openwrt/staging/blogic.git
5 years agoselftests/bpf: add kprobe/uprobe selftests
Andrii Nakryiko [Mon, 1 Jul 2019 23:59:02 +0000 (16:59 -0700)]
selftests/bpf: add kprobe/uprobe selftests

Add tests verifying kprobe/kretprobe/uprobe/uretprobe APIs work as
expected.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: switch test to new attach_perf_event API
Andrii Nakryiko [Mon, 1 Jul 2019 23:59:01 +0000 (16:59 -0700)]
selftests/bpf: switch test to new attach_perf_event API

Use new bpf_program__attach_perf_event() in test previously relying on
direct ioctl manipulations.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: add raw tracepoint attach API
Andrii Nakryiko [Mon, 1 Jul 2019 23:59:00 +0000 (16:59 -0700)]
libbpf: add raw tracepoint attach API

Add a wrapper utilizing bpf_link "infrastructure" to allow attaching BPF
programs to raw tracepoints.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: add tracepoint attach API
Andrii Nakryiko [Mon, 1 Jul 2019 23:58:59 +0000 (16:58 -0700)]
libbpf: add tracepoint attach API

Allow attaching BPF programs to kernel tracepoint BPF hooks specified by
category and name.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: add kprobe/uprobe attach API
Andrii Nakryiko [Mon, 1 Jul 2019 23:58:58 +0000 (16:58 -0700)]
libbpf: add kprobe/uprobe attach API

Add ability to attach to kernel and user probes and retprobes.
Implementation depends on perf event support for kprobes/uprobes.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: add ability to attach/detach BPF program to perf event
Andrii Nakryiko [Mon, 1 Jul 2019 23:58:57 +0000 (16:58 -0700)]
libbpf: add ability to attach/detach BPF program to perf event

bpf_program__attach_perf_event allows to attach BPF program to existing
perf event hook, providing most generic and most low-level way to attach BPF
programs. It returns struct bpf_link, which should be passed to
bpf_link__destroy to detach and free resources, associated with a link.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: introduce concept of bpf_link
Andrii Nakryiko [Mon, 1 Jul 2019 23:58:56 +0000 (16:58 -0700)]
libbpf: introduce concept of bpf_link

bpf_link is an abstraction of an association of a BPF program and one of
many possible BPF attachment points (hooks). This allows to have uniform
interface for detaching BPF programs regardless of the nature of link
and how it was created. Details of creation and setting up of a specific
bpf_link is handled by corresponding attachment methods
(bpf_program__attach_xxx) added in subsequent commits. Once successfully
created, bpf_link has to be eventually destroyed with
bpf_link__destroy(), at which point BPF program is disassociated from
a hook and all the relevant resources are freed.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: make libbpf_strerror_r agnostic to sign of error
Andrii Nakryiko [Mon, 1 Jul 2019 23:58:55 +0000 (16:58 -0700)]
libbpf: make libbpf_strerror_r agnostic to sign of error

It's often inconvenient to switch sign of error when passing it into
libbpf_strerror_r. It's better for it to handle that automatically.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Thu, 4 Jul 2019 19:48:21 +0000 (12:48 -0700)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2019-07-03

The following pull-request contains BPF updates for your *net-next* tree.

There is a minor merge conflict in mlx5 due to 8960b38932be ("linux/dim:
Rename externally used net_dim members") which has been pulled into your
tree in the meantime, but resolution seems not that bad ... getting current
bpf-next out now before there's coming more on mlx5. ;) I'm Cc'ing Saeed
just so he's aware of the resolution below:

** First conflict in drivers/net/ethernet/mellanox/mlx5/core/en_main.c:

  <<<<<<< HEAD
  static int mlx5e_open_cq(struct mlx5e_channel *c,
                           struct dim_cq_moder moder,
                           struct mlx5e_cq_param *param,
                           struct mlx5e_cq *cq)
  =======
  int mlx5e_open_cq(struct mlx5e_channel *c, struct net_dim_cq_moder moder,
                    struct mlx5e_cq_param *param, struct mlx5e_cq *cq)
  >>>>>>> e5a3e259ef239f443951d401db10db7d426c9497

Resolution is to take the second chunk and rename net_dim_cq_moder into
dim_cq_moder. Also the signature for mlx5e_open_cq() in ...

  drivers/net/ethernet/mellanox/mlx5/core/en.h +977

... and in mlx5e_open_xsk() ...

  drivers/net/ethernet/mellanox/mlx5/core/en/xsk/setup.c +64

... needs the same rename from net_dim_cq_moder into dim_cq_moder.

** Second conflict in drivers/net/ethernet/mellanox/mlx5/core/en_main.c:

  <<<<<<< HEAD
          int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(priv->mdev, ix));
          struct dim_cq_moder icocq_moder = {0, 0};
          struct net_device *netdev = priv->netdev;
          struct mlx5e_channel *c;
          unsigned int irq;
  =======
          struct net_dim_cq_moder icocq_moder = {0, 0};
  >>>>>>> e5a3e259ef239f443951d401db10db7d426c9497

Take the second chunk and rename net_dim_cq_moder into dim_cq_moder
as well.

Let me know if you run into any issues. Anyway, the main changes are:

1) Long-awaited AF_XDP support for mlx5e driver, from Maxim.

2) Addition of two new per-cgroup BPF hooks for getsockopt and
   setsockopt along with a new sockopt program type which allows more
   fine-grained pass/reject settings for containers. Also add a sock_ops
   callback that can be selectively enabled on a per-socket basis and is
   executed for every RTT to help tracking TCP statistics, both features
   from Stanislav.

3) Follow-up fix from loops in precision tracking which was not propagating
   precision marks and as a result verifier assumed that some branches were
   not taken and therefore wrongly removed as dead code, from Alexei.

4) Fix BPF cgroup release synchronization race which could lead to a
   double-free if a leaf's cgroup_bpf object is released and a new BPF
   program is attached to the one of ancestor cgroups in parallel, from Roman.

5) Support for bulking XDP_TX on veth devices which improves performance
   in some cases by around 9%, from Toshiaki.

6) Allow for lookups into BPF devmap and improve feedback when calling into
   bpf_redirect_map() as lookup is now performed right away in the helper
   itself, from Toke.

7) Add support for fq's Earliest Departure Time to the Host Bandwidth
   Manager (HBM) sample BPF program, from Lawrence.

8) Various cleanups and minor fixes all over the place from many others.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: mediatek: Fix overlapping capability bits.
René van Dorst [Wed, 3 Jul 2019 18:42:04 +0000 (20:42 +0200)]
net: ethernet: mediatek: Fix overlapping capability bits.

Both MTK_TRGMII_MT7621_CLK and MTK_PATH_BIT are defined as bit 10.

This can causes issues on non-MT7621 devices which has the
MTK_PATH_BIT(MTK_ETH_PATH_GMAC1_RGMII) and MTK_TRGMII capability set.
The wrong TRGMII setup code can be executed. The current wrongly executed
code doesn’t do any harm on MT7623 and the TRGMII setup for the MT7623
SOC side is done in MT7530 driver So it wasn’t noticed in the test.

Move all capability bits in one enum so that they are all unique and easy
to expand in the future.

Because mtk_eth_path enum is merged in to mkt_eth_capabilities, the
variable path value is no longer between 0 to number of paths,
mtk_eth_path_name can’t be used anymore in this form. Convert the
mtk_eth_path_name array to a function to lookup the pathname.

The old code walked thru the mtk_eth_path enum, which is also merged
with mkt_eth_capabilities. Expand array mtk_eth_muxc so it can store the
name and capability bit of the mux. Convert the code so it can walk thru
the mtk_eth_muxc array.

Fixes: 8efaa653a8a5 ("net: ethernet: mediatek: Add MT7621 TRGMII mode support")
Signed-off-by: René van Dorst <opensource@vdorst.com>
v1->v2:
- Move all capability bits in one enum, suggested by Willem de Bruijn
- Convert the mtk_eth_path_name array to a function to lookup the pathname
- Expand array mtk_eth_muxc so it can also store the name and capability
  bit of the mux
- Updated commit message

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Enable dwmac4 jumbo frame more than 8KiB
Weifeng Voon [Wed, 3 Jul 2019 16:59:10 +0000 (00:59 +0800)]
net: stmmac: Enable dwmac4 jumbo frame more than 8KiB

Enable GMAC v4.xx and beyond to support 16KiB buffer.

Signed-off-by: Weifeng Voon <weifeng.voon@intel.com>
Signed-off-by: Ong Boon Leong <boon.leong.ong@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobonding: add an option to specify a delay between peer notifications
Vincent Bernat [Tue, 2 Jul 2019 17:43:54 +0000 (19:43 +0200)]
bonding: add an option to specify a delay between peer notifications

Currently, gratuitous ARP/ND packets are sent every `miimon'
milliseconds. This commit allows a user to specify a custom delay
through a new option, `peer_notif_delay'.

Like for `updelay' and `downdelay', this delay should be a multiple of
`miimon' to avoid managing an additional work queue. The configuration
logic is copied from `updelay' and `downdelay'. However, the default
value cannot be set using a module parameter: Netlink or sysfs should
be used to configure this feature.

