diff mbox

[PATCHv4,DPDK,5/6] Buffer related changes

Message ID 1424345237-8974-1-git-send-email-venkatesh.vivekanandan@linaro.org
State New
Headers show

Commit Message

Venkatesh Vivekanandan Feb. 19, 2015, 11:27 a.m. UTC
From: Venkatesh Vivekanandan <venkatesh.vivekanandan@linaro.org>

- Modify buffer API's according to dpdk's requirement

Signed-off-by: Venkatesh Vivekanandan <venkatesh.vivekanandan@linaro.org>
---
 platform/linux-dpdk/include/api/odp_buffer.h       |  30 +-
 platform/linux-dpdk/include/api/odp_buffer_pool.h  | 124 ++++++--
 platform/linux-dpdk/include/odp_buffer_internal.h  | 134 +++++++--
 .../linux-dpdk/include/odp_buffer_pool_internal.h  | 333 +++++++++++++++++++--
 platform/linux-dpdk/odp_buffer.c                   |  14 +-
 platform/linux-dpdk/odp_buffer_pool.c              | 153 ++++++----
 6 files changed, 655 insertions(+), 133 deletions(-)
diff mbox

Patch

diff --git a/platform/linux-dpdk/include/api/odp_buffer.h b/platform/linux-dpdk/include/api/odp_buffer.h
index b2fbc76..3c23035 100644
--- a/platform/linux-dpdk/include/api/odp_buffer.h
+++ b/platform/linux-dpdk/include/api/odp_buffer.h
@@ -19,20 +19,13 @@  extern "C" {
 #endif
 
 
-
 #include <odp_std_types.h>
+#include <odp_platform_types.h>
 
-
-
-
-
-/**
- * ODP buffer
+/** @defgroup odp_buffer ODP BUFFER
+ *  Operations on a buffer.
+ *  @{
  */
-typedef unsigned long odp_buffer_t;
-
-
-#define ODP_BUFFER_INVALID (unsigned long)(-1L) /**< Invalid buffer */
 
 
 /**
@@ -74,11 +67,21 @@  int odp_buffer_type(odp_buffer_t buf);
  *
  * @param buf      Buffer handle
  *
- * @return 1 if valid, otherwise 0
+ * @retval 1 Buffer handle represents a valid buffer.
+ * @retval 0 Buffer handle does not represent a valid buffer.
  */
 int odp_buffer_is_valid(odp_buffer_t buf);
 
 /**
+ * Buffer pool of the buffer
+ *
+ * @param buf       Buffer handle
+ *
+ * @return Handle of buffer pool buffer belongs to
+ */
+odp_buffer_pool_t odp_buffer_pool(odp_buffer_t buf);
+
+/**
  * Print buffer metadata to STDOUT
  *
  * @param buf      Buffer handle
@@ -86,6 +89,9 @@  int odp_buffer_is_valid(odp_buffer_t buf);
  */
 void odp_buffer_print(odp_buffer_t buf);
 
+/**
+ * @}
+ */
 
 #ifdef __cplusplus
 }
diff --git a/platform/linux-dpdk/include/api/odp_buffer_pool.h b/platform/linux-dpdk/include/api/odp_buffer_pool.h
index 382f4f0..4da5f84 100644
--- a/platform/linux-dpdk/include/api/odp_buffer_pool.h
+++ b/platform/linux-dpdk/include/api/odp_buffer_pool.h
@@ -21,66 +21,143 @@  extern "C" {
 
 
 #include <odp_std_types.h>
+#include <odp_platform_types.h>
 #include <odp_buffer.h>
 
+/** @addtogroup odp_buffer
+ *  Operations on a buffer pool.
+ *  @{
+ */
+
 /** Maximum queue name lenght in chars */
 #define ODP_BUFFER_POOL_NAME_LEN  32
 
-/** Invalid buffer pool */
-#define ODP_BUFFER_POOL_INVALID  (unsigned long)(-1L)
-
-/** ODP buffer pool */
-typedef unsigned long odp_buffer_pool_t;
-
+/**
+ * Buffer pool parameters
+ * Used to communicate buffer pool creation options.
+ */
+typedef struct odp_buffer_pool_param_t {
+	uint32_t buf_size;  /**< Buffer size in bytes.  The maximum
+			       number of bytes application will
+			       store in each buffer. For packets, this
+			       is the maximum packet data length, and
+			       configured headroom and tailroom will be
+			       added to this number */
+	uint32_t buf_align; /**< Minimum buffer alignment in bytes.
+			       Valid values are powers of two.  Use 0
+			       for default alignment.  Default will
+			       always be a multiple of 8. */
+	uint32_t num_bufs;  /**< Number of buffers in the pool */
+	int      buf_type;  /**< Buffer type */
+} odp_buffer_pool_param_t;
 
 /**
  * Create a buffer pool
+ * This routine is used to create a buffer pool. It take three
+ * arguments: the optional name of the pool to be created, an optional shared
+ * memory handle, and a parameter struct that describes the pool to be
+ * created. If a name is not specified the result is an anonymous pool that
+ * cannot be referenced by odp_buffer_pool_lookup().
+ *
+ * @param name     Name of the pool, max ODP_BUFFER_POOL_NAME_LEN-1 chars.
+ *                 May be specified as NULL for anonymous pools.
+ *
+ * @param shm      The shared memory object in which to create the pool.
+ *                 Use ODP_SHM_NULL to reserve default memory type
+ *                 for the buffer type.
  *
- * @param name      Name of the pool (max ODP_BUFFER_POOL_NAME_LEN - 1 chars)
- * @param base_addr Pool base address
- * @param size      Pool size in bytes
- * @param buf_size  Buffer size in bytes
- * @param buf_align Minimum buffer alignment
- * @param buf_type  Buffer type
+ * @param params   Buffer pool parameters.
  *
- * @return Buffer pool handle
+ * @return Handle of the created buffer pool
+ * @retval ODP_BUFFER_POOL_INVALID  Buffer pool could not be created
  */
+
 odp_buffer_pool_t odp_buffer_pool_create(const char *name,
-					 void *base_addr, uint64_t size,
-					 size_t buf_size, size_t buf_align,
-					 int buf_type);
+					 odp_shm_t shm,
+					 odp_buffer_pool_param_t *params);
 
