platform/kernel/linux-rpi.git
4 years agotcp: ipv6: support RFC 6069 (TCP-LD)
Eric Dumazet [Thu, 28 May 2020 00:34:58 +0000 (17:34 -0700)]
tcp: ipv6: support RFC 6069 (TCP-LD)

Make tcp_ld_RTO_revert() helper available to IPv6, and
implement RFC 6069 :

Quoting this RFC :

3. Connectivity Disruption Indication

   For Internet Protocol version 6 (IPv6) [RFC2460], the counterpart of
   the ICMP destination unreachable message of code 0 (net unreachable)
   and of code 1 (host unreachable) is the ICMPv6 destination
   unreachable message of code 0 (no route to destination) [RFC4443].
   As with IPv4, a router should generate an ICMPv6 destination
   unreachable message of code 0 in response to a packet that cannot be
   delivered to its destination address because it lacks a matching
   entry in its routing table.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: offload the Credit-Based Shaper qdisc
Vladimir Oltean [Thu, 28 May 2020 00:27:58 +0000 (03:27 +0300)]
net: dsa: sja1105: offload the Credit-Based Shaper qdisc

SJA1105, being AVB/TSN switches, provide hardware assist for the
Credit-Based Shaper as described in the IEEE 8021Q-2018 document.

First generation has 10 shapers, freely assignable to any of the 4
external ports and 8 traffic classes, and second generation has 16
shapers.

The Credit-Based Shaper tables are accessed through the dynamic
reconfiguration interface, so we have to restore them manually after a
switch reset. The tables are backed up by the static config only on
P/Q/R/S, and we don't want to add custom code only for that family,
since the procedure that is in place now works for both.

Tested with the following commands:

data_rate_kbps=67000
port_transmit_rate_kbps=1000000
idleslope=$data_rate_kbps
sendslope=$(($idleslope - $port_transmit_rate_kbps))
locredit=$((-0x80000000))
hicredit=$((0x7fffffff))
tc qdisc add dev swp2 root handle 1: mqprio hw 0 num_tc 8 \
        map 0 1 2 3 4 5 6 7 \
        queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7
tc qdisc replace dev swp2 parent 1:1 cbs \
        idleslope $idleslope \
        sendslope $sendslope \
        hicredit $hicredit \
        locredit $locredit \
        offload 1

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: Add torture tests to nexthop tests
David Ahern [Thu, 28 May 2020 00:03:44 +0000 (18:03 -0600)]
selftests: Add torture tests to nexthop tests

Add Nik's torture tests as a new set to stress the replace and cleanup
paths.

Torture test created by Nikolay Aleksandrov and then I adapted to
selftest and added IPv6 version.

Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'remove-kernel_getsockopt'
David S. Miller [Wed, 27 May 2020 22:11:33 +0000 (15:11 -0700)]
Merge branch 'remove-kernel_getsockopt'

Christoph Hellwig says:

====================
remove kernel_getsockopt

this series reduces scope from the last round and just removes
kernel_getsockopt to avoid conflicting with the sctp cleanup series.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: remove kernel_getsockopt
Christoph Hellwig [Wed, 27 May 2020 18:22:29 +0000 (20:22 +0200)]
net: remove kernel_getsockopt

No users left.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodlm: use the tcp version of accept_from_sock for sctp as well
Christoph Hellwig [Wed, 27 May 2020 18:22:28 +0000 (20:22 +0200)]
dlm: use the tcp version of accept_from_sock for sctp as well

The only difference between a few missing fixes applied to the SCTP
one is that TCP uses ->getpeername to get the remote address, while
SCTP uses kernel_getsockopt(.. SCTP_PRIMARY_ADDR).  But given that
getpeername is defined to return the primary address for sctp, there
doesn't seem to be any reason for the different way of quering the
peername, or all the code duplication.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosctp: fix typo sctp_ulpevent_nofity_peer_addr_change
Jonas Falkevik [Wed, 27 May 2020 09:59:43 +0000 (11:59 +0200)]
sctp: fix typo sctp_ulpevent_nofity_peer_addr_change

change typo in function name "nofity" to "notify"
sctp_ulpevent_nofity_peer_addr_change ->
sctp_ulpevent_notify_peer_addr_change

Signed-off-by: Jonas Falkevik <jonas.falkevik@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/tls: Add force_resync for driver resync
Tariq Toukan [Wed, 27 May 2020 09:25:26 +0000 (12:25 +0300)]
net/tls: Add force_resync for driver resync

This patch adds a field to the tls rx offload context which enables
drivers to force a send_resync call.

This field can be used by drivers to request a resync at the next
possible tls record. It is beneficial for hardware that provides the
resync sequence number asynchronously. In such cases, the packet that
triggered the resync does not contain the information required for a
resync. Instead, the driver requests resync for all the following
TLS record until the asynchronous notification with the resync request
TCP sequence arrives.

A following series for mlx5e ConnectX-6DX TLS RX offload support will
use this mechanism.

Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net_sched-reduce-the-number-of-qdisc-resets'
David S. Miller [Wed, 27 May 2020 22:05:50 +0000 (15:05 -0700)]
Merge branch 'net_sched-reduce-the-number-of-qdisc-resets'

Cong Wang says:

====================
net_sched: reduce the number of qdisc resets

This patchset aims to reduce the number of qdisc resets during
qdisc tear down. Patch 1~3 are preparation for their following
patches, especially patch 2 and patch 3 add a few tracepoints
so that we can observe the whole lifetime of qdisc's. Patch 4
and patch 5 are the ones do the actual work. Please find more
details in each patch description.

Vaclav Zindulka tested this patchset and his large ruleset with
over 13k qdiscs defined got from 22s to 520ms.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: get rid of unnecessary dev_qdisc_reset()
Cong Wang [Wed, 27 May 2020 04:35:27 +0000 (21:35 -0700)]
net_sched: get rid of unnecessary dev_qdisc_reset()

Resetting old qdisc on dev_queue->qdisc_sleeping in
dev_qdisc_reset() is redundant, because this qdisc,
even if not same with dev_queue->qdisc, is reset via
qdisc_put() right after calling dev_graft_qdisc() when
hitting refcnt 0.

This is very easy to observe with qdisc_reset() tracepoint
and stack traces.

Reported-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Tested-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: avoid resetting active qdisc for multiple times
Cong Wang [Wed, 27 May 2020 04:35:26 +0000 (21:35 -0700)]
net_sched: avoid resetting active qdisc for multiple times

Except for sch_mq and sch_mqprio, each dev queue points to the
same root qdisc, so when we reset the dev queues with
netdev_for_each_tx_queue() we end up resetting the same instance
of the root qdisc for multiple times.

Avoid this by checking the __QDISC_STATE_DEACTIVATED bit in
each iteration, so for sch_mq/sch_mqprio, we still reset all
of them like before, for the rest, we only reset it once.

Reported-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Tested-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: add a tracepoint for qdisc creation
Cong Wang [Wed, 27 May 2020 04:35:25 +0000 (21:35 -0700)]
net_sched: add a tracepoint for qdisc creation

With this tracepoint, we could know when qdisc's are created,
especially those default qdisc's.

Sample output:

  tc-736   [001] ...1    56.230107: qdisc_create: dev=ens3 kind=pfifo parent=1:0
  tc-736   [001] ...1    56.230113: qdisc_create: dev=ens3 kind=hfsc parent=ffff:ffff
  tc-738   [001] ...1    56.256816: qdisc_create: dev=ens3 kind=pfifo parent=1:100
  tc-739   [001] ...1    56.267584: qdisc_create: dev=ens3 kind=pfifo parent=1:200
  tc-740   [001] ...1    56.279649: qdisc_create: dev=ens3 kind=fq_codel parent=1:100
  tc-741   [001] ...1    56.289996: qdisc_create: dev=ens3 kind=pfifo_fast parent=1:200
  tc-745   [000] .N.1   111.687483: qdisc_create: dev=ens3 kind=ingress parent=ffff:fff1

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: add tracepoints for qdisc_reset() and qdisc_destroy()
Cong Wang [Wed, 27 May 2020 04:35:24 +0000 (21:35 -0700)]
net_sched: add tracepoints for qdisc_reset() and qdisc_destroy()

Add two tracepoints for qdisc_reset() and qdisc_destroy() to track
qdisc resetting and destroying.

