[v2] kni: dynamically allocate memory for each KNI

Message ID 20180923191202.64896-1-iryzhov@nfware.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers
Series [v2] kni: dynamically allocate memory for each KNI |

Checks

Context Check Description
ci/checkpatch warning coding style issues
ci/Intel-compilation success Compilation OK

Commit Message

Igor Ryzhov Sept. 23, 2018, 7:12 p.m. UTC
  Long time ago preallocation of memory for KNI was introduced in commit
0c6bc8e. It was done because of lack of ability to free previously
allocated memzones, which led to memzone exhaustion. Currently memzones
can be freed and this patch uses this ability for dynamic KNI memory
allocation.

Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
---
v2:
 * allocate KNI using rte_zmalloc
 * swap reserve/release functions
 * use "kni" as a variable name
 * use macros for memzone names

 lib/librte_kni/rte_kni.c | 502 +++++++++++++++++----------------------
 lib/librte_kni/rte_kni.h |   6 +-
 test/test/test_kni.c     |   6 -
 3 files changed, 218 insertions(+), 296 deletions(-)
  

Comments

Ferruh Yigit Sept. 26, 2018, 10:41 a.m. UTC | #1
On 9/23/2018 8:12 PM, Igor Ryzhov wrote:
> Long time ago preallocation of memory for KNI was introduced in commit
> 0c6bc8e. It was done because of lack of ability to free previously
> allocated memzones, which led to memzone exhaustion. Currently memzones
> can be freed and this patch uses this ability for dynamic KNI memory
> allocation. 

Hi Igor,

Good cleanup, thanks.
+1 to eal_tailq for ctx

A few minor comments below, but they are not significant enough to block the
patch, please let us know if you don't have bandwidth for a new version.

> Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
<...>

> @@ -294,41 +180,52 @@ rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
>  {
>  	int ret;
>  	struct rte_kni_device_info dev_info;
> -	struct rte_kni *ctx;
> -	char intf_name[RTE_KNI_NAMESIZE];
> -	const struct rte_memzone *mz;
> -	struct rte_kni_memzone_slot *slot = NULL;
> +	struct rte_kni *kni;
> +	struct rte_tailq_entry *te = NULL;
> +	struct rte_kni_list *kni_list;
> +
> +	kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);

Can you move this below input validation, no need this assignment if API will
fail because of wrong input.

>  	if (!pktmbuf_pool || !conf || !conf->name[0])
>  		return NULL;
>  
>  	/* Check if KNI subsystem has been initialized */
> -	if (kni_memzone_pool.initialized != 1) {
> +	if (kni_fd < 0) {
>  		RTE_LOG(ERR, KNI, "KNI subsystem has not been initialized. Invoke rte_kni_init() first\n");
>  		return NULL;
>  	}
>  
> -	/* Get an available slot from the pool */
> -	slot = kni_memzone_pool_alloc();
> -	if (!slot) {
> -		RTE_LOG(ERR, KNI, "Cannot allocate more KNI interfaces; increase the number of max_kni_ifaces(current %d) or release unused ones.\n",
> -			kni_memzone_pool.max_ifaces);
> -		return NULL;
> +	rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
> +
> +	TAILQ_FOREACH(te, kni_list, next) {
> +		kni = (struct rte_kni *) te->data;
> +		if (strncmp(conf->name, kni->name, RTE_KNI_NAMESIZE) == 0)
> +			break;
>  	}

This is rte_kni_get(), why not reuse it. You can create an version of it without
lock.

like _rte_kni_get() which you can call here.

And
rte_kni_get()
	rte_rwlock_read_lock(RTE_EAL_TAILQ_RWLOCK);
	_rte_kni_get()
	rte_rwlock_read_unlock(RTE_EAL_TAILQ_RWLOCK);

<...>

> +
> +	if (te == NULL) {
> +		goto unlock;
> +	}

No need {} for single line.
One more below.
  
Igor Ryzhov Sept. 26, 2018, 1:31 p.m. UTC | #2
Hi Ferruh,

Will fix it today.

