linux-2.6-microblaze.git
4 years agor8169: improve rtl8169_runtime_resume
Heiner Kallweit [Sat, 20 Jun 2020 20:39:35 +0000 (22:39 +0200)]
r8169: improve rtl8169_runtime_resume

Simplify rtl8169_runtime_resume() by calling rtl8169_resume().

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: remove driver-specific mutex
Heiner Kallweit [Sat, 20 Jun 2020 20:38:55 +0000 (22:38 +0200)]
r8169: remove driver-specific mutex

Now that the critical sections are protected with RTNL lock, we don't
need a separate mutex any longer.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: use RTNL to protect critical sections
Heiner Kallweit [Sat, 20 Jun 2020 20:38:24 +0000 (22:38 +0200)]
r8169: use RTNL to protect critical sections

Most relevant ops (open, close, ethtool ops) are protected with RTNL
lock by net core. Make sure that such ops can't be interrupted by
e.g. (runtime-)suspending by taking the RTNL lock in suspend ops
and the PCI error handler.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: add rtl8169_up
Heiner Kallweit [Sat, 20 Jun 2020 20:37:50 +0000 (22:37 +0200)]
r8169: add rtl8169_up

Factor out bringing device up to a new function rtl8169_up(), similar
to rtl8169_down() for bringing the device down.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: remove no longer needed checks for device being runtime-active
Heiner Kallweit [Sat, 20 Jun 2020 20:37:01 +0000 (22:37 +0200)]
r8169: remove no longer needed checks for device being runtime-active

Because the netdevice is marked as detached now when parent is not
accessible we can remove quite some checks.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: mark device as not present when in PCI D3
Heiner Kallweit [Sat, 20 Jun 2020 20:36:26 +0000 (22:36 +0200)]
r8169: mark device as not present when in PCI D3

Mark the netdevice as detached whenever we go into PCI D3hot.
This allows to remove some checks e.g. from ethtool ops because
dev_ethtool() checks for netif_device_present() in the beginning.

In this context move waking up the queue out of rtl_reset_work()
because in cases where netif_device_attach() is called afterwards
the queue should be woken up by the latter function only.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: core: try to runtime-resume detached device in __dev_open
Heiner Kallweit [Sat, 20 Jun 2020 20:35:42 +0000 (22:35 +0200)]
net: core: try to runtime-resume detached device in __dev_open

A netdevice may be marked as detached because the parent is
runtime-suspended and not accessible whilst interface or link is down.
An example are PCI network devices that go into PCI D3hot, see e.g.
__igc_shutdown() or rtl8169_net_suspend().
If netdevice is down and marked as detached we can only open it if
we runtime-resume it before __dev_open() calls netif_device_present().

Therefore, if netdevice is detached, try to runtime-resume the parent
and only return with an error if it's still detached.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'prepare-dwmac-meson8b-for-G12A-specific-initialization'
David S. Miller [Mon, 22 Jun 2020 23:24:10 +0000 (16:24 -0700)]
Merge branch 'prepare-dwmac-meson8b-for-G12A-specific-initialization'

Martin Blumenstingl says:

====================
prepare dwmac-meson8b for G12A specific initialization

Some users are reporting that RGMII (and sometimes also RMII) Ethernet
is not working for them on G12A/G12B/SM1 boards. Upon closer inspection
of the vendor code for these SoCs new register bits are found.

It's not clear yet how these registers work. Add a new compatible string
as the first preparation step to improve Ethernet support on these SoCs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: stmmac: dwmac-meson8b: add a compatible string for G12A SoCs
Martin Blumenstingl [Sat, 20 Jun 2020 19:26:41 +0000 (21:26 +0200)]
net: stmmac: dwmac-meson8b: add a compatible string for G12A SoCs

Amlogic Meson G12A, G12B and SM1 have the same (at least as far as we
know at the time of writing) PRG_ETHERNET glue register implementation.
This implementation however is slightly different from AXG as it now has
an undocument "auto cali idx val" register in PRG_ETH1[17:16] which
seems to be related to RGMII Ethernet.

Add a new compatible string for G12A SoCs so the logic for this new
register can be implemented in the future.

Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodt-bindings: net: dwmac-meson: Add a compatible string for G12A onwards
Martin Blumenstingl [Sat, 20 Jun 2020 19:26:40 +0000 (21:26 +0200)]
dt-bindings: net: dwmac-meson: Add a compatible string for G12A onwards

Amlogic Meson G12A, G12B and SM1 have the same (at least as far as we
know at the time of writing) PRG_ETHERNET glue register implementation.
This implementation however is slightly different from AXG as it now has
an undocument "auto cali idx val" register in PRG_ETH1[17:16] which
seems to be related to RGMII Ethernet.

Add a compatible string for G12A and newer so the new registers can be
used.

Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'devlink-Add-board-serial_number-field-to-info_get-cb'
David S. Miller [Mon, 22 Jun 2020 23:15:22 +0000 (16:15 -0700)]
Merge branch 'devlink-Add-board-serial_number-field-to-info_get-cb'

Vasundhara Volam says:

====================
devlink: Add board.serial_number field to info_get cb.

This patchset adds support for board.serial_number to devlink info_get
cb and also use it in bnxt_en driver.

Sample output:

$ devlink dev info pci/0000:af:00.1
pci/0000:af:00.1:
  driver bnxt_en
  serial_number 00-10-18-FF-FE-AD-1A-00
  board.serial_number 433551F+172300000
  versions:
      fixed:
        board.id 7339763 Rev 0.
        asic.id 16D7
        asic.rev 1
      running:
        fw 216.1.216.0
        fw.psid 0.0.0
        fw.mgmt 216.1.192.0
        fw.mgmt.api 1.10.1
        fw.ncsi 0.0.0.0
        fw.roce 216.1.16.0

v2:
- Modify board_serial_number to board.serial_number for maintaining
consistency.
- Combine 2 lines in second patchset as column limit is 100 now
====================

Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobnxt_en: Add board.serial_number field to info_get cb
Vasundhara Volam [Sat, 20 Jun 2020 16:31:57 +0000 (22:01 +0530)]
bnxt_en: Add board.serial_number field to info_get cb

Add board.serial_number field info to info_get cb via devlink,
if driver can fetch the information from the device.

Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodevlink: Add support for board.serial_number to info_get cb.
Vasundhara Volam [Sat, 20 Jun 2020 16:31:56 +0000 (22:01 +0530)]
devlink: Add support for board.serial_number to info_get cb.

Board serial number is a serial number, often available in PCI
*Vital Product Data*.

Also, update devlink-info.rst documentation file.

Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Cosmetic-cleanup-in-SJA1105-DSA-driver'
David S. Miller [Mon, 22 Jun 2020 23:01:29 +0000 (16:01 -0700)]
Merge branch 'Cosmetic-cleanup-in-SJA1105-DSA-driver'

Vladimir Oltean says:

====================
Cosmetic cleanup in SJA1105 DSA driver

This removes the sparse warnings from the sja1105 driver and makes some
structures constant.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: make the instantiations of struct sja1105_info constant
Vladimir Oltean [Sat, 20 Jun 2020 17:18:32 +0000 (20:18 +0300)]
net: dsa: sja1105: make the instantiations of struct sja1105_info constant

Since struct sja1105_private only holds a const pointer to one of these
structures based on device tree compatible string, the structures
themselves can be made const.

Also add an empty line between each structure definition, to appease
checkpatch.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: make config table operation structures constant
Vladimir Oltean [Sat, 20 Jun 2020 17:18:31 +0000 (20:18 +0300)]
net: dsa: sja1105: make config table operation structures constant

The per-chip instantiations of struct sja1105_table_ops and struct
sja1105_dynamic_table_ops can be made constant, so do that.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: remove empty structures from config table ops
Vladimir Oltean [Sat, 20 Jun 2020 17:18:30 +0000 (20:18 +0300)]
net: dsa: sja1105: remove empty structures from config table ops

Sparse is complaining and giving the following warning message:
'Using plain integer as NULL pointer'.

This is not what's going on, instead {0} is used as a zero initializer
for the structure members, to indicate that the particular chip revision
does not support those particular config tables.

But since the config tables are declared globally, the unpopulated
elements are zero-initialized anyway. So, to make sparse shut up, let's
remove the zero initializers.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-dsa-qca8k-Improve-SGMII-interface-handling'
David S. Miller [Mon, 22 Jun 2020 22:54:34 +0000 (15:54 -0700)]
Merge branch 'net-dsa-qca8k-Improve-SGMII-interface-handling'

Jonathan McDowell says:

====================
net: dsa: qca8k: Improve SGMII interface handling

This 3 patch series migrates the qca8k switch driver over to PHYLINK,
and then adds the SGMII clean-ups (i.e. the missing initialisation) on
top of that as a second patch. The final patch is a simple spelling fix
in a comment.

As before, tested with a device where the CPU connection is RGMII (i.e.
the common current use case) + one where the CPU connection is SGMII. I
don't have any devices where the SGMII interface is brought out to
something other than the CPU.

v5:
- Move spelling fix to separate patch
- Use ds directly rather than ds->priv
v4:
- Enable pcs_poll so we keep phylink updated when doing in-band
  negotiation
- Explicitly check for PHY_INTERFACE_MODE_1000BASEX when setting SGMII
  port mode.
- Address Vladimir's review comments
v3:
- Move phylink changes to separate patch
- Address rmk review comments
v2:
- Switch to phylink
- Avoid need for device tree configuration options
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: qca8k: Minor comment spelling fix
Jonathan McDowell [Sat, 20 Jun 2020 10:31:16 +0000 (11:31 +0100)]
net: dsa: qca8k: Minor comment spelling fix

Signed-off-by: Jonathan McDowell <noodles@earth.li>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: qca8k: Improve SGMII interface handling
Jonathan McDowell [Sat, 20 Jun 2020 10:31:05 +0000 (11:31 +0100)]
net: dsa: qca8k: Improve SGMII interface handling

This patch improves the handling of the SGMII interface on the QCA8K
devices. Previously the driver did no configuration of the port, even if
it was selected. We now configure it up in the appropriate
PHY/MAC/Base-X mode depending on what phylink tells us we are connected
to and ensure it is enabled.

Tested with a device where the CPU connection is RGMII (i.e. the common
current use case) + one where the CPU connection is SGMII. I don't have
any devices where the SGMII interface is brought out to something other
than the CPU.

Signed-off-by: Jonathan McDowell <noodles@earth.li>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: qca8k: Switch to PHYLINK instead of PHYLIB
Jonathan McDowell [Sat, 20 Jun 2020 10:30:32 +0000 (11:30 +0100)]
net: dsa: qca8k: Switch to PHYLINK instead of PHYLIB

Update the driver to use the new PHYLINK callbacks, removing the
legacy adjust_link callback.

Signed-off-by: Jonathan McDowell <noodles@earth.li>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'bonding-initial-support-for-hardware-crypto-offload'
David S. Miller [Mon, 22 Jun 2020 22:38:57 +0000 (15:38 -0700)]
Merge branch 'bonding-initial-support-for-hardware-crypto-offload'

Jarod Wilson says:

====================
bonding: initial support for hardware crypto offload

This is an initial functional implementation for doing pass-through of
hardware encryption from bonding device to capable slaves, in active-backup
bond setups. This was developed and tested using ixgbe-driven Intel x520
interfaces with libreswan and a transport mode connection, primarily using
netperf, with assorted connection failures forced during transmission. The
failover works quite well in my testing, and overall performance is right
on par with offload when running on a bare interface, no bond involved.

Caveats: this is ONLY enabled for active-backup, because I'm not sure
how one would manage multiple offload handles for different devices all
running at the same time in the same xfrm, and it relies on some minor
changes to both the xfrm code and slave device driver code to get things
to behave, and I don't have immediate access to any other hardware that
could function similarly, but the NIC driver changes are minimal and
straight-forward enough that I've included what I think ought to be
enough for mlx5 devices too.

v2: reordered patches, switched (back) to using CONFIG_XFRM_OFFLOAD
to wrap the code additions and wrapped overlooked additions.
v3: rebase w/net-next open, add proper cc list to cover letter
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobonding: support hardware encryption offload to slaves
Jarod Wilson [Fri, 19 Jun 2020 14:31:55 +0000 (10:31 -0400)]
bonding: support hardware encryption offload to slaves

Currently, this support is limited to active-backup mode, as I'm not sure
about the feasilibity of mapping an xfrm_state's offload handle to
multiple hardware devices simultaneously, and we rely on being able to
pass some hints to both the xfrm and NIC driver about whether or not
they're operating on a slave device.

I've tested this atop an Intel x520 device (ixgbe) using libreswan in
transport mode, succesfully achieving ~4.3Gbps throughput with netperf
(more or less identical to throughput on a bare NIC in this system),
as well as successful failover and recovery mid-netperf.

v2: just use CONFIG_XFRM_OFFLOAD for wrapping, isolate more code with it

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlx5: become aware of when running as a bonding slave
Jarod Wilson [Fri, 19 Jun 2020 14:31:54 +0000 (10:31 -0400)]
mlx5: become aware of when running as a bonding slave

I've been unable to get my hands on suitable supported hardware to date,
but I believe this ought to be all that is needed to enable the mlx5
driver to also work with bonding active-backup crypto offload passthru.