Sample output:

  tc-756   [000] ...3   138.355662: qdisc_reset: dev=ens3 kind=pfifo_fast parent=ffff:ffff handle=0:0
  tc-756   [000] ...1   138.355720: qdisc_reset: dev=ens3 kind=pfifo_fast parent=ffff:ffff handle=0:0
  tc-756   [000] ...1   138.355867: qdisc_reset: dev=ens3 kind=pfifo_fast parent=ffff:ffff handle=0:0
  tc-756   [000] ...1   138.355930: qdisc_destroy: dev=ens3 kind=pfifo_fast parent=ffff:ffff handle=0:0
  tc-757   [000] ...2   143.073780: qdisc_reset: dev=ens3 kind=fq_codel parent=ffff:ffff handle=8001:0
  tc-757   [000] ...1   143.073878: qdisc_reset: dev=ens3 kind=fq_codel parent=ffff:ffff handle=8001:0
  tc-757   [000] ...1   143.074114: qdisc_reset: dev=ens3 kind=fq_codel parent=ffff:ffff handle=8001:0
  tc-757   [000] ...1   143.074228: qdisc_destroy: dev=ens3 kind=fq_codel parent=ffff:ffff handle=8001:0

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: use qdisc_reset() in qdisc_destroy()
Cong Wang [Wed, 27 May 2020 04:35:23 +0000 (21:35 -0700)]
net_sched: use qdisc_reset() in qdisc_destroy()

qdisc_destroy() calls ops->reset() and cleans up qdisc->gso_skb
and qdisc->skb_bad_txq, these are nearly same with qdisc_reset(),
so just call it directly, and cosolidate the code for the next
patch.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodrivers: ipa: remove discription of nonexistent element
Wang Wenhu [Wed, 27 May 2020 03:32:22 +0000 (20:32 -0700)]
drivers: ipa: remove discription of nonexistent element

No element named "client" exists within "struct ipa_endpoint".
It might be a heritage forgotten to be removed. Delete it now.

Signed-off-by: Wang Wenhu <wenhu.wang@vivo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodrivers: ipa: fix typoes for ipa
Wang Wenhu [Wed, 27 May 2020 03:19:24 +0000 (20:19 -0700)]
drivers: ipa: fix typoes for ipa

Change "transactio" -> "transaction". Also an alignment correction.

Signed-off-by: Wang Wenhu <wenhu.wang@vivo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'tcp-tcp_v4_err-cleanups'
David S. Miller [Wed, 27 May 2020 21:57:27 +0000 (14:57 -0700)]
Merge branch 'tcp-tcp_v4_err-cleanups'

Eric Dumazet says:

====================
tcp: tcp_v4_err() cleanups

This series is a followup of patch 239174945dac ("tcp: tcp_v4_err() icmp
skb is named icmp_skb").

Move the RFC 6069 code into a helper, and rename icmp_skb to standard
skb name so that tcp_v4_err() and tcp_v6_err() are using consistent names.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: rename tcp_v4_err() skb parameter
Eric Dumazet [Wed, 27 May 2020 02:48:50 +0000 (19:48 -0700)]
tcp: rename tcp_v4_err() skb parameter

This essentially reverts 4d1a2d9ec1c1 ("Revert Backoff [v3]:
Rename skb to icmp_skb in tcp_v4_err()")

Now we have tcp_ld_RTO_revert() helper, we can use the usual
name for sk_buff parameter, so that tcp_v4_err() and
tcp_v6_err() use similar names.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: add tcp_ld_RTO_revert() helper
Eric Dumazet [Wed, 27 May 2020 02:48:49 +0000 (19:48 -0700)]
tcp: add tcp_ld_RTO_revert() helper

RFC 6069 logic has been implemented for IPv4 only so far,
right in the middle of tcp_v4_err() and was error prone.

Move this code to one helper, to make tcp_v4_err() more
readable and to eventually expand RFC 6069 to IPv6 in
the future.

Also perform sock_owned_by_user() check a bit sooner.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hns3-next'
David S. Miller [Wed, 27 May 2020 21:56:08 +0000 (14:56 -0700)]
Merge branch 'hns3-next'

Huazhong Tan says:

====================
net: hns3: misc updates for -next

This patchset includes some misc updates for the HNS3 ethernet driver.

====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add a print for initializing CMDQ when reset pending
Huazhong Tan [Wed, 27 May 2020 00:59:17 +0000 (08:59 +0800)]
net: hns3: add a print for initializing CMDQ when reset pending

When initializing CMDQ fails because of reset pending,
there is no hint for debugging, so adds a log for it.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: remove unnecessary MAC enable in app loopback
Yufeng Mo [Wed, 27 May 2020 00:59:16 +0000 (08:59 +0800)]
net: hns3: remove unnecessary MAC enable in app loopback

Packets will not pass through MAC during app loopback.
Therefore, it is meaningless to enable MAC while doing
app loopback. This patch removes this unnecessary action.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: change the order of reinitializing RoCE and NIC client during reset
Yufeng Mo [Wed, 27 May 2020 00:59:15 +0000 (08:59 +0800)]
net: hns3: change the order of reinitializing RoCE and NIC client during reset

The HNS RDMA driver will support VF device later, whose
re-initialization should be done after PF's. This patch
changes the order of hclge_reset_prepare_up() and
hclge_notify_roce_client(), so that PF's RoCE client
will be reinitialized before VF's.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add a resetting check in hclgevf_init_nic_client_instance()
Guangbin Huang [Wed, 27 May 2020 00:59:14 +0000 (08:59 +0800)]
net: hns3: add a resetting check in hclgevf_init_nic_client_instance()

To prevent from initializing VF NIC client in reset handling state,
this patch adds resetting check in hclgevf_init_nic_client_instance().

Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-mscc-allow-forwarding-ioctl-operations-to-attached-PHYs'
David S. Miller [Wed, 27 May 2020 21:54:32 +0000 (14:54 -0700)]
Merge branch 'net-mscc-allow-forwarding-ioctl-operations-to-attached-PHYs'

Antoine Tenart says:

====================
net: mscc: allow forwarding ioctl operations to attached PHYs

These two patches allow forwarding ioctl to the PHY MII implementation,
and support is added for offloading timestamping operations to
compatible attached PHYs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: allow offloading timestamping operations to the PHY
Antoine Tenart [Tue, 26 May 2020 15:01:49 +0000 (17:01 +0200)]
net: mscc: allow offloading timestamping operations to the PHY

This patch adds support for offloading timestamping operations not only
to the Ocelot switch (as already supported) but to compatible PHYs.
When both the PHY and the Ocelot switch support timestamping operations,
the PHY implementation is chosen as the timestamp will happen closer to
the medium.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: use the PHY MII ioctl interface when possible
Antoine Tenart [Tue, 26 May 2020 15:01:48 +0000 (17:01 +0200)]
net: mscc: use the PHY MII ioctl interface when possible

Allow ioctl to be implemented by the PHY, when a PHY is attached to the
Ocelot switch. In case the ioctl is a request to set or get the hardware
timestamp, use the Ocelot switch implementation for now.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: felix: accept VLAN config regardless of bridge VLAN awareness state
Vladimir Oltean [Wed, 27 May 2020 16:45:38 +0000 (19:45 +0300)]
net: dsa: felix: accept VLAN config regardless of bridge VLAN awareness state

The ocelot core library is written with the idea in mind that the VLAN
table is populated by the bridge. Otherwise, not even a sane default
pvid is provided: in standalone mode, the default pvid is 0, and the
core expects the bridge layer to change it to 1.

So without this patch, the VLAN table is completely empty at the end of
the commands below, and traffic is broken as a result:

ip link add dev br0 type bridge vlan_filtering 0 && ip link set dev br0 up
for eth in $(ls /sys/bus/pci/devices/0000\:00\:00.5/net/); do
ip link set dev $eth master br0
ip link set dev $eth up
done
ip link set dev br0 type bridge vlan_filtering 1

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: add large ecmp group nexthop tests
Stephen Worley [Wed, 27 May 2020 16:41:42 +0000 (12:41 -0400)]
net: add large ecmp group nexthop tests

Add a couple large ecmp group nexthop selftests to cover
the remnant fixed by d69100b8eee27c2d60ee52df76e0b80a8d492d34.

The tests create 100 x32 ecmp groups of ipv4 and ipv6 and then
dump them. On kernels without the fix, they will fail due
to data remnant during the dump.

Signed-off-by: Stephen Worley <sworley@cumulusnetworks.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomtk-star-emac: mark PM functions as __maybe_unused
Arnd Bergmann [Wed, 27 May 2020 13:34:45 +0000 (15:34 +0200)]
mtk-star-emac: mark PM functions as __maybe_unused

Without CONFIG_PM, the compiler warns about two unused functions:

drivers/net/ethernet/mediatek/mtk_star_emac.c:1472:12: error: unused function 'mtk_star_suspend' [-Werror,-Wunused-function]
drivers/net/ethernet/mediatek/mtk_star_emac.c:1488:12: error: unused function 'mtk_star_resume' [-Werror,-Wunused-function]

Mark these as __maybe_unused.

