net: napi: add hard irqs deferral feature

Back in commit 3b47d30396ba ("net: gro: add a per device gro flush timer")
we added the ability to arm one high resolution timer, that we used
to keep not-complete packets in GRO engine a bit longer, hoping that further
frames might be added to them.

Since then, we added the napi_complete_done() interface, and commit
364b6055738b ("net: busy-poll: return busypolling status to drivers")
allowed drivers to avoid re-arming NIC interrupts if we made a promise
that their NAPI poll() handler would be called in the near future.

This infrastructure can be leveraged, thanks to a new device parameter,
which allows to arm the napi hrtimer, instead of re-arming the device
hard IRQ.

We have noticed that on some servers with 32 RX queues or more, the chit-chat
between the NIC and the host caused by IRQ delivery and re-arming could hurt
throughput by ~20% on 100Gbit NIC.

In contrast, hrtimers are using local (percpu) resources and might have lower
cost.

The new tunable, named napi_defer_hard_irqs, is placed in the same hierarchy
than gro_flush_timeout (/sys/class/net/ethX/)

By default, both gro_flush_timeout and napi_defer_hard_irqs are zero.

This patch does not change the prior behavior of gro_flush_timeout
if used alone : NIC hard irqs should be rearmed as before.

One concrete usage can be :

echo 20000 >/sys/class/net/eth1/gro_flush_timeout
echo 10 >/sys/class/net/eth1/napi_defer_hard_irqs

If at least one packet is retired, then we will reset napi counter
to 10 (napi_defer_hard_irqs), ensuring at least 10 periodic scans
of the queue.

On busy queues, this should avoid NIC hard IRQ, while before this patch IRQ
avoidance was only possible if napi->poll() was exhausting its budget
and not call napi_complete_done().

This feature also can be used to work around some non-optimal NIC irq
coalescing strategies.

Having the ability to insert XX usec delays between each napi->poll()
can increase cache efficiency, since we increase batch sizes.

It also keeps serving cpus not idle too long, reducing tail latencies.

Co-developed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 36 insertions, 11 deletions

diff --git a/net/core/dev.c b/net/core/dev.c
index fb61522b1ce1..67585484ad32 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -6227,7 +6227,8 @@ EXPORT_SYMBOL(__napi_schedule_irqoff);
 
 bool napi_complete_done(struct napi_struct *n, int work_done)
 {
-	unsigned long flags, val, new;
+	unsigned long flags, val, new, timeout = 0;
+	bool ret = true;
 
 	/*
 	 * 1) Don't let napi dequeue from the cpu poll list
@@ -6239,20 +6240,23 @@ bool napi_complete_done(struct napi_struct *n, int work_done)
 				 NAPIF_STATE_IN_BUSY_POLL)))
 		return false;
 
-	if (n->gro_bitmask) {
-		unsigned long timeout = 0;
-
-		if (work_done)
+	if (work_done) {
+		if (n->gro_bitmask)
 			timeout = n->dev->gro_flush_timeout;
-
+		n->defer_hard_irqs_count = n->dev->napi_defer_hard_irqs;
+	}
+	if (n->defer_hard_irqs_count > 0) {
+		n->defer_hard_irqs_count--;
+		timeout = n->dev->gro_flush_timeout;
+		if (timeout)
+			ret = false;
+	}
+	if (n->gro_bitmask) {
 		/* When the NAPI instance uses a timeout and keeps postponing
 		 * it, we need to bound somehow the time packets are kept in
 		 * the GRO layer
 		 */
 		napi_gro_flush(n, !!timeout);
-		if (timeout)
-			hrtimer_start(&n->timer, ns_to_ktime(timeout),
-				      HRTIMER_MODE_REL_PINNED);
 	}
 
 	gro_normal_list(n);
@@ -6284,7 +6288,10 @@ bool napi_complete_done(struct napi_struct *n, int work_done)
 		return false;
 	}
 
-	return true;
+	if (timeout)
+		hrtimer_start(&n->timer, ns_to_ktime(timeout),
+			      HRTIMER_MODE_REL_PINNED);
+	return ret;
 }
 EXPORT_SYMBOL(napi_complete_done);
 
@@ -6464,7 +6471,7 @@ static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
 	 */
-	if (napi->gro_bitmask && !napi_disable_pending(napi) &&
+	if (!napi_disable_pending(napi) &&
 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
 		__napi_schedule_irqoff(napi);
 
diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index 0d9e46de205e..f3b650cd0923 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -382,6 +382,23 @@ static ssize_t gro_flush_timeout_store(struct device *dev,
 }
 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 
+static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
+{
+	dev->napi_defer_hard_irqs = val;
+	return 0;
+}
+
+static ssize_t napi_defer_hard_irqs_store(struct device *dev,
+					  struct device_attribute *attr,
+					  const char *buf, size_t len)
+{
+	if (!capable(CAP_NET_ADMIN))
+		return -EPERM;
+
+	return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
+}
+NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
+
 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 			     const char *buf, size_t len)
 {
@@ -545,6 +562,7 @@ static struct attribute *net_class_attrs[] __ro_after_init = {
 	&dev_attr_flags.attr,
 	&dev_attr_tx_queue_len.attr,
 	&dev_attr_gro_flush_timeout.attr,
+	&dev_attr_napi_defer_hard_irqs.attr,
 	&dev_attr_phys_port_id.attr,
 	&dev_attr_phys_port_name.attr,
 	&dev_attr_phys_switch_id.attr,