From fad8bdfa40df8636a52d770bbab010086c1976ec Mon Sep 17 00:00:00 2001 From: Felix Fietkau Date: Thu, 12 May 2016 21:04:50 +0200 Subject: kernel: backport patches improving fq_codel drop behavior Signed-off-by: Felix Fietkau --- ..._codel-add-batch-ability-to-fq_codel_drop.patch | 189 +++++++++++++++++++++ 1 file changed, 189 insertions(+) create mode 100644 target/linux/generic/patches-4.4/032-fq_codel-add-batch-ability-to-fq_codel_drop.patch (limited to 'target/linux/generic/patches-4.4/032-fq_codel-add-batch-ability-to-fq_codel_drop.patch') diff --git a/target/linux/generic/patches-4.4/032-fq_codel-add-batch-ability-to-fq_codel_drop.patch b/target/linux/generic/patches-4.4/032-fq_codel-add-batch-ability-to-fq_codel_drop.patch new file mode 100644 index 0000000000..94c9c7fe27 --- /dev/null +++ b/target/linux/generic/patches-4.4/032-fq_codel-add-batch-ability-to-fq_codel_drop.patch @@ -0,0 +1,189 @@ +From: Eric Dumazet +Date: Sun, 1 May 2016 16:47:26 -0700 +Subject: [PATCH] fq_codel: add batch ability to fq_codel_drop() + +In presence of inelastic flows and stress, we can call +fq_codel_drop() for every packet entering fq_codel qdisc. + +fq_codel_drop() is quite expensive, as it does a linear scan +of 4 KB of memory to find a fat flow. +Once found, it drops the oldest packet of this flow. + +Instead of dropping a single packet, try to drop 50% of the backlog +of this fat flow, with a configurable limit of 64 packets per round. + +TCA_FQ_CODEL_DROP_BATCH_SIZE is the new attribute to make this +limit configurable. + +With this strategy the 4 KB search is amortized to a single cache line +per drop [1], so fq_codel_drop() no longer appears at the top of kernel +profile in presence of few inelastic flows. + +[1] Assuming a 64byte cache line, and 1024 buckets + +Signed-off-by: Eric Dumazet +Reported-by: Dave Taht +Cc: Jonathan Morton +Acked-by: Jesper Dangaard Brouer +Acked-by: Dave Taht +Signed-off-by: David S. Miller +--- + +--- a/include/uapi/linux/pkt_sched.h ++++ b/include/uapi/linux/pkt_sched.h +@@ -711,6 +711,7 @@ enum { + TCA_FQ_CODEL_FLOWS, + TCA_FQ_CODEL_QUANTUM, + TCA_FQ_CODEL_CE_THRESHOLD, ++ TCA_FQ_CODEL_DROP_BATCH_SIZE, + __TCA_FQ_CODEL_MAX + }; + +--- a/net/sched/sch_fq_codel.c ++++ b/net/sched/sch_fq_codel.c +@@ -57,6 +57,7 @@ struct fq_codel_sched_data { + u32 flows_cnt; /* number of flows */ + u32 perturbation; /* hash perturbation */ + u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ ++ u32 drop_batch_size; + struct codel_params cparams; + struct codel_stats cstats; + u32 drop_overlimit; +@@ -133,17 +134,20 @@ static inline void flow_queue_add(struct + skb->next = NULL; + } + +-static unsigned int fq_codel_drop(struct Qdisc *sch) ++static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets) + { + struct fq_codel_sched_data *q = qdisc_priv(sch); + struct sk_buff *skb; + unsigned int maxbacklog = 0, idx = 0, i, len; + struct fq_codel_flow *flow; ++ unsigned int threshold; + +- /* Queue is full! Find the fat flow and drop packet from it. ++ /* Queue is full! Find the fat flow and drop packet(s) from it. + * This might sound expensive, but with 1024 flows, we scan + * 4KB of memory, and we dont need to handle a complex tree + * in fast path (packet queue/enqueue) with many cache misses. ++ * In stress mode, we'll try to drop 64 packets from the flow, ++ * amortizing this linear lookup to one cache line per drop. + */ + for (i = 0; i < q->flows_cnt; i++) { + if (q->backlogs[i] > maxbacklog) { +@@ -151,15 +155,24 @@ static unsigned int fq_codel_drop(struct + idx = i; + } + } ++ ++ /* Our goal is to drop half