CC: Boris Pismenny <borisp@mellanox.com>
CC: Saeed Mahameed <saeedm@mellanox.com>
CC: Leon Romanovsky <leon@kernel.org>
CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoixgbe_ipsec: become aware of when running as a bonding slave
Jarod Wilson [Fri, 19 Jun 2020 14:31:53 +0000 (10:31 -0400)]
ixgbe_ipsec: become aware of when running as a bonding slave

Slave devices in a bond doing hardware encryption also need to be aware
that they're slaves, so we operate on the slave instead of the bonding
master to do the actual hardware encryption offload bits.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Acked-by: Jeff Kirsher <Jeffrey.t.kirsher@intel.com>
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoxfrm: bail early on slave pass over skb
Jarod Wilson [Fri, 19 Jun 2020 14:31:52 +0000 (10:31 -0400)]
xfrm: bail early on slave pass over skb

This is prep work for initial support of bonding hardware encryption
pass-through support. The bonding driver will fill in the slave_dev
pointer, and we use that to know not to skb_push() again on a given
skb that was already processed on the bond device.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'devlink-Support-get-set-mac-address-of-a-port-function'
David S. Miller [Mon, 22 Jun 2020 22:29:19 +0000 (15:29 -0700)]
Merge branch 'devlink-Support-get-set-mac-address-of-a-port-function'

Parav Pandit says:

====================
devlink: Support get,set mac address of a port function

Currently, ip link set dev <pfndev> vf <vf_num> <param> <value> has
below few limitations.

1. Command is limited to set VF parameters only.
It cannot set the default MAC address for the PCI PF.

2. It can be set only on system where PCI SR-IOV capability exists.
In smartnic based system, eswitch of a NIC resides on a different
embedded cpu which has the VF and PF representors for the SR-IOV
functions of a host system in which this smartnic is plugged-in.

3. It cannot setup the function attributes of sub-function described
in detail in comprehensive RFC [1] and [2].

This series covers the first small part to let user query and set MAC
address (hardware address) of a PCI PF/VF which is represented by
devlink port pcipf, pcivf port flavours respectively.

Whenever a devlink port manages a function connected to a devlink port,
it allows to query and set its hardware address.

Driver implements necessary get/set callback functions if it supports
port function for a given port type.

Patch summary:
Patch-1 Prepares devlink port fill routines for extack
Patch-2 and 3 extended devlink interface to get/set port function
attributes, mainly hardware address to start with.

Patch-2 Extended port dump command to query port function hardware
address
Patch-3 Introduces a command to set the hardware address of a port
function

Patch-4 to 9 refactors and implement devlink callbacks in mlx5_core
driver.
Patch-4 Constify the mac address pointer in set routines
Patch-5 Introduces eswich check helper to use in devlink facing
callbacks
Patch-6 Moves port index, port number conversion routine to eswitch
header file
Patch-7 Implements port function query devlink callback
Patch-8 Refactors mac address setting routine to uniformly use
state_lock
Patch-9 Implements port function set devlink callback

[1] https://lore.kernel.org/netdev/20200519092258.GF4655@nanopsycho/
[2] https://marc.info/?l=linux-netdev&m=158555928517777&w=2
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: E-switch, Supporting setting devlink port function mac address
Parav Pandit [Fri, 19 Jun 2020 03:32:55 +0000 (03:32 +0000)]
net/mlx5: E-switch, Supporting setting devlink port function mac address

Enable user to set mac address of the PCI PF and VF port function.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: Split mac address setting function for using state_lock
Parav Pandit [Fri, 19 Jun 2020 03:32:54 +0000 (03:32 +0000)]
net/mlx5: Split mac address setting function for using state_lock

Refactor mac address setting function to let caller hold the necessary
state_lock mutex, so that subsequent patch and use this helper routine.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: E-switch, Support querying port function mac address
Parav Pandit [Fri, 19 Jun 2020 03:32:53 +0000 (03:32 +0000)]
net/mlx5: E-switch, Support querying port function mac address

Support querying mac address of the eswitch devlink port function.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: Move helper to eswitch layer
Parav Pandit [Fri, 19 Jun 2020 03:32:52 +0000 (03:32 +0000)]
net/mlx5: Move helper to eswitch layer

To use port number to port index conversion at eswitch level, move it to
eswitch header.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: E-switch, Introduce and use eswitch support check helper
Parav Pandit [Fri, 19 Jun 2020 03:32:51 +0000 (03:32 +0000)]
net/mlx5: E-switch, Introduce and use eswitch support check helper

Introduce an helper routine to get esw from a devlink device and use it
at eswitch callbacks and in subsequent patch.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: Constify mac address pointer
Parav Pandit [Fri, 19 Jun 2020 03:32:50 +0000 (03:32 +0000)]
net/mlx5: Constify mac address pointer

Since none of the functions need to modify the input mac address,
constify them.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/devlink: Support setting hardware address of port function
Parav Pandit [Fri, 19 Jun 2020 03:32:49 +0000 (03:32 +0000)]
net/devlink: Support setting hardware address of port function

PCI PF and VF devlink port can manage the function represented by a
devlink port.

Allow users to set port function's hardware address.

Example of a PCI VF port which supports a 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

$ devlink port function set pci/0000:06:00.0/2 hw_addr 00:11:22:33:44:55

$ 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 00:11:22:33:44:55

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/devlink: Support querying hardware address of port function
Parav Pandit [Fri, 19 Jun 2020 03:32:48 +0000 (03:32 +0000)]
net/devlink: Support querying hardware address of port function

PCI PF and VF devlink port can manage the function represented by
a devlink port.

Enable users to query port function's hardware address.

Example of a PCI VF port which supports a 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 00:11:22:33:44:66

$ devlink port show pci/0000:06:00.0/2 -jp
{
    "port": {
        "pci/0000:06:00.0/2": {
            "type": "eth",
            "netdev": "enp6s0pf0vf1",
            "flavour": "pcivf",
            "pfnum": 0,
            "vfnum": 1,
            "function": {
                "hw_addr": "00:11:22:33:44:66"
            }
        }
    }
}

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/devlink: Prepare devlink port functions to fill extack
Parav Pandit [Fri, 19 Jun 2020 03:32:47 +0000 (03:32 +0000)]
net/devlink: Prepare devlink port functions to fill extack

Prepare devlink port related functions to optionally fill up
the extack information which will be used in subsequent patch by port
function attribute(s).

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Marvell-mvpp2-improvements'
David S. Miller [Sun, 21 Jun 2020 04:38:26 +0000 (21:38 -0700)]
Merge branch 'Marvell-mvpp2-improvements'

Russell King says:

====================
Marvell mvpp2 improvements

This series primarily cleans up mvpp2, but also fixes a left-over from
91a208f2185a ("net: phylink: propagate resolved link config via
mac_link_up()").

