openwrt/staging/blogic.git
5 years agonet: phy: tja11xx: Add TJA11xx PHY driver
Marek Vasut [Fri, 24 May 2019 14:22:28 +0000 (16:22 +0200)]
net: phy: tja11xx: Add TJA11xx PHY driver

Add driver for the NXP TJA1100 and TJA1101 PHYs. These PHYs are special
BroadRReach 100BaseT1 PHYs used in automotive.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Heiner Kallweit <hkallweit1@gmail.com>
Cc: Jean Delvare <jdelvare@suse.com>
Cc: linux-hwmon@vger.kernel.org
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-hns3-add-aRFS-feature-and-fix-FEC-bugs-for-HNS3-driver'
David S. Miller [Sun, 26 May 2019 20:24:56 +0000 (13:24 -0700)]
Merge branch 'net-hns3-add-aRFS-feature-and-fix-FEC-bugs-for-HNS3-driver'

Huazhong Tan says:

====================
net: hns3: add aRFS feature and fix FEC bugs for HNS3 driver

This patchset adds some new features support and fixes some bugs:
[Patch 1/4 - 3/4] adds support for aRFS
[Patch 4/4] fix FEC configuration issue
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix for FEC configuration
Jian Shen [Fri, 24 May 2019 11:19:48 +0000 (19:19 +0800)]
net: hns3: fix for FEC configuration

The FEC capbility may be changed with port speed changes. Driver
needs to read the active FEC mode, and update FEC capability
when port speed changes.

Fixes: 7e6ec9148a1d ("net: hns3: add support for FEC encoding control")
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add aRFS support for PF
Jian Shen [Fri, 24 May 2019 11:19:47 +0000 (19:19 +0800)]
net: hns3: add aRFS support for PF

This patch adds aRFS support for PF. The aRFS rules are also
stored in the hardware flow director table, Use the existing
filter management functions to insert TCPv4/UDPv4/TCPv6/UDPv6
flow director filters. To avoid rule conflict, once user adds
flow director rules with ethtool, the aRFS will be disabled,
and clear exist aRFS rules. Once all user configure rules were
removed, aRFS can work again.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: refine the flow director handle
Jian Shen [Fri, 24 May 2019 11:19:46 +0000 (19:19 +0800)]
net: hns3: refine the flow director handle

In order to be compatible with aRFS rules, this patch adds
spin_lock for flow director rule adding, deleting, querying,
and packages the rule configuration.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: initialize CPU reverse mapping
Jian Shen [Fri, 24 May 2019 11:19:45 +0000 (19:19 +0800)]
net: hns3: initialize CPU reverse mapping

Allocate CPU rmap and add entry for each irq. CPU rmap is
used in aRFS to get the queue number of the rx completion
interrupts.

In additional, remove the calling of
irq_set_affinity_notifier() in hns3_nic_init_irq(), because
we have registered notifier in irq_cpu_rmap_add() for each
vector, otherwise it may cause use-after-free issue.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ath79-add-ag71xx-support'
David S. Miller [Sun, 26 May 2019 20:22:50 +0000 (13:22 -0700)]
Merge branch 'ath79-add-ag71xx-support'

Oleksij Rempel says:

====================
MIPS: ath79: add ag71xx support

2019.05.24 v6:
- ag71xx: remove double union
- ag71xx: reverse Christmas tree for all functions
- ag71xx: add Reviewed-by: Andrew Lunn <andrew@lunn.ch>

2019.05.20 v5:
- ag71xx: remove MII_CMD_WRITE, the name is confusing. It is
  actually disables MII_CMD_READ.
- ag71xx: rework ag71xx_mdio_mii_read/write
- ag71xx: set proper mask for the addr in ag71xx_mdio_mii_read/write
- Kconfig: remove MDIO_BITBANG
- ag71xx: ./scripts/checkpatch.pl it.

2019.05.19 v4:
- DT: define eth and mdio clocks
- ag71xx: remove module parameters
- ag71xx: return proper error value on mdio_read/write
- ag71xx: use proper mdio clock registration
- ag71xx: add ag71xx_dma_wait_stop() for ag71xx_dma_reset()
- ag71xx: remove ag71xx_speed_str()
- ag71xx: use phydev->link/sped/duplex instead of ag-> variants
- ag71xx: use WARN() instead of BUG()
- ag71xx: drop big part of ag71xx_phy_link_adjust()
- ag71xx: drop most of ag71xx_do_ioctl()
- ag71xx: register eth clock
- ag71xx: remove AG71XX_ETH0_NO_MDIO quirk.

2019.04.22 v3:
- ag71xx: use phy_modes() instead of ag71xx_get_phy_if_mode_name()
- ag71xx: remove .ndo_poll_controller support
- ag71xx: unregister_netdev before disconnecting phy.

2019.04.18 v2:
- ag71xx: add list of openwrt authors
- ag71xx: remove redundant PHY_POLL assignment
- ag71xx: use phy_attached_info instead of netif_info
- ag71xx: remove redundant netif_carrier_off() on .stop.
- DT: use "ethernet" instead of "eth"

This patch series provide ethernet support for many Atheros/QCA
MIPS based SoCs.

I reworked ag71xx driver which was previously maintained within OpenWRT
repository. So far, following changes was made to make upstreaming
easier:
- everything what can be some how used in user space was removed. Most
  of it was debug functionality.
- most of deficetree bindings was removed. Not every thing made sense
  and most of it is SoC specific, so it is possible to detect it by
  compatible.
- mac and mdio parts are merged in to one driver. It makes easier to
  maintaine SoC specific quirks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: add ag71xx driver
Oleksij Rempel [Fri, 24 May 2019 11:12:24 +0000 (13:12 +0200)]
net: ethernet: add ag71xx driver

Add support for Atheros/QCA AR7XXX/AR9XXX/QCA95XX built-in ethernet mac support

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMIPS: ath79: ar9331: add Ethernet nodes
Oleksij Rempel [Fri, 24 May 2019 11:12:23 +0000 (13:12 +0200)]
MIPS: ath79: ar9331: add Ethernet nodes

Add ethernet nodes supported by ag71xx driver.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: net: add qca,ar71xx.txt documentation
Oleksij Rempel [Fri, 24 May 2019 11:12:22 +0000 (13:12 +0200)]
dt-bindings: net: add qca,ar71xx.txt documentation

Add binding documentation for Atheros/QCA networking IP core used
in many routers.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'qed-Improve-performance-on-100G-link-for-offload-protocols'
David S. Miller [Sun, 26 May 2019 20:04:12 +0000 (13:04 -0700)]
Merge branch 'qed-Improve-performance-on-100G-link-for-offload-protocols'

Michal Kalderon says:

====================
qed*: Improve performance on 100G link for offload protocols

This patch series modifies the current implementation of PF selection.
The refactoring of the llh code enables setting additional filters
(mac / protocol) per PF, and improves performance for offload protocols
(RoCE, iWARP, iSCSI, fcoe) on 100G link (was capped at 90G per single
PF).

