Merge ra.kernel.org:/pub/scm/linux/kernel/git/davem/net

Overlapping changes in RXRPC, changing to ktime_get_seconds() whilst
adding some tracepoints.

Signed-off-by: David S. Miller <davem@davemloft.net>

7 files changed, 109 insertions, 90 deletions

diff --git a/net/rxrpc/ar-internal.h b/net/rxrpc/ar-internal.h
index 9d9278a13d91..c97558710421 100644
--- a/net/rxrpc/ar-internal.h
+++ b/net/rxrpc/ar-internal.h
@@ -104,9 +104,9 @@ struct rxrpc_net {
 
 #define RXRPC_KEEPALIVE_TIME 20 /* NAT keepalive time in seconds */
 	u8			peer_keepalive_cursor;
-	ktime_t			peer_keepalive_base;
-	struct hlist_head	peer_keepalive[RXRPC_KEEPALIVE_TIME + 1];
-	struct hlist_head	peer_keepalive_new;
+	time64_t		peer_keepalive_base;
+	struct list_head	peer_keepalive[32];
+	struct list_head	peer_keepalive_new;
 	struct timer_list	peer_keepalive_timer;
 	struct work_struct	peer_keepalive_work;
 };
@@ -295,7 +295,7 @@ struct rxrpc_peer {
 	struct hlist_head	error_targets;	/* targets for net error distribution */
 	struct work_struct	error_distributor;
 	struct rb_root		service_conns;	/* Service connections */
-	struct hlist_node	keepalive_link;	/* Link in net->peer_keepalive[] */
+	struct list_head	keepalive_link;	/* Link in net->peer_keepalive[] */
 	time64_t		last_tx_at;	/* Last time packet sent here */
 	seqlock_t		service_conn_lock;
 	spinlock_t		lock;		/* access lock */
diff --git a/net/rxrpc/conn_event.c b/net/rxrpc/conn_event.c
index 84d40ba9856f..6df56ce68861 100644
--- a/net/rxrpc/conn_event.c
+++ b/net/rxrpc/conn_event.c
@@ -138,7 +138,7 @@ static void rxrpc_conn_retransmit_call(struct rxrpc_connection *conn,
 	}
 
 	ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, ioc, len);
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 	if (ret < 0)
 		trace_rxrpc_tx_fail(chan->call_debug_id, serial, ret,
 				    rxrpc_tx_point_call_final_resend);
@@ -252,7 +252,7 @@ static int rxrpc_abort_connection(struct rxrpc_connection *conn,
 
 	trace_rxrpc_tx_packet(conn->debug_id, &whdr, rxrpc_tx_point_conn_abort);
 
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 
 	_leave(" = 0");
 	return 0;
diff --git a/net/rxrpc/net_ns.c b/net/rxrpc/net_ns.c
index 5d6a773db973..417d80867c4f 100644
--- a/net/rxrpc/net_ns.c
+++ b/net/rxrpc/net_ns.c
@@ -85,12 +85,12 @@ static __net_init int rxrpc_init_net(struct net *net)
 	hash_init(rxnet->peer_hash);
 	spin_lock_init(&rxnet->peer_hash_lock);
 	for (i = 0; i < ARRAY_SIZE(rxnet->peer_keepalive); i++)
-		INIT_HLIST_HEAD(&rxnet->peer_keepalive[i]);
-	INIT_HLIST_HEAD(&rxnet->peer_keepalive_new);
+		INIT_LIST_HEAD(&rxnet->peer_keepalive[i]);
+	INIT_LIST_HEAD(&rxnet->peer_keepalive_new);
 	timer_setup(&rxnet->peer_keepalive_timer,
 		    rxrpc_peer_keepalive_timeout, 0);
 	INIT_WORK(&rxnet->peer_keepalive_work, rxrpc_peer_keepalive_worker);
-	rxnet->peer_keepalive_base = ktime_add(ktime_get_real(), NSEC_PER_SEC);
+	rxnet->peer_keepalive_base = ktime_get_seconds();
 
