openwrt/staging/blogic.git
4 years agoieee802154: 6lowpan: remove unnecessary comparison
Yang Yingliang [Fri, 8 May 2020 03:52:08 +0000 (11:52 +0800)]
ieee802154: 6lowpan: remove unnecessary comparison

The type of dispatch is u8 which is always '<=' 0xff, so the
dispatch <= 0xff is always true, we can remove this comparison.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
4 years agonet/dst: use a smaller percpu_counter batch for dst entries accounting
Eric Dumazet [Fri, 8 May 2020 01:58:10 +0000 (18:58 -0700)]
net/dst: use a smaller percpu_counter batch for dst entries accounting

percpu_counter_add() uses a default batch size which is quite big
on platforms with 256 cpus. (2*256 -> 512)

This means dst_entries_get_fast() can be off by +/- 2*(nr_cpus^2)
(131072 on servers with 256 cpus)

Reduce the batch size to something more reasonable, and
add logic to ip6_dst_gc() to call dst_entries_get_slow()
before calling the _very_ expensive fib6_run_gc() function.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
4 years agonet: relax SO_TXTIME CAP_NET_ADMIN check
Eric Dumazet [Thu, 7 May 2020 17:05:39 +0000 (10:05 -0700)]
net: relax SO_TXTIME CAP_NET_ADMIN check

Now sch_fq has horizon feature, we want to allow QUIC/UDP applications
to use EDT model so that pacing can be offloaded to the kernel (sch_fq)
or the NIC.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'bonding-report-transmit-status-to-callers'
David S. Miller [Fri, 8 May 2020 01:11:07 +0000 (18:11 -0700)]
Merge branch 'bonding-report-transmit-status-to-callers'

Eric Dumazet says:

====================
bonding: report transmit status to callers

First patches cleanup netpoll, and make sure it provides tx status to its users.

Last patch changes bonding to not pretend packets were sent without error.

By providing more accurate status, TCP stack can avoid adding more
packets if the slave qdisc is already full.

This came while testing latest horizon feature in sch_fq, with
very low pacing rate flows, but should benefit hosts under stress.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobonding: propagate transmit status
Eric Dumazet [Thu, 7 May 2020 16:32:22 +0000 (09:32 -0700)]
bonding: propagate transmit status

Currently, bonding always returns NETDEV_TX_OK to its caller.

It is worth trying to be more accurate : TCP for instance
can have different recovery strategies if it can have more
precise status, if packet was dropped by slave qdisc.

This is especially important when host is under stress.

Signed-off-by: Eric Dumazet <edumazet@google.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>
4 years agonetpoll: accept NULL np argument in netpoll_send_skb()
Eric Dumazet [Thu, 7 May 2020 16:32:21 +0000 (09:32 -0700)]
netpoll: accept NULL np argument in netpoll_send_skb()

netpoll_send_skb() callers seem to leak skb if
the np pointer is NULL. While this should not happen, we
can make the code more robust.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetpoll: netpoll_send_skb() returns transmit status
Eric Dumazet [Thu, 7 May 2020 16:32:20 +0000 (09:32 -0700)]
netpoll: netpoll_send_skb() returns transmit status

Some callers want to know if the packet has been sent or
dropped, to inform upper stacks.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetpoll: move netpoll_send_skb() out of line
Eric Dumazet [Thu, 7 May 2020 16:32:19 +0000 (09:32 -0700)]
netpoll: move netpoll_send_skb() out of line

There is no need to inline this helper, as we intend to add more
code in this function.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetpoll: remove dev argument from netpoll_send_skb_on_dev()
Eric Dumazet [Thu, 7 May 2020 16:32:18 +0000 (09:32 -0700)]
netpoll: remove dev argument from netpoll_send_skb_on_dev()

netpoll_send_skb_on_dev() can get the device pointer directly from np->dev

Rename it to __netpoll_send_skb()

Following patch will move netpoll_send_skb() out-of-line.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: fix less than zero comparison with unsigned variable val
Colin Ian King [Thu, 7 May 2020 14:34:30 +0000 (15:34 +0100)]
net: phy: fix less than zero comparison with unsigned variable val

The unsigned variable val is being checked for an error by checking
if it is less than zero. This can never occur because val is unsigned.
Fix this by making val a plain int.

Addresses-Coverity: ("Unsigned compared against zero")
Fixes: bdbdac7649fa ("ethtool: provide UAPI for PHY master/slave configuration.")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: remove set but not used variables 'del_llc, del_llc_resp'
YueHaibing [Thu, 7 May 2020 14:24:06 +0000 (16:24 +0200)]
net/smc: remove set but not used variables 'del_llc, del_llc_resp'

Fixes gcc '-Wunused-but-set-variable' warning:

net/smc/smc_llc.c: In function 'smc_llc_cli_conf_link':
net/smc/smc_llc.c:753:31: warning:
 variable 'del_llc' set but not used [-Wunused-but-set-variable]
  struct smc_llc_msg_del_link *del_llc;
                               ^
net/smc/smc_llc.c: In function 'smc_llc_process_srv_delete_link':
net/smc/smc_llc.c:1311:33: warning:
 variable 'del_llc_resp' set but not used [-Wunused-but-set-variable]
    struct smc_llc_msg_del_link *del_llc_resp;
                                 ^

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: tcp_mark_head_lost is only valid for sack-tcp
zhang kai [Thu, 7 May 2020 03:08:30 +0000 (11:08 +0800)]
tcp: tcp_mark_head_lost is only valid for sack-tcp

so tcp_is_sack/reno checks are removed from tcp_mark_head_lost.

Signed-off-by: zhang kai <zhangkaiheb@126.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: remove newlines in NL_SET_ERR_MSG_MOD
Jacob Keller [Thu, 7 May 2020 00:58:27 +0000 (17:58 -0700)]
net: remove newlines in NL_SET_ERR_MSG_MOD

The NL_SET_ERR_MSG_MOD macro is used to report a string describing an
error message to userspace via the netlink extended ACK structure. It
should not have a trailing newline.

Add a cocci script which catches cases where the newline marker is
present. Using this script, fix the handful of cases which accidentally
included a trailing new line.

I couldn't figure out a way to get a patch mode working, so this script
only implements context, report, and org.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'ti-am65x-cpts-follow-up-dt-bindings-update'
David S. Miller [Fri, 8 May 2020 00:51:03 +0000 (17:51 -0700)]
Merge branch 'ti-am65x-cpts-follow-up-dt-bindings-update'

Grygorii Strashko says:

====================
net: ethernet: ti: am65x-cpts: follow up dt bindings update

This series is follow update for  TI A65x/J721E Common platform time sync (CPTS)
driver [1] to implement  DT bindings review comments from
Rob Herring <robh@kernel.org> [2].
 - "reg" and "compatible" properties are made required for CPTS DT nodes which
   also required to change K3 CPSW driver to use of_platform_device_create()
   instead of of_platform_populate() for proper CPTS and MDIO initialization
 - minor DT bindings format changes
 - K3 CPTS example added to K3 MCU CPSW bindings

[1] https://lwn.net/Articles/819313/
[2] https://lwn.net/ml/linux-kernel/20200505040419.GA8509@bogus/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoarm64: dts: ti: k3-am65/j721e-mcu: update cpts node
Grygorii Strashko [Wed, 6 May 2020 18:14:01 +0000 (21:14 +0300)]
arm64: dts: ti: k3-am65/j721e-mcu: update cpts node

Update CPTS node following DT binding update:
 - add reg and compatible properties
 - fix node name

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodt-binding: net: ti: am65x-cpts: make reg and compatible required
Grygorii Strashko [Wed, 6 May 2020 18:14:00 +0000 (21:14 +0300)]
dt-binding: net: ti: am65x-cpts: make reg and compatible required

This patch follows K3 CPTS review comments from Rob Herring
<robh@kernel.org>.
 - "reg" and "compatible" properties are required now
 - minor format changes
 - K3 CPTS example added to K3 MCU CPSW bindings

Cc: Rob Herring <robh@kernel.org>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: ti: am65-cpsw-nuss: use of_platform_device_create() for mdio
Grygorii Strashko [Wed, 6 May 2020 18:13:59 +0000 (21:13 +0300)]
net: ethernet: ti: am65-cpsw-nuss: use of_platform_device_create() for mdio

The MCU CPSW expected to populate only MDIO device, but follow up patches
will add "compatible" property to the MCU CPSW CPTS node which will cause
creation of CPTS device and MCU CPSW init failure. Hence, switch to use
of_platform_device_create() instead of of_platform_populate() for MDIO
device population.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hsr-hsr-code-refactoring'
David S. Miller [Fri, 8 May 2020 00:40:02 +0000 (17:40 -0700)]
Merge branch 'hsr-hsr-code-refactoring'

Taehee Yoo says:

====================
hsr: hsr code refactoring

There are some unnecessary routine in the hsr module.
This patch removes these routines.

