staging: fsl-dpaa2/eth: Add driver specific stats

Add custom statistics to be reported via ethtool -S. These include
driver specific per-cpu statistics as well as queue and channel
counters.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.c

@@ -208,6 +208,7 @@ static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 	void *vaddr;
 	struct sk_buff *skb;
 	struct rtnl_link_stats64 *percpu_stats;
+	struct dpaa2_eth_drv_stats *percpu_extras;
 	struct device *dev = priv->net_dev->dev.parent;
 	struct dpaa2_fas *fas;
 	u32 status = 0;
@@ -219,6 +220,7 @@ static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 	prefetch(vaddr + dpaa2_fd_get_offset(fd));
 
 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
+	percpu_extras = this_cpu_ptr(priv->percpu_extras);
 
 	if (fd_format == dpaa2_fd_single) {
 		skb = build_linear_skb(priv, ch, fd, vaddr);
@@ -227,6 +229,8 @@ static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 				vaddr + dpaa2_fd_get_offset(fd);
 		skb = build_frag_skb(priv, ch, sgt);
 		skb_free_frag(vaddr);
+		percpu_extras->rx_sg_frames++;
+		percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
 	} else {
 		/* We don't support any other format */
 		goto err_frame_format;
@@ -291,6 +295,7 @@ static int consume_frames(struct dpaa2_eth_channel *ch)
 
 		fd = dpaa2_dq_fd(dq);
 		fq = (struct dpaa2_eth_fq *)dpaa2_dq_fqd_ctx(dq);
+		fq->stats.frames++;
 
 		fq->consume(priv, ch, fd, &ch->napi);
 		cleaned++;
@@ -532,11 +537,13 @@ static int dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
 	struct dpaa2_fd fd;
 	struct rtnl_link_stats64 *percpu_stats;
+	struct dpaa2_eth_drv_stats *percpu_extras;
 	struct dpaa2_eth_fq *fq;
 	u16 queue_mapping;
 	int err, i;
 
 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
+	percpu_extras = this_cpu_ptr(priv->percpu_extras);
 
 	if (unlikely(skb_headroom(skb) < DPAA2_ETH_NEEDED_HEADROOM(priv))) {
 		struct sk_buff *ns;
@@ -563,10 +570,14 @@ static int dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 	/* Setup the FD fields */
 	memset(&fd, 0, sizeof(fd));
 
-	if (skb_is_nonlinear(skb))
+	if (skb_is_nonlinear(skb)) {
 		err = build_sg_fd(priv, skb, &fd);
-	else
+		percpu_extras->tx_sg_frames++;
+		percpu_extras->tx_sg_bytes += skb->len;
+	} else {
 		err = build_single_fd(priv, skb, &fd);
+	}
+
 	if (unlikely(err)) {
 		percpu_stats->tx_dropped++;
 		goto err_build_fd;
@@ -583,6 +594,7 @@ static int dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 		if (err != -EBUSY)
 			break;
 	}
+	percpu_extras->tx_portal_busy += i;
 	if (unlikely(err < 0)) {
 		percpu_stats->tx_errors++;
 		/* Clean up everything, including freeing the skb */
@@ -608,8 +620,13 @@ static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 			      struct napi_struct *napi __always_unused)
 {
 	struct rtnl_link_stats64 *percpu_stats;
+	struct dpaa2_eth_drv_stats *percpu_extras;
 	u32 status = 0;
 
+	percpu_extras = this_cpu_ptr(priv->percpu_extras);
+	percpu_extras->tx_conf_frames++;
+	percpu_extras->tx_conf_bytes += dpaa2_fd_get_len(fd);
+
 	free_tx_fd(priv, fd, &status);
 
 	if (unlikely(status & DPAA2_ETH_TXCONF_ERR_MASK)) {
@@ -814,13 +831,19 @@ static int refill_pool(struct dpaa2_eth_priv *priv,
 static int pull_channel(struct dpaa2_eth_channel *ch)
 {
 	int err;
+	int dequeues = -1;
 
 	/* Retry while portal is busy */
 	do {
 		err = dpaa2_io_service_pull_channel(NULL, ch->ch_id, ch->store);
+		dequeues++;
 		cpu_relax();
 	} while (err == -EBUSY);
 
