openwrt/staging/blogic.git
6 years agortnetlink: Remove VLA usage
Kees Cook [Wed, 30 May 2018 22:20:52 +0000 (15:20 -0700)]
rtnetlink: Remove VLA usage

In the quest to remove all stack VLA usage from the kernel[1], this
allocates the maximum size expected for all possible types and adds
sanity-checks at both registration and usage to make sure nothing gets
out of sync. This matches the proposed VLA solution for nfnetlink[2]. The
values chosen here were based on finding assignments for .maxtype and
.slave_maxtype and manually counting the enums:

slave_maxtype (max 33):
IFLA_BRPORT_MAX     33
IFLA_BOND_SLAVE_MAX  9

maxtype (max 45):
IFLA_BOND_MAX       28
IFLA_BR_MAX         45
__IFLA_CAIF_HSI_MAX  8
IFLA_CAIF_MAX        4
IFLA_CAN_MAX        16
IFLA_GENEVE_MAX     12
IFLA_GRE_MAX        25
IFLA_GTP_MAX         5
IFLA_HSR_MAX         7
IFLA_IPOIB_MAX       4
IFLA_IPTUN_MAX      21
IFLA_IPVLAN_MAX      3
IFLA_MACSEC_MAX     15
IFLA_MACVLAN_MAX     7
IFLA_PPP_MAX         2
__IFLA_RMNET_MAX     4
IFLA_VLAN_MAX        6
IFLA_VRF_MAX         2
IFLA_VTI_MAX         7
IFLA_VXLAN_MAX      28
VETH_INFO_MAX        2
VXCAN_INFO_MAX       2

This additionally changes maxtype and slave_maxtype fields to unsigned,
since they're only ever using positive values.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com
[2] https://patchwork.kernel.org/patch/10439647/

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: Be explicit about DT or pdata
Andrew Lunn [Wed, 30 May 2018 22:15:42 +0000 (00:15 +0200)]
net: dsa: mv88e6xxx: Be explicit about DT or pdata

Make it explicit that either device tree is used or platform data.  If
neither is available, abort the probe.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 877b7cb0b6f2 ("net: dsa: mv88e6xxx: Add minimal platform_data support")
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ti: cpsw: include gpio/consumer.h
Arnd Bergmann [Wed, 30 May 2018 21:51:54 +0000 (23:51 +0200)]
net: ti: cpsw: include gpio/consumer.h

On platforms that don't always enable CONFIG_GPIOLIB, we run into
a build failure:

drivers/net/ethernet/ti/cpsw.c: In function 'cpsw_probe':
drivers/net/ethernet/ti/cpsw.c:3006:9: error: implicit declaration of function 'devm_gpiod_get_array_optional' [-Werror=implicit-function-declaration]
  mode = devm_gpiod_get_array_optional(&pdev->dev, "mode", GPIOD_OUT_LOW);
         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/ti/cpsw.c:3006:59: error: 'GPIOD_OUT_LOW' undeclared (first use in this function); did you mean 'GPIOF_INIT_LOW'?
  mode = devm_gpiod_get_array_optional(&pdev->dev, "mode", GPIOD_OUT_LOW);

Since we cannot rely on this to be visible from gpio.h, we have to include
gpio/consumer.h directly.

Fixes: 2652113ff043 ("net: ethernet: ti: Allow most drivers with COMPILE_TEST")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlx5-new-device-events'
David S. Miller [Thu, 31 May 2018 19:35:39 +0000 (15:35 -0400)]
Merge branch 'mlx5-new-device-events'

Saeed Mahameed says:

====================
Mellanox, mlx5 new device events

The following series is for mlx5-next tree [1], it adds the support of two
new device events, from Ilan Tayari:

1. High temperature warnings.
2. FPGA QP error event.

In case of no objection this series will be applied to mlx5-next tree
and will be sent later as a pull request to both rdma and net trees.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux.git/log/?h=mlx5-next

v1->v2:
  - improve commit message of the FPGA QP error event patch.
====================

Acked-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5: Add FPGA QP error event
Ilan Tayari [Wed, 30 May 2018 17:59:50 +0000 (10:59 -0700)]
net/mlx5: Add FPGA QP error event

The FPGA queue pair (QP) event fires whenever a QP on the FPGA
transitions to the error state.

At this stage, this event is unrecoverable, it may become recoverable
in the future.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5: Add temperature warning event to log
Ilan Tayari [Wed, 30 May 2018 17:59:49 +0000 (10:59 -0700)]
net/mlx5: Add temperature warning event to log

Temperature warning event is sent by FW to indicate high temperature
as detected by one of the sensors on the board.
Add handling of this event by writing the numbers of the alert sensors
to the kernel log.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agortnetlink: Add more well known protocol values
Donald Sharp [Wed, 30 May 2018 12:27:32 +0000 (08:27 -0400)]
rtnetlink: Add more well known protocol values

FRRouting installs routes into the kernel associated with
the originating protocol.  Add these values to the well
known values in rtnetlink.h.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: Add FORCE_PAUSE bit to 32 bit port caps
Ganesh Goudar [Wed, 30 May 2018 11:45:50 +0000 (17:15 +0530)]
cxgb4: Add FORCE_PAUSE bit to 32 bit port caps

Add FORCE_PAUSE bit to force local pause settings instead
of using auto negotiated values.

Signed-off-by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bridge-vlan-notify'
David S. Miller [Thu, 31 May 2018 18:13:45 +0000 (14:13 -0400)]
Merge branch 'bridge-vlan-notify'

Petr Machata says:

====================
net: bridge: Notify about bridge VLANs

In commit 946a11e7408e ("mlxsw: spectrum_span: Allow bridge for gretap
mirror"), mlxsw got support for offloading mirror-to-gretap such that
the underlay packet path involves a bridge. In that case, the offload is
also influenced by PVID setting of said bridge. However, changes to VLAN
configuration of the bridge itself do not generate switchdev
notifications, so there's no mechanism to prod mlxsw to update the
offload when these settings change.

In this patchset, the problem is resolved by distributing the switchdev
notification SWITCHDEV_OBJ_ID_PORT_VLAN also for configuration changes
on bridge VLANs. Since stacked devices distribute the notification to
lower devices, such event eventually reaches the driver, which can
determine whether it's a bridge or port VLAN by inspecting orig_dev.

To keep things consistent, the newly-distributed notifications observe
the same protocol as the existing ones: dual prepare/commit, with
-EOPNOTSUPP indicating lack of support, even though there's currently
nothing to prepare for and nothing to support. Correspondingly, all
switchdev drivers have been updated to return -EOPNOTSUPP for bridge
VLAN notifications.

In patches #1 and #2, the code base is changed to support the following
additions: functions br_switchdev_port_vlan_add() and
br_switchdev_port_vlan_del() are introduced to simplify sending
notifications; and br_vlan_add_existing() is introduced to later make it
simpler to add error-handling code for the case of configuring a
preexisting VLAN on bridge CPU port.

In patches #3-#6, respectively for mlxsw, rocker, DSA and DPAA2 ethsw,
the new notifications (which are not enabled yet) are ignored to
maintain the current behavior.

In patch #7, the notification is actually enabled.

In patch #8, mlxsw is changed to update offloads of mirror-to-gre also
for bridge-related notifications.

Changes from v3 to v4:

- In patch #1, separate variable declarations from program logic.
- Add patch #2.
- In patch #7, add error handling around a newly-introduced call to
  br_switchdev_port_vlan_add().
- Rephrase commit messages of patches #3-#6 to explain motivation for
  the change.

Changes from v2 to v3:

- Add a fallback definition for br_switchdev_port_obj_add() and
  br_switchdev_port_obj_del() when !CONFIG_NET_SWITCHDEV.