The first patch removes incorrect comment.
The second patch removes unnecessary WARN_ONCE() macro.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodpaa2-eth: create a function to flush the XDP fds
Ioana Ciornei [Wed, 6 May 2020 17:47:17 +0000 (20:47 +0300)]
dpaa2-eth: create a function to flush the XDP fds

Create an independent function that takes a particular frame queue and
an array of frame descriptors and tries to enqueue them until it hits
the maximum number fo retries. The same function will be used in the
next patch also on the XDP_TX path.

Also, create the dpaa2_eth_xdp_fds structure to incorporate the array of
FDs as well as the number of FDs already populated.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agohsr: remove WARN_ONCE() in hsr_fill_frame_info()
Taehee Yoo [Wed, 6 May 2020 15:47:45 +0000 (15:47 +0000)]
hsr: remove WARN_ONCE() in hsr_fill_frame_info()

When VLAN frame is being sent, hsr calls WARN_ONCE() because hsr doesn't
support VLAN. But using WARN_ONCE() is overdoing.
Using netdev_warn_once() is enough.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosoc: fsl: dpio: properly compute the consumer index
Ioana Ciornei [Tue, 5 May 2020 20:14:29 +0000 (23:14 +0300)]
soc: fsl: dpio: properly compute the consumer index

Mask the consumer index before using it. Without this, we would be
writing frame descriptors beyond the ring size supported by the QBMAN
block.

Fixes: 3b2abda7d28c ("soc: fsl: dpio: Replace QMAN array mode with ring mode enqueue")
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Li Yang <leoyang.li@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'tc-gate-offload-for-SJA1105-DSA-switch'
David S. Miller [Fri, 8 May 2020 00:31:57 +0000 (17:31 -0700)]
Merge branch 'tc-gate-offload-for-SJA1105-DSA-switch'

Vladimir Oltean says:

====================
tc-gate offload for SJA1105 DSA switch

Expose the TTEthernet hardware features of the switch using standard
tc-flower actions: trap, drop, redirect and gate.

v1 was submitted at:
https://patchwork.ozlabs.org/project/netdev/cover/20200503211035.19363-1-olteanv@gmail.com/

v2 was submitted at:
https://patchwork.ozlabs.org/project/netdev/cover/20200503211035.19363-1-olteanv@gmail.com/

Changes in v3:
Made sure there are no compilation warnings when
CONFIG_NET_DSA_SJA1105_TAS or CONFIG_NET_DSA_SJA1105_VL are disabled.

Changes in v2:
Using a newly introduced dsa_port_from_netdev public helper.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodocs: net: dsa: sja1105: document intended usage of virtual links
Vladimir Oltean [Tue, 5 May 2020 19:20:57 +0000 (22:20 +0300)]
docs: net: dsa: sja1105: document intended usage of virtual links

Add some verbiage describing how the hardware features of the switch are
exposed to users through tc-flower.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: implement tc-gate using time-triggered virtual links
Vladimir Oltean [Tue, 5 May 2020 19:20:56 +0000 (22:20 +0300)]
net: dsa: sja1105: implement tc-gate using time-triggered virtual links

Restrict the TTEthernet hardware support on this switch to operate as
closely as possible to IEEE 802.1Qci as possible. This means that it can
perform PTP-time-based ingress admission control on streams identified
by {DMAC, VID, PCP}, which is useful when trying to ensure the
determinism of traffic scheduled via IEEE 802.1Qbv.

The oddity comes from the fact that in hardware (and in TTEthernet at
large), virtual links always need a full-blown action, including not
only the type of policing, but also the list of destination ports. So in
practice, a single tc-gate action will result in all packets getting
dropped. Additional actions (either "trap" or "redirect") need to be
specified in the same filter rule such that the conforming packets are
actually forwarded somewhere.

Apart from the VL Lookup, Policing and Forwarding tables which need to
be programmed for each flow (virtual link), the Schedule engine also
needs to be told to open/close the admission gates for each individual
virtual link. A fairly accurate (and detailed) description of how that
works is already present in sja1105_tas.c, since it is already used to
trigger the egress gates for the tc-taprio offload (IEEE 802.1Qbv). Key
point here, we remember that the schedule engine supports 8
"subschedules" (execution threads that iterate through the global
schedule in parallel, and that no 2 hardware threads must execute a
schedule entry at the same time). For tc-taprio, each egress port used
one of these 8 subschedules, leaving a total of 4 subschedules unused.
In principle we could have allocated 1 subschedule for the tc-gate
offload of each ingress port, but actually the schedules of all virtual
links installed on each ingress port would have needed to be merged
together, before they could have been programmed to hardware. So
simplify our life and just merge the entire tc-gate configuration, for
all virtual links on all ingress ports, into a single subschedule. Be
sure to check that against the usual hardware scheduling conflicts, and
program it to hardware alongside any tc-taprio subschedule that may be
present.

The following scenarios were tested:

1. Quantitative testing:

   tc qdisc add dev swp2 clsact
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate index 1 base-time 0 \
           sched-entry OPEN 1200 -1 -1 \
           sched-entry CLOSE 1200 -1 -1 \
           action trap

   ping 192.168.1.2 -f
   PING 192.168.1.2 (192.168.1.2) 56(84) bytes of data.
   .............................
   --- 192.168.1.2 ping statistics ---
   948 packets transmitted, 467 received, 50.7384% packet loss, time 9671ms

2. Qualitative testing (with a phase-aligned schedule - the clocks are
   synchronized by ptp4l, not shown here):

   Receiver (sja1105):

   tc qdisc add dev swp2 clsact
   now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate base-time ${base_time} \
           sched-entry OPEN  60000 -1 -1 \
           sched-entry CLOSE 40000 -1 -1 \
           action trap

   Sender (enetc):
   now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc qdisc add dev eno0 parent root taprio \
           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 \
           base-time ${base_time} \
           sched-entry S 01  50000 \
           sched-entry S 00  50000 \
           flags 2

   ping -A 192.168.1.1
   PING 192.168.1.1 (192.168.1.1): 56 data bytes
   ...
   ^C
   --- 192.168.1.1 ping statistics ---
   1425 packets transmitted, 1424 packets received, 0% packet loss
   round-trip min/avg/max = 0.322/0.361/0.990 ms

   And just for comparison, with the tc-taprio schedule deleted:

   ping -A 192.168.1.1
   PING 192.168.1.1 (192.168.1.1): 56 data bytes
   ...
   ^C
   --- 192.168.1.1 ping statistics ---
   33 packets transmitted, 19 packets received, 42% packet loss
   round-trip min/avg/max = 0.336/0.464/0.597 ms

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: support flow-based redirection via virtual links
Vladimir Oltean [Tue, 5 May 2020 19:20:55 +0000 (22:20 +0300)]
net: dsa: sja1105: support flow-based redirection via virtual links

Implement tc-flower offloads for redirect, trap and drop using
non-critical virtual links.

Commands which were tested to work are:

  # Send frames received on swp2 with a DA of 42:be:24:9b:76:20 to the
  # CPU and to swp3. This type of key (DA only) when the port's VLAN
  # awareness state is off.
  tc qdisc add dev swp2 clsact
  tc filter add dev swp2 ingress flower skip_sw dst_mac 42:be:24:9b:76:20 \
          action mirred egress redirect dev swp3 \
          action trap

  # Drop frames received on swp2 with a DA of 42:be:24:9b:76:20, a VID
  # of 100 and a PCP of 0.
  tc filter add dev swp2 ingress protocol 802.1Q flower skip_sw \
          dst_mac 42:be:24:9b:76:20 vlan_id 100 vlan_prio 0 action drop

Under the hood, all rules match on DMAC, VID and PCP, but when VLAN
filtering is disabled, those are set internally by the driver to the
port-based defaults. Because we would be put in an awkward situation if
the user were to change the VLAN filtering state while there are active
rules (packets would no longer match on the specified keys), we simply
deny changing vlan_filtering unless the list of flows offloaded via
virtual links is empty. Then the user can re-add new rules.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: make room for virtual link parsing in flower offload
Vladimir Oltean [Tue, 5 May 2020 19:20:54 +0000 (22:20 +0300)]
net: dsa: sja1105: make room for virtual link parsing in flower offload

Virtual links are a sja1105 hardware concept of executing various flow
actions based on a key extracted from the frame's DMAC, VID and PCP.

Currently the tc-flower offload code supports only parsing the DMAC if
that is the broadcast MAC address, and the VLAN PCP. Extract the key
parsing logic from the L2 policers functionality and move it into its
own function, after adding extra logic for matching on any DMAC and VID.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: add static tables for virtual links
Vladimir Oltean [Tue, 5 May 2020 19:20:53 +0000 (22:20 +0300)]
net: dsa: sja1105: add static tables for virtual links

This patch adds the register definitions for the:
- VL Lookup Table
- VL Policing Table
- VL Forwarding Table
- VL Forwarding Parameters Table

These are needed in order to perform TTEthernet operations: QoS
classification, flow-based policing and/or frame redirecting with the
switch.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: introduce a dsa_port_from_netdev public helper
Vladimir Oltean [Tue, 5 May 2020 19:20:52 +0000 (22:20 +0300)]
net: dsa: introduce a dsa_port_from_netdev public helper

As its implementation shows, this is synonimous with calling
dsa_slave_dev_check followed by dsa_slave_to_port, so it is quite simple
already and provides functionality which is already there.