Improved performance on 100G link is achieved by configuring engine
affinty to each PF.
The engine affinity is read from the Management FW and hw is configured accordingly.
A new hw resource called PPFID is exposed and an API is introduced to utilize
it. This additional resource enables setting the affinity of a PF and providing
more classification rules per PF.
qedr,qedi,qedf are also modified as part of the series. Without the
changes functionality is broken.

v1 --> v2
---------
- Remove iWARP module parameter. Instead use devlink param infrastructure
  for setting the iwarp_cmt mode. Additional patch added to the series for
  adding the devlink support.

- Fix kbuild test robot warning on qed_llh_filter initialization.

- Remove comments inside function calls
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqedf: Use hwfns and affin_hwfn_idx to get MSI-X vector index to use
Chad Dupuis [Sun, 26 May 2019 12:22:30 +0000 (15:22 +0300)]
qedf: Use hwfns and affin_hwfn_idx to get MSI-X vector index to use

MSI-X vector index is determined using qed device information and
affinity to use.

Signed-off-by: Chad Dupuis <cdupuis@marvell.com>
Signed-off-by: Saurav Kashyap <skashyap@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqedi: Use hwfns and affin_hwfn_idx to get MSI-X vector index
Manish Rangankar [Sun, 26 May 2019 12:22:29 +0000 (15:22 +0300)]
qedi: Use hwfns and affin_hwfn_idx to get MSI-X vector index

MSI-X vector index is determined using qed device information and
affinity to use.

Signed-off-by: Manish Rangankar <mrangankar@marvell.com>
Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoRevert "scsi: qedi: Allocate IRQs based on msix_cnt"
Manish Rangankar [Sun, 26 May 2019 12:22:28 +0000 (15:22 +0300)]
Revert "scsi: qedi: Allocate IRQs based on msix_cnt"

 Always request for number of irqs equals to number of queues.

This reverts commit 1a291bce5eaf5374627d337157544aa6499ce34a.

Signed-off-by: Manish Rangankar <mrangankar@marvell.com>
Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed*: Add iWARP 100g support
Michal Kalderon [Sun, 26 May 2019 12:22:27 +0000 (15:22 +0300)]
qed*: Add iWARP 100g support

Add iWARP engine affinity setting for supporting iWARP over 100g.
iWARP cannot be distinguished by the LLH from L2, hence the
engine division will affect L2 as well. For this reason we add
a parameter to devlink to determine the engine division.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Add qed devlink parameters table
Michal Kalderon [Sun, 26 May 2019 12:22:26 +0000 (15:22 +0300)]
qed: Add qed devlink parameters table

The table currently contains a single parameter for
configuring whether iWARP should be enabled on a 100g
device. Enabling iWARP on a 100g device impacts L2
performance and is therefore not enabled by default.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Set the doorbell address correctly
Michal Kalderon [Sun, 26 May 2019 12:22:25 +0000 (15:22 +0300)]
qed: Set the doorbell address correctly

In 100g mode the doorbell bar is united for both engines. Set
the correct offset in the hwfn so that the doorbell returned
for RoCE is in the affined hwfn.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Denis Bolotin <denis.bolotin@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqedr: Change the MSI-X vectors selection to be based on affined engine
Michal Kalderon [Sun, 26 May 2019 12:22:24 +0000 (15:22 +0300)]
qedr: Change the MSI-X vectors selection to be based on affined engine

Use the msix vectors of the affined hwfn and not the
leading one.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Modify offload protocols to use the affined engine
Michal Kalderon [Sun, 26 May 2019 12:22:23 +0000 (15:22 +0300)]
qed: Modify offload protocols to use the affined engine

To enable 100g support for offload protocols each PF gets
a dedicated engine to work on from the MFW.
This patch modifies the code to use the affined hwfn instead
of the leading one.
The offload protocols require the ll2 to be opened on both
engines, and not just the affined hwfn.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed*: Change hwfn used for sb initialization
Michal Kalderon [Sun, 26 May 2019 12:22:22 +0000 (15:22 +0300)]
qed*: Change hwfn used for sb initialization

When initializing status blocks use the affined hwfn
instead of the leading one for RDMA / Storage

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Add llh ppfid interface and 100g support for offload protocols
Michal Kalderon [Sun, 26 May 2019 12:22:21 +0000 (15:22 +0300)]
qed: Add llh ppfid interface and 100g support for offload protocols

This patch refactors the current llh implementation. It exposes a hw
resource called ppfid (port-pfid) and implements an API for configuring
the resource. Default configuration which was used until now limited
the number of filters per PF and did not support engine affinity per
protocol. The new API enables allocating more filter rules per PF and
enables affinitizing protocol packets to a certain engine which
enables full 100g protocol offload support.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Modify api for performing a dmae to another PF
Michal Kalderon [Sun, 26 May 2019 12:22:20 +0000 (15:22 +0300)]
qed: Modify api for performing a dmae to another PF

This patch modifies the dmae API to enable performing a dmae operation
to another PF. This enables sharing between the llh entries between PFs
and thus increasing the amount of filters per PF under certain
configurations.
The llh entries require using the dmae since the memory is widebus,
which requires atomicity in access.

Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-mvpp2-Classifier-updates-RSS'
David S. Miller [Sat, 25 May 2019 23:38:16 +0000 (16:38 -0700)]
Merge branch 'net-mvpp2-Classifier-updates-RSS'

Maxime Chevallier says:

====================
net: mvpp2: Classifier updates, RSS

Here is a set of updates for the PPv2 classifier, the main feature being
the support for steering to RSS contexts, to leverage all the available
RSS tables in the controller.

The first two patches are non-critical fixes for the classifier, the
first one prevents us from allocating too much room to store the
classification rules, the second one configuring the C2 engine as
suggested by the PPv2 functionnal specs.

Patches 3 to 5 introduce support for RSS contexts in mvpp2, allowing us
to steer traffic to dedicated RSS tables.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvpp2: cls: Support steering to RSS contexts
Maxime Chevallier [Fri, 24 May 2019 10:05:54 +0000 (12:05 +0200)]
net: mvpp2: cls: Support steering to RSS contexts

When steering to an RXQ, we can perform an extra RSS step to assign a
queue from an RSS table.

This is done by setting the RSS_EN attribute in the C2 engine. In that
case, the RXQ that is assigned is the global RSS context id, that is
then translated to an RSS table using the RXQ2RSS table.

An example using ethtool to steer to RXQ 2 and 3 would be :

ethtool -X eth0 weight 0 0 1 1 context new

(This would print the allocated context id, let's say it's 1)

ethtool -N eth0 flow-type udp4 dst-port 1234 context 1 loc 0

The hash parameters are the ones that are globally configured for RSS :

ethtool -N eth0 rx-flow-hash udp4 sdfn

When an RSS context is removed while there are active classification
rules using this context, these rules are removed.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvpp2: cls: Extract the RSS context when parsing the ethtool rule
Maxime Chevallier [Fri, 24 May 2019 10:05:53 +0000 (12:05 +0200)]
net: mvpp2: cls: Extract the RSS context when parsing the ethtool rule

ethtool_rx_flow_rule_create takes into parameter the ethtool flow spec,
which doesn't contain the rss context id. We therefore need to extract
it ourself before parsing the ethtool rule.