Changes from v1 to v2:

- Rename br_switchdev_port_obj_add() and br_switchdev_port_obj_del() to
  br_switchdev_port_vlan_add() and br_switchdev_port_vlan_del(), and
  move from br_vlan.c to br_switchdev.c.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_switchdev: Schedule respin during trans prepare
Petr Machata [Wed, 30 May 2018 01:00:49 +0000 (03:00 +0200)]
mlxsw: spectrum_switchdev: Schedule respin during trans prepare

Since there's no special support for the bridge events, the driver
returns -EOPNOTSUPP, and thus the commit never happens. Therefore
schedule respin during the prepare stage: there's no real difference one
way or another.

This fixes the problem that mirror-to-gretap offload wouldn't adapt to
changes in bridge vlan configuration right away and another notification
would have to arrive for mlxsw to catch up.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bridge: Notify about bridge VLANs
Petr Machata [Wed, 30 May 2018 01:00:45 +0000 (03:00 +0200)]
net: bridge: Notify about bridge VLANs

A driver might need to react to changes in settings of brentry VLANs.
Therefore send switchdev port notifications for these as well. Reuse
SWITCHDEV_OBJ_ID_PORT_VLAN for this purpose. Listeners should use
netif_is_bridge_master() on orig_dev to determine whether the
notification is about a bridge port or a bridge.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agostaging: fsl-dpaa2: ethsw: Ignore bridge VLAN events
Petr Machata [Wed, 30 May 2018 01:00:07 +0000 (03:00 +0200)]
staging: fsl-dpaa2: ethsw: Ignore bridge VLAN events

A follow-up patch enables emitting VLAN notifications for the bridge CPU
port in addition to the existing slave port notifications. These
notifications have orig_dev set to the bridge in question.

Because there's no specific support for these VLANs, just ignore the
notifications to maintain the current behavior.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodsa: port: Ignore bridge VLAN events
Petr Machata [Wed, 30 May 2018 00:59:26 +0000 (02:59 +0200)]
dsa: port: Ignore bridge VLAN events

A follow-up patch enables emitting VLAN notifications for the bridge CPU
port in addition to the existing slave port notifications. These
notifications have orig_dev set to the bridge in question.

Because there's no specific support for these VLANs, just ignore the
notifications to maintain the current behavior.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agorocker: rocker_main: Ignore bridge VLAN events
Petr Machata [Wed, 30 May 2018 00:58:36 +0000 (02:58 +0200)]
rocker: rocker_main: Ignore bridge VLAN events

A follow-up patch enables emitting VLAN notifications for the bridge CPU
port in addition to the existing slave port notifications. These
notifications have orig_dev set to the bridge in question.

Because there's no specific support for these VLANs, just ignore the
notifications to maintain the current behavior.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_switchdev: Ignore bridge VLAN events
Petr Machata [Wed, 30 May 2018 00:57:46 +0000 (02:57 +0200)]
mlxsw: spectrum_switchdev: Ignore bridge VLAN events

A follow-up patch enables emitting VLAN notifications for the bridge CPU
port in addition to the existing slave port notifications. These
notifications have orig_dev set to the bridge in question.

Because there's no specific support for these VLANs, just ignore the
notifications to maintain the current behavior.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bridge: Extract br_vlan_add_existing()
Petr Machata [Wed, 30 May 2018 00:56:55 +0000 (02:56 +0200)]
net: bridge: Extract br_vlan_add_existing()

Extract the code that deals with adding a preexisting VLAN to bridge CPU
port to a separate function. A follow-up patch introduces a need to roll
back operations in this block due to an error, and this split will make
the error-handling code clearer.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bridge: Extract boilerplate around switchdev_port_obj_*()
Petr Machata [Wed, 30 May 2018 00:56:03 +0000 (02:56 +0200)]
net: bridge: Extract boilerplate around switchdev_port_obj_*()

A call to switchdev_port_obj_add() or switchdev_port_obj_del() involves
initializing a struct switchdev_obj_port_vlan, a piece of code that
repeats on each call site almost verbatim. While in the current codebase
there is just one duplicated add call, the follow-up patches add more of
both add and del calls.

Thus to remove the duplication, extract the repetition into named
functions and reuse.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed*: Add link change count value to ethtool statistics display.
Sudarsana Reddy Kalluru [Tue, 29 May 2018 09:31:24 +0000 (02:31 -0700)]
qed*: Add link change count value to ethtool statistics display.

This patch adds driver changes for capturing the link change count in
ethtool statistics display.

Please consider applying this to "net-next".

Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mlx5e-updates-2018-05-29' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Thu, 31 May 2018 17:47:30 +0000 (13:47 -0400)]
Merge tag 'mlx5e-updates-2018-05-29' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5e-updates-2018-05-29

This series includes mlx5 FPGA and mlx5e netdevice updates:

1) Print FPGA info such as device name, vendor id, etc.., from Ilan Tayari.
2) Abort FPGA if some essential capabilities are not supported, from Yevgeny Kliteynik.
3) Two FPGA dma related minor fixes, from Ilya Lesokhin.
4) Use the right table to report offloaded TC rules, from Or Gerlitz.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: remove bypassed check in sch_direct_xmit()
Song Liu [Tue, 29 May 2018 17:03:21 +0000 (10:03 -0700)]
net: remove bypassed check in sch_direct_xmit()

Checking netif_xmit_frozen_or_stopped() at the end of sch_direct_xmit()
is being bypassed. This is because "ret" from sch_direct_xmit() will be
either NETDEV_TX_OK or NETDEV_TX_BUSY, and only ret == NETDEV_TX_OK == 0
will reach the condition:

    if (ret && netif_xmit_frozen_or_stopped(txq))
        return false;

This patch cleans up the code by removing the whole condition.

For more discussion about this, please refer to
   https://marc.info/?t=152727195700008

Signed-off-by: Song Liu <songliubraving@fb.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: minor optimization around tcp_hdr() usage in receive path
Yafang Shao [Tue, 29 May 2018 15:27:31 +0000 (23:27 +0800)]
tcp: minor optimization around tcp_hdr() usage in receive path

This is additional to the
commit ea1627c20c34 ("tcp: minor optimizations around tcp_hdr() usage").
At this point, skb->data is same with tcp_hdr() as tcp header has not
been pulled yet. So use the less expensive one to get the tcp header.

Remove the third parameter of tcp_rcv_established() and put it into
the function body.

Furthermore, the local variables are listed as a reverse christmas tree :)

Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMAINTAINERS: add myself as maintainer for QorIQ PTP clock driver
Yangbo Lu [Tue, 29 May 2018 03:47:44 +0000 (11:47 +0800)]
MAINTAINERS: add myself as maintainer for QorIQ PTP clock driver

Added myself as maintainer for QorIQ PTP clock driver.
Since gianfar_ptp.c was renamed to ptp_qoriq.c, let's
maintain it under QorIQ PTP clock driver.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodrivers/net: Fix various unnecessary characters after logging newlines
Joe Perches [Tue, 29 May 2018 02:51:57 +0000 (19:51 -0700)]
drivers/net: Fix various unnecessary characters after logging newlines

Remove and coalesce formats when there is an unnecessary
character after a logging newline.  These extra characters
cause logging defects.

Miscellanea:

o Coalesce formats

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: davinci: fix building davinci mdio code without CONFIG_OF
Arnd Bergmann [Mon, 28 May 2018 15:50:20 +0000 (17:50 +0200)]
net: davinci: fix building davinci mdio code without CONFIG_OF

Test-building this driver on targets without CONFIG_OF revealed a build
failure:

drivers/net/ethernet/ti/davinci_mdio.c: In function 'davinci_mdio_probe':
drivers/net/ethernet/ti/davinci_mdio.c:380:9: error: implicit declaration of function 'davinci_mdio_probe_dt'; did you mean 'davinci_mdio_probe'? [-Werror=implicit-function-declaration]

This adjusts the #ifdef logic in the driver to make it build in
all configurations.

