Merge branch 'nfp-improve-signal-handing-on-FW-waits-and-flower-contr…

…ol-message-Jakub Kicinski says:

====================
nfp: improve signal handing on FW waits and flower control message processing

The first part of this set aims to improve handling of interrupted
waits.  Patch 1 makes waiting for management FW responses
uninterruptible while patch 2 adds a message when signal arrives
while waiting for an NFP mutex.  We can't interrupt execution of
FW commands so uninterruptible sleep seems reasonable there.
Exiting a wait for a mutex should be clean and have no side affects
so we are allowing to abort it.  Note that both waits have rather
large timeouts (tens of seconds).

Patches 3 and 4 improve flower offload operation under heavy load.
Currently there is no cap on the number of queued FW notifications.
Some of the notifications have to be processed from a workqueue
which may lead to very large number of messages getting queued
if workqueue never gets a chance to run.  Pieter puts a limit
on number of queued messages, tries to drop some messages we ignore
without queuing and process more important messages first.
====================

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

drivers/net/ethernet/netronome/nfp/flower/cmsg.c

@@ -258,9 +258,6 @@ nfp_flower_cmsg_process_one_rx(struct nfp_app *app, struct sk_buff *skb)
 	case NFP_FLOWER_CMSG_TYPE_ACTIVE_TUNS:
 		nfp_tunnel_keep_alive(app, skb);
 		break;
-	case NFP_FLOWER_CMSG_TYPE_TUN_NEIGH:
-		/* Acks from the NFP that the route is added - ignore. */
-		break;
 	default:
 		nfp_flower_cmsg_warn(app, "Cannot handle invalid repr control type %u\n",
 				     type);
@@ -275,18 +272,49 @@ nfp_flower_cmsg_process_one_rx(struct nfp_app *app, struct sk_buff *skb)
 
 void nfp_flower_cmsg_process_rx(struct work_struct *work)
 {
+	struct sk_buff_head cmsg_joined;
 	struct nfp_flower_priv *priv;
 	struct sk_buff *skb;
 
 	priv = container_of(work, struct nfp_flower_priv, cmsg_work);
+	skb_queue_head_init(&cmsg_joined);
+
+	spin_lock_bh(&priv->cmsg_skbs_high.lock);
+	skb_queue_splice_tail_init(&priv->cmsg_skbs_high, &cmsg_joined);
+	spin_unlock_bh(&priv->cmsg_skbs_high.lock);
 
-	while ((skb = skb_dequeue(&priv->cmsg_skbs)))
+	spin_lock_bh(&priv->cmsg_skbs_low.lock);
+	skb_queue_splice_tail_init(&priv->cmsg_skbs_low, &cmsg_joined);
+	spin_unlock_bh(&priv->cmsg_skbs_low.lock);
+
+	while ((skb = __skb_dequeue(&cmsg_joined)))
 		nfp_flower_cmsg_process_one_rx(priv->app, skb);
 }
 
-void nfp_flower_cmsg_rx(struct nfp_app *app, struct sk_buff *skb)
+static void
+nfp_flower_queue_ctl_msg(struct nfp_app *app, struct sk_buff *skb, int type)
 {
 	struct nfp_flower_priv *priv = app->priv;
+	struct sk_buff_head *skb_head;
+
+	if (type == NFP_FLOWER_CMSG_TYPE_PORT_REIFY ||
+	    type == NFP_FLOWER_CMSG_TYPE_PORT_MOD)
+		skb_head = &priv->cmsg_skbs_high;
+	else
+		skb_head = &priv->cmsg_skbs_low;
+
+	if (skb_queue_len(skb_head) >= NFP_FLOWER_WORKQ_MAX_SKBS) {
+		nfp_flower_cmsg_warn(app, "Dropping queued control messages\n");
+		dev_kfree_skb_any(skb);
+		return;
+	}
+
+	skb_queue_tail(skb_head, skb);
+	schedule_work(&priv->cmsg_work);
+}
+
+void nfp_flower_cmsg_rx(struct nfp_app *app, struct sk_buff *skb)
+{
 	struct nfp_flower_cmsg_hdr *cmsg_hdr;
 
