GitHub/moto-9609/android_kernel_motorola_exynos9610.git
7 years agoMerge branch 'liqudio-VF-data-path'
David S. Miller [Thu, 8 Dec 2016 18:21:40 +0000 (13:21 -0500)]
Merge branch 'liqudio-VF-data-path'

Raghu Vatsavayi says:

====================
liquidio VF data path

Following is V3 patch series that adds support for VF
data path related features. It also has following changes
related to previous comments:
1) Remove unnecessary "void *" casting.
2) Remove inline for functions and let gcc decide.

Please apply patches in following order as some of them
depend on earlier patches.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio VF rx data and ctl path
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:37 +0000 (08:54 -0800)]
liquidio VF rx data and ctl path

Adds support for VF receive data control path.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF TX buffers
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:36 +0000 (08:54 -0800)]
liquidio CN23XX: VF TX buffers

Adds support for freeing VF xmit buffers.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF xmit
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:35 +0000 (08:54 -0800)]
liquidio CN23XX: VF xmit

Adds support for transmit functionality in VF.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF scatter gather lists
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:34 +0000 (08:54 -0800)]
liquidio CN23XX: VF scatter gather lists

Adds support for VF scatter gather lists.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF mac address
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:33 +0000 (08:54 -0800)]
liquidio CN23XX: VF mac address

Adds support for configuring mtu, multicast and mac address.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF link status
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:32 +0000 (08:54 -0800)]
liquidio CN23XX: VF link status

Adds support for VF link status related changes.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio CN23XX: VF offload features
Raghu Vatsavayi [Wed, 7 Dec 2016 16:54:31 +0000 (08:54 -0800)]
liquidio CN23XX: VF offload features

Adds support for VF link initialization and offload features.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: rfs: add a jump label
Eric Dumazet [Wed, 7 Dec 2016 16:29:10 +0000 (08:29 -0800)]
net: rfs: add a jump label

RFS is not commonly used, so add a jump label to avoid some conditionals
in fast path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomacsec: remove first zero and add attribute name in comments
Zhang Shengju [Wed, 7 Dec 2016 15:02:09 +0000 (23:02 +0800)]
macsec: remove first zero and add attribute name in comments

Remove first zero for add, and use full attribute name in comments.

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'stmmac-DMA-burst'
David S. Miller [Thu, 8 Dec 2016 18:07:11 +0000 (13:07 -0500)]
Merge branch 'stmmac-DMA-burst'

Niklas Cassel says:

====================
net: stmmac: make DMA programmable burst length more configurable

Make DMA programmable burst length more configurable in the stmmac driver.

This is done by adding support for independent pbl for tx/rx through DT.
More fine grained tuning of pbl is possible thanks to a DT property saying
that we should NOT multiply pbl values by x8/x4 in hardware.

All new DT properties are optional, and created in a way that it will not
affect any existing DT configurations.

Changes since V1:
Created cover-letter.
Rebased patch set against next-20161205, since conflicting patches to
stmmac_platform.c has been merged since V1.

Changes since V2:
Moved default value initialization of pbl to stmmac_platform.c
and added a check for pbl != 0 in stmmac_main.c,
to catch a possble pbl == 0 from pci glue.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: smmac: allow configuring lower pbl values
Niklas Cassel [Wed, 7 Dec 2016 14:20:08 +0000 (15:20 +0100)]
net: smmac: allow configuring lower pbl values

The driver currently always sets the PBLx8/PBLx4 bit, which means that
the pbl values configured via the pbl/txpbl/rxpbl DT properties are
always multiplied by 8/4 in the hardware.

In order to allow the DT to configure lower pbl values, while at the
same time not changing behavior of any existing device trees using the
pbl/txpbl/rxpbl settings, add a property to disable the multiplication
of the pbl by 8/4 in the hardware.

Suggested-by: Rabin Vincent <rabinv@axis.com>
Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: add support for independent DMA pbl for tx/rx
Niklas Cassel [Wed, 7 Dec 2016 14:20:07 +0000 (15:20 +0100)]
net: stmmac: add support for independent DMA pbl for tx/rx

GMAC and newer supports independent programmable burst lengths for
DMA tx/rx. Add new optional devicetree properties representing this.

To be backwards compatible, snps,pbl will still be valid, but
snps,txpbl/snps,rxpbl will override the value in snps,pbl if set.

If the IP is synthesized to use the AXI interface, there is a register
and a matching DT property inside the optional stmmac-axi-config DT node
for controlling burst lengths, named snps,blen.
However, using this register, it is not possible to control tx and rx
independently. Also, this register is not available if the IP was
synthesized with, e.g., the AHB interface.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: dwmac1000: fix define DMA_BUS_MODE_RPBL_MASK
Niklas Cassel [Wed, 7 Dec 2016 14:20:06 +0000 (15:20 +0100)]
net: stmmac: dwmac1000: fix define DMA_BUS_MODE_RPBL_MASK

DMA_BUS_MODE_RPBL_MASK is really 6 bits,
just like DMA_BUS_MODE_PBL_MASK.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: stmmac_platform: fix parsing of DT binding
Niklas Cassel [Wed, 7 Dec 2016 14:20:05 +0000 (15:20 +0100)]
net: stmmac: stmmac_platform: fix parsing of DT binding

commit 64c3b252e9fc ("net: stmmac: fixed the pbl setting with DT")
changed the parsing of the DT binding.

