There are two peculiarities about offloading FIFO:

- sometimes the qdisc has an unspecified handle (it is "invisible")
- it may be created before the qdisc that it will be a child of

These features make the offload a bit more tricky. The approach chosen in
this patch is to make note of all the FIFOs that needed to be rejected
because their parents were not known. Later when the parent is created,
they are offloaded

FIFO is only offloaded for its counters, queue length is ignored.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

PRIO and ETS will need to check the value of qdisc handle in their
handlers. Add it to the callback and propagate through.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In order to have a tidy structure where to put information related to Qdisc
offloads, introduce a new structure. Move there the two existing pieces of
data: root_qdisc and tclass_qdiscs. Embed them directly, because there's no
reason to go through pointer anymore. Convert users, update init/fini
functions.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Set ethtool_ops->supported_coalesce_params to let
the core reject unsupported coalescing parameters.

This driver did not previously reject unsupported parameters.

v3: adjust commit message for new member name
Signed-off-by: NJakub Kicinski <kuba@kernel.org>
Acked-by: NSaeed Mahameed <saeedm@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:

struct foo {
        int stuff;
        struct boo array[];
};

By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.

Also, notice that, dynamic memory allocations won't be affected by
this change:

"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]

This issue was found with the help of Coccinelle.

[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 76497732 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: NGustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use newly introduce 'virtual' port flavour for devlink
port of PCI VF devlink device in non-representors mode.

While at it, remove recently introduced empty lines at end of the file.
Reviewed-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NParav Pandit <parav@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The code is self explanatory and makes the comment redundant.
Signed-off-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

mlx5e_tc_offload_to_slow_path() and mlx5e_tc_unoffload_from_slow_path()
take an extra argument allocated on the stack of the caller but not used
by the caller. Avoid the extra argument and use local variable in the
function itself.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

parse_attr is not used by parse_tc_pedit_action() so revmove it.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

This to be consistent and adds the module name to the error message.
Signed-off-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

This is for added netdev prefix that helps identify
the source of the message.
Signed-off-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

This helps identify the source of the message.
If netdev still doesn't exists use mlx5_core_warn().
Signed-off-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NVlad Buslov <vladbu@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Few print messages are in debug level where they should be in error, and
few messages are missing.
Signed-off-by: NErez Shitrit <erezsh@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Change matcher priority parameter type from u16 to u32,
this change is needed since sometimes upper levels
create a matcher with priority bigger than 2^16.
Signed-off-by: NHamdan Igbaria <hamdani@mellanox.com>
Reviewed-by: NAlex Vesker <valex@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Add a devlink parameter to control the number of large groups in a
autogrouped flow table. The default value is 15, and the range is between 1
and 1024.

The size of each large group can be calculated according to the following
formula: size = 4M / (fdb_large_groups + 1).

Examples:
- Set the number of large groups to 20.
    $ devlink dev param set pci/0000:82:00.0 name fdb_large_groups \
      cmode driverinit value 20

  Then run devlink reload command to apply the new value.
    $ devlink dev reload pci/0000:82:00.0

- Read the number of large groups in flow table.
    $ devlink dev param show pci/0000:82:00.0 name fdb_large_groups
    pci/0000:82:00.0:
      name fdb_large_groups type driver-specific
        values:
          cmode driverinit value 20
Signed-off-by: NJianbo Liu <jianbol@mellanox.com>
Reviewed-by: NVlad Buslov <vladbu@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

The prefix should be "MLX5_DEVLINK_PARAM_ID_" for all in
mlx5_devlink_param_id enum.
Signed-off-by: NJianbo Liu <jianbol@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Added devlink_port field to mlx5e_priv structure and a callback to
netdev ops to enable devlink to get info about the port. The port
registration happens at driver initialization.
Signed-off-by: NVladyslav Tarasiuk <vladyslavt@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>
Reviewed-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Rename representor's mlx5e_get_devlink_port() to
mlx5e_rep_get_devlink_port().
The downstream patch will add a non-representor mlx5e function called
mlx5e_get_devlink_phy_port().
Signed-off-by: NVladyslav Tarasiuk <vladyslavt@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Mellanox FW can support this if ignore_flow_level capability exists.
Signed-off-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NOz Shlomo <ozsh@mellanox.com>
Reviewed-by: NPaul Blakey <paulb@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

When using port mirroring, we forward the traffic to another table and
use that table to forward to the mirrored vport. Since the hardware
loses the values of reg c, and in particular reg c0, we fail the match
on the input vport which previously existed in reg c0. To overcome this
situation, we use a set of per vport tables, positioned at the lowest
priority, and forward traffic to those tables. Since these tables are
per vport, we can avoid matching on reg c0.

