platform/kernel/linux-starfive.git
3 years agonet/mlx5e: Allow RQ outside of channel context
Aya Levin [Tue, 1 Dec 2020 22:41:56 +0000 (14:41 -0800)]
net/mlx5e: Allow RQ outside of channel context

In order to be able to create an RQ outside of a channel context, remove
rq->channel direct pointer. This requires adding a direct pointer to:
ICOSQ and priv in order to support RQs that are part of mlx5e_channel.
Use channel_stats from the corresponding CQ.

Signed-off-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Eran Ben Elisha <eranbe@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Allow CQ outside of channel context
Aya Levin [Tue, 1 Dec 2020 22:41:55 +0000 (14:41 -0800)]
net/mlx5e: Allow CQ outside of channel context

In order to be able to create a CQ outside of a channel context, remove
cq->channel direct pointer. This requires adding a direct pointer to
channel statistics, netdevice, priv and to mlx5_core in order to support
CQs that are a part of mlx5e_channel.
In addition, parameters the were previously derived from the channel
like napi, NUMA node, channel stats and index are now assembled in
struct mlx5e_create_cq_param which is given to mlx5e_open_cq() instead
of channel pointer. Generalizing mlx5e_open_cq() allows opening CQ
outside of channel context which will be used in following patches in
the patch-set.

Signed-off-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Eran Ben Elisha <eranbe@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Free drop RQ in a dedicated function
Tariq Toukan [Tue, 1 Dec 2020 22:41:54 +0000 (14:41 -0800)]
net/mlx5e: Free drop RQ in a dedicated function

The drop RQ has very limited objects to be freed, and differs
from regular RQs in the context that it is freed from.
Add a dedicated function for it, use it where needed, and remove
the drop_rq-specific checks in the generic function.

Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Jakub Kicinski [Tue, 8 Dec 2020 02:36:21 +0000 (18:36 -0800)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

Saeed Mahameed says:

====================
mlx5-next auxbus support

This pull request is targeting net-next and rdma-next branches.

This series provides mlx5 support for auxiliary bus devices.

It starts with a merge commit of tag 'auxbus-5.11-rc1' from
gregkh/driver-core into mlx5-next, then the mlx5 patches that will convert
mlx5 ulp devices (netdev, rdma, vdpa) to use the proper auxbus
infrastructure instead of the internal mlx5 device and interface management
implementation, which Leon is deleting at the end of this patchset.

Link: https://lore.kernel.org/alsa-devel/20201026111849.1035786-1-leon@kernel.org/
Thanks to everyone for the joint effort !

* 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
  RDMA/mlx5: Remove IB representors dead code
  net/mlx5: Simplify eswitch mode check
  net/mlx5: Delete custom device management logic
  RDMA/mlx5: Convert mlx5_ib to use auxiliary bus
  net/mlx5e: Connect ethernet part to auxiliary bus
  vdpa/mlx5: Connect mlx5_vdpa to auxiliary bus
  net/mlx5: Register mlx5 devices to auxiliary virtual bus
  vdpa/mlx5: Make hardware definitions visible to all mlx5 devices
  net/mlx5_core: Clean driver version and name
  net/mlx5: Properly convey driver version to firmware
  driver core: auxiliary bus: minor coding style tweaks
  driver core: auxiliary bus: make remove function return void
  driver core: auxiliary bus: move slab.h from include file
  Add auxiliary bus support
====================

Link: https://lore.kernel.org/r/20201207053349.402772-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet/mlx4: Remove unused #define MAX_MSIX_P_PORT
Tariq Toukan [Sun, 6 Dec 2020 09:12:54 +0000 (11:12 +0200)]
net/mlx4: Remove unused #define MAX_MSIX_P_PORT

All usages of the definition MAX_MSIX_P_PORT were removed.
It's not in use anymore. Remove it.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Link: https://lore.kernel.org/r/20201206091254.12476-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agobonding: set xfrm feature flags more sanely
Jarod Wilson [Sat, 5 Dec 2020 17:40:03 +0000 (12:40 -0500)]
bonding: set xfrm feature flags more sanely

We can remove one of the ifdef blocks here, and instead of setting both
the xfrm hw_features and features flags, then unsetting the feature
flags if not in AB, wait to set the features flags if we're actually in AB
mode.

Signed-off-by: Jarod Wilson <jarod@redhat.com>
Link: https://lore.kernel.org/r/20201205174003.578267-1-jarod@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoibmvnic: add some debugs
Sukadev Bhattiprolu [Sat, 5 Dec 2020 02:22:35 +0000 (18:22 -0800)]
ibmvnic: add some debugs

We sometimes run into situations where a soft/hard reset of the adapter
takes a long time or fails to complete. Having additional messages that
include important adapter state info will hopefully help understand what
is happening, reduce the guess work and minimize requests to reproduce
problems with debug patches.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Link: https://lore.kernel.org/r/20201205022235.2414110-1-sukadev@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 's390-qeth-next'
David S. Miller [Mon, 7 Dec 2020 14:55:21 +0000 (06:55 -0800)]
Merge branch 's390-qeth-next'

Julian Wiedmann says:

====================
s390/qeth: updates 2020-12-07

please apply the following patch series for qeth to netdev's net-next tree.

Some sysfs cleanups (with the prep work in ccwgroup acked by Heiko), and
a few improvements to the code that deals with async TX completion
notifications for IQD devices.

This also brings the missing patch from the previous net-next submission.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: make qeth_qdio_handle_aob() more robust
Julian Wiedmann [Mon, 7 Dec 2020 13:12:33 +0000 (14:12 +0100)]
s390/qeth: make qeth_qdio_handle_aob() more robust

When qeth_qdio_handle_aob() frees dangling allocations in the notified
TX buffer, there are rare tear-down cases where
qeth_drain_output_queue() would later call qeth_clear_output_buffer()
for the same buffer - and thus end up walking the buffer a second time
to check for dangling kmem_cache allocations.

Luckily current code previously scrubs such a buffer, so
qeth_clear_output_buffer() would find buf->buffer->element[i].addr as
NULL and not do anything. But this is fragile, and we can easily improve
it by consistently clearing the ->is_header flag after freeing the
allocation.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: remove QETH_QDIO_BUF_HANDLED_DELAYED state
Julian Wiedmann [Mon, 7 Dec 2020 13:12:32 +0000 (14:12 +0100)]
s390/qeth: remove QETH_QDIO_BUF_HANDLED_DELAYED state

Reuse the QETH_QDIO_BUF_EMPTY state to indicate that a TX buffer has
been completed with a QAOB notification, and may be cleaned up by
qeth_cleanup_handled_pending().

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: don't replace a fully completed async TX buffer
Julian Wiedmann [Mon, 7 Dec 2020 13:12:31 +0000 (14:12 +0100)]
s390/qeth: don't replace a fully completed async TX buffer

For TX buffers that require an additional async notification via QAOB, the
TX completion code can now manage all the necessary processing if the
notification has already occurred (or is occurring concurrently).

In such cases we can avoid replacing the metadata that is associated
with the buffer's slot on the ring, and just keep using the current one.

As qeth_clear_output_buffer() will also handle any kmem cache-allocated
memory that was mapped into the TX buffer, qeth_qdio_handle_aob()
doesn't need to worry about it.

While at it, also remove the unneeded forward declaration for
qeth_init_qdio_out_buf().

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: use dev->groups for common sysfs attributes
Julian Wiedmann [Mon, 7 Dec 2020 13:12:30 +0000 (14:12 +0100)]
s390/qeth: use dev->groups for common sysfs attributes

All qeth devices have a minimum set of sysfs attributes, and non-OSN
devices share a group of additional attributes. Depending on whether
the device is forced to use a specific discipline, the device_type then
specifies further attributes.

Shift the common attributes into dev->groups, so that the device_type
only contains the discipline-specific attributes. This avoids exposing
the common attributes to the disciplines, and nicely cleans up our
sysfs code.

While replacing the qeth_l*_*_device_attributes() helpers, switch from
sysfs_*_groups() to the more generic device_*_groups().

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/ccwgroup: use bus->dev_groups for bus-based sysfs attributes
Julian Wiedmann [Mon, 7 Dec 2020 13:12:29 +0000 (14:12 +0100)]
s390/ccwgroup: use bus->dev_groups for bus-based sysfs attributes

Bus drivers have their own way of describing the sysfs attributes that
all devices on a bus should provide.
Switch ccwgroup_attr_groups over to use bus->dev_groups, and thus
free up dev->groups for usage by the ccwgroup device drivers.

While adjusting the attribute naming, use ATTRIBUTE_GROUPS() to get rid
of some boilerplate code.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: don't call INIT_LIST_HEAD() on iob's list entry
Julian Wiedmann [Mon, 7 Dec 2020 13:12:28 +0000 (14:12 +0100)]
s390/qeth: don't call INIT_LIST_HEAD() on iob's list entry

INIT_LIST_HEAD() only needs to be called on actual list heads.
While at it clarify the naming of the field.

Suggested-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 agoMerge branch 'mlxsw-Misc-updates'
David S. Miller [Mon, 7 Dec 2020 03:22:15 +0000 (19:22 -0800)]
Merge branch 'mlxsw-Misc-updates'
Ido Schimmel says:

====================
mlxsw: Misc updates

This patchset contains miscellaneous patches we gathered in our queue.
Some of them are dependencies of larger patchsets that I will submit
later this cycle.

Patches #1-#3 perform small non-functional changes in mlxsw.

Patch #4 adds more extended ack messages in mlxsw.

Patch #5 adds devlink parameters documentation for mlxsw. To be extended
with more parameters this cycle.