When setting `miimon' to 100 and `peer_notif_delay' to 500, we can
observe the 500 ms delay is respected:

    20:30:19.354693 ARP, Request who-has 203.0.113.10 tell 203.0.113.10, length 28
    20:30:19.874892 ARP, Request who-has 203.0.113.10 tell 203.0.113.10, length 28
    20:30:20.394919 ARP, Request who-has 203.0.113.10 tell 203.0.113.10, length 28
    20:30:20.914963 ARP, Request who-has 203.0.113.10 tell 203.0.113.10, length 28

In bond_mii_monitor(), I have tried to keep the lock logic readable.
The change is due to the fact we cannot rely on a notification to
lower the value of `bond->send_peer_notif' as `NETDEV_NOTIFY_PEERS' is
only triggered once every N times, while we need to decrement the
counter each time.

iproute2 also needs to be updated to be able to specify this new
attribute through `ip link'.

Signed-off-by: Vincent Bernat <vincent@bernat.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: sun: remove redundant assignment to variable err
Colin Ian King [Thu, 4 Jul 2019 12:36:51 +0000 (13:36 +0100)]
net: ethernet: sun: remove redundant assignment to variable err

The variable err is being assigned with a value that is never
read and it is being updated in the next statement with a new value.
The assignment is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: fix -ENOMEM null check on a page allocation
Colin Ian King [Wed, 3 Jul 2019 16:50:37 +0000 (17:50 +0100)]
gve: fix -ENOMEM null check on a page allocation

Currently the check to see if a page is allocated is incorrect
and is checking if the pointer page is null, not *page as
intended.  Fix this.

Addresses-Coverity: ("Dereference before null check")
Fixes: f5cedc84a30d ("gve: Add transmit and receive support")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-ICW-sendmsg-recvmsg'
David S. Miller [Wed, 3 Jul 2019 20:51:54 +0000 (13:51 -0700)]
Merge branch 'net-ICW-sendmsg-recvmsg'

Paolo Abeni says:

====================
net: use ICW for sk_proto->{send,recv}msg

This series extends ICW usage to one of the few remaining spots in fast-path
still hitting per packet retpoline overhead, namely the sk_proto->{send,recv}msg
calls.

The first 3 patches in this series refactor the existing code so that applying
the ICW macros is straight-forward: we demux inet_{recv,send}msg in ipv4 and
ipv6 variants so that each of them can easily select the appropriate TCP or UDP
direct call. While at it, a new helper is created to avoid excessive code
duplication, and the current ICWs for inet_{recv,send}msg are adjusted
accordingly.

The last 2 patches really introduce the new ICW use-case, respectively for the
ipv6 and the ipv4 code path.

This gives up to 5% performance improvement under UDP flood, and smaller but
measurable gains for TCP RR workloads.

v1 -> v2:
 - drop inet6_{recv,send}msg declaration from header file,
   prefer ICW macro instead
 - avoid unneeded reclaration for udp_sendmsg, as suggested by Willem
====================

Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: use indirect call wrappers for {tcp, udp}_{recv, send}msg()
Paolo Abeni [Wed, 3 Jul 2019 14:06:56 +0000 (16:06 +0200)]
ipv4: use indirect call wrappers for {tcp, udp}_{recv, send}msg()

This avoids an indirect call per syscall for common ipv4 transports

v1 -> v2:
 - avoid unneeded reclaration for udp_sendmsg, as suggested by Willem

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: use indirect call wrappers for {tcp, udpv6}_{recv, send}msg()
Paolo Abeni [Wed, 3 Jul 2019 14:06:55 +0000 (16:06 +0200)]
ipv6: use indirect call wrappers for {tcp, udpv6}_{recv, send}msg()

This avoids an indirect call per syscall for common ipv6 transports

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: adjust socket level ICW to cope with ipv6 variant of {recv, send}msg
Paolo Abeni [Wed, 3 Jul 2019 14:06:54 +0000 (16:06 +0200)]
net: adjust socket level ICW to cope with ipv6 variant of {recv, send}msg

After the previous patch we have ipv{6,4} variants for {recv,send}msg,
we should use the generic _INET ICW variant to call into the proper
build-in.

This also allows dropping the now unused and rather ugly _INET4 ICW macro

v1 -> v2:
 - use ICW macro to declare inet6_{recv,send}msg
 - fix a couple of checkpatch offender in the code context

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: provide and use ipv6 specific version for {recv, send}msg
Paolo Abeni [Wed, 3 Jul 2019 14:06:53 +0000 (16:06 +0200)]
ipv6: provide and use ipv6 specific version for {recv, send}msg

This will simplify indirect call wrapper invocation in the following
patch.

No functional change intended, any - out-of-tree - IPv6 user of
inet_{recv,send}msg can keep using the existing functions.

SCTP code still uses the existing version even for ipv6: as this series
will not add ICW for SCTP, moving to the new helper would not give
any benefit.

The only other in-kernel user of inet_{recv,send}msg is
pvcalls_conn_back_read(), but psvcalls explicitly creates only IPv4 socket,
so no need to update that code path, too.

v1 -> v2: drop inet6_{recv,send}msg declaration from header file,
   prefer ICW macro instead

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoinet: factor out inet_send_prepare()
Paolo Abeni [Wed, 3 Jul 2019 14:06:52 +0000 (16:06 +0200)]
inet: factor out inet_send_prepare()

The same code is replicated verbatim in multiple places, and the next
patches will introduce an additional user for it. Factor out a
helper and use it where appropriate. No functional change intended.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqlcnic: remove redundant assignment to variable err
Colin Ian King [Wed, 3 Jul 2019 08:32:14 +0000 (09:32 +0100)]
qlcnic: remove redundant assignment to variable err

The variable err is being initialized with a value that is never
read and it is being updated later with a new value. The
initialization is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoatl1c: remove redundant assignment to variable tpd_req
Colin Ian King [Wed, 3 Jul 2019 07:53:58 +0000 (08:53 +0100)]
atl1c: remove redundant assignment to variable tpd_req

The variable tpd_req is being initialized with a value that is never
read and it is being updated later with a new value. The
initialization is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Add support for Timestamping the unicast PTP packets.
Sudarsana Reddy Kalluru [Wed, 3 Jul 2019 06:01:59 +0000 (23:01 -0700)]
qed: Add support for Timestamping the unicast PTP packets.

This patch adds driver changes to detect/timestamp the unicast PTP packets.

Changes from previous version:
-------------------------------
v2: Defined a macro for unicast ptp param mask.

Please consider applying this to "net-next".

Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: Fix u64_stats_sync to initialize start
Catherine Sullivan [Tue, 2 Jul 2019 22:46:57 +0000 (15:46 -0700)]
gve: Fix u64_stats_sync to initialize start

u64_stats_fetch_begin needs to initialize start.

Signed-off-by: Catherine Sullivan <csully@google.com>
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoloopback: fix lockdep splat
Mahesh Bandewar [Wed, 3 Jul 2019 06:16:31 +0000 (23:16 -0700)]
loopback: fix lockdep splat

dev_init_scheduler() and dev_activate() expect the caller to
hold RTNL. Since we don't want blackhole device to be initialized
per ns, we are initializing at init.

[    3.855027] Call Trace:
[    3.855034]  dump_stack+0x67/0x95
[    3.855037]  lockdep_rcu_suspicious+0xd5/0x110
[    3.855044]  dev_init_scheduler+0xe3/0x120
[    3.855048]  ? net_olddevs_init+0x60/0x60
[    3.855050]  blackhole_netdev_init+0x45/0x6e
[    3.855052]  do_one_initcall+0x6c/0x2fa
[    3.855058]  ? rcu_read_lock_sched_held+0x8c/0xa0
[    3.855066]  kernel_init_freeable+0x1e5/0x288
[    3.855071]  ? rest_init+0x260/0x260
[    3.855074]  kernel_init+0xf/0x180
[    3.855076]  ? rest_init+0x260/0x260
[    3.855078]  ret_from_fork+0x24/0x30

Fixes: 4de83b88c66 ("loopback: create blackhole net device similar to loopack.")
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bpf-tcp-rtt-hook'
Daniel Borkmann [Wed, 3 Jul 2019 14:52:03 +0000 (16:52 +0200)]
Merge branch 'bpf-tcp-rtt-hook'

Stanislav Fomichev says:

====================
Congestion control team would like to have a periodic callback to
track some TCP statistics. Let's add a sock_ops callback that can be
selectively enabled on a socket by socket basis and is executed for
every RTT. BPF program frequency can be further controlled by calling
bpf_ktime_get_ns and bailing out early.