Igor

On Wed, Sep 26, 2018 at 1:41 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote:

> On 9/23/2018 8:12 PM, Igor Ryzhov wrote:
> > Long time ago preallocation of memory for KNI was introduced in commit
> > 0c6bc8e. It was done because of lack of ability to free previously
> > allocated memzones, which led to memzone exhaustion. Currently memzones
> > can be freed and this patch uses this ability for dynamic KNI memory
> > allocation.
>
> Hi Igor,
>
> Good cleanup, thanks.
> +1 to eal_tailq for ctx
>
> A few minor comments below, but they are not significant enough to block
> the
> patch, please let us know if you don't have bandwidth for a new version.
>
> > Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
> <...>
>
> > @@ -294,41 +180,52 @@ rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
> >  {
> >       int ret;
> >       struct rte_kni_device_info dev_info;
> > -     struct rte_kni *ctx;
> > -     char intf_name[RTE_KNI_NAMESIZE];
> > -     const struct rte_memzone *mz;
> > -     struct rte_kni_memzone_slot *slot = NULL;
> > +     struct rte_kni *kni;
> > +     struct rte_tailq_entry *te = NULL;
> > +     struct rte_kni_list *kni_list;
> > +
> > +     kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
>
> Can you move this below input validation, no need this assignment if API
> will
> fail because of wrong input.
>
> >       if (!pktmbuf_pool || !conf || !conf->name[0])
> >               return NULL;
> >
> >       /* Check if KNI subsystem has been initialized */
> > -     if (kni_memzone_pool.initialized != 1) {
> > +     if (kni_fd < 0) {
> >               RTE_LOG(ERR, KNI, "KNI subsystem has not been initialized.
> Invoke rte_kni_init() first\n");
> >               return NULL;
> >       }
> >
> > -     /* Get an available slot from the pool */
> > -     slot = kni_memzone_pool_alloc();
> > -     if (!slot) {
> > -             RTE_LOG(ERR, KNI, "Cannot allocate more KNI interfaces;
> increase the number of max_kni_ifaces(current %d) or release unused
> ones.\n",
> > -                     kni_memzone_pool.max_ifaces);
> > -             return NULL;
> > +     rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
> > +
> > +     TAILQ_FOREACH(te, kni_list, next) {
> > +             kni = (struct rte_kni *) te->data;
> > +             if (strncmp(conf->name, kni->name, RTE_KNI_NAMESIZE) == 0)
> > +                     break;
> >       }
>
> This is rte_kni_get(), why not reuse it. You can create an version of it
> without
> lock.
>
> like _rte_kni_get() which you can call here.
>
> And
> rte_kni_get()
>         rte_rwlock_read_lock(RTE_EAL_TAILQ_RWLOCK);
>         _rte_kni_get()
>         rte_rwlock_read_unlock(RTE_EAL_TAILQ_RWLOCK);
>
> <...>
>
> > +
> > +     if (te == NULL) {
> > +             goto unlock;
> > +     }
>
> No need {} for single line.
> One more below.
>
  

Patch

diff --git a/lib/librte_kni/rte_kni.c b/lib/librte_kni/rte_kni.c
index 8a8f6c1cc..6af6e7efe 100644
--- a/lib/librte_kni/rte_kni.c
+++ b/lib/librte_kni/rte_kni.c
@@ -18,6 +18,9 @@ 
 #include <rte_log.h>
 #include <rte_kni.h>
 #include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_rwlock.h>
+#include <rte_eal_memconfig.h>
 #include <exec-env/rte_kni_common.h>
 #include "rte_kni_fifo.h"
 
@@ -30,7 +33,23 @@ 
 
 #define KNI_REQUEST_MBUF_NUM_MAX      32
 
-#define KNI_MEM_CHECK(cond) do { if (cond) goto kni_fail; } while (0)
+#define KNI_MEM_CHECK(cond, fail) do { if (cond) goto fail; } while (0)
+
+#define KNI_MZ_NAME_FMT			"kni_info_%s"
+#define KNI_TX_Q_MZ_NAME_FMT		"kni_tx_%s"
+#define KNI_RX_Q_MZ_NAME_FMT		"kni_rx_%s"
+#define KNI_ALLOC_Q_MZ_NAME_FMT		"kni_alloc_%s"
+#define KNI_FREE_Q_MZ_NAME_FMT		"kni_free_%s"
+#define KNI_REQ_Q_MZ_NAME_FMT		"kni_req_%s"
+#define KNI_RESP_Q_MZ_NAME_FMT		"kni_resp_%s"
+#define KNI_SYNC_ADDR_MZ_NAME_FMT	"kni_sync_%s"
+
+TAILQ_HEAD(rte_kni_list, rte_tailq_entry);
+
+static struct rte_tailq_elem rte_kni_tailq = {
+	.name = "RTE_KNI",
+};
+EAL_REGISTER_TAILQ(rte_kni_tailq)
 
