DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH] mbuf:rearrange mbuf to be more mbuf chain friendly
@ 2016-06-25 15:29 Keith Wiles
  2016-06-25 15:48 ` Wiles, Keith
  2016-06-25 15:55 ` [dpdk-dev] [PATCH v2] " Keith Wiles
  0 siblings, 2 replies; 8+ messages in thread
From: Keith Wiles @ 2016-06-25 15:29 UTC (permalink / raw)
  To: dev

Move the next pointer to the first cacheline of the rte_mbuf structure
and move the offload values to the second cacheline to give better
performance to applications using chained mbufs.

Enabled by a configuration option CONFIG_RTE_MBUF_CHAIN_FRIENDLY default
is set to No.

Signed-off-by: Keith Wiles <keith.wiles@intel.com>
---
 config/common_base                                 |  2 +
 .../linuxapp/eal/include/exec-env/rte_kni_common.h |  8 +++
 lib/librte_mbuf/rte_mbuf.h                         | 67 +++++++++++++++-------
 3 files changed, 56 insertions(+), 21 deletions(-)

diff --git a/config/common_base b/config/common_base
index 3a04fba..bdde2e7 100644
--- a/config/common_base
+++ b/config/common_base
@@ -402,6 +402,8 @@ CONFIG_RTE_LIBRTE_MBUF=y
 CONFIG_RTE_LIBRTE_MBUF_DEBUG=n
 CONFIG_RTE_MBUF_REFCNT_ATOMIC=y
 CONFIG_RTE_PKTMBUF_HEADROOM=128
+# Set to y if needing to be mbuf chain friendly.
+CONFIG_RTE_MBUF_CHAIN_FRIENDLY=n
 
 #
 # Compile librte_timer
diff --git a/lib/librte_eal/linuxapp/eal/include/exec-env/rte_kni_common.h b/lib/librte_eal/linuxapp/eal/include/exec-env/rte_kni_common.h
index 2acdfd9..44d65cd 100644
--- a/lib/librte_eal/linuxapp/eal/include/exec-env/rte_kni_common.h
+++ b/lib/librte_eal/linuxapp/eal/include/exec-env/rte_kni_common.h
@@ -120,11 +120,19 @@ struct rte_kni_mbuf {
 	char pad2[4];
 	uint32_t pkt_len;       /**< Total pkt len: sum of all segment data_len. */
 	uint16_t data_len;      /**< Amount of data in segment buffer. */
+#ifdef RTE_MBUF_CHAIN_FRIENDLY
+	char pad3[8];
+	void *next;
 
 	/* fields on second cache line */
+	char pad4[16] __attribute__((__aligned__(RTE_CACHE_LINE_MIN_SIZE)));
+	void *pool;
+#else
+	/* fields on second cache line */
 	char pad3[8] __attribute__((__aligned__(RTE_CACHE_LINE_MIN_SIZE)));
 	void *pool;
 	void *next;
+#endif
 };
 
 /*
diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index 8798c41..d02ca28 100644
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -758,6 +758,28 @@ typedef uint8_t  MARKER8[0];  /**< generic marker with 1B alignment */
 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
                                * with a single assignment */
 
+typedef union {
+	uint32_t rss;     /**< RSS hash result if RSS enabled */
+	struct {
+		union {
+			struct {
+				uint16_t hash;
+				uint16_t id;
+			};
+			uint32_t lo;
+			/**< Second 4 flexible bytes */
+		};
+		uint32_t hi;
+		/**< First 4 flexible bytes or FD ID, dependent on
+			PKT_RX_FDIR_* flag in ol_flags. */
+	} fdir;           /**< Filter identifier if FDIR enabled */
+	struct {
+		uint32_t lo;
+		uint32_t hi;
+	} sched;          /**< Hierarchical scheduler */
+	uint32_t usr;	  /**< User defined tags. See rte_distributor_process() */
+} rss_hash_t;
+
 /**
  * The generic rte_mbuf, containing a packet mbuf.
  */
@@ -817,28 +839,31 @@ struct rte_mbuf {
 	uint16_t data_len;        /**< Amount of data in segment buffer. */
 	/** VLAN TCI (CPU order), valid if PKT_RX_VLAN_STRIPPED is set. */
 	uint16_t vlan_tci;
+#ifdef RTE_MBUF_CHAIN_FRIENDLY
+	/*
+	 * Move offload into the second cache line and next in the first.
+	 * Better performance for applications using chained mbufs to have
+	 * the next pointer in the first cache line.
+	 * If you change this structure, you must change the user-mode
+	 * version in rte_kni_common.h to match the new layout.
+	 */
+	uint32_t seqn; /**< Sequence number. See also rte_reorder_insert() */
+	uint16_t vlan_tci_outer;  /**< Outer VLAN Tag Control Identifier (CPU order) */
+	struct rte_mbuf *next;    /**< Next segment of scattered packet. */
+
+	/* second cache line - fields only used in slow path or on TX */
+	MARKER cacheline1 __rte_cache_min_aligned;
+
+	rss_hash_t hash;      /**< hash information */
 
 	union {
-		uint32_t rss;     /**< RSS hash result if RSS enabled */
-		struct {
-			union {
-				struct {
-					uint16_t hash;
-					uint16_t id;
-				};
-				uint32_t lo;
-				/**< Second 4 flexible bytes */
-			};
-			uint32_t hi;
-			/**< First 4 flexible bytes or FD ID, dependent on
-			     PKT_RX_FDIR_* flag in ol_flags. */
-		} fdir;           /**< Filter identifier if FDIR enabled */
-		struct {
-			uint32_t lo;
-			uint32_t hi;
-		} sched;          /**< Hierarchical scheduler */
-		uint32_t usr;	  /**< User defined tags. See rte_distributor_process() */
-	} hash;                   /**< hash information */
+		void *userdata;   /**< Can be used for external metadata */
+		uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
+	};
+
+	struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
+#else
+	rss_hash_t hash;      /**< hash information */
 
 	uint32_t seqn; /**< Sequence number. See also rte_reorder_insert() */
 
@@ -855,7 +880,7 @@ struct rte_mbuf {
 
 	struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
 	struct rte_mbuf *next;    /**< Next segment of scattered packet. */
-
+#endif
 	/* fields to support TX offloads */
 	union {
 		uint64_t tx_offload;       /**< combined for easy fetch */
-- 
2.1.4

^ permalink raw reply	[flat|nested] 8+ messages in thread

end of thread, other threads:[~2016-06-27 14:30 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-06-25 15:29 [dpdk-dev] [PATCH] mbuf:rearrange mbuf to be more mbuf chain friendly Keith Wiles
2016-06-25 15:48 ` Wiles, Keith
2016-06-25 15:55 ` [dpdk-dev] [PATCH v2] " Keith Wiles
2016-06-27  8:21   ` Ananyev, Konstantin
2016-06-27  8:27     ` Olivier Matz
2016-06-27  9:05       ` Thomas Monjalon
2016-06-27 13:06         ` Wiles, Keith
2016-06-27 14:30           ` Thomas Monjalon

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).