net: dev: BYPASS for lockless qdisc
Subsystems (3)
Paolo Abeni
maintainerv3 cover net: dev: BYPASS for lockless qdisc
This patch series is aimed at improving xmit performances of lockless qdisc in the uncontended scenario.
After the lockless refactor pfifo_fast can't leverage the BYPASS optimization. Due to retpolines the overhead for the avoidables enqueue and dequeue operations has increased and we see measurable regressions.
The first patch introduces the BYPASS code path for lockless qdisc, and the second one optimizes such path further. Overall this avoids up to 3 indirect calls per xmit packet. Detailed performance figures are reported in the 2nd patch.
v2 -> v3:
- qdisc_is_empty() has a const argument (Eric)
v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as suggested by Eric
Paolo Abeni (2):
net: sched: add empty status flag for NOLOCK qdisc
net: dev: introduce support for sch BYPASS for lockless qdisc
include/net/sch_generic.h | 11 +++++++++++ net/core/dev.c | 9 +++++++++ net/sched/sch_generic.c | 3 +++ 3 files changed, 23 insertions(+)
Signature · 1 line
2.20.1
Paolo Abeni
maintainerPATCH v3 1/2 net: sched: add empty status flag for NOLOCK qdisc
The queue is marked not empty after acquiring the seqlock, and it's up to the NOLOCK qdisc clearing such flag on dequeue. Since the empty status lays on the same cache-line of the seqlock, it's always hot on cache during the updates.
This makes the empty flag update a little bit loosy. Given the lack of synchronization between enqueue and dequeue, this is unavoidable.
v2 -> v3:
- qdisc_is_empty() has a const argument (Eric)
v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as suggested by Eric
--- include/net/sch_generic.h | 11 +++++++++++ net/sched/sch_generic.c | 3 +++ 2 files changed, 14 insertions(+)
include/net/sch_generic.h +11 -0
@@ -113,6 +113,9 @@ struct Qdisc {spinlock_t busylock____cacheline_aligned_in_smp ;spinlock_t seqlock; + +/* for NOLOCK qdisc, true if there are no enqueued skbs */ +bool empty;struct rcu_head rcu; };
@@ -143,11 +146,19 @@ static inline bool qdisc_is_running(struct Qdisc *qdisc)return (raw_read_seqcount (&qdisc->running) &1 ) ? true : false; } +static inline bool qdisc_is_empty (const struct Qdisc *qdisc) +{ +if (qdisc->flags &TCQ_F_NOLOCK ) +return qdisc->empty; +return !qdisc->q.qlen; +} +static inline bool qdisc_run_begin (struct Qdisc *qdisc) {if (qdisc->flags &TCQ_F_NOLOCK ) {if (!spin_trylock (&qdisc->seqlock))return false; + qdisc->empty = false; }else if (qdisc_is_running (qdisc)) {return false; }
net/sched/sch_generic.c +3 -0
@@ -671,6 +671,8 @@ static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)qdisc_qstats_cpu_backlog_dec (qdisc, skb);qdisc_bstats_cpu_update (qdisc, skb);qdisc_qstats_atomic_qlen_dec (qdisc); + }else { + qdisc->empty = true; }return skb;
@@ -880,6 +882,7 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, sch->enqueue = ops->enqueue; sch->dequeue = ops->dequeue; sch->dev_queue = dev_queue; + sch->empty = true;dev_hold (dev);refcount_set (&sch->refcnt,1 );
Signature · 1 line
2.20.1
Re: [PATCH net-next v3 1/2] net: sched: add empty status flag for NOLOCK qdisc
Quoted 9 lines · Paolo Abeni
The queue is marked not empty after acquiring the seqlock, and it's up to the NOLOCK qdisc clearing such flag on dequeue. Since the empty status lays on the same cache-line of the seqlock, it's always hot on cache during the updates.
This makes the empty flag update a little bit loosy. Given the lack of synchronization between enqueue and dequeue, this is unavoidable.
Re: [PATCH net-next v3 1/2] net: sched: add empty status flag for NOLOCK qdisc
In reply to Paolo AbeniQuoted 71 lines · Paolo Abeni
The queue is marked not empty after acquiring the seqlock, and it's up to the NOLOCK qdisc clearing such flag on dequeue. Since the empty status lays on the same cache-line of the seqlock, it's always hot on cache during the updates.