Fixes: 8c7bd5a454ff ("net: ethernet: mtk-star-emac: new driver")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobridge: mrp: Rework the MRP netlink interface
Horatiu Vultur [Wed, 27 May 2020 12:34:30 +0000 (12:34 +0000)]
bridge: mrp: Rework the MRP netlink interface

This patch reworks the MRP netlink interface. Before, each attribute
represented a binary structure which made it hard to be extended.
Therefore update the MRP netlink interface such that each existing
attribute to be a nested attribute which contains the fields of the
binary structures.
In this way the MRP netlink interface can be extended without breaking
the backwards compatibility. It is also using strict checking for
attributes under the MRP top attribute.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: b53: remove redundant premature assignment to new_pvid
Colin Ian King [Wed, 27 May 2020 12:01:29 +0000 (13:01 +0100)]
net: dsa: b53: remove redundant premature assignment to new_pvid

Variable new_pvid is being assigned with a value that is never read,
the following if statement updates new_pvid with a new value in both
of the if paths. The assignment is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: mtk-star-emac: fix error path in RX handling
Bartosz Golaszewski [Wed, 27 May 2020 09:24:04 +0000 (11:24 +0200)]
net: ethernet: mtk-star-emac: fix error path in RX handling

The dma_addr field in desc_data must not be overwritten until after the
new skb is mapped. Currently we do replace it with uninitialized value
in error path. This change fixes it by moving the assignment before the
label to which we jump after mapping or allocation errors.

Fixes: 8c7bd5a454ff ("net: ethernet: mtk-star-emac: new driver")
Reported-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Tested-by: Nathan Chancellor <natechancellor@gmail.com> # build
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: remove redundant initialization of pointer br_dev
Colin Ian King [Wed, 27 May 2020 08:15:55 +0000 (09:15 +0100)]
mlxsw: spectrum_router: remove redundant initialization of pointer br_dev

The pointer br_dev is being initialized with a value that is never read
and is being updated with a new value later on. The initialization
is redundant and can be removed.

Addresses-Coverity: ("Unused value")
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>
4 years agonexthop: Fix type of event_type in call_nexthop_notifiers
Nathan Chancellor [Wed, 27 May 2020 08:00:20 +0000 (01:00 -0700)]
nexthop: Fix type of event_type in call_nexthop_notifiers

Clang warns:

net/ipv4/nexthop.c:841:30: warning: implicit conversion from enumeration
type 'enum nexthop_event_type' to different enumeration type 'enum
fib_event_type' [-Wenum-conversion]
        call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh);
        ~~~~~~~~~~~~~~~~~~~~~~      ^~~~~~~~~~~~~~~~~
1 warning generated.

Use the right type for event_type so that clang does not warn.

Fixes: 8590ceedb701 ("nexthop: add support for notifiers")
Link: https://github.com/ClangBuiltLinux/linux/issues/1038
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: at803x: add cable diagnostics support for ATH9331 and ATH8032
Oleksij Rempel [Wed, 27 May 2020 05:08:43 +0000 (07:08 +0200)]
net: phy: at803x: add cable diagnostics support for ATH9331 and ATH8032

Add support for Atheros 100Base-T PHYs. The only difference seems to be
the ability to test 2 pairs instead of 4 and the lack of 1000Base-T
specific register.

Only the ATH9331 was tested with this patch.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Raw-PHY-TDR-data'
David S. Miller [Wed, 27 May 2020 06:22:28 +0000 (23:22 -0700)]
Merge branch 'Raw-PHY-TDR-data'

Andrew Lunn says:

====================
Raw PHY TDR data

Some ethernet PHYs allow access to raw TDR data in addition to summary
diagnostics information. Add support for retrieving this data via
netlink ethtool. The basic structure in the core is the same as for
normal phy diagnostics, the PHY driver simply uses different helpers
to fill the netlink message with different data.

There is a graphical tool under development, as well a ethtool(1)
which can dump the data as text and JSON.

A patched ethtool(1) can be found in
https://github.com/lunn/ethtool.git feature/cable-test-v5

Thanks for Chris Healy for lots of testing.

v2:
See the individual patches but:

Pass distances in centimeters, not meters

Allow the PHY to round distances to what it supports and report what
it actually used along with the results.

Make the Marvell PHY use steps a multiple of 0.805 meters, its native
step size.

v3:
Move the TDR configuration into a structure
Add a range check on step
Use NL_SET_ERR_MSG_ATTR() when appropriate
Move TDR configuration into a nest
Document attributes in the request
Unsquash the last two patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: Configure TDR pulse based on measurement length
Andrew Lunn [Tue, 26 May 2020 22:21:43 +0000 (00:21 +0200)]
net: phy: marvell: Configure TDR pulse based on measurement length

When performing a TDR measurement for a short distance, the pulse
width should be low, to help differentiate between the outgoing pulse
and any reflection. For longer distances, the pulse should be wider,
to help with attenuation.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet : phy: marvell: Speedup TDR data retrieval by only changing page once
Andrew Lunn [Tue, 26 May 2020 22:21:42 +0000 (00:21 +0200)]
net : phy: marvell: Speedup TDR data retrieval by only changing page once

Getting the TDR data requires a large number of MDIO bus
transactions. The number can however be reduced if the page is only
changed once. Add the needed locking to allow this, and make use of
unlocked read/write methods where needed.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethtool: Allow PHY cable test TDR data to configured
Andrew Lunn [Tue, 26 May 2020 22:21:41 +0000 (00:21 +0200)]
net: ethtool: Allow PHY cable test TDR data to configured

Allow the user to configure where on the cable the TDR data should be
retrieved, in terms of first and last sample, and the step between
samples. Also add the ability to ask for TDR data for just one pair.

If this configuration is not provided, it defaults to 1-150m at 1m
intervals for all pairs.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
v3:
Move the TDR configuration into a structure
Add a range check on step
Use NL_SET_ERR_MSG_ATTR() when appropriate
Move TDR configuration into a nest
Document attributes in the request

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: Add support for amplitude graph
Andrew Lunn [Tue, 26 May 2020 22:21:40 +0000 (00:21 +0200)]
net: phy: marvell: Add support for amplitude graph

The Marvell PHYs can measure the amplitude of the returned signal for
a given distance. Implement this option of the cable test
infrastructure. When reporting the step, convert the distance into cm.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
v2:
Step based on the measurement resolution, and convert this to cm.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethtool: Add helpers for cable test TDR data
Andrew Lunn [Tue, 26 May 2020 22:21:39 +0000 (00:21 +0200)]
net: ethtool: Add helpers for cable test TDR data

Add helpers for returning raw TDR helpers in netlink messages.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethtool: Add generic parts of cable test TDR
Andrew Lunn [Tue, 26 May 2020 22:21:38 +0000 (00:21 +0200)]
net: ethtool: Add generic parts of cable test TDR

Add the generic parts of the code used to trigger a cable test and
return raw TDR data. Any PHY driver which support this must implement
the new driver op.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
v2
Update nxp-tja11xx for API change.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethtool: Add attributes for cable test TDR data
Andrew Lunn [Tue, 26 May 2020 22:21:37 +0000 (00:21 +0200)]
net: ethtool: Add attributes for cable test TDR data

Some Ethernet PHYs can return the raw time domain reflectromatry data.
Add the attributes to allow this data to be requested and returned via
netlink ethtool.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
v2:
m -> cm
Report what the PHY actually used for start/stop/step.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agone2k-pci: Fix various coding-style issues and improve printk() usage
Armin Wolf [Tue, 26 May 2020 18:03:02 +0000 (20:03 +0200)]
ne2k-pci: Fix various coding-style issues and improve printk() usage

Fixed a ton of minor checkpatch errors/warnings and remove version
printing at module init/when device is found and use MODULE_VERSION
instead. Also modifying the RTL8029 PCI string to include the compatible
RTL8029AS nic.
The only mayor issue remaining is the missing SPDX tag, but since the
exact version of the GPL is not stated anywhere inside the file, its
impossible to add such a tag at the moment.
But maybe it is possible, since 8390.h states Donald Becker's 8390
drivers are licensed under GPL 2.2 only (= GPL-2.0-only ?).
The kernel module containing this patch compiles and runs without
problems on a RTL8029AS-based NE2000 clone card with kernel 5.7.0-rc6.