Patches #6-#7 perform small changes in forwarding selftests
infrastructure.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_router: Reduce mlxsw_sp_ipip_fib_entry_op_gre4()
Jiri Pirko [Sun, 6 Dec 2020 08:22:27 +0000 (10:22 +0200)]
mlxsw: spectrum_router: Reduce mlxsw_sp_ipip_fib_entry_op_gre4()

Turned out that mlxsw_sp_ipip_fib_entry_op_gre4() does not need to
figure out the IP address and virtual router id. Those are exactly
the same as in the fib_entry it is called for. So just use that and
reduce mlxsw_sp_ipip_fib_entry_op_gre4() function to only call
mlxsw_sp_ipip_fib_entry_op_gre4_rtdp() make the ipip decap op
code similar to nve.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum: Bump minimum FW version to xx.2008.2018
Petr Machata [Sun, 6 Dec 2020 08:22:26 +0000 (10:22 +0200)]
mlxsw: spectrum: Bump minimum FW version to xx.2008.2018

The indicated version fixes an issue whereby the MOMTE register would by
default enable mirroring of ECN-marked traffic from all traffic classes,
once the ECN mirroring was configured. This fix is necessary for offload
of RED "ecn_mark" qevent.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: core_acl: Use an array instead of a struct with a zero-length array
Ido Schimmel [Sun, 6 Dec 2020 08:22:25 +0000 (10:22 +0200)]
mlxsw: core_acl: Use an array instead of a struct with a zero-length array

Suppresses the following coccinelle warning:

drivers/net/ethernet/mellanox/mlxsw/core_acl_flex_keys.c:139:3-7:
WARNING use flexible-array member instead

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_mr: Use flexible-array member instead of zero-length array
Ido Schimmel [Sun, 6 Dec 2020 08:22:24 +0000 (10:22 +0200)]
mlxsw: spectrum_mr: Use flexible-array member instead of zero-length array

Suppresses the following coccinelle warning:

drivers/net/ethernet/mellanox/mlxsw/spectrum_mr.c:18:15-19: WARNING use flexible-array member instead

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: core: Trace EMAD events
Ido Schimmel [Sun, 6 Dec 2020 08:22:23 +0000 (10:22 +0200)]
mlxsw: core: Trace EMAD events

Currently, mlxsw triggers the 'devlink:devlink_hwmsg' tracepoint
whenever a request is sent to the device and whenever a response is
received from it. However, the tracepoint is not triggered when an event
(e.g., port up / down) is received from the device.

Also trace EMAD events in order to log a more complete picture of all
the exchanged hardware messages.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: mlxsw: Test RIF's reference count when joining a LAG
Ido Schimmel [Sun, 6 Dec 2020 08:22:22 +0000 (10:22 +0200)]
selftests: mlxsw: Test RIF's reference count when joining a LAG

Test that the reference count of a router interface (RIF) configured for
a LAG is incremented / decremented when ports join / leave the LAG. Use
the offload indication on routes configured on the RIF to understand if
it was created / destroyed.

The test fails without the previous patch.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum: Apply RIF configuration when joining a LAG
Ido Schimmel [Sun, 6 Dec 2020 08:22:21 +0000 (10:22 +0200)]
mlxsw: spectrum: Apply RIF configuration when joining a LAG

In case a router interface (RIF) is configured for a LAG, make sure its
configuration is applied on the new LAG member.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoRDMA/mlx5: Remove IB representors dead code
Leon Romanovsky [Sat, 10 Oct 2020 09:27:27 +0000 (12:27 +0300)]
RDMA/mlx5: Remove IB representors dead code

Delete dead code.

Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agonet/mlx5: Simplify eswitch mode check
Leon Romanovsky [Sat, 10 Oct 2020 08:57:26 +0000 (11:57 +0300)]
net/mlx5: Simplify eswitch mode check

Provide mlx5_core device instead of "priv" pointer while checking
eswith mode.

Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agonet/mlx5: Delete custom device management logic
Leon Romanovsky [Mon, 5 Oct 2020 08:13:56 +0000 (11:13 +0300)]
net/mlx5: Delete custom device management logic

After conversion to use auxiliary bus, all custom device management is
not needed anymore, delete it.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agoRDMA/mlx5: Convert mlx5_ib to use auxiliary bus
Leon Romanovsky [Sun, 4 Oct 2020 04:35:36 +0000 (07:35 +0300)]
RDMA/mlx5: Convert mlx5_ib to use auxiliary bus

The conversion to auxiliary bus solves long standing issue with
existing mlx5_ib<->mlx5_core coupling. It required to have both
modules in initramfs if one of them needed for the boot.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agonet/mlx5e: Connect ethernet part to auxiliary bus
Leon Romanovsky [Mon, 5 Oct 2020 05:08:25 +0000 (08:08 +0300)]
net/mlx5e: Connect ethernet part to auxiliary bus

Reuse auxiliary bus to perform device management of the
ethernet part of the mlx5 driver.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agovdpa/mlx5: Connect mlx5_vdpa to auxiliary bus
Leon Romanovsky [Sun, 4 Oct 2020 12:38:20 +0000 (15:38 +0300)]
vdpa/mlx5: Connect mlx5_vdpa to auxiliary bus

Change module registration logic to use auxiliary bus instead of custom
made mlx5 register interface.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agoMerge branch 'r8169-improve-rtl_rx-and-NUM_RX_DESC-handling'
David S. Miller [Sun, 6 Dec 2020 00:29:21 +0000 (16:29 -0800)]
Merge branch 'r8169-improve-rtl_rx-and-NUM_RX_DESC-handling'

Heiner Kallweit says:

====================
r8169: improve rtl_rx and NUM_RX_DESC handling

This series improves rtl_rx() and the handling of NUM_RX_DESC.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agor8169: make NUM_RX_DESC a signed int
Heiner Kallweit [Sun, 6 Dec 2020 00:00:54 +0000 (01:00 +0100)]
r8169: make NUM_RX_DESC a signed int

After recent changes there's no need any longer to define NUM_RX_DESC
as an unsigned value.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agor8169: improve rtl_rx
Heiner Kallweit [Sun, 6 Dec 2020 00:00:07 +0000 (01:00 +0100)]
r8169: improve rtl_rx

There's no need to check min(budget, NUM_RX_DESC). At first budget
(NAPI_POLL_WEIGHT = 64) is less then NUM_RX_DESC (256).
And more important: Even in case of budget > NUM_RX_DESC we could
safely continue processing descriptors as long as they are owned by
the CPU. In addition replace rx_left with a normal counter variable,
this allows to simplify the code. Last but not least there's no need
any longer to pass the budget as an u32.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: fix spelling mistake "wil" -> "will" in Kconfig
Colin Ian King [Fri, 4 Dec 2020 19:45:49 +0000 (19:45 +0000)]
net: fix spelling mistake "wil" -> "will" in Kconfig

There is a spelling mistake in the Kconfig help text. Fix it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Link: https://lore.kernel.org/r/20201204194549.1153063-1-colin.king@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge tag 'batadv-next-pullrequest-20201204' of git://git.open-mesh.org/linux-merge
Jakub Kicinski [Sat, 5 Dec 2020 23:08:05 +0000 (15:08 -0800)]
Merge tag 'batadv-next-pullrequest-20201204' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
This cleanup patchset includes the following patches:

 - bump version strings, by Simon Wunderlich

 - update include for min/max helpers, by Sven Eckelmann

 - add infrastructure and netlink functions for routing algo selection,
   by Sven Eckelmann (2 patches)

 - drop deprecated debugfs and sysfs support and obsoleted
   functionality, by Sven Eckelmann (3 patches)

 - drop unused include in fragmentation.c, by Simon Wunderlich

* tag 'batadv-next-pullrequest-20201204' of git://git.open-mesh.org/linux-merge:
  batman-adv: Drop unused soft-interface.h include in fragmentation.c
  batman-adv: Drop legacy code for auto deleting mesh interfaces
  batman-adv: Drop deprecated debugfs support
  batman-adv: Drop deprecated sysfs support
  batman-adv: Allow selection of routing algorithm over rtnetlink
  batman-adv: Prepare infrastructure for newlink settings
  batman-adv: Add new include for min/max helpers
  batman-adv: Start new development cycle
====================

Link: https://lore.kernel.org/r/20201204154631.21063-1-sw@simonwunderlich.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoenetc: Fix unused var build warning for CONFIG_OF
Arnd Bergmann [Fri, 4 Dec 2020 12:08:00 +0000 (14:08 +0200)]
enetc: Fix unused var build warning for CONFIG_OF

When CONFIG_OF is disabled, there is a harmless warning about
an unused variable:

enetc_pf.c: In function 'enetc_phylink_create':
enetc_pf.c:981:17: error: unused variable 'dev' [-Werror=unused-variable]

Slightly rearrange the code to pass around the of_node as a
function argument, which avoids the problem without hurting
readability.

Fixes: 71b77a7a27a3 ("enetc: Migrate to PHYLINK and PCS_LYNX")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Link: https://lore.kernel.org/r/20201204120800.17193-1-claudiu.manoil@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoptp: Add clock driver for the OpenCompute TimeCard.
Jonathan Lemon [Fri, 4 Dec 2020 03:51:28 +0000 (19:51 -0800)]
ptp: Add clock driver for the OpenCompute TimeCard.

The OpenCompute time card is an atomic clock along with
a GPS receiver that provides a Grandmaster clock source
for a PTP enabled network.

More information is available at http://www.timingcard.com/

Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Link: https://lore.kernel.org/r/20201204035128.2219252-2-jonathan.lemon@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet/nfc/nci: Support NCI 2.x initial sequence
Bongsu Jeon [Wed, 2 Dec 2020 22:31:47 +0000 (07:31 +0900)]
net/nfc/nci: Support NCI 2.x initial sequence

implement the NCI 2.x initial sequence to support NCI 2.x NFCC.
Since NCI 2.0, CORE_RESET and CORE_INIT sequence have been changed.
If NFCEE supports NCI 2.x, then NCI 2.x initial sequence will work.