Patch 1 introduces some port helpers:
  mvpp2_port_supports_xlg() - does the port support the XLG MAC
  mvpp2_port_supports_rgmii() - does the port support RGMII modes

Patch 2 introduces mvpp2_phylink_to_port(), rather than having repeated
  open coding of container_of().

Patch 3 introduces mvpp2_modify(), which reads-modifies-writes a
  register - I've converted the phylink specific code to use this
  helper.

Patch 4 moves the hardware control of the pause modes from
  mvpp2_xlg_config() (which is called via the phylink_config method)
  to mvpp2_mac_link_up() - a change that was missed in the above
  referenced commit.

v2: remove "inline" in patch 2.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: set xlg flow control in mvpp2_mac_link_up()
Russell King [Sat, 20 Jun 2020 09:21:42 +0000 (10:21 +0100)]
net: mvpp2: set xlg flow control in mvpp2_mac_link_up()

Set the flow control settings in mvpp2_mac_link_up() for 10G links
just as we do for 1G and slower links. This is now the preferred
location.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: add register modification helper
Russell King [Sat, 20 Jun 2020 09:21:37 +0000 (10:21 +0100)]
net: mvpp2: add register modification helper

Add a helper to read-modify-write a register, and use it in the phylink
helpers.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: add mvpp2_phylink_to_port() helper
Russell King [Sat, 20 Jun 2020 09:21:32 +0000 (10:21 +0100)]
net: mvpp2: add mvpp2_phylink_to_port() helper

Add a helper to convert the struct phylink_config pointer passed in
from phylink to the drivers internal struct mvpp2_port.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: add port support helpers
Russell King [Sat, 20 Jun 2020 09:21:26 +0000 (10:21 +0100)]
net: mvpp2: add port support helpers

The mvpp2 code has tests scattered amongst the code to determine
whether the port supports the XLG, and whether the port supports
RGMII mode.

Rather than having these tests scattered, provide a couple of helper
functions, so that future additions can ensure that they get these
tests correct.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoRemove redundant skb null check
Gaurav Singh [Fri, 19 Jun 2020 19:24:13 +0000 (15:24 -0400)]
Remove redundant skb null check

Remove the redundant null check for skb.

Signed-off-by: Gaurav Singh <gaurav1086@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'tcp-remove-two-indirect-calls-from-xmit-path'
David S. Miller [Sun, 21 Jun 2020 00:47:53 +0000 (17:47 -0700)]
Merge branch 'tcp-remove-two-indirect-calls-from-xmit-path'

Eric Dumazet says:

====================
tcp: remove two indirect calls from xmit path

__tcp_transmit_skb() does two indirect calls per packet, lets get rid
of them.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: remove indirect calls for icsk->icsk_af_ops->send_check
Eric Dumazet [Fri, 19 Jun 2020 19:12:35 +0000 (12:12 -0700)]
tcp: remove indirect calls for icsk->icsk_af_ops->send_check

Mitigate RETPOLINE costs in __tcp_transmit_skb()
by using INDIRECT_CALL_INET() wrapper.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: remove indirect calls for icsk->icsk_af_ops->queue_xmit
Eric Dumazet [Fri, 19 Jun 2020 19:12:34 +0000 (12:12 -0700)]
tcp: remove indirect calls for icsk->icsk_af_ops->queue_xmit

Mitigate RETPOLINE costs in __tcp_transmit_skb()
by using INDIRECT_CALL_INET() wrapper.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoof: mdio: preserve phy dev_flags in of_phy_connect()
Tao Ren [Thu, 18 Jun 2020 22:04:44 +0000 (15:04 -0700)]
of: mdio: preserve phy dev_flags in of_phy_connect()

Replace assignment "=" with OR "|=" for "phy->dev_flags" so "dev_flags"
configured in phy probe() function can be preserved.

The idea is similar to commit e7312efbd5de ("net: phy: modify assignment
to OR for dev_flags in phy_attach_direct").

Signed-off-by: Tao Ren <rentao.bupt@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: Avoid overwriting valid skb->napi_id
Amritha Nambiar [Thu, 18 Jun 2020 21:22:15 +0000 (14:22 -0700)]
net: Avoid overwriting valid skb->napi_id

This will be useful to allow busy poll for tunneled traffic. In case of
busy poll for sessions over tunnels, the underlying physical device's
queues need to be polled.

Tunnels schedule NAPI either via netif_rx() for backlog queue or
schedule the gro_cell_poll(). netif_rx() propagates the valid skb->napi_id
to the socket. OTOH, gro_cell_poll() stamps the skb->napi_id again by
calling skb_mark_napi_id() with the tunnel NAPI which is not a busy poll
candidate. This was preventing tunneled traffic to use busy poll. A valid
NAPI ID in the skb indicates it was already marked for busy poll by a
NAPI driver and hence needs to be copied into the socket.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoRemove redundant condition in qdisc_graft
Gaurav Singh [Thu, 18 Jun 2020 20:36:31 +0000 (16:36 -0400)]
Remove redundant condition in qdisc_graft

parent cannot be NULL here since its in the else part
of the if (parent == NULL) condition. Remove the extra
check on parent pointer.

Signed-off-by: Gaurav Singh <gaurav1086@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Ocelot-Felix-driver-cleanup'
David S. Miller [Sun, 21 Jun 2020 00:25:23 +0000 (17:25 -0700)]
Merge branch 'Ocelot-Felix-driver-cleanup'

Vladimir Oltean says:

====================
Ocelot/Felix driver cleanup

Some of the code in the mscc felix and ocelot drivers was added while in
a bit of a hurry. Let's take a moment and put things in relative order.

First 3 patches are sparse warning fixes.

Patches 4-9 perform some further splitting between mscc_felix,
mscc_ocelot, and the common hardware library they share. Meaning that
some code is being moved from the library into just the mscc_ocelot
module.

Patches 10-12 refactor the naming conventions in the existing VCAP code
(for tc-flower offload), since we're going to be adding some more code
for VCAP IS1 (previous tentatives already submitted here:
https://patchwork.ozlabs.org/project/netdev/cover/20200602051828.5734-1-xiaoliang.yang_1@nxp.com/),
and that code would be confusing to read and maintain using current
naming conventions.