+/**
+ * Destroy a buffer pool previously created by odp_buffer_pool_create()
+ *
+ * @param pool    Handle of the buffer pool to be destroyed
+ *
+ * @retval 0 Success
+ * @retval -1 Failure
+ *
+ * @note This routine destroys a previously created buffer pool. This call
+ * does not destroy any shared memory object passed to
+ * odp_buffer_pool_create() used to store the buffer pool contents. The caller
+ * takes responsibility for that. If no shared memory object was passed as
+ * part of the create call, then this routine will destroy any internal shared
+ * memory objects associated with the buffer pool. Results are undefined if
+ * an attempt is made to destroy a buffer pool that contains allocated or
+ * otherwise active buffers.
+ */
+int odp_buffer_pool_destroy(odp_buffer_pool_t pool);
 
 /**
  * Find a buffer pool by name
  *
  * @param name      Name of the pool
  *
- * @return Buffer pool handle, or ODP_BUFFER_POOL_INVALID if not found.
+ * @return Handle of found buffer pool
+ * @retval ODP_BUFFER_POOL_INVALID  Buffer pool could not be found
+ *
+ * @note This routine cannot be used to look up an anonymous pool (one created
+ * with no name).
  */
 odp_buffer_pool_t odp_buffer_pool_lookup(const char *name);
 
+/**
+ * Buffer pool information struct
+ * Used to get information about a buffer pool.
+ */
+typedef struct odp_buffer_pool_info_t {
+	const char *name;                 /**< pool name */
+	odp_shm_t shm;                    /**< handle of shared memory area
+					     supplied by application to
+					     contain buffer pool, or
+					     ODP_SHM_NULL if this pool is
+					     managed by ODP */
+	odp_buffer_pool_param_t params;   /**< pool parameters */
+} odp_buffer_pool_info_t;
+
+/**
+ * Retrieve information about a buffer pool
+ *
+ * @param pool         Buffer pool handle
+ *
+ * @param[out] info    Receives an odp_buffer_pool_info_t object
+ *                     that describes the pool.
+ *
+ * @retval 0 Success
+ * @retval -1 Failure.  Info could not be retrieved.
+ */
+
+int odp_buffer_pool_info(odp_buffer_pool_t pool,
+			 odp_buffer_pool_info_t *info);
 
 /**
  * Print buffer pool info
  *
  * @param pool      Pool handle
  *
+ * @note This routine writes implementation-defined information about the
+ * specified buffer pool to the ODP log. The intended use is for debugging.
  */
 void odp_buffer_pool_print(odp_buffer_pool_t pool);
 
-
-
 /**
  * Buffer alloc
  *
+ * The validity of a buffer can be cheked at any time with odp_buffer_is_valid()
  * @param pool      Pool handle
  *
- * @return Buffer handle or ODP_BUFFER_INVALID
+ * @return Handle of allocated buffer
+ * @retval ODP_BUFFER_INVALID  Buffer could not be allocated
  */
 odp_buffer_t odp_buffer_alloc(odp_buffer_pool_t pool);
 
-
 /**
  * Buffer free
  *
@@ -89,8 +166,9 @@  odp_buffer_t odp_buffer_alloc(odp_buffer_pool_t pool);
  */
 void odp_buffer_free(odp_buffer_t buf);
 
-
-
+/**
+ * @}
+ */
 
 #ifdef __cplusplus
 }
diff --git a/platform/linux-dpdk/include/odp_buffer_internal.h b/platform/linux-dpdk/include/odp_buffer_internal.h
index 5406606..acc958b 100644
--- a/platform/linux-dpdk/include/odp_buffer_internal.h
+++ b/platform/linux-dpdk/include/odp_buffer_internal.h
@@ -24,50 +24,144 @@  extern "C" {
 #include <odp_buffer.h>
 #include <odp_debug.h>
 #include <odp_align.h>
-#include <rte_mbuf.h>
+#include <odp_align_internal.h>
+#include <odp_config.h>
+#include <odp_byteorder.h>
+#include <odp_thread.h>
 
