Thanks!
I also noticed that there seems to be no kernel code to set the MPU
values.
# latest commit at the time of writing
administrator@UX32LN:~/src/linux$ git rev-parse HEAD
64291f7db5bd8150a74ad2036f1037e6a0428df2
Some cscope output below:
0 include/uapi/linux/pkt_sched.h 84 struct tc_ratespec {
A ratespec structure that holds an mpu value:
struct tc_ratespec {
unsigned char cell_log;
__u8 linklayer; /* lower 4 bits */
unsigned short overhead;
short cell_align;
unsigned short mpu;
__u32 rate;
};
0 include/net/sch_generic.h psched_l2t_ns 766 static inline u64
psched_l2t_ns(const struct psched_ratecfg *r,
An L2T (Length to Time) conversion function which returns an amount of
time in nanoseconds needed to send a packet of a specific size.
Here the length is first modified by a manually configured overhead
value - could be a good place to round the length up to mpu if it is
smaller than mpu:
static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
unsigned int len)
{
len += r->overhead;
if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r
->shift;
return ((u64)len * r->mult) >> r->shift;
}
Seems like psched_l2t_ns is used in both htb and tbf and affects the
amount of tokens that are going to be consumed by a packet of a certain
size:
C symbol: psched_l2t_ns
File Function Line
0 include/net/sch_generic.h psched_l2t_ns 766 static inline u64
psched_l2t_ns(const struct psched_ratecfg *r,
1 net/sched/act_police.c tcf_act_police_locate 221 police
->tcfp_mtu_ptoks = (s64) psched_l2t_ns(&police->peak,
2 net/sched/act_police.c tcf_act_police 289 ptoks -= (s64)
psched_l2t_ns(&police->peak,
3 net/sched/act_police.c tcf_act_police 295 toks -= (s64)
psched_l2t_ns(&police->rate, qdisc_pkt_len(skb));
4 net/sched/sch_htb.c htb_accnt_tokens 613 toks -= (s64)
psched_l2t_ns(&cl->rate, bytes);
5 net/sched/sch_htb.c htb_accnt_ctokens 626 toks -= (s64)
psched_l2t_ns(&cl->ceil, bytes);
6 net/sched/sch_tbf.c tbf_dequeue 249 ptoks -= (s64)
psched_l2t_ns(&q->peak, len);
7 net/sched/sch_tbf.c tbf_dequeue 254 toks -= (s64)
psched_l2t_ns(&q->rate, len);
8 net/sched/sch_tbf.c tbf_change 351 buffer = psched_l
2t_ns(&rate, max_size);
9 net/sched/sch_tbf.c tbf_change 370 mtu =
psched_l2t_ns(&peak, pburst);
Also found a patch (https://patchwork.ozlabs.org/patch/501666/) that
was intended to fix setting of mpu values which never got pushed.
Printing definitely deserved a fix but there is some more work to it at
the kernel side.
On Ср., 2016-02-17 at 17:51 -0800, Stephen Hemminger wrote:
> On Sat, 19 Dec 2015 18:25:52 +0300
> Dmitrii Shcherbakov <fw.dmit...@yandex.com> wrote:
>
> > Remove printing according to the previously used encoding of mpu
> > and overhead values within the tc_ratespec's mpu field. This
> > encoding is no longer being used as a separate 'overhead' field in
> > the ratespec structure has been introduced.
> >
> > Signed-off-by: Dmitrii Shcherbakov <fw.dmit...@yandex.com>
> > Acked-by: Jesper Dangaard Brouer <bro...@redhat.com>
> > Acked-by: Phil Sutter <p...@nwl.cc>
>
> Both applied.
>
> I had to fix up the commit logs.
> Many tools don't like long subject/summary lines, and it is standard
> practice
> to wrap text in commit body at 80 characters.