Signed-off-by: Armin Wolf <W_Armin@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomacvlan: Skip loopback packets in RX handler
Alexander Sverdlin [Tue, 26 May 2020 12:27:51 +0000 (14:27 +0200)]
macvlan: Skip loopback packets in RX handler

Ignore loopback-originatig packets soon enough and don't try to process L2
header where it doesn't exist. The very similar br_handle_frame() in bridge
code performs exactly the same check.

This is an example of such ICMPv6 packet:

skb len=96 headroom=40 headlen=96 tailroom=56
mac=(40,0) net=(40,40) trans=80
shinfo(txflags=0 nr_frags=0 gso(size=0 type=0 segs=0))
csum(0xae2e9a2f ip_summed=1 complete_sw=0 valid=0 level=0)
hash(0xc97ebd88 sw=1 l4=1) proto=0x86dd pkttype=5 iif=24
dev name=etha01.212 feat=0x0x0000000040005000
skb headroom: 00000000: 00 7c 86 52 84 88 ff ff 00 00 00 00 00 00 08 00
skb headroom: 00000010: 45 00 00 9e 5d 5c 40 00 40 11 33 33 00 00 00 01
skb headroom: 00000020: 02 40 43 80 00 00 86 dd
skb linear:   00000000: 60 09 88 bd 00 38 3a ff fe 80 00 00 00 00 00 00
skb linear:   00000010: 00 40 43 ff fe 80 00 00 ff 02 00 00 00 00 00 00
skb linear:   00000020: 00 00 00 00 00 00 00 01 86 00 61 00 40 00 00 2d
skb linear:   00000030: 00 00 00 00 00 00 00 00 03 04 40 e0 00 00 01 2c
skb linear:   00000040: 00 00 00 78 00 00 00 00 fd 5f 42 68 23 87 a8 81
skb linear:   00000050: 00 00 00 00 00 00 00 00 01 01 02 40 43 80 00 00
skb tailroom: 00000000: ...
skb tailroom: 00000010: ...
skb tailroom: 00000020: ...
skb tailroom: 00000030: ...

Call Trace, how it happens exactly:
 ...
 macvlan_handle_frame+0x321/0x425 [macvlan]
 ? macvlan_forward_source+0x110/0x110 [macvlan]
 __netif_receive_skb_core+0x545/0xda0
 ? enqueue_task_fair+0xe5/0x8e0
 ? __netif_receive_skb_one_core+0x36/0x70
 __netif_receive_skb_one_core+0x36/0x70
 process_backlog+0x97/0x140
 net_rx_action+0x1eb/0x350
 ? __hrtimer_run_queues+0x136/0x2e0
 __do_softirq+0xe3/0x383
 do_softirq_own_stack+0x2a/0x40
 </IRQ>
 do_softirq.part.4+0x4e/0x50
 netif_rx_ni+0x60/0xd0
 dev_loopback_xmit+0x83/0xf0
 ip6_finish_output2+0x575/0x590 [ipv6]
 ? ip6_cork_release.isra.1+0x64/0x90 [ipv6]
 ? __ip6_make_skb+0x38d/0x680 [ipv6]
 ? ip6_output+0x6c/0x140 [ipv6]
 ip6_output+0x6c/0x140 [ipv6]
 ip6_send_skb+0x1e/0x60 [ipv6]
 rawv6_sendmsg+0xc4b/0xe10 [ipv6]
 ? proc_put_long+0xd0/0xd0
 ? rw_copy_check_uvector+0x4e/0x110
 ? sock_sendmsg+0x36/0x40
 sock_sendmsg+0x36/0x40
 ___sys_sendmsg+0x2b6/0x2d0
 ? proc_dointvec+0x23/0x30
 ? addrconf_sysctl_forward+0x8d/0x250 [ipv6]
 ? dev_forward_change+0x130/0x130 [ipv6]
 ? _raw_spin_unlock+0x12/0x30
 ? proc_sys_call_handler.isra.14+0x9f/0x110
 ? __call_rcu+0x213/0x510
 ? get_max_files+0x10/0x10
 ? trace_hardirqs_on+0x2c/0xe0
 ? __sys_sendmsg+0x63/0xa0
 __sys_sendmsg+0x63/0xa0
 do_syscall_64+0x6c/0x1e0
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Various-trap-changes-part-2'
David S. Miller [Wed, 27 May 2020 03:33:59 +0000 (20:33 -0700)]
Merge branch 'mlxsw-Various-trap-changes-part-2'

Ido Schimmel says:

====================
mlxsw: Various trap changes - part 2

This patch set contains another set of small changes in mlxsw trap
configuration. It is the last set before exposing control traps (e.g.,
IGMP query, ARP request) via devlink-trap.

Tested with existing devlink-trap selftests. Please see individual
patches for a detailed changelog.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Allow programming link-local prefix routes
Ido Schimmel [Mon, 25 May 2020 23:05:56 +0000 (02:05 +0300)]
mlxsw: spectrum_router: Allow programming link-local prefix routes

The device has a trap for IPv6 packets that need be routed and have a
unicast link-local destination IP (i.e., fe80::/10). This allows mlxsw
to ignore link-local routes, as the packets will be trapped to the CPU
in any case.

However, since link-local routes are not programmed, it is possible for
routed packets to hit the default route which might also be programmed
to trap packets. This means that packets with a link-local destination
IP might be trapped for the wrong reason.

To overcome this, allow programming link-local prefix routes (usually
one fe80::/64 per-table), so that the packets will be forwarded until
reaching the link-local trap.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Add packet traps for BFD packets
Ido Schimmel [Mon, 25 May 2020 23:05:55 +0000 (02:05 +0300)]
mlxsw: spectrum: Add packet traps for BFD packets

Bidirectional Forwarding Detection (BFD) provides "low-overhead,
short-duration detection of failures in the path between adjacent
forwarding engines" (RFC 5880).

This is accomplished by exchanging BFD packets between the two
forwarding engines. Up until now these packets were trapped via the
general local delivery (i.e., IP2ME) trap which also traps a lot of
other packets that are not as time-sensitive as BFD packets.

Expose dedicated traps for BFD packets so that user space could
configure a dedicated policer for them.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Treat IPv6 link-local SIP as an exception
Ido Schimmel [Mon, 25 May 2020 23:05:54 +0000 (02:05 +0300)]
mlxsw: spectrum: Treat IPv6 link-local SIP as an exception

IPv6 packets that need to be forwarded and have a link-local source IP are
dropped by the kernel and an ICMPv6 "Destination unreachable" is sent to
the sending host.

As such, change the trap group of such packets so that they do not
interfere with IPv6 management packets. In the future this trap will be
exposed as an exception via devlink-trap.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Share one group for all locally delivered packets
Ido Schimmel [Mon, 25 May 2020 23:05:53 +0000 (02:05 +0300)]
mlxsw: spectrum: Share one group for all locally delivered packets

Routed IP packets with the Router Alert option need to be trapped to
the CPU as they might need to be locally delivered to raw sockets with
the IP_ROUTER_ALERT / IPV6_ROUTER_ALERT socket option.

Move them to the same group with other packets that might need to be
trapped following route lookup.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: reg: Move all trap groups under the same enum
Ido Schimmel [Mon, 25 May 2020 23:05:52 +0000 (02:05 +0300)]
mlxsw: reg: Move all trap groups under the same enum

After the previous patch the split is no longer necessary and all the
trap groups can be moved under the same enum.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_trap: Do not hard code "thin" policer identifier
Ido Schimmel [Mon, 25 May 2020 23:05:51 +0000 (02:05 +0300)]
mlxsw: spectrum_trap: Do not hard code "thin" policer identifier

As explained in commit e612523041ab ("mlxsw: spectrum_trap: Introduce
dummy group with thin policer"), the purpose of the "thin" policer is to
pass as less packets as possible to the CPU.

The identifier of this policer is currently set according to the maximum
number of used trap groups, but this is fragile: On Spectrum-1 the
maximum number of policers is less than the maximum number of trap
groups, which might result in an invalid policer identifier in case the
number of used trap groups grows beyond the policer limit.