-/* TODO: move these to correct files */
+/* DPDK */
+#include <rte_mbuf.h>
 
-typedef uint64_t odp_phys_addr_t;
+#define ODP_BITSIZE(x) \
+	((x) <=     2 ?  1 : \
+	((x) <=     4 ?  2 : \
+	((x) <=     8 ?  3 : \
+	((x) <=    16 ?  4 : \
+	((x) <=    32 ?  5 : \
+	((x) <=    64 ?  6 : \
+	((x) <=   128 ?  7 : \
+	((x) <=   256 ?  8 : \
+	((x) <=   512 ?  9 : \
+	((x) <=  1024 ? 10 : \
+	((x) <=  2048 ? 11 : \
+	((x) <=  4096 ? 12 : \
+	((x) <=  8196 ? 13 : \
+	((x) <= 16384 ? 14 : \
+	((x) <= 32768 ? 15 : \
+	((x) <= 65536 ? 16 : \
+	 (0/0)))))))))))))))))
+
+_ODP_STATIC_ASSERT(ODP_CONFIG_PACKET_BUF_LEN_MIN >= 256,
+		   "ODP Segment size must be a minimum of 256 bytes");
+
+_ODP_STATIC_ASSERT((ODP_CONFIG_PACKET_BUF_LEN_MIN % ODP_CACHE_LINE_SIZE) == 0,
+		   "ODP Segment size must be a multiple of cache line size");
+
+_ODP_STATIC_ASSERT((ODP_CONFIG_PACKET_BUF_LEN_MAX %
+		   ODP_CONFIG_PACKET_BUF_LEN_MIN) == 0,
+		   "Packet max size must be a multiple of segment size");
+
+#define ODP_BUFFER_MAX_SEG \
+	(ODP_CONFIG_PACKET_BUF_LEN_MAX / ODP_CONFIG_PACKET_BUF_LEN_MIN)
+
+/* We can optimize storage of small raw buffers within metadata area */
+#define ODP_MAX_INLINE_BUF     ((sizeof(void *)) * (ODP_BUFFER_MAX_SEG - 1))
+
+#define ODP_BUFFER_POOL_BITS   ODP_BITSIZE(ODP_CONFIG_BUFFER_POOLS)
+#define ODP_BUFFER_SEG_BITS    ODP_BITSIZE(ODP_BUFFER_MAX_SEG)
+#define ODP_BUFFER_INDEX_BITS  (32 - ODP_BUFFER_POOL_BITS - ODP_BUFFER_SEG_BITS)
+#define ODP_BUFFER_PREFIX_BITS (ODP_BUFFER_POOL_BITS + ODP_BUFFER_INDEX_BITS)
+#define ODP_BUFFER_MAX_POOLS   (1 << ODP_BUFFER_POOL_BITS)
+#define ODP_BUFFER_MAX_BUFFERS (1 << ODP_BUFFER_INDEX_BITS)
 
 #define ODP_BUFFER_MAX_INDEX     (ODP_BUFFER_MAX_BUFFERS - 2)
 #define ODP_BUFFER_INVALID_INDEX (ODP_BUFFER_MAX_BUFFERS - 1)
 
-#define ODP_BUFS_PER_CHUNK       16
-#define ODP_BUFS_PER_SCATTER      4
-
-#define ODP_BUFFER_TYPE_CHUNK    0xffff
-
-
-#define ODP_BUFFER_POOL_BITS   4
-#define ODP_BUFFER_INDEX_BITS  (32 - ODP_BUFFER_POOL_BITS)
-#define ODP_BUFFER_MAX_POOLS   (1 << ODP_BUFFER_POOL_BITS)
-#define ODP_BUFFER_MAX_BUFFERS (1 << ODP_BUFFER_INDEX_BITS)
-
 typedef union odp_buffer_bits_t {
 	uint32_t     u32;
 	odp_buffer_t handle;
 
 	struct {
-		uint32_t pool:ODP_BUFFER_POOL_BITS;
+#if ODP_BYTE_ORDER == ODP_BIG_ENDIAN
+		uint32_t pool_id:ODP_BUFFER_POOL_BITS;
+		uint32_t index:ODP_BUFFER_INDEX_BITS;
+		uint32_t seg:ODP_BUFFER_SEG_BITS;
+#else
+		uint32_t seg:ODP_BUFFER_SEG_BITS;
 		uint32_t index:ODP_BUFFER_INDEX_BITS;
+		uint32_t pool_id:ODP_BUFFER_POOL_BITS;
+#endif
 	};
-} odp_buffer_bits_t;
 
+	struct {
+#if ODP_BYTE_ORDER == ODP_BIG_ENDIAN
+		uint32_t prefix:ODP_BUFFER_PREFIX_BITS;
+		uint32_t pfxseg:ODP_BUFFER_SEG_BITS;
+#else
+		uint32_t pfxseg:ODP_BUFFER_SEG_BITS;
+		uint32_t prefix:ODP_BUFFER_PREFIX_BITS;
+#endif
+	};
+} odp_buffer_bits_t;
 
 /* forward declaration */
 struct odp_buffer_hdr_t;
 
-
+/* Common buffer header */
 typedef struct odp_buffer_hdr_t {
+	/* DPDK specific */
 	struct rte_mbuf mb;            /* Underlying DPDK rte_mbuf */
-	struct odp_buffer_hdr_t *next; /* Next buf in a list */
-	int type;                      /* ODP buffer type; not DPDK buf type */
 	uint32_t index;                /* Index in the rte_mempool */
+
+	struct odp_buffer_hdr_t *next;       /* next buf in a list */
+	int                      allocator;  /* allocating thread id */
+	odp_buffer_bits_t        handle;     /* handle */
+	union {
+		uint32_t all;
+		struct {
+			uint32_t zeroized:1; /* Zeroize buf data on free */
+			uint32_t hdrdata:1;  /* Data is in buffer hdr */
+		};
+	} flags;
+	int                      type;       /* buffer type */
+	size_t                   size;       /* max data size */
+	odp_atomic_u32_t         ref_count;  /* reference count */
+	odp_buffer_pool_t        pool_hdl;   /* buffer pool handle */
+	union {
+		uint64_t         buf_u64;    /* user u64 */
+		void            *buf_ctx;    /* user context */
+		const void      *buf_cctx;   /* const alias for ctx */
+		void            *udata_addr; /* user metadata addr */
+	};
+	size_t                   udata_size; /* size of user metadata */
+	uint32_t                 segcount;   /* segment count */
+	uint32_t                 segsize;    /* segment size */
+	void                    *addr[ODP_BUFFER_MAX_SEG]; /* block addrs */
+
 } odp_buffer_hdr_t;
 
-int odp_buffer_snprint(char *str, size_t n, odp_buffer_t buf);
+typedef struct odp_buffer_hdr_stride {
+	uint8_t pad[ODP_CACHE_LINE_SIZE_ROUNDUP(sizeof(odp_buffer_hdr_t))];
+} odp_buffer_hdr_stride;
+/* Ensure next header starts from 8 byte align */
+_ODP_STATIC_ASSERT((sizeof(odp_buffer_hdr_t) % 8) == 0,
+		   "ODP_BUFFER_HDR_T__SIZE_ERROR");
+
+typedef struct odp_buf_blk_t {
+	struct odp_buf_blk_t *next;
+	struct odp_buf_blk_t *prev;
+} odp_buf_blk_t;
+
+/* Raw buffer header */
+typedef struct {
+	odp_buffer_hdr_t buf_hdr;    /* common buffer header */
+} odp_raw_buffer_hdr_t;
+
+/* Free buffer marker */
+#define ODP_FREEBUF -1
 