 	ret = -ENOMEM;
 	rxnet->proc_net = proc_net_mkdir(net, "rxrpc", net->proc_net);
diff --git a/net/rxrpc/output.c b/net/rxrpc/output.c
index 801dbf3d3478..ccf5de160444 100644
--- a/net/rxrpc/output.c
+++ b/net/rxrpc/output.c
@@ -209,7 +209,7 @@ int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping,
 	now = ktime_get_real();
 	if (ping)
 		call->ping_time = now;
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 	if (ret < 0)
 		trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 				    rxrpc_tx_point_call_ack);
@@ -299,7 +299,7 @@ int rxrpc_send_abort_packet(struct rxrpc_call *call)
 
 	ret = kernel_sendmsg(conn->params.local->socket,
 			     &msg, iov, 1, sizeof(pkt));
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 	if (ret < 0)
 		trace_rxrpc_tx_fail(call->debug_id, serial, ret,
 				    rxrpc_tx_point_call_abort);
@@ -397,7 +397,7 @@ int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
 	 *     message and update the peer record
 	 */
 	ret = kernel_sendmsg(conn->params.local->socket, &msg, iov, 2, len);
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 
 	up_read(&conn->params.local->defrag_sem);
 	if (ret < 0)
@@ -466,7 +466,7 @@ send_fragmentable:
 		if (ret == 0) {
 			ret = kernel_sendmsg(conn->params.local->socket, &msg,
 					     iov, 2, len);
-			conn->params.peer->last_tx_at = ktime_get_real();
+			conn->params.peer->last_tx_at = ktime_get_seconds();
 
 			opt = IP_PMTUDISC_DO;
 			kernel_setsockopt(conn->params.local->socket, SOL_IP,
@@ -484,7 +484,7 @@ send_fragmentable:
 		if (ret == 0) {
 			ret = kernel_sendmsg(conn->params.local->socket, &msg,
 					     iov, 2, len);
-			conn->params.peer->last_tx_at = ktime_get_real();
+			conn->params.peer->last_tx_at = ktime_get_seconds();
 
 			opt = IPV6_PMTUDISC_DO;
 			kernel_setsockopt(conn->params.local->socket,
@@ -617,6 +617,6 @@ void rxrpc_send_keepalive(struct rxrpc_peer *peer)
 		trace_rxrpc_tx_packet(peer->debug_id, &whdr,
 				      rxrpc_tx_point_version_keepalive);
 
-	peer->last_tx_at = ktime_get_real();
+	peer->last_tx_at = ktime_get_seconds();
 	_leave("");
 }
diff --git a/net/rxrpc/peer_event.c b/net/rxrpc/peer_event.c
index 0ed8b651cec2..4f9da2f51c69 100644
--- a/net/rxrpc/peer_event.c
+++ b/net/rxrpc/peer_event.c
@@ -350,97 +350,117 @@ void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
 }
 
 /*
- * Perform keep-alive pings with VERSION packets to keep any NAT alive.
+ * Perform keep-alive pings.
  */
-void rxrpc_peer_keepalive_worker(struct work_struct *work)
+static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
+					  struct list_head *collector,
+					  time64_t base,
+					  u8 cursor)
 {
-	struct rxrpc_net *rxnet =
-		container_of(work, struct rxrpc_net, peer_keepalive_work);
 	struct rxrpc_peer *peer;
-	unsigned long delay;
-	ktime_t base, now = ktime_get_real();
-	s64 diff;
-	u8 cursor, slot;
+	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
+	time64_t keepalive_at;
+	int slot;
 
-	base = rxnet->peer_keepalive_base;
-	cursor = rxnet->peer_keepalive_cursor;
+	spin_lock_bh(&rxnet->peer_hash_lock);
 
-	_enter("%u,%lld", cursor, ktime_sub(now, base));
+	while (!list_empty(collector)) {
+		peer = list_entry(collector->next,
+				  struct rxrpc_peer, keepalive_link);
 
-next_bucket:
-	diff = ktime_to_ns(ktime_sub(now, base));
-	if (diff < 0)
-		goto resched;
+		list_del_init(&peer->keepalive_link);
+		if (!rxrpc_get_peer_maybe(peer))
+			continue;
 