No functional modification is intended. I checked that the VCAP IS2 code
still works by applying a tc ingress filter with an EtherType key and
'drop' action.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: unexpose ocelot_vcap_policer_{add,del}
Vladimir Oltean [Sat, 20 Jun 2020 15:43:47 +0000 (18:43 +0300)]
net: mscc: ocelot: unexpose ocelot_vcap_policer_{add,del}

Remove the function prototypes from ocelot_police.h and make these
functions static. We need to move them above their callers. Note that
moving the implementations to ocelot_police.c is not trivially possible
due to dependency on is2_entry_set() which is static to ocelot_vcap.c.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: generalize the "ACE/ACL" names
Vladimir Oltean [Sat, 20 Jun 2020 15:43:46 +0000 (18:43 +0300)]
net: mscc: ocelot: generalize the "ACE/ACL" names

Access Control Lists (and their respective Access Control Entries) are
specifically entries in the VCAP IS2, the security enforcement block,
according to the documentation.
Let's rename the structures and functions to something more generic, so
that VCAP IS1 structures (which would otherwise have to be called
Ingress Classification Entries) can reuse the same code without
confusion.

Some renaming that was done:

struct ocelot_ace_rule -> struct ocelot_vcap_filter
struct ocelot_acl_block -> struct ocelot_vcap_block
enum ocelot_ace_type -> enum ocelot_vcap_key_type
struct ocelot_ace_vlan -> struct ocelot_vcap_key_vlan
enum ocelot_ace_action -> enum ocelot_vcap_action
struct ocelot_ace_stats -> struct ocelot_vcap_stats
enum ocelot_ace_type -> enum ocelot_vcap_key_type
struct ocelot_ace_frame_* -> struct ocelot_vcap_key_*

No functional change is intended.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: rename ocelot_ace.{c, h} to ocelot_vcap.{c,h}
Vladimir Oltean [Sat, 20 Jun 2020 15:43:45 +0000 (18:43 +0300)]
net: mscc: ocelot: rename ocelot_ace.{c, h} to ocelot_vcap.{c,h}

Access Control Lists (and their respective Access Control Entries) are
specifically entries in the VCAP IS2, the security enforcement block,
according to the documentation.

Let's rename the files that deal with generic operations on the VCAP
TCAM, so that VCAP IS1 and ES0 can reuse the same code without
confusion.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: move net_device related functions to ocelot_net.c
Vladimir Oltean [Sat, 20 Jun 2020 15:43:44 +0000 (18:43 +0300)]
net: mscc: ocelot: move net_device related functions to ocelot_net.c

The ocelot hardware library shouldn't contain too much net_device
specific code, since it is shared with DSA which abstracts that
structure away. So much as much of this code as possible into the
mscc_ocelot driver and outside of the common library.

We're making an exception for MDB and LAG code. That is not yet exported
to DSA, but when it will, most of the code that's already in ocelot.c
will remain there. So, there's no point in moving code to ocelot_net.c
just to move it back later.

We could have moved all net_device code to ocelot_vsc7514.c directly,
but let's operate under the assumption that if a new switchdev ocelot
driver gets added, it'll define its SoC-specific stuff in a new
ocelot_vsc*.c file and it'll reuse the rest of the code.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: move ocelot_regs.c into ocelot_vsc7514.c
Vladimir Oltean [Sat, 20 Jun 2020 15:43:43 +0000 (18:43 +0300)]
net: mscc: ocelot: move ocelot_regs.c into ocelot_vsc7514.c

ocelot_regs.c actually shouldn't be part of the common library. It
describes the register map of the VSC7514 switch. The way ocelot
switches work, they'll have highly optimized register maps, so another
SoC will likely have the same registers but laid out completely
different in memory (so there's little room for reusing this structure).
So move it to ocelot_vsc7514.c instead.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: rename MSCC_OCELOT_SWITCH_OCELOT to MSCC_OCELOT_SWITCH
Vladimir Oltean [Sat, 20 Jun 2020 15:43:42 +0000 (18:43 +0300)]
net: mscc: ocelot: rename MSCC_OCELOT_SWITCH_OCELOT to MSCC_OCELOT_SWITCH

Putting 'ocelot' in the config's name twice just to say that 'it's the
ocelot driver running on the ocelot SoC' is a bit confusing. Instead,
it's just the ocelot driver. Now that we've renamed the previous symbol
that was holding the MSCC_OCELOT_SWITCH_OCELOT into *_LIB (because
that's what it is), we're free to use this name for the driver.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: convert MSCC_OCELOT_SWITCH into a library
Vladimir Oltean [Sat, 20 Jun 2020 15:43:41 +0000 (18:43 +0300)]
net: mscc: ocelot: convert MSCC_OCELOT_SWITCH into a library

Hide the CONFIG_MSCC_OCELOT_SWITCH option from users. It is meant to be
only a hardware library which is selected by the drivers that use it
(ocelot, felix).

Since it is "selected" from Kconfig, all its dependencies are manually
transferred to the driver that selects it. This is because "select" in
Kconfig language is a bit of a mess, and doesn't handle dependencies of
selected options quite right.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: rename module to mscc_ocelot
Vladimir Oltean [Sat, 20 Jun 2020 15:43:40 +0000 (18:43 +0300)]
net: mscc: ocelot: rename module to mscc_ocelot

mscc_ocelot is a slightly better name for a module than ocelot_board or
ocelot_vsc7514 is, so let's use that.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: rename ocelot_board.c to ocelot_vsc7514.c
Vladimir Oltean [Sat, 20 Jun 2020 15:43:39 +0000 (18:43 +0300)]
net: mscc: ocelot: rename ocelot_board.c to ocelot_vsc7514.c

To follow the model of felix and seville where we have one
platform-specific file, rename this file to the actual SoC it serves.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: access EtherType using __be16
Vladimir Oltean [Sat, 20 Jun 2020 15:43:38 +0000 (18:43 +0300)]
net: mscc: ocelot: access EtherType using __be16

Get rid of sparse "cast to restricted __be16" warnings.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: ocelot: use plain int when interacting with TCAM tables
Vladimir Oltean [Sat, 20 Jun 2020 15:43:37 +0000 (18:43 +0300)]
net: mscc: ocelot: use plain int when interacting with TCAM tables

sparse is rightfully complaining about the fact that:

warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
   26 |   __builtin_constant_p((l) > (h)), (l) > (h), 0)))
      |                            ^
note: in expansion of macro â€˜GENMASK_INPUT_CHECK’
   39 |  (GENMASK_INPUT_CHECK(h, l) + __GENMASK(h, l))
      |   ^~~~~~~~~~~~~~~~~~~
note: in expansion of macro â€˜GENMASK’
  127 |   mask = GENMASK(width, 0);
      |          ^~~~~~~

So replace the variables that go into GENMASK with plain, signed integer
types.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: felix: make vcap is2 keys and actions static
Vladimir Oltean [Sat, 20 Jun 2020 15:43:36 +0000 (18:43 +0300)]
net: dsa: felix: make vcap is2 keys and actions static

Get rid of some sparse warnings.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Strict-mode-for-VRF'
David S. Miller [Sun, 21 Jun 2020 00:22:23 +0000 (17:22 -0700)]
Merge branch 'Strict-mode-for-VRF'

Andrea Mayer says:

====================
Strict mode for VRF

This patch set adds the new "strict mode" functionality to the Virtual
Routing and Forwarding infrastructure (VRF). Hereafter we discuss the
requirements and the main features of the "strict mode" for VRF.