+/* Forward declarations */
+odp_buffer_t buffer_alloc(odp_buffer_pool_t pool, size_t size);
 
 #ifdef __cplusplus
 }
diff --git a/platform/linux-dpdk/include/odp_buffer_pool_internal.h b/platform/linux-dpdk/include/odp_buffer_pool_internal.h
index 1a36655..0f19f87 100644
--- a/platform/linux-dpdk/include/odp_buffer_pool_internal.h
+++ b/platform/linux-dpdk/include/odp_buffer_pool_internal.h
@@ -19,16 +19,48 @@  extern "C" {
 #endif
 
 #include <odp_std_types.h>
+#include <odp_align.h>
+#include <odp_align_internal.h>
 #include <odp_buffer_pool.h>
 #include <odp_buffer_internal.h>
-#include <odp_align.h>
 #include <odp_hints.h>
 #include <odp_config.h>
 #include <odp_debug.h>
+#include <odp_shared_memory.h>
+#include <odp_atomic.h>
+#include <odp_atomic_internal.h>
+#include <string.h>
 
-/* for DPDK */
+/* DPDK */
 #include <rte_mempool.h>
 
+/**
+ * Buffer initialization routine prototype
+ *
+ * @note Routines of this type MAY be passed as part of the
+ * _odp_buffer_pool_init_t structure to be called whenever a
+ * buffer is allocated to initialize the user metadata
+ * associated with that buffer.
+ */
+typedef void (_odp_buf_init_t)(odp_buffer_t buf, void *buf_init_arg);
+
+/**
+ * Buffer pool initialization parameters
+ * Used to communicate buffer pool initialization options. Internal for now.
+ */
+typedef struct _odp_buffer_pool_init_t {
+	size_t udata_size;         /**< Size of user metadata for each buffer */
+	_odp_buf_init_t *buf_init; /**< Buffer initialization routine to use */
+	void *buf_init_arg;        /**< Argument to be passed to buf_init() */
+} _odp_buffer_pool_init_t;         /**< Type of buffer initialization struct */
+
+/* Local cache for buffer alloc/free acceleration */
+typedef struct local_cache_t {
+	odp_buffer_hdr_t *buf_freelist;  /* The local cache */
+	uint64_t bufallocs;              /* Local buffer alloc count */
+	uint64_t buffrees;               /* Local buffer free count */
+} local_cache_t;
+
 /* Use ticketlock instead of spinlock */
 #define POOL_USE_TICKETLOCK
 
@@ -38,11 +70,16 @@  extern "C" {
 
 #ifdef POOL_USE_TICKETLOCK
 #include <odp_ticketlock.h>
+#define POOL_LOCK(a)      odp_ticketlock_lock(a)
+#define POOL_UNLOCK(a)    odp_ticketlock_unlock(a)
+#define POOL_LOCK_INIT(a) odp_ticketlock_init(a)
 #else
 #include <odp_spinlock.h>
+#define POOL_LOCK(a)      odp_spinlock_lock(a)
+#define POOL_UNLOCK(a)    odp_spinlock_unlock(a)
+#define POOL_LOCK_INIT(a) odp_spinlock_init(a)
 #endif
 
-
 struct pool_entry_s {
 #ifdef POOL_USE_TICKETLOCK
 	odp_ticketlock_t        lock ODP_ALIGNED_CACHE;
@@ -50,38 +87,294 @@  struct pool_entry_s {
 	odp_spinlock_t          lock ODP_ALIGNED_CACHE;
 #endif
 
-	uint64_t                free_bufs;
 	char                    name[ODP_BUFFER_POOL_NAME_LEN];
-
-
-	odp_buffer_pool_t       pool ODP_ALIGNED_CACHE;
-	uintptr_t               buf_base;
-	size_t                  buf_size;
-	size_t                  buf_offset;
-	uint64_t                num_bufs;
-	void                   *pool_base_addr;
-	uint64_t                pool_size;
-	size_t                  payload_size;
-	size_t                  payload_align;
-	int                     buf_type;
-	size_t                  hdr_size;
+	odp_buffer_pool_param_t params;
+	_odp_buffer_pool_init_t init_params;
+	odp_buffer_pool_t       pool_hdl;
+	uint32_t                pool_id;
+	odp_shm_t               pool_shm;
+	union {
+		uint32_t all;
+		struct {
+			uint32_t has_name:1;
+			uint32_t user_supplied_shm:1;
+			uint32_t unsegmented:1;
+			uint32_t zeroized:1;
+			uint32_t predefined:1;
+		};
+	} flags;
+	uint32_t                quiesced;
+	uint32_t                low_wm_assert;
+	uint8_t                *pool_base_addr;
+	uint8_t                *pool_mdata_addr;
+	size_t                  pool_size;
+	uint32_t                buf_align;
+	uint32_t                buf_stride;
+	_odp_atomic_ptr_t       buf_freelist;
+	_odp_atomic_ptr_t       blk_freelist;
+	odp_atomic_u32_t        bufcount;
+	odp_atomic_u32_t        blkcount;
+	odp_atomic_u64_t        bufallocs;
+	odp_atomic_u64_t        buffrees;
+	odp_atomic_u64_t        blkallocs;
+	odp_atomic_u64_t        blkfrees;
+	odp_atomic_u64_t        bufempty;
+	odp_atomic_u64_t        blkempty;
+	odp_atomic_u64_t        high_wm_count;
+	odp_atomic_u64_t        low_wm_count;
+	uint32_t                seg_size;
+	uint32_t                high_wm;
+	uint32_t                low_wm;
+	uint32_t                headroom;
+	uint32_t                tailroom;
 };
 