I run neper tcp_stream and tcp_rr tests with the sample program
from the last patch and didn't observe any noticeable performance
difference.

v2:
* add a comment about second accept() in selftest (Yonghong Song)
* refer to tcp_bpf.readme in sample program (Yonghong Song)
====================

Suggested-by: Eric Dumazet <edumazet@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Yonghong Song <yhs@fb.com>
Acked-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: fix tcp_bpf.readme detach command
Stanislav Fomichev [Tue, 2 Jul 2019 16:14:03 +0000 (09:14 -0700)]
samples/bpf: fix tcp_bpf.readme detach command

Copy-paste, should be detach, not attach.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agosamples/bpf: add sample program that periodically dumps TCP stats
Stanislav Fomichev [Tue, 2 Jul 2019 16:14:02 +0000 (09:14 -0700)]
samples/bpf: add sample program that periodically dumps TCP stats

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@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: test BPF_SOCK_OPS_RTT_CB
Stanislav Fomichev [Tue, 2 Jul 2019 16:14:01 +0000 (09:14 -0700)]
selftests/bpf: test BPF_SOCK_OPS_RTT_CB

Make sure the callback is invoked for syn-ack and data packet.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf/tools: sync bpf.h
Stanislav Fomichev [Tue, 2 Jul 2019 16:14:00 +0000 (09:14 -0700)]
bpf/tools: sync bpf.h

Sync new bpf_tcp_sock fields and new BPF_PROG_TYPE_SOCK_OPS RTT callback.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: add icsk_retransmits to bpf_tcp_sock
Stanislav Fomichev [Tue, 2 Jul 2019 16:13:59 +0000 (09:13 -0700)]
bpf: add icsk_retransmits to bpf_tcp_sock

Add some inet_connection_sock fields to bpf_tcp_sock that might be useful
for debugging congestion control issues.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: add dsack_dups/delivered{, _ce} to bpf_tcp_sock
Stanislav Fomichev [Tue, 2 Jul 2019 16:13:58 +0000 (09:13 -0700)]
bpf: add dsack_dups/delivered{, _ce} to bpf_tcp_sock

Add more fields to bpf_tcp_sock that might be useful for debugging
congestion control issues.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: split shared bpf_tcp_sock and bpf_sock_ops implementation
Stanislav Fomichev [Tue, 2 Jul 2019 16:13:57 +0000 (09:13 -0700)]
bpf: split shared bpf_tcp_sock and bpf_sock_ops implementation

We've added bpf_tcp_sock member to bpf_sock_ops and don't expect
any new tcp_sock fields in bpf_sock_ops. Let's remove
CONVERT_COMMON_TCP_SOCK_FIELDS so bpf_tcp_sock can be independently
extended.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: add BPF_CGROUP_SOCK_OPS callback that is executed on every RTT
Stanislav Fomichev [Tue, 2 Jul 2019 16:13:56 +0000 (09:13 -0700)]
bpf: add BPF_CGROUP_SOCK_OPS callback that is executed on every RTT

Performance impact should be minimal because it's under a new
BPF_SOCK_OPS_RTT_CB_FLAG flag that has to be explicitly enabled.

Suggested-by: Eric Dumazet <edumazet@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Priyaranjan Jha <priyarjha@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests: bpf: standardize to static __always_inline
Jiri Benc [Tue, 2 Jul 2019 18:26:51 +0000 (20:26 +0200)]
selftests: bpf: standardize to static __always_inline

The progs for bpf selftests use several different notations to force
function inlining. Standardize to what most of them use,
static __always_inline.

Suggested-by: Song Liu <liu.song.a23@gmail.com>
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: Add support for fq's EDT to HBM
brakmo [Tue, 2 Jul 2019 22:09:52 +0000 (15:09 -0700)]
bpf: Add support for fq's EDT to HBM

Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.

It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.

It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.

The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.

The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.

EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.

                         Aggr              10K  10K  10K   1MB  1MB  1MB
         Limit           rate drops  RTT  rate  P90  P99  rate  P90  P99
Test      rate  Flows    Mbps   %     us  Mbps   us   us  Mbps   ms   ms
--------  ----  -----    ---- -----  ---  ---- ---- ----  ---- ---- ----
cubic       1G    2       904  0.02  108   257  511  539   647 13.4 24.5
cubic-edt   1G    2       982  0.01  156   239  656  967   743 14.0 17.2
dctcp       1G    2       977  0.00  105   324  408  744   653 14.5 15.9
dctcp-edt   1G    2       981  0.01  142   321  417  811   660 15.7 17.0
cubic-htb   1G    2       919  0.00 1825    40 2822 4140   879  9.7  9.9

cubic     200M    2       155  0.30  220    81  532  655    74  283  450
cubic-edt 200M    2       188  0.02  222    87 1035 1095   101   84   85
dctcp     200M    2       188  0.03  111    77  912  939   111   76  325
dctcp-edt 200M    2       188  0.03  217    74 1416 1738   114   76   79
cubic-htb 200M    2       188  0.00 5015     8 14ms 15ms   180   48   50

cubic       1G    4       952  0.03  110   165  516  546   262   38  154
cubic-edt   1G    4       973  0.01  190   111 1034 1314   287   65   79
dctcp       1G    4       951  0.00  103   180  617  905   257   37   38
dctcp-edt   1G    4       967  0.00  163   151  732 1126   272   43   55
cubic-htb   1G    4       914  0.00 3249    13  7ms  8ms   300   29   34

cubic       5G    4      4236  0.00  134   305  490  624  1310   10   17
cubic-edt   5G    4      4865  0.00  156   306  425  759  1520   10   16
dctcp       5G    4      4936  0.00  128   485  221  409  1484    7    9
dctcp-edt   5G    4      4924  0.00  148   390  392  623  1508   11   26

v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed

Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf, libbpf, smatch: Fix potential NULL pointer dereference
Leo Yan [Tue, 2 Jul 2019 10:25:31 +0000 (18:25 +0800)]
bpf, libbpf, smatch: Fix potential NULL pointer dereference

Based on the following report from Smatch, fix the potential NULL
pointer dereference check:

  tools/lib/bpf/libbpf.c:3493
  bpf_prog_load_xattr() warn: variable dereferenced before check 'attr'
  (see line 3483)

  3479 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
  3480                         struct bpf_object **pobj, int *prog_fd)
  3481 {
  3482         struct bpf_object_open_attr open_attr = {
  3483                 .file           = attr->file,
  3484                 .prog_type      = attr->prog_type,
                                         ^^^^^^
  3485         };

At the head of function, it directly access 'attr' without checking
if it's NULL pointer. This patch moves the values assignment after
validating 'attr' and 'attr->file'.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: fix GCC8 warning for strncpy
Andrii Nakryiko [Tue, 2 Jul 2019 15:16:20 +0000 (08:16 -0700)]
libbpf: fix GCC8 warning for strncpy

GCC8 started emitting warning about using strncpy with number of bytes
exactly equal destination size, which is generally unsafe, as can lead
to non-zero terminated string being copied. Use IFNAMSIZ - 1 as number
of bytes to ensure name is always zero-terminated.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Cc: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Yonghong Song <yhs@fb.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: fix precision tracking
Alexei Starovoitov [Fri, 28 Jun 2019 16:24:09 +0000 (09:24 -0700)]
bpf: fix precision tracking

When equivalent state is found the current state needs to propagate precision marks.
Otherwise the verifier will prune the search incorrectly.

There is a price for correctness:
                      before      before    broken    fixed
                      cnst spill  precise   precise
bpf_lb-DLB_L3.o       1923        8128      1863      1898
bpf_lb-DLB_L4.o       3077        6707      2468      2666
bpf_lb-DUNKNOWN.o     1062        1062      544       544
bpf_lxc-DDROP_ALL.o   166729      380712    22629     36823
bpf_lxc-DUNKNOWN.o    174607      440652    28805     45325
bpf_netdev.o          8407        31904     6801      7002
bpf_overlay.o         5420        23569     4754      4858
bpf_lxc_jit.o         39389       359445    50925     69631
Overall precision tracking is still very effective.

Fixes: b5dc0163d8fd ("bpf: precise scalar_value tracking")
Reported-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Tested-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agomlxsw: spectrum_ptp: Fix validation in mlxsw_sp1_ptp_packet_finish()
Petr Machata [Tue, 2 Jul 2019 19:06:47 +0000 (19:06 +0000)]
mlxsw: spectrum_ptp: Fix validation in mlxsw_sp1_ptp_packet_finish()

Before mlxsw_sp1_ptp_packet_finish() sends the packet back, it validates
whether the corresponding port is still valid. However the condition is
incorrect: when mlxsw_sp_port == NULL, the code dereferences the port to
compare it to skb->dev.

The condition needs to check whether the port is present and skb->dev still
refers to that port (or else is NULL). If that does not hold, bail out.
Add a pair of parentheses to fix the condition.

Fixes: d92e4e6e33c8 ("mlxsw: spectrum: PTP: Support timestamping on Spectrum-1")
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: add random MAC address fallback
Heiner Kallweit [Tue, 2 Jul 2019 18:46:09 +0000 (20:46 +0200)]
r8169: add random MAC address fallback

It was reported that the GPD MicroPC is broken in a way that no valid
MAC address can be read from the network chip. The vendor driver deals
with this by assigning a random MAC address as fallback. So let's do
the same.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoRevert "r8169: improve handling VLAN tag"
Heiner Kallweit [Tue, 2 Jul 2019 05:59:17 +0000 (07:59 +0200)]
Revert "r8169: improve handling VLAN tag"

This reverts commit 759d095741721888b6ee51afa74e0a66ce65e974.

The patch was based on a misunderstanding. As Al Viro pointed out [0]
it's simply wrong on big endian. So let's revert it.

