Test against MLX4_MAX_DESC_TXBBS only matters if the TX
bounce buffer is going to be used.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Google production kernel has increased MAX_SKB_FRAGS to 45
for BIG-TCP rollout.

Unfortunately mlx4 TX bounce buffer is not big enough whenever
an skb has up to 45 page fragments.

This can happen often with TCP TX zero copy, as one frag usually
holds 4096 bytes of payload (order-0 page).

Tested:
 Kernel built with MAX_SKB_FRAGS=45
 ip link set dev eth0 gso_max_size 185000
 netperf -t TCP_SENDFILE

I made sure that "ethtool -G eth0 tx 64" was properly working,
ring->full_size being set to 15.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Wei Wang <weiwan@google.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

MAX_DESC_SIZE is really the size of the bounce buffer used
when reaching the right side of TX ring buffer.

MAX_DESC_TXBBS get a MLX4_ prefix.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Saeed Mahameed says:

====================
Support tc police jump conform-exceed attribute

The tc police action conform-exceed option defines how to handle
packets which exceed or conform to the configured bandwidth limit.
One of the possible conform-exceed values is jump, which skips over
a specified number of actions.
This series adds support for conform-exceed jump action.

The series adds platform support for branching actions by providing
true/false flow attributes to the branching action.
This is necessary for supporting police jump, as each branch may
execute a different action list.

The first five patches are preparation patches:
- Patches 1 and 2 add support for actions with no destinations (e.g. drop)
- Patch 3 refactor the code for subsequent function reuse
- Patch 4 defines an abstract way for identifying terminating actions
- Patch 5 updates action list validations logic considering branching actions

The following three patches introduce an interface for abstracting branching
actions:
- Patch 6 introduces an abstract api for defining branching actions
- Patch 7 generically instantiates the branching flow attributes using
  the abstract API

Patch 8 adds the platform support for jump actions, by executing the following
sequence:
  a. Store the jumping flow attr
  b. Identify the jump target action while iterating the actions list.
  c. Instantiate a new flow attribute after the jump target action.
     This is the flow attribute that the branching action should jump to.
  d. Set the target post action id on:
    d.1. The jumping attribute, thus realizing the jump functionality.
    d.2. The attribute preceding the target jump attr, if not terminating.

The next patches apply the platform's branching attributes to the police
action:
- Patch 9 is a refactor patch
- Patch 10 initializes the post meter table with the red/green flow attributes,
           as were initialized by the platform
- Patch 11 enables the offload of meter actions using jump conform-exceed
           value.
====================

Link: https://lore.kernel.org/all/20221203221337.29267-1-saeed@kernel.org/Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Separate the matchall police action validation from flower validation.
Isolate the action validation logic in the police action parser.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-12-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Instantiate the post meter actions with the platform initialized branching
action attributes.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-11-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Currently post meter supports only the pipe/drop conform-exceed policy.
This assumption is reflected in several variable names.
Rename the following variables as a pre-step for using the generalized
branching action platform.

Rename fwd_green_rule/drop_red_rule to green_rule/red_rule respectively.
Repurpose red_counter/green_counter to act_counter/drop_counter to allow
police conform-exceed configurations that do not drop.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-10-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Identify the jump target action when iterating the action list.
Initialize the jump target attr with the jumping attribute during the
parsing phase. Initialize the jumping attr post action with the target
during the offload phase.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-9-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Initialize flow attribute for drop, accept, pipe and jump branching actions.

Instantiate a flow attribute instance according to the specified branch
control action. Store the branching attributes on the branching action
flow attribute during the parsing phase. Then, during the offload phase,
allocate the relevant mod header objects to the branching actions.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-8-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Extend the act tc api to set the branch control params aligning with
the police conform/exceed use case.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-7-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Currently the entire flow action list is validate for offload limitations.
For example, flow with both forward and drop actions are declared invalid
due to hardware restrictions.
However, a multi-table hardware model changes the limitations from a flow
scope to a single flow attribute scope.