 /**
  * KNI context
@@ -42,18 +61,26 @@  struct rte_kni {
 	struct rte_mempool *pktmbuf_pool;   /**< pkt mbuf mempool */
 	unsigned mbuf_size;                 /**< mbuf size */
 
+	const struct rte_memzone *m_tx_q;   /**< TX queue memzone */
+	const struct rte_memzone *m_rx_q;   /**< RX queue memzone */
+	const struct rte_memzone *m_alloc_q;/**< Alloc queue memzone */
+	const struct rte_memzone *m_free_q; /**< Free queue memzone */
+
 	struct rte_kni_fifo *tx_q;          /**< TX queue */
 	struct rte_kni_fifo *rx_q;          /**< RX queue */
 	struct rte_kni_fifo *alloc_q;       /**< Allocated mbufs queue */
 	struct rte_kni_fifo *free_q;        /**< To be freed mbufs queue */
 
+	const struct rte_memzone *m_req_q;  /**< Request queue memzone */
+	const struct rte_memzone *m_resp_q; /**< Response queue memzone */
+	const struct rte_memzone *m_sync_addr;/**< Sync addr memzone */
+
 	/* For request & response */
 	struct rte_kni_fifo *req_q;         /**< Request queue */
 	struct rte_kni_fifo *resp_q;        /**< Response queue */
 	void * sync_addr;                   /**< Req/Resp Mem address */
 
 	struct rte_kni_ops ops;             /**< operations for request */
-	uint8_t in_use : 1;                 /**< kni in use */
 };
 
 enum kni_ops_status {
@@ -61,232 +88,91 @@  enum kni_ops_status {
 	KNI_REQ_REGISTERED,
 };
 
-/**
- * KNI memzone pool slot
- */
-struct rte_kni_memzone_slot {
-	uint32_t id;
-	uint8_t in_use : 1;                    /**< slot in use */
-
-	/* Memzones */
-	const struct rte_memzone *m_ctx;       /**< KNI ctx */
-	const struct rte_memzone *m_tx_q;      /**< TX queue */
-	const struct rte_memzone *m_rx_q;      /**< RX queue */
-	const struct rte_memzone *m_alloc_q;   /**< Allocated mbufs queue */
-	const struct rte_memzone *m_free_q;    /**< To be freed mbufs queue */
-	const struct rte_memzone *m_req_q;     /**< Request queue */
-	const struct rte_memzone *m_resp_q;    /**< Response queue */
-	const struct rte_memzone *m_sync_addr;
-
-	/* Free linked list */
-	struct rte_kni_memzone_slot *next;     /**< Next slot link.list */
-};
-
-/**
- * KNI memzone pool
- */
-struct rte_kni_memzone_pool {
-	uint8_t initialized : 1;            /**< Global KNI pool init flag */
-
-	uint32_t max_ifaces;                /**< Max. num of KNI ifaces */
-	struct rte_kni_memzone_slot *slots;        /**< Pool slots */
-	rte_spinlock_t mutex;               /**< alloc/release mutex */
-
-	/* Free memzone slots linked-list */
-	struct rte_kni_memzone_slot *free;         /**< First empty slot */
-	struct rte_kni_memzone_slot *free_tail;    /**< Last empty slot */
-};
-
-
 static void kni_free_mbufs(struct rte_kni *kni);
 static void kni_allocate_mbufs(struct rte_kni *kni);
 
 static volatile int kni_fd = -1;
-static struct rte_kni_memzone_pool kni_memzone_pool = {
-	.initialized = 0,
-};
 
