tcp/fq: move back to CLOCK_MONOTONIC
Subsystems (4)
PATCH 1/1 tcp/fq: move back to CLOCK_MONOTONIC
In the recent TCP/EDT patch series, I switched TCP and sch_fq clocks from MONOTONIC to TAI, in order to meet the choice done earlier for sch_etf packet scheduler.
But sure enough, this broke some setups were the TAI clock jumps forward (by almost 50 year...), as reported by Leonard Crestez.
If we want to converge later, we'll probably need to add an skb field to differentiate the clock bases, or a socket option.
In the meantime, an UDP application will need to use CLOCK_MONOTONIC base for its SCM_TXTIME timestamps if using fq packet scheduler.
--- include/net/tcp.h | 2 +- net/ipv4/tcp_timer.c | 2 +- net/sched/sch_fq.c | 6 +++--- 3 files changed, 5 insertions(+), 5 deletions(-)
include/net/tcp.h +1 -1
@@ -732,7 +732,7 @@ void tcp_send_window_probe(struct sock *sk);static inline u64 tcp_clock_ns (void ) { -return ktime_get_tai_ns (); +return ktime_get_ns (); }static inline u64 tcp_clock_us (void )
net/ipv4/tcp_timer.c +1 -1
@@ -758,7 +758,7 @@ void tcp_init_xmit_timers(struct sock *sk) {inet_csk_init_xmit_timers (sk, &tcp_write_timer, &tcp_delack_timer, &tcp_keepalive_timer); -hrtimer_init (&tcp_sk (sk)->pacing_timer,CLOCK_TAI , +hrtimer_init (&tcp_sk (sk)->pacing_timer,CLOCK_MONOTONIC ,HRTIMER_MODE_ABS_PINNED_SOFT );tcp_sk (sk)->pacing_timer.function = tcp_pace_kick;
net/sched/sch_fq.c +3 -3
@@ -412,7 +412,7 @@ static void fq_check_throttled(struct fq_sched_data *q, u64 now)static struct sk_buff *fq_dequeue (struct Qdisc *sch) {struct fq_sched_data *q =qdisc_priv (sch); -u64 now =ktime_get_tai_ns (); +u64 now =ktime_get_ns ();struct fq_flow_head *head;struct sk_buff *skb;struct fq_flow *f;
@@ -776,7 +776,7 @@ static int fq_init(struct Qdisc *sch, struct nlattr *opt, q->fq_trees_log =ilog2 (1024 ); q->orphan_mask =1024 -1 ; q->low_rate_threshold =550000 /8 ; -qdisc_watchdog_init_clockid (&q->watchdog, sch,CLOCK_TAI ); +qdisc_watchdog_init_clockid (&q->watchdog, sch,CLOCK_MONOTONIC );if (opt) err =fq_change (sch, opt, extack);
@@ -831,7 +831,7 @@ static int fq_dump_stats(struct Qdisc *sch, struct gnet_dump *d) st.flows_plimit = q->stat_flows_plimit; st.pkts_too_long = q->stat_pkts_too_long; st.allocation_errors = q->stat_allocation_errors; - st.time_next_delayed_flow = q->time_next_delayed_flow -ktime_get_tai_ns (); + st.time_next_delayed_flow = q->time_next_delayed_flow -ktime_get_ns (); st.flows = q->flows; st.inactive_flows = q->inactive_flows; st.throttled_flows = q->throttled_flows;
Signature · 1 line
2.19.0.605.g01d371f741-goog
Quoted 11 lines · Eric Dumazet
In the recent TCP/EDT patch series, I switched TCP and
sch_fqclocks from MONOTONIC to TAI, in order to meet the choice done earlier forsch_etfpacket scheduler.But sure enough, this broke some setups were the TAI clock jumps forward (by almost 50 year...), as reported by Leonard Crestez.
If we want to converge later, we'll probably need to add an skb field to differentiate the clock bases, or a socket option.
Quoted 8 lines
In the meantime, an UDP application will need to use
CLOCK_MONOTONICbase for itsSCM_TXTIMEtimestamps if using fq packet scheduler.Fixes 72b0094f9182 ("tcp: switch tcp_clock_ns() to CLOCK_TAI base")Fixes 142537e41923 ("net_sched: sch_fq: switch to CLOCK_TAI")Fixes fd2bca2aa789 ("tcp: switch internal pacing timer to CLOCK_TAI")Signed-off-by Eric Dumazet <[email protected]>Reported-by Leonard Crestez <[email protected]>
Fixes the problem reported earlier when applied on top of next-20180928
Quoted 21 lines · Leonard Crestez
Quoted 11 lines · Eric Dumazet
In the recent TCP/EDT patch series, I switched TCP and
sch_fqclocks from MONOTONIC to TAI, in order to meet the choice done earlier forsch_etfpacket scheduler.But sure enough, this broke some setups were the TAI clock jumps forward (by almost 50 year...), as reported by Leonard Crestez.
If we want to converge later, we'll probably need to add an skb field to differentiate the clock bases, or a socket option.
Quoted 8 lines
In the meantime, an UDP application will need to use
CLOCK_MONOTONICbase for itsSCM_TXTIMEtimestamps if using fq packet scheduler.Fixes 72b0094f9182 ("tcp: switch tcp_clock_ns() to CLOCK_TAI base")Fixes 142537e41923 ("net_sched: sch_fq: switch to CLOCK_TAI")Fixes fd2bca2aa789 ("tcp: switch internal pacing timer to CLOCK_TAI")Signed-off-by Eric Dumazet <[email protected]>Reported-by Leonard Crestez <[email protected]>
Quoted 4 lines
Tested-by Leonard Crestez <[email protected]>Fixes the problem reported earlier when applied on top of next-20180928
Thanks again ;)
From: Eric Dumazet <[email protected]>
Date: Fri, 28 Sep 2018 10:28:44 -0700
Quoted 11 lines
In the recent TCP/EDT patch series, I switched TCP and
sch_fqclocks from MONOTONIC to TAI, in order to meet the choice done earlier forsch_etfpacket scheduler.But sure enough, this broke some setups were the TAI clock jumps forward (by almost 50 year...), as reported by Leonard Crestez.
If we want to converge later, we'll probably need to add an skb field to differentiate the clock bases, or a socket option.
Quoted 8 lines
In the meantime, an UDP application will need to use
CLOCK_MONOTONICbase for itsSCM_TXTIMEtimestamps if using fq packet scheduler.Fixes 72b0094f9182 ("tcp: switch tcp_clock_ns() to CLOCK_TAI base")Fixes 142537e41923 ("net_sched: sch_fq: switch to CLOCK_TAI")Fixes fd2bca2aa789 ("tcp: switch internal pacing timer to CLOCK_TAI")Signed-off-by Eric Dumazet <[email protected]>Reported-by Leonard Crestez <[email protected]>
Applied.