In NCI 1.0, Initial sequence and payloads are as below:
(DH)                     (NFCC)
 |  -- CORE_RESET_CMD --> |
 |  <-- CORE_RESET_RSP -- |
 |  -- CORE_INIT_CMD -->  |
 |  <-- CORE_INIT_RSP --  |
 CORE_RESET_RSP payloads are Status, NCI version, Configuration Status.
 CORE_INIT_CMD payloads are empty.
 CORE_INIT_RSP payloads are Status, NFCC Features,
    Number of Supported RF Interfaces, Supported RF Interface,
    Max Logical Connections, Max Routing table Size,
    Max Control Packet Payload Size, Max Size for Large Parameters,
    Manufacturer ID, Manufacturer Specific Information.

In NCI 2.0, Initial Sequence and Parameters are as below:
(DH)                     (NFCC)
 |  -- CORE_RESET_CMD --> |
 |  <-- CORE_RESET_RSP -- |
 |  <-- CORE_RESET_NTF -- |
 |  -- CORE_INIT_CMD -->  |
 |  <-- CORE_INIT_RSP --  |
 CORE_RESET_RSP payloads are Status.
 CORE_RESET_NTF payloads are Reset Trigger,
    Configuration Status, NCI Version, Manufacturer ID,
    Manufacturer Specific Information Length,
    Manufacturer Specific Information.
 CORE_INIT_CMD payloads are Feature1, Feature2.
 CORE_INIT_RSP payloads are Status, NFCC Features,
    Max Logical Connections, Max Routing Table Size,
    Max Control Packet Payload Size,
    Max Data Packet Payload Size of the Static HCI Connection,
    Number of Credits of the Static HCI Connection,
    Max NFC-V RF Frame Size, Number of Supported RF Interfaces,
    Supported RF Interfaces.

Signed-off-by: Bongsu Jeon <bongsu.jeon@samsung.com>
Link: https://lore.kernel.org/r/20201202223147.3472-1-bongsu.jeon@samsung.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: forwarding: Add MPLS L2VPN test
Guillaume Nault [Wed, 2 Dec 2020 14:35:43 +0000 (15:35 +0100)]
selftests: forwarding: Add MPLS L2VPN test

Connect hosts H1 and H2 using two intermediate encapsulation routers
(LER1 and LER2). These routers encapsulate traffic from the hosts,
including the original Ethernet header, into MPLS.

Use ping to test reachability between H1 and H2.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Link: https://lore.kernel.org/r/625f5c1aafa3a8085f8d3e082d680a82e16ffbaa.1606918980.git.gnault@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: bna: remove trailing semicolon in macro definition
Tom Rix [Wed, 2 Dec 2020 16:36:22 +0000 (08:36 -0800)]
net: bna: remove trailing semicolon in macro definition

The macro use will already have a semicolon.
Clean up escaped newlines.

Signed-off-by: Tom Rix <trix@redhat.com>
Link: https://lore.kernel.org/r/20201202163622.3733506-1-trix@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agotipc: support 128bit node identity for peer removing
Hoang Le [Thu, 3 Dec 2020 03:50:45 +0000 (10:50 +0700)]
tipc: support 128bit node identity for peer removing

We add the support to remove a specific node down with 128bit
node identifier, as an alternative to legacy 32-bit node address.

example:
$tipc peer remove identiy <1001002|16777777>

Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Hoang Le <hoang.h.le@dektech.com.au>
Link: https://lore.kernel.org/r/20201203035045.4564-1-hoang.h.le@dektech.com.au
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonfp: Replace zero-length array with flexible-array member
Simon Horman [Fri, 4 Dec 2020 12:56:01 +0000 (13:56 +0100)]
nfp: Replace zero-length array with flexible-array member

There is a regular need in the kernel to provide a way to declare having a
dynamically sized set of trailing elements in a structure. Kernel code
should always use "flexible array members"[1] for these cases. The older
style of one-element or zero-length arrays should no longer be used[2].

[1] https://en.wikipedia.org/wiki/Flexible_array_member
[2] https://www.kernel.org/doc/html/v5.9/process/deprecated.html#zero-length-and-one-element-arrays

Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Louis Peens <louis.peens@netronome.com>
Link: https://lore.kernel.org/r/20201204125601.24876-1-simon.horman@netronome.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonfc: s3fwrn5: skip the NFC bootloader mode
Bongsu Jeon [Thu, 3 Dec 2020 22:52:57 +0000 (07:52 +0900)]
nfc: s3fwrn5: skip the NFC bootloader mode

If there isn't a proper NFC firmware image, Bootloader mode will be
skipped.

Signed-off-by: Bongsu Jeon <bongsu.jeon@samsung.com>
Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>
Link: https://lore.kernel.org/r/20201203225257.2446-1-bongsu.jeon@samsung.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'perf-optimizations-for-tcp-recv-zerocopy'
Jakub Kicinski [Fri, 4 Dec 2020 21:40:55 +0000 (13:40 -0800)]
Merge branch 'perf-optimizations-for-tcp-recv-zerocopy'

Arjun Roy says:

====================
Perf. optimizations for TCP Recv. Zerocopy

This patchset contains several optimizations for TCP Recv. Zerocopy.

Summarized:
1. It is possible that a read payload is not exactly page aligned -
that there may exist "straggler" bytes that we cannot map into the
caller's address space cleanly. For this, we allow the caller to
provide as argument a "hybrid copy buffer", turning
getsockopt(TCP_ZEROCOPY_RECEIVE) into a "hybrid" operation that allows
the caller to avoid a subsequent recvmsg() call to read the
stragglers.

2. Similarly, for "small" read payloads that are either below the size
of a page, or small enough that remapping pages is not a performance
win - we allow the user to short-circuit the remapping operations
entirely and simply copy into the buffer provided.

Some of the patches in the middle of this set are refactors to support
this "short-circuiting" optimization.

3. We allow the user to provide a hint that performing a page zap
operation (and the accompanying TLB shootdown) may not be necessary,
for the provided region that the kernel will attempt to map pages
into. This allows us to avoid this expensive operation while holding
the socket lock, which provides a significant performance advantage.

With all of these changes combined, "medium" sized receive traffic
(multiple tens to few hundreds of KB) see significant efficiency gains
when using TCP receive zerocopy instead of regular recvmsg(). For
example, with RPC-style traffic with 32KB messages, there is a roughly
15% efficiency improvement when using zerocopy. Without these changes,
there is a roughly 60-70% efficiency reduction with such messages when
employing zerocopy.
====================

Link: https://lore.kernel.org/r/20201202225349.935284-1-arjunroy.kdev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Defer vm zap unless actually needed.
Arjun Roy [Wed, 2 Dec 2020 22:53:49 +0000 (14:53 -0800)]
net-zerocopy: Defer vm zap unless actually needed.

Zapping pages is required only if we are calling vm_insert_page into a
region where pages had previously been mapped. Receive zerocopy allows
reusing such regions, and hitherto called zap_page_range() before
calling vm_insert_page() in that range.

zap_page_range() can also be triggered from userspace with
madvise(MADV_DONTNEED). If userspace is configured to call this before
reusing a segment, or if there was nothing mapped at this virtual
address to begin with, we can avoid calling zap_page_range() under the
socket lock. That said, if userspace does not do that, then we are
still responsible for calling zap_page_range().

This patch adds a flag that the user can use to hint to the kernel
that a zap is not required. If the flag is not set, or if an older
user application does not have a flags field at all, then the kernel
calls zap_page_range as before. Also, if the flag is set but a zap is
still required, the kernel performs that zap as necessary. Thus
incorrectly indicating that a zap can be avoided does not change the
correctness of operation. It also increases the batchsize for
vm_insert_pages and prefetches the page struct for the batch since
we're about to bump the refcount.

An alternative mechanism could be to not have a flag, assume by
default a zap is not needed, and fall back to zapping if needed.
However, this would harm performance for older applications for which
a zap is necessary, and thus we implement it with an explicit flag
so newer applications can opt in.

When using RPC-style traffic with medium sized (tens of KB) RPCs, this
change yields an efficency improvement of about 30% for QPS/CPU usage.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Set zerocopy hint when data is copied
Arjun Roy [Wed, 2 Dec 2020 22:53:48 +0000 (14:53 -0800)]
net-zerocopy: Set zerocopy hint when data is copied

Set zerocopy hint, event when falling back to copy, so that the
pending data can be efficiently received using zerocopy when
possible.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Introduce short-circuit small reads.
Arjun Roy [Wed, 2 Dec 2020 22:53:47 +0000 (14:53 -0800)]
net-zerocopy: Introduce short-circuit small reads.

Sometimes, we may call tcp receive zerocopy when inq is 0,
or inq < PAGE_SIZE, or inq is generally small enough that
it is cheaper to copy rather than remap pages.

In these cases, we may want to either return early (inq=0) or
attempt to use the provided copy buffer to simply copy
the received data.

This allows us to save both system call overhead and
the latency of acquiring mmap_sem in read mode for cases where
it would be useless to do so.

This patchset enables this behaviour by:
1. Returning quickly if inq is 0.
2. Attempting to perform a regular copy if a hybrid copybuffer is
   provided and it is large enough to absorb all available bytes.
3. Return quickly if no such buffer was provided and there are less
   than PAGE_SIZE bytes available.