of this fat flow backlog */ ++ threshold = maxbacklog >> 1; ++ + flow = &q->flows[idx]; +- skb = dequeue_head(flow); +- len = qdisc_pkt_len(skb); ++ len = 0; ++ i = 0; ++ do { ++ skb = dequeue_head(flow); ++ len += qdisc_pkt_len(skb); ++ kfree_skb(skb); ++ } while (++i < max_packets && len < threshold); ++ ++ flow->dropped += i; + q->backlogs[idx] -= len; +- sch->q.qlen--; +- qdisc_qstats_drop(sch); +- qdisc_qstats_backlog_dec(sch, skb); +- kfree_skb(skb); +- flow->dropped++; ++ sch->qstats.drops += i; ++ sch->qstats.backlog -= len; ++ sch->q.qlen -= i; + return idx; + } + +@@ -168,14 +181,14 @@ static unsigned int fq_codel_qdisc_drop( + unsigned int prev_backlog; + + prev_backlog = sch->qstats.backlog; +- fq_codel_drop(sch); ++ fq_codel_drop(sch, 1U); + return prev_backlog - sch->qstats.backlog; + } + + static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch) + { + struct fq_codel_sched_data *q = qdisc_priv(sch); +- unsigned int idx, prev_backlog; ++ unsigned int idx, prev_backlog, prev_qlen; + struct fq_codel_flow *flow; + int uninitialized_var(ret); + +@@ -204,16 +217,22 @@ static int fq_codel_enqueue(struct sk_bu + return NET_XMIT_SUCCESS; + + prev_backlog = sch->qstats.backlog; +- q->drop_overlimit++; +- /* Return Congestion Notification only if we dropped a packet +- * from this flow. ++ prev_qlen = sch->q.qlen; ++ ++ /* fq_codel_drop() is quite expensive, as it performs a linear search ++ * in q->backlogs[] to find a fat flow. ++ * So instead of dropping a single packet, drop half of its backlog ++ * with a 64 packets limit to not add a too big cpu spike here. + */ +- if (fq_codel_drop(sch) == idx) +- return NET_XMIT_CN; ++ ret = fq_codel_drop(sch, q->drop_batch_size); + +- /* As we dropped a packet, better let upper stack know this */ +- qdisc_tree_reduce_backlog(sch, 1, prev_backlog - sch->qstats.backlog); +- return NET_XMIT_SUCCESS; ++ q->drop_overlimit += prev_qlen - sch->q.qlen; ++ ++ /* As we dropped packet(s), better let upper stack know this */ ++ qdisc_tree_reduce_backlog(sch, prev_qlen - sch->q.qlen, ++ prev_backlog - sch->qstats.backlog); ++ ++ return ret == idx ? NET_XMIT_CN : NET_XMIT_SUCCESS; + } + + /* This is the specific function called from codel_dequeue() +@@ -323,6 +342,7 @@ static const struct nla_policy fq_codel_ + [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, + [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, + [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, ++ [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, + }; + + static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt) +@@ -374,6 +394,9 @@ static int fq_codel_change(struct Qdisc + if (tb[TCA_FQ_CODEL_QUANTUM]) + q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); + ++ if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]) ++ q->drop_batch_size = min(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])); ++ + while (sch->q.qlen > sch->limit) { + struct sk_buff *skb = fq_codel_dequeue(sch); + +@@ -419,6 +442,7 @@ static int fq_codel_init(struct Qdisc *s + + sch->limit = 10*1024; + q->flows_cnt = 1024; ++ q->drop_batch_size = 64; + q->quantum = psched_mtu(qdisc_dev(sch)); + q->perturbation = prandom_u32(); + INIT_LIST_HEAD(&q->new_flows); +@@ -476,6 +500,8 @@ static int fq_codel_dump(struct Qdisc *s + q->cparams.ecn) || + nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, + q->quantum) || ++ nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, ++ q->drop_batch_size) || + nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, + q->flows_cnt)) + goto nla_put_failure; -- cgit v1.2.3