Apply offload limitations to flow attributes instead of the entire flow.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-6-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Extend act api to identify actions that terminate action list.
Pre-step for terminating branching actions.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-5-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

After the tc action parsing phase the flow attribute is initialized with
relevant eswitch offload objects such as tunnel, vlan, header modify and
counter attributes. The post processing is done both for fdb and post-action
attributes.

Reuse the flow attribute post parsing logic by both fdb and post-action
offloads.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-4-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Currently create_flow_handle() assumes a null dest pointer when there
are no destinations.
This might not be the case as the caller may pass an allocated dest
array while setting the dest_num parameter to 0.

Assert null dest array for flow rules that have no destinations (e.g. drop
rule).
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-3-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Rules with drop action are not required to have a destination.
Currently the destination list is allocated with the maximum number of
destinations and passed to the fs_core layer along with the actual number
of destinations.

Remove redundant passing of dest pointer when count of dest is 0.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221203221337.29267-2-saeed@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Add stats64 support for ksz8xxx series of switches.
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Link: https://lore.kernel.org/r/20221205052904.2834962-1-o.rempel@pengutronix.deSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Roger Quadros says:

====================
net: ethernet: ti: am65-cpsw: Fix set channel operation

This contains a critical bug fix for the recently merged suspend/resume
support [1] that broke set channel operation. (ethtool -L eth0 tx <n>)

As there were 2 dependent patches on top of the offending commit [1]
first revert them and then apply them back after the correct fix.

[1] fd23df72 ("net: ethernet: ti: am65-cpsw: Add suspend/resume support")
====================

Link: https://lore.kernel.org/r/20221206094419.19478-1-rogerq@kernel.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

On low power during system suspend the ALE table context is lost.
Save the ALE context before suspend and restore it after resume.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

During suspend resume the context of PORT_VLAN_REG is lost so
save it during suspend and restore it during resume for
host port and slave ports.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Add PM handlers for System suspend/resume.

As DMA driver doesn't yet support suspend/resume we free up
the DMA channels at suspend and acquire and initialize them
at resume.

In this revised approach we do not free the TX/RX IRQs at
am65_cpsw_nuss_common_stop() as it causes problems.
We will now free them only on .suspend() as we need to release
the DMA channels (as DMA looses context) and re-acquiring
them on .resume() may not necessarily give us the same
IRQs.

To make this easier:
- introduce am65_cpsw_nuss_remove_rx_chns() which is
   similar to am65_cpsw_nuss_remove_tx_chns(). These will
   be invoked in pm.suspend() to release the DMA channels
   and free up the IRQs.
- move napi_add() and request_irq() calls to
   am65_cpsw_nuss_init_rx/tx_chns() so we can invoke them
   in pm.resume() to acquire the DMA channels and IRQs.

As CPTS looses contect during suspend/resume, invoke the
necessary CPTS suspend/resume helpers.

ALE_CLEAR command is issued in cpsw_ale_start() so no need
to issue it before the call to cpsw_ale_start().
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

This reverts commit fd23df72.

This commit broke set channel operation. Revert this and
implement it with a different approach in a separate patch.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

This reverts commit 643cf0e3.

This is to make it easier to revert the offending commit
fd23df72 ("net: ethernet: ti: am65-cpsw: Add suspend/resume support")
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

This reverts commit 1af3cb37.

This is to make it easier to revert the offending commit
fd23df72 ("net: ethernet: ti: am65-cpsw: Add suspend/resume support")
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Shay Drory says:

====================
devlink: Add port function attribute to enable/disable Roce and migratable

This series is a complete rewrite of the series "devlink: Add port
function attribute to enable/disable roce"
link:
https://lore.kernel.org/netdev/20221102163954.279266-1-danielj@nvidia.com/

Currently mlx5 PCI VF and SF are enabled by default for RoCE
functionality. And mlx5 PCI VF is disable by dafault for migratable
functionality.