+typedef union pool_entry_u {
+	struct pool_entry_s s;
+
+	uint8_t pad[ODP_CACHE_LINE_SIZE_ROUNDUP(sizeof(struct pool_entry_s))];
+} pool_entry_t;
 
 extern void *pool_entry_ptr[];
 
+#if defined(ODP_CONFIG_SECURE_POOLS) && (ODP_CONFIG_SECURE_POOLS == 1)
+#define buffer_is_secure(buf) (buf->flags.zeroized)
+#define pool_is_secure(pool) (pool->flags.zeroized)
+#else
+#define buffer_is_secure(buf) 0
+#define pool_is_secure(pool) 0
+#endif
+
+#define TAG_ALIGN ((size_t)16)
+
+#define odp_cs(ptr, old, new) \
+	_odp_atomic_ptr_cmp_xchg_strong(&ptr, (void **)&old, (void *)new, \
+					_ODP_MEMMODEL_SC, \
+					_ODP_MEMMODEL_SC)
 
-static inline void *get_pool_entry(odp_buffer_pool_t pool_id)
+/* Helper functions for pointer tagging to avoid ABA race conditions */
+#define odp_tag(ptr) \
+	(((size_t)ptr) & (TAG_ALIGN - 1))
+
+#define odp_detag(ptr) \
+	((void *)(((size_t)ptr) & -TAG_ALIGN))
+
+#define odp_retag(ptr, tag) \
+	((void *)(((size_t)ptr) | odp_tag(tag)))
+
+
+static inline void *get_blk(struct pool_entry_s *pool)
+{
+	void *oldhead, *myhead, *newhead;
+
+	oldhead = _odp_atomic_ptr_load(&pool->blk_freelist, _ODP_MEMMODEL_ACQ);
+
+	do {
+		size_t tag = odp_tag(oldhead);
+		myhead = odp_detag(oldhead);
+		if (odp_unlikely(myhead == NULL))
+			break;
+		newhead = odp_retag(((odp_buf_blk_t *)myhead)->next, tag + 1);
+	} while (odp_cs(pool->blk_freelist, oldhead, newhead) == 0);
+
+	if (odp_unlikely(myhead == NULL))
+		odp_atomic_inc_u64(&pool->blkempty);
+	else
+		odp_atomic_dec_u32(&pool->blkcount);
+
+	return (void *)myhead;
+}
+
+static inline void ret_blk(struct pool_entry_s *pool, void *block)
+{
+	void *oldhead, *myhead, *myblock;
+
+	oldhead = _odp_atomic_ptr_load(&pool->blk_freelist, _ODP_MEMMODEL_ACQ);
+
+	do {
+		size_t tag = odp_tag(oldhead);
+		myhead = odp_detag(oldhead);
+		((odp_buf_blk_t *)block)->next = myhead;
+		myblock = odp_retag(block, tag + 1);
+	} while (odp_cs(pool->blk_freelist, oldhead, myblock) == 0);
+
+	odp_atomic_inc_u32(&pool->blkcount);
+	odp_atomic_inc_u64(&pool->blkfrees);
+}
+
+static inline odp_buffer_hdr_t *get_buf(struct pool_entry_s *pool)
+{
+	odp_buffer_hdr_t *oldhead, *myhead, *newhead;
+
+	oldhead = _odp_atomic_ptr_load(&pool->buf_freelist, _ODP_MEMMODEL_ACQ);
+
+	do {
+		size_t tag = odp_tag(oldhead);
+		myhead = odp_detag(oldhead);
+		if (odp_unlikely(myhead == NULL))
+			break;
+		newhead = odp_retag(myhead->next, tag + 1);
+	} while (odp_cs(pool->buf_freelist, oldhead, newhead) == 0);
+
+	if (odp_unlikely(myhead == NULL)) {
+		odp_atomic_inc_u64(&pool->bufempty);
+	} else {
+		uint64_t bufcount =
+			odp_atomic_fetch_sub_u32(&pool->bufcount, 1) - 1;
+
+		/* Check for low watermark condition */
+		if (bufcount == pool->low_wm && !pool->low_wm_assert) {
+			pool->low_wm_assert = 1;
+			odp_atomic_inc_u64(&pool->low_wm_count);
+		}
+
+		odp_atomic_inc_u64(&pool->bufallocs);
+		myhead->next = myhead;  /* Mark buffer allocated */
+		myhead->allocator = odp_thread_id();
+	}
+
+	return (void *)myhead;
+}
+
+static inline void ret_buf(struct pool_entry_s *pool, odp_buffer_hdr_t *buf)
+{
+	odp_buffer_hdr_t *oldhead, *myhead, *mybuf;
+
+	buf->allocator = ODP_FREEBUF;  /* Mark buffer free */
+
+	if (!buf->flags.hdrdata && buf->type != ODP_BUFFER_TYPE_RAW) {
+		while (buf->segcount > 0) {
+			if (buffer_is_secure(buf) || pool_is_secure(pool))
+				memset(buf->addr[buf->segcount - 1],
+				       0, buf->segsize);
+			ret_blk(pool, buf->addr[--buf->segcount]);
+		}
+		buf->size = 0;
+	}
+
+	oldhead = _odp_atomic_ptr_load(&pool->buf_freelist, _ODP_MEMMODEL_ACQ);
+
+	do {
+		size_t tag = odp_tag(oldhead);
+		myhead = odp_detag(oldhead);
+		buf->next = myhead;
+		mybuf = odp_retag(buf, tag + 1);
+	} while (odp_cs(pool->buf_freelist, oldhead, mybuf) == 0);
+
+	uint64_t bufcount = odp_atomic_fetch_add_u32(&pool->bufcount, 1) + 1;
+
+	/* Check if low watermark condition should be deasserted */
+	if (bufcount == pool->high_wm && pool->low_wm_assert) {
+		pool->low_wm_assert = 0;
+		odp_atomic_inc_u64(&pool->high_wm_count);
+	}
+
+	odp_atomic_inc_u64(&pool->buffrees);
+}
+
+static inline void *get_local_buf(local_cache_t *buf_cache,
+				  struct pool_entry_s *pool,
+				  size_t totsize)
+{
+	odp_buffer_hdr_t *buf = buf_cache->buf_freelist;
+
+	if (odp_likely(buf != NULL)) {
+		buf_cache->buf_freelist = buf->next;
+
+		if (odp_unlikely(buf->size < totsize)) {
+			intmax_t needed = totsize - buf->size;
+
+			do {
+				void *blk = get_blk(pool);
+				if (odp_unlikely(blk == NULL)) {
+					ret_buf(pool, buf);
+					buf_cache->buffrees--;
+					return NULL;
+				}
+				buf->addr[buf->segcount++] = blk;
+				needed -= pool->seg_size;
+			} while (needed > 0);
+
+			buf->size = buf->segcount * pool->seg_size;
+		}
+
+		buf_cache->bufallocs++;
+		buf->allocator = odp_thread_id();  /* Mark buffer allocated */
+	}
+
+	return buf;
+}
+
+static inline void ret_local_buf(local_cache_t *buf_cache,
+				odp_buffer_hdr_t *buf)
+{
+	buf->allocator = ODP_FREEBUF;
+	buf->next = buf_cache->buf_freelist;
+	buf_cache->buf_freelist = buf;
+
+	buf_cache->buffrees++;
+}
+
+static inline void flush_cache(local_cache_t *buf_cache,
+			       struct pool_entry_s *pool)
+{
+	odp_buffer_hdr_t *buf = buf_cache->buf_freelist;
+	uint32_t flush_count = 0;
+
+	while (buf != NULL) {
+		odp_buffer_hdr_t *next = buf->next;
+		ret_buf(pool, buf);
+		buf = next;
+		flush_count++;
+	}
+
+	odp_atomic_add_u64(&pool->bufallocs, buf_cache->bufallocs);
+	odp_atomic_add_u64(&pool->buffrees, buf_cache->buffrees - flush_count);
+
+	buf_cache->buf_freelist = NULL;
+	buf_cache->bufallocs = 0;
+	buf_cache->buffrees = 0;
+}
+
+static inline odp_buffer_pool_t pool_index_to_handle(uint32_t pool_id)
+{
+	return pool_id;
+}
+
+static inline uint32_t pool_handle_to_index(odp_buffer_pool_t pool_hdl)
+{
+	return pool_hdl;
+}
+
+static inline void *get_pool_entry(uint32_t pool_id)
 {
 	return pool_entry_ptr[pool_id];
 }
 