For small RPC ping-pong workloads, normally we would have
1 getsockopt(), 1 recvmsg() and 1 sendmsg() call per RPC. With this
change, we remove the recvmsg() call entirely, reducing the syscall
overhead by about 33%. In testing with small (hundreds of bytes)
RPC traffic, this yields a syscall reduction of about 33% and
an efficiency gain of about 3-5% when defined as QPS/CPU Util.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Fast return if inq < PAGE_SIZE
Arjun Roy [Wed, 2 Dec 2020 22:53:46 +0000 (14:53 -0800)]
net-zerocopy: Fast return if inq < PAGE_SIZE

Sometimes, we may call tcp receive zerocopy when inq is 0,
or inq < PAGE_SIZE, in which case we cannot remap pages. In this case,
simply return the appropriate hint for regular copying without taking
mmap_sem.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Refactor frag-is-remappable test.
Arjun Roy [Wed, 2 Dec 2020 22:53:45 +0000 (14:53 -0800)]
net-zerocopy: Refactor frag-is-remappable test.

Refactor frag-is-remappable test for tcp receive zerocopy. This is
part of a patch set that introduces short-circuited hybrid copies
for small receive operations, which results in roughly 33% fewer
syscalls for small RPC scenarios.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Refactor skb frag fast-forward op.
Arjun Roy [Wed, 2 Dec 2020 22:53:44 +0000 (14:53 -0800)]
net-zerocopy: Refactor skb frag fast-forward op.

Refactor skb frag fast-forwarding for tcp receive zerocopy. This is
part of a patch set that introduces short-circuited hybrid copies
for small receive operations, which results in roughly 33% fewer
syscalls for small RPC scenarios.

skb_advance_to_frag(), given a skb and an offset into the skb,
iterates from the first frag for the skb until we're at the frag
specified by the offset. Assuming the offset provided refers to how
many bytes in the skb are already read, the returned frag points to
the next frag we may read from, while offset_frag is set to the number
of bytes from this frag that we have already read.

If frag is not null and offset_frag is equal to 0, then we may be able
to map this frag's page into the process address space with
vm_insert_page(). However, if offset_frag is not equal to 0, then we
cannot do so.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-tcp: Introduce tcp_recvmsg_locked().
Arjun Roy [Wed, 2 Dec 2020 22:53:43 +0000 (14:53 -0800)]
net-tcp: Introduce tcp_recvmsg_locked().

Refactor tcp_recvmsg() by splitting it into locked and unlocked
portions. Callers already holding the socket lock and not using
ERRQUEUE/cmsg/busy polling can simply call tcp_recvmsg_locked().
This is in preparation for a short-circuit copy performed by
TCP receive zerocopy for small (< PAGE_SIZE, or otherwise requested
by the user) reads.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet-zerocopy: Copy straggler unaligned data for TCP Rx. zerocopy.
Arjun Roy [Wed, 2 Dec 2020 22:53:42 +0000 (14:53 -0800)]
net-zerocopy: Copy straggler unaligned data for TCP Rx. zerocopy.

When TCP receive zerocopy does not successfully map the entire
requested space, it outputs a 'hint' that the caller should recvmsg().

Augment zerocopy to accept a user buffer that it tries to copy this
hint into - if it is possible to copy the entire hint, it will do so.
This elides a recvmsg() call for received traffic that isn't exactly
page-aligned in size.

This was tested with RPC-style traffic of arbitrary sizes. Normally,
each received message required at least one getsockopt() call, and one
recvmsg() call for the remaining unaligned data.

With this change, almost all of the recvmsg() calls are eliminated,
leading to a savings of about 25%-50% in number of system calls
for RPC-style workloads.

Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'seg6-add-support-for-srv6-end-dt4-dt6-behavior'
Jakub Kicinski [Fri, 4 Dec 2020 21:30:53 +0000 (13:30 -0800)]
Merge branch 'seg6-add-support-for-srv6-end-dt4-dt6-behavior'

Andrea Mayer says:

====================
seg6: add support for SRv6 End.DT4/DT6 behavior

This patchset provides support for the SRv6 End.DT4 and End.DT6 (VRF mode)
behaviors.

The SRv6 End.DT4 behavior is used to implement multi-tenant IPv4 L3 VPNs. It
decapsulates the received packets and performs IPv4 routing lookup in the
routing table of the tenant. The SRv6 End.DT4 Linux implementation leverages a
VRF device in order to force the routing lookup into the associated routing
table.
The SRv6 End.DT4 behavior is defined in the SRv6 Network Programming [1].

The Linux kernel already offers an implementation of the SRv6 End.DT6 behavior
which allows us to set up IPv6 L3 VPNs over SRv6 networks. This new
implementation of DT6 is based on the same VRF infrastructure already exploited
for implementing the SRv6 End.DT4 behavior. The aim of the new SRv6 End.DT6 in
VRF mode consists in simplifying the construction of IPv6 L3 VPN services in
the multi-tenant environment.
Currently, the two SRv6 End.DT6 implementations (legacy and VRF mode)
coexist seamlessly and can be chosen according to the context and the user
preferences.

- Patch 1 is needed to solve a pre-existing issue with tunneled packets
  when a sniffer is attached;

- Patch 2 improves the management of the seg6local attributes used by the
  SRv6 behaviors;

- Patch 3 adds support for optional attributes in SRv6 behaviors;

- Patch 4 introduces two callbacks used for customizing the
  creation/destruction of a SRv6 behavior;

- Patch 5 is the core patch that adds support for the SRv6 End.DT4
  behavior;

- Patch 6 introduces the VRF support for SRv6 End.DT6 behavior;

- Patch 7 adds the selftest for SRv6 End.DT4 behavior;

- Patch 8 adds the selftest for SRv6 End.DT6 (VRF mode) behavior.

Regarding iproute2, the support for the new "vrftable" attribute, required by
both SRv6 End.DT4 and End.DT6 (VRF mode) behaviors, is provided in a different
patchset that will follow shortly.

I would like to thank David Ahern for his support during the development of
this patchset.

[1] https://tools.ietf.org/html/draft-ietf-spring-srv6-network-programming
====================

Link: https://lore.kernel.org/r/20201202130517.4967-1-andrea.mayer@uniroma2.it
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: add selftest for the SRv6 End.DT6 (VRF) behavior
Andrea Mayer [Wed, 2 Dec 2020 13:05:17 +0000 (14:05 +0100)]
selftests: add selftest for the SRv6 End.DT6 (VRF) behavior

this selftest is designed for evaluating the new SRv6 End.DT6 (VRF) behavior
used, in this example, for implementing IPv6 L3 VPN use cases.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Paolo Lungaroni <paolo.lungaroni@cnit.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: add selftest for the SRv6 End.DT4 behavior
Andrea Mayer [Wed, 2 Dec 2020 13:05:16 +0000 (14:05 +0100)]
selftests: add selftest for the SRv6 End.DT4 behavior

this selftest is designed for evaluating the new SRv6 End.DT4 behavior
used, in this example, for implementing IPv4 L3 VPN use cases.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoseg6: add VRF support for SRv6 End.DT6 behavior
Andrea Mayer [Wed, 2 Dec 2020 13:05:15 +0000 (14:05 +0100)]
seg6: add VRF support for SRv6 End.DT6 behavior

SRv6 End.DT6 is defined in the SRv6 Network Programming [1].

The Linux kernel already offers an implementation of the SRv6
End.DT6 behavior which permits IPv6 L3 VPNs over SRv6 networks. This
implementation is not particularly suitable in contexts where we need to
deploy IPv6 L3 VPNs among different tenants which share the same network
address schemes. The underlying problem lies in the fact that the
current version of DT6 (called legacy DT6 from now on) needs a complex
configuration to be applied on routers which requires ad-hoc routes and
routing policy rules to ensure the correct isolation of tenants.

Consequently, a new implementation of DT6 has been introduced with the
aim of simplifying the construction of IPv6 L3 VPN services in the
multi-tenant environment using SRv6 networks. To accomplish this task,
we reused the same VRF infrastructure and SRv6 core components already
exploited for implementing the SRv6 End.DT4 behavior.

Currently the two End.DT6 implementations coexist seamlessly and can be
used depending on the context and the user preferences. So, in order to
support both versions of DT6 a new attribute (vrftable) has been
introduced which allows us to differentiate the implementation of the
behavior to be used.

A SRv6 End.DT6 legacy behavior is still instantiated using a command
like the following one:

 $ ip -6 route add 2001:db8::1 encap seg6local action End.DT6 table 100 dev eth0

While to instantiate the SRv6 End.DT6 in VRF mode, the command is still
pretty straight forward:

 $ ip -6 route add 2001:db8::1 encap seg6local action End.DT6 vrftable 100 dev eth0.

Obviously as in the case of SRv6 End.DT4, the VRF strict_mode parameter
must be set (net.vrf.strict_mode=1) and the VRF associated with table
100 must exist.

Please note that the instances of SRv6 End.DT6 legacy and End.DT6 VRF
mode can coexist in the same system/configuration without problems.

[1] https://tools.ietf.org/html/draft-ietf-spring-srv6-network-programming

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoseg6: add support for the SRv6 End.DT4 behavior
Andrea Mayer [Wed, 2 Dec 2020 13:05:14 +0000 (14:05 +0100)]
seg6: add support for the SRv6 End.DT4 behavior

SRv6 End.DT4 is defined in the SRv6 Network Programming [1].

The SRv6 End.DT4 is used to implement IPv4 L3VPN use-cases in
multi-tenants environments. It decapsulates the received packets and it
performs IPv4 routing lookup in the routing table of the tenant.

The SRv6 End.DT4 Linux implementation leverages a VRF device in order to
force the routing lookup into the associated routing table.

To make the End.DT4 work properly, it must be guaranteed that the routing
table used for routing lookup operations is bound to one and only one
VRF during the tunnel creation. Such constraint has to be enforced by
enabling the VRF strict_mode sysctl parameter, i.e:
 $ sysctl -wq net.vrf.strict_mode=1.

At JANOG44, LINE corporation presented their multi-tenant DC architecture
using SRv6 [2]. In the slides, they reported that the Linux kernel is
missing the support of SRv6 End.DT4 behavior.

