On Mon, Jul 1, 2019 at 5:31 PM Stanislav Fomichev <sdf@xxxxxxxxxxx> wrote:
>
> On 07/01, Y Song wrote:
> > On Mon, Jul 1, 2019 at 1:49 PM Stanislav Fomichev <sdf@xxxxxxxxxx> wrote:
> > >
> > > Uses new RTT callback to dump stats every second.
> > >
> > > $ mkdir -p /tmp/cgroupv2
> > > $ mount -t cgroup2 none /tmp/cgroupv2
> > > $ mkdir -p /tmp/cgroupv2/foo
> > > $ echo $$ >> /tmp/cgroupv2/foo/cgroup.procs
> > > $ bpftool prog load ./tcp_dumpstats_kern.o /sys/fs/bpf/tcp_prog
> > > $ bpftool cgroup attach /tmp/cgroupv2/foo sock_ops pinned /sys/fs/bpf/tcp_prog
> > > $ bpftool prog tracelog
> > > $ # run neper/netperf/etc
> > >
> > > Used neper to compare performance with and without this program attached
> > > and didn't see any noticeable performance impact.
> > >
> > > Sample output:
> > > <idle>-0 [015] ..s. 2074.128800: 0: dsack_dups=0 delivered=242526
> > > <idle>-0 [015] ..s. 2074.128808: 0: delivered_ce=0 icsk_retransmits=0
> > > <idle>-0 [015] ..s. 2075.130133: 0: dsack_dups=0 delivered=323599
> > > <idle>-0 [015] ..s. 2075.130138: 0: delivered_ce=0 icsk_retransmits=0
> > > <idle>-0 [005] .Ns. 2076.131440: 0: dsack_dups=0 delivered=404648
> > > <idle>-0 [005] .Ns. 2076.131447: 0: delivered_ce=0 icsk_retransmits=0
> > >
> > > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > Cc: Priyaranjan Jha <priyarjha@xxxxxxxxxx>
> > > Cc: Yuchung Cheng <ycheng@xxxxxxxxxx>
> > > Cc: Soheil Hassas Yeganeh <soheil@xxxxxxxxxx>
> > > Signed-off-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > > ---
> > > samples/bpf/Makefile | 1 +
> > > samples/bpf/tcp_dumpstats_kern.c | 65 ++++++++++++++++++++++++++++++++
> > > 2 files changed, 66 insertions(+)
> > > create mode 100644 samples/bpf/tcp_dumpstats_kern.c
> >
> > Currently, the bpf program into the repo. If we do not have another
> > script to use
> > this program for testing, the instructions in the commit message should be
> > added to the bpf program as comments so people know what to do with this file
> > without going through git commit message.
> >
> > Is it possible to create a script to run with this bpf program?
> There is a general instruction in samples/bpf/tcp_bpf.readme
> with bpftool examples/etc. Should I just a comment at the top
> of the BPF program to point people to that .readme file instead?
Referring to tcp_bpf.readme should work. Even simpler :-)
>
> > >
> > > diff --git a/samples/bpf/Makefile b/samples/bpf/Makefile
> > > index 0917f8cf4fab..eaebbeead42f 100644
> > > --- a/samples/bpf/Makefile
> > > +++ b/samples/bpf/Makefile
> > > @@ -154,6 +154,7 @@ always += tcp_iw_kern.o
> > > always += tcp_clamp_kern.o
> > > always += tcp_basertt_kern.o
> > > always += tcp_tos_reflect_kern.o
> > > +always += tcp_dumpstats_kern.o
> > > always += xdp_redirect_kern.o
> > > always += xdp_redirect_map_kern.o
> > > always += xdp_redirect_cpu_kern.o
> > > diff --git a/samples/bpf/tcp_dumpstats_kern.c b/samples/bpf/tcp_dumpstats_kern.c
> > > new file mode 100644
> > > index 000000000000..5d22bf61db65
> > > --- /dev/null
> > > +++ b/samples/bpf/tcp_dumpstats_kern.c
> > > @@ -0,0 +1,65 @@
> > > +// SPDX-License-Identifier: GPL-2.0
> > > +#include <linux/bpf.h>
> > > +
> > > +#include "bpf_helpers.h"
> > > +#include "bpf_endian.h"
> > > +
> > > +#define INTERVAL 1000000000ULL
> > > +
> > > +int _version SEC("version") = 1;
> > > +char _license[] SEC("license") = "GPL";
[...]