The FLOW_RSS flag is only set in info->fs.flow_type, and not
info->flow_type.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvpp2: cls: Use RSS contexts to handle RSS tables
Maxime Chevallier [Fri, 24 May 2019 10:05:52 +0000 (12:05 +0200)]
net: mvpp2: cls: Use RSS contexts to handle RSS tables

The PPv2 controller has 8 RSS tables that are shared across all ports on
a given PPv2 instance. The previous implementation allocated one table
per port, leaving others unused.

By using RSS contexts, we can make use of multiple RSS tables per
port, one being the default table (always id 0), the other ones being
used as destinations for flow steering, in the same way as rx rings.

This commit introduces RSS contexts management in the PPv2 driver. We
always reserve one table per port, allocated when the port is probed.

The global table list is stored in the struct mvpp2, as it's a global
resource. Each port then maintains a list of indices in that global
table, that way each port can have it's own numbering scheme starting
from 0.

One limitation that seems unavoidable is that the hashing parameters are
shared across all RSS contexts for a given port. Hashing parameters for
ctx 0 will be applied to all contexts.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvpp2: cls: Bypass C2 internals FIFOs at init
Maxime Chevallier [Fri, 24 May 2019 10:05:51 +0000 (12:05 +0200)]
net: mvpp2: cls: Bypass C2 internals FIFOs at init

The C2 TCAM has internal FIFOs that are only useful for the built-in
self-tests. Disable these FIFOS at init, as recommended in the
functionnal specs.

Suggested-by: Alan Winkowski <walan@marvell.com>
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvpp2: cls: Use the correct number of rules in various places
Maxime Chevallier [Fri, 24 May 2019 10:05:50 +0000 (12:05 +0200)]
net: mvpp2: cls: Use the correct number of rules in various places

As of today, the classification offload implementation only supports 4
different rules to be offloaded. This number has been hardcoded in the
rule insertion function, and the wrong define is being used elsewhere.

Use the correct #define everywhere to make sure we always check for the
correct number of rules.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoflow_offload: use struct_size() in kzalloc()
Gustavo A. R. Silva [Thu, 23 May 2019 22:56:53 +0000 (17:56 -0500)]
flow_offload: use struct_size() in kzalloc()

One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:

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

instance = kzalloc(sizeof(struct foo) + count * sizeof(struct boo), GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:

instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-stmmac-Improvements-and-Selftests'
David S. Miller [Fri, 24 May 2019 20:45:57 +0000 (13:45 -0700)]
Merge branch 'net-stmmac-Improvements-and-Selftests'

Jose Abreu says:

====================
net: stmmac: Improvements and Selftests

[ Thanks to the introducion of selftests this series ended up being a misc
of improvements and the selftests additions per-se. ]

This introduces selftests support in stmmac driver. We add 9 basic sanity
checks and MAC loopback support for all cores within the driver. This way
more tests can easily be added in the future and can be run in virtually
any MAC/GMAC/QoS/XGMAC platform.

Having this we can find regressions and missing features in the driver
while at the same time we can check if the IP is correctly working.

We have been using this for some time now and I do have more tests to
submit in the feature. My experience is that although writing the tests
adds more development time, the gain results are obvious.

I let this feature optional within the driver under a Kconfig option.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Prevent missing interrupts when running NAPI
Jose Abreu [Fri, 24 May 2019 08:20:26 +0000 (10:20 +0200)]
net: stmmac: Prevent missing interrupts when running NAPI

When we trigger NAPI we are disabling interrupts but in case we receive
or send a packet in the meantime, as interrupts are disabled, we will
miss this event.

Trigger both NAPI instances (RX and TX) when at least one event happens
so that we don't miss any interrupts.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4/5: Clear unused address entries
Jose Abreu [Fri, 24 May 2019 08:20:25 +0000 (10:20 +0200)]
net: stmmac: dwmac4/5: Clear unused address entries

In case we don't use a given address entry we need to clear it because
it could contain previous values that are no longer valid.

Found out while running stmmac selftests.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwxgmac2: Do not disable whole RX in dma_stop_rx()
Jose Abreu [Fri, 24 May 2019 08:20:24 +0000 (10:20 +0200)]
net: stmmac: dwxgmac2: Do not disable whole RX in dma_stop_rx()

We don't need to disable the whole RX when dma_stop_rx() is called
because there may be the need of just disabling 1 DMA channel.

This is also needed for stmmac Flow Control selftest.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4/5: Do not disable whole RX in dma_stop_rx()
Jose Abreu [Fri, 24 May 2019 08:20:23 +0000 (10:20 +0200)]
net: stmmac: dwmac4/5: Do not disable whole RX in dma_stop_rx()

We don't need to disable the whole RX when dma_stop_rx() is called
because there may be the need of just disabling 1 DMA channel.

This is also needed for stmmac Flow Control selftest.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4/5: Fix Hash Filter
Jose Abreu [Fri, 24 May 2019 08:20:22 +0000 (10:20 +0200)]
net: stmmac: dwmac4/5: Fix Hash Filter

In order for hash filter to work we need to set the HPF bit.

Fout out while running stmmac selftests

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac1000: Clear unused address entries
Jose Abreu [Fri, 24 May 2019 08:20:21 +0000 (10:20 +0200)]
net: stmmac: dwmac1000: Clear unused address entries

In case we don't use a given address entry we need to clear it because
it could contain previous values that are no longer valid.

Found out while running stmmac selftests.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac1000: Fix Hash Filter
Jose Abreu [Fri, 24 May 2019 08:20:20 +0000 (10:20 +0200)]
net: stmmac: dwmac1000: Fix Hash Filter

In order for hash filter to work we need to set the HPF bit.

Found out while running stmmac selftests.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Introduce selftests support
Jose Abreu [Fri, 24 May 2019 08:20:19 +0000 (10:20 +0200)]
net: stmmac: Introduce selftests support

We add support for selftests on stmmac driver with 9 basic sanity checks
for now:
- MAC Loopback
- PHY Loopback
- MMC Counters
- EEE
- Hash Filter Multicast
- Perfect Filter Unicast
- Multicast Filter All
- Unicast Filter All
- Flow Control

This allows for fast tracking of regressions in the driver and helps in
spotting mis-configuration of HW.

Changes from v1:
- Fix build error as module (David)
- Check for link status before running tests
Changes from RFC v2:
- Return proper error code in stmmac_test_mmc (Corentin)
- Use only 1 MMC counter in stmmac_test_mmc (Alexandre)
Changes from RFC v1:
- Change test_loopback to test_mac_loopback (Andrew)
- Change timeout to retries (Andrew)
- Add MC/UC filter tests (Andrew)
- Only test in offline mode (Andrew)
- Do not call phy_loopback twice (Alexandre)

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Cc: Corentin Labbe <clabbe.montjoie@gmail.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwxgmac2: Also pass control frames while in promisc mode
Jose Abreu [Fri, 24 May 2019 08:20:18 +0000 (10:20 +0200)]
net: stmmac: dwxgmac2: Also pass control frames while in promisc mode

In order for the selftests to run the Flow Control selftest we need to
also pass pause frames to the stack.

