ixgbevf: implement CONFIG_NET_RX_BUSY_POLL

This patch enables CONFIG_NET_RX_BUSY_POLL support in the VF code. This enables
sockets which have enabled the SO_BUSY_POLL socket option to use the
ndo_busy_poll_recv operation which could result in lower latency, at the cost
of higher CPU utilization, and increased power usage. This support is similar
to how the ixgbe driver works.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

2 files changed, 177 insertions, 0 deletions

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
index d7837dcc9897..85c7560e237b 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
@@ -38,6 +38,10 @@
 
 #include "vf.h"
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+#include <net/busy_poll.h>
+#endif
+
 /* wrapper around a pointer to a socket buffer,
  * so a DMA handle can be stored along with the buffer */
 struct ixgbevf_tx_buffer {
@@ -145,7 +149,112 @@ struct ixgbevf_q_vector {
 	struct napi_struct napi;
 	struct ixgbevf_ring_container rx, tx;
 	char name[IFNAMSIZ + 9];
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	unsigned int state;
+#define IXGBEVF_QV_STATE_IDLE		0
+#define IXGBEVF_QV_STATE_NAPI		1    /* NAPI owns this QV */
+#define IXGBEVF_QV_STATE_POLL		2    /* poll owns this QV */
+#define IXGBEVF_QV_STATE_DISABLED	4    /* QV is disabled */
+#define IXGBEVF_QV_OWNED (IXGBEVF_QV_STATE_NAPI | IXGBEVF_QV_STATE_POLL)
+#define IXGBEVF_QV_LOCKED (IXGBEVF_QV_OWNED | IXGBEVF_QV_STATE_DISABLED)
+#define IXGBEVF_QV_STATE_NAPI_YIELD	8    /* NAPI yielded this QV */
+#define IXGBEVF_QV_STATE_POLL_YIELD	16   /* poll yielded this QV */
+#define IXGBEVF_QV_YIELD (IXGBEVF_QV_STATE_NAPI_YIELD | IXGBEVF_QV_STATE_POLL_YIELD)
+#define IXGBEVF_QV_USER_PEND (IXGBEVF_QV_STATE_POLL | IXGBEVF_QV_STATE_POLL_YIELD)
+	spinlock_t lock;
+#endif /* CONFIG_NET_RX_BUSY_POLL */
 };
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static inline void ixgbevf_qv_init_lock(struct ixgbevf_q_vector *q_vector)
+{
+
+	spin_lock_init(&q_vector->lock);
+	q_vector->state = IXGBEVF_QV_STATE_IDLE;
+}
+
+/* called from the device poll routine to get ownership of a q_vector */
+static inline bool ixgbevf_qv_lock_napi(struct ixgbevf_q_vector *q_vector)
+{
+	int rc = true;
+	spin_lock_bh(&q_vector->lock);
+	if (q_vector->state & IXGBEVF_QV_LOCKED) {
+		WARN_ON(q_vector->state & IXGBEVF_QV_STATE_NAPI);
+		q_vector->state |= IXGBEVF_QV_STATE_NAPI_YIELD;
+		rc = false;
+	} else {
+		/* we don't care if someone yielded */
+		q_vector->state = IXGBEVF_QV_STATE_NAPI;
+	}
+	spin_unlock_bh(&q_vector->lock);
+	return rc;
+}
+
+/* returns true is someone tried to get the qv while napi had it */
+static inline bool ixgbevf_qv_unlock_napi(struct ixgbevf_q_vector *q_vector)
+{
+	int rc = false;
+	spin_lock_bh(&q_vector->lock);
+	WARN_ON(q_vector->state & (IXGBEVF_QV_STATE_POLL |
+				   IXGBEVF_QV_STATE_NAPI_YIELD));
+
+	if (q_vector->state & IXGBEVF_QV_STATE_POLL_YIELD)
+		rc = true;
+	/* reset state to idle, unless QV is disabled */
+	q_vector->state &= IXGBEVF_QV_STATE_DISABLED;
+	spin_unlock_bh(&q_vector->lock);
+	return rc;
+}
+
+/* called from ixgbevf_low_latency_poll() */
+static inline bool ixgbevf_qv_lock_poll(struct ixgbevf_q_vector *q_vector)
+{
+	int rc = true;
+	spin_lock_bh(&q_vector->lock);
+	if ((q_vector->state & IXGBEVF_QV_LOCKED)) {
+		q_vector->state |= IXGBEVF_QV_STATE_POLL_YIELD;
+		rc = false;
+	} else {
+		/* preserve yield marks */
+		q_vector->state |= IXGBEVF_QV_STATE_POLL;
+	}
+	spin_unlock_bh(&q_vector->lock);
+	return rc;
+}
+
+/* returns true if someone tried to get the qv while it was locked */
+static inline bool ixgbevf_qv_unlock_poll(struct ixgbevf_q_vector *q_vector)
+{
+	int rc = false;
+	spin_lock_bh(&q_vector->lock);
+	WARN_ON(q_vector->state & (IXGBEVF_QV_STATE_NAPI));
+
+	if (q_vector->state & IXGBEVF_QV_STATE_POLL_YIELD)
+		rc = true;
+	/* reset state to idle, unless QV is disabled */
+	q_vector->state &= IXGBEVF_QV_STATE_DISABLED;
+	spin_unlock_bh(&q_vector->lock);
+	return rc;
+}
+
+/* true if a socket is polling, even if it did not get the lock */
+static inline bool ixgbevf_qv_busy_polling(struct ixgbevf_q_vector *q_vector)
+{
+	WARN_ON(!(q_vector->state & IXGBEVF_QV_OWNED));
+	return q_vector->state & IXGBEVF_QV_USER_PEND;
+}
+
+/* false if QV is currently owned */
+static inline bool ixgbevf_qv_disable(struct ixgbevf_q_vector *q_vector)
+{
+	int rc = true;
+	spin_lock_bh(&q_vector->lock);
+	if (q_vector->state & IXGBEVF_QV_OWNED)
+		rc = false;
+	spin_unlock_bh(&q_vector->lock);
+	return rc;
+}
+
+#endif /* CONFIG_NET_RX_BUSY_POLL */
 