-static const struct rte_memzone *
-kni_memzone_reserve(const char *name, size_t len, int socket_id,
-						unsigned flags)
+/* Shall be called before any allocation happens */
+int
+rte_kni_init(unsigned int max_kni_ifaces __rte_unused)
 {
-	const struct rte_memzone *mz = rte_memzone_lookup(name);
-
-	if (mz == NULL)
-		mz = rte_memzone_reserve(name, len, socket_id, flags);
+	/* Check FD and open */
+	if (kni_fd < 0) {
+		kni_fd = open("/dev/" KNI_DEVICE, O_RDWR);
+		if (kni_fd < 0) {
+			RTE_LOG(ERR, KNI,
+				"Can not open /dev/%s\n", KNI_DEVICE);
+			return -1;
+		}
+	}
 
-	return mz;
+	return 0;
 }
 
-/* Pool mgmt */
-static struct rte_kni_memzone_slot*
-kni_memzone_pool_alloc(void)
+static int
+kni_reserve_mz(struct rte_kni *kni)
 {
-	struct rte_kni_memzone_slot *slot;
+	char mz_name[RTE_MEMZONE_NAMESIZE];
 
-	rte_spinlock_lock(&kni_memzone_pool.mutex);
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_TX_Q_MZ_NAME_FMT, kni->name);
+	kni->m_tx_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_tx_q == NULL, tx_q_fail);
 
-	if (!kni_memzone_pool.free) {
-		rte_spinlock_unlock(&kni_memzone_pool.mutex);
-		return NULL;
-	}
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_RX_Q_MZ_NAME_FMT, kni->name);
+	kni->m_rx_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_rx_q == NULL, rx_q_fail);
 
-	slot = kni_memzone_pool.free;
-	kni_memzone_pool.free = slot->next;
-	slot->in_use = 1;
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_ALLOC_Q_MZ_NAME_FMT, kni->name);
+	kni->m_alloc_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_alloc_q == NULL, alloc_q_fail);
 
-	if (!kni_memzone_pool.free)
-		kni_memzone_pool.free_tail = NULL;
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_FREE_Q_MZ_NAME_FMT, kni->name);
+	kni->m_free_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_free_q == NULL, free_q_fail);
 
-	rte_spinlock_unlock(&kni_memzone_pool.mutex);
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_REQ_Q_MZ_NAME_FMT, kni->name);
+	kni->m_req_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_req_q == NULL, req_q_fail);
 
-	return slot;
-}
-
-static void
-kni_memzone_pool_release(struct rte_kni_memzone_slot *slot)
-{
-	rte_spinlock_lock(&kni_memzone_pool.mutex);
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_RESP_Q_MZ_NAME_FMT, kni->name);
+	kni->m_resp_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_resp_q == NULL, resp_q_fail);
 
-	if (kni_memzone_pool.free)
-		kni_memzone_pool.free_tail->next = slot;
-	else
-		kni_memzone_pool.free = slot;
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_SYNC_ADDR_MZ_NAME_FMT, kni->name);
+	kni->m_sync_addr = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
+	KNI_MEM_CHECK(kni->m_sync_addr == NULL, sync_addr_fail);
 
-	kni_memzone_pool.free_tail = slot;
-	slot->next = NULL;
-	slot->in_use = 0;
+	return 0;
 
-	rte_spinlock_unlock(&kni_memzone_pool.mutex);
+sync_addr_fail:
+	rte_memzone_free(kni->m_resp_q);
+resp_q_fail:
+	rte_memzone_free(kni->m_req_q);
+req_q_fail:
+	rte_memzone_free(kni->m_free_q);
+free_q_fail:
+	rte_memzone_free(kni->m_alloc_q);
+alloc_q_fail:
+	rte_memzone_free(kni->m_rx_q);
+rx_q_fail:
+	rte_memzone_free(kni->m_tx_q);
+tx_q_fail:
+	return -1;
 }
 