+	ch->stats.dequeue_portal_busy += dequeues;
+	if (unlikely(err))
+		ch->stats.pull_err++;
+
 	return err;
 }
 
@@ -868,6 +891,8 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 		} while (err == -EBUSY);
 	}
 
+	ch->stats.frames += cleaned;
+
 	return cleaned;
 }
 
@@ -1320,6 +1345,10 @@ static void cdan_cb(struct dpaa2_io_notification_ctx *ctx)
 	struct dpaa2_eth_channel *ch;
 
 	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
+
+	/* Update NAPI statistics */
+	ch->stats.cdan++;
+
 	napi_schedule_irqoff(&ch->napi);
 }
 
@@ -2345,6 +2374,12 @@ static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
 		err = -ENOMEM;
 		goto err_alloc_percpu_stats;
 	}
+	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
+	if (!priv->percpu_extras) {
+		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
+		err = -ENOMEM;
+		goto err_alloc_percpu_extras;
+	}
 
 	err = netdev_init(net_dev);
 	if (err)
@@ -2387,6 +2422,8 @@ err_alloc_rings:
 err_csum:
 	unregister_netdev(net_dev);
 err_netdev_init:
+	free_percpu(priv->percpu_extras);
+err_alloc_percpu_extras:
 	free_percpu(priv->percpu_stats);
 err_alloc_percpu_stats:
 	del_ch_napi(priv);
@@ -2425,6 +2462,7 @@ static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
 
 	free_rings(priv);
 	free_percpu(priv->percpu_stats);
+	free_percpu(priv->percpu_extras);
 
 	del_ch_napi(priv);
 	free_dpbp(priv);

drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.h

@@ -197,6 +197,38 @@ struct dpaa2_fas {
  */
 #define DPAA2_ETH_ENQUEUE_RETRIES	10
 
+/* Driver statistics, other than those in struct rtnl_link_stats64.
+ * These are usually collected per-CPU and aggregated by ethtool.
+ */
+struct dpaa2_eth_drv_stats {
+	__u64	tx_conf_frames;
+	__u64	tx_conf_bytes;
+	__u64	tx_sg_frames;
+	__u64	tx_sg_bytes;
+	__u64	rx_sg_frames;
+	__u64	rx_sg_bytes;
+	/* Enqueues retried due to portal busy */
+	__u64	tx_portal_busy;
+};
+
+/* Per-FQ statistics */
+struct dpaa2_eth_fq_stats {
+	/* Number of frames received on this queue */
+	__u64 frames;
+};
+
+/* Per-channel statistics */
+struct dpaa2_eth_ch_stats {
+	/* Volatile dequeues retried due to portal busy */
+	__u64 dequeue_portal_busy;
+	/* Number of CDANs; useful to estimate avg NAPI len */
+	__u64 cdan;
+	/* Number of frames received on queues from this channel */
+	__u64 frames;
+	/* Pull errors */
+	__u64 pull_err;
+};
+
 /* Maximum number of queues associated with a DPNI */
 #define DPAA2_ETH_MAX_RX_QUEUES		16
 #define DPAA2_ETH_MAX_TX_QUEUES		NR_CPUS
@@ -224,6 +256,7 @@ struct dpaa2_eth_fq {
 			struct dpaa2_eth_channel *,
 			const struct dpaa2_fd *,
 			struct napi_struct *);
+	struct dpaa2_eth_fq_stats stats;
 };
 
 struct dpaa2_eth_channel {
@@ -236,6 +269,7 @@ struct dpaa2_eth_channel {
 	struct dpaa2_io_store *store;
 	struct dpaa2_eth_priv *priv;
 	int buf_count;
+	struct dpaa2_eth_ch_stats stats;
 };
 
 struct dpaa2_eth_hash_fields {
@@ -275,6 +309,8 @@ struct dpaa2_eth_priv {
 
 	/* Standard statistics */
 	struct rtnl_link_stats64 __percpu *percpu_stats;
+	/* Extra stats, in addition to the ones known by the kernel */
+	struct dpaa2_eth_drv_stats __percpu *percpu_extras;
 
 	u16 mc_token;
 

drivers/staging/fsl-dpaa2/ethernet/dpaa2-ethtool.c

@@ -56,6 +56,23 @@ char dpaa2_ethtool_stats[][ETH_GSTRING_LEN] = {
 
 #define DPAA2_ETH_NUM_STATS	ARRAY_SIZE(dpaa2_ethtool_stats)
 