Fixes: 2652113ff043 ("net: ethernet: ti: Allow most drivers with COMPILE_TEST")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: freescale: fix false-positive string overflow warning
Arnd Bergmann [Mon, 28 May 2018 15:49:46 +0000 (17:49 +0200)]
net: ethernet: freescale: fix false-positive string overflow warning

While compile-testing on arm64 with gcc-8.1, I ran into a build diagnostic:

drivers/net/ethernet/freescale/fec_main.c: In function 'fec_probe':
drivers/net/ethernet/freescale/fec_main.c:3517:25: error: '%d' directive writing between 1 and 10 bytes into a region of size 5 [-Werror=format-overflow=]
   sprintf(irq_name, "int%d", i);
                         ^~
drivers/net/ethernet/freescale/fec_main.c:3517:21: note: directive argument in the range [0, 2147483646]
   sprintf(irq_name, "int%d", i);
                     ^~~~~~~
drivers/net/ethernet/freescale/fec_main.c:3517:3: note: 'sprintf' output between 5 and 14 bytes into a destination of size 8
   sprintf(irq_name, "int%d", i);
   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~

It appears this has never shown on ppc32 or arm32 for an unknown reason, but
now gcc fails to identify that the 'irq_cnt' loop index has an upper bound
of 3, and instead uses a bogus range.

To work around the warning, this changes the sprintf to snprintf with the
correct buffer length.

Fixes: 78cc6e7ef957 ("net: ethernet: freescale: Allow FEC with COMPILE_TEST")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5e: Get the number of offloaded TC rules from the correct table
Or Gerlitz [Tue, 22 May 2018 17:18:36 +0000 (20:18 +0300)]
net/mlx5e: Get the number of offloaded TC rules from the correct table

As we keep the offloaded TC rules for NIC and e-switch in two different
places, make sure to return the number of offloaded flows according
to the use-case and not blindly from the priv.

Fixes: 655dc3d2b91b ('net/mlx5e: Use shared table for offloaded TC eswitch flows')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reported-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, Call DMA unmap with the right size
Ilya Lesokhin [Mon, 28 Aug 2017 06:47:32 +0000 (09:47 +0300)]
net/mlx5: FPGA, Call DMA unmap with the right size

When mlx5_fpga_conn_unmap_buf is called buf->sg[0].size
should equal the actual buffer size, not the message size.
Otherwise we will trigger the following dma debug warning
"DMA-API: device driver frees DMA memory with different size"

Fixes: 537a50574175 ('net/mlx5: FPGA, Add high-speed connection routines')
Signed-off-by: Ilya Lesokhin <ilyal@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, Properly initialize dma direction on fpga conn send
Ilya Lesokhin [Tue, 14 Nov 2017 08:30:55 +0000 (10:30 +0200)]
net/mlx5: FPGA, Properly initialize dma direction on fpga conn send

Properly initialize dma direction on fpga conn send.
Do not rely on dma_dir == 0 (DMA_BIDIRECTIONAL).

Signed-off-by: Ilya Lesokhin <ilyal@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, Abort FPGA init if the device reports no QP capability
Yevgeny Kliteynik [Wed, 8 Nov 2017 16:07:17 +0000 (18:07 +0200)]
net/mlx5: FPGA, Abort FPGA init if the device reports no QP capability

In the case that the reported max number of QPs capability
equals to zero, abort FPGA init.

Signed-off-by: Yevgeny Kliteynik <kliteyn@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, print SBU identification on init
Ilan Tayari [Tue, 4 Jul 2017 09:53:29 +0000 (12:53 +0300)]
net/mlx5: FPGA, print SBU identification on init

Add print of the following values on init:
1. ieee vendor id
2. sandbox product id
3. sandbox product version

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, Add device name
Ilan Tayari [Sun, 2 Jul 2017 07:47:24 +0000 (10:47 +0300)]
net/mlx5: FPGA, Add device name

Add device name for Mellanox FPGA devices.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: FPGA, Add doxygen for access type enum
Ilan Tayari [Tue, 25 Jul 2017 05:43:09 +0000 (08:43 +0300)]
net/mlx5: FPGA, Add doxygen for access type enum

Add doxygen comments for enum mlx5_fpga_access_type.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agobpfilter: fix a build err
YueHaibing [Fri, 25 May 2018 10:17:57 +0000 (18:17 +0800)]
bpfilter: fix a build err

gcc-7.3.0 report following err:

  HOSTCC  net/bpfilter/main.o
In file included from net/bpfilter/main.c:9:0:
./include/uapi/linux/bpf.h:12:10: fatal error: linux/bpf_common.h: No such file or directory
 #include <linux/bpf_common.h>

remove it by adding a include path.
Fixes: d2ba09c17a06 ("net: add skeleton of bpfilter kernel module")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: use data length instead of skb->len in tcp_probe
Yafang Shao [Fri, 25 May 2018 10:14:05 +0000 (18:14 +0800)]
tcp: use data length instead of skb->len in tcp_probe

skb->len is meaningless to user.
data length could be more helpful, with which we can easily filter out
the packet without payload.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocrypto: chtls: free beyond end rspq_skb_cache
Atul Gupta [Sun, 27 May 2018 15:45:22 +0000 (21:15 +0530)]
crypto: chtls: free beyond end rspq_skb_cache

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocrypto: chtls: kbuild warnings
Atul Gupta [Sun, 27 May 2018 15:45:21 +0000 (21:15 +0530)]
crypto: chtls: kbuild warnings

- unindented continue
- check for null page
- signed return

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocrypto: chtls: dereference null variable
Atul Gupta [Sun, 27 May 2018 15:45:20 +0000 (21:15 +0530)]
crypto: chtls: dereference null variable

skb dereferenced before check in sendpage

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocrypto: chtls: wait for memory sendmsg, sendpage
Atul Gupta [Sun, 27 May 2018 15:45:19 +0000 (21:15 +0530)]
crypto: chtls: wait for memory sendmsg, sendpage

address suspicious code <gustavo@embeddedor.com>

1210       set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1211       }

The issue is that in the code above, set_bit is never reached
due to the 'continue' statement at line 1208.

Also reported by bug report:<dan.carpenter@oracle.com>
1210       set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Not reachable.

Its required to wait for buffer in the send path and takes care of
unaddress and un-handled SOCK_NOSPACE.

v2: use csk_mem_free where appropriate
    proper indent of goto do_nonblock
    replace out with do_rm_wq

Reported-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocrypto:chtls: key len correction
Atul Gupta [Sun, 27 May 2018 15:45:18 +0000 (21:15 +0530)]
crypto:chtls: key len correction

corrected the key length to copy 128b key. Removed 192b and 256b
key as user input supports key of size 128b in gcm_ctx

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-Add-address-attribute-to-control-metric-of-prefix-route'
David S. Miller [Tue, 29 May 2018 14:12:46 +0000 (10:12 -0400)]
Merge branch 'net-Add-address-attribute-to-control-metric-of-prefix-route'

David Ahern says:

====================
net: Add address attribute to control metric of prefix route

For use cases such as VRR (Virtual Router Redundancy) interface managers
want efficient control over the order of prefix routes when multiple
interfaces have addresses with overlapping/duplicate subnets.