-
-/* Shall be called before any allocation happens */
-void
-rte_kni_init(unsigned int max_kni_ifaces)
+static void
+kni_release_mz(struct rte_kni *kni)
 {
-	uint32_t i;
-	struct rte_kni_memzone_slot *it;
-	const struct rte_memzone *mz;
-#define OBJNAMSIZ 32
-	char obj_name[OBJNAMSIZ];
-	char mz_name[RTE_MEMZONE_NAMESIZE];
-
-	/* Immediately return if KNI is already initialized */
-	if (kni_memzone_pool.initialized) {
-		RTE_LOG(WARNING, KNI, "Double call to rte_kni_init()");
-		return;
-	}
-
-	if (max_kni_ifaces == 0) {
-		RTE_LOG(ERR, KNI, "Invalid number of max_kni_ifaces %d\n",
-							max_kni_ifaces);
-		RTE_LOG(ERR, KNI, "Unable to initialize KNI\n");
-		return;
-	}
-
-	/* Check FD and open */
-	if (kni_fd < 0) {
-		kni_fd = open("/dev/" KNI_DEVICE, O_RDWR);
-		if (kni_fd < 0) {
-			RTE_LOG(ERR, KNI,
-				"Can not open /dev/%s\n", KNI_DEVICE);
-			return;
-		}
-	}
-
-	/* Allocate slot objects */
-	kni_memzone_pool.slots = (struct rte_kni_memzone_slot *)
-					rte_malloc(NULL,
-					sizeof(struct rte_kni_memzone_slot) *
-					max_kni_ifaces,
-					0);
-	KNI_MEM_CHECK(kni_memzone_pool.slots == NULL);
-
-	/* Initialize general pool variables */
-	kni_memzone_pool.initialized = 1;
-	kni_memzone_pool.max_ifaces = max_kni_ifaces;
-	kni_memzone_pool.free = &kni_memzone_pool.slots[0];
-	rte_spinlock_init(&kni_memzone_pool.mutex);
-
-	/* Pre-allocate all memzones of all the slots; panic on error */
-	for (i = 0; i < max_kni_ifaces; i++) {
-
-		/* Recover current slot */
-		it = &kni_memzone_pool.slots[i];
-		it->id = i;
-
-		/* Allocate KNI context */
-		snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "KNI_INFO_%d", i);
-		mz = kni_memzone_reserve(mz_name, sizeof(struct rte_kni),
-					SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_ctx = mz;
-
-		/* TX RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_tx_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_tx_q = mz;
-
-		/* RX RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_rx_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_rx_q = mz;
-
-		/* ALLOC RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_alloc_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_alloc_q = mz;
-
-		/* FREE RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_free_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_free_q = mz;
-
-		/* Request RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_req_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_req_q = mz;
-
-		/* Response RING */
-		snprintf(obj_name, OBJNAMSIZ, "kni_resp_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_resp_q = mz;
-
-		/* Req/Resp sync mem area */
-		snprintf(obj_name, OBJNAMSIZ, "kni_sync_%d", i);
-		mz = kni_memzone_reserve(obj_name, KNI_FIFO_SIZE,
-							SOCKET_ID_ANY, 0);
-		KNI_MEM_CHECK(mz == NULL);
-		it->m_sync_addr = mz;
-
-		if ((i+1) == max_kni_ifaces) {
-			it->next = NULL;
-			kni_memzone_pool.free_tail = it;
-		} else
-			it->next = &kni_memzone_pool.slots[i+1];
-	}
-
-	return;
-
-kni_fail:
-	RTE_LOG(ERR, KNI, "Unable to allocate memory for max_kni_ifaces:%d."
-		"Increase the amount of hugepages memory\n", max_kni_ifaces);
+	rte_memzone_free(kni->m_tx_q);
+	rte_memzone_free(kni->m_rx_q);
+	rte_memzone_free(kni->m_alloc_q);
+	rte_memzone_free(kni->m_free_q);
+	rte_memzone_free(kni->m_req_q);
+	rte_memzone_free(kni->m_resp_q);
+	rte_memzone_free(kni->m_sync_addr);
 }
 
-
 struct rte_kni *
 rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
 	      const struct rte_kni_conf *conf,
@@ -294,41 +180,52 @@  rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
 {
 	int ret;
 	struct rte_kni_device_info dev_info;
-	struct rte_kni *ctx;
-	char intf_name[RTE_KNI_NAMESIZE];
-	const struct rte_memzone *mz;
-	struct rte_kni_memzone_slot *slot = NULL;
+	struct rte_kni *kni;
+	struct rte_tailq_entry *te = NULL;
+	struct rte_kni_list *kni_list;
+
+	kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
 
 	if (!pktmbuf_pool || !conf || !conf->name[0])
 		return NULL;
 
 	/* Check if KNI subsystem has been initialized */
-	if (kni_memzone_pool.initialized != 1) {
+	if (kni_fd < 0) {
 		RTE_LOG(ERR, KNI, "KNI subsystem has not been initialized. Invoke rte_kni_init() first\n");
 		return NULL;
 	}
 
