nfp: add support for coalesce adaptive feature

Use dynamic interrupt moderation library to implement coalesce
adaptive feature for nfp driver.

Signed-off-by: Yinjun Zhang <yinjun.zhang@corigine.com>
Signed-off-by: Yu Xiao <yu.xiao@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 159 insertions, 13 deletions

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index df5b748be068..df203738511b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -17,6 +17,7 @@
 #include <linux/list.h>
 #include <linux/netdevice.h>
 #include <linux/pci.h>
+#include <linux/dim.h>
 #include <linux/io-64-nonatomic-hi-lo.h>
 #include <linux/semaphore.h>
 #include <linux/workqueue.h>
@@ -360,6 +361,9 @@ struct nfp_net_rx_ring {
  * @rx_ring:        Pointer to RX ring
  * @xdp_ring:	    Pointer to an extra TX ring for XDP
  * @irq_entry:      MSI-X table entry (use for talking to the device)
+ * @event_ctr:	    Number of interrupt
+ * @rx_dim:	    Dynamic interrupt moderation structure for RX
+ * @tx_dim:	    Dynamic interrupt moderation structure for TX
  * @rx_sync:	    Seqlock for atomic updates of RX stats
  * @rx_pkts:        Number of received packets
  * @rx_bytes:	    Number of received bytes
@@ -410,6 +414,10 @@ struct nfp_net_r_vector {
 
 	u16 irq_entry;
 
+	u16 event_ctr;
+	struct dim rx_dim;
+	struct dim tx_dim;
+
 	struct u64_stats_sync rx_sync;
 	u64 rx_pkts;
 	u64 rx_bytes;
@@ -571,6 +579,8 @@ struct nfp_net_dp {
  *			mailbox area, crypto TLV
  * @link_up:            Is the link up?
  * @link_status_lock:	Protects @link_* and ensures atomicity with BAR reading
+ * @rx_coalesce_adapt_on:   Is RX interrupt moderation adaptive?
+ * @tx_coalesce_adapt_on:   Is TX interrupt moderation adaptive?
  * @rx_coalesce_usecs:      RX interrupt moderation usecs delay parameter
  * @rx_coalesce_max_frames: RX interrupt moderation frame count parameter
  * @tx_coalesce_usecs:      TX interrupt moderation usecs delay parameter
@@ -654,6 +664,8 @@ struct nfp_net {
 
 	struct semaphore bar_lock;
 
+	bool rx_coalesce_adapt_on;
+	bool tx_coalesce_adapt_on;
 	u32 rx_coalesce_usecs;
 	u32 rx_coalesce_max_frames;
 	u32 tx_coalesce_usecs;
@@ -919,6 +931,14 @@ static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev)
 	return netdev->netdev_ops == &nfp_net_netdev_ops;
 }
 
+static inline int nfp_net_coalesce_para_check(u32 usecs, u32 pkts)
+{
+	if ((usecs >= ((1 << 16) - 1)) || (pkts >= ((1 << 16) - 1)))
+		return -EINVAL;
+
+	return 0;
+}
+
 /* Prototypes */
 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
 			    void __iomem *ctrl_bar);
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index ed2ade2a4f04..15078f9dc9f1 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -474,6 +474,12 @@ static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
 {
 	struct nfp_net_r_vector *r_vec = data;
 
+	/* Currently we cannot tell if it's a rx or tx interrupt,
+	 * since dim does not need accurate event_ctr to calculate,
+	 * we just use this counter for both rx and tx dim.
+	 */
+	r_vec->event_ctr++;
+
 	napi_schedule_irqoff(&r_vec->napi);
 
 	/* The FW auto-masks any interrupt, either via the MASK bit in
@@ -2061,6 +2067,36 @@ static int nfp_net_poll(struct napi_struct *napi, int budget)
 		if (napi_complete_done(napi, pkts_polled))
 			nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
 