[0] https://marc.info/?t=156200975600004&r=1&w=2

Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: make "snps,reset-delays-us" optional again
Martin Blumenstingl [Mon, 1 Jul 2019 22:42:25 +0000 (00:42 +0200)]
net: stmmac: make "snps,reset-delays-us" optional again

Commit 760f1dc2958022 ("net: stmmac: add sanity check to
device_property_read_u32_array call") introduced error checking of the
device_property_read_u32_array() call in stmmac_mdio_reset().
This results in the following error when the "snps,reset-delays-us"
property is not defined in devicetree:
  invalid property snps,reset-delays-us

This sanity check made sense until commit 84ce4d0f9f55b4 ("net: stmmac:
initialize the reset delay array") ensured that there are fallback
values for the reset delay if the "snps,reset-delays-us" property is
absent. That was at the cost of making that property mandatory though.

Drop the sanity check for device_property_read_u32_array() and thus make
the "snps,reset-delays-us" property optional again (avoiding the error
message while loading the stmmac driver with a .dtb where the property
is absent).

Fixes: 760f1dc2958022 ("net: stmmac: add sanity check to device_property_read_u32_array call")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobonding/main: fix NULL dereference in bond_select_active_slave()
Eric Dumazet [Mon, 1 Jul 2019 17:48:51 +0000 (10:48 -0700)]
bonding/main: fix NULL dereference in bond_select_active_slave()

A bonding master can be up while best_slave is NULL.

[12105.636318] BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
[12105.638204] mlx4_en: eth1: Linkstate event 1 -> 1
[12105.648984] IP: bond_select_active_slave+0x125/0x250
[12105.653977] PGD 0 P4D 0
[12105.656572] Oops: 0000 [#1] SMP PTI
[12105.660487] gsmi: Log Shutdown Reason 0x03
[12105.664620] Modules linked in: kvm_intel loop act_mirred uhaul vfat fat stg_standard_ftl stg_megablocks stg_idt stg_hdi stg elephant_dev_num stg_idt_eeprom w1_therm wire i2c_mux_pca954x i2c_mux mlx4_i2c i2c_usb cdc_acm ehci_pci ehci_hcd i2c_iimc mlx4_en mlx4_ib ib_uverbs ib_core mlx4_core [last unloaded: kvm_intel]
[12105.685686] mlx4_core 0000:03:00.0: dispatching link up event for port 2
[12105.685700] mlx4_en: eth2: Linkstate event 2 -> 1
[12105.685700] mlx4_en: eth2: Link Up (linkstate)
[12105.724452] Workqueue: bond0 bond_mii_monitor
[12105.728854] RIP: 0010:bond_select_active_slave+0x125/0x250
[12105.734355] RSP: 0018:ffffaf146a81fd88 EFLAGS: 00010246
[12105.739637] RAX: 0000000000000003 RBX: ffff8c62b03c6900 RCX: 0000000000000000
[12105.746838] RDX: 0000000000000000 RSI: ffffaf146a81fd08 RDI: ffff8c62b03c6000
[12105.754054] RBP: ffffaf146a81fdb8 R08: 0000000000000001 R09: ffff8c517d387600
[12105.761299] R10: 00000000001075d9 R11: ffffffffaceba92f R12: 0000000000000000
[12105.768553] R13: ffff8c8240ae4800 R14: 0000000000000000 R15: 0000000000000000
[12105.775748] FS:  0000000000000000(0000) GS:ffff8c62bfa40000(0000) knlGS:0000000000000000
[12105.783892] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[12105.789716] CR2: 0000000000000000 CR3: 0000000d0520e001 CR4: 00000000001626f0
[12105.796976] Call Trace:
[12105.799446]  [<ffffffffac31d387>] bond_mii_monitor+0x497/0x6f0
[12105.805317]  [<ffffffffabd42643>] process_one_work+0x143/0x370
[12105.811225]  [<ffffffffabd42c7a>] worker_thread+0x4a/0x360
[12105.816761]  [<ffffffffabd48bc5>] kthread+0x105/0x140
[12105.821865]  [<ffffffffabd42c30>] ? rescuer_thread+0x380/0x380
[12105.827757]  [<ffffffffabd48ac0>] ? kthread_associate_blkcg+0xc0/0xc0
[12105.834266]  [<ffffffffac600241>] ret_from_fork+0x51/0x60

Fixes: e2a7420df2e0 ("bonding/main: convert to using slave printk macros")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: John Sperbeck <jsperbeck@google.com>
Cc: Jarod Wilson <jarod@redhat.com>
CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotipc: remove ub->ubsock checks
Xin Long [Mon, 1 Jul 2019 16:57:19 +0000 (00:57 +0800)]
tipc: remove ub->ubsock checks

Both tipc_udp_enable and tipc_udp_disable are called under rtnl_lock,
ub->ubsock could never be NULL in tipc_udp_disable and cleanup_bearer,
so remove the check.

Also remove the one in tipc_udp_enable by adding "free" label.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: Fix off-by-one in route dump counter without netlink strict checking
Stefano Brivio [Sat, 29 Jun 2019 17:55:08 +0000 (19:55 +0200)]
ipv4: Fix off-by-one in route dump counter without netlink strict checking

In commit ee28906fd7a1 ("ipv4: Dump route exceptions if requested") I
added a counter of per-node dumped routes (including actual routes and
exceptions), analogous to the existing counter for dumped nodes. Dumping
exceptions means we need to also keep track of how many routes are dumped
for each node: this would be just one route per node, without exceptions.

When netlink strict checking is not enabled, we dump both routes and
exceptions at the same time: the RTM_F_CLONED flag is not used as a
filter. In this case, the per-node counter 'i_fa' is incremented by one
to track the single dumped route, then also incremented by one for each
exception dumped, and then stored as netlink callback argument as skip
counter, 's_fa', to be used when a partial dump operation restarts.

The per-node counter needs to be increased by one also when we skip a
route (exception) due to a previous non-zero skip counter, because it
needs to match the existing skip counter, if we are dumping both routes
and exceptions. I missed this, and only incremented the counter, for
regular routes, if the previous skip counter was zero. This means that,
in case of a mixed dump, partial dump operations after the first one
will start with a mismatching skip counter value, one less than expected.

This means in turn that the first exception for a given node is skipped
every time a partial dump operation restarts, if netlink strict checking
is not enabled (iproute < 5.0).

It turns out I didn't repeat the test in its final version, commit
de755a85130e ("selftests: pmtu: Introduce list_flush_ipv4_exception test
case"), which also counts the number of route exceptions returned, with
iproute2 versions < 5.0 -- I was instead using the equivalent of the IPv6
test as it was before commit b964641e9925 ("selftests: pmtu: Make
list_flush_ipv6_exception test more demanding").

Always increment the per-node counter by one if we previously dumped
a regular route, so that it matches the current skip counter.

Fixes: ee28906fd7a1 ("ipv4: Dump route exceptions if requested")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: mediatek: Allow non TRGMII mode with MT7621 DDR2 devices
René van Dorst [Sat, 29 Jun 2019 12:24:51 +0000 (14:24 +0200)]
net: ethernet: mediatek: Allow non TRGMII mode with MT7621 DDR2 devices

No reason to error out on a MT7621 device with DDR2 memory when non
TRGMII mode is selected.
Only MT7621 DDR2 clock setup is not supported for TRGMII mode.
But non TRGMII mode doesn't need any special clock setup.

Signed-off-by: René van Dorst <opensource@vdorst.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agorxrpc: Fix uninitialized error code in rxrpc_send_data_packet()
David Howells [Tue, 2 Jul 2019 14:55:28 +0000 (15:55 +0100)]
rxrpc: Fix uninitialized error code in rxrpc_send_data_packet()

With gcc 4.1:

    net/rxrpc/output.c: In function ‘rxrpc_send_data_packet’:
    net/rxrpc/output.c:338: warning: ‘ret’ may be used uninitialized in this function

Indeed, if the first jump to the send_fragmentable label is made, and
the address family is not handled in the switch() statement, ret will be
used uninitialized.

Fix this by BUG()'ing as is done in other places in rxrpc where internal
support for future address families will need adding.  It should not be
possible to reach this normally as the address families are checked
up-front.

Fixes: 5a924b8951f835b5 ("rxrpc: Don't store the rxrpc header in the Tx queue sk_buffs")
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfc: st-nci: remove redundant assignment to variable r
Colin Ian King [Tue, 2 Jul 2019 13:16:42 +0000 (14:16 +0100)]
nfc: st-nci: remove redundant assignment to variable r

The variable r is being initialized with a value that is never
read and it is being updated later with a new value. The
initialization is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohinic: remove standard netdev stats
Xue Chaojing [Mon, 1 Jul 2019 23:40:00 +0000 (23:40 +0000)]
hinic: remove standard netdev stats

This patch removes standard netdev stats in ethtool -S.

Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Xue Chaojing <xuechaojing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Re-word Kconfig entry
Jose Abreu [Tue, 2 Jul 2019 09:12:10 +0000 (11:12 +0200)]
net: stmmac: Re-word Kconfig entry

We support many speeds and it doesn't make much sense to list them all
in the Kconfig. Let's just call it Multi-Gigabit.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'Add-gve-driver'
David S. Miller [Tue, 2 Jul 2019 02:36:35 +0000 (19:36 -0700)]
Merge branch 'Add-gve-driver'

Catherine Sullivan says:

====================
Add gve driver

This patch series adds the gve driver which will support the
Compute Engine Virtual NIC that will be available in the future.

v2:
- Patch 1:
  - Remove gve_size_assert.h and use static_assert instead.
  - Loop forever instead of bugging if the device won't reset
  - Use module_pci_driver
- Patch 2:
  - Use be16_to_cpu in the RX Seq No define
  - Remove unneeded ndo_change_mtu
- Patch 3:
  - No Changes
- Patch 4:
  - Instead of checking netif_carrier_ok in ethtool stats, just make sure

v3:
- Patch 1:
  - Remove X86 dep
- Patch 2:
  - No changes
- Patch 3:
  - No changes
- Patch 4:
  - Remove unneeded memsets in ethtool stats

v4:
- Patch 1:
  - Use io[read|write]32be instead of [read|write]l(cpu_to_be32())
  - Explicitly add padding to gve_adminq_set_driver_parameter
  - Use static where appropriate
- Patch 2:
  - Use u64_stats_sync
  - Explicity add padding to gve_adminq_create_rx_queue
  - Fix some enianness typing issues found by kbuild
  - Use static where appropriate
  - Remove unused variables
- Patch 3:
  - Use io[read|write]32be instead of [read|write]l(cpu_to_be32())
- Patch 4:
  - Use u64_stats_sync
  - Use static where appropriate
Warnings reported by:
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: Add ethtool support
Catherine Sullivan [Mon, 1 Jul 2019 22:57:55 +0000 (15:57 -0700)]
gve: Add ethtool support

Add support for the following ethtool commands:

ethtool -s|--change devname [msglvl N] [msglevel type on|off]
ethtool -S|--statistics devname
ethtool -i|--driver devname
ethtool -l|--show-channels devname
ethtool -L|--set-channels devname
ethtool -g|--show-ring devname
ethtool --reset devname

Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: Sagi Shahar <sagis@google.com>
Signed-off-by: Jon Olson <jonolson@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: Add workqueue and reset support
Catherine Sullivan [Mon, 1 Jul 2019 22:57:54 +0000 (15:57 -0700)]
gve: Add workqueue and reset support

Add support for the workqueue to handle management interrupts and
support for resets.

Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: Sagi Shahar <sagis@google.com>
Signed-off-by: Jon Olson <jonolson@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: Add transmit and receive support
Catherine Sullivan [Mon, 1 Jul 2019 22:57:53 +0000 (15:57 -0700)]
gve: Add transmit and receive support

Add support for passing traffic.

Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: Sagi Shahar <sagis@google.com>
Signed-off-by: Jon Olson <jonolson@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogve: Add basic driver framework for Compute Engine Virtual NIC
Catherine Sullivan [Mon, 1 Jul 2019 22:57:52 +0000 (15:57 -0700)]
gve: Add basic driver framework for Compute Engine Virtual NIC

Add a driver framework for the Compute Engine Virtual NIC that will be
available in the future.

At this point the only functionality is loading the driver.

Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: Sagi Shahar <sagis@google.com>
Signed-off-by: Jon Olson <jonolson@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'blackhole-device-to-invalidate-dst'
David S. Miller [Tue, 2 Jul 2019 02:34:46 +0000 (19:34 -0700)]
Merge branch 'blackhole-device-to-invalidate-dst'

Mahesh Bandewar says:

====================
blackhole device to invalidate dst

When we invalidate dst or mark it "dead", we assign 'lo' to
dst->dev. First of all this assignment is racy and more over,
it has MTU implications.

The standard dev MTU is 1500 while the Loopback MTU is 64k. TCP
code when dereferencing the dst don't check if the dst is valid
or not. TCP when dereferencing a dead-dst while negotiating a
new connection, may use dst device which is 'lo' instead of
using the correct device. Consider the following scenario:

A SYN arrives on an interface and tcp-layer while processing
SYNACK finds a dst and associates it with SYNACK skb. Now before
skb gets passed to L3 for processing, if that dst gets "dead"
(because of the virtual device getting disappeared & then reappeared),
the 'lo' gets assigned to that dst (lo MTU = 64k). Let's assume
the SYN has ADV_MSS set as 9k while the output device through
which this SYNACK is going to go out has standard MTU of 1500.
The MTU check during the route check passes since MIN(9K, 64K)
is 9k and TCP successfully negotiates 9k MSS. The subsequent
data packet; bigger in size gets passed to the device and it
won't be marked as GSO since the assumed MTU of the device is
9k.

This either crashes the NIC and we have seen fixes that went
into drivers to handle this scenario. 8914a595110a ('bnx2x:
disable GSO where gso_size is too big for hardware') and
2b16f048729b ('net: create skb_gso_validate_mac_len()') and
with those fixes TCP eventually recovers but not before
few dropped segments.

Well, I'm not a TCP expert and though we have experienced
these corner cases in our environment, I could not reproduce
this case reliably in my test setup to try this fix myself.
However, Michael Chan <michael.chan@broadcom.com> had a setup
where these fixes helped him mitigate the issue and not cause
the crash.

The idea here is to not alter the data-path with additional
locks or smb()/rmb() barriers to avoid racy assignments but
to create a new device that has really low MTU that has
.ndo_start_xmit essentially a kfree_skb(). Make use of this
device instead of 'lo' when marking the dst dead.

First patch implements the blackhole device and second
patch uses it in IPv4 and IPv6 stack while the third patch
is the self test that ensures the sanity of this device.

v1->v2
  fixed the self-test patch to handle the conflict

v2 -> v3
  fixed Kconfig text/string.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoblackhole_dev: add a selftest
Mahesh Bandewar [Mon, 1 Jul 2019 21:39:01 +0000 (14:39 -0700)]
blackhole_dev: add a selftest

Since this is not really a device with all capabilities, this test
ensures that it has *enough* to make it through the data path
without causing unwanted side-effects (read crash!).

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoblackhole_netdev: use blackhole_netdev to invalidate dst entries
Mahesh Bandewar [Mon, 1 Jul 2019 21:38:57 +0000 (14:38 -0700)]
blackhole_netdev: use blackhole_netdev to invalidate dst entries

Use blackhole_netdev instead of 'lo' device with lower MTU when marking
dst "dead".

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Tested-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoloopback: create blackhole net device similar to loopack.
Mahesh Bandewar [Mon, 1 Jul 2019 21:38:49 +0000 (14:38 -0700)]
loopback: create blackhole net device similar to loopack.

Create a blackhole net device that can be used for "dead"
dst entries instead of loopback device. This blackhole device differs
from loopback in few aspects: (a) It's not per-ns. (b)  MTU on this
device is ETH_MIN_MTU (c) The xmit function is essentially kfree_skb().
and (d) since it's not registered it won't have ifindex.

Lower MTU effectively make the device not pass the MTU check during
the route check when a dst associated with the skb is dead.

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: broadcom: bcm63xx_enet: Remove unneeded memset
Hariprasad Kelam [Sun, 30 Jun 2019 14:29:49 +0000 (19:59 +0530)]
net: ethernet: broadcom: bcm63xx_enet: Remove unneeded memset

Remove unneeded memset as alloc_etherdev is using kvzalloc which uses
__GFP_ZERO flag

Signed-off-by: Hariprasad Kelam <hariprasad.kelam@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-netsec-Add-XDP-Support'
David S. Miller [Tue, 2 Jul 2019 02:27:08 +0000 (19:27 -0700)]
Merge branch 'net-netsec-Add-XDP-Support'

Ilias Apalodimas says:

====================
net: netsec: Add XDP Support

This is a respin of https://www.spinics.net/lists/netdev/msg526066.html
Since page_pool API fixes are merged into net-next we can now safely use
it's DMA mapping capabilities.

First patch changes the buffer allocation from napi/netdev_alloc_frag()
to page_pool API. Although this will lead to slightly reduced performance
(on raw packet drops only) we can use the API for XDP buffer recycling.
Another side effect is a slight increase in memory usage, due to using a
single page per packet.

The second patch adds XDP support on the driver.
There's a bunch of interesting options that come up due to the single
Tx queue.
Locking is needed(to avoid messing up the Tx queues since ndo_xdp_xmit
and the normal stack can co-exist). We also need to track down the
'buffer type' for TX and properly free or recycle the packet depending
on it's nature.

