net: mvneta: use per_cpu stats to fix an SMP lock up

commit 74c41b0 upstream.

Stats writers are mvneta_rx() and mvneta_tx(). They don't lock anything
when they update the stats, and as a result, it randomly happens that
the stats freeze on SMP if two updates happen during stats retrieval.
This is very easily reproducible by starting two HTTP servers and binding
each of them to a different CPU, then consulting /proc/net/dev in loops
during transfers, the interface should immediately lock up. This issue
also randomly happens upon link state changes during transfers, because
the stats are collected in this situation, but it takes more attempts to
reproduce it.

The comments in netdevice.h suggest using per_cpu stats instead to get
rid of this issue.

This patch implements this. It merges both rx_stats and tx_stats into
a single "stats" member with a single syncp. Both mvneta_rx() and
mvneta_rx() now only update the a single CPU's counters.

In turn, mvneta_get_stats64() does the summing by iterating over all CPUs
to get their respective stats.

With this change, stats are still correct and no more lockup is encountered.

Note that this bug was present since the first import of the mvneta
driver.  It might make sense to backport it to some stable trees. If
so, it depends on "d33dc73 net: mvneta: increase the 64-bit rx/tx stats
out of the hot path".

Cc: Thomas Petazzoni <[email protected]>
Cc: Gregory CLEMENT <[email protected]>
Reviewed-by: Eric Dumazet <[email protected]>
Tested-by: Arnaud Ebalard <[email protected]>
Signed-off-by: Willy Tarreau <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
[wt: port to 3.10 : u64_stats_init() does not exist in 3.10 and is not needed]
Signed-off-by: Willy Tarreau <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: wtarreau

drivers/net/ethernet/marvell/mvneta.c

@@ -219,10 +219,12 @@
 
 #define MVNETA_RX_BUF_SIZE(pkt_size)   ((pkt_size) + NET_SKB_PAD)
 
-struct mvneta_stats {
+struct mvneta_pcpu_stats {
 	struct	u64_stats_sync syncp;
-	u64	packets;
-	u64	bytes;
+	u64	rx_packets;
+	u64	rx_bytes;
+	u64	tx_packets;
+	u64	tx_bytes;
 };
 
 struct mvneta_port {
@@ -248,8 +250,7 @@ struct mvneta_port {
 	u8 mcast_count[256];
 	u16 tx_ring_size;
 	u16 rx_ring_size;
-	struct mvneta_stats tx_stats;
-	struct mvneta_stats rx_stats;
+	struct mvneta_pcpu_stats *stats;
 
 	struct mii_bus *mii_bus;
 	struct phy_device *phy_dev;
@@ -428,21 +429,29 @@ struct rtnl_link_stats64 *mvneta_get_stats64(struct net_device *dev,
 {
 	struct mvneta_port *pp = netdev_priv(dev);
 	unsigned int start;
+	int cpu;
 
-	memset(stats, 0, sizeof(struct rtnl_link_stats64));
-
-	do {
-		start = u64_stats_fetch_begin_bh(&pp->rx_stats.syncp);
-		stats->rx_packets = pp->rx_stats.packets;
-		stats->rx_bytes	= pp->rx_stats.bytes;
-	} while (u64_stats_fetch_retry_bh(&pp->rx_stats.syncp, start));
+	for_each_possible_cpu(cpu) {
+		struct mvneta_pcpu_stats *cpu_stats;
+		u64 rx_packets;
+		u64 rx_bytes;
+		u64 tx_packets;
+		u64 tx_bytes;
 
+		cpu_stats = per_cpu_ptr(pp->stats, cpu);
+		do {
+			start = u64_stats_fetch_begin_bh(&cpu_stats->syncp);
+			rx_packets = cpu_stats->rx_packets;
+			rx_bytes   = cpu_stats->rx_bytes;
+			tx_packets = cpu_stats->tx_packets;
+			tx_bytes   = cpu_stats->tx_bytes;
+		} while (u64_stats_fetch_retry_bh(&cpu_stats->syncp, start));
 