Currently, if two interfaces have addresses in the same subnet, the order
of the prefix route entries is determined by the order in which the
addresses are assigned or the links brought up. Any actions like cycling
an interface up and down changes that order. This set adds a new attribute
for addresses to allow a user to specify the metric of the prefix route
associated with an address giving interface managers better and more
efficient control of the order of prefix routes.

Patches 1-3 refactor IPv6 address add functions to pass an ifa6_config
struct. The functions currently have a long list of arguments and adding
the metric just makes it worse. Because of the overall diff size in
moving the arguments to a struct, the change is done in stages to make
it easier to review starting with the bottom function and pushing the
struct up to callers in each successive patch.

Patch 4 introduces the new attribute.

Patches 5 and 6 add support for the new attribute to IPv4 and IPv6
addresses.

Patch 7 adds a set of test cases.

Patch 8 adds support to iproute2

Changes since RFC
- collapsed patches 1 and 3 into patch 2
- simplified stack variables in fib_modify_prefix_metric in patch 5
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: fib_tests: Add prefix route tests with metric
David Ahern [Sun, 27 May 2018 15:09:59 +0000 (08:09 -0700)]
selftests: fib_tests: Add prefix route tests with metric

Add tests verifying prefix routes are inserted with expected metric.

IPv6 prefix route tests
    TEST: Default metric                                      [ OK ]
    TEST: User specified metric on first device               [ OK ]
    TEST: User specified metric on second device              [ OK ]
    TEST: Delete of address on first device                   [ OK ]
    TEST: Modify metric of address                            [ OK ]
    TEST: Prefix route removed on link down                   [ OK ]
    TEST: Prefix route with metric on link up                 [ OK ]

IPv4 prefix route tests
    TEST: Default metric                                      [ OK ]
    TEST: User specified metric on first device               [ OK ]
    TEST: User specified metric on second device              [ OK ]
    TEST: Delete of address on first device                   [ OK ]
    TEST: Modify metric of address                            [ OK ]
    TEST: Prefix route removed on link down                   [ OK ]
    TEST: Prefix route with metric on link up                 [ OK ]

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Add support for specifying metric of connected routes
David Ahern [Sun, 27 May 2018 15:09:58 +0000 (08:09 -0700)]
net/ipv6: Add support for specifying metric of connected routes

Add support for IFA_RT_PRIORITY to ipv6 addresses.

If the metric is changed on an existing address then the new route
is inserted before removing the old one. Since the metric is one
of the route keys, the prefix route can not be atomically replaced.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv4: Add support for specifying metric of connected routes
David Ahern [Sun, 27 May 2018 15:09:57 +0000 (08:09 -0700)]
net/ipv4: Add support for specifying metric of connected routes

Add support for IFA_RT_PRIORITY to ipv4 addresses.

If the metric is changed on an existing address then the new route
is inserted before removing the old one. Since the metric is one
of the route keys, the prefix route can not be replaced.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Add IFA_RT_PRIORITY address attribute
David Ahern [Sun, 27 May 2018 15:09:56 +0000 (08:09 -0700)]
net: Add IFA_RT_PRIORITY address attribute

Currently, if two interfaces have addresses in the same connected route,
then the order of the prefix route entries is determined by the order in
which the addresses are assigned or the links brought up.

Add IFA_RT_PRIORITY to allow user to specify the metric of the prefix
route associated with an address giving interface managers better
control of the order of prefix routes.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Pass ifa6_config struct to inet6_addr_modify
David Ahern [Sun, 27 May 2018 15:09:55 +0000 (08:09 -0700)]
net/ipv6: Pass ifa6_config struct to inet6_addr_modify

Update inet6_addr_modify to take ifa6_config argument versus a parameter
list. This is an argument move only; no functional change intended.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Pass ifa6_config struct to inet6_addr_add
David Ahern [Sun, 27 May 2018 15:09:54 +0000 (08:09 -0700)]
net/ipv6: Pass ifa6_config struct to inet6_addr_add

Move the creation of struct ifa6_config up to callers of inet6_addr_add.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Convert ipv6_add_addr to struct ifa6_config
David Ahern [Sun, 27 May 2018 15:09:53 +0000 (08:09 -0700)]
net/ipv6: Convert ipv6_add_addr to struct ifa6_config

Move config parameters for adding an ipv6 address to a struct. struct
names stem from inet6_rtm_newaddr which is the modern handler for
adding an address.

Start the conversion to ifa6_config with ipv6_add_addr. This is an argument
move only; no functional change intended. Mapping of variable changes:

    addr      -->  cfg->pfx
    peer_addr -->  cfg->peer_pfx
    pfxlen    -->  cfg->plen
    flags     -->  cfg->ifa_flags

scope, valid_lft, prefered_lft have the same names within cfg
(with corrected spelling).

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-use-MRSR-register-for-FW-reset'
David S. Miller [Tue, 29 May 2018 14:10:04 +0000 (10:10 -0400)]
Merge branch 'mlxsw-use-MRSR-register-for-FW-reset'

Ido Schimmel says:

====================
mlxsw: use MRSR register for FW reset

Jiri says:

Introduce a MRSR register definition and use it to do FW reset instead
of existing mechanism using PCI BAR0 register.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: pci: Utilize MRSR register to perform FW reset
Jiri Pirko [Sun, 27 May 2018 06:56:15 +0000 (09:56 +0300)]
mlxsw: pci: Utilize MRSR register to perform FW reset

So far, the PCI BAR0 register is used for triggering FW reset. However,
that is a legacy attitude and it is recommended to use MRSR to perform
reset instead. So do that. Move the reset into init() function as
the cmd interface needs to be used. With that, IRQ initialization needs
to be moved as well. As a side effect, the reset move simplifies
the devlink reload flow.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: cmd: Handle error after reset gracefully
Jiri Pirko [Sun, 27 May 2018 06:56:14 +0000 (09:56 +0300)]
mlxsw: cmd: Handle error after reset gracefully

There is an exception in command interface processing in case the MRSR
register is written to. The register triggers FW reset and during the
reset FW returns an error. So handle this by ignoring this error while
writing to MRSR register.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: reg: Add Management Reset and Shutdown Register
Jiri Pirko [Sun, 27 May 2018 06:56:13 +0000 (09:56 +0300)]
mlxsw: reg: Add Management Reset and Shutdown Register

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: qcom/emac: fix device tree initialization
Timur Tabi [Sun, 27 May 2018 01:29:14 +0000 (20:29 -0500)]
net: qcom/emac: fix device tree initialization

Commit "net: qcom/emac: Encapsulate sgmii ops under one structure"
introduced the sgmii_ops structure, but did not correctly initialize
it on device tree platforms.  This resulted in compiler warnings when
ACPI is not enabled.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5e: fix TLS dependency
Arnd Bergmann [Fri, 25 May 2018 21:36:06 +0000 (23:36 +0200)]
net/mlx5e: fix TLS dependency

With CONFIG_TLS=m and MLX5_CORE_EN=y, we get a link failure:

drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.o: In function `mlx5e_tls_handle_ooo':
tls_rxtx.c:(.text+0x24c): undefined reference to `tls_get_record'
drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.o: In function `mlx5e_tls_handle_tx_skb':
tls_rxtx.c:(.text+0x9a8): undefined reference to `tls_device_sk_destruct'

This narrows down the dependency to only allow the configurations
that will actually work. The existing dependency on TLS_DEVICE is
not sufficient here since MLX5_EN_TLS is a 'bool' symbol.

Fixes: c83294b9efa5 ("net/mlx5e: TLS, Add Innova TLS TX support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: remove unnecessary genlmsg_cancel() calls
YueHaibing [Sat, 26 May 2018 11:15:48 +0000 (19:15 +0800)]
net: remove unnecessary genlmsg_cancel() calls

the message be freed immediately, no need to trim it
back to the previous size.