Changes since RFC:
- Bug fixes from Jesper and Maciej
- Added page pool API to retrieve the DMA direction

Changes since v1:
- Use page_pool_free correctly if xdp_rxq_info_reg() failed
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: netsec: add XDP support
Ilias Apalodimas [Sat, 29 Jun 2019 05:23:25 +0000 (08:23 +0300)]
net: netsec: add XDP support

The interface only supports 1 Tx queue so locking is introduced on
the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
.ndo_xdp_xmit won't corrupt Tx ring

- Performance (SMMU off)

Benchmark   XDP_SKB     XDP_DRV
xdp1        291kpps     344kpps
rxdrop      282kpps     342kpps

- Performance (SMMU on)
Benchmark   XDP_SKB     XDP_DRV
xdp1        167kpps     324kpps
rxdrop      164kpps     323kpps

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: page_pool: add helper function for retrieving dma direction
Ilias Apalodimas [Sat, 29 Jun 2019 05:23:24 +0000 (08:23 +0300)]
net: page_pool: add helper function for retrieving dma direction

Since the dma direction is stored in page pool params, offer an API
helper for driver that choose not to keep track of it locally

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: netsec: Use page_pool API
Ilias Apalodimas [Sat, 29 Jun 2019 05:23:23 +0000 (08:23 +0300)]
net: netsec: Use page_pool API

Use page_pool and it's DMA mapping capabilities for Rx buffers instead
of netdev/napi_alloc_frag()

Although this will result in a slight performance penalty on small sized
packets (~10%) the use of the API will allow to easily add XDP support.
The penalty won't be visible in network testing i.e ipef/netperf etc, it
only happens during raw packet drops.
Furthermore we intend to add recycling capabilities on the API
in the future. Once the recycling is added the performance penalty will
go away.
The only 'real' penalty is the slightly increased memory usage, since we
now allocate a page per packet instead of the amount of bytes we need +
skb metadata (difference is roughly 2kb per packet).
With a minimum of 4BG of RAM on the only SoC that has this NIC the
extra memory usage is negligible (a bit more on 64K pages)

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotc-testing: added tdc tests for prio qdisc
Roman Mashak [Fri, 28 Jun 2019 21:32:01 +0000 (17:32 -0400)]
tc-testing: added tdc tests for prio qdisc

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mirred-batch-fixes'
David S. Miller [Tue, 2 Jul 2019 02:18:04 +0000 (19:18 -0700)]
Merge branch 'mirred-batch-fixes'

Roman Mashak says:

====================
Fix batched event generation for mirred action

When adding or deleting a batch of entries, the kernel sends upto
TCA_ACT_MAX_PRIO entries in an event to user space. However it does not
consider that the action sizes may vary and require different skb sizes.

For example :

% cat tc-batch.sh
TC="sudo /mnt/iproute2.git/tc/tc"

$TC actions flush action mirred
for i in `seq 1 $1`;
do
   cmd="action mirred egress redirect dev lo index $i "
   args=$args$cmd
done
$TC actions add $args
%
% ./tc-batch.sh 32
Error: Failed to fill netlink attributes while adding TC action.
We have an error talking to the kernel
%

patch 1 adds callback in tc_action_ops of mirred action, which calculates
the action size, and passes size to tcf_add_notify()/tcf_del_notify().

patch 2 updates the TDC test suite with relevant test cases.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotc-testing: updated mirred action tests with batch create/delete
Roman Mashak [Fri, 28 Jun 2019 18:30:18 +0000 (14:30 -0400)]
tc-testing: updated mirred action tests with batch create/delete

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet sched: update mirred action for batched events operations
Roman Mashak [Fri, 28 Jun 2019 18:30:17 +0000 (14:30 -0400)]
net sched: update mirred action for batched events operations

Add get_fill_size() routine used to calculate the action size
when building a batch of events.

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonetlink: use 48 byte ctx instead of 6 signed longs for callback
Jason A. Donenfeld [Fri, 28 Jun 2019 14:40:21 +0000 (16:40 +0200)]
netlink: use 48 byte ctx instead of 6 signed longs for callback

People are inclined to stuff random things into cb->args[n] because it
looks like an array of integers. Sometimes people even put u64s in there
with comments noting that a certain member takes up two slots. The
horror! Really this should mirror the usage of skb->cb, which are just
48 opaque bytes suitable for casting a struct. Then people can create
their usual casting macros for accessing strongly typed members of a
struct.

As a plus, this also gives us the same amount of space on 32bit and 64bit.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotipc: embed jiffies in macro TIPC_BC_RETR_LIM
Jon Maloy [Fri, 28 Jun 2019 15:06:20 +0000 (17:06 +0200)]
tipc: embed jiffies in macro TIPC_BC_RETR_LIM

The macro TIPC_BC_RETR_LIM is always used in combination with 'jiffies',
so we can just as well perform the addition in the macro itself. This
way, we get a few shorter code lines and one less line break.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/ipv6: Fix misuse of proc_dointvec "flowlabel_reflect"
Eiichi Tsukata [Fri, 28 Jun 2019 02:37:14 +0000 (11:37 +0900)]
net/ipv6: Fix misuse of proc_dointvec "flowlabel_reflect"

/proc/sys/net/ipv6/flowlabel_reflect assumes written value to be in the
range of 0 to 3. Use proc_dointvec_minmax instead of proc_dointvec.

Fixes: 323a53c41292 ("ipv6: tcp: enable flowlabel reflection in some RST packets")
Signed-off-by: Eiichi Tsukata <devel@etsukata.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: link_watch: prevent starvation when processing linkwatch wq
Yunsheng Lin [Fri, 28 Jun 2019 01:13:19 +0000 (09:13 +0800)]
net: link_watch: prevent starvation when processing linkwatch wq

When user has configured a large number of virtual netdev, such
as 4K vlans, the carrier on/off operation of the real netdev
will also cause it's virtual netdev's link state to be processed
in linkwatch. Currently, the processing is done in a work queue,
which may cause rtnl locking starvation problem and worker
starvation problem for other work queue, such as irqfd_inject wq.

This patch releases the cpu when link watch worker has processed
a fixed number of netdev' link watch event, and schedule the
work queue again when there is still link watch event remaining.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mlxsw-PTP-timestamping-support'
David S. Miller [Tue, 2 Jul 2019 01:58:35 +0000 (18:58 -0700)]
Merge branch 'mlxsw-PTP-timestamping-support'

Ido Schimmel says:

====================
mlxsw: PTP timestamping support

This is the second patchset adding PTP support in mlxsw. Next patchset
will add PTP shapers which are required to maintain accuracy under rates
lower than 40Gb/s, while subsequent patchsets will add tracepoints and
selftests.

Petr says:

This patch set introduces support for retrieving and processing hardware
timestamps for PTP packets.

The way PTP timestamping works on Spectrum-1 is that there are two queues
associated with each front panel port. When a packet is timestamped, the
timestamp is put to one of the queues: timestamps for transmitted packets
to one and for received packets to the other. Activity on these queues is
signaled through the events PTP_ING_FIFO and PTP_EGR_FIFO.

Packets themselves arrive through two traps: PTP0 and PTP1. It is possible
to configure which PTP messages should be trapped under which PTP trap. On
Spectrum systems, mlxsw will use PTP0 for event messages (which need
timestamping), and PTP1 for general messages (which do not).

There are therefore four relevant traps: receive of PTP event resp. general
message, and receive of timestamp for a transmitted resp. received PTP
packet. The obvious point where to put the new logic is a custom listener
to the mentioned traps.

Besides handling ingress traffic (be in packets or timestamps), the driver
also needs to handle timestamping of transmitted packets. One option would
be to invoke the relevant logic from mlxsw_core_ptp_transmitted(). However
on Spectrum-2, the timestamps are actually delivered through the completion
queue, and for that reason this patchset opts to invoke the logic from the
PCI code, via core and the driver, to a chip-specific operation. That way
the invocation will be done in a place where a Spectrum-2 implementation
will have an opportunity to extract the timestamp.

As indicated above, the PTP FIFO signaling happens independently from
packet delivery. A packet corresponding to any given timestamp could be
delivered sooner or later than the timestamp itself. Additionally, the
queues are only four elements deep, and it is therefore possible that the
timestamp for a delivered packet never arrives at all. Similarly a PTP
packet might be dropped due to CPU traffic pressure, and never be delivered
even if the corresponding timestamp was.

The driver thus needs to hold a cache of as-yet-unmatched SKBs and
timestamps. The first piece to arrive (be it timestamp or SKB) is put to
this cache. When the other piece arrives, the timestamp is attached to the
SKB and that is passed on. A delayed work is run at regular intervals to
prune the old unmatched entries.

As mentioned above, the mechanism for timestamp delivery changes on
Spectrum-2, where timestamps are part of completion queue elements, and all
packets are timestamped. All this bookkeeping is therefore unnecessary on
Spectrum-2. For this reason, this patchset spends some time introducing
Spectrum-1 specific artifacts such as a possibility to register a given
trap only on Spectrum-1.

Patches #1-#4 describe new registers.

Patches #5 and #6 introduce the possibility to register certain traps
only on some systems. The list of Spectrum-1 specific traps is left empty
at this point.

Patch #7 hooks into packet receive path by registering PTP traps
and appropriate handlers (that however do nothing of substance yet).

Patch #8 adds a helper to allow storing custom data to SKB->cb.

Patch #9 adds a call into the PCI completion queue handler that invokes,
via core and spectrum code, a PTP transmit handler. (Which also does not do
anything interesting yet.)

Patch #10 introduces code to invoke PTP initialization and adds data types
for the cache of unmatched entries.

