Add support for rewriting of DSCP part of ToS field.
Next commands, for example, can be used to offload rewrite action:

OVS:
 $ ovs-ofctl add-flow ovs-sriov "ip, in_port=REP, \
       actions=mod_nw_tos:68, output:NIC"

iproute2 (used retain mask, as tc command rewrite whole ToS field):
 $ tc filter add dev REP ingress protocol ip prio 1 flower skip_sw \
       ip_proto icmp action pedit munge ip tos set 68 retain 0xfc pipe \
       action mirred egress redirect dev NIC
Signed-off-by: Dmytro Linkin <dmitrolin@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

This patch doesn't change any functionality, but is a pre-step for
adding support for rewriting of bit-sized fields, like DSCP and ECN
in IPv4 header, similar fields in IPv6, etc.
Signed-off-by: Dmytro Linkin <dmitrolin@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Move short Work Queue API getter functions into the WQ
header file.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Dump the Work Queue's TX WQE descriptor when a completion with
error is received.

Example:
[5.331832] mlx5_core 0000:00:04.0 enp0s4: Error cqe on cqn 0xa, ci 0x1, TXQ-SQ qpn 0xe, opcode 0xd, syndrome 0x2, vendor syndrome 0x0
[5.333127] 00000000: 55 65 02 75 31 fe c2 d2 6b 6c 62 1e f9 e1 d8 5c
[5.333837] 00000010: d3 b2 6c b8 89 e4 84 20 0b f4 3c e0 f3 75 41 ca
[5.334568] 00000020: 46 00 00 00 cd 70 a0 92 18 3a 01 de 00 00 00 00
[5.335313] 00000030: 7d bc 05 89 b2 e9 00 02 1e 00 00 0e 00 00 30 d2
[5.335972] WQE DUMP: WQ size 1024 WQ cur size 0, WQE index 0x0, len: 64
[5.336710] 00000000: 00 00 00 1e 00 00 0e 04 00 00 00 08 00 00 00 00
[5.337524] 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 12 33 33
[5.338151] 00000020: 00 00 00 16 52 54 00 00 00 01 86 dd 60 00 00 00
[5.338740] 00000030: 00 00 00 48 00 00 00 00 00 00 00 00 66 ba 58 14
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

During connection tracking offloads with high number of connections,
(40K connections per second), flow table group lock contention is
observed.
To improve the performance by reducing lock contention, lockless
FTE read lookup is performed as described below.

Each flow table entry is refcounted.
Flow table entry is removed when refcount drops to zero.
rhash table allows rcu protected lookup.
Each hash table entry insertion and removal is write lock protected.

Hence, it is possible to perform lockless lookup in rhash table using
following scheme.

(a) Guard FTE entry lookup per group using rcu read lock.
(b) Before freeing the FTE entry, wait for all readers to finish
accessing the FTE.

Below example of one reader and write in parallel racing, shows
protection in effect with rcu lock.

lookup_fte_locked()
  rcu_read_lock();
  search_hash_table()
                                  existing_flow_group_write_lock();
                                  tree_put_node(fte)
                                    drop_ref_cnt(fte)
                                    del_sw_fte(fte)
                                    del_hash_table_entry();
                                    call_rcu();
                                  existing_flow_group_write_unlock();
  get_ref_cnt(fte) fails
  rcu_read_unlock();
                                  rcu grace period();
                                    [..]
                                    kmem_cache_free(fte);
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

FTE memory allocation using alloc_fte() doesn't have any dependency
on the flow group.
Hence, do not hold flow group lock while performing alloc_fte().
This helps to reduce contention of flow group lock.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Currently, mlx5 tc layer doesn't verify that rule has at least one forward
or drop action which leads to following firmware syndrome when user tries
to offload such action:

[ 1824.860501] mlx5_core 0000:81:00.0: mlx5_cmd_check:753:(pid 29458): SET_FLOW_TABLE_ENTRY(0x936) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0x144b7a)

Add check at the end of parse_tc_fdb_actions() that verifies that resulting
attribute has action fwd or drop flag set.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

