		return new_idx - 1;
+		/* Primary bucket full, need to make space for new entry
+		 * After recursive function.
+		 * Insert the new entry in the position of the pushed entry
+		 * if successful or return error and
+		 * store the new slot back in the ring
+		 */
+		ret = make_space_bucket(h, prim_bkt);
+		if (ret >= 0) {
+			prim_bkt->signatures[ret].current = sig;
+			prim_bkt->signatures[ret].alt = alt_hash;
+			prim_bkt->key_idx[ret] = new_idx;
+			if (h->add_key == ADD_KEY_MULTIWRITER)
+				rte_spinlock_unlock(h->multiwriter_lock);
+			return new_idx - 1;
+		}
+#if defined(RTE_ARCH_X86)
 	}
-
+#endif
 	/* Error in addition, store new slot back in the ring and return error */
 	enqueue_slot_back(h, cached_free_slots, (void *)((uintptr_t) new_idx));
 
+	if (h->add_key == ADD_KEY_MULTIWRITER)
+		rte_spinlock_unlock(h->multiwriter_lock);
 	return ret;
 }
 
diff --git a/lib/librte_hash/rte_cuckoo_hash.h b/lib/librte_hash/rte_cuckoo_hash.h
new file mode 100644
index 0000000..6c76700
--- /dev/null
+++ b/lib/librte_hash/rte_cuckoo_hash.h
@@ -0,0 +1,219 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* rte_cuckoo_hash.h
+ * This file hold Cuckoo Hash private data structures to allows include from
+ * platform specific files like rte_cuckoo_hash_x86.h
+ */
+
+#ifndef _RTE_CUCKOO_HASH_H_
+#define _RTE_CUCKOO_HASH_H_
+
+#if defined(RTE_ARCH_X86)
+#include "rte_cmp_x86.h"
+#endif
+
+#if defined(RTE_ARCH_ARM64)
+#include "rte_cmp_arm64.h"
+#endif
+
+/* Macro to enable/disable run-time checking of function parameters */
+#if defined(RTE_LIBRTE_HASH_DEBUG)
+#define RETURN_IF_TRUE(cond, retval) do { \
+	if (cond) \
+		return retval; \
+} while (0)
+#else
+#define RETURN_IF_TRUE(cond, retval)
+#endif
+
+/* Hash function used if none is specified */
+#if defined(RTE_MACHINE_CPUFLAG_SSE4_2) || defined(RTE_MACHINE_CPUFLAG_CRC32)
+#include <rte_hash_crc.h>
+#define DEFAULT_HASH_FUNC       rte_hash_crc
+#else
+#include <rte_jhash.h>
+#define DEFAULT_HASH_FUNC       rte_jhash
+#endif
+
+#if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
+/*
+ * All different options to select a key compare function,
+ * based on the key size and custom function.
+ */
+enum cmp_jump_table_case {
+	KEY_CUSTOM = 0,
+	KEY_16_BYTES,
+	KEY_32_BYTES,
+	KEY_48_BYTES,
+	KEY_64_BYTES,
+	KEY_80_BYTES,
+	KEY_96_BYTES,
+	KEY_112_BYTES,
+	KEY_128_BYTES,
+	KEY_OTHER_BYTES,
+	NUM_KEY_CMP_CASES,
+};
+
+/*
+ * Table storing all different key compare functions
+ * (multi-process supported)
+ */
+const rte_hash_cmp_eq_t cmp_jump_table[NUM_KEY_CMP_CASES] = {
+	NULL,
+	rte_hash_k16_cmp_eq,
+	rte_hash_k32_cmp_eq,
+	rte_hash_k48_cmp_eq,
+	rte_hash_k64_cmp_eq,
+	rte_hash_k80_cmp_eq,
+	rte_hash_k96_cmp_eq,
+	rte_hash_k112_cmp_eq,
+	rte_hash_k128_cmp_eq,
+	memcmp
+};
+#else
+/*
+ * All different options to select a key compare function,
+ * based on the key size and custom function.
+ */
+enum cmp_jump_table_case {
+	KEY_CUSTOM = 0,
+	KEY_OTHER_BYTES,
+	NUM_KEY_CMP_CASES,
+};
+
+/*
+ * Table storing all different key compare functions
+ * (multi-process supported)
+ */
+const rte_hash_cmp_eq_t cmp_jump_table[NUM_KEY_CMP_CASES] = {
+	NULL,
+	memcmp
+};
+
+#endif
+
+enum add_key_case {
+	ADD_KEY_SINGLEWRITER = 0,
+	ADD_KEY_MULTIWRITER,
+	ADD_KEY_MULTIWRITER_TM,
+};
+
+/** Number of items per bucket. */
+#define RTE_HASH_BUCKET_ENTRIES		4
+
+#define NULL_SIGNATURE			0
+
+#define KEY_ALIGNMENT			16
+
+#define LCORE_CACHE_SIZE		64
+
+#define RTE_HASH_BFS_QUEUE_MAX_LEN       1000
+
+#define RTE_XABORT_CUCKOO_PATH_INVALIDED 0x4
+
+#define RTE_HASH_TSX_MAX_RETRY  10
+
+struct lcore_cache {