The SRv6 End.DT4 behavior can be instantiated using a command similar to
the following:

 $ ip route add 2001:db8::1 encap seg6local action End.DT4 vrftable 100 dev eth0

We introduce the "vrftable" extension in iproute2 in a following patch.

[1] https://tools.ietf.org/html/draft-ietf-spring-srv6-network-programming
[2] https://speakerdeck.com/line_developers/line-data-center-networking-with-srv6

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoseg6: add callbacks for customizing the creation/destruction of a behavior
Andrea Mayer [Wed, 2 Dec 2020 13:05:13 +0000 (14:05 +0100)]
seg6: add callbacks for customizing the creation/destruction of a behavior

We introduce two callbacks used for customizing the creation/destruction of
a SRv6 behavior. Such callbacks are defined in the new struct
seg6_local_lwtunnel_ops and hereafter we provide a brief description of
them:

 - build_state(...): used for calling the custom constructor of the
   behavior during its initialization phase and after all the attributes
   have been parsed successfully;

 - destroy_state(...): used for calling the custom destructor of the
   behavior before it is completely destroyed.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoseg6: add support for optional attributes in SRv6 behaviors
Andrea Mayer [Wed, 2 Dec 2020 13:05:12 +0000 (14:05 +0100)]
seg6: add support for optional attributes in SRv6 behaviors

Before this patch, each SRv6 behavior specifies a set of required
attributes that must be provided by the userspace application when such
behavior is going to be instantiated. If at least one of the required
attributes is not provided, the creation of the behavior fails.

The SRv6 behavior framework lacks a way to manage optional attributes.
By definition, an optional attribute for a SRv6 behavior consists of an
attribute which may or may not be provided by the userspace. Therefore,
if an optional attribute is missing (and thus not supplied by the user)
the creation of the behavior goes ahead without any issue.

This patch explicitly differentiates the required attributes from the
optional attributes. In particular, each behavior can declare a set of
required attributes and a set of optional ones.

The semantic of the required attributes remains *totally* unaffected by
this patch. The introduction of the optional attributes does NOT impact
on the backward compatibility of the existing SRv6 behaviors.

It is essential to note that if an (optional or required) attribute is
supplied to a SRv6 behavior which does not expect it, the behavior
simply discards such attribute without generating any error or warning.
This operating mode remained unchanged both before and after the
introduction of the optional attributes extension.

The optional attributes are one of the key components used to implement
the SRv6 End.DT6 behavior based on the Virtual Routing and Forwarding
(VRF) framework. The optional attributes make possible the coexistence
of the already existing SRv6 End.DT6 implementation with the new SRv6
End.DT6 VRF-based implementation without breaking any backward
compatibility. Further details on the SRv6 End.DT6 behavior (VRF mode)
are reported in subsequent patches.

From the userspace point of view, the support for optional attributes DO
NOT require any changes to the userspace applications, i.e: iproute2
unless new attributes (required or optional) are needed.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoseg6: improve management of behavior attributes
Andrea Mayer [Wed, 2 Dec 2020 13:05:11 +0000 (14:05 +0100)]
seg6: improve management of behavior attributes

Depending on the attribute (i.e.: SEG6_LOCAL_SRH, SEG6_LOCAL_TABLE, etc),
the parse() callback performs some validity checks on the provided input
and updates the tunnel state (slwt) with the result of the parsing
operation. However, an attribute may also need to reserve some additional
resources (i.e.: memory or setting up an eBPF program) in the parse()
callback to complete the parsing operation.

The parse() callbacks are invoked by the parse_nla_action() for each
attribute belonging to a specific behavior. Given a behavior with N
attributes, if the parsing of the i-th attribute fails, the
parse_nla_action() returns immediately with an error. Nonetheless, the
resources acquired during the parsing of the i-1 attributes are not freed
by the parse_nla_action().

Attributes which acquire resources must release them *in an explicit way*
in both the seg6_local_{build/destroy}_state(). However, adding a new
attribute of this type requires changes to
seg6_local_{build/destroy}_state() to release the resources correctly.

The seg6local infrastructure still lacks a simple and structured way to
release the resources acquired in the parse() operations.

We introduced a new callback in the struct seg6_action_param named
destroy(). This callback releases any resource which may have been acquired
in the parse() counterpart. Each attribute may or may not implement the
destroy() callback depending on whether it needs to free some acquired
resources.

The destroy() callback comes with several of advantages:

 1) we can have many attributes as we want for a given behavior with no
    need to explicitly free the taken resources;

 2) As in case of the seg6_local_build_state(), the
    seg6_local_destroy_state() does not need to handle the release of
    resources directly. Indeed, it calls the destroy_attrs() function which
    is in charge of calling the destroy() callback for every set attribute.
    We do not need to patch seg6_local_{build/destroy}_state() anymore as
    we add new attributes;

 3) the code is more readable and better structured. Indeed, all the
    information needed to handle a given attribute are contained in only
    one place;

 4) it facilitates the integration with new features introduced in further
    patches.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agovrf: add mac header for tunneled packets when sniffer is attached
Andrea Mayer [Wed, 2 Dec 2020 13:05:10 +0000 (14:05 +0100)]
vrf: add mac header for tunneled packets when sniffer is attached

Before this patch, a sniffer attached to a VRF used as the receiving
interface of L3 tunneled packets detects them as malformed packets and
it complains about that (i.e.: tcpdump shows bogus packets).

The reason is that a tunneled L3 packet does not carry any L2
information and when the VRF is set as the receiving interface of a
decapsulated L3 packet, no mac header is currently set or valid.
Therefore, the purpose of this patch consists of adding a MAC header to
any packet which is directly received on the VRF interface ONLY IF:

 i) a sniffer is attached on the VRF and ii) the mac header is not set.

In this case, the mac address of the VRF is copied in both the
destination and the source address of the ethernet header. The protocol
type is set either to IPv4 or IPv6, depending on which L3 packet is
received.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge tag 'wireless-drivers-next-2020-12-03' of git://git.kernel.org/pub/scm/linux...
Jakub Kicinski [Fri, 4 Dec 2020 18:56:37 +0000 (10:56 -0800)]
Merge tag 'wireless-drivers-next-2020-12-03' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for v5.11

First set of patches for v5.11. rtw88 getting improvements to work
better with Bluetooth and other driver also getting some new features.
mhi-ath11k-immutable branch was pulled from mhi tree to avoid
conflicts with mhi tree.

Major changes:

rtw88
 * major bluetooth co-existance improvements
wilc1000
 * Wi-Fi Multimedia (WMM) support
ath11k
 * Fast Initial Link Setup (FILS) discovery and unsolicited broadcast
   probe response support
 * qcom,ath11k-calibration-variant Device Tree setting
 * cold boot calibration support
 * new DFS region: JP
wnc36xx
 * enable connection monitoring and keepalive in firmware
ath10k
 * firmware IRAM recovery feature
mhi
 * merge mhi-ath11k-immutable branch to make MHI API change go smoothly

* tag 'wireless-drivers-next-2020-12-03' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next: (180 commits)
  wl1251: remove trailing semicolon in macro definition
  airo: remove trailing semicolon in macro definition
  wilc1000: added queue support for WMM
  wilc1000: call complete() for failure in wilc_wlan_txq_add_cfg_pkt()
  wilc1000: free resource in wilc_wlan_txq_add_mgmt_pkt() for failure path
  wilc1000: free resource in wilc_wlan_txq_add_net_pkt() for failure path
  wilc1000: added 'ndo_set_mac_address' callback support
  brcmfmac: expose firmware config files through modinfo
  wlcore: Switch to using the new API kobj_to_dev()
  rtw88: coex: add feature to enhance HID coexistence performance
  rtw88: coex: upgrade coexistence A2DP mechanism
  rtw88: coex: add action for coexistence in hardware initial
  rtw88: coex: add function to avoid cck lock
  rtw88: coex: change the coexistence mechanism for WLAN connected
  rtw88: coex: change the coexistence mechanism for HID
  rtw88: coex: update AFH information while in free-run mode
  rtw88: coex: update the mechanism for A2DP + PAN
  rtw88: coex: add debug message
  rtw88: coex: run coexistence when WLAN entering/leaving LPS
  Revert "rtl8xxxu: Add Buffalo WI-U3-866D to list of supported devices"
  ...
====================

Link: https://lore.kernel.org/r/20201203185732.9CFA5C433ED@smtp.codeaurora.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agodpaa_eth: fix build errorr in dpaa_fq_init
Anders Roxell [Thu, 3 Dec 2020 14:43:43 +0000 (15:43 +0100)]
dpaa_eth: fix build errorr in dpaa_fq_init

When building FSL_DPAA_ETH the following build error shows up:

/tmp/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c: In function â€˜dpaa_fq_init’:
/tmp/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c:1135:9: error: too few arguments to function â€˜xdp_rxq_info_reg’
 1135 |   err = xdp_rxq_info_reg(&dpaa_fq->xdp_rxq, dpaa_fq->net_dev,
      |         ^~~~~~~~~~~~~~~~

Commit b02e5a0ebb17 ("xsk: Propagate napi_id to XDP socket Rx path")
added an extra argument to function xdp_rxq_info_reg and commit
d57e57d0cd04 ("dpaa_eth: add XDP_TX support") didn't know about that
extra argument.

Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Link: https://lore.kernel.org/r/20201203144343.790719-1-anders.roxell@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Jakub Kicinski [Fri, 4 Dec 2020 15:48:11 +0000 (07:48 -0800)]
Merge https://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2020-12-03

The main changes are:

1) Support BTF in kernel modules, from Andrii.

2) Introduce preferred busy-polling, from Björn.

3) bpf_ima_inode_hash() and bpf_bprm_opts_set() helpers, from KP Singh.

