platform/kernel/linux-starfive.git
3 years agodrivers/net/wan/hdlc_fr: Improvements to the code of pvc_xmit
Xie He [Sat, 3 Oct 2020 22:41:05 +0000 (15:41 -0700)]
drivers/net/wan/hdlc_fr: Improvements to the code of pvc_xmit

1. Keep the code for the normal (non-error) flow at the lowest
indentation level. And use "goto drop" for all error handling.

2. Replace code that pads short Ethernet frames with a "__skb_pad" call.

3. Change "dev_kfree_skb" to "kfree_skb" in error handling code.
"kfree_skb" is the correct function to call when dropping an skb due to
an error. "dev_kfree_skb", which is an alias of "consume_skb", is for
dropping skbs normally (not due to an error).

Cc: Krzysztof Halasa <khc@pm.waw.pl>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Xie He <xie.he.0141@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bnxt_en-net-next-updates'
David S. Miller [Sun, 4 Oct 2020 21:41:05 +0000 (14:41 -0700)]
Merge branch 'bnxt_en-net-next-updates'

Michael Chan says:

====================
bnxt_en: net-next updates.

This series starts off with the usual update of the firmware interface
spec.  A new firmware status bit in the interface will be used in patch
add the infrastructure to read the firmware status very early during
driver probe and this will allow patch #4 to do the recovery if needed.

The rest of the patches add improvements to the current RX reset
logic by localizing the reset to the affected RX ring only and to
reset only if firmware has determined that the RX ring is in permanent
error state.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Eliminate unnecessary RX resets.
Michael Chan [Sun, 4 Oct 2020 19:23:01 +0000 (15:23 -0400)]
bnxt_en: Eliminate unnecessary RX resets.

Currently, the driver will schedule RX ring reset when we get a buffer
error in the RX completion record.  These RX buffer errors can be due
to normal out-of-buffer conditions or a permanent error in the RX
ring.  Because the driver cannot distinguish between these 2
conditions, we assume all these buffer errors require reset.

This is very disruptive when it is just a normal out-of-buffer
condition.  Newer firmware will now monitor the rings for the permanent
failure and will send a notification to the driver when it happens.
This allows the driver to reset only when such a notification is
received.  In environments where we have predominently out-of-buffer
conditions, we now can avoid these unnecessary resets.

Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Reduce unnecessary message log during RX errors.
Michael Chan [Sun, 4 Oct 2020 19:23:00 +0000 (15:23 -0400)]
bnxt_en: Reduce unnecessary message log during RX errors.

There is logic in the RX path to detect unexpected handles in the
RX completion.  We'll print a warning and schedule a reset.  The
next expected handle is then set to 0xffff which is guaranteed to
not match any valid handle.  This will force all remaining packets in
the ring to be discarded before the reset.  There can be hundreds of
these packets remaining in the ring and there is no need to print the
warnings for these forced errors.

Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Add a software counter for RX ring reset.
Michael Chan [Sun, 4 Oct 2020 19:22:59 +0000 (15:22 -0400)]
bnxt_en: Add a software counter for RX ring reset.

Add a per ring rx_resets counter to count these RX resets.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Implement RX ring reset in response to buffer errors.
Michael Chan [Sun, 4 Oct 2020 19:22:58 +0000 (15:22 -0400)]
bnxt_en: Implement RX ring reset in response to buffer errors.

On some older chips, it is necessary to do a reset when we get buffer
errors associated with an RX ring.  These buffer errors may become
frequent if the RX ring underruns under heavy traffic.  The current
code does a global reset of all reasources when this happens.  This
works but creates a big disruption of all rings when one RX ring is
having problem.  This patch implements a localized RX ring reset of
just the RX ring having the issue.  All other rings including all
TX rings will not be affected by this single RX ring reset.

Only the older chips prior to the P5 class supports this reset.
Because it is not a global reset, packets may still be arriving
while we are calling firmware to reset that ring.  We need to be
sure that we don't post any buffers during this time while the
ring is undergoing reset.  After firmware completes successfully,
the ring will be in the reset state with no buffers and we can start
filling it with new buffers and posting them.

Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Refactor bnxt_init_one_rx_ring().
Michael Chan [Sun, 4 Oct 2020 19:22:57 +0000 (15:22 -0400)]
bnxt_en: Refactor bnxt_init_one_rx_ring().

bnxt_init_one_rx_ring() includes logic to initialize the BDs for one RX
ring and to allocate the buffers.  Separate the allocation logic into a
new bnxt_alloc_one_rx_ring() function.  The allocation function will be
used later to allocate new buffers for one specified RX ring when we
reset that RX ring.

Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Refactor bnxt_free_rx_skbs().
Michael Chan [Sun, 4 Oct 2020 19:22:56 +0000 (15:22 -0400)]
bnxt_en: Refactor bnxt_free_rx_skbs().

bnxt_free_rx_skbs() frees all the allocated buffers and SKBs for
every RX ring.  Refactor this function by calling a new function
bnxt_free_one_rx_ring_skbs() to free these buffers on one specified
RX ring at a time.  This is preparation work for resetting one RX
ring during run-time.

Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Log FW health status info, if reset is aborted.
Michael Chan [Sun, 4 Oct 2020 19:22:55 +0000 (15:22 -0400)]
bnxt_en: Log FW health status info, if reset is aborted.

If firmware does not come out of reset, log FW health status info
to provide more information on firmware status.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: perform no master recovery during startup
Edwin Peer [Sun, 4 Oct 2020 19:22:54 +0000 (15:22 -0400)]
bnxt_en: perform no master recovery during startup

The NS3 SoC platforms require assistance from the OP-TEE to recover
firmware if a crash occurs while no driver is bound. The
CRASHED_NO_MASTER condition is recorded in the firmware status register
during the crash to indicate when driver intervension is needed to
coordinate a firmware reload. This condition is detected during early
driver initialization in order to effect a firmware fastboot on
supported platforms when necessary.

Reviewed-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: log firmware status on firmware init failure
Edwin Peer [Sun, 4 Oct 2020 19:22:53 +0000 (15:22 -0400)]
bnxt_en: log firmware status on firmware init failure

Firmware now supports device independent discovery of the status
register location. This status register can provide more detailed
information about firmware errors, especially if problems occur
before the HWRM interface is functioning. Attempt to map this
register if it is present and report the firmware status on firmware
init failures.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: refactor bnxt_alloc_fw_health()
Edwin Peer [Sun, 4 Oct 2020 19:22:52 +0000 (15:22 -0400)]
bnxt_en: refactor bnxt_alloc_fw_health()

The allocator for the firmware health structure conflates allocation
and capability checks, limiting the reusability of the code. This patch
separates out the capability check and disablement and improves the
warning message to better describe the consequences of an allocation
failure.

Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Update firmware interface spec to 1.10.1.68.
Vasundhara Volam [Sun, 4 Oct 2020 19:22:51 +0000 (15:22 -0400)]
bnxt_en: Update firmware interface spec to 1.10.1.68.

Main changes is to extend hwrm_nvm_get_dev_info_output() for stored
firmware versions and a new flag is added to fw_status_reg.

Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'mv88e6xxx-Add-per-port-devlink-regions'
David S. Miller [Sun, 4 Oct 2020 21:38:53 +0000 (14:38 -0700)]
Merge branch 'mv88e6xxx-Add-per-port-devlink-regions'

Andrew Lunn says:

====================
mv88e6xxx: Add per port devlink regions

This patchset extends devlink regions to support per port regions, and
them makes use of them to support the ports of the mv88e6xxx switches.

root@rap:~# devlink region show
mdio_bus/gpio-0:00/global1: size 64 snapshot []
mdio_bus/gpio-0:00/global2: size 64 snapshot []
mdio_bus/gpio-0:00/atu: size 49152 snapshot []
mdio_bus/gpio-0:00/0/port: size 64 snapshot []
mdio_bus/gpio-0:00/1/port: size 64 snapshot []
mdio_bus/gpio-0:00/2/port: size 64 snapshot []
mdio_bus/gpio-0:00/3/port: size 64 snapshot []
mdio_bus/gpio-0:00/4/port: size 64 snapshot []
mdio_bus/gpio-0:00/5/port: size 64 snapshot []
mdio_bus/gpio-0:00/6/port: size 64 snapshot []
mdio_bus/gpio-0:00/7/port: size 64 snapshot []
mdio_bus/gpio-0:00/8/port: size 64 snapshot []
mdio_bus/gpio-0:00/9/port: size 64 snapshot []
mdio_bus/gpio-0:00/10/port: size 64 snapshot []

root@rap:~# devlink region new mdio_bus/gpio-0:00/1/port snapshot 42
root@rap:~# devlink region dump mdio_bus/gpio-0:00/1/port snapshot 42
0000000000000000 4f 1e 3e 20 00 01 01 39 3f 05 00 00 fd 07 00 00
0000000000000010 80 00 01 00 00 00 00 00 00 00 00 00 00 00 00 91
0000000000000020 00 00 00 00 00 00 00 00 00 00 00 00 22 00 00 00
0000000000000030 07 3e 00 00 00 00 00 80 00 00 00 00 00 00 5b 00

In order to support all ports of the switch, a new devlink flavour has
been added for unused ports:

mdio_bus/gpio-0:00/0: type notset flavour unused splittable false
mdio_bus/gpio-0:00/1: type notset flavour cpu port 1 splittable false
mdio_bus/gpio-0:00/2: type eth netdev red flavour physical port 2 splittable fae
mdio_bus/gpio-0:00/3: type eth netdev blue flavour physical port 3 splittable fe
mdio_bus/gpio-0:00/4: type eth netdev green flavour physical port 4 splittable e
mdio_bus/gpio-0:00/5: type notset flavour unused splittable false
mdio_bus/gpio-0:00/6: type notset flavour unused splittable false
mdio_bus/gpio-0:00/7: type notset flavour unused splittable false
mdio_bus/gpio-0:00/8: type eth netdev waic0 flavour physical port 8 splittable e
mdio_bus/gpio-0:00/9: type notset flavour unused splittable false
mdio_bus/gpio-0:00/10: type notset flavour unused splittable false

The DSA core now creates the devlink port instances earlier, so that
the driver setup function can make use of them.

v3:
Whitespace cleanup
Added justification for devlink unused flavour
Added Tested-by, Reviewed-by:
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: mv88e6xxx: Add per port devlink regions
Andrew Lunn [Sun, 4 Oct 2020 16:12:57 +0000 (18:12 +0200)]
net: dsa: mv88e6xxx: Add per port devlink regions

Add a devlink region to return the per port registers.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: Add helper for converting devlink port to ds and port
Andrew Lunn [Sun, 4 Oct 2020 16:12:56 +0000 (18:12 +0200)]
net: dsa: Add helper for converting devlink port to ds and port

Hide away from DSA drivers how devlink works.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: Add devlink port regions support to DSA
Andrew Lunn [Sun, 4 Oct 2020 16:12:55 +0000 (18:12 +0200)]
net: dsa: Add devlink port regions support to DSA

Allow DSA drivers to make use of devlink port regions, via simple
wrappers.

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: devlink: Add support for port regions
Andrew Lunn [Sun, 4 Oct 2020 16:12:54 +0000 (18:12 +0200)]
net: devlink: Add support for port regions

Allow regions to be registered to a devlink port. The same netlink API
is used, but the port index is provided to indicate when a region is a
port region as opposed to a device region.

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: Register devlink ports before calling DSA driver setup()
Andrew Lunn [Sun, 4 Oct 2020 16:12:53 +0000 (18:12 +0200)]
net: dsa: Register devlink ports before calling DSA driver setup()

DSA drivers want to create regions on devlink ports as well as the
devlink device instance, in order to export registers and other tables
per port. To keep all this code together in the drivers, have the
devlink ports registered early, so the setup() method can setup both
device and port devlink regions.

v3:
Remove dp->setup
Move common code out of switch statement.
Fix wrong goto

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: Make use of devlink port flavour unused
Andrew Lunn [Sun, 4 Oct 2020 16:12:52 +0000 (18:12 +0200)]
net: dsa: Make use of devlink port flavour unused

If a port is unused, still create a devlink port for it, but set the
flavour to unused. This allows us to attach devlink regions to the
port, etc.

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: devlink: Add unused port flavour
Andrew Lunn [Sun, 4 Oct 2020 16:12:51 +0000 (18:12 +0200)]
net: devlink: Add unused port flavour

Not all ports of a switch need to be used, particularly in embedded
systems. Add a port flavour for ports which physically exist in the
switch, but are not connected to the front panel etc, and so are
unused. By having unused ports present in devlink, it gives a more
accurate representation of the hardware. It also allows regions to be
associated to such ports, so allowing, for example, to determine
unused ports are correctly powered off, or to compare probable reset
defaults of unused ports to used ports experiences issues.

Actually registering unused ports and setting the flavour to unused is
optional. The DSA core will register all such switch ports, but such
ports are expected to be limited in number. Bigger ASICs may decide
not to list unused ports.

v2:
Expand the description about why it is useful

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Sun, 4 Oct 2020 21:35:53 +0000 (14:35 -0700)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next:

1) Rename 'searched' column to 'clashres' in conntrack /proc/ stats
   to amend a recent patch, from Florian Westphal.

2) Remove unused nft_data_debug(), from YueHaibing.

3) Remove unused definitions in IPVS, also from YueHaibing.

4) Fix user data memleak in tables and objects, this is also amending
   a recent patch, from Jose M. Guisado.

5) Use nla_memdup() to allocate user data in table and objects, also
   from Jose M. Guisado

6) User data support for chains, from Jose M. Guisado

7) Remove unused definition in nf_tables_offload, from YueHaibing.

8) Use kvzalloc() in ip_set_alloc(), from Vasily Averin.

9) Fix false positive reported by lockdep in nfnetlink mutexes,
   from Florian Westphal.

10) Extend fast variant of cmp for neq operation, from Phil Sutter.

11) Implement fast bitwise variant, also from Phil Sutter.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetfilter: nf_tables: Implement fast bitwise expression
Phil Sutter [Thu, 1 Oct 2020 16:57:44 +0000 (18:57 +0200)]
netfilter: nf_tables: Implement fast bitwise expression

A typical use of bitwise expression is to mask out parts of an IP
address when matching on the network part only. Optimize for this common
use with a fast variant for NFT_BITWISE_BOOL-type expressions operating
on 32bit-sized values.

Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nf_tables: Enable fast nft_cmp for inverted matches
Phil Sutter [Fri, 2 Oct 2020 13:50:56 +0000 (15:50 +0200)]
netfilter: nf_tables: Enable fast nft_cmp for inverted matches

Add a boolean indicating NFT_CMP_NEQ. To include it into the match
decision, it is sufficient to XOR it with the data comparison's result.

While being at it, store the mask that is calculated during expression
init and free the eval routine from having to recalculate it each time.

Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nfnetlink: place subsys mutexes in distinct lockdep classes
Florian Westphal [Fri, 2 Oct 2020 11:51:29 +0000 (13:51 +0200)]
netfilter: nfnetlink: place subsys mutexes in distinct lockdep classes

From time to time there are lockdep reports similar to this one:

 WARNING: possible circular locking dependency detected
 ------------------------------------------------------
 000000004f61aa56 (&table[i].mutex){+.+.}, at: nfnl_lock [nfnetlink]
 but task is already holding lock:
 [..] (&net->nft.commit_mutex){+.+.}, at: nf_tables_valid_genid [nf_tables]
 which lock already depends on the new lock.
 the existing dependency chain (in reverse order) is:
 -> #1 (&net->nft.commit_mutex){+.+.}:
 [..]
        nf_tables_valid_genid+0x18/0x60 [nf_tables]
        nfnetlink_rcv_batch+0x24c/0x620 [nfnetlink]
        nfnetlink_rcv+0x110/0x140 [nfnetlink]
        netlink_unicast+0x12c/0x1e0
 [..]
        sys_sendmsg+0x18/0x40
        linux_sparc_syscall+0x34/0x44
 -> #0 (&table[i].mutex){+.+.}:
 [..]
        nfnl_lock+0x24/0x40 [nfnetlink]
        ip_set_nfnl_get_byindex+0x19c/0x280 [ip_set]
        set_match_v1_checkentry+0x14/0xc0 [xt_set]
        xt_check_match+0x238/0x260 [x_tables]
        __nft_match_init+0x160/0x180 [nft_compat]
 [..]
        sys_sendmsg+0x18/0x40
        linux_sparc_syscall+0x34/0x44
 other info that might help us debug this:
  Possible unsafe locking scenario:
        CPU0                    CPU1
        ----                    ----
   lock(&net->nft.commit_mutex);
                                lock(&table[i].mutex);
                                lock(&net->nft.commit_mutex);
   lock(&table[i].mutex);

Lockdep considers this an ABBA deadlock because the different nfnl subsys
mutexes reside in the same lockdep class, but this is a false positive.

CPU1 table[i] refers to the nftables subsys mutex, whereas CPU1 locks
the ipset subsys mutex.

Yi Che reported a similar lockdep splat, this time between ipset and
ctnetlink subsys mutexes.

Time to place them in distinct classes to avoid these warnings.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: ipset: enable memory accounting for ipset allocations
Vasily Averin [Fri, 25 Sep 2020 08:56:02 +0000 (11:56 +0300)]
netfilter: ipset: enable memory accounting for ipset allocations

Currently netadmin inside non-trusted container can quickly allocate
whole node's memory via request of huge ipset hashtable.
Other ipset-related memory allocations should be restricted too.

v2: fixed typo ALLOC -> ACCOUNT

Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nf_tables_offload: Remove unused macro FLOW_SETUP_BLOCK
YueHaibing [Fri, 18 Sep 2020 13:17:29 +0000 (21:17 +0800)]
netfilter: nf_tables_offload: Remove unused macro FLOW_SETUP_BLOCK

commit 9a32669fecfb ("netfilter: nf_tables_offload: support indr block call")
left behind this.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agomptcp: ADD_ADDRs with echo bit are smaller
Matthieu Baerts [Sat, 3 Oct 2020 15:36:56 +0000 (17:36 +0200)]
mptcp: ADD_ADDRs with echo bit are smaller

The MPTCP ADD_ADDR suboption with echo-flag=1 has no HMAC, the size is
smaller than the one initially sent without echo-flag=1. We then need to
use the correct size everywhere when we need this echo bit.

Before this patch, the wrong size was reserved but the correct amount of
bytes were written (and read): the remaining bytes contained garbage.

Fixes: 6a6c05a8b016 ("mptcp: send out ADD_ADDR with echo flag")
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/95
Reported-and-tested-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodt-bindings: net: dsa: b53: Fix full duplex in example
Kurt Kanzenbach [Sat, 3 Oct 2020 09:30:51 +0000 (11:30 +0200)]
dt-bindings: net: dsa: b53: Fix full duplex in example

There is no such property as duplex-full. It's called full-duplex. Leading to
reduced speed when using the example as base for a real device tree.

Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodt-bindings: net: dsa: b53: Specify unit address in hex
Kurt Kanzenbach [Sat, 3 Oct 2020 09:30:50 +0000 (11:30 +0200)]
dt-bindings: net: dsa: b53: Specify unit address in hex

The unit address should be 1e, because the unit address is supposed
to be in hexadecimal.

Suggested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: remove duplicate prefix for VL Lookup dynamic config
Vladimir Oltean [Sat, 3 Oct 2020 08:18:36 +0000 (11:18 +0300)]
net: dsa: sja1105: remove duplicate prefix for VL Lookup dynamic config

This is a strictly cosmetic change that renames some macros in
sja1105_dynamic_config.c. They were copy-pasted in haste and this has
resulted in them having the driver prefix twice.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnx2x: Use fallthrough pseudo-keyword
Gustavo A. R. Silva [Fri, 2 Oct 2020 23:26:11 +0000 (18:26 -0500)]
bnx2x: Use fallthrough pseudo-keyword

Replace /* no break */ comments with the new pseudo-keyword macro
fallthrough[1].

[1] https://www.kernel.org/doc/html/v5.7/process/deprecated.html?highlight=fallthrough#implicit-switch-case-fall-through

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ksz884x: Use fallthrough pseudo-keyword
Gustavo A. R. Silva [Fri, 2 Oct 2020 23:02:59 +0000 (18:02 -0500)]
net: ksz884x: Use fallthrough pseudo-keyword

Replace /* Fallthrough... */ comment with the new pseudo-keyword macro
fallthrough[1].

[1] https://www.kernel.org/doc/html/v5.7/process/deprecated.html?highlight=fallthrough#implicit-switch-case-fall-through

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bna: Use fallthrough pseudo-keyword
Gustavo A. R. Silva [Fri, 2 Oct 2020 23:00:01 +0000 (18:00 -0500)]
net: bna: Use fallthrough pseudo-keyword

Replace /* !!! fall through !!! */ comments with the new pseudo-keyword
macro fallthrough[1].

[1] https://www.kernel.org/doc/html/v5.7/process/deprecated.html?highlight=fallthrough#implicit-switch-case-fall-through

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agousbnet: Use fallthrough pseudo-keyword
Gustavo A. R. Silva [Fri, 2 Oct 2020 22:53:15 +0000 (17:53 -0500)]
usbnet: Use fallthrough pseudo-keyword

Replace // FALLTHROUGH comment with the new pseudo-keyword macro
fallthrough[1].

[1] https://www.kernel.org/doc/html/v5.7/process/deprecated.html?highlight=fallthrough#implicit-switch-case-fall-through

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'net-sched-Add-actions-for-MPLS-L2-VPNs'
David S. Miller [Sun, 4 Oct 2020 00:28:45 +0000 (17:28 -0700)]
Merge branch 'net-sched-Add-actions-for-MPLS-L2-VPNs'

Guillaume Nault says:

====================
net/sched: Add actions for MPLS L2 VPNs

This patch series adds the necessary TC actions for supporting layer 2
MPLS VPNs (VPLS).

The objective is to give the possibility to add an MPLS header right
before an skb's mac header, then to prepend this MPLS packet with a
new Ethernet header with the MAC address of the next hop.

Patch 1 implements the actions for adding and removing the external
Ethernet header.
Patch 2 adds the possibility to push an MPLS header before the mac
header.

Most of the code already exists as these operations were first
implemented in openvswitch.