Inspired by commit 7a9b3ec1e19f ("nl80211: remove unnecessary genlmsg_cancel() calls")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bpfilter: make function bpfilter_mbox_request() static
Wei Yongjun [Sat, 26 May 2018 09:47:26 +0000 (09:47 +0000)]
net: bpfilter: make function bpfilter_mbox_request() static

Fixes the following sparse warnings:

net/ipv4/bpfilter/sockopt.c:13:5: warning:
 symbol 'bpfilter_mbox_request' was not declared. Should it be static?

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'nfp-abm-RED-MQ-qdisc-offload'
David S. Miller [Tue, 29 May 2018 13:49:17 +0000 (09:49 -0400)]
Merge branch 'nfp-abm-RED-MQ-qdisc-offload'

Jakub Kicinski says:

====================
nfp: abm: RED/MQ qdisc offload

This is second batch of advanced buffer management nfp driver
changes.  This series adds the qdisc offload.  Support for
a very simple subset of RED qdisc offload is added as needed
for DCTCP ECN marking (min and max thresholds set to the same
value).

The first two patches fix glitches introduced by the previous
series.  We have to be careful about phys_port_name handling,
because VFs share the same code path, and some user space may
get confused by the names we chose.

Since unlike previous offloads we can report the queue backlog
both in bytes and packets we need to adjust how statistics are
added up in the core (patch 6).

There are some extra statistics we want to expose which don't
fit into TC stats, namely counts of packets which have been fast-
-forwarded without getting enqueued because there was no
contention and number of packets that were ever queued (sum of
all momentary backlogs).  We expose those through ethtool stats
(patches 8 and 9).

Remaining 5 patches add MQ offload - to be able to set different
configurations on different queues.  Representors are made multi-
-queue and we add offload support to MQ.  MQ stats are added up
before calling ->dump qdiscs on the children, and therefore don't
include updated offload values.  To avoid clearly incorrect stats
MQ is made to also request stats update from offloads.  This way
we can correct the diff at the driver level.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: report correct MQ stats
Jakub Kicinski [Sat, 26 May 2018 04:53:38 +0000 (21:53 -0700)]
nfp: abm: report correct MQ stats

Report the stat diff to make sure MQ stats add up to child stats.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: mq: request stats from offloads
Jakub Kicinski [Sat, 26 May 2018 04:53:37 +0000 (21:53 -0700)]
net: sched: mq: request stats from offloads

MQ doesn't hold any statistics on its own, however, statistic
from offloads are requested starting from the root, hence MQ
will read the old values for its sums.  Call into the drivers,
because of the additive nature of the stats drivers are aware
of how much "pending updates" they have to children of the MQ.
Since MQ reset its stats on every dump we can simply offset
the stats, predicting how stats of offloaded children will
change.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: multi-queue RED offload
Jakub Kicinski [Sat, 26 May 2018 04:53:36 +0000 (21:53 -0700)]
nfp: abm: multi-queue RED offload

Add support for MQ offload and setting RED parameters
on queue-by-queue basis.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: mq: add simple offload notification
Jakub Kicinski [Sat, 26 May 2018 04:53:35 +0000 (21:53 -0700)]
net: sched: mq: add simple offload notification

mq offload is trivial, we just need to let the device know
that the root qdisc is mq.  Alternative approach would be
to export qdisc_lookup() and make drivers check the root
type themselves, but notification via ndo_setup_tc is more
in line with other qdiscs.

Note that mq doesn't hold any stats on it's own, it just
adds up stats of its children.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: expose all PF queues
Jakub Kicinski [Sat, 26 May 2018 04:53:34 +0000 (21:53 -0700)]
nfp: abm: expose all PF queues

Allocate the PF representor as multi-queue to allow setting
the configuration per-queue.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: expose the internal stats in ethtool
Jakub Kicinski [Sat, 26 May 2018 04:53:33 +0000 (21:53 -0700)]
nfp: abm: expose the internal stats in ethtool

There is a handful of statistics exposing some internal details
of the implementation.  Expose those via ethtool.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: allow apps to add extra stats to ports
Jakub Kicinski [Sat, 26 May 2018 04:53:32 +0000 (21:53 -0700)]
nfp: allow apps to add extra stats to ports

Allow nfp apps to add extra ethtool stats.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: report statistics from RED offload
Jakub Kicinski [Sat, 26 May 2018 04:53:31 +0000 (21:53 -0700)]
nfp: abm: report statistics from RED offload

Report basic and extended RED statistics back to TC.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: add qstats.qlen to qlen
Jakub Kicinski [Sat, 26 May 2018 04:53:30 +0000 (21:53 -0700)]
net: sched: add qstats.qlen to qlen

AFAICT struct gnet_stats_queue.qlen is not used in Qdiscs.
It may, however, be useful for offloads to report HW queue
length there.  Add that value to the result of qdisc_qlen_sum().

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: add simple RED offload
Jakub Kicinski [Sat, 26 May 2018 04:53:29 +0000 (21:53 -0700)]
nfp: abm: add simple RED offload

Offload simple RED configurations.  For now support only DCTCP
like scenarios where min and max are the same.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: add helpers for configuring queue marking levels
Jakub Kicinski [Sat, 26 May 2018 04:53:28 +0000 (21:53 -0700)]
nfp: abm: add helpers for configuring queue marking levels

Queue levels for simple ECN marking are stored in _abi_nfd_out_q_lvls_X
symbol, where X is the PCIe PF id.  Find out the location of that symbol
and add helpers for modifying it.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: abm: enable advanced queuing on demand
Jakub Kicinski [Sat, 26 May 2018 04:53:27 +0000 (21:53 -0700)]
nfp: abm: enable advanced queuing on demand

ABM NIC FW has a cut-through mode where the PCIe queuing
is bypassed, thus working like our standard NIC FWs.  Use this
mode by default and only enable queuing in switchdev mode where
users can configure it.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: prefix vNIC phys_port_name with 'n'
Jakub Kicinski [Sat, 26 May 2018 04:53:26 +0000 (21:53 -0700)]
nfp: prefix vNIC phys_port_name with 'n'

Some drivers are using a bare number inside phys_port_name
as VF id and OpenStack's regexps will pick it up.  We can't
use a bare number for your vNICs, prefix the names with 'n'.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: return -EOPNOTSUPP from .ndo_get_phys_port_name for VFs
Jakub Kicinski [Sat, 26 May 2018 04:53:25 +0000 (21:53 -0700)]
nfp: return -EOPNOTSUPP from .ndo_get_phys_port_name for VFs

After recent change we started returning 0 from
ndo_get_phys_port_name for VFs.  The name parameter for
ndo_get_phys_port_name is not initialized by the stack so
this can lead to a crash.  We should have kept returning
-EOPNOTSUPP in the first place.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mlx5e-updates-2018-05-25' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Tue, 29 May 2018 13:45:13 +0000 (09:45 -0400)]
Merge tag 'mlx5e-updates-2018-05-25' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5e-updates-2018-05-25

This series includes updates for mlx5e netdev driver.

1) Allowr flow based VF vport mirroring under sriov switchdev scheme,
added support for offloading the TC mirred mirror sub-action, from
Chris Mi.

=================
From: Or Gerlitz <ogerlitz@mellanox.com>

The user will typically set the actions order such that the mirror
port (mirror VF) sees packets as the original port (VF under
mirroring) sent them or as it will receive them. In the general case,
it means that packets are potentially sent to the mirror port before
or after some actions were applied on them.

To properly do that, we follow on the exact action order as set for
the flow and make sure this will also be the case when we program the
HW offload.