-	/* Get an available slot from the pool */
-	slot = kni_memzone_pool_alloc();
-	if (!slot) {
-		RTE_LOG(ERR, KNI, "Cannot allocate more KNI interfaces; increase the number of max_kni_ifaces(current %d) or release unused ones.\n",
-			kni_memzone_pool.max_ifaces);
-		return NULL;
+	rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
+
+	TAILQ_FOREACH(te, kni_list, next) {
+		kni = (struct rte_kni *) te->data;
+		if (strncmp(conf->name, kni->name, RTE_KNI_NAMESIZE) == 0)
+			break;
 	}
 
-	/* Recover ctx */
-	ctx = slot->m_ctx->addr;
-	snprintf(intf_name, RTE_KNI_NAMESIZE, "%s", conf->name);
+	if (te != NULL) {
+		RTE_LOG(ERR, KNI, "KNI already exists\n");
+		goto unlock;
+	}
 
-	if (ctx->in_use) {
-		RTE_LOG(ERR, KNI, "KNI %s is in use\n", ctx->name);
-		return NULL;
+	te = rte_zmalloc("KNI_TAILQ_ENTRY", sizeof(*te), 0);
+	if (te == NULL) {
+		RTE_LOG(ERR, KNI, "Failed to allocate tailq entry\n");
+		goto unlock;
+	}
+
+	kni = rte_zmalloc("KNI", sizeof(struct rte_kni), RTE_CACHE_LINE_SIZE);
+	if (kni == NULL) {
+		RTE_LOG(ERR, KNI, "KNI memory allocation failed\n");
+		goto kni_fail;
 	}
-	memset(ctx, 0, sizeof(struct rte_kni));
+
+	snprintf(kni->name, RTE_KNI_NAMESIZE, "%s", conf->name);
+
 	if (ops)
-		memcpy(&ctx->ops, ops, sizeof(struct rte_kni_ops));
+		memcpy(&kni->ops, ops, sizeof(struct rte_kni_ops));
 	else
-		ctx->ops.port_id = UINT16_MAX;
+		kni->ops.port_id = UINT16_MAX;
 
 	memset(&dev_info, 0, sizeof(dev_info));
 	dev_info.bus = conf->addr.bus;
@@ -344,72 +241,76 @@  rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
 
 	memcpy(dev_info.mac_addr, conf->mac_addr, ETHER_ADDR_LEN);
 
-	snprintf(ctx->name, RTE_KNI_NAMESIZE, "%s", intf_name);
-	snprintf(dev_info.name, RTE_KNI_NAMESIZE, "%s", intf_name);
+	snprintf(dev_info.name, RTE_KNI_NAMESIZE, "%s", conf->name);
 
 	RTE_LOG(INFO, KNI, "pci: %02x:%02x:%02x \t %02x:%02x\n",
 		dev_info.bus, dev_info.devid, dev_info.function,
 			dev_info.vendor_id, dev_info.device_id);
+
+	ret = kni_reserve_mz(kni);
+	if (ret < 0)
+		goto mz_fail;
+
 	/* TX RING */
-	mz = slot->m_tx_q;
-	ctx->tx_q = mz->addr;
-	kni_fifo_init(ctx->tx_q, KNI_FIFO_COUNT_MAX);
-	dev_info.tx_phys = mz->phys_addr;
+	kni->tx_q = kni->m_tx_q->addr;
+	kni_fifo_init(kni->tx_q, KNI_FIFO_COUNT_MAX);
+	dev_info.tx_phys = kni->m_tx_q->phys_addr;
 
 	/* RX RING */
-	mz = slot->m_rx_q;
-	ctx->rx_q = mz->addr;
-	kni_fifo_init(ctx->rx_q, KNI_FIFO_COUNT_MAX);
-	dev_info.rx_phys = mz->phys_addr;
+	kni->rx_q = kni->m_rx_q->addr;
+	kni_fifo_init(kni->rx_q, KNI_FIFO_COUNT_MAX);
+	dev_info.rx_phys = kni->m_rx_q->phys_addr;
 