+	unsigned len; /**< Cache len */
+	void *objs[LCORE_CACHE_SIZE]; /**< Cache objects */
+} __rte_cache_aligned;
+
+/* Structure storing both primary and secondary hashes */
+struct rte_hash_signatures {
+	union {
+		struct {
+			hash_sig_t current;
+			hash_sig_t alt;
+		};
+		uint64_t sig;
+	};
+};
+
+/* Structure that stores key-value pair */
+struct rte_hash_key {
+	union {
+		uintptr_t idata;
+		void *pdata;
+	};
+	/* Variable key size */
+	char key[0];
+} __attribute__((aligned(KEY_ALIGNMENT)));
+
+/** Bucket structure */
+struct rte_hash_bucket {
+	struct rte_hash_signatures signatures[RTE_HASH_BUCKET_ENTRIES];
+	/* Includes dummy key index that always contains index 0 */
+	uint32_t key_idx[RTE_HASH_BUCKET_ENTRIES + 1];
+	uint8_t flag[RTE_HASH_BUCKET_ENTRIES];
+} __rte_cache_aligned;
+
+/** A hash table structure. */
+struct rte_hash {
+	char name[RTE_HASH_NAMESIZE];   /**< Name of the hash. */
+	uint32_t entries;               /**< Total table entries. */
+	uint32_t num_buckets;           /**< Number of buckets in table. */
+	uint32_t key_len;               /**< Length of hash key. */
+	rte_hash_function hash_func;    /**< Function used to calculate hash. */
+	uint32_t hash_func_init_val;    /**< Init value used by hash_func. */
+	rte_hash_cmp_eq_t rte_hash_custom_cmp_eq;
+	/**< Custom function used to compare keys. */
+	enum cmp_jump_table_case cmp_jump_table_idx;
+	/**< Indicates which compare function to use. */
+	uint32_t bucket_bitmask;        /**< Bitmask for getting bucket index
+						from hash signature. */
+	uint32_t key_entry_size;         /**< Size of each key entry. */
+
+	struct rte_ring *free_slots;    /**< Ring that stores all indexes
+						of the free slots in the key table */
+	void *key_store;                /**< Table storing all keys and data */
+	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 */
+	enum add_key_case add_key; /**< Multi-writer hash add behavior */
+
+	rte_spinlock_t *multiwriter_lock; /**< Multi-writer spinlock for w/o TM */
+} __rte_cache_aligned;
+
+struct queue_node {
+	struct rte_hash_bucket *bkt; /* Current bucket on the bfs search */
+
+	struct queue_node *prev;     /* Parent(bucket) in search path */
+	int prev_slot;               /* Parent(slot) in search path */
+};
+
+#endif
diff --git a/lib/librte_hash/rte_cuckoo_hash_x86.h b/lib/librte_hash/rte_cuckoo_hash_x86.h
new file mode 100644
index 0000000..fa5630b
--- /dev/null
+++ b/lib/librte_hash/rte_cuckoo_hash_x86.h
@@ -0,0 +1,193 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2016 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* rte_cuckoo_hash_x86.h
+ * This file holds all x86 specific Cuckoo Hash functions
+ */
+
+/* Only tries to insert at one bucket (@prim_bkt) without trying to push
+ * buckets around
+ */
+static inline unsigned
+rte_hash_cuckoo_insert_mw_tm(struct rte_hash_bucket *prim_bkt,
+		hash_sig_t sig, hash_sig_t alt_hash, uint32_t new_idx)
+{
+	unsigned i, status;
+	unsigned try = 0;
+
+	while (try < RTE_HASH_TSX_MAX_RETRY) {
+		status = rte_xbegin();
+		if (likely(status == RTE_XBEGIN_STARTED)) {
+			/* Insert new entry if there is room in the primary
+			* bucket.
+			*/
+			for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+				/* Check if slot is available */
+				if (likely(prim_bkt->signatures[i].sig ==
+						NULL_SIGNATURE)) {
+					prim_bkt->signatures[i].current = sig;
+					prim_bkt->signatures[i].alt = alt_hash;
+					prim_bkt->key_idx[i] = new_idx;
+					break;
+				}
+			}
+			rte_xend();
+
+			if (i != RTE_HASH_BUCKET_ENTRIES)
+				return 0;
+