Solve this by dynamically allocating the policer identifier.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: switchx2: Move SwitchX-2 trap groups out of main enum
Ido Schimmel [Mon, 25 May 2020 23:05:50 +0000 (02:05 +0300)]
mlxsw: switchx2: Move SwitchX-2 trap groups out of main enum

The number of Spectrum trap groups is not infinite, but two identifiers
are occupied by SwitchX-2 specific trap groups. Free these identifiers
by moving them out of the main enum.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Reduce priority of locally delivered packets
Ido Schimmel [Mon, 25 May 2020 23:05:49 +0000 (02:05 +0300)]
mlxsw: spectrum: Reduce priority of locally delivered packets

To align with recent recommended values. Will be configurable by future
patches.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Use same trap group for local routes and link-local destination
Ido Schimmel [Mon, 25 May 2020 23:05:48 +0000 (02:05 +0300)]
mlxsw: spectrum: Use same trap group for local routes and link-local destination

Packets with an IPv6 link-local destination (i.e., fe80::/10) should not
be forwarded and are therefore trapped to the CPU for local delivery.
Since these packets are trapped for the same logical reason as packets
hitting local routes, associate both traps with the same group.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Use separate trap group for FID miss
Ido Schimmel [Mon, 25 May 2020 23:05:47 +0000 (02:05 +0300)]
mlxsw: spectrum: Use separate trap group for FID miss

When a packet enters the device it is classified to a filtering
identifier (FID) based on the ingress port and VLAN. The FID miss trap
is used to trap packets for which a FID could not be found.

In mlxsw this trap should only be triggered when a port is enslaved to
an OVS bridge and a matching ACL rule could not be found, so as to
trigger learning.

These packets are therefore completely unrelated to packets hitting
local routes and should be in a different group. Move them.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Use same trap group for various IPv6 packets
Ido Schimmel [Mon, 25 May 2020 23:05:46 +0000 (02:05 +0300)]
mlxsw: spectrum: Use same trap group for various IPv6 packets

Group these various IPv6 packets (e.g., router solicitations, router
advertisement) together and subject them to the same policer.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Rename IPv6 ND trap group
Ido Schimmel [Mon, 25 May 2020 23:05:45 +0000 (02:05 +0300)]
mlxsw: spectrum: Rename IPv6 ND trap group

The IPv6 Neighbour Discovery (ND) group will be used for various IPv6
packets, not all of which fall under the definition of ND, so rename it
to "IPV6" which is more appropriate.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Use same switch case for identical groups
Ido Schimmel [Mon, 25 May 2020 23:05:44 +0000 (02:05 +0300)]
mlxsw: spectrum: Use same switch case for identical groups

Trap groups that use the same policer settings can share the same switch
case.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Use dedicated trap group for ACL trap
Ido Schimmel [Mon, 25 May 2020 23:05:43 +0000 (02:05 +0300)]
mlxsw: spectrum: Use dedicated trap group for ACL trap

Packets that are trapped via tc's trap action are currently subject to
the same policer as packets hitting local routes. The latter are
critical to the correct functioning of the control plane, while the
former are mainly used for traffic inspection.

Split the ACL trap to a separate group with its own policer. Use a
higher priority for these traps than for traps using mirror action
(e.g., ARP, IGMP). Otherwise, packets matching both traps will not be
forwarded in hardware (because of trap action) and also not forwarded in
software because they will be marked with 'offload_fwd_mark'.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: attempt coalescing when moving skbs to mptcp rx queue
Florian Westphal [Mon, 25 May 2020 21:41:13 +0000 (23:41 +0200)]
mptcp: attempt coalescing when moving skbs to mptcp rx queue

We can try to coalesce skbs we take from the subflows rx queue with the
tail of the mptcp rx queue.

If successful, the skb head can be discarded early.

We can also free the skb extensions, we do not access them after this.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: improve rtl_remove_one
Heiner Kallweit [Mon, 25 May 2020 19:54:00 +0000 (21:54 +0200)]
r8169: improve rtl_remove_one

Don't call netif_napi_del() manually, free_netdev() does this for us.
In addition reorder calls to match reverse order of calls in probe().

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ethernet-fec-move-GPR-register-offset-and-bit-into-DT'
David S. Miller [Wed, 27 May 2020 03:21:43 +0000 (20:21 -0700)]
Merge branch 'net-ethernet-fec-move-GPR-register-offset-and-bit-into-DT'

Fugang Duan says:

====================
net: ethernet: fec: move GPR register offset and bit into DT

The commit da722186f654 (net: fec: set GPR bit on suspend by
DT configuration) set the GPR reigster offset and bit in driver
for wol feature support.

It brings trouble to enable wol feature on imx6sx/imx6ul/imx7d
platforms that have multiple ethernet instances with different
GPR bit for stop mode control. So the patch set is to move GPR
register offset and bit define into DT, and enable imx6q/imx6dl
imx6qp/imx6sx/imx6ul/imx7d stop mode support.

Currently, below NXP i.MX boards support wol:
- imx6q/imx6dl/imx6qp sabresd
- imx6sx sabreauto
- imx7d sdb

imx6q/imx6dl/imx6qp sabresd board dts file miss the property
"fsl,magic-packet;", so patch#4 is to add the property for stop
mode support.

v1 -> v2:
 - driver: switch back to store the quirks bitmask in driver_data
 - dt-bindings: rename 'gpr' property string to 'fsl,stop-mode'
 - imx6/7 dtsi: add imx6sx/imx6ul/imx7d ethernet stop mode property
v2 -> v3:
 - driver: suggested by Sascha Hauer, use a struct fec_devinfo for
   abstracting differences between different hardware variants,
   it can give more freedom to describe the differences.
 - imx6/7 dtsi: correct one typo pointed out by Andrew.

Thanks Martin, Andrew and Sascha Hauer for the review.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoARM: dts: imx6qdl-sabresd: enable fec wake-on-lan
Fugang Duan [Mon, 25 May 2020 16:27:13 +0000 (00:27 +0800)]
ARM: dts: imx6qdl-sabresd: enable fec wake-on-lan

Enable ethernet wake-on-lan feature for imx6q/dl/qp sabresd
boards since the PHY clock is supplied by external osc.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoARM: dts: imx: add ethernet stop mode property
Fugang Duan [Mon, 25 May 2020 16:27:12 +0000 (00:27 +0800)]
ARM: dts: imx: add ethernet stop mode property

- Update the imx6qdl gpr property to define gpr register
  offset and bit in DT.
- Add imx6sx/imx6ul/imx7d ethernet stop mode property.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodt-bindings: fec: update the gpr property
Fugang Duan [Mon, 25 May 2020 16:27:11 +0000 (00:27 +0800)]
dt-bindings: fec: update the gpr property

- rename the 'gpr' property string to 'fsl,stop-mode'.
- Update the property to define gpr register offset and
bit in DT, since different instance have different gpr bit.

v2:
 * rename 'gpr' property string to 'fsl,stop-mode'.

Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: fec: move GPR register offset and bit into DT
Fugang Duan [Mon, 25 May 2020 16:27:10 +0000 (00:27 +0800)]
net: ethernet: fec: move GPR register offset and bit into DT

The commit da722186f654 (net: fec: set GPR bit on suspend by DT
configuration) set the GPR reigster offset and bit in driver for
wake on lan feature.

But it introduces two issues here:
- one SOC has two instances, they have different bit
- different SOCs may have different offset and bit

So to support wake-on-lan feature on other i.MX platforms, it should
configure the GPR reigster offset and bit from DT.

So the patch is to improve the commit da722186f654 (net: fec: set GPR
bit on suspend by DT configuration) to support multiple ethernet
instances on i.MX series.

v2:
 * switch back to store the quirks bitmask in driver_data
v3:
 * suggested by Sascha Hauer, use a struct fec_devinfo for
   abstracting differences between different hardware variants,
   it can give more freedom to describe the differences.

Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: mark smc_pnet_policy as const
Dmitry Vyukov [Mon, 25 May 2020 15:31:58 +0000 (17:31 +0200)]
net/smc: mark smc_pnet_policy as const

Netlink policies are generally declared as const.
This is safer and prevents potential bugs.

Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mac80211-next-for-net-next-2020-04-25' of git://git.kernel.org/pub/scm...
David S. Miller [Wed, 27 May 2020 03:17:35 +0000 (20:17 -0700)]
Merge tag 'mac80211-next-for-net-next-2020-04-25' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
One batch of changes, containing:
 * hwsim improvements from Jouni and myself, to be able to
   test more scenarios easily
 * some more HE (802.11ax) support
 * some initial S1G (sub 1 GHz) work for fractional MHz channels
 * some (action) frame registration updates to help DPP support
 * along with other various improvements/fixes
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-phy-mscc-miim-reduce-waiting-time-between-MDIO-transactions'
David S. Miller [Tue, 26 May 2020 22:33:57 +0000 (15:33 -0700)]
Merge branch 'net-phy-mscc-miim-reduce-waiting-time-between-MDIO-transactions'

Antoine Tenart says:

====================
net: phy: mscc-miim: reduce waiting time between MDIO transactions

This series aims at reducing the waiting time between MDIO transactions
when using the MSCC MIIM MDIO controller.

I'm not sure we need patch 4/4 and we could reasonably drop it from the
series. I'm including the patch as it could help to ensure the system
is functional with a non optimal configuration.