Pass this type of frames while in promiscuous mode.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4/5: Also pass control frames while in promisc mode
Jose Abreu [Fri, 24 May 2019 08:20:17 +0000 (10:20 +0200)]
net: stmmac: dwmac4/5: Also pass control frames while in promisc mode

In order for the selftests to run the Flow Control selftest we need to
also pass pause frames to the stack.

Pass this type of frames while in promiscuous mode.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac1000: Also pass control frames while in promisc mode
Jose Abreu [Fri, 24 May 2019 08:20:16 +0000 (10:20 +0200)]
net: stmmac: dwmac1000: Also pass control frames while in promisc mode

In order for the selftests to run the Flow Control selftest we need to
also pass pause frames to the stack.

Pass this type of frames while in promiscuous mode.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Switch MMC functions to HWIF callbacks
Jose Abreu [Fri, 24 May 2019 08:20:15 +0000 (10:20 +0200)]
net: stmmac: Switch MMC functions to HWIF callbacks

XGMAC has a different MMC module. Lets use HWIF callbacks for MMC module
so that correct callbacks are automatically selected.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: stmmac: dwmac-sun8i: Enable control of loopback
Corentin Labbe [Fri, 24 May 2019 08:20:14 +0000 (10:20 +0200)]
net: ethernet: stmmac: dwmac-sun8i: Enable control of loopback

This patch enable use of set_mac_loopback in dwmac-sun8i

Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwxgmac2: Add MAC loopback support
Jose Abreu [Fri, 24 May 2019 08:20:13 +0000 (10:20 +0200)]
net: stmmac: dwxgmac2: Add MAC loopback support

In preparation for the addition of stmmac selftests we implement the MAC
loopback callback in dwxgmac2 core.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4/5: Add MAC loopback support
Jose Abreu [Fri, 24 May 2019 08:20:12 +0000 (10:20 +0200)]
net: stmmac: dwmac4/5: Add MAC loopback support

In preparation for the addition of stmmac selftests we implement the MAC
loopback callback in dwmac4/5 cores.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac1000: Add MAC loopback support
Jose Abreu [Fri, 24 May 2019 08:20:11 +0000 (10:20 +0200)]
net: stmmac: dwmac1000: Add MAC loopback support

In preparation for the addition of stmmac selftests we implement the MAC
loopback callback in dwmac1000 core.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac100: Add MAC loopback support
Jose Abreu [Fri, 24 May 2019 08:20:10 +0000 (10:20 +0200)]
net: stmmac: dwmac100: Add MAC loopback support

In preparation for the addition of stmmac selftests we implement the MAC
loopback callback in dwmac100 core.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Add MAC loopback callback to HWIF
Jose Abreu [Fri, 24 May 2019 08:20:09 +0000 (10:20 +0200)]
net: stmmac: Add MAC loopback callback to HWIF

In preparation for the addition of selftests support for stmmac we add a
new callback to HWIF that can be used to set the controller in loopback
mode.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Cc: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-phy-add-interface-mode-PHY_INTERFACE_MODE_USXGMII'
David S. Miller [Fri, 24 May 2019 20:39:34 +0000 (13:39 -0700)]
Merge branch 'net-phy-add-interface-mode-PHY_INTERFACE_MODE_USXGMII'

Heiner Kallweit says:

====================
net: phy: add interface mode PHY_INTERFACE_MODE_USXGMII

Add support for interface mode USXGMII.

On Freescale boards LS1043A and LS1046A a warning may pop up now
because mode xgmii should be changed to usxgmii (as the used
Aquantia PHY doesn't support XGMII).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: aquantia: add USXGMII support and warn if XGMII mode is set
Heiner Kallweit [Thu, 23 May 2019 18:09:08 +0000 (20:09 +0200)]
net: phy: aquantia: add USXGMII support and warn if XGMII mode is set

So far we didn't support mode USXGMII, and in order to not break few
boards mode XGMII was accepted for the AQR107 family even though it
doesn't support XGMII. Add USXGMII support to the Aquantia PHY driver
and warn if XGMII mode is set.

v2:
- add warning if XGMII mode is set

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: net: document new usxgmii phy mode
Heiner Kallweit [Thu, 23 May 2019 18:07:56 +0000 (20:07 +0200)]
dt-bindings: net: document new usxgmii phy mode

Add new interface mode USXGMII to binding documentation.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: add interface mode PHY_INTERFACE_MODE_USXGMII
Heiner Kallweit [Thu, 23 May 2019 18:06:49 +0000 (20:06 +0200)]
net: phy: add interface mode PHY_INTERFACE_MODE_USXGMII

Add support for interface mode PHY_INTERFACE_MODE_USXGMII.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests/net: SO_TXTIME with ETF and FQ
Willem de Bruijn [Thu, 23 May 2019 17:48:46 +0000 (13:48 -0400)]
selftests/net: SO_TXTIME with ETF and FQ

The SO_TXTIME API enables packet tranmission with delayed delivery.
This is currently supported by the ETF and FQ packet schedulers.

Evaluate the interface with both schedulers. Install the scheduler
and send a variety of packets streams: without delay, with one
delayed packet, with multiple ordered delays and with reordering.
Verify that packets are released by the scheduler in expected order.

The ETF qdisc requires a timestamp in the future on every packet. It
needs a delay on the qdisc else the packet is dropped on dequeue for
having a delivery time in the past. The test value is experimentally
derived. ETF requires clock_id CLOCK_TAI. It checks this base and
drops for non-conformance.

The FQ qdisc expects clock_id CLOCK_MONOTONIC, the base used by TCP
as of commit fb420d5d91c1 ("tcp/fq: move back to CLOCK_MONOTONIC").
Within a flow there is an expecation of ordered delivery, as shown by
delivery times of test 4. The FQ qdisc does not require all packets to
have timestamps and does not drop for non-conformance.

The large (msec) delays are chosen to avoid flakiness.

Output:

SO_TXTIME ipv6 clock monotonic
payload:a delay:28 expected:0 (us)

SO_TXTIME ipv4 clock monotonic
payload:a delay:38 expected:0 (us)

SO_TXTIME ipv6 clock monotonic
payload:a delay:40 expected:0 (us)

SO_TXTIME ipv4 clock monotonic
payload:a delay:33 expected:0 (us)

SO_TXTIME ipv6 clock monotonic
payload:a delay:10120 expected:10000 (us)

SO_TXTIME ipv4 clock monotonic
payload:a delay:10102 expected:10000 (us)

[.. etc ..]

OK. All tests passed

Changes v1->v2: update commit message output

Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ipv6-Move-exceptions-to-fib6_nh-and-make-it-optional-in-a-fib6_info'
David S. Miller [Fri, 24 May 2019 20:26:44 +0000 (13:26 -0700)]
Merge branch 'ipv6-Move-exceptions-to-fib6_nh-and-make-it-optional-in-a-fib6_info'

David Ahern says:

====================
ipv6: Move exceptions to fib6_nh and make it optional in a fib6_info

Patches 1 and 4 move pcpu and exception caches from fib6_info to fib6_nh.
With respect to the current FIB entries this is only a movement from one
struct to another contained within the first.

Patch 2 refactors the core logic of fib6_drop_pcpu_from into a helper
that is invoked per fib6_nh.

