Merge r254336 from user/np/cxl_tuning.

Add a last-modified timestamp to each LRO entry and provide an interface
to flush all inactive entries.  Drivers decide when to flush and what
the inactivity threshold should be.

Network drivers that process an rx queue to completion can enter a
livelock type situation when the rate at which packets are received
reaches equilibrium with the rate at which the rx thread is processing
them.  When this happens the final LRO flush (normally when the rx
routine is done) does not occur.  Pure ACKs and segments with total
payload < 64K can get stuck in an LRO entry.  Symptoms are that TCP
tx-mostly connections' performance falls off a cliff during heavy,
unrelated rx on the interface.

Flushing only inactive LRO entries works better than any of these
alternates that I tried:
- don't LRO pure ACKs
- flush _all_ LRO entries periodically (every 'x' microseconds or every
  'y' descriptors)
- stop rx processing in the driver periodically and schedule remaining
  work for later.

Reviewed by:	andre

2 files changed, 26 insertions, 1 deletions

diff --git a/sys/netinet/tcp_lro.c b/sys/netinet/tcp_lro.c
index 032d47c..63a6bba 100644
--- a/sys/netinet/tcp_lro.c
+++ b/sys/netinet/tcp_lro.c
@@ -194,6 +194,25 @@ tcp_lro_rx_csum_fixup(struct lro_entry *le, void *l3hdr, struct tcphdr *th,
 #endif
 
 void
+tcp_lro_flush_inactive(struct lro_ctrl *lc, const struct timeval *timeout)
+{
+	struct lro_entry *le, *le_tmp;
+	struct timeval tv;
+
+	if (SLIST_EMPTY(&lc->lro_active))
+		return;
+
+	getmicrotime(&tv);
+	timevalsub(&tv, timeout);
+	SLIST_FOREACH_SAFE(le, &lc->lro_active, next, le_tmp) {
+		if (timevalcmp(&tv, &le->mtime, >=)) {
+			SLIST_REMOVE(&lc->lro_active, le, lro_entry, next);
+			tcp_lro_flush(lc, le);
+		}
+	}
+}
+
+void
 tcp_lro_flush(struct lro_ctrl *lc, struct lro_entry *le)
 {
 
@@ -543,7 +562,8 @@ tcp_lro_rx(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum)
 		if (le->p_len > (65535 - lc->ifp->if_mtu)) {
 			SLIST_REMOVE(&lc->lro_active, le, lro_entry, next);
 			tcp_lro_flush(lc, le);
-		}
+		} else
+			getmicrotime(&le->mtime);
 
 		return (0);
 	}
@@ -556,6 +576,7 @@ tcp_lro_rx(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum)
 	le = SLIST_FIRST(&lc->lro_free);
 	SLIST_REMOVE_HEAD(&lc->lro_free, next);
 	SLIST_INSERT_HEAD(&lc->lro_active, le, next);
+	getmicrotime(&le->mtime);
 
 	/* Start filling in details. */
 	switch (eh_type) {
diff --git a/sys/netinet/tcp_lro.h b/sys/netinet/tcp_lro.h
index b3a5017..ab6d74a 100644
--- a/sys/netinet/tcp_lro.h
+++ b/sys/netinet/tcp_lro.h
@@ -30,6 +30,8 @@
 #ifndef _TCP_LRO_H_
 #define _TCP_LRO_H_
 
+#include <sys/time.h>
+
 struct lro_entry
 {
 	SLIST_ENTRY(lro_entry)	next;
@@ -59,6 +61,7 @@ struct lro_entry
 	uint32_t		tsecr;
 	uint16_t		window;
 	uint16_t		timestamp;	/* flag, not a TCP hdr field. */
+	struct timeval		mtime;
 };
 SLIST_HEAD(lro_head, lro_entry);
 
@@ -83,6 +86,7 @@ struct lro_ctrl {
 
 int tcp_lro_init(struct lro_ctrl *);
 void tcp_lro_free(struct lro_ctrl *);
+void tcp_lro_flush_inactive(struct lro_ctrl *, const struct timeval *);
 void tcp_lro_flush(struct lro_ctrl *, struct lro_entry *);
 int tcp_lro_rx(struct lro_ctrl *, struct mbuf *, uint32_t);