We needed to improve the driver's performances as when using a PHY
requiring lots of registers accesses (such as the VSC85xx family),
delays would add up and ended up to be quite large which would cause
issues such as: a slow initialization of the PHY, and issues when using
timestamping operations (this feature will be sent quite soon to the
mailing lists).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc-miim: read poll when high resolution timers are disabled
Antoine Tenart [Tue, 26 May 2020 16:22:56 +0000 (18:22 +0200)]
net: phy: mscc-miim: read poll when high resolution timers are disabled

The driver uses a read polling mechanism to check the status of the MDIO
bus, to know if it is ready to accept next commands. This polling
mechanism uses usleep_delay() under the hood between reads which is fine
as long as high resolution timers are enabled. Otherwise the delays will
end up to be much longer than expected.

This patch fixes this by using udelay() under the hood when
CONFIG_HIGH_RES_TIMERS isn't enabled. This increases CPU usage.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc-miim: improve waiting logic
Antoine Tenart [Tue, 26 May 2020 16:22:55 +0000 (18:22 +0200)]
net: phy: mscc-miim: improve waiting logic

The MSCC MIIM MDIO driver uses a waiting logic to wait for the MDIO bus
to be ready to accept next commands. It does so by polling the BUSY
status bit which indicates the MDIO bus has completed all pending
operations. This can take time, and the controller supports writing the
next command as soon as there are no pending commands (which happens
while the MDIO bus is busy completing its current command).

This patch implements this improved logic by adding an helper to poll
the PENDING status bit, and by adjusting where we should wait for the
bus to not be busy or to not be pending.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc-miim: remove redundant timeout check
Antoine Tenart [Tue, 26 May 2020 16:22:54 +0000 (18:22 +0200)]
net: phy: mscc-miim: remove redundant timeout check

readl_poll_timeout already returns -ETIMEDOUT if the condition isn't
satisfied, there's no need to check again the condition after calling
it. Remove the redundant timeout check.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc-miim: use more reasonable delays
Antoine Tenart [Tue, 26 May 2020 16:22:53 +0000 (18:22 +0200)]
net: phy: mscc-miim: use more reasonable delays

The MSCC MIIM MDIO driver uses delays to read poll a status register. I
made multiple tests on a Ocelot PCS120 platform which led me to reduce
those delays. The delay in between which the polling function is allowed
to sleep is reduced from 100us to 50us which in almost all cases is a
good value to succeed at the first retry. The overall delay is also
lowered as the prior value was really way to high, 10000us is large
enough.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mdiobus: add clause 45 mdiobus accessors
Russell King [Tue, 26 May 2020 15:29:36 +0000 (16:29 +0100)]
net: mdiobus: add clause 45 mdiobus accessors

There is a recurring pattern throughout some of the PHY code converting
a devad and regnum to our packed clause 45 representation. Rather than
having this scattered around the code, let's put a common translation
function in mdio.h, and provide some register accessors.

Convert the phylib core, phylink, bcm87xx and cortina to use these.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'flow-mpls'
David S. Miller [Tue, 26 May 2020 22:22:58 +0000 (15:22 -0700)]
Merge branch 'flow-mpls'

Guillaume Nault says:

====================
flow_dissector, cls_flower: Add support for multiple MPLS Label Stack Entries

Currently, the flow dissector and the Flower classifier can only handle
the first entry of an MPLS label stack. This patch series generalises
the code to allow parsing and matching the Label Stack Entries that
follow.

Patch 1 extends the flow dissector to parse MPLS LSEs until the Bottom
Of Stack bit is reached. The number of parsed LSEs is capped at
FLOW_DIS_MPLS_MAX (arbitrarily set to 7). Flower and the NFP driver
are updated to take into account the new layout of struct
flow_dissector_key_mpls.

Patch 2 extends Flower. It defines new netlink attributes, which are
independent from the previous MPLS ones. Mixing the old and the new
attributes in a same filter is not allowed. For backward compatibility,
the old attributes are used when dumping filters that don't require the
new ones.

Changes since v2:
  * Fix compilation with the new MLX5 bareudp tunnel code.

Changes since v1:
  * Fix compilation of NFP driver (kbuild test robot).
  * Fix sparse warning with entropy label (kbuild test robot).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocls_flower: Support filtering on multiple MPLS Label Stack Entries
Guillaume Nault [Tue, 26 May 2020 12:29:04 +0000 (14:29 +0200)]
cls_flower: Support filtering on multiple MPLS Label Stack Entries

With struct flow_dissector_key_mpls now recording the first
FLOW_DIS_MPLS_MAX labels, we can extend Flower to filter on any of
these LSEs independently.

In order to avoid creating new netlink attributes for every possible
depth, let's define a new TCA_FLOWER_KEY_MPLS_OPTS nested attribute
that contains the list of LSEs to match. Each LSE is represented by
another attribute, TCA_FLOWER_KEY_MPLS_OPTS_LSE, which then contains
the attributes representing the depth and the MPLS fields to match at
this depth (label, TTL, etc.).

For each MPLS field, the mask is always set to all-ones, as this is
what the original API did. We could allow user configurable masks in
the future if there is demand for more flexibility.

The new API also allows to only specify an LSE depth. In that case,
Flower only verifies that the MPLS label stack depth is greater or
equal to the provided depth (that is, an LSE exists at this depth).

Filters that only match on one (or more) fields of the first LSE are
dumped using the old netlink attributes, to avoid confusing user space
programs that don't understand the new API.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoflow_dissector: Parse multiple MPLS Label Stack Entries
Guillaume Nault [Tue, 26 May 2020 12:29:00 +0000 (14:29 +0200)]
flow_dissector: Parse multiple MPLS Label Stack Entries

The current MPLS dissector only parses the first MPLS Label Stack
Entry (second LSE can be parsed too, but only to set a key_id).

This patch adds the possibility to parse several LSEs by making
__skb_flow_dissect_mpls() return FLOW_DISSECT_RET_PROTO_AGAIN as long
as the Bottom Of Stack bit hasn't been seen, up to a maximum of
FLOW_DIS_MPLS_MAX entries.

FLOW_DIS_MPLS_MAX is arbitrarily set to 7. This should be enough for
many practical purposes, without wasting too much space.

To record the parsed values, flow_dissector_key_mpls is modified to
store an array of stack entries, instead of just the values of the
first one. A bit field, "used_lses", is also added to keep track of
the LSEs that have been set. The objective is to avoid defining a
new FLOW_DISSECTOR_KEY_MPLS_XX for each level of the MPLS stack.

TC flower is adapted for the new struct flow_dissector_key_mpls layout.
Matching on several MPLS Label Stack Entries will be added in the next
patch.

The NFP and MLX5 drivers are also adapted: nfp_flower_compile_mac() and
mlx5's parse_tunnel() now verify that the rule only uses the first LSE
and fail if it doesn't.

Finally, the behaviour of the FLOW_DISSECTOR_KEY_MPLS_ENTROPY key is
slightly modified. Instead of recording the first Entropy Label, it
now records the last one. This shouldn't have any consequences since
there doesn't seem to have any user of FLOW_DISSECTOR_KEY_MPLS_ENTROPY
in the tree. We'd probably better do a hash of all parsed MPLS labels
instead (excluding reserved labels) anyway. That'd give better entropy
and would probably also simplify the code. But that's not the purpose
of this patch, so I'm keeping that as a future possible improvement.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'batadv-next-for-davem-20200526' of git://git.open-mesh.org/linux-merge
David S. Miller [Tue, 26 May 2020 22:19:29 +0000 (15:19 -0700)]
Merge tag 'batadv-next-for-davem-20200526' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
This cleanup patchset includes the following patches:

 - Fix revert dynamic lockdep key changes for batman-adv,
   by Sven Eckelmann

 - use rcu_replace_pointer() where appropriate, by Antonio Quartulli

 - Revert "disable ethtool link speed detection when auto negotiation
   off", by Sven Eckelmann
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'tipc-add-some-improvements'
David S. Miller [Tue, 26 May 2020 22:16:52 +0000 (15:16 -0700)]
Merge branch 'tipc-add-some-improvements'

Tuong Lien says:

====================
tipc: add some improvements

This series adds some improvements to TIPC.

The first patch improves the TIPC broadcast's performance with the 'Gap
ACK blocks' mechanism similar to unicast before, while the others give
support on tracing & statistics for broadcast links, and an alternative
to carry broadcast retransmissions via unicast which might be useful in
some cases.