Patch 3 refactors exception handling in a similar way - creating a bunch
of helpers that can be invoked per fib6_nh with the goal of making patch
4 easier to review as well as creating the code needed for nexthop
objects.

Patch 5 makes a fib6_nh at the end of a fib6_info an array similar to
IPv4 and its fib_info. For the current fib entry model, all fib6_info
will have a fib6_nh allocated for it.

Patch 6 refactors ip6_route_del moving the code for deleting an
exception entry into a new function.

Patch 7 adds tests for redirect route exceptions. The new test was
written against 5.1 (before any of the nexthop refactoring). It and the
pmtu.sh selftest exercise the exception code paths - from creating
exceptions to cleaning them up on device delete. All tests pass without
any rcu locking or memleak warnings.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: Add redirect tests
David Ahern [Thu, 23 May 2019 03:28:01 +0000 (20:28 -0700)]
selftests: Add redirect tests

Add test for ICMP redirects and exception processing. Test is setup
for later addition of tests using nexthop objects for routing.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Refactor ip6_route_del for cached routes
David Ahern [Thu, 23 May 2019 03:28:00 +0000 (20:28 -0700)]
ipv6: Refactor ip6_route_del for cached routes

Move the removal of cached routes to a helper, ip6_del_cached_rt, that
can be invoked per nexthop. Rename the existig ip6_del_cached_rt to
__ip6_del_cached_rt since it is called by ip6_del_cached_rt.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Make fib6_nh optional at the end of fib6_info
David Ahern [Thu, 23 May 2019 03:27:59 +0000 (20:27 -0700)]
ipv6: Make fib6_nh optional at the end of fib6_info

Move fib6_nh to the end of fib6_info and make it an array of
size 0. Pass a flag to fib6_info_alloc indicating if the
allocation needs to add space for a fib6_nh.

The current code path always has a fib6_nh allocated with a
fib6_info; with nexthop objects they will be separate.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Move exception bucket to fib6_nh
David Ahern [Thu, 23 May 2019 03:27:58 +0000 (20:27 -0700)]
ipv6: Move exception bucket to fib6_nh

Similar to the pcpu routes exceptions are really per nexthop, so move
rt6i_exception_bucket from fib6_info to fib6_nh.

To avoid additional increases to the size of fib6_nh for a 1-bit flag,
use the lowest bit in the allocated memory pointer for the flushed flag.
Add helpers for retrieving the bucket pointer to mask off the flag.

The cleanup of the exception bucket is moved to fib6_nh_release.

fib6_nh_flush_exceptions can now be called from 2 contexts:
1. deleting a fib entry
2. deleting a fib6_nh

For 1., fib6_nh_flush_exceptions is called for a specific fib6_info that
is getting deleted. All exceptions in the cache using the entry are
deleted. For 2, the fib6_nh itself is getting destroyed so
fib6_nh_flush_exceptions is called for a NULL fib6_info which means
flush all entries.

The pmtu.sh selftest exercises the affected code paths - from creating
exceptions to cleaning them up on device delete. All tests pass without
any rcu locking or memleak warnings.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Refactor exception functions
David Ahern [Thu, 23 May 2019 03:27:57 +0000 (20:27 -0700)]
ipv6: Refactor exception functions

Before moving exception bucket from fib6_info to fib6_nh, refactor
rt6_flush_exceptions, rt6_remove_exception_rt, rt6_mtu_change_route,
and rt6_update_exception_stamp_rt. In all 3 cases, move the primary
logic into a new helper that starts with fib6_nh_. The latter 3
functions still take a fib6_info; this will be changed to fib6_nh
in the next patch.

In the case of rt6_mtu_change_route, move the fib6_metric_locked
out as a standalone check - no need to call the new function if
the fib entry has the mtu locked. Also, add fib6_info to
rt6_mtu_change_arg as a way of passing the fib entry to the new
helper.

No functional change intended. The goal here is to make the next
patch easier to review by moving existing lookup logic for each to
new helpers.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Refactor fib6_drop_pcpu_from
David Ahern [Thu, 23 May 2019 03:27:56 +0000 (20:27 -0700)]
ipv6: Refactor fib6_drop_pcpu_from

Move the existing pcpu walk in fib6_drop_pcpu_from to a new
helper, __fib6_drop_pcpu_from, that can be invoked per fib6_nh with a
reference to the from entries that need to be evicted. If the passed
in 'from' is non-NULL then only entries associated with that fib6_info
are removed (e.g., case where fib entry is deleted); if the 'from' is
NULL are entries are flushed (e.g., fib6_nh is deleted).

For fib6_info entries with builtin fib6_nh (ie., current code) there
is no change in behavior.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Move pcpu cached routes to fib6_nh
David Ahern [Thu, 23 May 2019 03:27:55 +0000 (20:27 -0700)]
ipv6: Move pcpu cached routes to fib6_nh

rt6_info are specific instances of a fib entry and are tied to a
device and gateway - ie., a nexthop. Before nexthop objects, IPv6 fib
entries have separate fib6_info for each nexthop in a multipath route,
so the location of the pcpu cache in the fib6_info struct worked.
However, with nexthop objects a fib6_info can point to a set of nexthops
(yet another alignment of ipv6 with ipv4). Accordingly, the pcpu
cache needs to be moved to the fib6_nh struct so the cached entries
are local to the nexthop specification used to create the rt6_info.

Initialization and free of the pcpu entries moved to fib6_nh_init and
fib6_nh_release.

Change in location only, from fib6_info down to fib6_nh; no other
functional change intended.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ENETC-support-hardware-timestamping'
David S. Miller [Fri, 24 May 2019 20:16:33 +0000 (13:16 -0700)]
Merge branch 'ENETC-support-hardware-timestamping'

Y.b. Lu says:

====================
ENETC: support hardware timestamping

This patch-set is to support hardware timestamping for ENETC
and also to add ENETC 1588 timer device tree node for ls1028a.

Because the ENETC RX BD ring dynamic allocation has not been
supported and it is too expensive to use extended RX BDs
if timestamping is not used, a Kconfig option is used to
enable extended RX BDs in order to support hardware
timestamping. This option will be removed once RX BD
ring dynamic allocation is implemented.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoarm64: dts: fsl: ls1028a: add ENETC 1588 timer node
Y.b. Lu [Thu, 23 May 2019 02:33:41 +0000 (02:33 +0000)]
arm64: dts: fsl: ls1028a: add ENETC 1588 timer node

Add ENETC 1588 timer node which is ENETC PF 4 (Physiscal Function 4).

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-binding: ptp_qoriq: support ENETC PTP compatible
Y.b. Lu [Thu, 23 May 2019 02:33:37 +0000 (02:33 +0000)]
dt-binding: ptp_qoriq: support ENETC PTP compatible

Add a new compatible for ENETC PTP.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoenetc: add get_ts_info interface for ethtool
Y.b. Lu [Thu, 23 May 2019 02:33:33 +0000 (02:33 +0000)]
enetc: add get_ts_info interface for ethtool

This patch is to add get_ts_info interface for ethtool
to support getting timestamping capability.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoenetc: add hardware timestamping support
Y.b. Lu [Thu, 23 May 2019 02:33:29 +0000 (02:33 +0000)]
enetc: add hardware timestamping support

This patch is to add hardware timestamping support
for ENETC. On Rx, timestamping is enabled for all
frames. On Tx, we only instruct the hardware to
timestamp the frames marked accordingly by the stack.