Currently a user does not have the ability to disable RoCE for a PCI
VF/SF device before such device is enumerated by the driver.

User is also incapable to do such setting from smartnic scenario for a
VF from the smartnic.

Current 'enable_roce' device knob is limited to do setting only at
driverinit time. By this time device is already created and firmware has
already allocated necessary system memory for supporting RoCE.

Also, Currently a user does not have the ability to enable migratable
for a PCI VF.

The above are a hyper visor level control, to set the functionality of
devices passed through to guests.

This is achieved by extending existing 'port function' object to control
capabilities of a function. This enables users to control capability of
the device before enumeration.

Examples when user prefers to disable RoCE for a VF when using switchdev
mode:

$ devlink port show pci/0000:06:00.0/1
pci/0000:06:00.0/1: type eth netdev pf0vf0 flavour pcivf controller 0
pfnum 0 vfnum 0 external false splittable false
  function:
    hw_addr 00:00:00:00:00:00 roce enable

$ devlink port function set pci/0000:06:00.0/1 roce disable

$ devlink port show pci/0000:06:00.0/1
pci/0000:06:00.0/1: type eth netdev pf0vf0 flavour pcivf controller 0
pfnum 0 vfnum 0 external false splittable false
  function:
    hw_addr 00:00:00:00:00:00 roce disable

FAQs:
-----
1. What does roce enable/disable do?
Ans: It disables RoCE capability of the function before its enumerated,
so when driver reads the capability from the device firmware, it is
disabled.
At this point RDMA stack will not be able to create UD, QP1, RC, XRC
type of QPs. When RoCE is disabled, the GID table of all ports of the
device is disabled in the device and software stack.

2. How is the roce 'port function' option different from existing
devlink param?
Ans: RoCE attribute at the port function level disables the RoCE
capability at the specific function level; while enable_roce only does
at the software level.

3. Why is this option for disabling only RoCE and not the whole RDMA
device?
Ans: Because user still wants to use the RDMA device for non RoCE
commands in more memory efficient way.
====================

Link: https://lore.kernel.org/r/20221206185119.380138-1-shayd@nvidia.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Implement devlink port function commands to enable / disable migratable.
This is used to control the migratable capability of the device.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Acked-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Expose port function commands to enable / disable migratable
capability, this is used to set the port function as migratable.

Live migration is the process of transferring a live virtual machine
from one physical host to another without disrupting its normal
operation.

In order for a VM to be able to perform LM, all the VM components must
be able to perform migration. e.g.: to be migratable.
In order for VF to be migratable, VF must be bound to VFIO driver with
migration support.

When migratable capability is enabled for a function of the port, the
device is making the necessary preparations for the function to be
migratable, which might include disabling features which cannot be
migrated.

Example of LM with migratable function configuration:
Set migratable of the VF's port function.

$ devlink port show pci/0000:06:00.0/2
pci/0000:06:00.0/2: type eth netdev enp6s0pf0vf1 flavour pcivf pfnum 0
vfnum 1
    function:
        hw_addr 00:00:00:00:00:00 migratable disable

$ devlink port function set pci/0000:06:00.0/2 migratable enable

$ devlink port show pci/0000:06:00.0/2
pci/0000:06:00.0/2: type eth netdev enp6s0pf0vf1 flavour pcivf pfnum 0
vfnum 1
    function:
        hw_addr 00:00:00:00:00:00 migratable enable

Bind VF to VFIO driver with migration support:
$ echo <pci_id> > /sys/bus/pci/devices/0000:08:00.0/driver/unbind
$ echo mlx5_vfio_pci > /sys/bus/pci/devices/0000:08:00.0/driver_override
$ echo <pci_id> > /sys/bus/pci/devices/0000:08:00.0/driver/bind

Attach VF to the VM.
Start the VM.
Perform LM.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Acked-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Implement devlink port function commands to enable / disable RoCE.
This is used to control the RoCE device capabilities.