 /*
  * microsecond values for various ITR rates shifted by 2 to fit itr register
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 06e29bd73777..bc03a2ec5c0a 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -310,6 +310,16 @@ static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 			   struct sk_buff *skb, u8 status,
 			   union ixgbe_adv_rx_desc *rx_desc)
 {
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	skb_mark_napi_id(skb, &q_vector->napi);
+
+	if (ixgbevf_qv_busy_polling(q_vector)) {
+		netif_receive_skb(skb);
+		/* exit early if we busy polled */
+		return;
+	}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
 	ixgbevf_receive_skb(q_vector, skb, status, rx_desc);
 }
 
@@ -563,6 +573,11 @@ static int ixgbevf_poll(struct napi_struct *napi, int budget)
 	ixgbevf_for_each_ring(ring, q_vector->tx)
 		clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	if (!ixgbevf_qv_lock_napi(q_vector))
+		return budget;
+#endif
+
 	/* attempt to distribute budget to each queue fairly, but don't allow
 	 * the budget to go below 1 because we'll exit polling */
 	if (q_vector->rx.count > 1)
@@ -577,6 +592,10 @@ static int ixgbevf_poll(struct napi_struct *napi, int budget)
 				   < per_ring_budget);
 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	ixgbevf_qv_unlock_napi(q_vector);
+#endif
+
 	/* If all work not completed, return budget and keep polling */
 	if (!clean_complete)
 		return budget;
@@ -611,6 +630,34 @@ void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
 }
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+/* must be called with local_bh_disable()d */
+static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
+{
+	struct ixgbevf_q_vector *q_vector =
+			container_of(napi, struct ixgbevf_q_vector, napi);
+	struct ixgbevf_adapter *adapter = q_vector->adapter;
+	struct ixgbevf_ring  *ring;
+	int found = 0;
+
+	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
+		return LL_FLUSH_FAILED;
+
+	if (!ixgbevf_qv_lock_poll(q_vector))
+		return LL_FLUSH_BUSY;
+
+	ixgbevf_for_each_ring(ring, q_vector->rx) {
+		found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
+		if (found)
+			break;
+	}
+
+	ixgbevf_qv_unlock_poll(q_vector);
+
+	return found;
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
 /**
  * ixgbevf_configure_msix - Configure MSI-X hardware
  * @adapter: board private structure
@@ -1297,6 +1344,9 @@ static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
 
 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
 		q_vector = adapter->q_vector[q_idx];
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
+#endif
 		napi_enable(&q_vector->napi);
 	}
 }
@@ -1310,6 +1360,12 @@ static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
 		q_vector = adapter->q_vector[q_idx];
 		napi_disable(&q_vector->napi);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
+			pr_info("QV %d locked\n", q_idx);
+			usleep_range(1000, 20000);
+		}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
 	}
 }
 
@@ -1960,6 +2016,9 @@ static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
 		q_vector->v_idx = q_idx;
 		netif_napi_add(adapter->netdev, &q_vector->napi,
 			       ixgbevf_poll, 64);
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		napi_hash_add(&q_vector->napi);
+#endif
 		adapter->q_vector[q_idx] = q_vector;
 	}
 
@@ -1969,6 +2028,9 @@ err_out:
 	while (q_idx) {
 		q_idx--;
 		q_vector = adapter->q_vector[q_idx];
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		napi_hash_del(&q_vector->napi);
+#endif
 		netif_napi_del(&q_vector->napi);
 		kfree(q_vector);
 		adapter->q_vector[q_idx] = NULL;
@@ -1992,6 +2054,9 @@ static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
 
 		adapter->q_vector[q_idx] = NULL;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+		napi_hash_del(&q_vector->napi);
+#endif
 		netif_napi_del(&q_vector->napi);
 		kfree(q_vector);
 	}
@@ -3323,6 +3388,9 @@ static const struct net_device_ops ixgbevf_netdev_ops = {
 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	.ndo_busy_poll		= ixgbevf_busy_poll_recv,
+#endif
 };
 
 static void ixgbevf_assign_netdev_ops(struct net_device *dev)