Practical example, with encap on Host-A and decap on Host-B:

 Host-A# tc filter add dev ethAx ingress matchall         \
           action mpls mac_push label 20                  \
           action vlan push_eth dst_mac 02:00:00:00:00:02 \
                                src_mac 02:00:00:00:00:01 \
           action mirred egress redirect dev ethAy

 Host-B# tc filter add dev ethBx ingress protocol mpls_uc \
           flower mpls_label 20 mpls_bos 1                \
           action vlan pop_eth                            \
           action mpls pop proto teb                      \
           action mirred egress redirect dev ethBy
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/sched: act_mpls: Add action to push MPLS LSE before Ethernet header
Guillaume Nault [Fri, 2 Oct 2020 22:44:31 +0000 (00:44 +0200)]
net/sched: act_mpls: Add action to push MPLS LSE before Ethernet header

Define the MAC_PUSH action which pushes an MPLS LSE before the mac
header (instead of between the mac and the network headers as the
plain PUSH action does).

The only special case is when the skb has an offloaded VLAN. In that
case, it has to be inlined before pushing the MPLS header.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/sched: act_vlan: Add {POP,PUSH}_ETH actions
Guillaume Nault [Fri, 2 Oct 2020 22:44:28 +0000 (00:44 +0200)]
net/sched: act_vlan: Add {POP,PUSH}_ETH actions

Implement TCA_VLAN_ACT_POP_ETH and TCA_VLAN_ACT_PUSH_ETH, to
respectively pop and push a base Ethernet header at the beginning of a
frame.

POP_ETH is just a matter of pulling ETH_HLEN bytes. VLAN tags, if any,
must be stripped before calling POP_ETH.

PUSH_ETH is restricted to skbs with no mac_header, and only the MAC
addresses can be configured. The Ethertype is automatically set from
skb->protocol. These restrictions ensure that all skb's fields remain
consistent, so that this action can't confuse other part of the
networking stack (like GSO).

Since openvswitch already had these actions, consolidate the code in
skbuff.c (like for vlan and mpls push/pop).

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: dp83869: fix unsigned comparisons against less than zero values
Colin Ian King [Fri, 2 Oct 2020 16:54:22 +0000 (17:54 +0100)]
net: phy: dp83869: fix unsigned comparisons against less than zero values

Currently the comparisons of u16 integers value and sopass_val with
less than zero for error checking is always false because the values
are unsigned. Fix this by making these variables int.  This does not
affect the shift and mask operations performed on these variables

Addresses-Coverity: ("Unsigned compared against zero")
Fixes: 49fc23018ec6 ("net: phy: dp83869: support Wake on LAN")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Dan Murphy <dmurphy@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smc: use an array to check fields in system EID
Karsten Graul [Fri, 2 Oct 2020 15:09:27 +0000 (17:09 +0200)]
net/smc: use an array to check fields in system EID

The check for old hardware versions that did not have SMCDv2 support was
using suspicious pointer magic. Address the fields using an array.

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smc: send ISM devices with unique chid in CLC proposal
Karsten Graul [Fri, 2 Oct 2020 15:09:26 +0000 (17:09 +0200)]
net/smc: send ISM devices with unique chid in CLC proposal

When building a CLC proposal message then the list of ISM devices does
not need to contain multiple devices that have the same chid value,
all these devices use the same function at the end.
Improve smc_find_ism_v2_device_clnt() to collect only ISM devices that
have unique chid values.

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'Add-Seville-Ethernet-switch-to-T1040RDB'
David S. Miller [Sun, 4 Oct 2020 00:02:42 +0000 (17:02 -0700)]
Merge branch 'Add-Seville-Ethernet-switch-to-T1040RDB'

Vladimir Oltean says:

===================
Add Seville Ethernet switch to T1040RDB

Seville is a DSA switch that is embedded inside the T1040 SoC, and
supported by the mscc_seville DSA driver inside drivers/net/dsa/ocelot.

This series adds this switch to the SoC's dtsi files and to the T1040RDB
board file.

I would like to send this series through net-next. There is no conflict
with other patches submitted to T1040 device tree. Maybe this could at
least get an ACK from devicetree maintainers.
===================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agopowerpc: dts: t1040rdb: add ports for Seville Ethernet switch
Vladimir Oltean [Fri, 2 Oct 2020 13:41:06 +0000 (16:41 +0300)]
powerpc: dts: t1040rdb: add ports for Seville Ethernet switch

Define the network interface names for the switch ports and hook them up
to the 2 QSGMII PHYs that are onboard.

A conscious decision was taken to go along with the numbers that are
written on the front panel of the board and not with the hardware
numbers of the switch chip ports.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Maxim Kochetkov <fido_max@inbox.ru>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agopowerpc: dts: t1040: add bindings for Seville Ethernet switch
Vladimir Oltean [Fri, 2 Oct 2020 13:41:05 +0000 (16:41 +0300)]
powerpc: dts: t1040: add bindings for Seville Ethernet switch

Add the description of the embedded L2 switch inside the SoC dtsi file
for NXP T1040.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Maxim Kochetkov <fido_max@inbox.ru>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agotcp: account total lost packets properly
Yuchung Cheng [Thu, 1 Oct 2020 21:05:18 +0000 (14:05 -0700)]
tcp: account total lost packets properly

The retransmission refactoring patch
686989700cab ("tcp: simplify tcp_mark_skb_lost")
does not properly update the total lost packet counter which may
break the policer mode in BBR. This patch fixes it.

Fixes: 686989700cab ("tcp: simplify tcp_mark_skb_lost")
Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'net-iucv-next'
David S. Miller [Sat, 3 Oct 2020 23:51:07 +0000 (16:51 -0700)]
Merge branch 'net-iucv-next'

Julian Wiedmann says:

====================
net/iucv: updates 2020-10-01

Just two (rare) patches, and both deal with smatch warnings.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/iucv: fix indentation in __iucv_message_receive()
Julian Wiedmann [Thu, 1 Oct 2020 17:21:27 +0000 (19:21 +0200)]
net/iucv: fix indentation in __iucv_message_receive()

smatch complains about
net/iucv/iucv.c:1119 __iucv_message_receive() warn: inconsistent indenting

While touching this line, also make the return logic consistent and thus
get rid of a goto label.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/af_iucv: right-size the uid variable in iucv_sock_bind()
Julian Wiedmann [Thu, 1 Oct 2020 17:21:26 +0000 (19:21 +0200)]
net/af_iucv: right-size the uid variable in iucv_sock_bind()

smatch complains about
net/iucv/af_iucv.c:624 iucv_sock_bind() error: memcpy() 'sa->siucv_user_id' too small (8 vs 9)

Which is absolutely correct - the memcpy() takes 9 bytes (sizeof(uid))
from an 8-byte field (sa->siucv_user_id).
Luckily the sockaddr_iucv struct contains more data after the
.siucv_user_id field, and we checked the size of the passed data earlier
on. So the memcpy() won't accidentally read from an invalid location.

Fix the warning by reducing the size of the uid variable to what's
actually needed, and thus reducing the amount of copied data.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: remove NETDEV_HW_ADDR_T_SLAVE
Taehee Yoo [Thu, 1 Oct 2020 17:12:50 +0000 (17:12 +0000)]
net: remove NETDEV_HW_ADDR_T_SLAVE

NETDEV_HW_ADDR_T_SLAVE is not used anymore, remove it.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'genetlink-per-op-policy-export'
David S. Miller [Sat, 3 Oct 2020 21:18:29 +0000 (14:18 -0700)]
Merge branch 'genetlink-per-op-policy-export'

Johannes Berg says:

====================
genetlink per-op policy export

Here's a respin, now including Jakub's patch last so that it will
do the right thing from the start.