4) Memcg-based memory accounting for bpf objects, from Roman.

5) Allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks, from Stanislav.

* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (118 commits)
  selftests/bpf: Fix invalid use of strncat in test_sockmap
  libbpf: Use memcpy instead of strncpy to please GCC
  selftests/bpf: Add fentry/fexit/fmod_ret selftest for kernel module
  selftests/bpf: Add tp_btf CO-RE reloc test for modules
  libbpf: Support attachment of BPF tracing programs to kernel modules
  libbpf: Factor out low-level BPF program loading helper
  bpf: Allow to specify kernel module BTFs when attaching BPF programs
  bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
  selftests/bpf: Add CO-RE relocs selftest relying on kernel module BTF
  selftests/bpf: Add support for marking sub-tests as skipped
  selftests/bpf: Add bpf_testmod kernel module for testing
  libbpf: Add kernel module BTF support for CO-RE relocations
  libbpf: Refactor CO-RE relocs to not assume a single BTF object
  libbpf: Add internal helper to load BTF data by FD
  bpf: Keep module's btf_data_size intact after load
  bpf: Fix bpf_put_raw_tracepoint()'s use of __module_address()
  selftests/bpf: Add Userspace tests for TCP_WINDOW_CLAMP
  bpf: Adds support for setting window clamp
  samples/bpf: Fix spelling mistake "recieving" -> "receiving"
  bpf: Fix cold build of test_progs-no_alu32
  ...
====================

Link: https://lore.kernel.org/r/20201204021936.85653-1-alexei.starovoitov@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet/mlx5: Register mlx5 devices to auxiliary virtual bus
Leon Romanovsky [Thu, 8 Oct 2020 13:06:37 +0000 (16:06 +0300)]
net/mlx5: Register mlx5 devices to auxiliary virtual bus

Create auxiliary devices under new virtual bus. This will replace
the custom-made mlx5 ->add()/->remove() interfaces and next patches
will fill the missing callback and remove the old interface logic.

The attachment of auxiliary drivers to the devices is possible in
1-to-1 manner only and it requires us to create device for every protocol,
so that device (module) will be able to connect to it.

System with 2 IB and 1 RoCE cards:
[leonro@vm ~]$ lspci |grep nox
00:09.0 Ethernet controller: Mellanox Technologies MT27800 Family [ConnectX-5]
00:0a.0 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6]
00:0b.0 Ethernet controller: Mellanox Technologies MT2910 Family [ConnectX-7]
[leonro@vm ~]$ ls -l /sys/bus/auxiliary/devices/
 mlx5_core.eth.2 -> ../../../devices/pci0000:00/0000:00:0b.0/mlx5_core.eth.2
 mlx5_core.rdma.0 -> ../../../devices/pci0000:00/0000:00:09.0/mlx5_core.rdma.0
 mlx5_core.rdma.1 -> ../../../devices/pci0000:00/0000:00:0a.0/mlx5_core.rdma.1
 mlx5_core.rdma.2 -> ../../../devices/pci0000:00/0000:00:0b.0/mlx5_core.rdma.2
 mlx5_core.vdpa.1 -> ../../../devices/pci0000:00/0000:00:0a.0/mlx5_core.vdpa.1
 mlx5_core.vdpa.2 -> ../../../devices/pci0000:00/0000:00:0b.0/mlx5_core.vdpa.2
[leonro@vm ~]$ rdma dev
0: ibp0s9: node_type ca fw 4.6.9999 node_guid 5254:00c0:fe12:3455 sys_image_guid 5254:00c0:fe12:3455
1: ibp0s10: node_type ca fw 4.6.9999 node_guid 5254:00c0:fe12:3456 sys_image_guid 5254:00c0:fe12:3456
2: rdmap0s11: node_type ca fw 4.6.9999 node_guid 5254:00c0:fe12:3457 sys_image_guid 5254:00c0:fe12:3457

System with RoCE SR-IOV card with 4 VFs:
[leonro@vm ~]$ lspci |grep nox
01:00.0 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6]
01:00.1 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6 Virtual Function]
01:00.2 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6 Virtual Function]
01:00.3 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6 Virtual Function]
01:00.4 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6 Virtual Function]
[leonro@vm ~]$ ls -l /sys/bus/auxiliary/devices/
 mlx5_core.eth.0 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.0/mlx5_core.eth.0
 mlx5_core.eth.1 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.1/mlx5_core.eth.1
 mlx5_core.eth.2 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.2/mlx5_core.eth.2
 mlx5_core.eth.3 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.3/mlx5_core.eth.3
 mlx5_core.eth.4 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.4/mlx5_core.eth.4
 mlx5_core.rdma.0 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.0/mlx5_core.rdma.0
 mlx5_core.rdma.1 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.1/mlx5_core.rdma.1
 mlx5_core.rdma.2 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.2/mlx5_core.rdma.2
 mlx5_core.rdma.3 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.3/mlx5_core.rdma.3
 mlx5_core.rdma.4 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.4/mlx5_core.rdma.4
 mlx5_core.vdpa.1 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.1/mlx5_core.vdpa.1
 mlx5_core.vdpa.2 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.2/mlx5_core.vdpa.2
 mlx5_core.vdpa.3 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.3/mlx5_core.vdpa.3
 mlx5_core.vdpa.4 -> ../../../devices/pci0000:00/0000:00:09.0/0000:01:00.4/mlx5_core.vdpa.4
[leonro@vm ~]$ rdma dev
0: rocep1s0f0: node_type ca fw 4.6.9999 node_guid 5254:00c0:fe12:3455 sys_image_guid 5254:00c0:fe12:3455
1: rocep1s0f0v0: node_type ca fw 4.6.9999 node_guid 0000:0000:0000:0000 sys_image_guid 5254:00c0:fe12:3456
2: rocep1s0f0v1: node_type ca fw 4.6.9999 node_guid 0000:0000:0000:0000 sys_image_guid 5254:00c0:fe12:3457
3: rocep1s0f0v2: node_type ca fw 4.6.9999 node_guid 0000:0000:0000:0000 sys_image_guid 5254:00c0:fe12:3458
4: rocep1s0f0v3: node_type ca fw 4.6.9999 node_guid 0000:0000:0000:0000 sys_image_guid 5254:00c0:fe12:3459

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agovdpa/mlx5: Make hardware definitions visible to all mlx5 devices
Leon Romanovsky [Tue, 6 Oct 2020 09:34:25 +0000 (12:34 +0300)]
vdpa/mlx5: Make hardware definitions visible to all mlx5 devices

Move mlx5_vdpa IFC header file to the general include folder, so
mlx5_core will be able to reuse it to check if VDPA is supported
prior to creating an auxiliary device.

As part of this move, update the header file name to mlx5 general
naming scheme.

Reviewed-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agonet/mlx5_core: Clean driver version and name
Leon Romanovsky [Sun, 4 Oct 2020 11:30:58 +0000 (14:30 +0300)]
net/mlx5_core: Clean driver version and name

Remove exposed driver version as it was done in other drivers,
so module version will work correctly by displaying the kernel
version for which it is compiled.

And move mlx5_core module name to general include, so auxiliary drivers
will be able to use it as a basis for a name in their device ID tables.

Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agonet/mlx5: Properly convey driver version to firmware
Leon Romanovsky [Thu, 15 Oct 2020 11:52:00 +0000 (14:52 +0300)]
net/mlx5: Properly convey driver version to firmware

mlx5 firmware expects driver version in specific format X.X.X, so
make it always correct and based on real kernel version aligned with
the driver.

Fixes: 012e50e109fd ("net/mlx5: Set driver version into firmware")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agoMerge tag 'auxbus-5.11-rc1' of https://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Leon Romanovsky [Fri, 4 Dec 2020 12:46:02 +0000 (14:46 +0200)]
Merge tag 'auxbus-5.11-rc1' of https://git./linux/kernel/git/gregkh/driver-core into mlx5-next

Auxiliary Bus support tag for 5.11-rc1

This is a signed tag for other subsystems to be able to pull in the
auxiliary bus support into their trees for the 5.11-rc1 merge.

Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
* tag 'auxbus-5.11-rc1' of https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  driver core: auxiliary bus: minor coding style tweaks
  driver core: auxiliary bus: make remove function return void
  driver core: auxiliary bus: move slab.h from include file
  Add auxiliary bus support

3 years agodriver core: auxiliary bus: minor coding style tweaks
Greg Kroah-Hartman [Fri, 4 Dec 2020 11:49:28 +0000 (12:49 +0100)]
driver core: auxiliary bus: minor coding style tweaks

For some reason, the original aux bus patch had some really long lines
in a few places, probably due to it being a very long-lived patch in
development by many different people.  Fix that up so that the two files
all have the same length lines and function formatting styles.

Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Ertman <david.m.ertman@intel.com>
Cc: Fred Oh <fred.oh@linux.intel.com>
Cc: Kiran Patil <kiran.patil@intel.com>
Cc: Leon Romanovsky <leonro@nvidia.com>
Cc: Martin Habets <mhabets@solarflare.com>
Cc: Parav Pandit <parav@mellanox.com>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Cc: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Cc: Shiraz Saleem <shiraz.saleem@intel.com>
Link: https://lore.kernel.org/r/X8oiSFTpYHw1xE/o@kroah.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agodriver core: auxiliary bus: make remove function return void
Greg Kroah-Hartman [Fri, 4 Dec 2020 11:44:07 +0000 (12:44 +0100)]
driver core: auxiliary bus: make remove function return void

There's an effort to move the remove() callback in the driver core to
not return an int, as nothing can be done if this function fails.  To
make that effort easier, make the aux bus remove function void to start
with so that no users have to be changed sometime in the future.

Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Ertman <david.m.ertman@intel.com>
Cc: Fred Oh <fred.oh@linux.intel.com>
Cc: Kiran Patil <kiran.patil@intel.com>
Cc: Leon Romanovsky <leonro@nvidia.com>
Cc: Martin Habets <mhabets@solarflare.com>
Cc: Parav Pandit <parav@mellanox.com>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Cc: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Cc: Shiraz Saleem <shiraz.saleem@intel.com>
Link: https://lore.kernel.org/r/X8ohB1ks1NK7kPop@kroah.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agodriver core: auxiliary bus: move slab.h from include file
Greg Kroah-Hartman [Fri, 4 Dec 2020 11:43:47 +0000 (12:43 +0100)]
driver core: auxiliary bus: move slab.h from include file

No need to include slab.h in include/linux/auxiliary_bus.h, as it is not
needed there.  Move it to drivers/base/auxiliary.c instead.

Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Ertman <david.m.ertman@intel.com>
Cc: Fred Oh <fred.oh@linux.intel.com>
Cc: Kiran Patil <kiran.patil@intel.com>
Cc: Leon Romanovsky <leonro@nvidia.com>
Cc: Martin Habets <mhabets@solarflare.com>
Cc: Parav Pandit <parav@mellanox.com>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Cc: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Cc: Shiraz Saleem <shiraz.saleem@intel.com>
Link: https://lore.kernel.org/r/X8og8xi3WkoYXet9@kroah.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agoAdd auxiliary bus support
Dave Ertman [Thu, 3 Dec 2020 00:54:24 +0000 (16:54 -0800)]
Add auxiliary bus support

Add support for the Auxiliary Bus, auxiliary_device and auxiliary_driver.
It enables drivers to create an auxiliary_device and bind an
auxiliary_driver to it.

The bus supports probe/remove shutdown and suspend/resume callbacks.
Each auxiliary_device has a unique string based id; driver binds to
an auxiliary_device based on this id through the bus.

Co-developed-by: Kiran Patil <kiran.patil@intel.com>
Co-developed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Co-developed-by: Fred Oh <fred.oh@linux.intel.com>
Co-developed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Fred Oh <fred.oh@linux.intel.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Martin Habets <mhabets@solarflare.com>
Link: https://lore.kernel.org/r/20201113161859.1775473-2-david.m.ertman@intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Link: https://lore.kernel.org/r/160695681289.505290.8978295443574440604.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agobatman-adv: Drop unused soft-interface.h include in fragmentation.c
Simon Wunderlich [Fri, 4 Dec 2020 07:41:16 +0000 (08:41 +0100)]
batman-adv: Drop unused soft-interface.h include in fragmentation.c

The commit 992b03b88e36 ("batman-adv: Don't always reallocate the
fragmentation skb head") removed the last user of functions from
soft-interface.h.

Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Drop legacy code for auto deleting mesh interfaces
Sven Eckelmann [Mon, 17 Aug 2020 12:37:13 +0000 (14:37 +0200)]
batman-adv: Drop legacy code for auto deleting mesh interfaces

The only way to automatically drop batadv mesh interfaces when all soft
interfaces were removed was dropped with the sysfs support. It is no longer
needed to have them handled by kernel anymore.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Drop deprecated debugfs support
Sven Eckelmann [Mon, 17 Aug 2020 11:42:29 +0000 (13:42 +0200)]
batman-adv: Drop deprecated debugfs support

The debugfs support in batman-adv was marked as deprecated by the commit
00caf6a2b318 ("batman-adv: Mark debugfs functionality as deprecated") and
scheduled for removal in 2021.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Drop deprecated sysfs support
Sven Eckelmann [Mon, 17 Aug 2020 11:42:29 +0000 (13:42 +0200)]
batman-adv: Drop deprecated sysfs support

The sysfs in batman-adv support was marked as deprecated by the commit
42cdd521487f ("batman-adv: ABI: Mark sysfs files as deprecated") and
scheduled for removal in 2021.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Allow selection of routing algorithm over rtnetlink
Sven Eckelmann [Sun, 11 Oct 2020 10:25:24 +0000 (12:25 +0200)]
batman-adv: Allow selection of routing algorithm over rtnetlink

A batadv net_device is associated to a B.A.T.M.A.N. routing algorithm. This
algorithm has to be selected before the interface is initialized and cannot
be changed after that. The only way to select this algorithm was a module
parameter which specifies the default algorithm used during the creation of
the net_device.

This module parameter is writeable over
/sys/module/batman_adv/parameters/routing_algo and thus allows switching of
the routing algorithm:

1. change routing_algo parameter
2. create new batadv net_device

But this is not race free because another process can be scheduled between
1 + 2 and in that time frame change the routing_algo parameter again.

It is much cleaner to directly provide this information inside the
rtnetlink's RTM_NEWLINK message. The two processes would be (in regards of
the creation parameter of their batadv interfaces) be isolated. This also
eases the integration of batadv devices inside tools like network-manager
or systemd-networkd which are not expecting to operate on /sys before a new
net_device is created.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Prepare infrastructure for newlink settings
Sven Eckelmann [Sun, 11 Oct 2020 10:25:23 +0000 (12:25 +0200)]
batman-adv: Prepare infrastructure for newlink settings

The batadv generic netlink family can be used to retrieve the current state
and set various configuration settings. But there are also settings which
must be set before the actual interface is created.

The rtnetlink already uses IFLA_INFO_DATA to allow net_device families to
transfer such configurations. The minimal required functionality for this
is now available for the batadv rtnl_link_ops. Also a new IFLA class of
attributes will be attached to it because rtnetlink only allows 51
different attributes but batadv_nl_attrs already contains 62 attributes.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Add new include for min/max helpers
Sven Eckelmann [Mon, 26 Oct 2020 20:01:59 +0000 (21:01 +0100)]
batman-adv: Add new include for min/max helpers

The commit b296a6d53339 ("kernel.h: split out min()/max() et al. helpers")
moved the min/max helper functionality from kernel.h to minmax.h. Adjust
the kernel code accordingly to avoid fragile indirect includes.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agobatman-adv: Start new development cycle
Simon Wunderlich [Tue, 27 Oct 2020 13:58:32 +0000 (14:58 +0100)]
batman-adv: Start new development cycle

Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
3 years agoselftests/bpf: Fix invalid use of strncat in test_sockmap
Andrii Nakryiko [Thu, 3 Dec 2020 23:54:40 +0000 (15:54 -0800)]
selftests/bpf: Fix invalid use of strncat in test_sockmap

strncat()'s third argument is how many bytes will be added *in addition* to
already existing bytes in destination. Plus extra zero byte will be added
after that. So existing use in test_sockmap has many opportunities to overflow
the string and cause memory corruptions. And in this case, GCC complains for
a good reason.

Fixes: 16962b2404ac ("bpf: sockmap, add selftests")
Fixes: 73563aa3d977 ("selftests/bpf: test_sockmap, print additional test options")
Fixes: 1ade9abadfca ("bpf: test_sockmap, add options for msg_pop_data() helper")
Fixes: 463bac5f1ca7 ("bpf, selftests: Add test for ktls with skb bpf ingress policy")
Fixes: e9dd904708c4 ("bpf: add tls support for testing in test_sockmap")
Fixes: 753fb2ee0934 ("bpf: sockmap, add msg_peek tests to test_sockmap")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203235440.2302137-2-andrii@kernel.org
3 years agolibbpf: Use memcpy instead of strncpy to please GCC
Andrii Nakryiko [Thu, 3 Dec 2020 23:54:39 +0000 (15:54 -0800)]
libbpf: Use memcpy instead of strncpy to please GCC

Some versions of GCC are really nit-picky about strncpy() use. Use memcpy(),
as they are pretty much equivalent for the case of fixed length strings.

Fixes: e459f49b4394 ("libbpf: Separate XDP program load with xsk socket creation")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203235440.2302137-1-andrii@kernel.org
3 years agoMerge branch 'Support BTF-powered BPF tracing programs for kernel modules'
Alexei Starovoitov [Fri, 4 Dec 2020 01:38:22 +0000 (17:38 -0800)]
Merge branch 'Support BTF-powered BPF tracing programs for kernel modules'

Andrii Nakryiko says:

====================

This patch sets extends kernel and libbpf with support for attaching
BTF-powered raw tracepoint (tp_btf) and tracing (fentry/fexit/fmod_ret/lsm)
BPF programs to BPF hooks defined in kernel modules. As part of that, libbpf
now supports performing CO-RE relocations against types in kernel module BTFs,
in addition to existing vmlinux BTF support.

Kernel UAPI for BPF_PROG_LOAD now allows to specify kernel module (or vmlinux)
BTF object FD in attach_btf_obj_fd field, aliased to attach_prog_fd. This is
used to identify which BTF object needs to be used for finding BTF type by
provided attach_btf_id.

This patch set also sets up a convenient and fully-controlled custom kernel
module (called "bpf_testmod"), that is a predictable playground for all the
BPF selftests, that rely on module BTFs. Currently pahole doesn't generate
BTF_KIND_FUNC info for ftrace-able static functions in kernel modules, so
expose traced function in bpf_sidecar.ko. Once pahole is enhanced, we can go
back to static function.

From end user perspective there are no extra actions that need to happen.
Libbpf will continue searching across all kernel module BTFs, if desired
attach BTF type is not found in vmlinux. That way it doesn't matter if BPF
hook that user is trying to attach to is built into vmlinux image or is
loaded in kernel module.

v5->v6:
  - move btf_put() back to syscall.c (kernel test robot);
  - added close(fd) in patch #5 (John);
v4->v5:
  - use FD to specify BTF object (Alexei);
  - move prog->aux->attach_btf putting into bpf_prog_free() for consistency
    with putting prog->aux->dst_prog;
  - fix BTF FD leak(s) in libbpf;
v3->v4:
  - merge together patch sets [0] and [1];
  - avoid increasing bpf_reg_state by reordering fields (Alexei);
  - preserve btf_data_size in struct module;
v2->v3:
  - fix subtle uninitialized variable use in BTF ID iteration code;
v1->v2:
  - module_put() inside preempt_disable() region (Alexei);
  - bpf_sidecar -> bpf_testmod rename (Alexei);
  - test_progs more relaxed handling of bpf_testmod;
  - test_progs marks skipped sub-tests properly as SKIP now.