However there is now a need for these functions outside dsa_priv.h, for
example in drivers that perform mirroring and redirection through
tc-flower offloads (they are given raw access to the flow_cls_offload
structure), where they need to call this function on act->dev.

But simply exporting dsa_slave_to_port would make it non-inline and
would result in an extra function call in the hotpath, as can be seen
for example in sja1105:

Before:

000006dc <sja1105_xmit>:
{
 6dc: e92d4ff0  push {r4, r5, r6, r7, r8, r9, sl, fp, lr}
 6e0: e1a04000  mov r4, r0
 6e4: e591958c  ldr r9, [r1, #1420] ; 0x58c <- Inline dsa_slave_to_port
 6e8: e1a05001  mov r5, r1
 6ec: e24dd004  sub sp, sp, #4
u16 tx_vid = dsa_8021q_tx_vid(dp->ds, dp->index);
 6f0: e1c901d8  ldrd r0, [r9, #24]
 6f4: ebfffffe  bl 0 <dsa_8021q_tx_vid>
6f4: R_ARM_CALL dsa_8021q_tx_vid
u8 pcp = netdev_txq_to_tc(netdev, queue_mapping);
 6f8: e1d416b0  ldrh r1, [r4, #96] ; 0x60
u16 tx_vid = dsa_8021q_tx_vid(dp->ds, dp->index);
 6fc: e1a08000  mov r8, r0

After:

000006e4 <sja1105_xmit>:
{
 6e4: e92d4ff0  push {r4, r5, r6, r7, r8, r9, sl, fp, lr}
 6e8: e1a04000  mov r4, r0
 6ec: e24dd004  sub sp, sp, #4
struct dsa_port *dp = dsa_slave_to_port(netdev);
 6f0: e1a00001  mov r0, r1
{
 6f4: e1a05001  mov r5, r1
struct dsa_port *dp = dsa_slave_to_port(netdev);
 6f8: ebfffffe  bl 0 <dsa_slave_to_port>
6f8: R_ARM_CALL dsa_slave_to_port
 6fc: e1a09000  mov r9, r0
u16 tx_vid = dsa_8021q_tx_vid(dp->ds, dp->index);
 700: e1c001d8  ldrd r0, [r0, #24]
 704: ebfffffe  bl 0 <dsa_8021q_tx_vid>
704: R_ARM_CALL dsa_8021q_tx_vid

Because we want to avoid possible performance regressions, introduce
this new function which is designed to be public.

Suggested-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: remove duplicate headers
Chen Zhou [Thu, 7 May 2020 13:26:39 +0000 (21:26 +0800)]
cxgb4: remove duplicate headers

Remove duplicate headers which are included twice.

Signed-off-by: Chen Zhou <chenzhou10@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'wireless-drivers-next-2020-05-07' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Thu, 7 May 2020 20:22:35 +0000 (13:22 -0700)]
Merge tag 'wireless-drivers-next-2020-05-07' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

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

First set of patches for v5.8. Changes all over, ath10k apparently
seeing most new features this time. rtw88 also had lots of changes due
to preparation for new hardware support.

In this pull request there's also a new macro to include/linux/iopoll:
read_poll_timeout_atomic(). This is needed by rtw88 for atomic
polling.

Major changes:

ath11k

* add debugfs file for testing ADDBA and DELBA

* add 802.11 encapsulation offload on hardware support

* add htt_peer_stats_reset debugfs file

ath10k

* enable VHT160 and VHT80+80 modes

* enable radar detection in secondary segment

* sdio: disable TX complete indication to improve throughput

* sdio: decrease power consumption

* sdio: add HTT TX bundle support to increase throughput

* sdio: add rx bitrate reporting

ath9k

* improvements to AR9002 calibration logic

carl9170

* remove buggy P2P_GO support

p54usb

* add support for AirVasT USB stick

rtw88

* add support for antenna configuration

ti wlcore

* add support for AES_CMAC cipher

iwlwifi

* support for a few new FW API versions

* new hw configs
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Add-QRTR-MHI-client-driver'
David S. Miller [Thu, 7 May 2020 20:21:12 +0000 (13:21 -0700)]
Merge branch 'Add-QRTR-MHI-client-driver'

Manivannan Sadhasivam says:

====================
Add QRTR MHI client driver

Here is the series adding MHI client driver support to Qualcomm IPC router
protocol. MHI is a newly added bus to kernel which is used to communicate to
external modems over a physical interface like PCI-E. This driver is used to
transfer the QMI messages between the host processor and external modems over
the "IPCR" channel.

For QRTR, this driver is just another driver acting as a transport layer like
SMD.

Currently this driver is needed to control the QCA6390 WLAN device from ath11k.
The ath11k MHI controller driver will take care of booting up QCA6390 and
bringing it to operating state. Later, this driver will be used to transfer QMI
messages over the MHI-IPCR channel.

The second patch of this series removes the ARCH_QCOM dependency for QRTR. This
is needed because the QRTR driver will be used with x86 machines as well to talk
to devices like QCA6390.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qrtr: Do not depend on ARCH_QCOM
Manivannan Sadhasivam [Thu, 7 May 2020 12:53:06 +0000 (18:23 +0530)]
net: qrtr: Do not depend on ARCH_QCOM

IPC Router protocol is also used by external modems for exchanging the QMI
messages. Hence, it doesn't always depend on Qualcomm platforms. One such
instance is the QCA6390 WLAN device connected to x86 machine.

Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qrtr: Add MHI transport layer
Manivannan Sadhasivam [Thu, 7 May 2020 12:53:05 +0000 (18:23 +0530)]
net: qrtr: Add MHI transport layer

MHI is the transport layer used for communicating to the external modems.
Hence, this commit adds MHI transport layer support to QRTR for
transferring the QMI messages over IPC Router.

Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovia-rhine: Add platform dependencies
Geert Uytterhoeven [Thu, 7 May 2020 11:42:05 +0000 (13:42 +0200)]
via-rhine: Add platform dependencies

The VIA Rhine Ethernet interface is only present on PCI devices or
VIA/WonderMedia VT8500/WM85xx SoCs.  Add platform dependencies to the
VIA_RHINE config symbol, to avoid asking the user about it when
configuring a kernel without PCI or VT8500/WM85xx support.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet:enetc: bug fix for qos sfi operate space after freed
Po Liu [Thu, 7 May 2020 10:57:38 +0000 (18:57 +0800)]
net:enetc: bug fix for qos sfi operate space after freed

'Dan Carpenter' reported:
This code frees "sfi" and then dereferences it on the next line:
>                 kfree(sfi);
>                 clear_bit(sfi->index, epsfp.psfp_sfi_bitmap);

This "sfi->index" should be "index".

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Po Liu <Po.Liu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: encx24j600: make encx24j600_hw_init() return void
Jason Yan [Thu, 7 May 2020 11:09:05 +0000 (19:09 +0800)]
net: encx24j600: make encx24j600_hw_init() return void

This function always return 0 now, we can make it return void to
simplify the code. This fixes the following coccicheck warning:

drivers/net/ethernet/microchip/encx24j600.c:609:5-8: Unneeded variable:
"ret". Return "0" on line 653

Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mlx4: remove unneeded variable "err" in mlx4_en_ethtool_add_mac_rule()
Jason Yan [Thu, 7 May 2020 11:08:57 +0000 (19:08 +0800)]
net: mlx4: remove unneeded variable "err" in mlx4_en_ethtool_add_mac_rule()

Fix the following coccicheck warning:

drivers/net/ethernet/mellanox/mlx4/en_ethtool.c:1396:5-8: Unneeded
variable: "err". Return "0" on line 1411

Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: tulip: de4x5: make PCI_signature() return void
Jason Yan [Thu, 7 May 2020 11:08:47 +0000 (19:08 +0800)]
net: tulip: de4x5: make PCI_signature() return void

This function always return 0 now, we can make it return void to
simplify the code. This fixes the following coccicheck warning:

drivers/net/ethernet/dec/tulip/de4x5.c:3908:11-17: Unneeded variable:
"status". Return "0" on line 3912

Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atheros: remove dead code in atl1c_resume()
Jason Yan [Thu, 7 May 2020 11:08:36 +0000 (19:08 +0800)]
net: atheros: remove dead code in atl1c_resume()

This code has been marked dead for nearly 10 years. Remove it.

Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agob43: remove dead function b43_rssinoise_postprocess()
Jason Yan [Thu, 7 May 2020 11:07:41 +0000 (19:07 +0800)]
b43: remove dead function b43_rssinoise_postprocess()

This function is dead for more than 10 years. Remove it.

Signed-off-by: Jason Yan <yanaijie@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: Make iproc_mdio_resume static
Zheng Zengkai [Thu, 7 May 2020 08:03:26 +0000 (16:03 +0800)]
net: phy: Make iproc_mdio_resume static

Fix sparse warnings:

drivers/net/phy/mdio-bcm-iproc.c:182:5: warning:
 symbol 'iproc_mdio_resume' was not declared. Should it be static?

