Add optimizing LRO wrapper:

- Add optimizing LRO wrapper which pre-sorts all incoming packets
  according to the hash type and flowid. This prevents exhaustion of
  the LRO entries due to too many connections at the same time.
  Testing using a larger number of higher bandwidth TCP connections
  showed that the incoming ACK packet aggregation rate increased from
  ~1.3:1 to almost 3:1. Another test showed that for a number of TCP
  connections greater than 16 per hardware receive ring, where 8 TCP
  connections was the LRO active entry limit, there was a significant
  improvement in throughput due to being able to fully aggregate more
  than 8 TCP stream. For very few very high bandwidth TCP streams, the
  optimizing LRO wrapper will add CPU usage instead of reducing CPU
  usage. This is expected. Network drivers which want to use the
  optimizing LRO wrapper needs to call "tcp_lro_queue_mbuf()" instead
  of "tcp_lro_rx()" and "tcp_lro_flush_all()" instead of
  "tcp_lro_flush()". Further the LRO control structure must be
  initialized using "tcp_lro_init_args()" passing a non-zero number
  into the "lro_mbufs" argument.

- Make LRO statistics 64-bit. Previously 32-bit integers were used for
  statistics which can be prone to wrap-around. Fix this while at it
  and update all SYSCTL's which expose LRO statistics.

- Ensure all data is freed when destroying a LRO control structures,
  especially leftover LRO entries.

- Reduce number of memory allocations needed when setting up a LRO
  control structure by precomputing the total amount of memory needed.

- Add own memory allocation counter for LRO.

- Bump the FreeBSD version to force recompilation of all KLDs due to
  change of the LRO control structure size.

Sponsored by:	Mellanox Technologies
Reviewed by:	gallatin, sbruno, rrs, gnn, transport
Tested by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D4914

1 files changed, 162 insertions, 22 deletions

diff --git a/sys/netinet/tcp_lro.c b/sys/netinet/tcp_lro.c
index 83ab2e2072d4..566334348810 100644
--- a/sys/netinet/tcp_lro.c
+++ b/sys/netinet/tcp_lro.c
@@ -2,6 +2,7 @@
  * Copyright (c) 2007, Myricom Inc.
  * Copyright (c) 2008, Intel Corporation.
  * Copyright (c) 2012 The FreeBSD Foundation
+ * Copyright (c) 2016 Mellanox Technologies.
  * All rights reserved.
  *
  * Portions of this software were developed by Bjoern Zeeb
@@ -58,9 +59,7 @@ __FBSDID("$FreeBSD$");
 
 #include <machine/in_cksum.h>
 
-#ifndef LRO_ENTRIES
-#define	LRO_ENTRIES	8	/* # of LRO entries per RX queue. */
-#endif
+static MALLOC_DEFINE(M_LRO, "LRO", "LRO control structures");
 
 #define	TCP_LRO_UPDATE_CSUM	1
 #ifndef	TCP_LRO_UPDATE_CSUM
@@ -70,42 +69,73 @@ __FBSDID("$FreeBSD$");
 int
 tcp_lro_init(struct lro_ctrl *lc)
 {
+	return (tcp_lro_init_args(lc, NULL, TCP_LRO_ENTRIES, 0));
+}
+
+int
+tcp_lro_init_args(struct lro_ctrl *lc, struct ifnet *ifp,
+    unsigned lro_entries, unsigned lro_mbufs)
+{
 	struct lro_entry *le;
-	int error, i;
+	size_t size;
+	unsigned i;
 
 	lc->lro_bad_csum = 0;
 	lc->lro_queued = 0;
 	lc->lro_flushed = 0;
 	lc->lro_cnt = 0;
+	lc->lro_mbuf_count = 0;
+	lc->lro_mbuf_max = lro_mbufs;
+	lc->lro_cnt = lro_entries;
+	lc->ifp = ifp;
 	SLIST_INIT(&lc->lro_free);
 	SLIST_INIT(&lc->lro_active);
 
-	error = 0;
-	for (i = 0; i < LRO_ENTRIES; i++) {
-		le = (struct lro_entry *)malloc(sizeof(*le), M_DEVBUF,
-		    M_NOWAIT | M_ZERO);
-                if (le == NULL) {
-			if (i == 0)
-				error = ENOMEM;
-                        break;
-                }
-		lc->lro_cnt = i + 1;
-		SLIST_INSERT_HEAD(&lc->lro_free, le, next);
-        }
-
-	return (error);
+	/* compute size to allocate */
+	size = (lro_mbufs * sizeof(struct mbuf *)) +
+	    (lro_entries * sizeof(*le));
+	lc->lro_mbuf_data = (struct mbuf **)
+	    malloc(size, M_LRO, M_NOWAIT | M_ZERO);
+
+	/* check for out of memory */
+	if (lc->lro_mbuf_data == NULL) {
+		memset(lc, 0, sizeof(*lc));
+		return (ENOMEM);
+	}
+	/* compute offset for LRO entries */
+	le = (struct lro_entry *)
+	    (lc->lro_mbuf_data + lro_mbufs);
+
+	/* setup linked list */
+	for (i = 0; i != lro_entries; i++)
+		SLIST_INSERT_HEAD(&lc->lro_free, le + i, next);
+
+	return (0);
 }
 
 void
 tcp_lro_free(struct lro_ctrl *lc)
 {
 	struct lro_entry *le;
+	unsigned x;
+
+	/* reset LRO free list */
+	SLIST_INIT(&lc->lro_free);
 