Fixes: c01cfd0f ("net/mlx5: E-Switch, Add match on vport metadata for rule in fast path")
Signed-off-by: NEli Cohen <eli@mellanox.com>
Reviewed-by: NMark Bloch <markb@mellanox.com>
Reviewed-by: NPaul Blakey <paulb@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

misc_params.source_port is a 16 bit field already so no need for
redundant masking against 0xffff. Also change local variables type to
u16.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

We can install forwarding packets rule between uplink
in switchdev mode, as show below. But the hardware does
not do that as expected (mlnx_perf -i $PF1, we can't get
the counter of the PF1). By the way, if we add the uplink
PF0, PF1 to Open vSwitch and enable hw-offload, the rules
can be offloaded but not work fine too. This patch add a
check and if so return -EOPNOTSUPP.

$ tc filter add dev $PF0 protocol all parent ffff: prio 1 handle 1 \
    flower skip_sw action mirred egress redirect dev $PF1

$ tc -d -s filter show dev $PF0 ingress
    skip_sw
    in_hw in_hw_count 1
    action order 1: mirred (Egress Redirect to device enp130s0f1) stolen
    ...
    Sent hardware 408954 bytes 4173 pkt
    ...
Signed-off-by: NTonghao Zhang <xiangxia.m.yue@gmail.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

During initialization the driver issues a reset to the device and waits
for 100ms before checking if the firmware is ready. The waiting is
necessary because before that the device is irresponsive and the first
read can result in a completion timeout.

While 100ms is sufficient for Spectrum-1 and Spectrum-2, it is
insufficient for Spectrum-3.

Fix this by increasing the timeout to 200ms.

Fixes: da382875 ("mlxsw: spectrum: Extend to support Spectrum-3 ASIC")
Signed-off-by: NAmit Cohen <amitc@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There are couple new values that PMTM register can return
in module_type field. Add them and map them to module width in
mlxsw_core_module_max_width(). Fix the existing names on the way.
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The current code causes problems when the unregistering netdevice could
be different then the registering one.

Since the check in mlx5_lag_netdev_event() does not allow any other
network namespace anyway, fix this by registerting the lag notifier
per init network namespace only.

Fixes: d48834f9 ("mlx5: Use dev_net netdevice notifier registrations")
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Tested-by: NAya Levin <ayal@mellanox.com>
Acked-by: NSaeed Mahameed <saeedm@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The buffer factor on Spectrum-3 is larger than on Spectrum-2. Add a new
callback and use it for mlxsw_sp->span_ops on Spectrum-3.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

During port initialization the driver instructs the device to only
advertise speeds that can be supported by the port's current width.

Since the device now returns the supported speeds based on the port's
current width, the driver no longer needs to compute the speeds that can
be advertised.

Simplify port initialization by setting the advertised speeds to the
queried supported speeds.
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Spectrum-1 and Spectrum-2 do not have a per-TC counter of number of packets
marked by ECN. The value reported currently is the total number of marked
packets. Showing this value at individual TC Qdiscs is misleading.

Move the counter to ethtool instead.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently, only one SFN query is done from repetitive work at a time,
processing 64 entries. Another work iteration is scheduled in 100ms,
that means that the max rate of learned FDB entries is limited to 6400/s.
That is slow. Fix this by doing 2 optimizations:
1) Run 10 SFN queries at a time.
2) In case the SFN is not drained, schedule work with 0 delay to allow
   to continue processing rest of the records.

On a testing setup with 500K entries the time to process decreased
from 870secs to 10secs.
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Tested-by: NAlex Kushnarov <alexanderk@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Return NULL instead of 0.
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>

drivers/net/ethernet/mellanox/mlx5/core/diag/fw_tracer.c:191:13:
sparse: warning: incorrect type in assignment (different base types)
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>

Clang warns:

 In file included from
 ../drivers/net/ethernet/mellanox/mlx5/core/main.c:73:
 ../drivers/net/ethernet/mellanox/mlx5/core/diag/rsc_dump.h:4:9: warning:
 '__MLX5_RSC_DUMP_H' is used as a header guard here, followed by #define
 of a different macro [-Wheader-guard]
 #ifndef __MLX5_RSC_DUMP_H
         ^~~~~~~~~~~~~~~~~
 ../drivers/net/ethernet/mellanox/mlx5/core/diag/rsc_dump.h:5:9: note:
 '__MLX5_RSC_DUMP__H' is defined here; did you mean '__MLX5_RSC_DUMP_H'?
 #define __MLX5_RSC_DUMP__H
         ^~~~~~~~~~~~~~~~~~
         __MLX5_RSC_DUMP_H
 1 warning generated.