+char dpaa2_ethtool_extras[][ETH_GSTRING_LEN] = {
+	/* per-cpu stats */
+	"tx conf frames",
+	"tx conf bytes",
+	"tx sg frames",
+	"tx sg bytes",
+	"rx sg frames",
+	"rx sg bytes",
+	"enqueue portal busy",
+	/* Channel stats */
+	"dequeue portal busy",
+	"channel pull errors",
+	"cdan",
+};
+
+#define DPAA2_ETH_NUM_EXTRA_STATS	ARRAY_SIZE(dpaa2_ethtool_extras)
+
 static void dpaa2_eth_get_drvinfo(struct net_device *net_dev,
 				  struct ethtool_drvinfo *drvinfo)
 {
@@ -147,6 +164,10 @@ static void dpaa2_eth_get_strings(struct net_device *netdev, u32 stringset,
 			strlcpy(p, dpaa2_ethtool_stats[i], ETH_GSTRING_LEN);
 			p += ETH_GSTRING_LEN;
 		}
+		for (i = 0; i < DPAA2_ETH_NUM_EXTRA_STATS; i++) {
+			strlcpy(p, dpaa2_ethtool_extras[i], ETH_GSTRING_LEN);
+			p += ETH_GSTRING_LEN;
+		}
 		break;
 	}
 }
@@ -155,7 +176,7 @@ static int dpaa2_eth_get_sset_count(struct net_device *net_dev, int sset)
 {
 	switch (sset) {
 	case ETH_SS_STATS: /* ethtool_get_stats(), ethtool_get_drvinfo() */
-		return DPAA2_ETH_NUM_STATS;
+		return DPAA2_ETH_NUM_STATS + DPAA2_ETH_NUM_EXTRA_STATS;
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -171,9 +192,14 @@ static void dpaa2_eth_get_ethtool_stats(struct net_device *net_dev,
 	int j, k, err;
 	int num_cnt;
 	union dpni_statistics dpni_stats;
+	u64 cdan = 0;
+	u64 portal_busy = 0, pull_err = 0;
 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
+	struct dpaa2_eth_drv_stats *extras;
+	struct dpaa2_eth_ch_stats *ch_stats;
 
-	memset(data, 0, sizeof(u64) * DPAA2_ETH_NUM_STATS);
+	memset(data, 0,
+	       sizeof(u64) * (DPAA2_ETH_NUM_STATS + DPAA2_ETH_NUM_EXTRA_STATS));
 
 	/* Print standard counters, from DPNI statistics */
 	for (j = 0; j <= 2; j++) {
@@ -197,6 +223,25 @@ static void dpaa2_eth_get_ethtool_stats(struct net_device *net_dev,
 		for (k = 0; k < num_cnt; k++)
 			*(data + i++) = dpni_stats.raw.counter[k];
 	}
+
+	/* Print per-cpu extra stats */
+	for_each_online_cpu(k) {
+		extras = per_cpu_ptr(priv->percpu_extras, k);
+		for (j = 0; j < sizeof(*extras) / sizeof(__u64); j++)
+			*((__u64 *)data + i + j) += *((__u64 *)extras + j);
+	}
+	i += j;
+
+	for (j = 0; j < priv->num_channels; j++) {
+		ch_stats = &priv->channel[j]->stats;
+		cdan += ch_stats->cdan;
+		portal_busy += ch_stats->dequeue_portal_busy;
+		pull_err += ch_stats->pull_err;
+	}
+
+	*(data + i++) = portal_busy;
+	*(data + i++) = pull_err;
+	*(data + i++) = cdan;
 }
 
 static int dpaa2_eth_get_rxnfc(struct net_device *net_dev,