The new wrappers tun_ring_consume/tap_ring_consume deal with consuming an entry of the ptr_ring and then waking the netdev queue when entries got invalidated to be used again by the producer. To avoid waking the netdev queue when the ptr_ring is full, it is checked if the netdev queue is stopped before invalidating entries. Like that the netdev queue can be safely woken after invalidating entries. The READ_ONCE in __ptr_ring_peek, paired with the smp_wmb() in __ptr_ring_produce within tun_net_xmit guarantees that the information about the netdev queue being stopped is visible after __ptr_ring_peek is called. The netdev queue is also woken after resizing the ptr_ring. Co-developed-by: Tim Gebauer Signed-off-by: Tim Gebauer Signed-off-by: Simon Schippers --- drivers/net/tap.c | 44 +++++++++++++++++++++++++++++++++++++++++++- drivers/net/tun.c | 47 +++++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 88 insertions(+), 3 deletions(-) diff --git a/drivers/net/tap.c b/drivers/net/tap.c index 1197f245e873..f8292721a9d6 100644 --- a/drivers/net/tap.c +++ b/drivers/net/tap.c @@ -753,6 +753,46 @@ static ssize_t tap_put_user(struct tap_queue *q, return ret ? ret : total; } +static struct sk_buff *tap_ring_consume(struct tap_queue *q) +{ + struct netdev_queue *txq; + struct net_device *dev; + bool will_invalidate; + bool stopped; + void *ptr; + + spin_lock(&q->ring.consumer_lock); + ptr = __ptr_ring_peek(&q->ring); + if (!ptr) { + spin_unlock(&q->ring.consumer_lock); + return ptr; + } + + /* Check if the queue stopped before zeroing out, so no ptr get + * produced in the meantime, because this could result in waking + * even though the ptr_ring is full. The order of the operations + * is ensured by barrier(). + */ + will_invalidate = __ptr_ring_will_invalidate(&q->ring); + if (unlikely(will_invalidate)) { + rcu_read_lock(); + dev = rcu_dereference(q->tap)->dev; + txq = netdev_get_tx_queue(dev, q->queue_index); + stopped = netif_tx_queue_stopped(txq); + } + barrier(); + __ptr_ring_discard_one(&q->ring, will_invalidate); + + if (unlikely(will_invalidate)) { + if (stopped) + netif_tx_wake_queue(txq); + rcu_read_unlock(); + } + spin_unlock(&q->ring.consumer_lock); + + return ptr; +} + static ssize_t tap_do_read(struct tap_queue *q, struct iov_iter *to, int noblock, struct sk_buff *skb) @@ -774,7 +814,7 @@ static ssize_t tap_do_read(struct tap_queue *q, TASK_INTERRUPTIBLE); /* Read frames from the queue */ - skb = ptr_ring_consume(&q->ring); + skb = tap_ring_consume(q); if (skb) break; if (noblock) { @@ -1207,6 +1247,8 @@ int tap_queue_resize(struct tap_dev *tap) ret = ptr_ring_resize_multiple_bh(rings, n, dev->tx_queue_len, GFP_KERNEL, __skb_array_destroy_skb); + if (netif_running(dev)) + netif_tx_wake_all_queues(dev); kfree(rings); return ret; diff --git a/drivers/net/tun.c b/drivers/net/tun.c index c6b22af9bae8..682df8157b55 100644 --- a/drivers/net/tun.c +++ b/drivers/net/tun.c @@ -2114,13 +2114,53 @@ static ssize_t tun_put_user(struct tun_struct *tun, return total; } +static void *tun_ring_consume(struct tun_file *tfile) +{ + struct netdev_queue *txq; + struct net_device *dev; + bool will_invalidate; + bool stopped; + void *ptr; + + spin_lock(&tfile->tx_ring.consumer_lock); + ptr = __ptr_ring_peek(&tfile->tx_ring); + if (!ptr) { + spin_unlock(&tfile->tx_ring.consumer_lock); + return ptr; + } + + /* Check if the queue stopped before zeroing out, so no ptr get + * produced in the meantime, because this could result in waking + * even though the ptr_ring is full. The order of the operations + * is ensured by barrier(). + */ + will_invalidate = __ptr_ring_will_invalidate(&tfile->tx_ring); + if (unlikely(will_invalidate)) { + rcu_read_lock(); + dev = rcu_dereference(tfile->tun)->dev; + txq = netdev_get_tx_queue(dev, tfile->queue_index); + stopped = netif_tx_queue_stopped(txq); + } + barrier(); + __ptr_ring_discard_one(&tfile->tx_ring, will_invalidate); + + if (unlikely(will_invalidate)) { + if (stopped) + netif_tx_wake_queue(txq); + rcu_read_unlock(); + } + spin_unlock(&tfile->tx_ring.consumer_lock); + + return ptr; +} + static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err) { DECLARE_WAITQUEUE(wait, current); void *ptr = NULL; int error = 0; - ptr = ptr_ring_consume(&tfile->tx_ring); + ptr = tun_ring_consume(tfile); if (ptr) goto out; if (noblock) { @@ -2132,7 +2172,7 @@ static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err) while (1) { set_current_state(TASK_INTERRUPTIBLE); - ptr = ptr_ring_consume(&tfile->tx_ring); + ptr = tun_ring_consume(tfile); if (ptr) break; if (signal_pending(current)) { @@ -3621,6 +3661,9 @@ static int tun_queue_resize(struct tun_struct *tun) dev->tx_queue_len, GFP_KERNEL, tun_ptr_free); + if (netif_running(dev)) + netif_tx_wake_all_queues(dev); + kfree(rings); return ret; } -- 2.43.0