On VRF creation, it is necessary to specify the associated routing table used
during the lookup operations. Currently, there is no mechanism that avoids
creating multiple VRFs sharing the same routing table. In other words, it is not
possible to force a one-to-one relationship between a specific VRF and the table
associated with it.

The "strict mode" imposes that each VRF can be associated to a routing table
only if such routing table is not already in use by any other VRF.
In particular, the strict mode ensures that:

 1) given a specific routing table, the VRF (if exists) is uniquely identified;
 2) given a specific VRF, the related table is not shared with any other VRF.

Constraints (1) and (2) force a one-to-one relationship between each VRF and the
corresponding routing table.

The strict mode feature is designed to be network-namespace aware and it can be
directly enabled/disabled acting on the "strict_mode" parameter.
Read and write operations are carried out through the classic sysctl command on
net.vrf.strict_mode path, i.e: sysctl -w net.vrf.strict_mode=1.

Only two distinct values {0,1} are accepted by the strict_mode parameter:

 - with strict_mode=0, multiple VRFs can be associated with the same table.
   This is the (legacy) default kernel behavior, the same that we experience
   when the strict mode patch set is not applied;

 - with strict_mode=1, the one-to-one relationship between the VRFs and the
   associated tables is guaranteed. In this configuration, the creation of a VRF
   which refers to a routing table already associated with another VRF fails and
   the error is returned to the user.

The kernel keeps track of the associations between a VRF and the routing table
during the VRF setup, in the "management" plane. Therefore, the strict mode does
not impact the performance or the intrinsic functionality of the data plane in
any way.

When the strict mode is active it is always possible to disable the strict mode,
while the reverse operation is not always allowed.
Setting the strict_mode parameter to 0 is equivalent to removing the one-to-one
constraint between any single VRF and its associated routing table.

Conversely, if the strict mode is disabled and there are multiple VRFs that
refer to the same routing table, then it is prohibited to set the strict_mode
parameter to 1. In this configuration, any attempt to perform the operation will
lead to an error and it will be reported to the user.
To enable strict mode once again (by setting the strict_mode parameter to 1),
you must first remove all the VRFs that share common tables.

There are several use cases which can take advantage from the introduction of
the strict mode feature. In particular, the strict mode allows us to:

  i) guarantee the proper functioning of some applications which deal with
     routing protocols;

 ii) perform some tunneling decap operations which require to use specific
     routing tables for segregating and forwarding the traffic.

Considering (i), the creation of different VRFs that point to the same table
leads to the situation where two different routing entities believe they have
exclusive access to the same table. This leads to the situation where different
routing daemons can conflict for gaining routes control due to overlapping
tables. By enabling strict mode it is possible to prevent this situation which
often occurs due to incorrect configurations done by the users.
The ability to enable/disable the strict mode functionality does not depend on
the tool used for configuring the networking. In essence, the strict mode patch
solves, at the kernel level, what some other patches [1] had tried to solve at
the userspace level (using only iproute2) with all the related problems.

Considering (ii), the introduction of the strict mode functionality allows us
implementing the SRv6 End.DT4 behavior. Such behavior terminates a SR tunnel and
it forwards the IPv4 traffic according to the routes present in the routing
table supplied during the configuration. The SRv6 End.DT4 can be realized
exploiting the routing capabilities made available by the VRF infrastructure.
This behavior could leverage a specific VRF for forcing the traffic to be
forwarded in accordance with the routes available in the VRF table.
Anyway, in order to make the End.DT4 properly work, it must be guaranteed that
the table used for the route lookup operations is bound to one and only one VRF.
In this way, it is possible to use the table for uniquely retrieving the
associated VRF and for routing packets.

I would like to thank David Ahern for his constant and valuable support during
the design and development phases of this patch set.

Comments, suggestions and improvements are very welcome!
====================

Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: add selftest for the VRF strict mode
Andrea Mayer [Fri, 19 Jun 2020 22:54:47 +0000 (00:54 +0200)]
selftests: add selftest for the VRF strict mode

The new strict mode functionality is tested in different configurations and
on different network namespaces.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovrf: add l3mdev registration for table to VRF device lookup
Andrea Mayer [Fri, 19 Jun 2020 22:54:46 +0000 (00:54 +0200)]
vrf: add l3mdev registration for table to VRF device lookup

During the initialization phase of the VRF module, the callback for table
to VRF device lookup is registered in l3mdev.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovrf: add sysctl parameter for strict mode
Andrea Mayer [Fri, 19 Jun 2020 22:54:45 +0000 (00:54 +0200)]
vrf: add sysctl parameter for strict mode

Add net.vrf.strict_mode sysctl parameter.

When net.vrf.strict_mode=0 (default) it is possible to associate multiple
VRF devices to the same table. Conversely, when net.vrf.strict_mode=1 a
table can be associated to a single VRF device.

When switching from net.vrf.strict_mode=0 to net.vrf.strict_mode=1, a check
is performed to verify that all tables have at most one VRF associated,
otherwise the switch is not allowed.

The net.vrf.strict_mode parameter is per network namespace.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovrf: track associations between VRF devices and tables
Andrea Mayer [Fri, 19 Jun 2020 22:54:44 +0000 (00:54 +0200)]
vrf: track associations between VRF devices and tables

Add the data structures and the processing logic to keep track of the
associations between VRF devices and routing tables.
When a VRF is instantiated, it needs to refer to a given routing table.
For each table, we explicitly keep track of the existing VRFs that refer to
the table.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agol3mdev: add infrastructure for table to VRF mapping
Andrea Mayer [Fri, 19 Jun 2020 22:54:43 +0000 (00:54 +0200)]
l3mdev: add infrastructure for table to VRF mapping

Add infrastructure to l3mdev (the core code for Layer 3 master devices) in
order to find out the corresponding VRF device for a given table id.
Therefore, the l3mdev implementations:
 - can register a callback that returns the device index of the l3mdev
   associated with a given table id;
 - can offer the lookup function (table to VRF device).