The first patch remains the same, of course; the others have mostly
some rebasing going on, except for the actual export patch (patch 4)
which is adjusted per Jakub's review comments about exporting the
policy only if it's actually used for do/dump.

To see that, the dump for "nlctrl" (i.e. the generic netlink control)
is instructive, because the ops are this:

        {
                .cmd            = CTRL_CMD_GETFAMILY,
                .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
                .policy         = ctrl_policy_family,
                .maxattr        = ARRAY_SIZE(ctrl_policy_family) - 1,
                .doit           = ctrl_getfamily,
                .dumpit         = ctrl_dumpfamily,
        },
        {
                .cmd            = CTRL_CMD_GETPOLICY,
                .policy         = ctrl_policy_policy,
                .maxattr        = ARRAY_SIZE(ctrl_policy_policy) - 1,
                .start          = ctrl_dumppolicy_start,
                .dumpit         = ctrl_dumppolicy,
                .done           = ctrl_dumppolicy_done,
        },

So we exercise both "don't have doit" and "GENL_DONT_VALIDATE_DUMP"
parts, and get (with the current genl patch):

$ genl ctrl policy name nlctrl
ID: 0x10  op 3 policies: do=0
ID: 0x10  op 10 policies: dump=1
ID: 0x10  policy[0]:attr[1]: type=U16 range:[0,65535]
ID: 0x10  policy[0]:attr[2]: type=NUL_STRING max len:15
ID: 0x10  policy[1]:attr[1]: type=U16 range:[0,65535]
ID: 0x10  policy[1]:attr[2]: type=NUL_STRING max len:15
ID: 0x10  policy[1]:attr[10]: type=U32 range:[0,4294967295]
====================

Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: allow dumping command-specific policy
Jakub Kicinski [Sat, 3 Oct 2020 08:44:46 +0000 (10:44 +0200)]
genetlink: allow dumping command-specific policy

Right now CTRL_CMD_GETPOLICY can only dump the family-wide
policy. Support dumping policy of a specific op.

v3:
 - rebase after per-op policy export and handle that
v2:
 - make cmd U32, just in case.
v1:
 - don't echo op in the output in a naive way, this should
   make it cleaner to extend the output format for dumping
   policies for all the commands at once in the future.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20201001225933.1373426-11-kuba@kernel.org
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: properly support per-op policy dumping
Johannes Berg [Sat, 3 Oct 2020 08:44:45 +0000 (10:44 +0200)]
genetlink: properly support per-op policy dumping

Add support for per-op policy dumping. The data is pretty much
as before, except that now the assumption that the policy with
index 0 is "the" policy no longer holds - you now need to look
at the new CTRL_ATTR_OP_POLICY attribute which is a nested attr
(indexed by op) containing attributes for do and dump policies.

When a single op is requested, the CTRL_ATTR_OP_POLICY will be
added in the same way, since do and dump policies may differ.

v2:
 - conditionally advertise per-command policies only if there
   actually is a policy being used for the do/dump and it's
   present at all

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: factor skb preparation out of ctrl_dumppolicy()
Johannes Berg [Sat, 3 Oct 2020 08:44:44 +0000 (10:44 +0200)]
genetlink: factor skb preparation out of ctrl_dumppolicy()

We'll need this later for the per-op policy index dump.

Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetlink: rework policy dump to support multiple policies
Johannes Berg [Sat, 3 Oct 2020 08:44:43 +0000 (10:44 +0200)]
netlink: rework policy dump to support multiple policies

Rework the policy dump code a bit to support adding multiple
policies to a single dump, in order to e.g. support per-op
policies in generic netlink.

v2:
 - move kernel-doc to implementation [Jakub]
 - squash the first patch to not flip-flop on the prototype
   [Jakub]
 - merge netlink_policy_dump_get_policy_idx() with the old
   get_policy_idx() we already had
 - rebase without Jakub's patch to have per-op dump

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetlink: compare policy more accurately
Johannes Berg [Sat, 3 Oct 2020 08:44:42 +0000 (10:44 +0200)]
netlink: compare policy more accurately

The maxtype is really an integral part of the policy, and while we
haven't gotten into a situation yet where this happens, it seems
that some developer might eventually have two places pointing to
identical policies, with different maxattr to exclude some attrs
in one of the places.

Even if not, it's really the right thing to compare both since the
two data items fundamentally belong together.

v2:
 - also do the proper comparison in get_policy_idx()

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'genetlink-support-per-command-policy-dump'
David S. Miller [Sat, 3 Oct 2020 02:11:12 +0000 (19:11 -0700)]
Merge branch 'genetlink-support-per-command-policy-dump'

Jakub Kicinski says:

====================
genetlink: support per-command policy dump

The objective of this series is to dump ethtool policies
to be able to tell which flags are supported by the kernel.
Current release adds ETHTOOL_FLAG_STATS for dumping extra
stats, but because of strict checking we need to make sure
that the flag is actually supported before setting it in
a request.

Ethtool policies are per command, and so far only dumping
family policies was supported.

The series adds new set of "light" ops to genl families which
don't have all the callbacks, and won't have the policy.
Most of families are then moved to these ops. This gives
us 4096B in savings on an allyesconfig build (not counting
the growth that would have happened when policy is added):

     text       data       bss        dec       hex
244415581  227958581  78372980  550747142  20d3bc06
244415581  227962677  78372980  550751238  20d3cc06

Next 5 patches deal the dumping per-op policy.

v3:

The actually patch to dump per-op policy was taken out and
will come in a series from Johannes, to make sure uAPI is
consistent from the start.

For dump-specific policies I think it should be fine to add
a new pair of members to the "full" ops, and not overthink it.

v2:
 - remove the stale comment in taskstats
 - split patch 8 -> 8, 9
 - now the getfamily policy is also in the op
 - make cmd u32
v1:
 - replace remaining uses of "light" with "small"
 - fix dump (ops can't be on the stack there)
 - coding changes in patch 4
 - new patch 7
 - don't echo op in responses - to make dump all easier

Dave - this series will cause a very trivial conflict with
the patch I sent to net. Both sides add some kdoc to struct
genl_ops so we'll need to keep it all.  I'm sending this
already because I also need to restructure ethool policies
in time for 5.10 if we want to use it for the stats flag.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: switch control commands to per-op policies
Jakub Kicinski [Fri, 2 Oct 2020 21:50:00 +0000 (14:50 -0700)]
genetlink: switch control commands to per-op policies

In preparation for adding a new attribute to CTRL_CMD_GETPOLICY
split the policies for getpolicy and getfamily apart.

This will cause a slight user-visible change in that dumping
the policies will switch from per family to per op, but
supposedly sniffer-type applications (which are the main use
case for policy dumping thus far) should support both, anyway.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: use parsed attrs in dumppolicy
Jakub Kicinski [Fri, 2 Oct 2020 21:49:59 +0000 (14:49 -0700)]
genetlink: use parsed attrs in dumppolicy

Attributes are already parsed based on the policy specified
in the family and ready-to-use in info->attrs. No need to
call genlmsg_parse() again.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agotaskstats: move specifying netlink policy back to ops
Jakub Kicinski [Fri, 2 Oct 2020 21:49:58 +0000 (14:49 -0700)]
taskstats: move specifying netlink policy back to ops

commit 3b0f31f2b8c9 ("genetlink: make policy common to family")
had to work around removal of policy from ops by parsing in
the pre_doit callback. Now that policy is back in full ops
we can switch again. Set maxattr to actual size of the policies
- both commands set GENL_DONT_VALIDATE_STRICT so out of range
attributes will be silently ignored, anyway.

v2:
 - remove stale comment

Suggested-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: bring back per op policy
Jakub Kicinski [Fri, 2 Oct 2020 21:49:57 +0000 (14:49 -0700)]
genetlink: bring back per op policy

Add policy to the struct genl_ops structure, this time
with maxattr, so it can be used properly.