Because the RX BD ring dynamic allocation has not been
supported and it is too expensive to use extended RX BDs
if timestamping is not used, a Kconfig option is used to
enable extended RX BDs in order to support hardware
timestamping. This option will be removed once RX BD
ring dynamic allocation is implemented.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ll_temac: Fix compile error
Esben Haabendal [Fri, 24 May 2019 05:24:42 +0000 (07:24 +0200)]
net: ll_temac: Fix compile error

Fixes: 1b3fa5cf859b ("net: ll_temac: Cleanup multicast filter on change")
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Fri, 24 May 2019 00:52:43 +0000 (17:52 -0700)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2019-05-23

This series contains updates to ice driver only.

Anirudh cleans up white space issues and other code formatting issues in the
driver.  Also implemented LLDP persistence across reboots and start/stop of the
LLDP agent.  Updated print statements for driver capabilities to include
if it is a device or function capability.

Bruce cleaned up variable declarations by removing unneeded assignment.

Dave fixes a potential hang due to a couple of flows that recursively
acquire the RTNL lock which results in a deadlock.

Tony updates the driver to advertise what link modes we are capable of
when the user does not request a specific link mode.

Usha fixes up the LLDP MIB change event handling by cleaning up
workarounds and print the DCB configuration changes detected.

Brett fixes the driver to handle failures in the VF reset path, which
was failing to free resources upon an error.

Richard fixed the reported of stats via ethtool to align with our other
Intel drivers.

Jesse optimizes the transmit buffer and ring structures to have more
efficient ordering to get hot cache lines to have packed data.  Also
optimized the VF structure to use less memory, since it is used hundreds
of times throughout the driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoice: Silence semantic parser warnings
Bruce Allan [Tue, 16 Apr 2019 17:24:38 +0000 (10:24 -0700)]
ice: Silence semantic parser warnings

Recent versions of sparse warn about casting pointers to/from restricted
endian types in the Linux driver.  Silence those with the compiler
attribute __force macro from the Linux kernel to force casts to/from
restricted endian types.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Fix couple of issues in ice_vsi_release
Brett Creeley [Tue, 16 Apr 2019 17:24:37 +0000 (10:24 -0700)]
ice: Fix couple of issues in ice_vsi_release

Currently the driver is calling ice_napi_del() and then
unregister_netdev(). The call to unregister_netdev() will result in a
call to ice_stop() and then ice_vsi_close(). This is where we call
napi_disable() for all the MSI-X vectors. This flow is reversed so make
the changes to ensure napi_disable() happens prior to napi_del().

Before calling napi_del() and free_netdev() make sure
unregister_netdev() was called. This is done by making sure the
__ICE_DOWN bit is set in the vsi->state for the interested VSI.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Reorganize ice_vf struct
Jesse Brandeburg [Tue, 16 Apr 2019 17:24:36 +0000 (10:24 -0700)]
ice: Reorganize ice_vf struct

The ice_vf struct can be used hundreds of times in our
driver so it pays to use less memory per struct.

ice_vf prior to this commit:
  /* size: 112, cachelines: 2, members: 25 */
  /* sum members: 101, holes: 4, sum holes: 8 */
  /* bit holes: 2, sum bit holes: 11 bits */
  /* padding: 3 */
  /* last cacheline: 48 bytes */

ice_vf after this commit:
  /* size: 104, cachelines: 2, members: 25 */
  /* sum members: 100, holes: 3, sum holes: 4 */
  /* bit holes: 1, sum bit holes: 3 bits */
  /* last cacheline: 40 bytes */

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Use bitfields when possible
Jesse Brandeburg [Tue, 16 Apr 2019 17:24:35 +0000 (10:24 -0700)]
ice: Use bitfields when possible

We can use bit fields to store boolean values and when the
bit fields are next to each other, the compiler will combine them
(as long as the size holds enough).

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Reorganize tx_buf and ring structs
Jesse Brandeburg [Tue, 16 Apr 2019 17:24:34 +0000 (10:24 -0700)]
ice: Reorganize tx_buf and ring structs

Use more efficient structure ordering by using the pahole tool
and a lot of code inspection to get hot cache lines to have
packed data (no holes if possible) and adjacent warm data.

ice_ring prior to this change:
  /* size: 192, cachelines: 3, members: 23 */
  /* sum members: 158, holes: 4, sum holes: 12 */
  /* padding: 22 */

ice_ring after this change:
  /* size: 192, cachelines: 3, members: 25 */
  /* sum members: 162, holes: 1, sum holes: 1 */
  /* padding: 29 */

ice_tx_buf prior to this change:
  /* size: 48, cachelines: 1, members: 7 */
  /* sum members: 38, holes: 2, sum holes: 6 */
  /* padding: 4 */
  /* last cacheline: 48 bytes */

ice_tx_buf after this change:
  /* size: 40, cachelines: 1, members: 7 */
  /* sum members: 38, holes: 1, sum holes: 2 */
  /* last cacheline: 40 bytes */

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Format ethtool reported stats
Richard Rodriguez [Tue, 16 Apr 2019 17:24:33 +0000 (10:24 -0700)]
ice: Format ethtool reported stats

Fixes ethtool -S reported stats in ice driver to match
format and nomenclature of the ixgbe driver.

Signed-off-by: Richard Rodriguez <richard.rodriguez@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Gracefully handle reset failure in ice_alloc_vfs()
Brett Creeley [Tue, 16 Apr 2019 17:24:32 +0000 (10:24 -0700)]
ice: Gracefully handle reset failure in ice_alloc_vfs()

Currently if ice_reset_all_vfs() fails in ice_alloc_vfs() we fail to
free some resources, reset variables, and return an error value.
Fix this by adding another unroll case to free the pf->vf array, set
the pf->num_alloc_vfs to 0, and return an error code.

Without this, if ice_reset_all_vfs() fails in ice_alloc_vfs() we will
not be able to do SRIOV without hard rebooting the system because
rmmod'ing the driver does not work.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Refactor the LLDP MIB change event handling
Usha Ketineni [Tue, 16 Apr 2019 17:24:31 +0000 (10:24 -0700)]
ice: Refactor the LLDP MIB change event handling

This patch fixes the LLDP MIB change event handling code by removing
the workarounds in the current code. Added ice_dcb_need_recfg() to
print the DCB configuration changes detected via MIB change event.