If all the actions should apply before forwarding to the mirror and dest port,
mirroring is just multicasting to the two vports. Otherwise, we split
the TC flow to two HW rules, where the 1st applies only the actions
needed up to the mirror (if there are such) and the 2nd the rest of
the actions plus the forwarding to the dest vport.
=================

2) Move to order-0 only allocations (using fragmented work queues) for all
work queues used by the driver, RX and TX descriptor rings
(RQs, SQs and Completion Queues (CQs)), from Tariq Toukan.

3) Avoid resetting netdevice statistics on netdevice
state changes, from Eran Ben Elisha.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpfilter: fix building without CONFIG_INET
Arnd Bergmann [Tue, 29 May 2018 09:55:06 +0000 (11:55 +0200)]
bpfilter: fix building without CONFIG_INET

bpfilter_process_sockopt is a callback that gets called from
ip_setsockopt() and ip_getsockopt(). However, when CONFIG_INET is
disabled, it never gets called at all, and assigning a function to the
callback pointer results in a link failure:

net/bpfilter/bpfilter_kern.o: In function `__stop_umh':
bpfilter_kern.c:(.text.unlikely+0x3): undefined reference to `bpfilter_process_sockopt'
net/bpfilter/bpfilter_kern.o: In function `load_umh':
bpfilter_kern.c:(.init.text+0x73): undefined reference to `bpfilter_process_sockopt'

Since there is no caller in this configuration, I assume we can
simply make the assignment conditional.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'hns3-next'
David S. Miller [Tue, 29 May 2018 04:04:27 +0000 (00:04 -0400)]
Merge branch 'hns3-next'

Salil Mehta says:

====================
Misc. bug fixes & some minor additions to HNS3 driver

This patch-set provides some bug fixes figured out during testing
and review. It also provides some additions due to running of the
existing code on the new revision of the HNS3 hardware.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Fixes initalization of RoCE handle and makes it conditional
Lijun Ou [Fri, 25 May 2018 18:43:07 +0000 (19:43 +0100)]
net: hns3: Fixes initalization of RoCE handle and makes it conditional

When register a RoCE client with hnae3vf device, it needs to judge
the device whether support RoCE vf function. Otherwise, it will
lead to calltrace when RoCE is not support vf function and remove
roce device.

The calltrace as follows:
[   93.156614] Unable to handle kernel NULL pointer dereference at virtual address 00000015
 <SNIP>
[   93.278784] Call trace:
[   93.278788]  hnae3_match_n_instantiate+0x24/0xd8 [hnae3]
[   93.278790]  hnae3_register_client+0xcc/0x150 [hnae3]
[   93.278801]  hns_roce_hw_v2_init+0x18/0x1000 [hns_roce_hw_v2]
[   93.278805]  do_one_initcall+0x58/0x160
[   93.278807]  do_init_module+0x64/0x1d8
[   93.278809]  load_module+0x135c/0x15c8
[   93.278811]  SyS_finit_module+0x100/0x118
[   93.278816]  __sys_trace_return+0x0/0x4
[   93.278827] Code: aa0003f5 12001c56 aa1e03e0 d503201f (b9402660)

Fixes: e2cb1dec9779 ("net: hns3: Add HNS3 VF HCL(Hardware Compatibility Layer) Support")
Reported-by: Xinwei Kong <kong.kongxinwei@hisilicon.com>
Reported-by: Zhou Wang <wangzhou1@hisilicon.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Zhou Wang <wangzhou1@hisilicon.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Adds support for led locate command for copper port
Jian Shen [Fri, 25 May 2018 18:43:06 +0000 (19:43 +0100)]
net: hns3: Adds support for led locate command for copper port

Firmware now supports control of all leds. Existing HNS3 driver code
only supported led locate command over SFP Fibre ports. But now it
is also supported over copper port.
This patch removes existing not needed code for the led locate
command and updates the led control command between driver and
firmware.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Remove unused led control code
Jian Shen [Fri, 25 May 2018 18:43:05 +0000 (19:43 +0100)]
net: hns3: Remove unused led control code

In the previous implementation of led control for fibre port , parses the
port speed configuration, checks the link status and traffic status per
second, and updates the blink status of link led, traffic led and speed
led.

Now, the firmware takes responsibility to handle the led, the dirver just
needs to deal with locate command.

So the codes for link led, traffic led and speed led are useless now. This
patch removes these redundant codes.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Clear TX/RX rings when stopping port & un-initializing client
Fuyun Liang [Fri, 25 May 2018 18:43:04 +0000 (19:43 +0100)]
net: hns3: Clear TX/RX rings when stopping port & un-initializing client

When we down the port, some packets are left in TX/RX buffer. When we
up the port again, these old packets are forwarded to protocol stack
or are sent to internet. It will make some problem. TX/RX buffer should
be cleared when stopping port. This patch adds some function to ensure
the buffer is clean when port is started. We should clear the rings
when clients are being un-initialized as well.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Fuyun Liang <liangfuyun1@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Removes unnecessary check when clearing TX/RX rings
Fuyun Liang [Fri, 25 May 2018 18:43:03 +0000 (19:43 +0100)]
net: hns3: Removes unnecessary check when clearing TX/RX rings

Our code will ensure that hns3_clear_tx_ring is not used to cleared
RX rings and hns3_clear_rx_ring is not used to cleared TX rings. So
the ring type check is unnecessary.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Fuyun Liang <liangfuyun1@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Fixes the init of the VALID BD info in the descriptor
Fuyun Liang [Fri, 25 May 2018 18:43:02 +0000 (19:43 +0100)]
net: hns3: Fixes the init of the VALID BD info in the descriptor

RX Buffer Descriptor contains a VALID bit which indicates if the BD
is valid and has some data. This field is set by HNS3 hardware to
intimate the driver of some valid data present in the BD. nd should
be reset by the driver when BD is being used again. In the existing
code this bit was not being (re-)initialized properly and hence was
causing problems.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Fuyun Liang <liangfuyun1@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Fixes the state to indicate client-type initialization
Peng Li [Fri, 25 May 2018 18:43:01 +0000 (19:43 +0100)]
net: hns3: Fixes the state to indicate client-type initialization

HNAE3 module supports kernel nic driver, user nic driver and roce driver,
and there are 3 client types. Driver uses one bit(HNAE3_CLIENT_INITED_B)
to indicate the client initialization state, it will cause confusion
for 3 client types. This patch fixes it by use 3 bits to indicate the
initialization state.

Fixes: 38caee9d3ee8 ("net: hns3: Add support of the HNAE3 framework")
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Fix for PF mailbox receving unknown message
Xi Wang [Fri, 25 May 2018 18:43:00 +0000 (19:43 +0100)]
net: hns3: Fix for PF mailbox receving unknown message

Before the firmware updates the crq's tail pointer, if the PF driver
reads the data in the crq, the data may be incomplete at this time,
which will lead to the driver read an unknown message.

This patch fixes it by checking if crq is not empty before reading the
message.

Fixes: c1a81619d73a ("net: hns3: Add mailbox interrupt handling to PF driver")
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Add support to enable TX/RX promisc mode for H/W rev(0x21)
Peng Li [Fri, 25 May 2018 18:42:59 +0000 (19:42 +0100)]
net: hns3: Add support to enable TX/RX promisc mode for H/W rev(0x21)

HCLGE_PROMISC_TX_EN_B and HCLGE_PROMISC_RX_EN_B are not supported
on pdev revision(0x20), new revision(0x21) supports them.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Add STRP_TAGP field support for hardware revision 0x21
Peng Li [Fri, 25 May 2018 18:42:58 +0000 (19:42 +0100)]
net: hns3: Add STRP_TAGP field support for hardware revision 0x21

Hardware Revision(0x21) Buffer Descriptor adds a field STRP_TAGP
for vlan stripped processed indication. STRP_TAGP field has 2 bits,
bit 0 is stripped indication of the vlan tag in outer vlan tag
field, bit 1 is stripped indication of the vlan tag in inner vlan
tag field. For each bit, 0 indicates the tag is not stripped and
1 indicates the tag is stripped.