Besides, the Nagle algorithm can now automatically 'adjust' itself
depending on the specific network condition a stream connection runs by
the last patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: add test for Nagle algorithm effectiveness
Tuong Lien [Tue, 26 May 2020 09:38:38 +0000 (16:38 +0700)]
tipc: add test for Nagle algorithm effectiveness

When streaming in Nagle mode, we try to bundle small messages from user
as many as possible if there is one outstanding buffer, i.e. not ACK-ed
by the receiving side, which helps boost up the overall throughput. So,
the algorithm's effectiveness really depends on when Nagle ACK comes or
what the specific network latency (RTT) is, compared to the user's
message sending rate.

In a bad case, the user's sending rate is low or the network latency is
small, there will not be many bundles, so making a Nagle ACK or waiting
for it is not meaningful.
For example: a user sends its messages every 100ms and the RTT is 50ms,
then for each messages, we require one Nagle ACK but then there is only
one user message sent without any bundles.

In a better case, even if we have a few bundles (e.g. the RTT = 300ms),
but now the user sends messages in medium size, then there will not be
any difference at all, that says 3 x 1000-byte data messages if bundled
will still result in 3 bundles with MTU = 1500.

When Nagle is ineffective, the delay in user message sending is clearly
wasted instead of sending directly.

Besides, adding Nagle ACKs will consume some processor load on both the
sending and receiving sides.

This commit adds a test on the effectiveness of the Nagle algorithm for
an individual connection in the network on which it actually runs.
Particularly, upon receipt of a Nagle ACK we will compare the number of
bundles in the backlog queue to the number of user messages which would
be sent directly without Nagle. If the ratio is good (e.g. >= 2), Nagle
mode will be kept for further message sending. Otherwise, we will leave
Nagle and put a 'penalty' on the connection, so it will have to spend
more 'one-way' messages before being able to re-enter Nagle.

In addition, the 'ack-required' bit is only set when really needed that
the number of Nagle ACKs will be reduced during Nagle mode.

Testing with benchmark showed that with the patch, there was not much
difference in throughput for small messages since the tool continuously
sends messages without a break, so Nagle would still take in effect.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: add support for broadcast rcv stats dumping
Tuong Lien [Tue, 26 May 2020 09:38:37 +0000 (16:38 +0700)]
tipc: add support for broadcast rcv stats dumping

This commit enables dumping the statistics of a broadcast-receiver link
like the traditional 'broadcast-link' one (which is for broadcast-
sender). The link dumping can be triggered via netlink (e.g. the
iproute2/tipc tool) by the link flag - 'TIPC_NLA_LINK_BROADCAST' as the
indicator.

The name of a broadcast-receiver link of a specific peer will be in the
format: 'broadcast-link:<peer-id>'.

For example:

Link <broadcast-link:1001002>
  Window:50 packets
  RX packets:7841 fragments:2408/440 bundles:0/0
  TX packets:0 fragments:0/0 bundles:0/0
  RX naks:0 defs:124 dups:0
  TX naks:21 acks:0 retrans:0
  Congestion link:0  Send queue max:0 avg:0

In addition, the broadcast-receiver link statistics can be reset in the
usual way via netlink by specifying that link name in command.

Note: the 'tipc_link_name_ext()' is removed because the link name can
now be retrieved simply via the 'l->name'.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: enable broadcast retrans via unicast
Tuong Lien [Tue, 26 May 2020 09:38:36 +0000 (16:38 +0700)]
tipc: enable broadcast retrans via unicast

In some environment, broadcast traffic is suppressed at high rate (i.e.
a kind of bandwidth limit setting). When it is applied, TIPC broadcast
can still run successfully. However, when it comes to a high load, some
packets will be dropped first and TIPC tries to retransmit them but the
packet retransmission is intentionally broadcast too, so making things
worse and not helpful at all.

This commit enables the broadcast retransmission via unicast which only
retransmits packets to the specific peer that has really reported a gap
i.e. not broadcasting to all nodes in the cluster, so will prevent from
being suppressed, and also reduce some overheads on the other peers due
to duplicates, finally improve the overall TIPC broadcast performance.

Note: the functionality can be turned on/off via the sysctl file:

echo 1 > /proc/sys/net/tipc/bc_retruni
echo 0 > /proc/sys/net/tipc/bc_retruni

Default is '0', i.e. the broadcast retransmission still works as usual.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: add back link trace events
Tuong Lien [Tue, 26 May 2020 09:38:35 +0000 (16:38 +0700)]
tipc: add back link trace events

In the previous commit ("tipc: add Gap ACK blocks support for broadcast
link"), we have removed the following link trace events due to the code
changes:

- tipc_link_bc_ack
- tipc_link_retrans

This commit adds them back along with some minor changes to adapt to
the new code.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: introduce Gap ACK blocks for broadcast link
Tuong Lien [Tue, 26 May 2020 09:38:34 +0000 (16:38 +0700)]
tipc: introduce Gap ACK blocks for broadcast link

As achieved through commit 9195948fbf34 ("tipc: improve TIPC throughput
by Gap ACK blocks"), we apply the same mechanism for the broadcast link
as well. The 'Gap ACK blocks' data field in a 'PROTOCOL/STATE_MSG' will
consist of two parts built for both the broadcast and unicast types:

 31                       16 15                        0
+-------------+-------------+-------------+-------------+
|  bgack_cnt  |  ugack_cnt  |            len            |
+-------------+-------------+-------------+-------------+  -
|            gap            |            ack            |   |
+-------------+-------------+-------------+-------------+    > bc gacks
:                           :                           :   |
+-------------+-------------+-------------+-------------+  -
|            gap            |            ack            |   |
+-------------+-------------+-------------+-------------+    > uc gacks
:                           :                           :   |
+-------------+-------------+-------------+-------------+  -

which is "automatically" backward-compatible.

We also increase the max number of Gap ACK blocks to 128, allowing upto
64 blocks per type (total buffer size = 516 bytes).

Besides, the 'tipc_link_advance_transmq()' function is refactored which
is applicable for both the unicast and broadcast cases now, so some old
functions can be removed and the code is optimized.

With the patch, TIPC broadcast is more robust regardless of packet loss
or disorder, latency, ... in the underlying network. Its performance is
boost up significantly.
For example, experiment with a 5% packet loss rate results:

$ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000
real    0m 42.46s
user    0m 1.16s
sys     0m 17.67s

Without the patch:

$ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000
real    8m 27.94s
user    0m 0.55s
sys     0m 2.38s

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoqed: Add EDPM mode type for user-fw compatibility
Yuval Basson [Tue, 26 May 2020 06:41:20 +0000 (09:41 +0300)]
qed: Add EDPM mode type for user-fw compatibility

In older FW versions the completion flag was treated as the ack flag in
edpm messages. Expose the FW option of setting which mode the QP is in
by adding a flag to the qedr <-> qed API.

Flag is added for backward compatibility with libqedr.
This flag will be set by qedr after determining whether the libqedr is
using the updated version.

Fixes: f10939403352 ("qed: Add support for QP verbs")
Signed-off-by: Yuval Basson <yuval.bason@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: tcp_v4_err() icmp skb is named icmp_skb
Eric Dumazet [Tue, 26 May 2020 03:15:24 +0000 (20:15 -0700)]
tcp: tcp_v4_err() icmp skb is named icmp_skb

I missed the fact that tcp_v4_err() differs from tcp_v6_err().

After commit 4d1a2d9ec1c1 ("Rename skb to icmp_skb in tcp_v4_err()")
the skb argument has been renamed to icmp_skb only in one function.

I will in a future patch reconciliate these functions to avoid
this kind of confusion.

Fixes: 45af29ca761c ("tcp: allow traceroute -Mtcp for unpriv users")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobatman-adv: Revert "disable ethtool link speed detection when auto negotiation off"
Sven Eckelmann [Mon, 25 Nov 2019 09:46:50 +0000 (10:46 +0100)]
batman-adv: Revert "disable ethtool link speed detection when auto negotiation off"

The commit 8c46fcd78308 ("batman-adv: disable ethtool link speed detection
when auto negotiation off") disabled the usage of ethtool's link_ksetting
when auto negotation was enabled due to invalid values when used with
tun/tap virtual net_devices. According to the patch, automatic measurements
should be used for these kind of interfaces.

But there are major flaws with this argumentation:

* automatic measurements are not implemented
* auto negotiation has nothing to do with the validity of the retrieved
  values

The first point has to be fixed by a longer patch series. The "validity"
part of the second point must be addressed in the same patch series by
dropping the usage of ethtool's link_ksetting (thus always doing automatic
measurements over ethernet).

Drop the patch again to have more default values for various net_device
types/configurations. The user can still overwrite them using the
batadv_hardif's BATADV_ATTR_THROUGHPUT_OVERRIDE.