+static inline pool_entry_t *odp_pool_to_entry(odp_buffer_pool_t pool)
+{
+	return (pool_entry_t *)get_pool_entry(pool_handle_to_index(pool));
+}
+
+static inline pool_entry_t *odp_buf_to_pool(odp_buffer_hdr_t *buf)
+{
+	return odp_pool_to_entry(buf->pool_hdl);
+}
 
-static inline odp_buffer_hdr_t *odp_buf_to_hdr(odp_buffer_t buf)
+static inline uint32_t odp_buffer_pool_segment_size(odp_buffer_pool_t pool)
 {
-	return (odp_buffer_hdr_t *)buf;
+	return odp_pool_to_entry(pool)->s.seg_size;
 }
 
+static inline uint32_t odp_buffer_pool_headroom(odp_buffer_pool_t pool)
+{
+	return odp_pool_to_entry(pool)->s.headroom;
+}
+
+static inline uint32_t odp_buffer_pool_tailroom(odp_buffer_pool_t pool)
+{
+	return odp_pool_to_entry(pool)->s.tailroom;
+}
 
 #ifdef __cplusplus
 }
diff --git a/platform/linux-dpdk/odp_buffer.c b/platform/linux-dpdk/odp_buffer.c
index 4914ca2..d89f2af 100644
--- a/platform/linux-dpdk/odp_buffer.c
+++ b/platform/linux-dpdk/odp_buffer.c
@@ -5,8 +5,10 @@ 
  */
 
 #include <odp_buffer.h>
-#include <odp_buffer_internal.h>
 #include <odp_buffer_pool_internal.h>
+#include <odp_buffer_internal.h>
+#include <odp_buffer_inlines.h>
+#include <odp_debug_internal.h>
 
 #include <string.h>
 #include <stdio.h>
@@ -44,13 +46,13 @@  int odp_buffer_is_valid(odp_buffer_t buf)
 }
 
 
-int odp_buffer_snprint(char *str, size_t n, odp_buffer_t buf)
+int odp_buffer_snprint(char *str, uint32_t n, odp_buffer_t buf)
 {
 	odp_buffer_hdr_t *hdr;
 	int len = 0;
 
 	if (!odp_buffer_is_valid(buf)) {
-		printf("Buffer is not valid.\n");
+		ODP_PRINT("Buffer is not valid.\n");
 		return len;
 	}
 
@@ -67,7 +69,9 @@  int odp_buffer_snprint(char *str, size_t n, odp_buffer_t buf)
 	len += snprintf(&str[len], n-len,
 			"  size         %u\n",        hdr->mb.buf_len);
 	len += snprintf(&str[len], n-len,
-			"  ref_count    %i\n",        hdr->mb.refcnt);
+			"  ref_count    %i\n",
+			odp_atomic_load_u32((odp_atomic_u32_t *)
+					    &hdr->mb.refcnt));
 	len += snprintf(&str[len], n-len,
 			"  dpdk type    %i\n",        hdr->mb.type);
 	len += snprintf(&str[len], n-len,
@@ -86,5 +90,5 @@  void odp_buffer_print(odp_buffer_t buf)
 	len = odp_buffer_snprint(str, max_len-1, buf);
 	str[len] = 0;
 