When setting number of VFs to 0 (disable SRIOV), clear VF's
configuration.
Signed-off-by: Aya Levin <ayal@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

mlx5_unload_one do not need local variable to store different value,
Hence just remove it.
Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Not all mlx5 cards with FPGA device use it for network processing.

mlx5_core driver configures network connection to FPGA device
for all mlx5 cards with installed FPGA. If FPGA is not a part of
network path, driver crashes in this case

Check FPGA name in function mlx5_fpga_device_start() and continue
integrate FPGA into packets flow only for dedicated cards.
Currently there are Newton and Edison cards.
Signed-off-by: Igor Leshenko <igorle@mellanox.com>
Reviewed-by: Meir Lichtinger <meirl@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Use kernel function to calculate crc32 Instead of dr implementation
since it has the same algorithm "slice by 8".

Fixes: 26d688e3 ("net/mlx5: DR, Add Steering entry (STE) utilities")
Signed-off-by: Hamdan Igbaria <hamdani@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

100BaseVG AnyLAN hasn't been useful since 1996 or so and even then
didn't sell many devices.  It's unlikely any are still in use.

Move the driver to staging with the intent of removing it altogether
one day.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Chan says:

====================
bnxt_en: Updates for net-next.

This patch series adds TC Flower tunnel decap and rewrite actions in
the first 4 patches.  The next 3 patches integrates the recently
added error recovery with the RDMA driver by calling the proper
hooks to stop and start.

v2: Fix pointer alignment issue in patch #1.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Inform the RDMA driver to stop/start during suspend/resume.  The
RDMA driver needs to stop and start just like error recovery.
Signed-off-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>

Notify the RDMA driver by calling the bnxt_ulp_stop()/bnxt_ulp_start()
hooks during error recovery.  The current ULP IRQ start/stop
sequence in error recovery (which is insufficient) is replaced with the
full reset sequence when we call bnxt_ulp_stop()/bnxt_ulp_start().
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>

We call bnxt_ulp_stop() to notify the RDMA driver that some error or
imminent reset is about to happen.  After that we always call
some variants of bnxt_close().

In the next patch, we will integrate the recently added error
recovery with the RDMA driver.  In response to ulp_stop, the
RDMA driver may free MSIX vectors and that will also trigger
bnxt_close().  To avoid bnxt_close() from being called twice,
we set a new flag after ulp_stop is called.  If the RDMA driver
frees MSIX vectors while the new flag is set, we will not call
bnxt_close(), knowing that it will happen in due course.

With this change, we must make sure that the bnxt_close() call
after ulp_stop will reset IRQ.  Modify bnxt_reset_task()
accordingly if we call ulp_stop.
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>

The decap (VXLAN tunnel) flow rules are not getting offloaded with
upstream kernel. This is because TC block callback infrastructure has
been updated to use indirect callbacks to get offloaded rules from
other higher level devices (such as tunnels), instead of ndo_setup_tc().
Since the decap rules are applied to the tunnel devices (e.g, vxlan_sys),
the driver should register for indirect TC callback with tunnel devices
to get the rules for offloading. This patch updates the driver to
register and process indirect TC block callbacks from VXLAN tunnels.
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Provides support for modifying L3/L4 Header parameters to support NAT.
Sets the appropriate fields/bits in cfa_flow_alloc cmd.

Sample cmd for offloading an IPv4 flow with SNAT:

ovs-ofctl add-flow ovsbr0 "ip,nw_src=192.168.201.44 \
actions=mod_nw_src:203.31.220.144,output:p7p1"

Replace 'nw_src' with 'nw_dst' in above cmd for DNAT with IPv4

Sample cmd for offloading an IPv4 flow with SNAPT:

ovs-ofctl add-flow ovsbr0 "ip,nw_src=192.168.201.44 \
actions=mod_nw_src:203.31.220.144, mod_tp_src:6789,output:p7p1"

