@@ -171,31 +171,38 @@ static inline bool dev_xmit_complete(int rc)
* (unsigned long) so they can be read and written atomically.
*/
+#define NET_DEV_STAT(FIELD) \
+ union { \
+ unsigned long FIELD; \
+ atomic_long_t __##FIELD; \
+ }
+
struct net_device_stats {
- unsigned long rx_packets;
- unsigned long tx_packets;
- unsigned long rx_bytes;
- unsigned long tx_bytes;
- unsigned long rx_errors;
- unsigned long tx_errors;
- unsigned long rx_dropped;
- unsigned long tx_dropped;
- unsigned long multicast;
- unsigned long collisions;
- unsigned long rx_length_errors;
- unsigned long rx_over_errors;
- unsigned long rx_crc_errors;
- unsigned long rx_frame_errors;
- unsigned long rx_fifo_errors;
- unsigned long rx_missed_errors;
- unsigned long tx_aborted_errors;
- unsigned long tx_carrier_errors;
- unsigned long tx_fifo_errors;
- unsigned long tx_heartbeat_errors;
- unsigned long tx_window_errors;
- unsigned long rx_compressed;
- unsigned long tx_compressed;
+ NET_DEV_STAT(rx_packets);
+ NET_DEV_STAT(tx_packets);
+ NET_DEV_STAT(rx_bytes);
+ NET_DEV_STAT(tx_bytes);
+ NET_DEV_STAT(rx_errors);
+ NET_DEV_STAT(tx_errors);
+ NET_DEV_STAT(rx_dropped);
+ NET_DEV_STAT(tx_dropped);
+ NET_DEV_STAT(multicast);
+ NET_DEV_STAT(collisions);
+ NET_DEV_STAT(rx_length_errors);
+ NET_DEV_STAT(rx_over_errors);
+ NET_DEV_STAT(rx_crc_errors);
+ NET_DEV_STAT(rx_frame_errors);
+ NET_DEV_STAT(rx_fifo_errors);
+ NET_DEV_STAT(rx_missed_errors);
+ NET_DEV_STAT(tx_aborted_errors);
+ NET_DEV_STAT(tx_carrier_errors);
+ NET_DEV_STAT(tx_fifo_errors);
+ NET_DEV_STAT(tx_heartbeat_errors);
+ NET_DEV_STAT(tx_window_errors);
+ NET_DEV_STAT(rx_compressed);
+ NET_DEV_STAT(tx_compressed);
};
+#undef NET_DEV_STAT
/* per-cpu stats, allocated on demand.
* Try to fit them in a single cache line, for dev_get_stats() sake.
@@ -5171,4 +5178,9 @@ extern struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
extern struct net_device *blackhole_netdev;
+/* Note: Avoid these macros in fast path, prefer per-cpu or per-queue counters. */
+#define DEV_STATS_INC(DEV, FIELD) atomic_long_inc(&(DEV)->stats.__##FIELD)
+#define DEV_STATS_ADD(DEV, FIELD, VAL) \
+ atomic_long_add((VAL), &(DEV)->stats.__##FIELD)
+
#endif /* _LINUX_NETDEVICE_H */
@@ -356,9 +356,8 @@ static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
struct net *net)
{
- /* TODO : stats should be SMP safe */
- dev->stats.rx_packets++;
- dev->stats.rx_bytes += skb->len;
+ DEV_STATS_INC(dev, rx_packets);
+ DEV_STATS_ADD(dev, rx_bytes, skb->len);
__skb_tunnel_rx(skb, dev, net);
}
@@ -10381,24 +10381,16 @@ void netdev_run_todo(void)
void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
const struct net_device_stats *netdev_stats)
{
-#if BITS_PER_LONG == 64
- BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
- memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
- /* zero out counters that only exist in rtnl_link_stats64 */
- memset((char *)stats64 + sizeof(*netdev_stats), 0,
- sizeof(*stats64) - sizeof(*netdev_stats));
-#else
- size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
- const unsigned long *src = (const unsigned long *)netdev_stats;
+ size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
+ const atomic_long_t *src = (atomic_long_t *)netdev_stats;
u64 *dst = (u64 *)stats64;
BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
for (i = 0; i < n; i++)
- dst[i] = src[i];
+ dst[i] = atomic_long_read(&src[i]);
/* zero out counters that only exist in rtnl_link_stats64 */
memset((char *)stats64 + n * sizeof(u64), 0,
sizeof(*stats64) - n * sizeof(u64));
-#endif
}
EXPORT_SYMBOL(netdev_stats_to_stats64);
Long standing KCSAN issues are caused by data-race around some dev->stats changes. Most performance critical paths already use per-cpu variables, or per-queue ones. It is reasonable (and more correct) to use atomic operations for the slow paths. This patch adds an union for each field of net_device_stats, so that we can convert paths that are not yet protected by a spinlock or a mutex. netdev_stats_to_stats64() no longer has an #if BITS_PER_LONG==64 Note that the memcpy() we were using on 64bit arches had no provision to avoid load-tearing, while atomic_long_read() is providing the needed protection at no cost. Signed-off-by: Eric Dumazet <edumazet@google.com> --- include/linux/netdevice.h | 58 +++++++++++++++++++++++---------------- include/net/dst.h | 5 ++-- net/core/dev.c | 14 ++-------- 3 files changed, 40 insertions(+), 37 deletions(-)