-	printf("\n%s\n", str);
+	ODP_PRINT("\n%s\n", str);
 }
diff --git a/platform/linux-dpdk/odp_buffer_pool.c b/platform/linux-dpdk/odp_buffer_pool.c
index ca25ace..6636b8d 100644
--- a/platform/linux-dpdk/odp_buffer_pool.c
+++ b/platform/linux-dpdk/odp_buffer_pool.c
@@ -6,16 +6,19 @@ 
 
 #include <odp_std_types.h>
 #include <odp_buffer_pool.h>
-#include <odp_buffer_pool_internal.h>
 #include <odp_buffer_internal.h>
+#include <odp_buffer_pool_internal.h>
+#include <odp_buffer_inlines.h>
 #include <odp_packet_internal.h>
 #include <odp_timer_internal.h>
+#include <odp_align_internal.h>
 #include <odp_shared_memory.h>
 #include <odp_align.h>
 #include <odp_internal.h>
 #include <odp_config.h>
 #include <odp_hints.h>
-#include <odp_debug.h>
+#include <odp_debug_internal.h>
+#include <odp_atomic_internal.h>
 
 #include <string.h>
 #include <stdlib.h>
@@ -26,44 +29,27 @@ 
 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 #define NB_MBUF   32768
 
-#ifdef POOL_USE_TICKETLOCK
-#include <odp_ticketlock.h>
-#define LOCK(a)      odp_ticketlock_lock(a)
-#define UNLOCK(a)    odp_ticketlock_unlock(a)
-#define LOCK_INIT(a) odp_ticketlock_init(a)
-#else
-#include <odp_spinlock.h>
-#define LOCK(a)      odp_spinlock_lock(a)
-#define UNLOCK(a)    odp_spinlock_unlock(a)
-#define LOCK_INIT(a) odp_spinlock_init(a)
-#endif
-
-
 #if ODP_CONFIG_BUFFER_POOLS > ODP_BUFFER_MAX_POOLS
 #error ODP_CONFIG_BUFFER_POOLS > ODP_BUFFER_MAX_POOLS
 #endif
 
-#define NULL_INDEX ((uint32_t)-1)
-
-union buffer_type_any_u {
+typedef union buffer_type_any_u {
 	odp_buffer_hdr_t  buf;
 	odp_packet_hdr_t  pkt;
 	odp_timeout_hdr_t tmo;
-};
-
-typedef union buffer_type_any_u odp_any_buffer_hdr_t;
+} odp_anybuf_t;
 
-typedef union pool_entry_u {
-	struct pool_entry_s s;
-
-	uint8_t pad[ODP_CACHE_LINE_SIZE_ROUNDUP(sizeof(struct pool_entry_s))];
-
-} pool_entry_t;
+/* Any buffer type header */
+typedef struct {
+	odp_anybuf_t any_hdr;    /* any buffer type */
+} odp_any_buffer_hdr_t;
 
+typedef struct odp_any_hdr_stride {
+	uint8_t pad[ODP_CACHE_LINE_SIZE_ROUNDUP(sizeof(odp_any_buffer_hdr_t))];
+} odp_any_hdr_stride;
 
 typedef struct pool_table_t {
 	pool_entry_t pool[ODP_CONFIG_BUFFER_POOLS];
-
 } pool_table_t;
 
 
@@ -73,6 +59,8 @@  static pool_table_t *pool_tbl;
 /* Pool entry pointers (for inlining) */
 void *pool_entry_ptr[ODP_CONFIG_BUFFER_POOLS];
 
+/* Local cache for buffer alloc/free acceleration */
+static __thread local_cache_t local_cache[ODP_CONFIG_BUFFER_POOLS];
 
 int odp_buffer_pool_init_global(void)
 {
@@ -94,9 +82,9 @@  int odp_buffer_pool_init_global(void)
 	for (i = 0; i < ODP_CONFIG_BUFFER_POOLS; i++) {
 		/* init locks */
 		pool_entry_t *pool = &pool_tbl->pool[i];
-		LOCK_INIT(&pool->s.lock);
-		pool->s.pool = i;
-
+		POOL_LOCK_INIT(&pool->s.lock);
+		pool->s.pool_hdl = pool_index_to_handle(i);
+		pool->s.pool_id = i;
 		pool_entry_ptr[i] = pool;
 	}
 
@@ -187,10 +175,13 @@  odp_dpdk_mbuf_ctor(struct rte_mempool *mp,
 	buf_hdr->index = i;
 }
 
+/**
+ * Buffer pool creation
+ */
+
 odp_buffer_pool_t odp_buffer_pool_create(const char *name,
-					 void *base_addr, uint64_t size,
-					 size_t buf_size, size_t buf_align,
-					 int buf_type)
+					odp_shm_t shm,
+					odp_buffer_pool_param_t *params)
 {
 	struct rte_mempool *pool = NULL;
 	struct mbuf_pool_ctor_arg mbp_ctor_arg;
@@ -198,47 +189,39 @@  odp_buffer_pool_t odp_buffer_pool_create(const char *name,
 	unsigned mb_size;
 	size_t hdr_size;
 
-	/* Not used for rte_mempool; the new ODP buffer management introduces
-	 * rte_mempool_create_from_region where base_addr makes sense */
-	(void)base_addr;
-
-	/* buf_align will be removed soon, no need to wory about it */
-	(void)buf_align;
-
-	ODP_DBG("odp_buffer_pool_create: %s, %lx, %u, %u, %u, %d\n", name,
-		(uint64_t) base_addr, (unsigned) size,
-		(unsigned) buf_size, (unsigned) buf_align,
-		buf_type);
+	ODP_DBG("odp_buffer_pool_create: %s, %lx, %u, %u, %d\n", name,
+		(uint64_t) shm, (unsigned) params->buf_size,
+		(unsigned) params->buf_align, params->buf_type);
 