Signed-off-by: Usha Ketineni <usha.k.ketineni@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Advertise supported link modes if none requested
Tony Nguyen [Tue, 16 Apr 2019 17:24:30 +0000 (10:24 -0700)]
ice: Advertise supported link modes if none requested

User requested link modes affect what is returned as an advertised
link mode.  If no modes have been requested, we are not advertising
any link modes.  Advertise what we are capable of supporting if no
link modes have been requested.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Fix hang when ethtool disables FW LLDP
Dave Ertman [Tue, 16 Apr 2019 17:24:29 +0000 (10:24 -0700)]
ice: Fix hang when ethtool disables FW LLDP

When disabling and enabling VSIs, there are a couple of flows
that recursively acquire the RTNL lock which causes a deadlock.
Fix that.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Call out dev/func caps when printing
Anirudh Venkataramanan [Tue, 16 Apr 2019 17:24:28 +0000 (10:24 -0700)]
ice: Call out dev/func caps when printing

ice_parse_caps is used to parse both device and function capabilities.
Currently, capabilities are printed with a cryptic "HW caps" prefix,
which makes it difficult to distinguish whether the capabilities being
printed are device or function capabilities.

This patch makes a change to add a "func cap" prefix when printing
function capabilities, and a "dev cap" prefix when printing device
capabilities.

This patch also changes some of the capability print strings for
consistency.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Remove braces for single statement blocks
Anirudh Venkataramanan [Tue, 16 Apr 2019 17:24:27 +0000 (10:24 -0700)]
ice: Remove braces for single statement blocks

Fix checkpatch warning "WARNING:BRACES: braces {} are not necessary
for single statement blocks"

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Cleanup an unnecessary variable initialization
Bruce Allan [Tue, 16 Apr 2019 17:24:26 +0000 (10:24 -0700)]
ice: Cleanup an unnecessary variable initialization

Commit 3463688e6ced ("ice: Add more validation in ice_vc_cfg_irq_map_msg")
added an assignment of vsi making the assignment during declaration
unnecessary.

Also, cleanup the declaration and assignment of irqmap_info to not use two
lines in the variable declaration section.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Implement LLDP persistence
Anirudh Venkataramanan [Tue, 16 Apr 2019 17:24:25 +0000 (10:24 -0700)]
ice: Implement LLDP persistence

Implement LLDP persistence across reboots, start and stop of LLDP agent.
Add additional parameter to ice_aq_start_lldp and ice_aq_stop_lldp.

Also change the ethtool private flag from "disable-fw-lldp" to
"enable-fw-lldp". This change will flip the boolean logic of the
functionality of the flag (on = enable, off = disable). The change
in name and functionality is to differentiate between the
pre-persistence and post-persistence states.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Fix double spacing
Anirudh Venkataramanan [Tue, 16 Apr 2019 17:21:28 +0000 (10:21 -0700)]
ice: Fix double spacing

Fix double spacing in ice_napi_disable_all

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agonet: qualcomm: rmnet: Move common struct definitions to include
Subash Abhinov Kasiviswanathan [Wed, 22 May 2019 20:21:07 +0000 (14:21 -0600)]
net: qualcomm: rmnet: Move common struct definitions to include

Create if_rmnet.h and move the rmnet MAP packet structs to this
common include file. To account for portablity, add little and
big endian bitfield definitions similar to the ip & tcp headers.

The definitions in the headers can now be re-used by the
upcoming ipa driver series as well as qmi_wwan.

Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoRevert "dpaa2-eth: configure the cache stashing amount on a queue"
Ioana Radulescu [Thu, 23 May 2019 14:38:22 +0000 (17:38 +0300)]
Revert "dpaa2-eth: configure the cache stashing amount on a queue"

This reverts commit f8b995853444aba9c16c1ccdccdd397527fde96d.

The reverted change instructed the QMan hardware block to fetch
RX frame annotation and beginning of frame data to cache before
the core would read them.

It turns out that in rare cases, it's possible that a QMan
stashing transaction is delayed long enough such that, by the time
it gets executed, the frame in question had already been dequeued
by the core and software processing began on it. If the core
manages to unmap the frame buffer _before_ the stashing transaction
is executed, an SMMU exception will be raised.

Unfortunately there is no easy way to work around this while keeping
the performance advantages brought by QMan stashing, so disable
it altogether.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: use firmware API for validating filter spec
Raju Rangoju [Thu, 23 May 2019 13:51:21 +0000 (19:21 +0530)]
cxgb4: use firmware API for validating filter spec

Adds support for validating hardware filter spec configured in firmware
before offloading exact match flows.

Use the new fw api FW_PARAM_DEV_FILTER_MODE_MASK to read the filter mode
and mask from firmware. If the api isn't supported, then fall-back to
older way of reading just the mode from indirect register.

Signed-off-by: Raju Rangoju <rajur@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-ll_temac-Fix-and-enable-multicast-support'
David S. Miller [Thu, 23 May 2019 16:33:57 +0000 (09:33 -0700)]
Merge branch 'net-ll_temac-Fix-and-enable-multicast-support'

Esben Haabendal says:

====================
net: ll_temac: Fix and enable multicast support

This patch series makes the necessary fixes to ll_temac driver to make
multicast work, and enables support for it.so that multicast support can

The main change is the change from mutex to spinlock of the lock used to
synchronize access to the shared indirect register access.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ll_temac: Enable multicast support
Esben Haabendal [Thu, 23 May 2019 12:02:22 +0000 (14:02 +0200)]
net: ll_temac: Enable multicast support

Multicast support have been tested and is working now.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ll_temac: Cleanup multicast filter on change
Esben Haabendal [Thu, 23 May 2019 12:02:21 +0000 (14:02 +0200)]
net: ll_temac: Cleanup multicast filter on change

Avoid leaving old address table entries when using multicast.  If more than
one multicast address were removed, only the first removed address would
actually be cleared.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ll_temac: Prepare indirect register access for multicast support
Esben Haabendal [Thu, 23 May 2019 12:02:20 +0000 (14:02 +0200)]
net: ll_temac: Prepare indirect register access for multicast support

With .ndo_set_rx_mode/temac_set_multicast_list() being called in atomic
context (holding addr_list_lock), and temac_set_multicast_list() needing
to access temac indirect registers, the mutex used to synchronize indirect
register is a no-no.

Replace it with a spinlock, and avoid sleeping in
temac_indirect_busywait().

To avoid excessive holding of the lock, which is now a spinlock, the
temac_device_reset() function is changed to only hold the lock for short
periods.  With timeouts, it could be holding the spinlock for more than
2 seconds.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ll_temac: Do not make promiscuous mode sticky on multicast
Esben Haabendal [Thu, 23 May 2019 12:02:19 +0000 (14:02 +0200)]
net: ll_temac: Do not make promiscuous mode sticky on multicast

When user has requested IFF_ALLMULTI or have set more than 4 multicast
addresses, we should just use promiscuous mode, but not set it in flags,
as it causes the interface to stay in promiscuous mode even when the
non-IFF_PROMISC condition that caused promiscuous mode to be enabled
has gone away.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: lxt: Add suspend/resume support to LXT971 and LXT973.
Christophe Leroy [Thu, 23 May 2019 08:55:32 +0000 (08:55 +0000)]
net: phy: lxt: Add suspend/resume support to LXT971 and LXT973.

All LXT PHYs implement the standard "power down" bit 11 of
BMCR, so this patch adds support using the generic
genphy_{suspend,resume} functions added by
commit 0f0ca340e57b ("phy: power management support").

LXT970 is left aside because all registers get cleared upon
"power down" exit.