-	_debug("at %u", cursor);
-	spin_lock_bh(&rxnet->peer_hash_lock);
-next_peer:
-	if (!rxnet->live) {
 		spin_unlock_bh(&rxnet->peer_hash_lock);
-		goto out;
-	}
 
-	/* Everything in the bucket at the cursor is processed this second; the
-	 * bucket at cursor + 1 goes now + 1s and so on...
-	 */
-	if (hlist_empty(&rxnet->peer_keepalive[cursor])) {
-		if (hlist_empty(&rxnet->peer_keepalive_new)) {
-			spin_unlock_bh(&rxnet->peer_hash_lock);
-			goto emptied_bucket;
+		keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
+		slot = keepalive_at - base;
+		_debug("%02x peer %u t=%d {%pISp}",
+		       cursor, peer->debug_id, slot, &peer->srx.transport);
+
+		if (keepalive_at <= base ||
+		    keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
+			rxrpc_send_keepalive(peer);
+			slot = RXRPC_KEEPALIVE_TIME;
 		}
 
-		hlist_move_list(&rxnet->peer_keepalive_new,
-				&rxnet->peer_keepalive[cursor]);
+		/* A transmission to this peer occurred since last we examined
+		 * it so put it into the appropriate future bucket.
+		 */
+		slot += cursor;
+		slot &= mask;
+		spin_lock_bh(&rxnet->peer_hash_lock);
+		list_add_tail(&peer->keepalive_link,
+			      &rxnet->peer_keepalive[slot & mask]);
+		rxrpc_put_peer(peer);
 	}
 
-	peer = hlist_entry(rxnet->peer_keepalive[cursor].first,
-			   struct rxrpc_peer, keepalive_link);
-	hlist_del_init(&peer->keepalive_link);
-	if (!rxrpc_get_peer_maybe(peer))
-		goto next_peer;
-
 	spin_unlock_bh(&rxnet->peer_hash_lock);
+}
 
-	_debug("peer %u {%pISp}", peer->debug_id, &peer->srx.transport);
+/*
+ * Perform keep-alive pings with VERSION packets to keep any NAT alive.
+ */
+void rxrpc_peer_keepalive_worker(struct work_struct *work)
+{
+	struct rxrpc_net *rxnet =
+		container_of(work, struct rxrpc_net, peer_keepalive_work);
+	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
+	time64_t base, now, delay;
+	u8 cursor, stop;
+	LIST_HEAD(collector);
 
-recalc:
-	diff = ktime_divns(ktime_sub(peer->last_tx_at, base), NSEC_PER_SEC);
-	if (diff < -30 || diff > 30)
-		goto send; /* LSW of 64-bit time probably wrapped on 32-bit */
-	diff += RXRPC_KEEPALIVE_TIME - 1;
-	if (diff < 0)
-		goto send;
+	now = ktime_get_seconds();
+	base = rxnet->peer_keepalive_base;
+	cursor = rxnet->peer_keepalive_cursor;
+	_enter("%lld,%u", base - now, cursor);
 
-	slot = (diff > RXRPC_KEEPALIVE_TIME - 1) ? RXRPC_KEEPALIVE_TIME - 1 : diff;
-	if (slot == 0)
-		goto send;
+	if (!rxnet->live)
+		return;
 
-	/* A transmission to this peer occurred since last we examined it so
-	 * put it into the appropriate future bucket.
+	/* Remove to a temporary list all the peers that are currently lodged
+	 * in expired buckets plus all new peers.
+	 *
+	 * Everything in the bucket at the cursor is processed this
+	 * second; the bucket at cursor + 1 goes at now + 1s and so
+	 * on...
 	 */
-	slot = (slot + cursor) % ARRAY_SIZE(rxnet->peer_keepalive);
 	spin_lock_bh(&rxnet->peer_hash_lock);
-	hlist_add_head(&peer->keepalive_link, &rxnet->peer_keepalive[slot]);
-	rxrpc_put_peer(peer);
-	goto next_peer;
-
-send:
-	rxrpc_send_keepalive(peer);
-	now = ktime_get_real();
-	goto recalc;
+	list_splice_init(&rxnet->peer_keepalive_new, &collector);
+
+	stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
+	while (base <= now && (s8)(cursor - stop) < 0) {
+		list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
+				      &collector);
+		base++;
+		cursor++;
+	}
 