Propagate .policy and .maxattr from the family
in genl_get_cmd() if needed, this way the rest of the
code does not have to worry if the policy is per op
or global.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: use .start callback for dumppolicy
Jakub Kicinski [Fri, 2 Oct 2020 21:49:56 +0000 (14:49 -0700)]
genetlink: use .start callback for dumppolicy

The structure of ctrl_dumppolicy() is clearly split into
init and dumping. Move the init to a .start callback
for clarity, it's a more idiomatic netlink dump code structure.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: add a structure for dump state
Jakub Kicinski [Fri, 2 Oct 2020 21:49:55 +0000 (14:49 -0700)]
genetlink: add a structure for dump state

Whenever netlink dump uses more than 2 cb->args[] entries
code gets hard to read. We're about to add more state to
ctrl_dumppolicy() so create a structure.

Since the structure is typed and clearly named we can remove
the local fam_id variable and use ctx->fam_id directly.

v3:
 - rebase onto explicit free fix
v1:
 - s/nl_policy_dump/netlink_policy_dump_state/
 - forward declare struct netlink_policy_dump_state,
   and move from passing unsigned long to actual pointer type
 - add build bug on
 - u16 fam_id
 - s/args/ctx/

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: move to smaller ops wherever possible
Jakub Kicinski [Fri, 2 Oct 2020 21:49:54 +0000 (14:49 -0700)]
genetlink: move to smaller ops wherever possible

Bulk of the genetlink users can use smaller ops, move them.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: add small version of ops
Jakub Kicinski [Fri, 2 Oct 2020 21:49:53 +0000 (14:49 -0700)]
genetlink: add small version of ops

We want to add maxattr and policy back to genl_ops, to enable
dumping per command policy to user space. This, however, would
cause bloat for all the families with global policies. Introduce
smaller version of ops (half the size of genl_ops). Translate
these smaller ops into a full blown struct before use in the
core.

v1:
 - use struct assignment
 - put a full copy of the op in struct genl_dumpit_info
 - s/light/small/

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agogenetlink: reorg struct genl_family
Jakub Kicinski [Fri, 2 Oct 2020 21:49:52 +0000 (14:49 -0700)]
genetlink: reorg struct genl_family

There are holes and oversized members in struct genl_family.

Before: /* size: 104, cachelines: 2, members: 16 */
After:  /* size:  88, cachelines: 2, members: 16 */

The command field in struct genlmsghdr is a u8, so no point
in the operation count being 32 bit. Also operation 0 is
usually undefined, so we only need 255 entries.

netnsok and parallel_ops are only ever initialized to true.

We can grow the fields as needed, compiler should warn us
if someone tries to assign larger constants.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'dpaa2-eth-add-devlink-parser-error-drop-trap-support'
David S. Miller [Fri, 2 Oct 2020 23:31:56 +0000 (16:31 -0700)]
Merge branch 'dpaa2-eth-add-devlink-parser-error-drop-trap-support'

Ioana Ciornei says:

====================
dpaa2-eth: add devlink parser error drop trap support

This patch set adds support in the dpaa2-eth driver for a new group of
devlink drop traps - PARSER_ERROR_DROPS.

The first patch adds a new generic trap group and associated traps,
their definitions in devlink and their corresponding entries in the
Documentation.

Because there might be more devices (besides DPAA2) which do not support
changing the action independently on each trap, a nre devlink callback
is introduced - .trap_group_action_set(). If this callback is populated,
it will take precedence over .trap_action_set() when the user requests
changing the action on all the traps in a group.

The next patches add basic linkage with devlink for the dpaa2-eth driver
and support for the newly added PARSER_ERROR_DROPS. Nothing special
here, just setting up the Rx error queue, interpreting the parse result,
and then reporting any frame received on that queue to devlink.

Changes in v2:
 - fix build error in 3/4

Changes in v3:
 - removed a commented line in 4/4
 - added an extack in 4/4
 - fixed up a warning on 32bit in 4/4
 - reworded the trap_group_action_set() description in 2/4
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: add support for devlink parser error drop traps
Ioana Ciornei [Thu, 1 Oct 2020 15:11:48 +0000 (18:11 +0300)]
dpaa2-eth: add support for devlink parser error drop traps

Add support for the new group of devlink traps - PARSER_ERROR_DROPS.
This consists of registering the array of parser error drops supported,
controlling their action through the .trap_group_action_set() callback
and reporting an erroneous skb received on the error queue
appropriately.
DPAA2 devices do not support controlling the action of independent
parser error traps, thus the .trap_action_set() callback just returns an
EOPNOTSUPP while .trap_group_action_set() actually notifies the hardware
what it should do with a frame marked as having a header error.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: add basic devlink support
Ioana Ciornei [Thu, 1 Oct 2020 15:11:47 +0000 (18:11 +0300)]
dpaa2-eth: add basic devlink support

Add basic support in dpaa2-eth for devlink. For the moment, just
register the device with devlink, add the corresponding devlink port and
implement the .info_get() callback.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: add .trap_group_action_set() callback
Ioana Ciornei [Thu, 1 Oct 2020 15:11:46 +0000 (18:11 +0300)]
devlink: add .trap_group_action_set() callback

Add a new devlink callback, .trap_group_action_set(), which can be used
by device drivers which do not support controlling the action (drop,
trap) on each trap but rather on the entire group trap.
If this new callback is populated, it will take precedence over the
.trap_action_set() callback when the user requests a change of all the
traps in a group.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: add parser error drop packet traps
Ioana Ciornei [Thu, 1 Oct 2020 15:11:45 +0000 (18:11 +0300)]
devlink: add parser error drop packet traps

Add parser error drop packet traps, so that capable device driver could
register them with devlink. The new packet trap group holds any drops of
packets which were marked by the device as erroneous during header
parsing. Add documentation for every added packet trap and packet trap
group.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'ionic-error-recovery'
David S. Miller [Fri, 2 Oct 2020 23:30:01 +0000 (16:30 -0700)]
Merge branch 'ionic-error-recovery'

Shannon Nelson says:

====================
ionic error recovery

This set of patches comes mostly from error recovery path testing,
as well as a couple of upstream review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: add new bad firmware error code
Shannon Nelson [Thu, 1 Oct 2020 16:22:46 +0000 (09:22 -0700)]
ionic: add new bad firmware error code

If the new firmware image downladed for update is corrupted
or is a bad format, the download process will report a status
code specifically for that.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: use lif ident for filter count
Shannon Nelson [Thu, 1 Oct 2020 16:22:45 +0000 (09:22 -0700)]
ionic: use lif ident for filter count

Use the lif's ident information for the uc and mc filter
counts rather than the ionic's version, to be sure
we're getting the info that is specific to this lif.

While we're thinking about it, add some missing error
checking where we get the lif's identity information.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: refill lif identity after fw_up
Shannon Nelson [Thu, 1 Oct 2020 16:22:44 +0000 (09:22 -0700)]
ionic: refill lif identity after fw_up

After we do a fw upgrade and refill the ionic->ident.dev, we
also need to update the other identity info.  Since the lif
identity needs to be updated each time the ionic identity is
refreshed, we can pull it into ionic_identify().