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
David S. Miller [Thu, 7 May 2020 05:10:13 +0000 (22:10 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Conflicts were all overlapping changes.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 7 May 2020 03:53:22 +0000 (20:53 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix reference count leaks in various parts of batman-adv, from Xiyu
    Yang.

 2) Update NAT checksum even when it is zero, from Guillaume Nault.

 3) sk_psock reference count leak in tls code, also from Xiyu Yang.

 4) Sanity check TCA_FQ_CODEL_DROP_BATCH_SIZE netlink attribute in
    fq_codel, from Eric Dumazet.

 5) Fix panic in choke_reset(), also from Eric Dumazet.

 6) Fix VLAN accel handling in bnxt_fix_features(), from Michael Chan.

 7) Disallow out of range quantum values in sch_sfq, from Eric Dumazet.

 8) Fix crash in x25_disconnect(), from Yue Haibing.

 9) Don't pass pointer to local variable back to the caller in
    nf_osf_hdr_ctx_init(), from Arnd Bergmann.

10) Wireguard should use the ECN decap helper functions, from Toke
    Høiland-Jørgensen.

11) Fix command entry leak in mlx5 driver, from Moshe Shemesh.

12) Fix uninitialized variable access in mptcp's
    subflow_syn_recv_sock(), from Paolo Abeni.

13) Fix unnecessary out-of-order ingress frame ordering in macsec, from
    Scott Dial.

14) IPv6 needs to use a global serial number for dst validation just
    like ipv4, from David Ahern.

15) Fix up PTP_1588_CLOCK deps, from Clay McClure.

16) Missing NLM_F_MULTI flag in gtp driver netlink messages, from
    Yoshiyuki Kurauchi.

17) Fix a regression in that dsa user port errors should not be fatal,
    from Florian Fainelli.

18) Fix iomap leak in enetc driver, from Dejin Zheng.

19) Fix use after free in lec_arp_clear_vccs(), from Cong Wang.

20) Initialize protocol value earlier in neigh code paths when
    generating events, from Roman Mashak.

21) netdev_update_features() must be called with RTNL mutex in macsec
    driver, from Antoine Tenart.

22) Validate untrusted GSO packets even more strictly, from Willem de
    Bruijn.

23) Wireguard decrypt worker needs a cond_resched(), from Jason
    Donenfeld.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (111 commits)
  net: flow_offload: skip hw stats check for FLOW_ACTION_HW_STATS_DONT_CARE
  MAINTAINERS: put DYNAMIC INTERRUPT MODERATION in proper order
  wireguard: send/receive: use explicit unlikely branch instead of implicit coalescing
  wireguard: selftests: initalize ipv6 members to NULL to squelch clang warning
  wireguard: send/receive: cond_resched() when processing worker ringbuffers
  wireguard: socket: remove errant restriction on looping to self
  wireguard: selftests: use normal kernel stack size on ppc64
  net: ethernet: ti: am65-cpsw-nuss: fix irqs type
  ionic: Use debugfs_create_bool() to export bool
  net: dsa: Do not leave DSA master with NULL netdev_ops
  net: dsa: remove duplicate assignment in dsa_slave_add_cls_matchall_mirred
  net: stricter validation of untrusted gso packets
  seg6: fix SRH processing to comply with RFC8754
  net: mscc: ocelot: ANA_AUTOAGE_AGE_PERIOD holds a value in seconds, not ms
  net: dsa: ocelot: the MAC table on Felix is twice as large
  net: dsa: sja1105: the PTP_CLK extts input reacts on both edges
  selftests: net: tcp_mmap: fix SO_RCVLOWAT setting
  net: hsr: fix incorrect type usage for protocol variable
  net: macsec: fix rtnl locking issue
  net: mvpp2: cls: Prevent buffer overflow in mvpp2_ethtool_cls_rule_del()
  ...

4 years agonet: flow_offload: skip hw stats check for FLOW_ACTION_HW_STATS_DONT_CARE
Pablo Neira Ayuso [Wed, 6 May 2020 18:34:50 +0000 (20:34 +0200)]
net: flow_offload: skip hw stats check for FLOW_ACTION_HW_STATS_DONT_CARE

This patch adds FLOW_ACTION_HW_STATS_DONT_CARE which tells the driver
that the frontend does not need counters, this hw stats type request
never fails. The FLOW_ACTION_HW_STATS_DISABLED type explicitly requests
the driver to disable the stats, however, if the driver cannot disable
counters, it bails out.

TCA_ACT_HW_STATS_* maintains the 1:1 mapping with FLOW_ACTION_HW_STATS_*
except by disabled which is mapped to FLOW_ACTION_HW_STATS_DISABLED
(this is 0 in tc). Add tc_act_hw_stats() to perform the mapping between
TCA_ACT_HW_STATS_* and FLOW_ACTION_HW_STATS_*.

Fixes: 319a1d19471e ("flow_offload: check for basic action hw stats type")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMAINTAINERS: put DYNAMIC INTERRUPT MODERATION in proper order
Lukas Bulwahn [Wed, 6 May 2020 20:29:06 +0000 (22:29 +0200)]
MAINTAINERS: put DYNAMIC INTERRUPT MODERATION in proper order

Commit 9b038086f06b ("docs: networking: convert DIM to RST") added a new
file entry to DYNAMIC INTERRUPT MODERATION to the end, and not following
alphabetical order.

So, ./scripts/checkpatch.pl -f MAINTAINERS complains:

  WARNING: Misordered MAINTAINERS entry - list file patterns in alphabetic
  order
  #5966: FILE: MAINTAINERS:5966:
  +F:      lib/dim/
  +F:      Documentation/networking/net_dim.rst

Reorder the file entries to keep MAINTAINERS nicely ordered.

Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'wireguard-fixes'
David S. Miller [Thu, 7 May 2020 03:03:48 +0000 (20:03 -0700)]
Merge branch 'wireguard-fixes'

Jason A. Donenfeld says:

====================
wireguard fixes for 5.7-rc5

With Ubuntu and Debian having backported this into their kernels, we're
finally seeing testing from places we hadn't seen prior, which is nice.
With that comes more fixes:

1) The CI for PPC64 was running with extremely small stacks for 64-bit,
   causing spurious crashes in surprising places.

2) There's was an old leftover routing loop restriction, which no longer
   makes sense given the queueing architecture, and was causing problems
   for people who really did want nested routing.

3) Not yielding our kthread on CONFIG_PREEMPT_VOLUNTARY systems caused
   RCU stalls and other issues, reported by Wang Jian, with the fix
   suggested by Sultan Alsawaf.

4) Clang spewed warnings in a selftest for CONFIG_IPV6=n, reported by
   Arnd Bergmann.

5) A complicated if statement was simplified to an assignment while also
   making the likely/unlikely hinting more correct and simple, and
   increasing readability, suggested by Sultan.

Patches (2) and (3) have Fixes: lines and are probably good candidates
for stable.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: send/receive: use explicit unlikely branch instead of implicit coalescing
Jason A. Donenfeld [Wed, 6 May 2020 21:33:06 +0000 (15:33 -0600)]
wireguard: send/receive: use explicit unlikely branch instead of implicit coalescing

It's very unlikely that send will become true. It's nearly always false
between 0 and 120 seconds of a session, and in most cases becomes true
only between 120 and 121 seconds before becoming false again. So,
unlikely(send) is clearly the right option here.

What happened before was that we had this complex boolean expression
with multiple likely and unlikely clauses nested. Since this is
evaluated left-to-right anyway, the whole thing got converted to
unlikely. So, we can clean this up to better represent what's going on.

The generated code is the same.

Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: selftests: initalize ipv6 members to NULL to squelch clang warning
Jason A. Donenfeld [Wed, 6 May 2020 21:33:05 +0000 (15:33 -0600)]
wireguard: selftests: initalize ipv6 members to NULL to squelch clang warning

Without setting these to NULL, clang complains in certain
configurations that have CONFIG_IPV6=n:

In file included from drivers/net/wireguard/ratelimiter.c:223:
drivers/net/wireguard/selftest/ratelimiter.c:173:34: error: variable 'skb6' is uninitialized when used here [-Werror,-Wuninitialized]
                ret = timings_test(skb4, hdr4, skb6, hdr6, &test_count);
                                               ^~~~
drivers/net/wireguard/selftest/ratelimiter.c:123:29: note: initialize the variable 'skb6' to silence this warning
        struct sk_buff *skb4, *skb6;
                                   ^
                                    = NULL
drivers/net/wireguard/selftest/ratelimiter.c:173:40: error: variable 'hdr6' is uninitialized when used here [-Werror,-Wuninitialized]
                ret = timings_test(skb4, hdr4, skb6, hdr6, &test_count);
                                                     ^~~~
drivers/net/wireguard/selftest/ratelimiter.c:125:22: note: initialize the variable 'hdr6' to silence this warning
        struct ipv6hdr *hdr6;
                            ^

We silence this warning by setting the variables to NULL as the warning
suggests.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: send/receive: cond_resched() when processing worker ringbuffers
Jason A. Donenfeld [Wed, 6 May 2020 21:33:04 +0000 (15:33 -0600)]
wireguard: send/receive: cond_resched() when processing worker ringbuffers

Users with pathological hardware reported CPU stalls on CONFIG_
PREEMPT_VOLUNTARY=y, because the ringbuffers would stay full, meaning
these workers would never terminate. That turned out not to be okay on
systems without forced preemption, which Sultan observed. This commit
adds a cond_resched() to the bottom of each loop iteration, so that
these workers don't hog the core. Note that we don't need this on the
napi poll worker, since that terminates after its budget is expended.

Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reported-by: Wang Jian <larkwang@gmail.com>
Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: socket: remove errant restriction on looping to self
Jason A. Donenfeld [Wed, 6 May 2020 21:33:03 +0000 (15:33 -0600)]
wireguard: socket: remove errant restriction on looping to self

It's already possible to create two different interfaces and loop
packets between them. This has always been possible with tunnels in the
kernel, and isn't specific to wireguard. Therefore, the networking stack
already needs to deal with that. At the very least, the packet winds up
exceeding the MTU and is discarded at that point. So, since this is
already something that happens, there's no need to forbid the not very
exceptional case of routing a packet back to the same interface; this
loop is no different than others, and we shouldn't special case it, but
rather rely on generic handling of loops in general. This also makes it
easier to do interesting things with wireguard such as onion routing.

At the same time, we add a selftest for this, ensuring that both onion
routing works and infinite routing loops do not crash the kernel. We
also add a test case for wireguard interfaces nesting packets and
sending traffic between each other, as well as the loop in this case
too. We make sure to send some throughput-heavy traffic for this use
case, to stress out any possible recursion issues with the locks around
workqueues.

Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: selftests: use normal kernel stack size on ppc64
Jason A. Donenfeld [Wed, 6 May 2020 21:33:02 +0000 (15:33 -0600)]
wireguard: selftests: use normal kernel stack size on ppc64

While at some point it might have made sense to be running these tests
on ppc64 with 4k stacks, the kernel hasn't actually used 4k stacks on
64-bit powerpc in a long time, and more interesting things that we test
don't really work when we deviate from the default (16k). So, we stop
pushing our luck in this commit, and return to the default instead of
the minimum.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: ti: am65-cpsw-nuss: fix irqs type
Grygorii Strashko [Tue, 5 May 2020 16:31:26 +0000 (19:31 +0300)]
net: ethernet: ti: am65-cpsw-nuss: fix irqs type

The K3 INTA driver, which is source TX/RX IRQs for CPSW NUSS, defines IRQs
triggering type as EDGE by default, but triggering type for CPSW NUSS TX/RX
IRQs has to be LEVEL as the EDGE triggering type may cause unnecessary IRQs
triggering and NAPI scheduling for empty queues. It was discovered with
RT-kernel.

Fix it by explicitly specifying CPSW NUSS TX/RX IRQ type as
IRQF_TRIGGER_HIGH.

Fixes: 93a76530316a ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver")
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodsa: sja1105: dynamically allocate stats structure
Arnd Bergmann [Tue, 5 May 2020 15:38:19 +0000 (17:38 +0200)]
dsa: sja1105: dynamically allocate stats structure

The addition of sja1105_port_status_ether structure into the
statistics causes the frame size to go over the warning limit:

drivers/net/dsa/sja1105/sja1105_ethtool.c:421:6: error: stack frame size of 1104 bytes in function 'sja1105_get_ethtool_stats' [-Werror,-Wframe-larger-than=]

Use dynamic allocation to avoid this.

Fixes: 336aa67bd027 ("net: dsa: sja1105: show more ethtool statistics counters for P/Q/R/S")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: Use debugfs_create_bool() to export bool
Geert Uytterhoeven [Tue, 5 May 2020 13:28:09 +0000 (15:28 +0200)]
ionic: Use debugfs_create_bool() to export bool

Currently bool ionic_cq.done_color is exported using
debugfs_create_u8(), which requires a cast, preventing further compiler
checks.

Fix this by switching to debugfs_create_bool(), and dropping the cast.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'ethtool-master-slave'
David S. Miller [Thu, 7 May 2020 00:45:45 +0000 (17:45 -0700)]
Merge branch 'ethtool-master-slave'

Oleksij Rempel says:

====================
provide support for PHY master/slave configuration

changes v6:
- use NL_SET_ERR_MSG_ATTR in ethnl_update_linkmodes
- add sanity checks in the ioctl interface
- use bool for ethnl_validate_master_slave_cfg()

changes v5:
- set MASTER_SLAVE_CFG_UNSUPPORTED as default value
- send a netlink error message on validation error
- more code fixes

changes v4:
- rename port_mode to master_slave
- move validation code to net/ethtool/linkmodes.c
- add UNSUPPORTED state and avoid sending unsupported fields
- more formatting and naming fixes
- tja11xx: support only force mode
- tja11xx: mark state as unsupported

changes v3:
- provide separate field for config and state.
- make state rejected on set
- add validation

changes v2:
- change names. Use MASTER_PREFERRED instead of MULTIPORT
- configure master/slave only on request. Default configuration can be
  provided by PHY or eeprom
- status and configuration to the user space.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: tja11xx: add support for master-slave configuration
Oleksij Rempel [Tue, 5 May 2020 06:35:06 +0000 (08:35 +0200)]
net: phy: tja11xx: add support for master-slave configuration

The TJA11xx PHYs have a vendor specific Master/Slave configuration bit,
which is not compatible with IEEE 803.2-2018 spec for 100Base-T1
devices. So, provide a custom config_ange call back to solve this
problem.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethtool: provide UAPI for PHY master/slave configuration.
Oleksij Rempel [Tue, 5 May 2020 06:35:05 +0000 (08:35 +0200)]
ethtool: provide UAPI for PHY master/slave configuration.

This UAPI is needed for BroadR-Reach 100BASE-T1 devices. Due to lack of
auto-negotiation support, we needed to be able to configure the
MASTER-SLAVE role of the port manually or from an application in user
space.

The same UAPI can be used for 1000BASE-T or MultiGBASE-T devices to
force MASTER or SLAVE role. See IEEE 802.3-2018:
22.2.4.3.7 MASTER-SLAVE control register (Register 9)
22.2.4.3.8 MASTER-SLAVE status register (Register 10)
40.5.2 MASTER-SLAVE configuration resolution
45.2.1.185.1 MASTER-SLAVE config value (1.2100.14)
45.2.7.10 MultiGBASE-T AN control 1 register (Register 7.32)

The MASTER-SLAVE role affects the clock configuration:

-------------------------------------------------------------------------------
When the  PHY is configured as MASTER, the PMA Transmit function shall
source TX_TCLK from a local clock source. When configured as SLAVE, the
PMA Transmit function shall source TX_TCLK from the clock recovered from
data stream provided by MASTER.

iMX6Q                     KSZ9031                XXX
------\                /-----------\        /------------\
      |                |           |        |            |
 MAC  |<----RGMII----->| PHY Slave |<------>| PHY Master |
      |<--- 125 MHz ---+-<------/  |        | \          |
------/                \-----------/        \------------/
                                               ^
                                                \-TX_TCLK

-------------------------------------------------------------------------------

Since some clock or link related issues are only reproducible in a
specific MASTER-SLAVE-role, MAC and PHY configuration, it is beneficial
to provide generic (not 100BASE-T1 specific) interface to the user space
for configuration flexibility and trouble shooting.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'chcr-next'
David S. Miller [Thu, 7 May 2020 00:44:11 +0000 (17:44 -0700)]
Merge branch 'chcr-next'

Devulapally Shiva Krishna says:

====================
Crypto/chcr: Fix issues regarding algorithm implementation in driver

The following series of patches fixes the issues which came during
self-tests with CONFIG_CRYPTO_MANAGER_EXTRA_TESTS enabled.

Patch 1: Fixes gcm(aes) hang issue and rfc4106-gcm encryption issue.
Patch 2: Fixes ctr, cbc, xts and rfc3686-ctr extra test failures.
Patch 3: Fixes ccm(aes) extra test failures.
Patch 4: Added support for 48 byte-key_len in aes_xts.
Patch 5: fix for hmac(sha) extra test failure.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoCrypto/chcr: fix for hmac(sha) test fails
Devulapally Shiva Krishna [Tue, 5 May 2020 03:12:57 +0000 (08:42 +0530)]
Crypto/chcr: fix for hmac(sha) test fails

The hmac(sha) test fails for a zero length source text data.
For hmac(sha) minimum length of the data must be of block-size.
So fix this by including the data_len for the last block.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoCrypto/chcr: support for 48 byte key_len in aes-xts
Devulapally Shiva Krishna [Tue, 5 May 2020 03:12:56 +0000 (08:42 +0530)]
Crypto/chcr: support for 48 byte key_len in aes-xts

Added support for 48 byte key length for aes-xts.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoCrypto/chcr: fix for ccm(aes) failed test
Devulapally Shiva Krishna [Tue, 5 May 2020 03:12:55 +0000 (08:42 +0530)]
Crypto/chcr: fix for ccm(aes) failed test

The ccm(aes) test fails when req->assoclen > ~240bytes.