Make them match to get the intended behavior and remove the warning.

Fixes: 12206b17 ("net/mlx5: Add support for resource dump")
Link: https://github.com/ClangBuiltLinux/linux/issues/897Signed-off-by: NNathan Chancellor <natechancellor@gmail.com>
Reviewed-by: NJason Gunthorpe <jgg@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

We can avoid an indirect call per compressed completion wrapping the
completion handling call with the appropriate helper.
Signed-off-by: NTariq Toukan <tariqt@mellanox.com>
Reviewed-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

We can avoid an indirect call per NAPI cycle wrapping the RX descriptors
posting call with the appropriate helper.
Signed-off-by: NTariq Toukan <tariqt@mellanox.com>
Reviewed-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

The current steps that are performed when the trust state changes, if
the channels are active:

1. The trust state is changed in hardware.

2. The new inline mode is calculated.

3. If the new inline mode is different, the channels are recreated using
the new inline mode.

This approach has some issues:

1. There is a time gap between changing trust state in hardware and
starting sending enough inline headers (the latter happens after
recreation of channels). It leads to failed transmissions and error
CQEs.

2. If the new channels fail to open, we'll be left with the old ones,
but the hardware will be configured for the new trust state, so the
interval when we can see TX errors never ends.

This patch fixes the issues above by moving the trust state change into
the preactivate hook that runs during the recreation of the channels
when no channels are active, so it eliminates the gap of partially
applied configuration. If the inline mode doesn't change with the change
of the trust state, the channels won't be recreated, just like before
this patch.
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Sometimes the preactivate hook of mlx5e_safe_switch_channels needs more
parameters than just struct mlx5e_priv *. For such cases, a new
parameter (void *context) is added to preactivate hooks.

Some of the existing normal functions are currently used as preactivate
callbacks. To avoid adding an extra unused parameter, they are wrapped
in an automatic way using the MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX
macro.
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Currently mlx5e_switch_priv_channels expects that the preactivate hook
doesn't fail, however, it can fail, because it may set hardware
parameters. This commit addresses this issue and provides a way to
recover from failures of the preactivate hook: the old channels are not
closed until the point where nothing can fail anymore, so in case
preactivate fails, the driver can roll back the old channels and
activate them again.
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

The number of queues is now updated by mlx5e_update_netdev_queues in a
centralized way, when no channels are active. Remove an extra occurrence
of netif_set_real_num_tx_queues to prepare it for the next commit.
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

Currently, mlx5e notifies the kernel about the number of queues and sets
the default XPS cpumasks when channels are activated. This
implementation has several corner cases, in which the kernel may not be
updated on time, or XPS cpumasks may be reset when not directly touched
by the user.

This commit fixes these corner cases to match the following expected
behavior:

1. The number of queues always corresponds to the number of channels
configured.

2. XPS cpumasks are set to driver's defaults on netdev attach.

3. XPS cpumasks set by user are not reset, unless the number of channels
changes. If the number of channels changes, they are reset to driver's
defaults. (In general case, when the number of channels increases or
decreases, it's not possible to guess how to convert the current XPS
cpumasks to work with the new number of channels, so we let the user
reconfigure it if they change the number of channels.)

XPS cpumasks are no longer stored per channel. Only one temporary
cpumask is used. The old stored cpumasks didn't reflect the user's
changes and were not used after applying them.

A scratchpad area is added to struct mlx5e_priv. As cpumask_var_t
requires allocation, and the preactivate hook can't fail, we need to
preallocate the temporary cpumask in advance. It's stored in the
scratchpad.

Fixes: 149e566f ("net/mlx5e: Expand XPS cpumask to cover all online cpus")
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>

mlx5e_ethtool_set_channels updates the indirection table before
switching to the new channels. If the switch fails, the indirection
table is new, but the channels are old, which is wrong. Fix it by using
the preactivate hook of mlx5e_safe_switch_channels to update the
indirection table at the stage when nothing can fail anymore.

As the code that updates the indirection table is now encapsulated into
a new function, use that function in the attach flow when the driver has
to reduce the number of channels, and prepare the code for the next
commit.

Fixes: 85082dba ("net/mlx5e: Correctly handle RSS indirection table when changing number of channels")
Signed-off-by: NMaxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>