This patch adds STRP_TAGP support for revision(0x21), and does not
change the revision(0x20) action.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Add support for tx_accept_tag2 and tx_accept_untag2 config
Peng Li [Fri, 25 May 2018 18:42:57 +0000 (19:42 +0100)]
net: hns3: Add support for tx_accept_tag2 and tx_accept_untag2 config

HNS3 Hardware can support up to two VLAN tags in transmit leg, the PPP
module can handle the packets based on the tag1 and tag2 config. This
patch adds support for tag2 config for vlan handling

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: Updates RX packet info fetch in case of multi BD
Peng Li [Fri, 25 May 2018 18:42:56 +0000 (19:42 +0100)]
net: hns3: Updates RX packet info fetch in case of multi BD

In the latest revision of the hardware, if a packet is spanning
across multiple BDs then only VLD bit and current data size info
is valid in each BD, and rest of the information is only valid
in the last BD of the packet. In such case we should make sure
we are fetching RX packet size from the first descriptor and
information like VLAN should be fetched from last BD.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: shrink struct Qdisc
Paolo Abeni [Fri, 25 May 2018 14:28:44 +0000 (16:28 +0200)]
net: sched: shrink struct Qdisc

The struct Qdisc has a lot of holes, especially after commit
a53851e2c321 ("net: sched: explicit locking in gso_cpu fallback"),
which as a side effect, moved the fields just after 'busylock'
on a new cacheline.

Since both 'padded' and 'refcnt' are not updated frequently, and
there is a hole before 'gso_skb', we can move such fields there,
saving a cacheline without any performance side effect.

Before this commit:

pahole -C Qdisc net/sche/sch_generic.o
# ...
        /* size: 384, cachelines: 6, members: 25 */
        /* sum members: 236, holes: 3, sum holes: 92 */
        /* padding: 56 */

After this commit:
pahole -C Qdisc net/sche/sch_generic.o
# ...
/* size: 320, cachelines: 5, members: 25 */
/* sum members: 236, holes: 2, sum holes: 28 */
/* padding: 56 */

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqmi_wwan: apply SET_DTR quirk to the SIMCOM shared device ID
Bjørn Mork [Fri, 25 May 2018 13:00:20 +0000 (15:00 +0200)]
qmi_wwan: apply SET_DTR quirk to the SIMCOM shared device ID

SIMCOM are reusing a single device ID for many (all of their?)
different modems, based on different chipsets and firmwares. Newer
Qualcomm chipset generations require setting DTR to wake the QMI
function.  The SIM7600E modem is using such a chipset, making it
fail to work with this driver despite the device ID match.

Fix by unconditionally enabling the SET_DTR quirk for all SIMCOM
modems using this specific device ID.  This is similar to what
we already have done for another case of device IDs recycled over
multiple chipset generations: 14cf4a771b30 ("drivers: net: usb:
qmi_wwan: add QMI_QUIRK_SET_DTR for Telit PID 0x1201")

Initial testing on an older SIM7100 modem shows no immediate side
effects.

Reported-by: Sebastian Sjoholm <sebastian.sjoholm@gmail.com>
Cc: Reinhard Speyerer <rspmn@arcor.de>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'stmmac-add-support-for-stm32mp1'
David S. Miller [Tue, 29 May 2018 03:08:30 +0000 (23:08 -0400)]
Merge branch 'stmmac-add-support-for-stm32mp1'

Christophe Roullier says:

====================
net: ethernet: stmmac: add support for stm32mp1

Patches to have Ethernet support on stm32mp1
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodt-bindings: stm32: add compatible for syscon
Christophe Roullier [Fri, 25 May 2018 07:46:41 +0000 (09:46 +0200)]
dt-bindings: stm32: add compatible for syscon

This patch describes syscon DT bindings.

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: stmmac: add dwmac-4.20a compatible
Christophe Roullier [Fri, 25 May 2018 07:46:40 +0000 (09:46 +0200)]
net: stmmac: add dwmac-4.20a compatible

Manage dwmac-4.20a version from synopsys

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodt-bindings: stm32-dwmac: add support of MPU families
Christophe Roullier [Fri, 25 May 2018 07:46:39 +0000 (09:46 +0200)]
dt-bindings: stm32-dwmac: add support of MPU families

Add description for Ethernet MPU families fields

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: stmmac: add adaptation for stm32mp157c.
Christophe Roullier [Fri, 25 May 2018 07:46:38 +0000 (09:46 +0200)]
net: ethernet: stmmac: add adaptation for stm32mp157c.

Glue codes to support stm32mp157c device and stay
compatible with stm32 mcu familly

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Acked-by: Alexandre TORGUE <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodt-bindings: ptp: add ptp-qoriq.txt
Yangbo Lu [Fri, 25 May 2018 04:40:37 +0000 (12:40 +0800)]
dt-bindings: ptp: add ptp-qoriq.txt

This patch is to add a documentation for ptp_qoriq dt-bindings.
The description for ptp_qoriq dt-bindings was actually moved
from Documentation/devicetree/bindings/net/fsl-tsec-phy.txt,
since gianfar_ptp driver was moved to ptp_qoriq driver.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: gianfar_ethtool: get phc index through drvdata
Yangbo Lu [Fri, 25 May 2018 04:40:36 +0000 (12:40 +0800)]
net: ethernet: gianfar_ethtool: get phc index through drvdata

Global variable gfar_phc_index was used to get and store
phc index through gianfar_ptp driver. However gianfar_ptp
had been renamed as ptp_qoriq for QorIQ common PTP driver.
This gfar_phc_index doesn't work any more, and the phc index
is stored in drvdata now. This patch is to support getting
phc index through ptp_qoriq drvdata.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoptp_qoriq: move some definitions to header file
Yangbo Lu [Fri, 25 May 2018 04:40:35 +0000 (12:40 +0800)]
ptp_qoriq: move some definitions to header file

This patch is to move some definitions in ptp_qoriq.c
to the header file.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoptp: rework gianfar_ptp as QorIQ common PTP driver
Yangbo Lu [Fri, 25 May 2018 04:40:34 +0000 (12:40 +0800)]
ptp: rework gianfar_ptp as QorIQ common PTP driver

gianfar_ptp was the PTP clock driver for 1588 timer
module of Freescale QorIQ eTSEC (Enhanced Three-Speed
Ethernet Controllers) platforms. Actually QorIQ DPAA
(Data Path Acceleration Architecture) platforms is
also using the same 1588 timer module in hardware.

This patch is to rework gianfar_ptp as QorIQ common
PTP driver to support both DPAA and eTSEC. Moved
gianfar_ptp.c to drivers/ptp/, renamed it as
ptp_qoriq.c, and renamed many variables. There were
not any function changes.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoifb: fix packets checksum
Jon Maxwell [Thu, 24 May 2018 21:38:29 +0000 (07:38 +1000)]
ifb: fix packets checksum

Fixup the checksum for CHECKSUM_COMPLETE when pulling skbs on RX path.
Otherwise we get splats when tc mirred is used to redirect packets to ifb.