+	if (r_vec->nfp_net->rx_coalesce_adapt_on) {
+		struct dim_sample dim_sample = {};
+		unsigned int start;
+		u64 pkts, bytes;
+
+		do {
+			start = u64_stats_fetch_begin(&r_vec->rx_sync);
+			pkts = r_vec->rx_pkts;
+			bytes = r_vec->rx_bytes;
+		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
+
+		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
+		net_dim(&r_vec->rx_dim, dim_sample);
+	}
+
+	if (r_vec->nfp_net->tx_coalesce_adapt_on) {
+		struct dim_sample dim_sample = {};
+		unsigned int start;
+		u64 pkts, bytes;
+
+		do {
+			start = u64_stats_fetch_begin(&r_vec->tx_sync);
+			pkts = r_vec->tx_pkts;
+			bytes = r_vec->tx_bytes;
+		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
+
+		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
+		net_dim(&r_vec->tx_dim, dim_sample);
+	}
+
 	return pkts_polled;
 }
 
@@ -2873,6 +2909,7 @@ static int nfp_net_set_config_and_enable(struct nfp_net *nn)
  */
 static void nfp_net_close_stack(struct nfp_net *nn)
 {
+	struct nfp_net_r_vector *r_vec;
 	unsigned int r;
 
 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
@@ -2880,8 +2917,16 @@ static void nfp_net_close_stack(struct nfp_net *nn)
 	nn->link_up = false;
 
 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
-		disable_irq(nn->r_vecs[r].irq_vector);
-		napi_disable(&nn->r_vecs[r].napi);
+		r_vec = &nn->r_vecs[r];
+
+		disable_irq(r_vec->irq_vector);
+		napi_disable(&r_vec->napi);
+
+		if (r_vec->rx_ring)
+			cancel_work_sync(&r_vec->rx_dim.work);
+
+		if (r_vec->tx_ring)
+			cancel_work_sync(&r_vec->tx_dim.work);
 	}
 
 	netif_tx_disable(nn->dp.netdev);
@@ -2948,17 +2993,92 @@ void nfp_ctrl_close(struct nfp_net *nn)
 	rtnl_unlock();
 }
 
+static void nfp_net_rx_dim_work(struct work_struct *work)
+{
+	struct nfp_net_r_vector *r_vec;
+	unsigned int factor, value;
+	struct dim_cq_moder moder;
+	struct nfp_net *nn;
+	struct dim *dim;
+
+	dim = container_of(work, struct dim, work);
+	moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
+	r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim);
+	nn = r_vec->nfp_net;
+
+	/* Compute factor used to convert coalesce '_usecs' parameters to
+	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
+	 * count.
+	 */
+	factor = nn->tlv_caps.me_freq_mhz / 16;
+	if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts))
+		return;
+
+	/* copy RX interrupt coalesce parameters */
+	value = (moder.pkts << 16) | (factor * moder.usec);
+	rtnl_lock();
+	nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value);
+	(void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD);
+	rtnl_unlock();
+
+	dim->state = DIM_START_MEASURE;
+}
+
+static void nfp_net_tx_dim_work(struct work_struct *work)
+{
+	struct nfp_net_r_vector *r_vec;
+	unsigned int factor, value;
+	struct dim_cq_moder moder;
+	struct nfp_net *nn;
+	struct dim *dim;
+
+	dim = container_of(work, struct dim, work);
+	moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
+	r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim);
+	nn = r_vec->nfp_net;
+
+	/* Compute factor used to convert coalesce '_usecs' parameters to
+	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
+	 * count.
+	 */
+	factor = nn->tlv_caps.me_freq_mhz / 16;
+	if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts))
+		return;
+
+	/* copy TX interrupt coalesce parameters */
+	value = (moder.pkts << 16) | (factor * moder.usec);
+	rtnl_lock();
+	nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value);
+	(void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD);
+	rtnl_unlock();
+
+	dim->state = DIM_START_MEASURE;
+}
+
 /**
  * nfp_net_open_stack() - Start the device from stack's perspective
  * @nn:      NFP Net device to reconfigure
  */
 static void nfp_net_open_stack(struct nfp_net *nn)
 {
+	struct nfp_net_r_vector *r_vec;
 	unsigned int r;
 