The debugfs entry is moved so that it doesn't cause an
error message when the data is refreshed after the fw upgrade.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: disable all queue napi contexts on timeout
Shannon Nelson [Thu, 1 Oct 2020 16:22:43 +0000 (09:22 -0700)]
ionic: disable all queue napi contexts on timeout

Some time ago we short-circuited the queue disables on a timeout
error in order to not have to wait on every queue when we already
know it will time out.  However, this meant that we're not
properly stopping all the interrupts and napi contexts.  This
changes queue disable to always call ionic_qcq_disable() and to
give it an argument to know when to not do the adminq request.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: check qcq ptr in ionic_qcq_disable
Shannon Nelson [Thu, 1 Oct 2020 16:22:42 +0000 (09:22 -0700)]
ionic: check qcq ptr in ionic_qcq_disable

There are a couple of error recovery paths that can come through
ionic_qcq_disable() without having set up the qcq, so we need
to make sure we have a valid qcq pointer before using it.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: clear linkcheck bit on alloc fail
Shannon Nelson [Thu, 1 Oct 2020 16:22:41 +0000 (09:22 -0700)]
ionic: clear linkcheck bit on alloc fail

Clear our link check requested flag on an allocation error.
We end up dropping this link check request, but that should
be fine as our watchdog will come back a few seconds later
and request it again.

Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: drain the work queue
Shannon Nelson [Thu, 1 Oct 2020 16:22:40 +0000 (09:22 -0700)]
ionic: drain the work queue

Check through our work list for additional items.  This normally
will only have one item, but occasionally may have another
job waiting.  There really is no need reschedule ourself here.

Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: contiguous memory for notifyq
Shannon Nelson [Thu, 1 Oct 2020 16:22:39 +0000 (09:22 -0700)]
ionic: contiguous memory for notifyq

The event notification queue is set up a little differently in the
NIC and so the notifyq q and cq descriptor structures need to be
contiguous, which got missed in an earlier patch that separated
out the q and cq descriptor allocations.  That patch was aimed at
making the big tx and rx descriptor queue allocations easier to
manage - the notifyq is much smaller and doesn't need to be split.
This patch simply adds an if/else and slightly different code for
the notifyq descriptor allocation.

Fixes: ea5a8b09dc3a ("ionic: reduce contiguous memory allocation requirement")
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 's390-net-next'
David S. Miller [Fri, 2 Oct 2020 23:22:49 +0000 (16:22 -0700)]
Merge branch 's390-net-next'

Julian Wiedmann says:

===================
s390/net: updates 2020-10-01

Patches 1-3 enable qeth to also support the .set_channels() ethtool
callback for OSA devices. This completes support for the full range
of device types.

The other patches are just the usual mix of cleanups.
(Even one for ctcm!)
===================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/ctcm: remove orphaned function declarations
Vasily Gorbik [Thu, 1 Oct 2020 17:11:36 +0000 (19:11 +0200)]
s390/ctcm: remove orphaned function declarations

drivers/s390/net/ctcm_fsms.h: fsm_action_nop - only declaration left
after commit 04885948b101 ("ctc: removal of the old ctc driver")

drivers/s390/net/ctcm_mpc.h: ctcmpc_open - only declaration left after
commit 293d984f0e36 ("ctcm: infrastructure for replaced ctc driver")

Reviewed-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: static checker cleanups
Julian Wiedmann [Thu, 1 Oct 2020 17:11:35 +0000 (19:11 +0200)]
s390/qeth: static checker cleanups

- Add/delete some blanks, white spaces and braces.
- Fix misindentations.
- Adjust a deprecated header include, and htons() conversion.
- Remove extra 'return' statements.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: use netdev_name()
Julian Wiedmann [Thu, 1 Oct 2020 17:11:34 +0000 (19:11 +0200)]
s390/qeth: use netdev_name()

Replace our custom version of netdev_name().

Once we started to allocate the netdev at probe time with
commit d3d1b205e89f ("s390/qeth: allocate netdevice early"), this
stopped working as intended anyway.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: constify the disciplines
Julian Wiedmann [Thu, 1 Oct 2020 17:11:33 +0000 (19:11 +0200)]
s390/qeth: constify the disciplines

The discipline struct is a fixed group of function pointers.
So declare the L2 and L3 disciplines as constant.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: allow configuration of TX queues for OSA devices
Julian Wiedmann [Thu, 1 Oct 2020 17:11:32 +0000 (19:11 +0200)]
s390/qeth: allow configuration of TX queues for OSA devices

For OSA devices that are _not_ configured in prio-queue mode, give users
the option of selecting the number of active TX queues.
This requires setting up the HW queues with a reasonable default QoS
value in the QIB's PQUE parm area.

As with the other device types, we bring up the device with a minimal
number of TX queues for compatibility reasons.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: de-magic the QIB parm area
Julian Wiedmann [Thu, 1 Oct 2020 17:11:31 +0000 (19:11 +0200)]
s390/qeth: de-magic the QIB parm area

Use a proper struct, and only program the QIB extensions for devices
where they are supported.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: keep track of wanted TX queues
Julian Wiedmann [Thu, 1 Oct 2020 17:11:30 +0000 (19:11 +0200)]
s390/qeth: keep track of wanted TX queues

When re-initializing a device, we can hit a situation where
qeth_osa_set_output_queues() detects that it supports more or less
HW TX queues than before. Right now we adjust dev->real_num_tx_queues
from right there, but
1. it's getting more & more complicated to cover all cases, and
2. we can't re-enable the actually expected number of TX queues later
because we lost the needed information.

So keep track of the wanted TX queues (on initial setup, and whenever
its changed via .set_channels), and later use that information when
re-enabling the netdevice.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'wireless-drivers-next-2020-10-02' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Fri, 2 Oct 2020 22:42:40 +0000 (15:42 -0700)]
Merge tag 'wireless-drivers-next-2020-10-02' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for v5.10

Third set of patches for v5.10. Lots of iwlwifi patches this time, but
also few patches ath11k and of course smaller changes to other
drivers.

Major changes:

rtw88

* properly recover from firmware crashes on 8822c

* dump firmware crash log

iwlwifi

* protected Target Wake Time (TWT) implementation

* support disabling 5.8GHz channels via ACPI

* support VHT extended NSS capability

* enable Target Wake Time (TWT) by default

ath11k

* improvements to QCA6390 PCI support to make it more usable
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'Offload-tc-flower-to-mscc_ocelot-switch-using-VCAP-chains'
David S. Miller [Fri, 2 Oct 2020 22:40:30 +0000 (15:40 -0700)]
Merge branch 'Offload-tc-flower-to-mscc_ocelot-switch-using-VCAP-chains'

Vladimir Oltean says:

====================
Offload tc-flower to mscc_ocelot switch using VCAP chains

The purpose of this patch is to add more comprehensive support for flow
offloading in the mscc_ocelot library and switch drivers.

The design (with chains) is the result of this discussion:
https://lkml.org/lkml/2020/6/2/203

I have tested it on Seville VSC9953 and Felix VSC9959, but it should
also work on Ocelot-1 VSC7514.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: ocelot: add some example VCAP IS1, IS2 and ES0 tc offloads
Vladimir Oltean [Fri, 2 Oct 2020 12:02:28 +0000 (15:02 +0300)]
selftests: ocelot: add some example VCAP IS1, IS2 and ES0 tc offloads

Provide an example script which can be used as a skeleton for offloading
TCAM rules in the Ocelot switches.

Not all actions are demoed, mostly because of difficulty to automate
this from a single board.