Patches #11 and #12 implement the timestamping itself. In #11, the PHC
spin_locks are converted to _bh variants, because unlike normal PHC path,
which runs in process context, timestamp processing runs as soft interrupt.
Then #12 introduces the code for saving and retrieval of unmatched entries,
invokes PTP classifier to identify packets of interest, registers timestamp
FIFO events, and handles decoding and attaching timestamps to packets.

Patch #13 introduces a garbage collector for left-behind entries that have
not been matched for about a second.

In patch #14, PTP message types are configured to arrive as PTP0
(events) or PTP1 (everything else) as appropriate. At this point, the PTP
packets start arriving through the traps, but because PTP is disabled and
there is no way to enable it yet, they are always just passed to the usual
receive path right away.

Finally patches #15 and #16 add the plumbing to actually make it possible
to enable this code through SIOCSHWTSTAMP ioctl, and to advertise the
hardware timestamping capabilities through ethtool.

v2:
- Patch #12:
    - In mlxsw_sp1_ptp_fifo_event_func(), post-increment when iterating over PTP
      FIFO records.
- Patch #14:
    - Change namespace of message type enumerators from MLXSW_ to MLXSW_SP_.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Support ethtool get_ts_info
Petr Machata [Sun, 30 Jun 2019 06:05:00 +0000 (09:05 +0300)]
mlxsw: spectrum: PTP: Support ethtool get_ts_info

The get_ts_info callback is used for obtaining information about
timestamping capabilities of a network device. On Spectrum-1, implement
it to advertise the PHC and the capability to do HW timestamping, and
the supported RX and TX filters.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Support SIOCGHWTSTAMP, SIOCSHWTSTAMP ioctls
Petr Machata [Sun, 30 Jun 2019 06:04:59 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Support SIOCGHWTSTAMP, SIOCSHWTSTAMP ioctls

The SIOCSHWTSTAMP ioctl configures HW timestamping on a given port.
Dispatch the ioctls to per-chip handler (which add to ptp_ops). Find
which PTP messages need to be timestamped and configure MTPPPC
accordingly.

The SIOCGHWTSTAMP ioctl is getter for the current configuration.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Configure PTP traps and FIFO events
Petr Machata [Sun, 30 Jun 2019 06:04:58 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Configure PTP traps and FIFO events

Configure MTPTPT to set which message types should arrive under which
PTP trap, and MOGCR to clear the timestamp queue after its contents are
reported through PTP_ING_FIFO or PTP_EGR_FIFO.

With this configuration, PTP packets start arriving through the PTP
traps. However since timestamping is disabled by default and there is
currently no way to enable it, they will not be timestamped.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Garbage-collect unmatched entries
Petr Machata [Sun, 30 Jun 2019 06:04:57 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Garbage-collect unmatched entries

On Spectrum-1, timestamped PTP packets and the corresponding timestamps
need to be kept in caches until both are available, at which point they are
matched up and packets forwarded as appropriate. However, not all packets
will ever see their timestamp, and not all timestamps will ever see their
packet. It is therefore necessary to dispose of such abandoned entries.

To that end, introduce a garbage collector to collect entries that have
not had their counterpart turn up within about a second. The GC
maintains a monotonously-increasing value of GC cycle. Every entry that
is put to the hash table is annotated with the GC cycle at which it
should be collected. When the GC runs, it walks the hash table, and
collects the objects according to their GC cycle annotation.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Support timestamping on Spectrum-1
Petr Machata [Sun, 30 Jun 2019 06:04:56 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Support timestamping on Spectrum-1

On Spectrum-1, timestamps arrive through a pair of dedicated events:
MLXSW_TRAP_ID_PTP_ING_FIFO and _EGR_FIFO. The payload delivered with
those traps is contents of the timestamp FIFO at a given port in a given
direction. Add a Spectrum-1-specific handler for these two events which
decodes the timestamps and forwards them to the PTP module.

Add a function that parses a packet, dispatching to ptp_classify_raw(),
and decodes PTP message type, domain number, and sequence ID. Add a new
mlxsw dependency on the PTP classifier.

Add helpers that can store and retrieve unmatched timestamps and SKBs to
the hash table added in a preceding patch.

Add the matching code itself: upon arrival of a timestamp or a packet,
look up the corresponding unmatched entry, and match it up. If there is
none, add a new unmatched entry. This logic is the same on ingress as on
egress.

Packets and timestamps that never matched need to be eventually disposed
of. A garbage collector added in a follow-up patch will take care of
that. Since currently all this code is turned off, no crud will
accumulate in the hash table.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Disable BH when working with PHC
Petr Machata [Sun, 30 Jun 2019 06:04:55 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Disable BH when working with PHC

Up until now, the PTP hardware clock code was only invoked in the process
context (SYS_clock_adjtime -> do_clock_adjtime -> k_clock::clock_adj ->
pc_clock_adjtime -> posix_clock_operations::clock_adjtime ->
ptp_clock_info::adjtime -> mlxsw_spectrum).

In order to enable HW timestamping, which is tied into trap handling, it
will be necessary to take the clock lock from the PCI queue handler
tasklets as well.

Therefore use the _bh variants when handling the clock lock. Incidentally,
Documentation/ptp/ptp.txt recommends _irqsave variants, but that's
unnecessarily strong for our needs.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Add PTP initialization / finalization
Petr Machata [Sun, 30 Jun 2019 06:04:54 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Add PTP initialization / finalization

Add two ptp_ops: init and fini, to initialize and finalize the PTP
subsystem. Call as appropriate from mlxsw_sp_init() and _fini().

Lay the groundwork for Spectrum-1 support. On Spectrum-1, the received
timestamped packets and their corresponding timestamps arrive
independently, and need to be matched up. Introduce the related data types
and add to struct mlxsw_sp_ptp_state the hash table that will keep the
unmatched entries.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: pci: PTP: Hook into packet transmit path
Petr Machata [Sun, 30 Jun 2019 06:04:53 +0000 (09:04 +0300)]
mlxsw: pci: PTP: Hook into packet transmit path

On Spectrum-1, timestamps are delivered separately from the packets, and
need to paired up. Therefore, at some point after mlxsw_sp_port_xmit()
is invoked, it is necessary to involve the chip-specific driver code to
allow it to do the necessary bookkeeping and matching.

On Spectrum-2, timestamps are delivered in CQE. For that reason,
position the point of driver involvement into mlxsw_pci_cqe_sdq_handle()
to make it hopefully easier to extend for Spectrum-2 in the future.

To tell the driver what port the packet was sent on, keep tx_info
in SKB control buffer.

Introduce a new driver core interface mlxsw_core_ptp_transmitted(), a
driver callback ptp_transmitted, and a PTP op transmitted. The callee is
responsible for taking care of releasing the SKB passed to the new
interfaces, and correspondingly have the new stub callbacks just call
dev_kfree_skb_any().

Follow-up patches will introduce the actual content into
mlxsw_sp1_ptp_transmitted() in particular.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: core: Add support for using SKB control buffer
Petr Machata [Sun, 30 Jun 2019 06:04:52 +0000 (09:04 +0300)]
mlxsw: core: Add support for using SKB control buffer

The SKB control buffer is useful (and used) for bookkeeping of information
related to that SKB. Add helpers so that the mlxsw driver(s) can safely use
the buffer as well. The structure is currently empty, individual users will
add members to it as necessary.

Note that SKB allocation functions already clear the buffer, so the cleanup
is only necessary when ndo_start_xmit is called.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: PTP: Hook into packet receive path
Petr Machata [Sun, 30 Jun 2019 06:04:51 +0000 (09:04 +0300)]
mlxsw: spectrum: PTP: Hook into packet receive path

When configured, the Spectrum hardware can recognize PTP packets and
trap them to the CPU using dedicated traps, PTP0 and PTP1.

One reason to get PTP packets under dedicated traps is to have a
separate policer suitable for the amount of PTP traffic expected when
switch is operated as a boundary clock. For this, add two new trap
groups, MLXSW_REG_HTGT_TRAP_GROUP_SP_PTP0 and _PTP1, and associate the
two PTP traps with these two groups.

In the driver, specifically for Spectrum-1, event PTP packets will need
to be paired up with their timestamps. Those arrive through a different
set of traps, added later in the patch set. To support this future use,
introduce a new PTP op, ptp_receive.

It is possible to configure which PTP messages should be trapped under
which PTP trap. On Spectrum systems, we will use PTP0 for event
packets (which need timestamping), and PTP1 for control packets (which
do not). Thus configure PTP0 trap with a custom callback that defers to
the ptp_receive op.

Additionally, L2 PTP packets are actually trapped through the LLDP trap,
not through any of the PTP traps. So treat the LLDP trap the same way as
the PTP0 trap. Unlike PTP traps, which are currently still disabled,
LLDP trap is active. Correspondingly, have all the implementations of
the ptp_receive op return true, which the handler treats as a signal to
forward the packet immediately.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: Add support for traps specific to Spectrum-1
Petr Machata [Sun, 30 Jun 2019 06:04:50 +0000 (09:04 +0300)]
mlxsw: spectrum: Add support for traps specific to Spectrum-1

On Spectrum-1, timestamps for PTP packets are delivered through queues
of ingress and egress timestamps. There are two event traps
corresponding to activity on each of those queues. This mechanism is
absent on Spectrum-2, and therefore the traps should only be registered
on Spectrum-1.