This patch implement infrastructure which will be used by downstream
patches that will add additional capabilities.
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Acked-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Downstream patch requires to get other function GENERAL2 caps while
mlx5_vport_get_other_func_cap() gets only one type of caps (general).
Rename it to represent this and introduce a generic implementation
of mlx5_vport_get_other_func_cap().
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Acked-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Expose port function commands to enable / disable RoCE, this is used to
control the port RoCE device capabilities.

When RoCE is disabled for a function of the port, function cannot create
any RoCE specific resources (e.g GID table).
It also saves system memory utilization. For example disabling RoCE enable a
VF/SF saves 1 Mbytes of system memory per function.

Example of a PCI VF port which supports function configuration:
Set RoCE of the VF's port function.

$ devlink port show pci/0000:06:00.0/2
pci/0000:06:00.0/2: type eth netdev enp6s0pf0vf1 flavour pcivf pfnum 0
vfnum 1
    function:
        hw_addr 00:00:00:00:00:00 roce enable

$ devlink port function set pci/0000:06:00.0/2 roce disable

$ devlink port show pci/0000:06:00.0/2
pci/0000:06:00.0/2: type eth netdev enp6s0pf0vf1 flavour pcivf pfnum 0
vfnum 1
    function:
        hw_addr 00:00:00:00:00:00 roce disable
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink port function hw_addr attr documentation is in mlx5 specific
file while there is nothing mlx5 specific about it.
Move it to devlink-port.rst.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

In order to avoid partial request processing, validate the request
before processing it.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Introduce IFC related capabilities to enable setting VF to be able to
perform live migration. e.g.: to be migratable.
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Acked-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Ido Schimmel says:

====================
bridge: mcast: Preparations for EVPN extensions

This patchset was split from [1] and includes non-functional changes
aimed at making it easier to add additional netlink attributes later on.
Future extensions are available here [2].

The idea behind these patches is to create an MDB configuration
structure into which netlink messages are parsed into. The structure is
then passed in the entry creation / deletion call chain instead of
passing the netlink attributes themselves. The same pattern is used by
other rtnetlink objects such as routes and nexthops.

I initially tried to extend the current code, but it proved to be too
difficult, which is why I decided to refactor it to the extensible and
familiar pattern used by other rtnetlink objects.

Tested using existing selftests and using a new selftest that will be
submitted together with the planned extensions.

[1] https://lore.kernel.org/netdev/20221018120420.561846-1-idosch@nvidia.com/
[2] https://github.com/idosch/linux/commits/submit/mdb_v1
====================

Link: https://lore.kernel.org/r/20221206105809.363767-1-idosch@nvidia.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

The 'group' argument is not modified, so mark it as 'const'. It will
allow us to constify arguments of the callers of this function in future
patches.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Drop the first three arguments and instead extract them from the MDB
configuration structure.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The checks only require information parsed from the RTM_NEWMDB netlink
message and do not rely on any state stored in the bridge driver.
Therefore, there is no need to perform the checks in the critical
section under the multicast lock.

Move the checks out of the critical section.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The parsing of the netlink messages and the validity checks are now
performed in br_mdb_config_init() so we can remove br_mdb_parse().

This finally allows us to stop passing netlink attributes deep in the
MDB control path and only use the MDB configuration structure.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The MDB group key (i.e., {source, destination, protocol, VID}) is
currently determined under the multicast lock from the netlink
attributes. Instead, use the group key from the MDB configuration
structure that was prepared before acquiring the lock.

No functional changes intended.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

As an intermediate step towards only using the new MDB configuration
structure, pass it further in the control path instead of passing
individual attributes.

No functional changes intended.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The MDB configuration structure (i.e., struct br_mdb_config) now
includes all the necessary information from the parsed RTM_{NEW,DEL}MDB
netlink messages, so use it.

This will later allow us to delete the calls to br_mdb_parse() from
br_mdb_add() and br_mdb_del().

No functional changes intended.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>