The problem is the value assigned to auth_offset is wrong.
As auth_offset is unsigned char, it can take max value as 255.
So fix it by making it unsigned int.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoCrypto/chcr: fix ctr, cbc, xts and rfc3686-ctr failed tests
Devulapally Shiva Krishna [Tue, 5 May 2020 03:12:54 +0000 (08:42 +0530)]
Crypto/chcr: fix ctr, cbc, xts and rfc3686-ctr failed tests

This solves the following issues observed during self test when
CONFIG_CRYPTO_MANAGER_EXTRA_TESTS is enabled.

1. Added fallback for cbc, ctr and rfc3686 if req->nbytes is zero
and for xts added a fallback case if req->nbytes is not multiple of 16.

2. In case of cbc-aes, solved wrong iv update. When
chcr_cipher_fallback() is called, used req->info pointer instead of
reqctx->iv.

3. In cbc-aes decryption there was a wrong result. This occurs when
chcr_cipher_fallback() is called from chcr_handle_cipher_resp().
In the fallback function iv(req->info) used is wrongly updated.
So use the initial iv for this case.

4)In case of ctr-aes encryption observed wrong result. In adjust_ctr_overflow()
there is condition which checks if ((bytes / AES_BLOCK_SIZE) > c),
where c is the number of blocks which can be processed without iv overflow,
but for the above bytes (req->nbytes < 32 , not a multiple of 16) this
condition fails and the 2nd block is corrupted as it requires the rollover iv.
So added a '=' condition in this to take care of this.

5)In rfc3686-ctr there was wrong result observed. This occurs when
chcr_cipher_fallback() is called from chcr_handle_cipher_resp().
Here also copying initial_iv in init_iv pointer for handling the fallback
case correctly.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoCrypto/chcr: fix gcm-aes and rfc4106-gcm failed tests
Devulapally Shiva Krishna [Tue, 5 May 2020 03:12:53 +0000 (08:42 +0530)]
Crypto/chcr: fix gcm-aes and rfc4106-gcm failed tests

This patch fixes two issues observed during self tests with
CONFIG_CRYPTO_MANAGER_EXTRA_TESTS enabled.

1. gcm(aes) hang issue , that happens during decryption.
2. rfc4106-gcm-aes-chcr encryption unexpectedly succeeded.

For gcm-aes decryption , authtag is not mapped due to
sg_nents_for_len(upto size: assoclen+ cryptlen - authsize).
So fix it by dma_mapping authtag.
Also replaced sg_nents() to sg_nents_for_len() in case of aead_dma_unmap().

For rfc4106-gcm-aes-chcr, used crypto_ipsec_check_assoclen() for checking
the validity of assoclen.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ipa-kill-endpoint-stop-workaround'
David S. Miller [Thu, 7 May 2020 00:41:42 +0000 (17:41 -0700)]
Merge branch 'net-ipa-kill-endpoint-stop-workaround'

Alex Elder says:

====================
net: ipa: kill endpoint stop workaround

It turns out that a workaround that performs a small DMA operation
between retried attempts to stop a GSI channel is not needed for any
supported hardware.  The hardware quirk that required the extra DMA
operation was fixed after IPA v3.1.  So this series gets rid of that
workaround code, along with some other code that was only present to
support it.

NOTE:  This series depends on (and includes/duplicates) another patch
       that has already been committed in the net tree:
         713b6ebb4c37 net: ipa: fix a bug in ipa_endpoint_stop()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: kill ipa_cmd_dma_task_32b_addr_add()
Alex Elder [Mon, 4 May 2020 23:53:45 +0000 (18:53 -0500)]
net: ipa: kill ipa_cmd_dma_task_32b_addr_add()

A recent commit removed the only use of ipa_cmd_dma_task_32b_addr_add().
This function (and the IPA immediate command it implements) is no
longer needed, so get rid of it, along with all of the definitions
associated with it.  Isolate its removal in a commit so it can be
easily added back again if needed.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: kill ipa_endpoint_stop()
Alex Elder [Mon, 4 May 2020 23:53:44 +0000 (18:53 -0500)]
net: ipa: kill ipa_endpoint_stop()

The previous commit made ipa_endpoint_stop() be a trivial wrapper
around gsi_channel_stop().  Since it no longer does anything
special, just open-code it in the three places it's used.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: don't retry in ipa_endpoint_stop()
Alex Elder [Mon, 4 May 2020 23:53:43 +0000 (18:53 -0500)]
net: ipa: don't retry in ipa_endpoint_stop()

The only reason ipa_endpoint_stop() had a retry loop was that the
just-removed workaround required an IPA DMA command to occur between
attempts.  The gsi_channel_stop() call that implements the stop does
its own retry loop, to cover a channel's transition from started to
stop-in-progress to stopped state.

Get rid of the unnecessary retry loop in ipa_endpoint_stop().

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: get rid of workaround in ipa_endpoint_stop()
Alex Elder [Mon, 4 May 2020 23:53:42 +0000 (18:53 -0500)]
net: ipa: get rid of workaround in ipa_endpoint_stop()

In ipa_endpoint_stop(), a workaround is used for IPA version 3.5.1
where a 1-byte DMA request is issued between GSI channel stop
retries.

It turns out that this workaround is only required for IPA versions
3.1 and 3.2, and we don't support those.  So remove the call to
ipa_endpoint_stop_rx_dma() in that function.  That leaves that
function unused, so get rid of it.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: fix a bug in ipa_endpoint_stop()
Alex Elder [Mon, 4 May 2020 23:53:41 +0000 (18:53 -0500)]
net: ipa: fix a bug in ipa_endpoint_stop()

In ipa_endpoint_stop(), for TX endpoints we set the number of retries
to 0.  When we break out of the loop, retries being 0 means we return
EIO rather than the value of ret (which should be 0).

Fix this by using a non-zero retry count for both RX and TX
channels, and just break out of the loop after calling
gsi_channel_stop() for TX channels.  This way only RX channels
will retry, and the retry count will be non-zero at the end
for TX channels (so the proper value gets returned).

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit 713b6ebb4c376b3fb65fdceb3b59e401c93248f9)
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ipa-kill-endpoint-delay-mode-workaround'
David S. Miller [Thu, 7 May 2020 00:38:06 +0000 (17:38 -0700)]
Merge branch 'net-ipa-kill-endpoint-delay-mode-workaround'

Alex Elder says:

====================
net: ipa: kill endpoint delay mode workaround

A "delay mode" feature was put in place to work around a problem
where packets could passed to the modem before it was ready to
handle them.  That problem no longer exists, and we don't need the
workaround any more so get rid of it.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: remove endpoint delay mode feature
Alex Elder [Mon, 4 May 2020 23:37:13 +0000 (18:37 -0500)]
net: ipa: remove endpoint delay mode feature

A "delay mode" feature was put in place to work around a problem
that was observed during development of the upstream IPA driver.  It
used TX endpoint "delay mode" in order to prevent transmitting
packets toward the modem before it was ready.

A race condition that would explain the problem has long since been
fixed, and we have concluded that the "delay mode" feature is no
longer required.  So get rid of it.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: introduce ipa_endpoint_program_suspend()
Alex Elder [Mon, 4 May 2020 23:37:12 +0000 (18:37 -0500)]
net: ipa: introduce ipa_endpoint_program_suspend()

Create a new helper function that encapsulates enabling or disabling
suspend on an RX endpoint.  It returns the previous state of the
endpoint (true means suspend mode was enabled).

Create another function that handles enabling or disabling delay mode
on a TX endpoint.  Delay mode does not work correctly on IPA version
4.2, so we don't currently use it (and shouldn't).

We only set delay mode in one case, and although we don't expect an
endpoint to already be in delay mode, it doesn't really matter if it
was.  So the delay function doesn't return a value.

Stop issuing warnings if the previous suspend or delay mode state
differs from what is expected.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: have ipa_endpoint_init_ctrl() return previous state
Alex Elder [Mon, 4 May 2020 23:37:11 +0000 (18:37 -0500)]
net: ipa: have ipa_endpoint_init_ctrl() return previous state

Change ipa_endpoint_init_ctrl() so it returns the previous state
(whether suspend or delay mode was enabled) rather than indicating
whether the request caused a change in state.  This makes it easier
to understand what's happening where called.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ipa-limit-special-reset-handling'
David S. Miller [Thu, 7 May 2020 00:36:04 +0000 (17:36 -0700)]
Merge branch 'net-ipa-limit-special-reset-handling'

Alex Elder says:

====================
net: ipa: limit special reset handling

Some special handling done during channel reset should only be done
for IPA hardare version 3.5.1.  This series generalizes the meaning
of a flag passed to indicate special behavior, then has the special
handling be used only when appropriate.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: only reset channel twice for IPA v3.5.1
Alex Elder [Mon, 4 May 2020 23:30:03 +0000 (18:30 -0500)]
net: ipa: only reset channel twice for IPA v3.5.1

In gsi_channel_reset(), RX channels are subjected to two consecutive
CHANNEL_RESET commands.  This workaround should only be used for IPA
version 3.5.1, and for newer hardware "can lead to unwanted behavior."