Carry a chip-specific listener array in mlxsw_sp->listeners and
listeners_count. Register listeners from that array in
mlxsw_sp_traps_init(). Add a new listener array for Spectrum-1 traps and
configure the newly-added mlxsw_sp->listeners with this array.

The listener array is empty for now, the events will be added in a later
patch.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: Extract a helper for trap registration
Petr Machata [Sun, 30 Jun 2019 06:04:49 +0000 (09:04 +0300)]
mlxsw: spectrum: Extract a helper for trap registration

On Spectrum-1, timestamps for PTP packets are delivered through queues
of ingress and egress timestamps. There are two event traps
corresponding to activity on each of those queues. This mechanism is
absent on Spectrum-2, and therefore the traps should only be registered
on Spectrum-1.

Extract out of mlxsw_sp_traps_init() a generic helper,
mlxsw_sp_traps_register(), and likewise with _unregister(). The new helpers
will later be called with Spectrum-1-specific traps.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: reg: Add Monitoring Global Configuration Register
Petr Machata [Sun, 30 Jun 2019 06:04:48 +0000 (09:04 +0300)]
mlxsw: reg: Add Monitoring Global Configuration Register

This register serves to configure global parameters of certain
monitoring operations. The following patches will use it to configure
that when PTP timestamps are delivered through the PTP FIFO traps, the
FIFO in question is cleared as well.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: reg: Add Time Precision Packet Timestamping Reading
Petr Machata [Sun, 30 Jun 2019 06:04:47 +0000 (09:04 +0300)]
mlxsw: reg: Add Time Precision Packet Timestamping Reading

The MTPPTR is used for reading the per port PTP timestamp FIFO.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: reg: Add Monitoring Precision Time Protocol Trap Register
Petr Machata [Sun, 30 Jun 2019 06:04:46 +0000 (09:04 +0300)]
mlxsw: reg: Add Monitoring Precision Time Protocol Trap Register

This register is used for configuring under which trap to deliver PTP
packets depending on type of the packet.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: reg: Add Monitoring Time Precision Packet Port Configuration Register
Petr Machata [Sun, 30 Jun 2019 06:04:45 +0000 (09:04 +0300)]
mlxsw: reg: Add Monitoring Time Precision Packet Port Configuration Register

This register serves for configuration of which PTP messages should be
timestamped. This is a global configuration, despite the register name.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosamples: pktgen: allow to specify destination port
Daniel T. Lee [Sat, 29 Jun 2019 13:33:58 +0000 (22:33 +0900)]
samples: pktgen: allow to specify destination port

Currently, kernel pktgen has the feature to specify udp destination port
for sending packet. (e.g. pgset "udp_dst_min 9")

But on samples, each of the scripts doesn't have any option to achieve this.

This commit adds the DST_PORT option to specify the target port(s) in the script.

    -p : ($DST_PORT)  destination PORT range (e.g. 433-444) is also allowed

Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosamples: pktgen: add some helper functions for port parsing
Daniel T. Lee [Sat, 29 Jun 2019 13:33:57 +0000 (22:33 +0900)]
samples: pktgen: add some helper functions for port parsing

This commit adds port parsing and port validate helper function to parse
single or range of port(s) from a given string. (e.g. 1234, 443-444)

Helpers will be used in prior to set target port(s) in samples/pktgen.

Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: icmp: allow flowlabel reflection in echo replies
Eric Dumazet [Mon, 1 Jul 2019 13:39:36 +0000 (06:39 -0700)]
ipv6: icmp: allow flowlabel reflection in echo replies

Extend flowlabel_reflect bitmask to allow conditional
reflection of incoming flowlabels in echo replies.

Note this has precedence against auto flowlabels.

Add flowlabel_reflect enum to replace hard coded
values.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'mlx5e-updates-2019-06-28' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Mon, 1 Jul 2019 01:41:13 +0000 (18:41 -0700)]
Merge tag 'mlx5e-updates-2019-06-28' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5e-updates-2019-06-28

This series adds some misc updates for mlx5e driver

1) Allow adding the same mac more than once in MPFS table
2) Move to HW checksumming advertising
3) Report netdevice MPLS features
4) Correct physical port name of the PF representor
5) Reduce stack usage in mlx5_eswitch_termtbl_create
6) Refresh TIR improvement for representors
7) Expose same physical switch_id for all representors
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 30 Jun 2019 23:03:35 +0000 (16:03 -0700)]
Merge branch '10GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2019-06-28

This series contains a smorgasbord of updates to many of the Intel
drivers.

Gustavo A. R. Silva updates the ice and iavf drivers to use the
strcut_size() helper where possible.

Miguel increases the pause and refresh time for flow control in the
e1000e driver during reset for certain devices.

Dann Frazier fixes a potential NULL pointer dereference in ixgbe driver
when using non-IPSec enabled devices.

Colin Ian King fixes a potential overflow during a shift in the ixgbe
driver.  Also fixes a potential NULL pointer dereference in the iavf
driver by adding a check.

Venkatesh Srinivas converts the e1000 driver to use dma_wmb() instead of
wmb() for doorbell writes to avoid SFENCEs in the transmit and receive
paths.

Arjan updates the e1000e driver to improve boot time by over 100 msec by
reducing the usleep ranges suring system startup.

Artem updates the igb driver register dump in ethtool, first prepares
the register dump for future additions of registers in the dump, then
secondly, adds the RR2DCDELAY register to the dump.  When dealing with
time-sensitive networks, this register is helpful in determining your
latency from the device to the ring.

Alex fixes the ixgbevf driver to use the current cached link state,
rather than trying to re-check the value from the PF.

Harshitha adds support for MACVLAN offloads in i40e by using channels as
MACVLAN interfaces.

Detlev Casanova updates the e1000e driver to use delayed work instead of
timers to run the watchdog.

Vitaly fixes an issue in e1000e, where when disconnecting and
reconnecting the physical cable connection, the NIC enters a DMoff
state.  This state causes a mismatch in link and duplexing, so check the
PCIm function state and perform a PHY reset when in this state to
resolve the issue.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: remove not needed call to dma_sync_single_for_device
Heiner Kallweit [Thu, 27 Jun 2019 21:19:09 +0000 (23:19 +0200)]
r8169: remove not needed call to dma_sync_single_for_device

DMA_API_HOWTO.txt includes an example explaining when
dma_sync_single_for_device() is not needed, and that example matches
our use case. The buffer isn't changed by the CPU and direction is
DMA_FROM_DEVICE, so we can remove the call to
dma_sync_single_for_device().

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: consider that 32 Bit DMA is the default
Heiner Kallweit [Thu, 27 Jun 2019 21:12:39 +0000 (23:12 +0200)]
r8169: consider that 32 Bit DMA is the default

Documentation/DMA-API-HOWTO.txt states:
By default, the kernel assumes that your device can address 32-bits of
DMA addressing. For a 64-bit capable device, this needs to be increased,
and for a device with limitations, it needs to be decreased.

Therefore we don't need the 32 Bit DMA fallback configuration and can
remove it.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: improve handling VLAN tag
Heiner Kallweit [Thu, 27 Jun 2019 21:06:33 +0000 (23:06 +0200)]
r8169: improve handling VLAN tag

The VLAN tag is stored in the descriptor in network byte order.
Using swab16 works on little endian host systems only. Better play safe
and use ntohs or htons respectively.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: rtnetlink: skip ipsec offload tests if netdevsim isn't present
Florian Westphal [Thu, 27 Jun 2019 15:12:42 +0000 (17:12 +0200)]
selftests: rtnetlink: skip ipsec offload tests if netdevsim isn't present

running the script on systems without netdevsim now prints:

SKIP: ipsec_offload can't load netdevsim

instead of error message & failed status.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'em_ipt-add-support-for-addrtype'
David S. Miller [Sat, 29 Jun 2019 18:15:12 +0000 (11:15 -0700)]
Merge branch 'em_ipt-add-support-for-addrtype'

Nikolay Aleksandrov says:

====================
em_ipt: add support for addrtype

We would like to be able to use the addrtype from tc for ACL rules and
em_ipt seems the best place to add support for the already existing xt
match. The biggest issue is that addrtype revision 1 (with ipv6 support)
is NFPROTO_UNSPEC and currently em_ipt can't differentiate between v4/v6
if such xt match is used because it passes the match's family instead of
the packet one. The first 3 patches make em_ipt match only on IP
traffic (currently both policy and addrtype recognize such traffic
only) and make it pass the actual packet's protocol instead of the xt
match family when it's unspecified. They also add support for NFPROTO_UNSPEC
xt matches. The last patch allows to add addrtype rules via em_ipt.
We need to keep the user-specified nfproto for dumping in order to be
compatible with libxtables, we cannot dump NFPROTO_UNSPEC as the nfproto
or we'll get an error from libxtables, thus the nfproto is limited to
ipv4/ipv6 in patch 03 and is recorded.

v3: don't use the user nfproto for matching, only for dumping, more
    information is available in the commit message in patch 03
v2: change patch 02 to set the nfproto only when unspecified and drop
    patch 04 from v1 (Eyal Birger)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>