 	/* ALLOC RING */
-	mz = slot->m_alloc_q;
-	ctx->alloc_q = mz->addr;
-	kni_fifo_init(ctx->alloc_q, KNI_FIFO_COUNT_MAX);
-	dev_info.alloc_phys = mz->phys_addr;
+	kni->alloc_q = kni->m_alloc_q->addr;
+	kni_fifo_init(kni->alloc_q, KNI_FIFO_COUNT_MAX);
+	dev_info.alloc_phys = kni->m_alloc_q->phys_addr;
 
 	/* FREE RING */
-	mz = slot->m_free_q;
-	ctx->free_q = mz->addr;
-	kni_fifo_init(ctx->free_q, KNI_FIFO_COUNT_MAX);
-	dev_info.free_phys = mz->phys_addr;
+	kni->free_q = kni->m_free_q->addr;
+	kni_fifo_init(kni->free_q, KNI_FIFO_COUNT_MAX);
+	dev_info.free_phys = kni->m_free_q->phys_addr;
 
 	/* Request RING */
-	mz = slot->m_req_q;
-	ctx->req_q = mz->addr;
-	kni_fifo_init(ctx->req_q, KNI_FIFO_COUNT_MAX);
-	dev_info.req_phys = mz->phys_addr;
+	kni->req_q = kni->m_req_q->addr;
+	kni_fifo_init(kni->req_q, KNI_FIFO_COUNT_MAX);
+	dev_info.req_phys = kni->m_req_q->phys_addr;
 
 	/* Response RING */
-	mz = slot->m_resp_q;
-	ctx->resp_q = mz->addr;
-	kni_fifo_init(ctx->resp_q, KNI_FIFO_COUNT_MAX);
-	dev_info.resp_phys = mz->phys_addr;
+	kni->resp_q = kni->m_resp_q->addr;
+	kni_fifo_init(kni->resp_q, KNI_FIFO_COUNT_MAX);
+	dev_info.resp_phys = kni->m_resp_q->phys_addr;
 
 	/* Req/Resp sync mem area */
-	mz = slot->m_sync_addr;
-	ctx->sync_addr = mz->addr;
-	dev_info.sync_va = mz->addr;
-	dev_info.sync_phys = mz->phys_addr;
+	kni->sync_addr = kni->m_sync_addr->addr;
+	dev_info.sync_va = kni->m_sync_addr->addr;
+	dev_info.sync_phys = kni->m_sync_addr->phys_addr;
 
-	ctx->pktmbuf_pool = pktmbuf_pool;
-	ctx->group_id = conf->group_id;
-	ctx->slot_id = slot->id;
-	ctx->mbuf_size = conf->mbuf_size;
+	kni->pktmbuf_pool = pktmbuf_pool;
+	kni->group_id = conf->group_id;
+	kni->mbuf_size = conf->mbuf_size;
 
 	ret = ioctl(kni_fd, RTE_KNI_IOCTL_CREATE, &dev_info);
-	KNI_MEM_CHECK(ret < 0);
+	if (ret < 0)
+		goto ioctl_fail;
 
-	ctx->in_use = 1;
+	te->data = kni;
+	TAILQ_INSERT_TAIL(kni_list, te, next);
+	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 
 	/* Allocate mbufs and then put them into alloc_q */
-	kni_allocate_mbufs(ctx);
+	kni_allocate_mbufs(kni);
 
-	return ctx;
+	return kni;
 
+ioctl_fail:
+	kni_release_mz(kni);
+mz_fail:
+	rte_free(kni);
 kni_fail:
-	if (slot)
-		kni_memzone_pool_release(&kni_memzone_pool.slots[slot->id]);
+	rte_free(te);
+unlock:
+	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 
 	return NULL;
 }
@@ -462,19 +363,37 @@  kni_free_fifo_phy(struct rte_mempool *mp, struct rte_kni_fifo *fifo)
 int
 rte_kni_release(struct rte_kni *kni)
 {
+	struct rte_tailq_entry *te;
+	struct rte_kni_list *kni_list;
 	struct rte_kni_device_info dev_info;
-	uint32_t slot_id;
 	uint32_t retry = 5;
 
-	if (!kni || !kni->in_use)
+	if (!kni)
 		return -1;
 