 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
-		napi_enable(&nn->r_vecs[r].napi);
-		enable_irq(nn->r_vecs[r].irq_vector);
+		r_vec = &nn->r_vecs[r];
+
+		if (r_vec->rx_ring) {
+			INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work);
+			r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+		}
+
+		if (r_vec->tx_ring) {
+			INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work);
+			r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
+		}
+
+		napi_enable(&r_vec->napi);
+		enable_irq(r_vec->irq_vector);
 	}
 
 	netif_tx_wake_all_queues(nn->dp.netdev);
@@ -3893,6 +4013,9 @@ static void nfp_net_irqmod_init(struct nfp_net *nn)
 	nn->rx_coalesce_max_frames = 64;
 	nn->tx_coalesce_usecs      = 50;
 	nn->tx_coalesce_max_frames = 64;
+
+	nn->rx_coalesce_adapt_on   = true;
+	nn->tx_coalesce_adapt_on   = true;
 }
 
 static void nfp_net_netdev_init(struct nfp_net *nn)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
index 1b482446536d..a213784ffa54 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c
@@ -1083,6 +1083,9 @@ static int nfp_net_get_coalesce(struct net_device *netdev,
 	if (!(nn->cap & NFP_NET_CFG_CTRL_IRQMOD))
 		return -EINVAL;
 
+	ec->use_adaptive_rx_coalesce = nn->rx_coalesce_adapt_on;
+	ec->use_adaptive_tx_coalesce = nn->tx_coalesce_adapt_on;
+
 	ec->rx_coalesce_usecs       = nn->rx_coalesce_usecs;
 	ec->rx_max_coalesced_frames = nn->rx_coalesce_max_frames;
 	ec->tx_coalesce_usecs       = nn->tx_coalesce_usecs;
@@ -1359,19 +1362,18 @@ static int nfp_net_set_coalesce(struct net_device *netdev,
 	if (!ec->tx_coalesce_usecs && !ec->tx_max_coalesced_frames)
 		return -EINVAL;
 
-	if (ec->rx_coalesce_usecs * factor >= ((1 << 16) - 1))
-		return -EINVAL;
-
-	if (ec->tx_coalesce_usecs * factor >= ((1 << 16) - 1))
+	if (nfp_net_coalesce_para_check(ec->rx_coalesce_usecs * factor,
+					ec->rx_max_coalesced_frames))
 		return -EINVAL;
 
-	if (ec->rx_max_coalesced_frames >= ((1 << 16) - 1))
-		return -EINVAL;
-
-	if (ec->tx_max_coalesced_frames >= ((1 << 16) - 1))
+	if (nfp_net_coalesce_para_check(ec->tx_coalesce_usecs * factor,
+					ec->tx_max_coalesced_frames))
 		return -EINVAL;
 
 	/* configuration is valid */
+	nn->rx_coalesce_adapt_on = !!ec->use_adaptive_rx_coalesce;
+	nn->tx_coalesce_adapt_on = !!ec->use_adaptive_tx_coalesce;
+
 	nn->rx_coalesce_usecs      = ec->rx_coalesce_usecs;
 	nn->rx_coalesce_max_frames = ec->rx_max_coalesced_frames;
 	nn->tx_coalesce_usecs      = ec->tx_coalesce_usecs;
@@ -1443,7 +1445,8 @@ static int nfp_net_set_channels(struct net_device *netdev,
 
 static const struct ethtool_ops nfp_net_ethtool_ops = {
 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
-				     ETHTOOL_COALESCE_MAX_FRAMES,
+				     ETHTOOL_COALESCE_MAX_FRAMES |
+				     ETHTOOL_COALESCE_USE_ADAPTIVE,
 	.get_drvinfo		= nfp_net_get_drvinfo,
 	.get_link		= ethtool_op_get_link,
 	.get_ringparam		= nfp_net_get_ringparam,