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/sched: cls_u32: Use struct_size() in kzalloc()
Gustavo A. R. Silva [Thu, 18 Jun 2020 14:53:42 +0000 (09:53 -0500)]
net/sched: cls_u32: Use struct_size() in kzalloc()

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotaprio: Use struct_size() in kzalloc()
Gustavo A. R. Silva [Thu, 18 Jun 2020 14:46:48 +0000 (09:46 -0500)]
taprio: Use struct_size() in kzalloc()

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes. Also, remove unnecessary
variable _size_.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Clause-45-PHY-probing-improvements'
David S. Miller [Sat, 20 Jun 2020 03:17:15 +0000 (20:17 -0700)]
Merge branch 'Clause-45-PHY-probing-improvements'

Russell King says:

====================
Clause 45 PHY probing improvements

Last time this series was posted back in May, Florian reviewed the
patches, which was the only feedback I received.  I'm now posting
them without the RFC tag.

This series aims to improve the probing for Clause 45 PHYs.

The first four patches clean up get_phy_device() and called functions,
updating the kernel doc, adding information about the various error
return values.

We then provide better kerneldoc for get_phy_device(), describing what
is going on, and more importantly what the various return codes mean.

Patch 6 adds support for probing MMDs >= 8 to check for their presence.

Patch 7 changes get_phy_c45_ids() to only set the returned
devices_in_package if we successfully find a PHY.

Patch 8 splits the use of "devices in package" from the "mmds present".

Patch 9 expands our ID reading to cover the other MMDs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: read MMD ID from all present MMDs
Russell King [Thu, 18 Jun 2020 13:46:13 +0000 (14:46 +0100)]
net: phy: read MMD ID from all present MMDs

Expand the device_ids[] array to allow all MMD IDs to be read rather
than just the first 8 MMDs, but only read the ID if the MDIO_STAT2
register reports that a device really is present here for these new
devices to maintain compatibility with our current behaviour.  Note
that only a limited number of devices have MDIO_STAT2.

88X3310 PHY vendor MMDs do are marked as present in the
devices_in_package, but do not contain IEE 802.3 compatible register
sets in their lower space.  This avoids reading incorrect values as MMD
identifiers.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: split devices_in_package
Russell King [Thu, 18 Jun 2020 13:46:08 +0000 (14:46 +0100)]
net: phy: split devices_in_package

We have two competing requirements for the devices_in_package field.
We want to use it as a bit array indicating which MMDs are present, but
we also want to know if the Clause 22 registers are present.

Since "devices in package" is a term used in the 802.3 specification,
keep this as the as-specified values read from the PHY, and introduce
a new member "mmds_present" to indicate which MMDs are actually
present in the PHY, derived from the "devices in package" value.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: set devices_in_package only after validation
Russell King [Thu, 18 Jun 2020 13:46:03 +0000 (14:46 +0100)]
net: phy: set devices_in_package only after validation

Only set the devices_in_package to a non-zero value if we find a valid
value for this field, so we avoid leaving it set to e.g. 0x1fffffff.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: add support for probing MMDs >= 8 for devices-in-package
Russell King [Thu, 18 Jun 2020 13:45:58 +0000 (14:45 +0100)]
net: phy: add support for probing MMDs >= 8 for devices-in-package

Add support for probing MMDs above 7 for a valid devices-in-package
specifier, but only probe the vendor MMDs for this if they also report
that there the device is present in status register 2.  This avoids
issues where the MMD is implemented, but does not provide IEEE 802.3
compliant registers (such as the MV88X3310 PHY.)

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: reword get_phy_device() kerneldoc
Russell King [Thu, 18 Jun 2020 13:45:53 +0000 (14:45 +0100)]
net: phy: reword get_phy_device() kerneldoc

Reword the get_phy_device() kerneldoc to be more explicit about how
we probe for the PHY, and what the various return conditions signify.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: clean up get_phy_c22_id() invalid ID handling
Russell King [Thu, 18 Jun 2020 13:45:48 +0000 (14:45 +0100)]
net: phy: clean up get_phy_c22_id() invalid ID handling

Move the ID check from get_phy_device() into get_phy_c22_id(), which
simplifies get_phy_device(). The ID reading functions are now
responsible for indicating whether they found a PHY or not via their
return code - they must return -ENODEV when a PHY is not present.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: clean up get_phy_c45_ids() failure handling
Russell King [Thu, 18 Jun 2020 13:45:43 +0000 (14:45 +0100)]
net: phy: clean up get_phy_c45_ids() failure handling

When we decide that a PHY is not present, we do not need to go through
the hoops of setting *phy_id to 0xffffffff, and then return zero to
make get_phy_device() fail - we can return -ENODEV which will have the
same effect.

Doing so means we no longer have to pass a pointer to phy_id in, and
we can then clean up the clause 22 path in a similar way.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: clean up PHY ID reading
Russell King [Thu, 18 Jun 2020 13:45:37 +0000 (14:45 +0100)]
net: phy: clean up PHY ID reading

Rearrange the code to read the PHY IDs, so we don't call get_phy_id()
only to immediately call get_phy_c45_ids().  Move that logic into
get_phy_device(), which results in better readability.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: clean up cortina workaround
Russell King [Thu, 18 Jun 2020 13:45:32 +0000 (14:45 +0100)]
net: phy: clean up cortina workaround

Move the Cortina PHY workaround out of the "devices in package" loop;
it doesn't need to be in there as the control flow will terminate the
loop once we enter the workaround irrespective of the workaround's
outcome. The workaround is triggered by the ID being mostly 1's, which
will in any case terminate the loop.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: Use struct_size() helper
Gustavo A. R. Silva [Thu, 18 Jun 2020 13:35:00 +0000 (08:35 -0500)]
tipc: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: sja1105: Use struct_size() in kzalloc()
Gustavo A. R. Silva [Fri, 19 Jun 2020 18:10:07 +0000 (13:10 -0500)]
net: dsa: sja1105: Use struct_size() in kzalloc()

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Addresses-KSPP-ID: https://github.com/KSPP/linux/issues/83
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: Use struct_size() helper
Gustavo A. R. Silva [Fri, 19 Jun 2020 18:01:49 +0000 (13:01 -0500)]
cxgb4: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Addresses-KSPP-ID: https://github.com/KSPP/linux/issues/83
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethernet: ti: am65-cpsw-qos: Use struct_size() in devm_kzalloc()
Gustavo A. R. Silva [Fri, 19 Jun 2020 17:37:15 +0000 (12:37 -0500)]
ethernet: ti: am65-cpsw-qos: Use struct_size() in devm_kzalloc()

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes. Also, remove unnecessary
variable _size_.

This code was detected with the help of Coccinelle and, audited and
fixed manually.