-	switch (buf_type) {
+	switch (params->buf_type) {
 	case ODP_BUFFER_TYPE_RAW:
 		hdr_size = sizeof(odp_buffer_hdr_t);
-		mbp_ctor_arg.seg_buf_size = (uint16_t) buf_size;
+		mbp_ctor_arg.seg_buf_size = (uint16_t) params->buf_size;
 		break;
 	case ODP_BUFFER_TYPE_PACKET:
 		hdr_size = sizeof(odp_packet_hdr_t);
 		mbp_ctor_arg.seg_buf_size =
-			(uint16_t) (RTE_PKTMBUF_HEADROOM + buf_size);
+			(uint16_t) (RTE_PKTMBUF_HEADROOM + params->buf_size);
 		break;
 	case ODP_BUFFER_TYPE_TIMEOUT:
 		hdr_size = sizeof(odp_timeout_hdr_t);
-		mbp_ctor_arg.seg_buf_size = (uint16_t) buf_size;
+		mbp_ctor_arg.seg_buf_size = (uint16_t) params->buf_size;
 		break;
 	case ODP_BUFFER_TYPE_ANY:
 		hdr_size = sizeof(odp_any_buffer_hdr_t);
 		mbp_ctor_arg.seg_buf_size =
-			(uint16_t) (RTE_PKTMBUF_HEADROOM + buf_size);
+			(uint16_t) (RTE_PKTMBUF_HEADROOM + params->buf_size);
 		break;
 	default:
-		ODP_ERR("odp_buffer_pool_create: Bad type %i\n", buf_type);
+		ODP_ERR("odp_buffer_pool_create: Bad type %i\n",
+			params->buf_type);
 		return ODP_BUFFER_POOL_INVALID;
-		break;
 	}
 
 	mb_ctor_arg.seg_buf_offset =
 		(uint16_t) ODP_CACHE_LINE_SIZE_ROUNDUP(hdr_size);
 	mb_ctor_arg.seg_buf_size = mbp_ctor_arg.seg_buf_size;
-	mb_ctor_arg.buf_type = buf_type;
+	mb_ctor_arg.buf_type = params->buf_type;
 	mb_size = mb_ctor_arg.seg_buf_offset + mb_ctor_arg.seg_buf_size;
 
 	pool = rte_mempool_create(name, NB_MBUF,
@@ -267,6 +250,61 @@  odp_buffer_pool_t odp_buffer_pool_lookup(const char *name)
 	return (odp_buffer_pool_t)mp;
 }
 
+odp_buffer_t buffer_alloc(odp_buffer_pool_t pool_hdl, size_t size)
+{
+	uint32_t pool_id = pool_handle_to_index(pool_hdl);
+	pool_entry_t *pool = get_pool_entry(pool_id);
+	uintmax_t totsize = pool->s.headroom + size + pool->s.tailroom;
+	odp_anybuf_t *buf;
+
+	/* Reject oversized allocation requests */
+	if ((pool->s.flags.unsegmented && totsize > pool->s.seg_size) ||
+	    (!pool->s.flags.unsegmented &&
+	     totsize > ODP_CONFIG_PACKET_BUF_LEN_MAX))
+		return ODP_BUFFER_INVALID;
+
+	/* Try to satisfy request from the local cache */
+	buf = (odp_anybuf_t *)(void *)get_local_buf(&local_cache[pool_id],
+						    &pool->s, totsize);
+
+	/* If cache is empty, satisfy request from the pool */
+	if (odp_unlikely(buf == NULL)) {
+		buf = (odp_anybuf_t *)(void *)get_buf(&pool->s);
+
+		if (odp_unlikely(buf == NULL))
+			return ODP_BUFFER_INVALID;
+
+		/* Get blocks for this buffer, if pool uses application data */
+		if (buf->buf.size < totsize) {
+			intmax_t needed = totsize - buf->buf.size;
+			do {
+				uint8_t *blk = get_blk(&pool->s);
+				if (blk == NULL) {
+					ret_buf(&pool->s, &buf->buf);
+					return ODP_BUFFER_INVALID;
+				}
+				buf->buf.addr[buf->buf.segcount++] = blk;
+				needed -= pool->s.seg_size;
+			} while (needed > 0);
+			buf->buf.size = buf->buf.segcount * pool->s.seg_size;
+		}
+	}
+
+	/* By default, buffers inherit their pool's zeroization setting */
+	buf->buf.flags.zeroized = pool->s.flags.zeroized;
+
+	if (buf->buf.type == ODP_BUFFER_TYPE_PACKET) {
+		packet_init(pool, &buf->pkt, size);
+
+		if (pool->s.init_params.buf_init != NULL)
+			(*pool->s.init_params.buf_init)
+				(buf->buf.handle.handle,
+				 pool->s.init_params.buf_init_arg);
+	}
+
+	return odp_hdr_to_buf(&buf->buf);
+}
+
 
 odp_buffer_t odp_buffer_alloc(odp_buffer_pool_t pool_id)
 {
@@ -279,6 +317,15 @@  void odp_buffer_free(odp_buffer_t buf)
 	rte_pktmbuf_free((struct rte_mbuf *)buf);
 }
 
+void _odp_flush_caches(void)
+{
+	int i;
+
+	for (i = 0; i < ODP_CONFIG_BUFFER_POOLS; i++) {
+		pool_entry_t *pool = get_pool_entry(i);
+		flush_cache(&local_cache[i], &pool->s);
+	}
+}
 
 void odp_buffer_pool_print(odp_buffer_pool_t pool_id)
 {