Reported-by: Matthias Schiffer <mschiffer@universe-factory.net>
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
4 years agoMerge branch 'r8169-sync-hw-config-for-few-chip-versions-with-r8168-vendor-driver'
David S. Miller [Tue, 26 May 2020 01:21:10 +0000 (18:21 -0700)]
Merge branch 'r8169-sync-hw-config-for-few-chip-versions-with-r8168-vendor-driver'

Heiner Kallweit says:

====================
r8169: sync hw config for few chip versions with r8168 vendor driver

Sync hw config for few chip versions with r8168 vendor driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: sync RTL8168f/RTL8411 hw config with vendor driver
Heiner Kallweit [Mon, 25 May 2020 17:52:27 +0000 (19:52 +0200)]
r8169: sync RTL8168f/RTL8411 hw config with vendor driver

Sync hw config for RTL8168f/RTL8411 with r8168 vendor driver.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: sync RTL8168evl hw config with vendor driver
Heiner Kallweit [Mon, 25 May 2020 17:50:38 +0000 (19:50 +0200)]
r8169: sync RTL8168evl hw config with vendor driver

Sync hw config for RTL8168evl with r8168 vendor driver.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: sync RTL8168h hw config with vendor driver
Heiner Kallweit [Mon, 25 May 2020 17:49:55 +0000 (19:49 +0200)]
r8169: sync RTL8168h hw config with vendor driver

Sync hw config for RTL8168h with r8168 vendor driver.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: sync RTL8168g hw config with vendor driver
Heiner Kallweit [Mon, 25 May 2020 17:49:15 +0000 (19:49 +0200)]
r8169: sync RTL8168g hw config with vendor driver

Sync hw config for RTL8168g with r8168 vendor driver.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'wireless-drivers-next-2020-05-25' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Tue, 26 May 2020 01:15:16 +0000 (18:15 -0700)]
Merge tag 'wireless-drivers-next-2020-05-25' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for v5.8

Second set of patches for v5.8. Lots of new features and new supported
hardware for mt76. Also rtw88 got new hardware support.

Major changes:

rtw88

* add support for Realtek 8723DE PCI adapter

* rename rtw88.ko/rtwpci.ko to rtw88_core.ko/rtw88_pci.ko

iwlwifi

* stop supporting swcrypto and bt_coex_active module parameters on
  mvm devices

* enable A-AMSDU in low latency

mt76

* new devices for mt76x0/mt76x2

* support for non-offload firmware on mt7663

* hw/sched scan support for mt7663

* mt7615/mt7663 MSI support

* TDLS support

* mt7603/mt7615 rate control fixes

* new driver for mt7915

* wowlan support for mt7663

* suspend/resume support for mt7663
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobridge: mrp: Fix out-of-bounds read in br_mrp_parse
Horatiu Vultur [Mon, 25 May 2020 09:55:41 +0000 (09:55 +0000)]
bridge: mrp: Fix out-of-bounds read in br_mrp_parse

The issue was reported by syzbot. When the function br_mrp_parse was
called with a valid net_bridge_port, the net_bridge was an invalid
pointer. Therefore the check br->stp_enabled could pass/fail
depending where it was pointing in memory.
The fix consists of setting the net_bridge pointer if the port is a
valid pointer.

Reported-by: syzbot+9c6f0f1f8e32223df9a4@syzkaller.appspotmail.com
Fixes: 6536993371fa ("bridge: mrp: Integrate MRP into the bridge")
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodrivers: ipa: print dev_err info accurately
Wang Wenhu [Mon, 25 May 2020 06:29:51 +0000 (23:29 -0700)]
drivers: ipa: print dev_err info accurately

Print certain name string instead of hard-coded "memory" for dev_err
output, which would be more accurate and helpful for debugging.

Signed-off-by: Wang Wenhu <wenhu.wang@vivo.com>
Cc: Alex Elder <elder@kernel.org>
Reviewed-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoptp_clock: Let the ADJ_OFFSET interface respect the ADJ_NANO flag for PHC devices.
Richard Cochran [Sun, 24 May 2020 18:27:10 +0000 (11:27 -0700)]
ptp_clock: Let the ADJ_OFFSET interface respect the ADJ_NANO flag for PHC devices.

In commit 184ecc9eb260d5a3bcdddc5bebd18f285ac004e9 ("ptp: Add adjphase
function to support phase offset control.") the PTP Hardware Clock
interface expanded to support the ADJ_OFFSET offset mode.  However,
the implementation did not respect the traditional yet pedantic
distinction between units of microseconds and nanoseconds signaled by
the ADJ_NANO flag.  This patch fixes the issue by adding logic to
handle that flag.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Reviewed-by: Vincent Cheng <vincent.cheng.xh@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: allow traceroute -Mtcp for unpriv users
Eric Dumazet [Sun, 24 May 2020 18:00:02 +0000 (11:00 -0700)]
tcp: allow traceroute -Mtcp for unpriv users

Unpriv users can use traceroute over plain UDP sockets, but not TCP ones.

$ traceroute -Mtcp 8.8.8.8
You do not have enough privileges to use this traceroute method.

$ traceroute -n -Mudp 8.8.8.8
traceroute to 8.8.8.8 (8.8.8.8), 30 hops max, 60 byte packets
 1  192.168.86.1  3.631 ms  3.512 ms  3.405 ms
 2  10.1.10.1  4.183 ms  4.125 ms  4.072 ms
 3  96.120.88.125  20.621 ms  19.462 ms  20.553 ms
 4  96.110.177.65  24.271 ms  25.351 ms  25.250 ms
 5  69.139.199.197  44.492 ms  43.075 ms  44.346 ms
 6  68.86.143.93  27.969 ms  25.184 ms  25.092 ms
 7  96.112.146.18  25.323 ms 96.112.146.22  25.583 ms 96.112.146.26  24.502 ms
 8  72.14.239.204  24.405 ms 74.125.37.224  16.326 ms  17.194 ms
 9  209.85.251.9  18.154 ms 209.85.247.55  14.449 ms 209.85.251.9  26.296 ms^C

We can easily support traceroute over TCP, by queueing an error message
into socket error queue.

Note that applications need to set IP_RECVERR/IPV6_RECVERR option to
enable this feature, and that the error message is only queued
while in SYN_SNT state.

socket(AF_INET6, SOCK_STREAM, IPPROTO_IP) = 3
setsockopt(3, SOL_IPV6, IPV6_RECVERR, [1], 4) = 0
setsockopt(3, SOL_SOCKET, SO_TIMESTAMP_OLD, [1], 4) = 0
setsockopt(3, SOL_IPV6, IPV6_UNICAST_HOPS, [5], 4) = 0
connect(3, {sa_family=AF_INET6, sin6_port=htons(8787), sin6_flowinfo=htonl(0),
        inet_pton(AF_INET6, "2002:a05:6608:297::", &sin6_addr), sin6_scope_id=0}, 28) = -1 EHOSTUNREACH (No route to host)
recvmsg(3, {msg_name={sa_family=AF_INET6, sin6_port=htons(8787), sin6_flowinfo=htonl(0),
        inet_pton(AF_INET6, "2002:a05:6608:297::", &sin6_addr), sin6_scope_id=0},
        msg_namelen=1024->28, msg_iov=[{iov_base="`\r\337\320\0004\6\1&\7\370\260\200\231\16\27\0\0\0\0\0\0\0\0 \2\n\5f\10\2\227"..., iov_len=1024}],
        msg_iovlen=1, msg_control=[{cmsg_len=32, cmsg_level=SOL_SOCKET, cmsg_type=SO_TIMESTAMP_OLD, cmsg_data={tv_sec=1590340680, tv_usec=272424}},
                                   {cmsg_len=60, cmsg_level=SOL_IPV6, cmsg_type=IPV6_RECVERR}],
        msg_controllen=96, msg_flags=MSG_ERRQUEUE}, MSG_ERRQUEUE) = 144

Suggested-by: Maciej Żenczykowski <maze@google.com
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobnx2x: allow bnx2x_bsc_read() to schedule
Eric Dumazet [Sat, 23 May 2020 19:46:49 +0000 (12:46 -0700)]
bnx2x: allow bnx2x_bsc_read() to schedule

bnx2x_warpcore_read_sfp_module_eeprom() can call bnx2x_bsc_read()
three times before giving up.

This causes latency blips of at least 31 ms (58 ms being reported
by our teams)

Convert the long lasting loops of udelay() to usleep_range() ones,
and breaks the loops on precise time tracking.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ariel Elior <aelior@marvell.com>
Cc: Sudarsana Kalluru <skalluru@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>