Only issue the second CHANNEL_RESET command for legacy hardware.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: rename db_enable flag
Alex Elder [Mon, 4 May 2020 23:30:02 +0000 (18:30 -0500)]
net: ipa: rename db_enable flag

In several places, a Boolean flag is used in the GSI code to
indicate whether the "doorbell engine" should be enabled or not
when a channel is configured.  This is basically done to abstract
this property from the IPA version; the GSI code doesn't otherwise
"know" what the IPA hardware version is.  The doorbell engine is
enabled only for IPA v3.5.1, not for IPA v4.0 and later.

The next patch makes another change that affects behavior during
channel reset (which also involves programming the channel).  It
also distinguishes IPA v3.5.1 hardware from newer hardware.

Rather than creating another flag whose value matches the "db_enable"
value, just rename "db_enable" to be "legacy" so it can be used to
signal more than just the special doorbell handling.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: Do not leave DSA master with NULL netdev_ops
Florian Fainelli [Mon, 4 May 2020 20:18:06 +0000 (13:18 -0700)]
net: dsa: Do not leave DSA master with NULL netdev_ops

When ndo_get_phys_port_name() for the CPU port was added we introduced
an early check for when the DSA master network device in
dsa_master_ndo_setup() already implements ndo_get_phys_port_name(). When
we perform the teardown operation in dsa_master_ndo_teardown() we would
not be checking that cpu_dp->orig_ndo_ops was successfully allocated and
non-NULL initialized.

With network device drivers such as virtio_net, this leads to a NPD as
soon as the DSA switch hanging off of it gets torn down because we are
now assigning the virtio_net device's netdev_ops a NULL pointer.

Fixes: da7b9e9b00d4 ("net: dsa: Add ndo_get_phys_port_name() for CPU port")
Reported-by: Allen Pais <allen.pais@oracle.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Allen Pais <allen.pais@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: remove duplicate assignment in dsa_slave_add_cls_matchall_mirred
Vladimir Oltean [Mon, 4 May 2020 19:58:56 +0000 (22:58 +0300)]
net: dsa: remove duplicate assignment in dsa_slave_add_cls_matchall_mirred

This was caused by a poor merge conflict resolution on my side. The
"act = &cls->rule->action.entries[0];" assignment was already present in
the code prior to the patch mentioned below.

Fixes: e13c2075280e ("net: dsa: refactor matchall mirred action to separate function")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'tcp-minor-adjustments-for-low-pacing-rates'
David S. Miller [Thu, 7 May 2020 00:29:38 +0000 (17:29 -0700)]
Merge branch 'tcp-minor-adjustments-for-low-pacing-rates'

Eric Dumazet says:

====================
tcp: minor adjustments for low pacing rates

After pacing horizon addition, we have to adjust how we arm rto
timer, otherwise we might freeze very low pacing rate flows.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: defer xmit timer reset in tcp_xmit_retransmit_queue()
Eric Dumazet [Mon, 4 May 2020 18:27:50 +0000 (11:27 -0700)]
tcp: defer xmit timer reset in tcp_xmit_retransmit_queue()

As hinted in prior change ("tcp: refine tcp_pacing_delay()
for very low pacing rates"), it is probably best arming
the xmit timer only when all the packets have been scheduled,
rather than when the head of rtx queue has been re-sent.

This does matter for flows having extremely low pacing rates,
since their tp->tcp_wstamp_ns could be far in the future.

Note that the regular xmit path has a stronger limit
in tcp_small_queue_check(), meaning it is less likely to
go beyond the pacing horizon.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: refine tcp_pacing_delay() for very low pacing rates
Eric Dumazet [Mon, 4 May 2020 18:27:49 +0000 (11:27 -0700)]
tcp: refine tcp_pacing_delay() for very low pacing rates

With the addition of horizon feature to sch_fq, we noticed some
suboptimal behavior of extremely low pacing rate TCP flows, especially
when TCP is not aware of a drop happening in lower stacks.

Back in commit 3f80e08f40cd ("tcp: add tcp_reset_xmit_timer() helper"),
tcp_pacing_delay() was added to estimate an extra delay to add to standard
rto timers.

This patch removes the skb argument from this helper and
tcp_reset_xmit_timer() because it makes more sense to simply
consider the time at which next packet is allowed to be sent,
instead of the time of whatever packet has been sent.

This avoids arming RTO timer too soon and removes
spurious horizon drops.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoarm64: dts: sdm845: add IPA iommus property
Alex Elder [Mon, 4 May 2020 18:13:50 +0000 (13:13 -0500)]
arm64: dts: sdm845: add IPA iommus property

Add an "iommus" property to the IPA node in "sdm845.dtsi".  It is
required because there are two regions of memory the IPA accesses
through an SMMU.  The next few patches define and map those regions.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stricter validation of untrusted gso packets
Willem de Bruijn [Mon, 4 May 2020 16:48:54 +0000 (12:48 -0400)]
net: stricter validation of untrusted gso packets

Syzkaller again found a path to a kernel crash through bad gso input:
a packet with transport header extending beyond skb_headlen(skb).

Tighten validation at kernel entry:

- Verify that the transport header lies within the linear section.

    To avoid pulling linux/tcp.h, verify just sizeof tcphdr.
    tcp_gso_segment will call pskb_may_pull (th->doff * 4) before use.

- Match the gso_type against the ip_proto found by the flow dissector.

Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoseg6: fix SRH processing to comply with RFC8754
Ahmed Abdelsalam [Mon, 4 May 2020 14:42:11 +0000 (14:42 +0000)]
seg6: fix SRH processing to comply with RFC8754

The Segment Routing Header (SRH) which defines the SRv6 dataplane is defined
in RFC8754.

RFC8754 (section 4.1) defines the SR source node behavior which encapsulates
packets into an outer IPv6 header and SRH. The SR source node encodes the
full list of Segments that defines the packet path in the SRH. Then, the
first segment from list of Segments is copied into the Destination address
of the outer IPv6 header and the packet is sent to the first hop in its path
towards the destination.

If the Segment list has only one segment, the SR source node can omit the SRH
as he only segment is added in the destination address.

RFC8754 (section 4.1.1) defines the Reduced SRH, when a source does not
require the entire SID list to be preserved in the SRH. A reduced SRH does
not contain the first segment of the related SR Policy (the first segment is
the one already in the DA of the IPv6 header), and the Last Entry field is
set to n-2, where n is the number of elements in the SR Policy.

RFC8754 (section 4.3.1.1) defines the SRH processing and the logic to
validate the SRH (S09, S10, S11) which works for both reduced and
non-reduced behaviors.

This patch updates seg6_validate_srh() to validate the SRH as per RFC8754.

Signed-off-by: Ahmed Abdelsalam <ahabdels@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'FDB-fixes-for-Felix-and-Ocelot-switches'
David S. Miller [Thu, 7 May 2020 00:15:38 +0000 (17:15 -0700)]
Merge branch 'FDB-fixes-for-Felix-and-Ocelot-switches'

Vladimir Oltean says:

====================
FDB fixes for Felix and Ocelot switches

This series fixes the following problems:
- Dynamically learnt addresses never expiring (neither for Ocelot nor
  for Felix)
- Half of the FDB not visible in 'bridge fdb show' (for Felix only)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: ANA_AUTOAGE_AGE_PERIOD holds a value in seconds, not ms
Vladimir Oltean [Sun, 3 May 2020 22:20:27 +0000 (01:20 +0300)]
net: mscc: ocelot: ANA_AUTOAGE_AGE_PERIOD holds a value in seconds, not ms

One may notice that automatically-learnt entries 'never' expire, even
though the bridge configures the address age period at 300 seconds.

Actually the value written to hardware corresponds to a time interval
1000 times higher than intended, i.e. 83 hours.

Fixes: a556c76adc05 ("net: mscc: Add initial Ocelot switch support")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Faineli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: ocelot: the MAC table on Felix is twice as large
Vladimir Oltean [Sun, 3 May 2020 22:20:26 +0000 (01:20 +0300)]
net: dsa: ocelot: the MAC table on Felix is twice as large

When running 'bridge fdb dump' on Felix, sometimes learnt and static MAC
addresses would appear, sometimes they wouldn't.

Turns out, the MAC table has 4096 entries on VSC7514 (Ocelot) and 8192
entries on VSC9959 (Felix), so the existing code from the Ocelot common
library only dumped half of Felix's MAC table. They are both organized
as a 4-way set-associative TCAM, so we just need a single variable
indicating the correct number of rows.

Fixes: 56051948773e ("net: dsa: ocelot: add driver for Felix switch family")
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 agoMerge branch 'timer-add-fsleep-for-flexible-sleeping'
David S. Miller [Thu, 7 May 2020 00:03:34 +0000 (17:03 -0700)]
Merge branch 'timer-add-fsleep-for-flexible-sleeping'
Heiner Kallweit says:

====================
timer: add fsleep for flexible sleeping

Sleeping for a certain amount of time requires use of different
functions, depending on the time period.
Documentation/timers/timers-howto.rst explains when to use which
function, and also checkpatch checks for some potentially
problematic cases.

