Merge branch 'net-rps-rfs-improvements'

Eric Dumazet says:

====================
net: rps/rfs improvements

Jason Xing attempted to optimize napi_schedule_rps() by avoiding
unneeded NET_RX_SOFTIRQ raises: [1], [2]

This is quite complex to implement properly. I chose to implement
the idea, and added a similar optimization in ____napi_schedule()

Overall, in an intensive RPC workload, with 32 TX/RX queues with RFS
I was able to observe a ~10% reduction of NET_RX_SOFTIRQ
invocations.

While this had no impact on throughput or cpu costs on this synthetic
benchmark, we know that firing NET_RX_SOFTIRQ from softirq handler
can force __do_softirq() to wakeup ksoftirqd when need_resched() is true.
This can have a latency impact on stressed hosts.

[1] https://lore.kernel.org/lkml/20230325152417.5403-1-kerneljasonxing@gmail.com/
[2] https://lore.kernel.org/netdev/20230328142112.12493-1-kerneljasonxing@gmail.com/
====================

Link: https://lore.kernel.org/r/20230328235021.1048163-1-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

include/linux/netdevice.h

@@ -3188,6 +3188,7 @@ struct softnet_data {
 #ifdef CONFIG_RPS
 	struct softnet_data	*rps_ipi_list;
 #endif
+	bool			in_net_rx_action;
 #ifdef CONFIG_NET_FLOW_LIMIT
 	struct sd_flow_limit __rcu *flow_limit;
 #endif

net/core/dev.c

@@ -4360,7 +4360,11 @@ static inline void ____napi_schedule(struct softnet_data *sd,
 	}
 
 	list_add_tail(&napi->poll_list, &sd->poll_list);
-	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
+	/* If not called from net_rx_action()
+	 * we have to raise NET_RX_SOFTIRQ.
+	 */
+	if (!sd->in_net_rx_action)
+		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
 }
 
 #ifdef CONFIG_RPS
@@ -4582,11 +4586,16 @@ static void trigger_rx_softirq(void *data)
 }
 
 /*
- * Check if this softnet_data structure is another cpu one
- * If yes, queue it to our IPI list and return 1
- * If no, return 0
+ * After we queued a packet into sd->input_pkt_queue,
+ * we need to make sure this queue is serviced soon.
+ *
+ * - If this is another cpu queue, link it to our rps_ipi_list,
+ *   and make sure we will process rps_ipi_list from net_rx_action().
+ *
+ * - If this is our own queue, NAPI schedule our backlog.
+ *   Note that this also raises NET_RX_SOFTIRQ.
  */
-static int napi_schedule_rps(struct softnet_data *sd)
+static void napi_schedule_rps(struct softnet_data *sd)
 {
 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
 
@@ -4595,12 +4604,15 @@ static int napi_schedule_rps(struct softnet_data *sd)
 		sd->rps_ipi_next = mysd->rps_ipi_list;
 		mysd->rps_ipi_list = sd;
 
-		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
-		return 1;
+		/* If not called from net_rx_action()
+		 * we have to raise NET_RX_SOFTIRQ.
+		 */
+		if (!mysd->in_net_rx_action)
+			__raise_softirq_irqoff(NET_RX_SOFTIRQ);
+		return;
 	}
 #endif /* CONFIG_RPS */
 	__napi_schedule_irqoff(&mysd->backlog);
-	return 0;
 }
 
 #ifdef CONFIG_NET_FLOW_LIMIT
@@ -6640,6 +6652,8 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
 	LIST_HEAD(list);
 	LIST_HEAD(repoll);
 
+start:
+	sd->in_net_rx_action = true;
 	local_irq_disable();
 	list_splice_init(&sd->poll_list, &list);
 	local_irq_enable();
@@ -6650,8 +6664,18 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
 		skb_defer_free_flush(sd);
 
 		if (list_empty(&list)) {
-			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
-				goto end;
+			if (list_empty(&repoll)) {
+				sd->in_net_rx_action = false;
+				barrier();
+				/* We need to check if ____napi_schedule()
+				 * had refilled poll_list while
+				 * sd->in_net_rx_action was true.
+				 */
+				if (!list_empty(&sd->poll_list))
+					goto start;
+				if (!sd_has_rps_ipi_waiting(sd))
+					goto end;
+			}
 			break;
 		}
 
@@ -6676,6 +6700,8 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
 	list_splice(&list, &sd->poll_list);
 	if (!list_empty(&sd->poll_list))
 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
+	else
+		sd->in_net_rx_action = false;
 
 	net_rps_action_and_irq_enable(sd);
 end:;