+			break; /* break off try loop if transaction commits */
+		} else {
+			/* If we abort we give up this cuckoo path. */
+			try++;
+			rte_pause();
+		}
+	}
+
+	return -1;
+}
+
+/* Shift buckets along provided cuckoo_path (@leaf and @leaf_slot) and fill
+ * the path head with new entry (sig, alt_hash, new_idx)
+ */
+static inline int
+rte_hash_cuckoo_move_insert_mw_tm(const struct rte_hash *h,
+			struct queue_node *leaf, uint32_t leaf_slot,
+			hash_sig_t sig, hash_sig_t alt_hash, uint32_t new_idx)
+{
+	unsigned try = 0;
+	unsigned status;
+	uint32_t prev_alt_bkt_idx;
+
+	struct queue_node *prev_node, *curr_node = leaf;
+	struct rte_hash_bucket *prev_bkt, *curr_bkt = leaf->bkt;
+	uint32_t prev_slot, curr_slot = leaf_slot;
+
+	while (try < RTE_HASH_TSX_MAX_RETRY) {
+		status = rte_xbegin();
+		if (likely(status == RTE_XBEGIN_STARTED)) {
+			while (likely(curr_node->prev != NULL)) {
+				prev_node = curr_node->prev;
+				prev_bkt = prev_node->bkt;
+				prev_slot = curr_node->prev_slot;
+
+				prev_alt_bkt_idx
+					= prev_bkt->signatures[prev_slot].alt
+					    & h->bucket_bitmask;
+
+				if (unlikely(&h->buckets[prev_alt_bkt_idx]
+					     != curr_bkt)) {
+					rte_xabort(RTE_XABORT_CUCKOO_PATH_INVALIDED);
+				}
+
+				/* Need to swap current/alt sig to allow later
+				 * Cuckoo insert to move elements back to its
+				 * primary bucket if available
+				 */
+				curr_bkt->signatures[curr_slot].alt =
+				    prev_bkt->signatures[prev_slot].current;
+				curr_bkt->signatures[curr_slot].current =
+				    prev_bkt->signatures[prev_slot].alt;
+				curr_bkt->key_idx[curr_slot]
+				    = prev_bkt->key_idx[prev_slot];
+
+				curr_slot = prev_slot;
+				curr_node = prev_node;
+				curr_bkt = curr_node->bkt;
+			}
+
+			curr_bkt->signatures[curr_slot].current = sig;
+			curr_bkt->signatures[curr_slot].alt = alt_hash;
+			curr_bkt->key_idx[curr_slot] = new_idx;
+
+			rte_xend();
+
+			return 0;
+		}
+
+		/* If we abort we give up this cuckoo path, since most likely it's
+		 * no longer valid as TSX detected data conflict
+		 */
+		try++;
+		rte_pause();
+	}
+
+	return -1;
+}
+
+/*
+ * Make space for new key, using bfs Cuckoo Search and Multi-Writer safe
+ * Cuckoo
+ */
+static inline int
+rte_hash_cuckoo_make_space_mw_tm(const struct rte_hash *h,
+			struct rte_hash_bucket *bkt,
+			hash_sig_t sig, hash_sig_t alt_hash,
+			uint32_t new_idx)
+{
+	unsigned i;
+	struct queue_node queue[RTE_HASH_BFS_QUEUE_MAX_LEN];
+	struct queue_node *tail, *head;
+	struct rte_hash_bucket *curr_bkt, *alt_bkt;
+
+	tail = queue;
+	head = queue + 1;
+	tail->bkt = bkt;
+	tail->prev = NULL;
+	tail->prev_slot = -1;
+
+	/* Cuckoo bfs Search */
+	while (likely(tail != head && head <
+					queue + RTE_HASH_BFS_QUEUE_MAX_LEN - 4)) {
+		curr_bkt = tail->bkt;
+		for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
+			if (curr_bkt->signatures[i].sig == NULL_SIGNATURE) {
+				if (likely(rte_hash_cuckoo_move_insert_mw_tm(h,
+						tail, i, sig,
+						alt_hash, new_idx) == 0))
+					return 0;
+			}
+
+			/* Enqueue new node and keep prev node info */
+			alt_bkt = &(h->buckets[curr_bkt->signatures[i].alt
+						    & h->bucket_bitmask]);
+			head->bkt = alt_bkt;
+			head->prev = tail;
+			head->prev_slot = i;
+			head++;
+		}
+		tail++;
+	}
+
+	return -ENOSPC;
+}
diff --git a/lib/librte_hash/rte_hash.h b/lib/librte_hash/rte_hash.h
index 724315a..c9612fb 100644
--- a/lib/librte_hash/rte_hash.h
+++ b/lib/librte_hash/rte_hash.h
@@ -60,6 +60,9 @@ extern "C" {
 /** Enable Hardware transactional memory support. */
 #define RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT	0x01
 