-	do {
-		start = u64_stats_fetch_begin_bh(&pp->tx_stats.syncp);
-		stats->tx_packets = pp->tx_stats.packets;
-		stats->tx_bytes	= pp->tx_stats.bytes;
-	} while (u64_stats_fetch_retry_bh(&pp->tx_stats.syncp, start));
+		stats->rx_packets += rx_packets;
+		stats->rx_bytes   += rx_bytes;
+		stats->tx_packets += tx_packets;
+		stats->tx_bytes   += tx_bytes;
+	}
 
 	stats->rx_errors	= dev->stats.rx_errors;
 	stats->rx_dropped	= dev->stats.rx_dropped;
@@ -1416,10 +1425,12 @@ static int mvneta_rx(struct mvneta_port *pp, int rx_todo,
 	}
 
 	if (rcvd_pkts) {
-		u64_stats_update_begin(&pp->rx_stats.syncp);
-		pp->rx_stats.packets += rcvd_pkts;
-		pp->rx_stats.bytes   += rcvd_bytes;
-		u64_stats_update_end(&pp->rx_stats.syncp);
+		struct mvneta_pcpu_stats *stats = this_cpu_ptr(pp->stats);
+
+		u64_stats_update_begin(&stats->syncp);
+		stats->rx_packets += rcvd_pkts;
+		stats->rx_bytes   += rcvd_bytes;
+		u64_stats_update_end(&stats->syncp);
 	}
 
 	/* Update rxq management counters */
@@ -1552,11 +1563,12 @@ static int mvneta_tx(struct sk_buff *skb, struct net_device *dev)
 
 out:
 	if (frags > 0) {
-		u64_stats_update_begin(&pp->tx_stats.syncp);
-		pp->tx_stats.packets++;
-		pp->tx_stats.bytes += skb->len;
-		u64_stats_update_end(&pp->tx_stats.syncp);
+		struct mvneta_pcpu_stats *stats = this_cpu_ptr(pp->stats);
 
+		u64_stats_update_begin(&stats->syncp);
+		stats->tx_packets++;
+		stats->tx_bytes  += skb->len;
+		u64_stats_update_end(&stats->syncp);
 	} else {
 		dev->stats.tx_dropped++;
 		dev_kfree_skb_any(skb);
@@ -2758,6 +2770,13 @@ static int mvneta_probe(struct platform_device *pdev)
 
 	clk_prepare_enable(pp->clk);
 
+	/* Alloc per-cpu stats */
+	pp->stats = alloc_percpu(struct mvneta_pcpu_stats);
+	if (!pp->stats) {
+		err = -ENOMEM;
+		goto err_clk;
+	}
+
 	pp->tx_done_timer.data = (unsigned long)dev;
 
 	pp->tx_ring_size = MVNETA_MAX_TXD;
@@ -2769,7 +2788,7 @@ static int mvneta_probe(struct platform_device *pdev)
 	err = mvneta_init(pp, phy_addr);
 	if (err < 0) {
 		dev_err(&pdev->dev, "can't init eth hal\n");
-		goto err_clk;
+		goto err_free_stats;
 	}
 	mvneta_port_power_up(pp, phy_mode);
 
@@ -2798,6 +2817,8 @@ static int mvneta_probe(struct platform_device *pdev)
 
 err_deinit:
 	mvneta_deinit(pp);
+err_free_stats:
+	free_percpu(pp->stats);
 err_clk:
 	clk_disable_unprepare(pp->clk);
 err_unmap:
@@ -2818,6 +2839,7 @@ static int mvneta_remove(struct platform_device *pdev)
 	unregister_netdev(dev);
 	mvneta_deinit(pp);
 	clk_disable_unprepare(pp->clk);
+	free_percpu(pp->stats);
 	iounmap(pp->base);
 	irq_dispose_mapping(dev->irq);
 	free_netdev(dev);