So let's create a helper that automatically chooses the appropriate
sleep function -> fsleep(), for flexible sleeping
Not sure why such a helper doesn't exist yet, or where the pitfall is,
because it's a quite obvious idea.

If the delay is a constant, then the compiler should be able to ensure
that the new helper doesn't create overhead. If the delay is not
constant, then the new helper can save some code.

First user is the r8169 network driver. If nothing speaks against it,
then this series could go through the netdev tree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: use fsleep in polling functions
Heiner Kallweit [Fri, 1 May 2020 21:29:12 +0000 (23:29 +0200)]
r8169: use fsleep in polling functions

Use new flexible sleep function fsleep() to merge the udelay and msleep
polling functions. We can safely do this because no polling function
is used in atomic context in this driver.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotimer: add fsleep for flexible sleeping
Heiner Kallweit [Fri, 1 May 2020 21:27:21 +0000 (23:27 +0200)]
timer: add fsleep for flexible sleeping

Sleeping for a certain amount of time requires use of different
functions, depending on the time period.
Documentation/timers/timers-howto.rst explains when to use which
function, and also checkpatch checks for some potentially
problematic cases.

So let's create a helper that automatically chooses the appropriate
sleep function -> fsleep(), for flexible sleeping

If the delay is a constant, then the compiler should be able to ensure
that the new helper doesn't create overhead. If the delay is not
constant, then the new helper can save some code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: Implement draft-ietf-6man-rfc4941bis
Fernando Gont [Fri, 1 May 2020 03:51:47 +0000 (00:51 -0300)]
ipv6: Implement draft-ietf-6man-rfc4941bis

Implement the upcoming rev of RFC4941 (IPv6 temporary addresses):
https://tools.ietf.org/html/draft-ietf-6man-rfc4941bis-09

* Reduces the default Valid Lifetime to 2 days
  The number of extra addresses employed when Valid Lifetime was
  7 days exacerbated the stress caused on network
  elements/devices. Additionally, the motivation for temporary
  addresses is indeed privacy and reduced exposure. With a
  default Valid Lifetime of 7 days, an address that becomes
  revealed by active communication is reachable and exposed for
  one whole week. The only use case for a Valid Lifetime of 7
  days could be some application that is expecting to have long
  lived connections. But if you want to have a long lived
  connections, you shouldn't be using a temporary address in the
  first place. Additionally, in the era of mobile devices, general
  applications should nevertheless be prepared and robust to
  address changes (e.g. nodes swap wifi <-> 4G, etc.)

* Employs different IIDs for different prefixes
  To avoid network activity correlation among addresses configured
  for different prefixes

* Uses a simpler algorithm for IID generation
  No need to store "history" anywhere

Signed-off-by: Fernando Gont <fgont@si6networks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'tag-chrome-platform-fixes-for-v5.7-rc5' of git://git.kernel.org/pub/scm...
Linus Torvalds [Wed, 6 May 2020 23:40:14 +0000 (16:40 -0700)]
Merge tag 'tag-chrome-platform-fixes-for-v5.7-rc5' of git://git./linux/kernel/git/chrome-platform/linux

Pull chrome platform fix from Benson Leung:
 "Fix a resource allocation issue in cros_ec_sensorhub.c"

* tag 'tag-chrome-platform-fixes-for-v5.7-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/chrome-platform/linux:
  platform/chrome: cros_ec_sensorhub: Allocate sensorhub resource before claiming sensors

4 years agonet: dsa: sja1105: the PTP_CLK extts input reacts on both edges
Vladimir Oltean [Wed, 6 May 2020 17:48:13 +0000 (20:48 +0300)]
net: dsa: sja1105: the PTP_CLK extts input reacts on both edges

It looks like the sja1105 external timestamping input is not as generic
as we thought. When fed a signal with 50% duty cycle, it will timestamp
both the rising and the falling edge. When fed a short pulse signal,
only the timestamp of the falling edge will be seen in the PTPSYNCTS
register, because that of the rising edge had been overwritten. So the
moral is: don't feed it short pulse inputs.

Luckily this is not a complete deal breaker, as we can still work with
1 Hz square waves. But the problem is that the extts polling period was
not dimensioned enough for this input signal. If we leave the period at
half a second, we risk losing timestamps due to jitter in the measuring
process. So we need to increase it to 4 times per second.

Also, the very least we can do to inform the user is to deny any other
flags combination than with PTP_RISING_EDGE and PTP_FALLING_EDGE both
set.

Fixes: 747e5eb31d59 ("net: dsa: sja1105: configure the PTP_CLK pin as EXT_TS or PER_OUT")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: net: tcp_mmap: fix SO_RCVLOWAT setting
Eric Dumazet [Wed, 6 May 2020 16:21:15 +0000 (09:21 -0700)]
selftests: net: tcp_mmap: fix SO_RCVLOWAT setting

Since chunk_size is no longer an integer, we can not
use it directly as an argument of setsockopt().

This patch should fix tcp_mmap for Big Endian kernels.

Fixes: 597b01edafac ("selftests: net: avoid ptl lock contention in tcp_mmap")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hsr: fix incorrect type usage for protocol variable
Murali Karicheri [Wed, 6 May 2020 15:41:07 +0000 (11:41 -0400)]
net: hsr: fix incorrect type usage for protocol variable

Fix following sparse checker warning:-

net/hsr/hsr_slave.c:38:18: warning: incorrect type in assignment (different base types)
net/hsr/hsr_slave.c:38:18:    expected unsigned short [unsigned] [usertype] protocol
net/hsr/hsr_slave.c:38:18:    got restricted __be16 [usertype] h_proto
net/hsr/hsr_slave.c:39:25: warning: restricted __be16 degrades to integer
net/hsr/hsr_slave.c:39:57: warning: restricted __be16 degrades to integer

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'add-phy-shared-storage'
David S. Miller [Wed, 6 May 2020 21:53:30 +0000 (14:53 -0700)]
Merge branch 'add-phy-shared-storage'

Michael Walle says:

====================
add phy shared storage

Introduce the concept of a shared PHY storage which can be used by some
QSGMII PHYs to ease initialization and access to global per-package
registers.

Changes since v2:
 - restore page to standard after reading the base address in the mscc
   driver, thanks Antoine.

Changes since v1:
 - fix typos and add a comment, thanks Florian.
 - check for "addr < 0" in phy_package_join()
 - remove multiple blank lines and make "checkpatch.pl --strict" happy

Changes since RFC:
 - check return code of kzalloc()
 - fix local variable ordering (reverse christmas tree)
 - add priv_size argument to phy_package_join()
 - add Tested-by tag, thanks Vladimir.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: mscc: use phy_package_shared
Michael Walle [Wed, 6 May 2020 14:53:15 +0000 (16:53 +0200)]
net: phy: mscc: use phy_package_shared

Use the new phy_package_shared common storage to ease the package
initialization and to access the global registers.

Signed-off-by: Michael Walle <michael@walle.cc>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
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 agonet: phy: bcm54140: use phy_package_shared
Michael Walle [Wed, 6 May 2020 14:53:14 +0000 (16:53 +0200)]
net: phy: bcm54140: use phy_package_shared

Use the new phy_package_shared common storage to ease the package
initialization and to access the global registers.

Signed-off-by: Michael Walle <michael@walle.cc>
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 agonet: phy: add concept of shared storage for PHYs
Michael Walle [Wed, 6 May 2020 14:53:13 +0000 (16:53 +0200)]
net: phy: add concept of shared storage for PHYs

There are packages which contain multiple PHY devices, eg. a quad PHY
transceiver. Provide functions to allocate and free shared storage.

Usually, a quad PHY contains global registers, which don't belong to any
PHY. Provide convenience functions to access these registers.

Signed-off-by: Michael Walle <michael@walle.cc>
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 agoRevert "crypto: chelsio - Inline single pdu only"
Ayush Sawal [Wed, 6 May 2020 14:47:19 +0000 (20:17 +0530)]
Revert "crypto: chelsio - Inline single pdu only"

This reverts commit 27c6feb0fb33a665a746346e76714826a5be5d10.

For ipsec offload the chelsio's ethernet driver expects a single mtu
sized packet.

But when ipsec traffic is running using iperf, most of the packets in
that traffic are gso packets(large sized skbs) because GSO is enabled by
default in TCP, due to this commit 0a6b2a1dc2a2 ("tcp: switch to GSO
being always on"), so chcr_ipsec_offload_ok() receives a gso
skb(with gso_size non zero).

Due to the check in chcr_ipsec_offload_ok(), this function returns false
for most of the packet, then ipsec offload is skipped and the skb goes
out taking the coprocessor path which reduces the bandwidth for inline
ipsec.

If this check is removed then for most of the packets(large sized skbs)
the chcr_ipsec_offload_ok() returns true and then as GSO is on, the
segmentation of the packet happens in the kernel and then finally the
driver_xmit is called, which receives a segmented mtu sized packet which
is what the driver expects for ipsec offload. So this case becomes
unnecessary here, therefore removing it.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>