Reviewed-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodevlink: add warning in case driver does not set port type
Jiri Pirko [Thu, 23 May 2019 08:43:35 +0000 (10:43 +0200)]
devlink: add warning in case driver does not set port type

Prevent misbehavior of drivers who would not set port type for longer
period of time. Drivers should always set port type. Do WARN if that
happens.

Note that it is perfectly fine to temporarily not have the type set,
during initialization and port type change.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohv_sock: perf: loop in send() to maximize bandwidth
Sunil Muthuswamy [Wed, 22 May 2019 23:10:44 +0000 (23:10 +0000)]
hv_sock: perf: loop in send() to maximize bandwidth

Currently, the hv_sock send() iterates once over the buffer, puts data into
the VMBUS channel and returns. It doesn't maximize on the case when there
is a simultaneous reader draining data from the channel. In such a case,
the send() can maximize the bandwidth (and consequently minimize the cpu
cycles) by iterating until the channel is found to be full.

Perf data:
Total Data Transfer: 10GB/iteration
Single threaded reader/writer, Linux hvsocket writer with Windows hvsocket
reader
Packet size: 64KB
CPU sys time was captured using the 'time' command for the writer to send
10GB of data.
'Send Buffer Loop' is with the patch applied.
The values below are over 10 iterations.

|--------------------------------------------------------|
|        |        Current        |   Send Buffer Loop    |
|--------------------------------------------------------|
|        | Throughput | CPU sys  | Throughput | CPU sys  |
|        | (MB/s)     | time (s) | (MB/s)     | time (s) |
|--------------------------------------------------------|
| Min    |     407    |   7.048  |    401     |  5.958   |
|--------------------------------------------------------|
| Max    |     455    |   7.563  |    542     |  6.993   |
|--------------------------------------------------------|
| Avg    |     440    |   7.411  |    451     |  6.639   |
|--------------------------------------------------------|
| Median |     446    |   7.417  |    447     |  6.761   |
|--------------------------------------------------------|

Observation:
1. The avg throughput doesn't really change much with this change for this
scenario. This is most probably because the bottleneck on throughput is
somewhere else.
2. The average system (or kernel) cpu time goes down by 10%+ with this
change, for the same amount of data transfer.

Signed-off-by: Sunil Muthuswamy <sunilmut@microsoft.com>
Reviewed-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohv_sock: perf: Allow the socket buffer size options to influence the actual socket...
Sunil Muthuswamy [Wed, 22 May 2019 22:56:07 +0000 (22:56 +0000)]
hv_sock: perf: Allow the socket buffer size options to influence the actual socket buffers

Currently, the hv_sock buffer size is static and can't scale to the
bandwidth requirements of the application. This change allows the
applications to influence the socket buffer sizes using the SO_SNDBUF and
the SO_RCVBUF socket options.

Few interesting points to note:
1. Since the VMBUS does not allow a resize operation of the ring size, the
socket buffer size option should be set prior to establishing the
connection for it to take effect.
2. Setting the socket option comes with the cost of that much memory being
reserved/allocated by the kernel, for the lifetime of the connection.

Perf data:
Total Data Transfer: 1GB
Single threaded reader/writer
Results below are summarized over 10 iterations.

Linux hvsocket writer + Windows hvsocket reader:
|---------------------------------------------------------------------------------------------|
|Packet size ->   |      128B       |       1KB       |       4KB       |        64KB         |
|---------------------------------------------------------------------------------------------|
|SO_SNDBUF size | |                 Throughput in MB/s (min/max/avg/median):                  |
|               v |                                                                           |
|---------------------------------------------------------------------------------------------|
|      Default    | 109/118/114/116 | 636/774/701/700 | 435/507/480/476 |   410/491/462/470   |
|      16KB       | 110/116/112/111 | 575/705/662/671 | 749/900/854/869 |   592/824/692/676   |
|      32KB       | 108/120/115/115 | 703/823/767/772 | 718/878/850/866 | 1593/2124/2000/2085 |
|      64KB       | 108/119/114/114 | 592/732/683/688 | 805/934/903/911 | 1784/1943/1862/1843 |
|---------------------------------------------------------------------------------------------|

Windows hvsocket writer + Linux hvsocket reader:
|---------------------------------------------------------------------------------------------|
|Packet size ->   |     128B    |      1KB        |          4KB        |        64KB         |
|---------------------------------------------------------------------------------------------|
|SO_RCVBUF size | |               Throughput in MB/s (min/max/avg/median):                    |
|               v |                                                                           |
|---------------------------------------------------------------------------------------------|
|      Default    | 69/82/75/73 | 313/343/333/336 |   418/477/446/445   |   659/701/676/678   |
|      16KB       | 69/83/76/77 | 350/401/375/382 |   506/548/517/516   |   602/624/615/615   |
|      32KB       | 62/83/73/73 | 471/529/496/494 |   830/1046/935/939  | 944/1180/1070/1100  |
|      64KB       | 64/70/68/69 | 467/533/501/497 | 1260/1590/1430/1431 | 1605/1819/1670/1660 |
|---------------------------------------------------------------------------------------------|

Signed-off-by: Sunil Muthuswamy <sunilmut@microsoft.com>
Reviewed-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4/igmp: shrink struct ip_sf_list
Eric Dumazet [Wed, 22 May 2019 22:00:25 +0000 (15:00 -0700)]
ipv4/igmp: shrink struct ip_sf_list

Removing two 4 bytes holes allows to use kmalloc-32
kmem cache instead of kmalloc-64 on 64bit kernels.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoneighbor: Add tracepoint to __neigh_create
David Ahern [Wed, 22 May 2019 19:22:21 +0000 (12:22 -0700)]
neighbor: Add tracepoint to __neigh_create

Add tracepoint to __neigh_create to enable debugging of new entries.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: pmtu: Simplify cleanup and namespace names
David Ahern [Wed, 22 May 2019 19:11:06 +0000 (12:11 -0700)]
selftests: pmtu: Simplify cleanup and namespace names

The point of the pause-on-fail argument is to leave the setup as is after
a test fails to allow a user to debug why it failed. Move the cleanup
after posting the result to the user to make it so.

Random names for the namespaces are not user friendly when trying to
debug a failure. Make them simpler and more direct for the tests. Run
cleanup at the beginning to ensure they are cleaned up if they already
exist.

Remove cleanup_done. There is no harm in doing cleanup twice; just
ignore any errors related to not existing - which is already done.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: fib-onlink: Make quiet by default
David Ahern [Wed, 22 May 2019 19:09:16 +0000 (12:09 -0700)]
selftests: fib-onlink: Make quiet by default

Add VERBOSE argument to fib-onlink-tests.sh and make output quiet by
default. Add getopt parsing of inputs and support for -v (verbose) and
-p (pause on fail).

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: Set strict_start_type for routes and rules
David Ahern [Wed, 22 May 2019 19:07:43 +0000 (12:07 -0700)]
net: Set strict_start_type for routes and rules

New userspace on an older kernel can send unknown and unsupported
attributes resulting in an incompelete config which is almost
always wrong for routing (few exceptions are passthrough settings
like the protocol that installed the route).

Set strict_start_type in the policies for IPv4 and IPv6 routes and
rules to detect new, unsupported attributes and fail the route add.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>