This makes the empty flag update a little bit loosy. Given the lack of synchronization between enqueue and dequeue, this is unavoidable.
v2 -> v3:
-qdisc_is_empty()has a const argument (Eric)v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as suggested by EricSigned-off-by Paolo Abeni <[email protected]>--- include/net/sch_generic.h | 11 +++++++++++ net/sched/sch_generic.c | 3 +++ 2 files changed, 14 insertions(+)include/net/sch_generic.h +11
-0@@ -113,6 +113,9 @@ struct Qdisc {spinlock_t busylock____cacheline_aligned_in_smp ;spinlock_t seqlock; + +/* for NOLOCK qdisc, true if there are no enqueued skbs */ +bool empty;struct rcu_head rcu; };@@ -143,11 +146,19 @@ static inline bool qdisc_is_running(struct Qdisc *qdisc)return (raw_read_seqcount (&qdisc->running) &1 ) ? true : false; } +static inline bool qdisc_is_empty (const struct Qdisc *qdisc) +{ +if (qdisc->flags &TCQ_F_NOLOCK ) +return qdisc->empty; +return !qdisc->q.qlen; +} +static inline bool qdisc_run_begin (struct Qdisc *qdisc) {if (qdisc->flags &TCQ_F_NOLOCK ) {if (!spin_trylock (&qdisc->seqlock))return false; + qdisc->empty = false; }else if (qdisc_is_running (qdisc)) {return false; }net/sched/sch_generic.c +2
-0@@ -671,6 +671,8 @@ static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)qdisc_qstats_cpu_backlog_dec (qdisc, skb);qdisc_bstats_cpu_update (qdisc, skb);qdisc_qstats_atomic_qlen_dec (qdisc); + }else { + qdisc->empty = true; }return skb;
net/sched/sch_generic.c struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, · line 888
Quoted 7 lines
net/sched/sch_generic.c +1
-0@@ -880,6 +882,7 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, sch->enqueue = ops->enqueue; sch->dequeue = ops->dequeue; sch->dev_queue = dev_queue; + sch->empty = true;dev_hold (dev);refcount_set (&sch->refcnt,1 );
Paolo Abeni
maintainerPATCH v3 2/2 net: dev: introduce support for sch BYPASS for lockless qdisc
In reply to Paolo AbeniWith commit c5ad119fb6c0 ("net: sched: pfifo_fast use skb_array") pfifo_fast no longer benefit from the TCQ_F_CAN_BYPASS optimization. Due to retpolines the cost of the enqueue()/dequeue() pair has become relevant and we observe measurable regression for the uncontended scenario when the packet-rate is below line rate.
After commit 46b1c18f9deb ("net: sched: put back q.qlen into a single location") we can check for empty qdisc with a reasonably fast operation even for nolock qdiscs.
This change extends TCQ_F_CAN_BYPASS support to nolock qdisc. The new chunk of code mirrors closely the existing one for traditional qdisc, leveraging a newly introduced helper to read atomically the qdisc length.
Tested with pktgen in queue xmit mode, with pfifo_fast, a MQ device, and MQ root qdisc:
threads vanilla patched
kpps kpps
1 2465 2889
2 4304 5188
4 7898 9589Same as above, but with a single queue device:
threads vanilla patched
kpps kpps
1 2556 2827
2 2900 2900
4 5000 5000
8 4700 4700No mesaurable changes in the contended scenarios, and more 10% improvement in the uncontended ones.
v1 -> v2:
- rebased after flag name change
--- net/core/dev.c | 9 +++++++++ 1 file changed, 9 insertions(+)
net/core/dev.c +9 -0
@@ -3468,6 +3468,15 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,if (unlikely (test_bit (__QDISC_STATE_DEACTIVATED, &q->state))) {__qdisc_drop (skb, &to_free); rc =NET_XMIT_DROP ; + }else if ((q->flags &TCQ_F_CAN_BYPASS ) && q->empty && +qdisc_run_begin (q)) { +qdisc_bstats_cpu_update (q, skb); + +if (sch_direct_xmit (skb, q, dev, txq,NULL , true)) +__qdisc_run (q); + +qdisc_run_end (q); + rc =NET_XMIT_SUCCESS ; }else { rc = q->enqueue (skb, q, &to_free) &NET_XMIT_MASK ;qdisc_run (q);
Signature · 1 line
2.20.1
David Miller
maintainerFrom: Paolo Abeni <[email protected]>
Date: Fri, 22 Mar 2019 16:01:54 +0100
Quoted 12 lines
This patch series is aimed at improving xmit performances of lockless qdisc in the uncontended scenario.
After the lockless refactor
pfifo_fastcan't leverage the BYPASS optimization. Due to retpolines the overhead for the avoidables enqueue and dequeue operations has increased and we see measurable regressions.The first patch introduces the BYPASS code path for lockless qdisc, and the second one optimizes such path further. Overall this avoids up to 3 indirect calls per xmit packet. Detailed performance figures are reported in the 2nd patch.
Quoted 6 lines
v2 -> v3:
-qdisc_is_empty()has a const argument (Eric)v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as suggested by Eric
Looks great, series applied, thanks Paolo.