-	while (!SLIST_EMPTY(&lc->lro_free)) {
-		le = SLIST_FIRST(&lc->lro_free);
-		SLIST_REMOVE_HEAD(&lc->lro_free, next);
-		free(le, M_DEVBUF);
+	/* free active mbufs, if any */
+	while ((le = SLIST_FIRST(&lc->lro_active)) != NULL) {
+		SLIST_REMOVE_HEAD(&lc->lro_active, next);
+		m_freem(le->m_head);
 	}
+
+	/* free mbuf array, if any */
+	for (x = 0; x != lc->lro_mbuf_count; x++)
+		m_freem(lc->lro_mbuf_data[x]);
+	lc->lro_mbuf_count = 0;
+	
+	/* free allocated memory, if any */
+	free(lc->lro_mbuf_data, M_LRO);
+	lc->lro_mbuf_data = NULL;
 }
 
 #ifdef TCP_LRO_UPDATE_CSUM
@@ -305,6 +335,83 @@ tcp_lro_flush(struct lro_ctrl *lc, struct lro_entry *le)
 	SLIST_INSERT_HEAD(&lc->lro_free, le, next);
 }
 
+static int
+tcp_lro_mbuf_compare_header(const void *ppa, const void *ppb)
+{
+	const struct mbuf *ma = *((const struct mbuf * const *)ppa);
+	const struct mbuf *mb = *((const struct mbuf * const *)ppb);
+	int ret;
+
+	ret = M_HASHTYPE_GET(ma) - M_HASHTYPE_GET(mb);
+	if (ret != 0)
+		goto done;
+
+	ret = ma->m_pkthdr.flowid - mb->m_pkthdr.flowid;
+	if (ret != 0)
+		goto done;
+
+	ret = TCP_LRO_SEQUENCE(ma) - TCP_LRO_SEQUENCE(mb);
+done:
+	return (ret);
+}
+
+void
+tcp_lro_flush_all(struct lro_ctrl *lc)
+{
+	struct lro_entry *le;
+	uint32_t hashtype;
+	uint32_t flowid;
+	unsigned x;
+
+	/* check if no mbufs to flush */
+	if (__predict_false(lc->lro_mbuf_count == 0))
+		goto done;
+
+	/* sort all mbufs according to stream */
+	qsort(lc->lro_mbuf_data, lc->lro_mbuf_count, sizeof(struct mbuf *),
+	    &tcp_lro_mbuf_compare_header);
+
+	/* input data into LRO engine, stream by stream */
+	flowid = 0;
+	hashtype = M_HASHTYPE_NONE;
+	for (x = 0; x != lc->lro_mbuf_count; x++) {
+		struct mbuf *mb;
+
+		mb = lc->lro_mbuf_data[x];
+
+		/* check for new stream */
+		if (mb->m_pkthdr.flowid != flowid ||
+		    M_HASHTYPE_GET(mb) != hashtype) {
+			flowid = mb->m_pkthdr.flowid;
+			hashtype = M_HASHTYPE_GET(mb);
+
+			/* flush active streams */
+			while ((le = SLIST_FIRST(&lc->lro_active)) != NULL) {
+				SLIST_REMOVE_HEAD(&lc->lro_active, next);
+				tcp_lro_flush(lc, le);
+			}
+		}
+#ifdef TCP_LRO_RESET_SEQUENCE
+		/* reset sequence number */
+		TCP_LRO_SEQUENCE(mb) = 0;
+#endif
+		/* add packet to LRO engine */
+		if (tcp_lro_rx(lc, mb, 0) != 0) {
+			/* input packet to network layer */
+			(*lc->ifp->if_input)(lc->ifp, mb);
+			lc->lro_queued++;
+			lc->lro_flushed++;
+		}
+	}
+done:
+	/* flush active streams */
+	while ((le = SLIST_FIRST(&lc->lro_active)) != NULL) {
+		SLIST_REMOVE_HEAD(&lc->lro_active, next);
+		tcp_lro_flush(lc, le);
+	}
+	lc->lro_mbuf_count = 0;
+}
+
 #ifdef INET6
 static int
 tcp_lro_rx_ipv6(struct lro_ctrl *lc, struct mbuf *m, struct ip6_hdr *ip6,
@@ -633,4 +740,37 @@ tcp_lro_rx(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum)
 	return (0);
 }
 
+void
+tcp_lro_queue_mbuf(struct lro_ctrl *lc, struct mbuf *mb)
+{
+	/* sanity checks */
+	if (__predict_false(lc->ifp == NULL || lc->lro_mbuf_data == NULL ||
+	    lc->lro_mbuf_max == 0)) {
+		/* packet drop */
+		m_freem(mb);
+		return;
+	}
+
+	/* check if packet is not LRO capable */
+	if (__predict_false(mb->m_pkthdr.csum_flags == 0 ||
+	    (lc->ifp->if_capenable & IFCAP_LRO) == 0)) {
+		lc->lro_flushed++;
+		lc->lro_queued++;
+
+		/* input packet to network layer */
+		(*lc->ifp->if_input) (lc->ifp, mb);
+		return;
+	}
+
+	/* check if array is full */
+	if (__predict_false(lc->lro_mbuf_count == lc->lro_mbuf_max))
+		tcp_lro_flush_all(lc);
+
+	/* store sequence number */
+	TCP_LRO_SEQUENCE(mb) = lc->lro_mbuf_count;
+
+	/* enter mbuf */
+	lc->lro_mbuf_data[lc->lro_mbuf_count++] = mb;
+}
+
 /* end */