For example, policing. We can set up an iperf3 UDP server and client and
measure throughput at destination. But at least with DSA setups, network
namespacing the individual ports is not possible because all switch
ports are handled by the same DSA master. And we cannot assume that the
target platform (an embedded board) has 2 other non-switch generator
ports, we need to work with the generator ports as switch ports (this is
the reason why mausezahn is used, and not IP traffic like ping). When
somebody has an idea how to test policing, that can be added to this
test.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: offload redirect action to VCAP IS2
Vladimir Oltean [Fri, 2 Oct 2020 12:02:27 +0000 (15:02 +0300)]
net: mscc: ocelot: offload redirect action to VCAP IS2

Via the OCELOT_MASK_MODE_REDIRECT flag put in the IS2 action vector, it
is possible to replace previous forwarding decisions with the port mask
installed in this rule.

I have studied Table 54 "MASK_MODE and PORT_MASK Combinations" from the
VSC7514 documentation and it appears to behave sanely when this rule is
installed in either lookup 0 or 1. Namely, a redirect in lookup 1 will
overwrite the forwarding decision taken by any entry in lookup 0.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: relax ocelot_exclusive_mac_etype_filter_rules()
Vladimir Oltean [Fri, 2 Oct 2020 12:02:26 +0000 (15:02 +0300)]
net: mscc: ocelot: relax ocelot_exclusive_mac_etype_filter_rules()

The issue which led to the introduction of this check was that MAC_ETYPE
rules, such as filters on dst_mac and src_mac, would only match non-IP
frames. There is a knob in VCAP_S2_CFG which forces all IP frames to be
treated as non-IP, which is what we're currently doing if the user
requested a dst_mac filter, in order to maintain sanity.

But that knob is actually per IS2 lookup. And the good thing with
exposing the lookups to the user via tc chains is that we're now able to
offload MAC_ETYPE keys to one lookup, and IP keys to the other lookup.
So let's do that.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: only install TCAM entries into a specific lookup and PAG
Vladimir Oltean [Fri, 2 Oct 2020 12:02:25 +0000 (15:02 +0300)]
net: mscc: ocelot: only install TCAM entries into a specific lookup and PAG

We were installing TCAM rules with the LOOKUP field as unmasked, meaning
that all entries were matching on all lookups. Now that lookups are
exposed as individual chains, let's make the LOOKUP explicit when
offloading TCAM entries.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: offload egress VLAN rewriting to VCAP ES0
Xiaoliang Yang [Fri, 2 Oct 2020 12:02:24 +0000 (15:02 +0300)]
net: mscc: ocelot: offload egress VLAN rewriting to VCAP ES0

VCAP ES0 is an egress VCAP operating on all outgoing frames.
This patch added ES0 driver to support vlan push action of tc filter.
Usage:

tc filter add dev swp1 egress protocol 802.1Q flower indev swp0 skip_sw \
        vlan_id 1 vlan_prio 1 action vlan push id 2 priority 2

Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: offload ingress skbedit and vlan actions to VCAP IS1
Xiaoliang Yang [Fri, 2 Oct 2020 12:02:23 +0000 (15:02 +0300)]
net: mscc: ocelot: offload ingress skbedit and vlan actions to VCAP IS1

VCAP IS1 is a VCAP module which can filter on the most common L2/L3/L4
Ethernet keys, and modify the results of the basic QoS classification
and VLAN classification based on those flow keys.

There are 3 VCAP IS1 lookups, mapped over chains 10000, 11000 and 12000.
Currently the driver is hardcoded to use IS1_ACTION_TYPE_NORMAL half
keys.

Note that the VLAN_MANGLE has been omitted for now. In hardware, the
VCAP_IS1_ACT_VID_REPLACE_ENA field replaces the classified VLAN
(metadata associated with the frame) and not the VLAN from the header
itself. There are currently some issues which need to be addressed when
operating in standalone, or in bridge with vlan_filtering=0 modes,
because in those cases the switch ports have VLAN awareness disabled,
and changing the classified VLAN to anything other than the pvid causes
the packets to be dropped. Another issue is that on egress, we expect
port tagging to push the classified VLAN, but port tagging is disabled
in the modes mentioned above, so although the classified VLAN is
replaced, it is not visible in the packet transmitted by the switch.

Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: create TCAM skeleton from tc filter chains
Vladimir Oltean [Fri, 2 Oct 2020 12:02:22 +0000 (15:02 +0300)]
net: mscc: ocelot: create TCAM skeleton from tc filter chains

For Ocelot switches, there are 2 ingress pipelines for flow offload
rules: VCAP IS1 (Ingress Classification) and IS2 (Security Enforcement).
IS1 and IS2 support different sets of actions. The pipeline order for a
packet on ingress is:

Basic classification -> VCAP IS1 -> VCAP IS2

Furthermore, IS1 is looked up 3 times, and IS2 is looked up twice (each
TCAM entry can be configured to match only on the first lookup, or only
on the second, or on both etc).

Because the TCAMs are completely independent in hardware, and because of
the fixed pipeline, we actually have very limited options when it comes
to offloading complex rules to them while still maintaining the same
semantics with the software data path.

This patch maps flow offload rules to ingress TCAMs according to a
predefined chain index number. There is going to be a script in
selftests that clarifies the usage model.

There is also an egress TCAM (VCAP ES0, the Egress Rewriter), which is
modeled on top of the default chain 0 of the egress qdisc, because it
doesn't have multiple lookups.

Suggested-by: Allan W. Nielsen <allan.nielsen@microchip.com>
Co-developed-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: introduce conversion helpers between port and netdev
Vladimir Oltean [Fri, 2 Oct 2020 12:02:21 +0000 (15:02 +0300)]
net: mscc: ocelot: introduce conversion helpers between port and netdev

Since the mscc_ocelot_switch_lib is common between a pure switchdev and
a DSA driver, the procedure of retrieving a net_device for a certain
port index differs, as those are registered by their individual
front-ends.

Up to now that has been dealt with by always passing the port index to
the switch library, but now, we're going to need to work with net_device
pointers from the tc-flower offload, for things like indev, or mirred.
It is not desirable to refactor that, so let's make sure that the flower
offload core has the ability to translate between a net_device and a
port index properly.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: offload multiple tc-flower actions in same rule
Vladimir Oltean [Fri, 2 Oct 2020 12:02:20 +0000 (15:02 +0300)]
net: mscc: ocelot: offload multiple tc-flower actions in same rule

At this stage, the tc-flower offload of mscc_ocelot can only delegate
rules to the VCAP IS2 security enforcement block. These rules have, in
hardware, separate bits for policing and for overriding the destination
port mask and/or copying to the CPU. So it makes sense that we attempt
to expose some more of that low-level complexity instead of simply
choosing between a single type of action.

Something similar happens with the VCAP IS1 block, where the same action
can contain enable bits for VLAN classification and for QoS
classification at the same time.

So model the action structure after the hardware description, and let
the high-level ocelot_flower.c construct an action vector from multiple
tc actions.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'mac80211-next-for-net-next-2020-10-02' of git://git.kernel.org/pub/scm...
David S. Miller [Fri, 2 Oct 2020 22:33:13 +0000 (15:33 -0700)]
Merge tag 'mac80211-next-for-net-next-2020-10-02' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
Another set of changes, this time with:
 * lots more S1G band support
 * 6 GHz scanning, finally
 * kernel-doc fixes
 * non-split wiphy dump fixes in nl80211
 * various other small cleanups/features
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smscx5xx: change to of_get_mac_address() eth_platform_get_mac_address()
Łukasz Stelmach [Wed, 30 Sep 2020 14:25:25 +0000 (16:25 +0200)]
net/smscx5xx: change to of_get_mac_address() eth_platform_get_mac_address()

Use more generic eth_platform_get_mac_address() which can get a MAC
address from other than DT platform specific sources too. Check if the
obtained address is valid.

Signed-off-by: Łukasz Stelmach <l.stelmach@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>