Message-ID: <CAEf4BzZ4c1VwPf9oBRRdN7jdBWrk4pg=mw_50LMjLr99Mb0yfw@mail.gmail.com>
Date:   Wed, 12 Jan 2022 11:20:48 -0800
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc:     Lorenzo Bianconi <lorenzo@...nel.org>, bpf <bpf@...r.kernel.org>,
        Networking <netdev@...r.kernel.org>,
        Lorenzo Bianconi <lorenzo.bianconi@...hat.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Shay Agroskin <shayagr@...zon.com>,
        john fastabend <john.fastabend@...il.com>,
        David Ahern <dsahern@...nel.org>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Eelco Chaudron <echaudro@...hat.com>,
        Jason Wang <jasowang@...hat.com>,
        Alexander Duyck <alexander.duyck@...il.com>,
        Saeed Mahameed <saeed@...nel.org>,
        Maciej Fijalkowski <maciej.fijalkowski@...el.com>,
        Magnus Karlsson <magnus.karlsson@...el.com>,
        tirthendu.sarkar@...el.com,
        Toke Høiland-Jørgensen <toke@...hat.com>
Subject: Re: [PATCH v21 bpf-next 18/23] libbpf: Add SEC name for xdp_mb programs

On Wed, Jan 12, 2022 at 11:17 AM Alexei Starovoitov
<alexei.starovoitov@...il.com> wrote:
>
> On Wed, Jan 12, 2022 at 10:24 AM Andrii Nakryiko
> <andrii.nakryiko@...il.com> wrote:
> >
> > On Wed, Jan 12, 2022 at 10:18 AM Lorenzo Bianconi <lorenzo@...nel.org> wrote:
> > >
> > > > On Sun, Jan 9, 2022 at 7:05 AM Lorenzo Bianconi <lorenzo@...nel.org> wrote:
> > > > >
> > > > > Introduce support for the following SEC entries for XDP multi-buff
> > > > > property:
> > > > > - SEC("xdp_mb/")
> > > > > - SEC("xdp_devmap_mb/")
> > > > > - SEC("xdp_cpumap_mb/")
> > > >
> > > > Libbpf seemed to went with .<suffix> rule (e.g., fentry.s for
> > > > sleepable, seems like we'll have kprobe.multi or  something along
> > > > those lines as well), so let's stay consistent and call this "xdp_mb",
> > > > "xdp_devmap.mb", "xdp_cpumap.mb" (btw, is "mb" really all that
> > > > recognizable? would ".multibuf" be too verbose?). Also, why the "/"
> > > > part? Also it shouldn't be "sloppy" either. Neither expected attach
> > > > type should be optional.  Also not sure SEC_ATTACHABLE is needed. So
> > > > at most it should be SEC_XDP_MB, probably.
> > >
> > > ack, I fine with it. Something like:
> > >
> > >         SEC_DEF("lsm.s/",               LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
> > >         SEC_DEF("iter/",                TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
> > >         SEC_DEF("syscall",              SYSCALL, 0, SEC_SLEEPABLE),
> > > +       SEC_DEF("xdp_devmap.multibuf",  XDP, BPF_XDP_DEVMAP, 0),
> > >         SEC_DEF("xdp_devmap/",          XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
> > > +       SEC_DEF("xdp_cpumap.multibuf",  XDP, BPF_XDP_CPUMAP, 0),
> > >         SEC_DEF("xdp_cpumap/",          XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
> > > +       SEC_DEF("xdp.multibuf",         XDP, BPF_XDP, 0),
> >
> > yep, but please use SEC_NONE instead of zero
> >
> > >         SEC_DEF("xdp",                  XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
> > >         SEC_DEF("perf_event",           PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX),
> > >         SEC_DEF("lwt_in",               LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX),
> > >
> > > >
> > > > >
> > > > > Acked-by: Toke Hoiland-Jorgensen <toke@...hat.com>
> > > > > Acked-by: John Fastabend <john.fastabend@...il.com>
> > > > > Signed-off-by: Lorenzo Bianconi <lorenzo@...nel.org>
> > > > > ---
> > > > >  tools/lib/bpf/libbpf.c | 8 ++++++++
> > > > >  1 file changed, 8 insertions(+)
> > > > >
> > > > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> > > > > index 7f10dd501a52..c93f6afef96c 100644
> > > > > --- a/tools/lib/bpf/libbpf.c
> > > > > +++ b/tools/lib/bpf/libbpf.c
> > > > > @@ -235,6 +235,8 @@ enum sec_def_flags {
> > > > >         SEC_SLEEPABLE = 8,
> > > > >         /* allow non-strict prefix matching */
> > > > >         SEC_SLOPPY_PFX = 16,
> > > > > +       /* BPF program support XDP multi-buff */
> > > > > +       SEC_XDP_MB = 32,
> > > > >  };
> > > > >
> > > > >  struct bpf_sec_def {
> > > > > @@ -6562,6 +6564,9 @@ static int libbpf_preload_prog(struct bpf_program *prog,
> > > > >         if (def & SEC_SLEEPABLE)
> > > > >                 opts->prog_flags |= BPF_F_SLEEPABLE;
> > > > >
> > > > > +       if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_MB))
> > > > > +               opts->prog_flags |= BPF_F_XDP_MB;
> > > >
> > > > I'd say you don't even need SEC_XDP_MB flag at all, you can just check
> > > > that prog->sec_name is one of "xdp.mb", "xdp_devmap.mb" or
> > > > "xdp_cpumap.mb" and add the flag. SEC_XDP_MB doesn't seem generic
> > > > enough to warrant a flag.
> > >
> > > ack, something like:
> > >
> > > +       if (prog->type == BPF_PROG_TYPE_XDP &&
> > > +           (!strcmp(prog->sec_name, "xdp_devmap.multibuf") ||
> > > +            !strcmp(prog->sec_name, "xdp_cpumap.multibuf") ||
> > > +            !strcmp(prog->sec_name, "xdp.multibuf")))
> > > +               opts->prog_flags |= BPF_F_XDP_MB;
> >
> > yep, can also simplify it a bit with strstr(prog->sec_name,
> > ".multibuf") instead of three strcmp
>
> Maybe ".mb" ?
> ".multibuf" is too verbose.
> We're fine with ".s" for sleepable :)


I had reservations about "mb" because the first and strong association
is "megabyte", not "multibuf". And it's not like anyone would have
tens of those programs in a single file so that ".multibuf" becomes
way too verbose. But I don't feel too strongly about this, if the
consensus is on ".mb".

Powered by blists - more mailing lists