Addresses-KSPP-ID: https://github.com/KSPP/linux/issues/83
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: icmp6: avoid indirect call for icmpv6_send()
Eric Dumazet [Fri, 19 Jun 2020 19:02:59 +0000 (12:02 -0700)]
ipv6: icmp6: avoid indirect call for icmpv6_send()

If IPv6 is builtin, we do not need an expensive indirect call
to reach icmp6_send().

v2: put inline keyword before the type to avoid sparse warnings.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'cxgb4-add-support-for-ethtool-n-tuple-filters'
David S. Miller [Fri, 19 Jun 2020 20:17:32 +0000 (13:17 -0700)]
Merge branch 'cxgb4-add-support-for-ethtool-n-tuple-filters'

Vishal Kulkarni says:

====================
cxgb4: add support for ethtool n-tuple filters

Patch 1: Adds data structure to maintain list of filters and handles init/dinit
 of the same.

Patch 2: Handles addition of filters via ETHTOOL_SRXCLSRLINS.

Patch 3: Handles deletion of filtes via ETHTOOL_SRXCLSRLDEL.

Patch 4: Handles viewing of added filters.

Patch 5: Adds FLOW_ACTION_QUEUE support.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add action to steer flows to specific Rxq
Vishal Kulkarni [Fri, 19 Jun 2020 14:21:39 +0000 (19:51 +0530)]
cxgb4: add action to steer flows to specific Rxq

Add support for queue action to steer Rx traffic
hitting the flows to specified Rxq.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add support to fetch ethtool n-tuple filters
Vishal Kulkarni [Fri, 19 Jun 2020 14:21:38 +0000 (19:51 +0530)]
cxgb4: add support to fetch ethtool n-tuple filters

Add support to fetch the requested ethtool n-tuple filters by
translating them from hardware spec to ethtool n-tuple spec.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add ethtool n-tuple filter deletion
Vishal Kulkarni [Fri, 19 Jun 2020 14:21:37 +0000 (19:51 +0530)]
cxgb4: add ethtool n-tuple filter deletion

Add support to delete ethtool n-tuple filter. Fetch the appropriate
filter region (HPFILTER, HASH, NORMAL) in which the filter exists,
and delete it from the respective region, accordingly.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add ethtool n-tuple filter insertion
Vishal Kulkarni [Fri, 19 Jun 2020 14:21:36 +0000 (19:51 +0530)]
cxgb4: add ethtool n-tuple filter insertion

Add support to parse and insert ethtool n-tuple filters.
Translate n-tuple spec to flow spec and use the existing tc-flower
offload infra to insert ethtool n-tuple filters.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add skeleton for ethtool n-tuple filters
Vishal Kulkarni [Fri, 19 Jun 2020 14:21:35 +0000 (19:51 +0530)]
cxgb4: add skeleton for ethtool n-tuple filters

Allocate and manage resources required for ethtool n-tuple filters.
Also fetch the HASH filter region size and calculate nhash entries.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qos offload add flow status with dropped count
Po Liu [Fri, 19 Jun 2020 06:01:07 +0000 (14:01 +0800)]
net: qos offload add flow status with dropped count

This patch adds a drop frames counter to tc flower offloading.
Reporting h/w dropped frames is necessary for some actions.
Some actions like police action and the coming introduced stream gate
action would produce dropped frames which is necessary for user. Status
update shows how many filtered packets increasing and how many dropped
in those packets.

v2: Changes
 - Update commit comments suggest by Jiri Pirko.

Signed-off-by: Po Liu <Po.Liu@nxp.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'cxgb4-add-support-to-read-write-flash'
David S. Miller [Fri, 19 Jun 2020 03:49:56 +0000 (20:49 -0700)]
Merge branch 'cxgb4-add-support-to-read-write-flash'

Vishal Kulkarni says:

====================
cxgb4: add support to read/write flash

This series of patches adds support to read/write different binary images
of serial flash present in Chelsio terminator.

V2 changes:
Patch 1: No change
Patch 2: No change
Patch 3: Fix 4 compilation warnings reported by C=1, W=1 flags
Patch 4: No change
Patch 5: No change
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add support to read serial flash
Vishal Kulkarni [Thu, 18 Jun 2020 06:05:56 +0000 (11:35 +0530)]
cxgb4: add support to read serial flash

This patch adds support to dump flash memory via
ethtool --get-dump

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add support to flash boot cfg image
Vishal Kulkarni [Thu, 18 Jun 2020 06:05:55 +0000 (11:35 +0530)]
cxgb4: add support to flash boot cfg image

Update set_flash to flash boot cfg image to flash region

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add support to flash boot image
Vishal Kulkarni [Thu, 18 Jun 2020 06:05:54 +0000 (11:35 +0530)]
cxgb4: add support to flash boot image

Update set_flash to flash boot image to flash region

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: add support to flash PHY image
Vishal Kulkarni [Thu, 18 Jun 2020 06:05:53 +0000 (11:35 +0530)]
cxgb4: add support to flash PHY image

Update set_flash to flash PHY image to flash region

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: update set_flash to flash different images
Vishal Kulkarni [Thu, 18 Jun 2020 06:05:52 +0000 (11:35 +0530)]
cxgb4: update set_flash to flash different images

Chelsio adapter contains different flash regions and each
region is used by different binary files. This patch adds
support to flash images like PHY firmware, boot and boot config
using ethtool -f N.

The N value mapping is as follows.
N = 0 : Parse image and decide which region to flash
N = 1 : Firmware
N = 2 : PHY firmware
N = 3 : boot image
N = 4 : boot cfg

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>"
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-tso-expand-to-UDP-support'
David S. Miller [Fri, 19 Jun 2020 03:46:23 +0000 (20:46 -0700)]
Merge branch 'net-tso-expand-to-UDP-support'

Eric Dumazet says:

====================
net: tso: expand to UDP support

With QUIC getting more attention these days, it is worth
implementing UDP direct segmentation, the same we did for TCP.

Drivers will need to advertize NETIF_F_GSO_UDP_L4 so that
GSO stack does not do the (more expensive) segmentation.

Note the two first patches are stable candidates, after
tests confirm they do not add regressions.

v2: addressed Jakub feedback :
   1) Added a prep patch for octeontx2-af
   2) calls tso_start() earlier in otx2_sq_append_tso()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: tso: add UDP segmentation support
Eric Dumazet [Thu, 18 Jun 2020 03:53:26 +0000 (20:53 -0700)]
net: tso: add UDP segmentation support

Note that like TCP, we do not support additional encapsulations,
and that checksums must be offloaded to the NIC.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: tso: cache transport header length
Eric Dumazet [Thu, 18 Jun 2020 03:53:25 +0000 (20:53 -0700)]
net: tso: cache transport header length

Add tlen field into struct tso_t, and change tso_start()
to return skb_transport_offset(skb) + tso->tlen

This removes from callers the need to use tcp_hdrlen(skb) and
will ease UDP segmentation offload addition.

v2: calls tso_start() earlier in otx2_sq_append_tso() [Jakub]

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>