+/** Default behavior of insertion, single writer/multi writer */
+#define RTE_HASH_EXTRA_FLAGS_MULTI_WRITER_ADD 0x02
+
 /** Signature of key that is stored internally. */
 typedef uint32_t hash_sig_t;
 
-- 
2.5.5

Re: [PATCH v3] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[De Lara Guarch, Pablo]

> -----Original Message-----
> From: Shen, Wei1
> Sent: Thursday, June 16, 2016 11:14 PM
> To: dev@dpdk.org
> Cc: De Lara Guarch, Pablo; Ananyev, Konstantin;
> stephen@networkplumber.org; Tai, Charlie; Maciocco, Christian; Gobriel,
> Sameh; Shen, Wei1
> Subject: [PATCH v3] rte_hash: add scalable multi-writer insertion w/ Intel TSX
> 
> This patch introduced scalable multi-writer Cuckoo Hash insertion
> based on a split Cuckoo Search and Move operation using Intel
> TSX. It can do scalable hash insertion with 22 cores with little
> performance loss and negligible TSX abortion rate.
> 
> * Added an extra rte_hash flag definition to switch default single writer
>   Cuckoo Hash behavior to multiwriter.
>     - If HTM is available, it would use hardware feature for concurrency.
>     - If HTM is not available, it would fall back to spinlock.
> 
> * Created a rte_cuckoo_hash_x86.h file to hold all x86-arch related
>   cuckoo_hash functions. And rte_cuckoo_hash.c uses compile time flag to
>   select x86 file or other platform-specific implementations. While HTM check
>   is still done at runtime (same idea with
>   RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT)
> 
> * Moved rte_hash private struct definitions to rte_cuckoo_hash.h, to allow
>   rte_cuckoo_hash_x86.h or future platform dependent functions to include.
> 
> * Following new functions are created for consistent names when new
> platform
>   TM support are added.
>     - rte_hash_cuckoo_move_insert_mw_tm: do insertion with bucket
> movement.
>     - rte_hash_cuckoo_insert_mw_tm: do insertion without bucket movement.
> 
> * One extra multi-writer test case is added.
> 
> Signed-off-by: Shen Wei <wei1.shen@intel.com>
> Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>

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

Re: [PATCH v3] rte_hash: add scalable multi-writer insertion w/ Intel TSX

[Thomas Monjalon]

> > This patch introduced scalable multi-writer Cuckoo Hash insertion
> > based on a split Cuckoo Search and Move operation using Intel
> > TSX. It can do scalable hash insertion with 22 cores with little
> > performance loss and negligible TSX abortion rate.
> > 
> > * Added an extra rte_hash flag definition to switch default single writer
> >   Cuckoo Hash behavior to multiwriter.
> >     - If HTM is available, it would use hardware feature for concurrency.
> >     - If HTM is not available, it would fall back to spinlock.
> > 
> > * Created a rte_cuckoo_hash_x86.h file to hold all x86-arch related
> >   cuckoo_hash functions. And rte_cuckoo_hash.c uses compile time flag to
> >   select x86 file or other platform-specific implementations. While HTM check
> >   is still done at runtime (same idea with
> >   RTE_HASH_EXTRA_FLAGS_TRANS_MEM_SUPPORT)
> > 
> > * Moved rte_hash private struct definitions to rte_cuckoo_hash.h, to allow
> >   rte_cuckoo_hash_x86.h or future platform dependent functions to include.
> > 
> > * Following new functions are created for consistent names when new
> > platform
> >   TM support are added.
> >     - rte_hash_cuckoo_move_insert_mw_tm: do insertion with bucket
> > movement.
> >     - rte_hash_cuckoo_insert_mw_tm: do insertion without bucket movement.
> > 
> > * One extra multi-writer test case is added.
> > 
> > Signed-off-by: Shen Wei <wei1.shen@intel.com>
> > Signed-off-by: Sameh Gobriel <sameh.gobriel@intel.com>
> 
> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>

Applied, thanks