Similar to DNAT, replace 'tp_src' with 'tp_dst' for offloading flow
with DNAPT

Sample cmd for offloading an IPv6 flow with SNAT:

ovs-ofctl add-flow ovsbr0 "ipv6, ipv6_src=2001:5c0:9168::2/64 \
actions=load:0x1->NXM_NX_IPV6_SRC[0..63], \
load:0x20010db801920000->NXM_NX_IPV6_SRC[64..127],output:p7p1"

Replace 'SRC' with DST' above for IPv6 DNAT
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For every single case where bnxt_tc_can_offload() can fail, we are
logging a user friendly descriptive message anyway, but because of the
path it would take in case of failure, another redundant error message
would get logged. Just freeing the node and returning from the point of
failure should suffice.
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds support for packet edit offload of L2 fields (src mac &
dst mac, also referred as L2 rewrite). Only when the mask is fully exact
match for a field, the command is sent down to the adapter to offload
such a flow. Otherwise, an error is returned.

v2: Fix pointer alignment issue in bnxt_fill_l2_rewrite_fields() [MChan]
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vivien Didelot says:

====================
net: dsa: replace routing tables with a list

This branch gets rid of the ds->rtable static arrays in favor of
a single dst->rtable list. This allows us to move away from the
DSA_MAX_SWITCHES limitation and simplify the switch fabric setup.

Changes in v2:
  - fix the reverse christmas for David
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Now that there's no restriction from the DSA core side regarding
the switch IDs and port numbers, only tag_8021q which is currently
reserving 3 bits for the switch ID and 4 bits for the port number, has
limitation for these values. Update their descriptions to reflect that.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Because there is no static array describing the links between switches
anymore, we have no reason to force a limitation of the index value
set by the device tree.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The DSA fabric setup code has been simplified a lot so get rid of
the dsa_tree_remove_switch, dsa_tree_add_switch and dsa_switch_add
helpers, and keep the code simple with only the dsa_switch_probe and
dsa_switch_remove functions.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Now that the DSA ports are listed in the switch fabric, there is
no need to store the dsa_switch structures from the drivers in the
fabric anymore. So get rid of the dst->ds static array.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The dsa_switch structure has no routing table specific data to setup,
so the switch fabric can directly walk its ports and initialize its
routing table from them.

This allows us to remove the dsa_switch_setup_routing_table function.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Drivers do not use the ds->rtable static arrays anymore, get rid of it.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Implement a new list of DSA links in the switch fabric itself, to
provide an alterative to the ds->rtable static arrays.

At the same time, provide a new dsa_routing_port() helper to abstract
the usage of ds->rtable in drivers. If there's no port to reach a
given device, return the first invalid port, ds->num_ports. This avoids
potential signedness errors or the need to define special values.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ioana Ciornei says:

====================
dpaa2-eth: add MAC/PHY support through phylink

The dpaa2-eth driver now has support for connecting to its associated PHY
device found through standard OF bindings. The PHY interraction is handled
by PHYLINK and even though, at the moment, only RGMII_* phy modes are
supported by the driver, this is just the first step into adding the
necessary changes to support the entire spectrum of capabilities.

This comes after feedback on the initial DPAA2 MAC RFC submitted here:
https://lwn.net/Articles/791182/

The notable change is that now, the DPMAC is not a separate driver, and
communication between the DPMAC and DPNI no longer happens through
firmware. Rather, the DPMAC is now a set of API functions that other
net_device drivers (DPNI, DPSW, etc) can use for PHY management.

The change is incremental, because the DPAA2 architecture has many modes of
connecting net devices in hardware loopback (for example DPNI to DPNI).
Those operating modes do not have a DPMAC and phylink instance.

The documentation patch provides a more complete view of the software
architecture and the current implementation.

Changes in v2:
 - added patch 1/5 in order to fix module build
 - use -ENOTCONN as a proper return error of dprc_get_connection()
 - move the locks to rtnl outside of dpaa2_eth_[dis]connect_mac functions
 - remove setting supported/advertised from .validate()

Changes in v3:
 - remove an unused variable

Changes in v4:
 - use ERR_PTR instead of plain NULL
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add documentation file for the MAC/PHY support in the DPAA2
architecture. This describes the architecture and implementation of the
interface between phylink and a DPAA2 network driver.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The dpaa2-eth driver now has support for connecting to its associated
PHY device found through standard OF bindings.