-emptied_bucket:
-	cursor++;
-	if (cursor >= ARRAY_SIZE(rxnet->peer_keepalive))
-		cursor = 0;
-	base = ktime_add_ns(base, NSEC_PER_SEC);
-	goto next_bucket;
+	base = now;
+	spin_unlock_bh(&rxnet->peer_hash_lock);
 
-resched:
 	rxnet->peer_keepalive_base = base;
 	rxnet->peer_keepalive_cursor = cursor;
-	delay = nsecs_to_jiffies(-diff) + 1;
-	timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
-out:
+	rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
+	ASSERT(list_empty(&collector));
+
+	/* Schedule the timer for the next occupied timeslot. */
+	cursor = rxnet->peer_keepalive_cursor;
+	stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
+	for (; (s8)(cursor - stop) < 0; cursor++) {
+		if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
+			break;
+		base++;
+	}
+
+	now = ktime_get_seconds();
+	delay = base - now;
+	if (delay < 1)
+		delay = 1;
+	delay *= HZ;
+	if (rxnet->live)
+		timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
+
 	_leave("");
 }
diff --git a/net/rxrpc/peer_object.c b/net/rxrpc/peer_object.c
index 1b7e8107b3ae..24ec7cdcf332 100644
--- a/net/rxrpc/peer_object.c
+++ b/net/rxrpc/peer_object.c
@@ -322,7 +322,7 @@ struct rxrpc_peer *rxrpc_lookup_incoming_peer(struct rxrpc_local *local,
 	if (!peer) {
 		peer = prealloc;
 		hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key);
-		hlist_add_head(&peer->keepalive_link, &rxnet->peer_keepalive_new);
+		list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new);
 	}
 
 	spin_unlock(&rxnet->peer_hash_lock);
@@ -367,8 +367,8 @@ struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local,
 		if (!peer) {
 			hash_add_rcu(rxnet->peer_hash,
 				     &candidate->hash_link, hash_key);
-			hlist_add_head(&candidate->keepalive_link,
-				       &rxnet->peer_keepalive_new);
+			list_add_tail(&candidate->keepalive_link,
+				      &rxnet->peer_keepalive_new);
 		}
 
 		spin_unlock_bh(&rxnet->peer_hash_lock);
@@ -441,7 +441,7 @@ static void __rxrpc_put_peer(struct rxrpc_peer *peer)
 
 	spin_lock_bh(&rxnet->peer_hash_lock);
 	hash_del_rcu(&peer->hash_link);
-	hlist_del_init(&peer->keepalive_link);
+	list_del_init(&peer->keepalive_link);
 	spin_unlock_bh(&rxnet->peer_hash_lock);
 
 	kfree_rcu(peer, rcu);
diff --git a/net/rxrpc/rxkad.c b/net/rxrpc/rxkad.c
index eaf8f4f446b0..cea16838d588 100644
--- a/net/rxrpc/rxkad.c
+++ b/net/rxrpc/rxkad.c
@@ -670,7 +670,7 @@ static int rxkad_issue_challenge(struct rxrpc_connection *conn)
 		return -EAGAIN;
 	}
 
-	conn->params.peer->last_tx_at = ktime_get_real();
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 	trace_rxrpc_tx_packet(conn->debug_id, &whdr,
 			      rxrpc_tx_point_rxkad_challenge);
 	_leave(" = 0");
@@ -728,8 +728,7 @@ static int rxkad_send_response(struct rxrpc_connection *conn,
 		return -EAGAIN;
 	}
 
-	conn->params.peer->last_tx_at = ktime_get_real();
-	trace_rxrpc_tx_packet(0, &whdr, rxrpc_tx_point_rxkad_response);
+	conn->params.peer->last_tx_at = ktime_get_seconds();
 	_leave(" = 0");
 	return 0;
 }