 	cmsg_hdr = nfp_flower_cmsg_get_hdr(skb);
@@ -306,8 +334,10 @@ void nfp_flower_cmsg_rx(struct nfp_app *app, struct sk_buff *skb)
 		   nfp_flower_process_mtu_ack(app, skb)) {
 		/* Handle MTU acks outside wq to prevent RTNL conflict. */
 		dev_consume_skb_any(skb);
+	} else if (cmsg_hdr->type == NFP_FLOWER_CMSG_TYPE_TUN_NEIGH) {
+		/* Acks from the NFP that the route is added - ignore. */
+		dev_consume_skb_any(skb);
 	} else {
-		skb_queue_tail(&priv->cmsg_skbs, skb);
-		schedule_work(&priv->cmsg_work);
+		nfp_flower_queue_ctl_msg(app, skb, cmsg_hdr->type);
 	}
 }

drivers/net/ethernet/netronome/nfp/flower/cmsg.h

@@ -108,6 +108,8 @@
 #define NFP_FL_IPV4_TUNNEL_TYPE		GENMASK(7, 4)
 #define NFP_FL_IPV4_PRE_TUN_INDEX	GENMASK(2, 0)
 
+#define NFP_FLOWER_WORKQ_MAX_SKBS	30000
+
 #define nfp_flower_cmsg_warn(app, fmt, args...)                         \
 	do {                                                            \
 		if (net_ratelimit())                                    \

drivers/net/ethernet/netronome/nfp/flower/main.c

@@ -519,7 +519,8 @@ static int nfp_flower_init(struct nfp_app *app)
 
 	app->priv = app_priv;
 	app_priv->app = app;
-	skb_queue_head_init(&app_priv->cmsg_skbs);
+	skb_queue_head_init(&app_priv->cmsg_skbs_high);
+	skb_queue_head_init(&app_priv->cmsg_skbs_low);
 	INIT_WORK(&app_priv->cmsg_work, nfp_flower_cmsg_process_rx);
 	init_waitqueue_head(&app_priv->reify_wait_queue);
 
@@ -549,7 +550,8 @@ static void nfp_flower_clean(struct nfp_app *app)
 {
 	struct nfp_flower_priv *app_priv = app->priv;
 
-	skb_queue_purge(&app_priv->cmsg_skbs);
+	skb_queue_purge(&app_priv->cmsg_skbs_high);
+	skb_queue_purge(&app_priv->cmsg_skbs_low);
 	flush_work(&app_priv->cmsg_work);
 
 	nfp_flower_metadata_cleanup(app);

drivers/net/ethernet/netronome/nfp/flower/main.h

@@ -107,7 +107,10 @@ struct nfp_mtu_conf {
  * @mask_table:		Hash table used to store masks
  * @flow_table:		Hash table used to store flower rules
  * @cmsg_work:		Workqueue for control messages processing
- * @cmsg_skbs:		List of skbs for control message processing
+ * @cmsg_skbs_high:	List of higher priority skbs for control message
+ *			processing
+ * @cmsg_skbs_low:	List of lower priority skbs for control message
+ *			processing
  * @nfp_mac_off_list:	List of MAC addresses to offload
  * @nfp_mac_index_list:	List of unique 8-bit indexes for non NFP netdevs
  * @nfp_ipv4_off_list:	List of IPv4 addresses to offload
@@ -136,7 +139,8 @@ struct nfp_flower_priv {
 	DECLARE_HASHTABLE(mask_table, NFP_FLOWER_MASK_HASH_BITS);
 	DECLARE_HASHTABLE(flow_table, NFP_FLOWER_HASH_BITS);
 	struct work_struct cmsg_work;
-	struct sk_buff_head cmsg_skbs;
+	struct sk_buff_head cmsg_skbs_high;
+	struct sk_buff_head cmsg_skbs_low;
 	struct list_head nfp_mac_off_list;
 	struct list_head nfp_mac_index_list;
 	struct list_head nfp_ipv4_off_list;

drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c

@@ -211,8 +211,11 @@ int nfp_cpp_mutex_lock(struct nfp_cpp_mutex *mutex)
 			break;
 
 		err = msleep_interruptible(timeout_ms);
-		if (err != 0)
+		if (err != 0) {
+			nfp_info(mutex->cpp,
+				 "interrupted waiting for NFP mutex\n");
 			return -ERESTARTSYS;
+		}
 
 		if (time_is_before_eq_jiffies(warn_at)) {
 			warn_at = jiffies + NFP_MUTEX_WAIT_NEXT_WARN * HZ;

drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c

@@ -281,8 +281,7 @@ nfp_nsp_wait_reg(struct nfp_cpp *cpp, u64 *reg, u32 nsp_cpp, u64 addr,
 		if ((*reg & mask) == val)
 			return 0;
 
-		if (msleep_interruptible(25))
-			return -ERESTARTSYS;
+		msleep(25);
 
 		if (time_after(start_time, wait_until))
 			return -ETIMEDOUT;