Before fix:

nic: hw csum failure

Signed-off-by: Jon Maxwell <jmaxwell37@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: realtek: add suspend/resume callbacks for RTL8211B
Heiner Kallweit [Thu, 24 May 2018 20:40:12 +0000 (22:40 +0200)]
net: phy: realtek: add suspend/resume callbacks for RTL8211B

Add RTL8211B suspend / resume callbacks.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'Enable-virtio_net-to-act-as-a-standby-for-a-passthru-device'
David S. Miller [Tue, 29 May 2018 02:59:55 +0000 (22:59 -0400)]
Merge branch 'Enable-virtio_net-to-act-as-a-standby-for-a-passthru-device'

Sridhar Samudrala says:

====================
Enable virtio_net to act as a standby for a passthru device

The main motivation for this patch is to enable cloud service providers
to provide an accelerated datapath to virtio-net enabled VMs in a
transparent manner with no/minimal guest userspace changes. This also
enables hypervisor controlled live migration to be supported with VMs that
have direct attached SR-IOV VF devices.

Patch 1 introduces a failover module that provides a generic interface for
paravirtual drivers to listen for netdev register/unregister/link change
events from pci ethernet devices with the same MAC and takeover their
datapath. The notifier and event handling code is based on the existing
netvsc implementation.

Patch 2 refactors netvsc to use the registration/notification framework
introduced by failover module.

Patch 3 introduces a net_failover driver that provides an automated
failover mechanism to paravirtual drivers via APIs to create and destroy
a failover master netdev and mananges a primary and standby slave netdevs
that get registered via the generic failover infrastructure.

Patch 4 introduces a new feature bit VIRTIO_NET_F_STANDBY to virtio-net
that can be used by hypervisor to indicate that virtio_net interface
should act as a standby for another device with the same MAC address.

Patch 5 extends virtio_net to use alternate datapath when available and
registered. When STANDBY feature is enabled, virtio_net driver uese the
net_failover API to create an additional 'failover' netdev that acts as
a master device and controls 2 slave devices.  The original virtio_net
netdev is registered as 'standby' netdev and a passthru/vf device with
the same MAC gets registered as 'primary' netdev. Both 'standby' and
'failover' netdevs are associated with the same 'pci' device.  The user
accesses the network interface via 'failover' netdev. The 'failover'
netdev chooses 'primary' netdev as default for transmits when it is
available with link up and running.

As this patch series is initially focusing on usecases where hypervisor
fully controls the VM networking and the guest is not expected to directly
configure any hardware settings, it doesn't expose all the ndo/ethtool ops
that are supported by virtio_net at this time. To support additional usecases,
it should be possible to enable additional ops later by caching the state
in failover netdev and replaying when the 'primary' netdev gets registered.

At the time of live migration, the hypervisor needs to unplug the VF device
from the guest on the source host and reset the MAC filter of the VF to
initiate failover of datapath to virtio before starting the migration. After
the migration is completed, the destination hypervisor sets the MAC filter
on the VF and plugs it back to the guest to switch over to VF datapath.

This patch is based on the discussion initiated by Jesse on this thread.
https://marc.info/?l=linux-virtualization&m=151189725224231&w=2

v12:
- Tested live migration with virtio-net/AVF(i40evf) configured in failover
  mode while running iperf in background. Tried static ip and dhcp
  configurations using 'network' scripts and Network Manager.
- Build tested netvsc module.
Updates:
- Extended generic failover module to do common functions like setting
  FAILOVER_SLAVE flag, registering rx-handler and linking to upper dev in
  the generic register/unregister handlers.
  This required adding 3 additional failover ops pre_register, pre_unregister
  and handle_frame.  netvsc and net_failover drivers are updated to support
  these ops.

v11:
- Split net_failover module into 2 components.
  1. 'failover' module that provides generic failover infrastructure
  to register a failover instance and listen for slave events.
  2. 'net_failover' driver that provides APIs to create/destroy upper
  netdev and supports 3-netdev model used by virtio-net.
- Added documentation

v10:
- fix net_failover_open() to update failover CARRIER correctly based on
  standby and primary states.
- fix net_failover_handle_frame() to handle frames received on standby
  when primary is present.
- replace netdev_upper_dev_link with netdev_master_upper_dev_link and
  handle lower dev state changes.
- fix net_failver_create() and net_failover_register() interfaces to
  use ERR_PTR and avoid arg **
- disable setting mac address when virtio-net in STANDBY mode
- document exported symbols
- added entry to MAINTAINERS file

v9:
Select NET_FAILOVER automatically when VIRTIO_NET/HYPERV_NET
are enabled. (stephen)

v8:
- Made the failover managment routines more robust by updating the feature
  bits/other fields in the failover netdev when slave netdevs are
  registered/unregistered. (mst)
- added support for handling vlans.
- Limited the changes in netvsc to only use the notifier/event/lookups
  from the failover module. The slave register/unregister/link-change
  handlers are only updated to use the getbymac routine to get the
  upper netdev. There is no change in their functionality. (stephen)
- renamed structs/function/file names to use net_failover prefix. (mst)

v7
- Rename 'bypass/active/backup' terminology with 'failover/primary/standy'
  (jiri, mst)
- re-arranged dev_open() and dev_set_mtu() calls in the register routines
  so that they don't get called for 2-netdev model. (stephen)
- fixed select_queue() routine to do queue selection based on VF if it is
  registered as primary. (stephen)
-  minor bugfixes

v6 RFC:
  Simplified virtio_net changes by moving all the ndo_ops of the
  bypass_netdev and create/destroy of bypass_netdev to 'bypass' module.
  avoided 2 phase registration(driver + instances).
  introduced IFF_BYPASS/IFF_BYPASS_SLAVE dev->priv_flags
  replaced mutex with a spinlock

v5 RFC:
  Based on Jiri's comments, moved the common functionality to a 'bypass'
  module so that the same notifier and event handlers to handle child
  register/unregister/link change events can be shared between virtio_net
  and netvsc.
  Improved error handling based on Siwei's comments.
v4:
- Based on the review comments on the v3 version of the RFC patch and
  Jakub's suggestion for the naming issue with 3 netdev solution,
  proposed 3 netdev in-driver bonding solution for virtio-net.
v3 RFC:
- Introduced 3 netdev model and pointed out a couple of issues with
  that model and proposed 2 netdev model to avoid these issues.
- Removed broadcast/multicast optimization and only use virtio as
  backup path when VF is unplugged.
v2 RFC:
- Changed VIRTIO_NET_F_MASTER to VIRTIO_NET_F_BACKUP (mst)
- made a small change to the virtio-net xmit path to only use VF datapath
  for unicasts. Broadcasts/multicasts use virtio datapath. This avoids
  east-west broadcasts to go over the PCI link.
- added suppport for the feature bit in qemu
====================

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio_net: Extend virtio to use VF datapath when available
Sridhar Samudrala [Thu, 24 May 2018 16:55:17 +0000 (09:55 -0700)]
virtio_net: Extend virtio to use VF datapath when available

This patch enables virtio_net to switch over to a VF datapath when STANDBY
feature is enabled and a VF netdev is present with the same MAC address.
It allows live migration of a VM with a direct attached VF without the need
to setup a bond/team between a VF and virtio net device in the guest.

It uses the API that is exported by the net_failover driver to create and
and destroy a master failover netdev. When STANDBY feature is enabled, an
additional netdev(failover netdev) is created that acts as a master device
and tracks the state of the 2 lower netdevs. The original virtio_net netdev
is marked as 'standby' netdev and a passthru device with the same MAC is
registered as 'primary' netdev.

The hypervisor needs to unplug the VF device from the guest on the source
host and reset the MAC filter of the VF to initiate failover of datapath
to virtio before starting the migration. After the migration is completed,
the destination hypervisor sets the MAC filter on the VF and plugs it back
to the guest to switch over to VF datapath.

This patch is based on the discussion initiated by Jesse on this thread.
https://marc.info/?l=linux-virtualization&m=151189725224231&w=2

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>