On 01/19/2018 03:09 PM, Cong Wang wrote: > Introduce a new qdisc ops ->change_tx_queue_len() so that > each qdisc could decide how to implement this if it wants. > Previously we simply read dev->tx_queue_len, after pfifo_fast > switches to skb array, we need this API to resize the skb array > when we change dev->tx_queue_len. > > To avoid handling race conditions with TX BH, we need to > deactivate all TX queues before change the value and bring them > back after we are done, this also makes implementation easier. > > Cc: John Fastabend <john.fastab...@gmail.com> > Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com> > --- > include/net/sch_generic.h | 2 ++ > net/core/dev.c | 1 + > net/sched/sch_generic.c | 22 ++++++++++++++++++++++ > 3 files changed, 25 insertions(+) > > diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h > index cd1be1f25c36..aae1baa1c30f 100644 > --- a/include/net/sch_generic.h > +++ b/include/net/sch_generic.h > @@ -200,6 +200,7 @@ struct Qdisc_ops { > struct nlattr *arg, > struct netlink_ext_ack *extack); > void (*attach)(struct Qdisc *sch); > + void (*change_tx_queue_len)(struct Qdisc *, unsigned > int); > > int (*dump)(struct Qdisc *, struct sk_buff *); > int (*dump_stats)(struct Qdisc *, struct gnet_dump > *); > @@ -488,6 +489,7 @@ void qdisc_class_hash_remove(struct Qdisc_class_hash *, > void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); > void qdisc_class_hash_destroy(struct Qdisc_class_hash *); > > +void dev_qdisc_change_tx_queue_len(struct net_device *dev); > void dev_init_scheduler(struct net_device *dev); > void dev_shutdown(struct net_device *dev); > void dev_activate(struct net_device *dev); > diff --git a/net/core/dev.c b/net/core/dev.c > index 99d353e4cbb2..24809d858a64 100644 > --- a/net/core/dev.c > +++ b/net/core/dev.c > @@ -7058,6 +7058,7 @@ int dev_change_tx_queue_len(struct net_device *dev, > unsigned long new_len) > dev->tx_queue_len = orig_len; > return res; > } > + dev_qdisc_change_tx_queue_len(dev); > } > > return 0; > diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c > index ef8b4ecde2ac..30aaeb3c1bf1 100644 > --- a/net/sched/sch_generic.c > +++ b/net/sched/sch_generic.c > @@ -1178,6 +1178,28 @@ void dev_deactivate(struct net_device *dev) > } > EXPORT_SYMBOL(dev_deactivate); > > +static void qdisc_change_tx_queue_len(struct net_device *dev, > + struct netdev_queue *dev_queue, > + void *unused) > +{ > + struct Qdisc *qdisc = dev_queue->qdisc_sleeping; > + const struct Qdisc_ops *ops = qdisc->ops; > + > + if (ops->change_tx_queue_len) > + ops->change_tx_queue_len(qdisc, dev->tx_queue_len);
hmm what happens if the resize fails in the next patch, > > +static void pfifo_fast_change_tx_queue_len(struct Qdisc *sch, unsigned int > new_len) > +{ > + struct pfifo_fast_priv *priv = qdisc_priv(sch); > + int prio; > + > + for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { > + struct skb_array *q = band2list(priv, prio); > + > + skb_array_resize(q, new_len, GFP_KERNEL); > + } > +} > + Here skb_array_resize() can fail with ENOMEM, do we need to unwind the change and push the error up the stack? Thanks, John