  [0] https://patchwork.kernel.org/project/netdevbpf/list/?series=393677&state=*
  [1] https://patchwork.kernel.org/project/netdevbpf/list/?series=393679&state=*
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agoselftests/bpf: Add fentry/fexit/fmod_ret selftest for kernel module
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:34 +0000 (12:46 -0800)]
selftests/bpf: Add fentry/fexit/fmod_ret selftest for kernel module

Add new selftest checking attachment of fentry/fexit/fmod_ret (and raw
tracepoint ones for completeness) BPF programs to kernel module function.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-15-andrii@kernel.org
3 years agoselftests/bpf: Add tp_btf CO-RE reloc test for modules
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:33 +0000 (12:46 -0800)]
selftests/bpf: Add tp_btf CO-RE reloc test for modules

Add another CO-RE relocation test for kernel module relocations. This time for
tp_btf with direct memory reads.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-14-andrii@kernel.org
3 years agolibbpf: Support attachment of BPF tracing programs to kernel modules
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:32 +0000 (12:46 -0800)]
libbpf: Support attachment of BPF tracing programs to kernel modules

Teach libbpf to search for BTF types in kernel modules for tracing BPF
programs. This allows attachment of raw_tp/fentry/fexit/fmod_ret/etc BPF
program types to tracepoints and functions in kernel modules.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-13-andrii@kernel.org
3 years agolibbpf: Factor out low-level BPF program loading helper
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:31 +0000 (12:46 -0800)]
libbpf: Factor out low-level BPF program loading helper

Refactor low-level API for BPF program loading to not rely on public API
types. This allows painless extension without constant efforts to cleverly not
break backwards compatibility.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-12-andrii@kernel.org
3 years agobpf: Allow to specify kernel module BTFs when attaching BPF programs
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:30 +0000 (12:46 -0800)]
bpf: Allow to specify kernel module BTFs when attaching BPF programs

Add ability for user-space programs to specify non-vmlinux BTF when attaching
BTF-powered BPF programs: raw_tp, fentry/fexit/fmod_ret, LSM, etc. For this,
attach_prog_fd (now with the alias name attach_btf_obj_fd) should specify FD
of a module or vmlinux BTF object. For backwards compatibility reasons,
0 denotes vmlinux BTF. Only kernel BTF (vmlinux or module) can be specified.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-11-andrii@kernel.org
3 years agobpf: Remove hard-coded btf_vmlinux assumption from BPF verifier
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:29 +0000 (12:46 -0800)]
bpf: Remove hard-coded btf_vmlinux assumption from BPF verifier

Remove a permeating assumption thoughout BPF verifier of vmlinux BTF. Instead,
wherever BTF type IDs are involved, also track the instance of struct btf that
goes along with the type ID. This allows to gradually add support for kernel
module BTFs and using/tracking module types across BPF helper calls and
registers.

This patch also renames btf_id() function to btf_obj_id() to minimize naming
clash with using btf_id to denote BTF *type* ID, rather than BTF *object*'s ID.

Also, altough btf_vmlinux can't get destructed and thus doesn't need
refcounting, module BTFs need that, so apply BTF refcounting universally when
BPF program is using BTF-powered attachment (tp_btf, fentry/fexit, etc). This
makes for simpler clean up code.

Now that BTF type ID is not enough to uniquely identify a BTF type, extend BPF
trampoline key to include BTF object ID. To differentiate that from target
program BPF ID, set 31st bit of type ID. BTF type IDs (at least currently) are
not allowed to take full 32 bits, so there is no danger of confusing that bit
with a valid BTF type ID.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-10-andrii@kernel.org
3 years agoselftests/bpf: Add CO-RE relocs selftest relying on kernel module BTF
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:28 +0000 (12:46 -0800)]
selftests/bpf: Add CO-RE relocs selftest relying on kernel module BTF

Add a self-tests validating libbpf is able to perform CO-RE relocations
against the type defined in kernel module BTF. if bpf_testmod.o is not
supported by the kernel (e.g., due to version mismatch), skip tests, instead
of failing.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-9-andrii@kernel.org
3 years agoselftests/bpf: Add support for marking sub-tests as skipped
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:27 +0000 (12:46 -0800)]
selftests/bpf: Add support for marking sub-tests as skipped

Previously skipped sub-tests would be counted as passing with ":OK" appened
in the log. Change that to be accounted as ":SKIP".

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-8-andrii@kernel.org
3 years agoselftests/bpf: Add bpf_testmod kernel module for testing
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:26 +0000 (12:46 -0800)]
selftests/bpf: Add bpf_testmod kernel module for testing

Add bpf_testmod module, which is conceptually out-of-tree module and provides
ways for selftests/bpf to test various kernel module-related functionality:
raw tracepoint, fentry/fexit/fmod_ret, etc. This module will be auto-loaded by
test_progs test runner and expected by some of selftests to be present and
loaded.

Pahole currently isn't able to generate BTF for static functions in kernel
modules, so make sure traced function is global.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-7-andrii@kernel.org
3 years agolibbpf: Add kernel module BTF support for CO-RE relocations
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:25 +0000 (12:46 -0800)]
libbpf: Add kernel module BTF support for CO-RE relocations

Teach libbpf to search for candidate types for CO-RE relocations across kernel
modules BTFs, in addition to vmlinux BTF. If at least one candidate type is
found in vmlinux BTF, kernel module BTFs are not iterated. If vmlinux BTF has
no matching candidates, then find all kernel module BTFs and search for all
matching candidates across all of them.

Kernel's support for module BTFs are inferred from the support for BTF name
pointer in BPF UAPI.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-6-andrii@kernel.org
3 years agolibbpf: Refactor CO-RE relocs to not assume a single BTF object
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:24 +0000 (12:46 -0800)]
libbpf: Refactor CO-RE relocs to not assume a single BTF object

Refactor CO-RE relocation candidate search to not expect a single BTF, rather
return all candidate types with their corresponding BTF objects. This will
allow to extend CO-RE relocations to accommodate kernel module BTFs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-5-andrii@kernel.org
3 years agolibbpf: Add internal helper to load BTF data by FD
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:23 +0000 (12:46 -0800)]
libbpf: Add internal helper to load BTF data by FD

Add a btf_get_from_fd() helper, which constructs struct btf from in-kernel BTF
data by FD. This is used for loading module BTFs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-4-andrii@kernel.org
3 years agobpf: Keep module's btf_data_size intact after load
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:22 +0000 (12:46 -0800)]
bpf: Keep module's btf_data_size intact after load

Having real btf_data_size stored in struct module is benefitial to quickly
determine which kernel modules have associated BTF object and which don't.
There is no harm in keeping this info, as opposed to keeping invalid pointer.

Fixes: 607c543f939d ("bpf: Sanitize BTF data pointer after module is loaded")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-3-andrii@kernel.org
3 years agobpf: Fix bpf_put_raw_tracepoint()'s use of __module_address()
Andrii Nakryiko [Thu, 3 Dec 2020 20:46:21 +0000 (12:46 -0800)]
bpf: Fix bpf_put_raw_tracepoint()'s use of __module_address()

__module_address() needs to be called with preemption disabled or with
module_mutex taken. preempt_disable() is enough for read-only uses, which is
what this fix does. Also, module_put() does internal check for NULL, so drop
it as well.

Fixes: a38d1107f937 ("bpf: support raw tracepoints in modules")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20201203204634.1325171-2-andrii@kernel.org
3 years agoMerge branch 'Add support to set window_clamp from bpf setsockops'
Alexei Starovoitov [Fri, 4 Dec 2020 01:23:24 +0000 (17:23 -0800)]
Merge branch 'Add support to set window_clamp from bpf setsockops'

Prankur gupta says:

====================

This patch contains support to set tcp window_field field from bpf setsockops.

v2: Used TCP_WINDOW_CLAMP setsockopt logic for bpf_setsockopt (review comment addressed)

v3: Created a common function for duplicated code (review comment addressed)

v4: Removing logic to pass struct sock and struct tcp_sock together (review comment addressed)
====================

Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agoselftests/bpf: Add Userspace tests for TCP_WINDOW_CLAMP
Prankur gupta [Wed, 2 Dec 2020 21:31:52 +0000 (13:31 -0800)]
selftests/bpf: Add Userspace tests for TCP_WINDOW_CLAMP

Adding selftests for new added functionality to set TCP_WINDOW_CLAMP
from bpf setsockopt.

Signed-off-by: Prankur gupta <prankgup@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201202213152.435886-3-prankgup@fb.com
3 years agobpf: Adds support for setting window clamp
Prankur gupta [Wed, 2 Dec 2020 21:31:51 +0000 (13:31 -0800)]
bpf: Adds support for setting window clamp

Adds a new bpf_setsockopt for TCP sockets, TCP_BPF_WINDOW_CLAMP,
which sets the maximum receiver window size. It will be useful for
limiting receiver window based on RTT.

Signed-off-by: Prankur gupta <prankgup@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201202213152.435886-2-prankgup@fb.com
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 3 Dec 2020 23:42:13 +0000 (15:42 -0800)]
Merge git://git./linux/kernel/git/netdev/net

Conflicts:
drivers/net/ethernet/ibm/ibmvnic.c

Signed-off-by: Jakub Kicinski <kuba@kernel.org>