This happens when the DPNI object (that the driver probes on) gets
connected to a DPMAC. When that happens, the device tree is looked up by
the DPMAC ID, and the associated PHY bindings are found.

The old logic of handling the net device's link state by hand still
needs to be kept, as the DPNI can be connected to other devices on the
bus than a DPMAC: other DPNI, DPSW ports, etc. This logic is only
engaged when there is no DPMAC (and therefore no phylink instance)
attached.

The MC firmware support multiple type of DPMAC links: TYPE_FIXED,
TYPE_PHY. The TYPE_FIXED mode does not require any DPMAC management from
Linux side, and as such, the driver will not handle such a DPMAC.

Although PHYLINK typically handles SFP cages and in-band AN modes, for
the moment the driver only supports the RGMII interfaces found on the
LX2160A. Support for other modes will come later.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently the function is called at every link up event, although the
FQID values will only change when the DPNI is disconnected from the
current object and reconnected to a different one.

The patch also avoids the forward declaration of update_tx_fqids.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Using the newly added fsl_mc_get_endpoint function a fsl-mc driver can
find its associated endpoint (another object at the other link of a MC
firmware link).

The API will be used in the following patch in order to discover the
connected DPMAC object of a DPNI.

Also, the fsl_mc_device_lookup function is made available to the entire
fsl-mc bus driver and not just for the dprc driver.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Export all device types present on the fsl-mc bus in order to be able to
actually use the is_fsl_mc_bus_*() functions from drivers on the bus.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Charles McLachlan says:

====================
sfc: Add XDP support

Supply the XDP callbacks in netdevice ops that enable lower level processing
of XDP frames.

Changes in v4:
- Handle the failure to send some frames in efx_xdp_tx_buffers() properly.

Changes in v3:
- Fix a BUG_ON when trying to allocate piobufs to xdp queues.
- Add a missed trace_xdp_exception.

Changes in v2:
- Use of xdp_return_frame_rx_napi() in tx.c
- Addition of xdp_rxq_info_valid and xdp_rxq_info_failed to track when
  xdp_rxq_info failures occur.
- Renaming of rc to err and more use of unlikely().
- Cut some duplicated code and fix an array overrun.
- Actually increment n_rx_xdp_tx when packets are transmitted.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Count XDP packet drops, error drops, transmissions and redirects and
expose these counters via the ethtool stats command.
Signed-off-by: Charles McLachlan <cmclachlan@solarflare.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Provide an ndo_xdp_xmit function that uses the XDP tx queue for this
CPU to send the packet.
Signed-off-by: Charles McLachlan <cmclachlan@solarflare.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Each CPU needs access to its own queue to allow uncontested
transmission of XDP_TX packets. This means we need to allocate (up
front) enough channels ("xdp transmit channels") to provide at least
one extra tx queue per CPU. These tx queues should not do TSO.
Signed-off-by: Charles McLachlan <cmclachlan@solarflare.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Provide an ndo_bpf function to efx_netdev_ops that allows setting and
querying of xdp programs on an interface.

Also check that the MTU size isn't too big when setting a program or
when the MTU is explicitly set.
Signed-off-by: Charles McLachlan <cmclachlan@solarflare.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Adds a field to hold an attached xdp_prog, but never populates it (see
following patch).  Also, XDP_TX support is deferred to a later patch
in the series.

Track failures of xdp_rxq_info_reg() via per-queue xdp_rxq_info_valid
flags and a per-nic xdp_rxq_info_failed flag. The per-queue flags are
needed to prevent attempts to xdp_rxq_info_unreg() structs that failed
to register.  Possibly the API could be changed in the future to avoid
the need for these flags.
Signed-off-by: Charles McLachlan <cmclachlan@solarflare.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>