+	kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
+
+	rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
+
+	TAILQ_FOREACH(te, kni_list, next) {
+		if (te->data == kni)
+			break;
+	}
+
+	if (te == NULL) {
+		goto unlock;
+	}
+
 	snprintf(dev_info.name, sizeof(dev_info.name), "%s", kni->name);
 	if (ioctl(kni_fd, RTE_KNI_IOCTL_RELEASE, &dev_info) < 0) {
 		RTE_LOG(ERR, KNI, "Fail to release kni device\n");
-		return -1;
+		goto unlock;
 	}
 
+	TAILQ_REMOVE(kni_list, te, next);
+
+	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
+
 	/* mbufs in all fifo should be released, except request/response */
 
 	/* wait until all rxq packets processed by kernel */
@@ -488,20 +407,18 @@  rte_kni_release(struct rte_kni *kni)
 	kni_free_fifo(kni->tx_q);
 	kni_free_fifo(kni->free_q);
 
-	slot_id = kni->slot_id;
+	kni_release_mz(kni);
 
-	/* Memset the KNI struct */
-	memset(kni, 0, sizeof(struct rte_kni));
+	rte_free(kni);
 
-	/* Release memzone */
-	if (slot_id > kni_memzone_pool.max_ifaces) {
-		RTE_LOG(ERR, KNI, "KNI pool: corrupted slot ID: %d, max: %d\n",
-			slot_id, kni_memzone_pool.max_ifaces);
-		return -1;
-	}
-	kni_memzone_pool_release(&kni_memzone_pool.slots[slot_id]);
+	rte_free(te);
 
 	return 0;
+
+unlock:
+	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
+
+	return -1;
 }
 
 /* default callback for request of configuring device mac address */
@@ -711,21 +628,28 @@  kni_allocate_mbufs(struct rte_kni *kni)
 struct rte_kni *
 rte_kni_get(const char *name)
 {
-	uint32_t i;
-	struct rte_kni_memzone_slot *it;
-	struct rte_kni *kni;
+	struct rte_kni *kni = NULL;
+	struct rte_tailq_entry *te;
+	struct rte_kni_list *kni_list;
+
+	if (!name || !name[0])
+		return NULL;
 
-	/* Note: could be improved perf-wise if necessary */
-	for (i = 0; i < kni_memzone_pool.max_ifaces; i++) {
-		it = &kni_memzone_pool.slots[i];
-		if (it->in_use == 0)
-			continue;
-		kni = it->m_ctx->addr;
-		if (strncmp(kni->name, name, RTE_KNI_NAMESIZE) == 0)
-			return kni;
+	kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
+
+	rte_rwlock_read_lock(RTE_EAL_TAILQ_RWLOCK);
+	TAILQ_FOREACH(te, kni_list, next) {
+		kni = te->data;
+		if (strncmp(name, kni->name, RTE_KNI_NAMESIZE) == 0)
+			break;
 	}
+	rte_rwlock_read_unlock(RTE_EAL_TAILQ_RWLOCK);
 
-	return NULL;
+	if (te == NULL) {
+		return NULL;
+	}
+
+	return kni;
 }
 
 const char *
diff --git a/lib/librte_kni/rte_kni.h b/lib/librte_kni/rte_kni.h
index 99055e2c2..601abdfc6 100644
--- a/lib/librte_kni/rte_kni.h
+++ b/lib/librte_kni/rte_kni.h
@@ -81,8 +81,12 @@  struct rte_kni_conf {
  *
  * @param max_kni_ifaces
  *  The maximum number of KNI interfaces that can coexist concurrently
+ *
+ * @return
+ *  - 0 indicates success.
+ *  - negative value indicates failure.
  */
-void rte_kni_init(unsigned int max_kni_ifaces);
+int rte_kni_init(unsigned int max_kni_ifaces);
 
 
 /**
diff --git a/test/test/test_kni.c b/test/test/test_kni.c
index 1b876719a..56c98513a 100644
--- a/test/test/test_kni.c
+++ b/test/test/test_kni.c
@@ -429,12 +429,6 @@  test_kni_processing(uint16_t port_id, struct rte_mempool *mp)
 	}
 	test_kni_ctx = NULL;
 
-	/* test of releasing a released kni device */
-	if (rte_kni_release(kni) == 0) {
-		printf("should not release a released kni device\n");
-		return -1;
-	}
-
 	/* test of reusing memzone */
 	kni = rte_kni_alloc(mp, &conf, &ops);
 	if (!kni) {