Before 64c3b252e9fc, snps,fixed-burst and snps,mixed-burst were parsed
regardless if the property snps,pbl existed or not.
After the commit, fixed burst and mixed burst are only parsed if
snps,pbl exists. Now when snps,aal has been added, it too is only
parsed if snps,pbl exists.

Since the DT binding does not specify that fixed burst, mixed burst
or aal depend on snps,pbl being specified, undo changes introduced
by 64c3b252e9fc.

The issue commit 64c3b252e9fc ("net: stmmac: fixed the pbl setting with
DT") tries to address is solved in another way:
The databook specifies that all values other than
1, 2, 4, 8, 16, or 32 results in undefined behavior,
so snps,pbl = <0> is invalid.

If pbl is 0 after parsing, set pbl to DEFAULT_DMA_PBL.
This handles the case where the property is omitted, and also handles
the case where the property is specified without any data.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: simplify the common DMA init API
Niklas Cassel [Wed, 7 Dec 2016 14:20:04 +0000 (15:20 +0100)]
net: stmmac: simplify the common DMA init API

Use struct stmmac_dma_cfg *dma_cfg as an argument rather
than using all the struct members as individual arguments.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: return error if no DMA configuration is found
Niklas Cassel [Wed, 7 Dec 2016 14:20:03 +0000 (15:20 +0100)]
net: stmmac: return error if no DMA configuration is found

All drivers except pci glue layer calls stmmac_probe_config_dt.
stmmac_probe_config_dt does a kzalloc dma_cfg.

pci glue layer does kzalloc dma_cfg explicitly, so all current
drivers does a kzalloc dma_cfg.

Return an error if no DMA configuration is found, that way
we can assume that the DMA configuration always exists.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: slicoss: use module_pci_driver()
Tobias Klauser [Wed, 7 Dec 2016 13:43:30 +0000 (14:43 +0100)]
net: ethernet: slicoss: use module_pci_driver()

Use module_pci_driver() to get rid of some boilerplate code.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Acked-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'cls_flower-ICMP'
David S. Miller [Thu, 8 Dec 2016 16:47:14 +0000 (11:47 -0500)]
Merge branch 'cls_flower-ICMP'

Simon Horman says:

====================
net/sched: cls_flower: Support matching on ICMP

this series adds support for matching on ICMP type and code to cls_flower.

Changes v5->v6:
* Restore missing signed-off-by

Changes v4->v5:
* Drop all helpers

Changes v3->v4:
* Do not add icmp to struct flow_keys, it is not needed
* Do not test for ICMP protocols in packet in __skb_flow_dissect,
  this is also not needed

Changes v2->v3:
* Add FLOW_DISSECTOR_KEY_ICMP and use separate structure for ICMP

Changes v1->v2:
* Include all dissector helpers in first patch
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_flower: Support matching on ICMP type and code
Simon Horman [Wed, 7 Dec 2016 12:48:28 +0000 (13:48 +0100)]
net/sched: cls_flower: Support matching on ICMP type and code

Support matching on ICMP type and code.

Example usage:

tc qdisc add dev eth0 ingress

tc filter add dev eth0 protocol ip parent ffff: flower \
indev eth0 ip_proto icmp type 8 code 0 action drop

tc filter add dev eth0 protocol ipv6 parent ffff: flower \
indev eth0 ip_proto icmpv6 type 128 code 0 action drop

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoflow dissector: ICMP support
Simon Horman [Wed, 7 Dec 2016 12:48:27 +0000 (13:48 +0100)]
flow dissector: ICMP support

Allow dissection of ICMP(V6) type and code. This should only occur
if a packet is ICMP(V6) and the dissector has FLOW_DISSECTOR_KEY_ICMP set.

There are currently no users of FLOW_DISSECTOR_KEY_ICMP.
A follow-up patch will allow FLOW_DISSECTOR_KEY_ICMP to be used by
the flower classifier.

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: stmmac_platform: use correct setup function for gmac4
Niklas Cassel [Wed, 7 Dec 2016 12:41:08 +0000 (13:41 +0100)]
net: stmmac: stmmac_platform: use correct setup function for gmac4

devicetree binding for stmmac states:
- compatible: Should be "snps,dwmac-<ip_version>", "snps,dwmac"
For backwards compatibility: "st,spear600-gmac" is also supported.

Previously, when specifying "snps,dwmac-4.10a", "snps,dwmac" as your
compatible string, plat_stmmacenet_data would have both has_gmac and
has_gmac4 set.

This would lead to stmmac_hw_init calling dwmac1000_setup rather than
dwmac4_setup, resulting in a non-functional driver.
This happened since the check for has_gmac is done before the check for
has_gmac4. However, the order should not matter, so it does not make sense
to have both set.

If something is valid for both, you should do as the stmmac_interrupt does:
if (priv->plat->has_gmac || priv->plat->has_gmac4) ...

The places where it was obvious that the author actually meant
if (has_gmac || has_gmac4) rather than if (has_gmac) has been updated.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre TORGUE <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: dwmac-generic: add missing compatible strings
Niklas Cassel [Wed, 7 Dec 2016 12:41:07 +0000 (13:41 +0100)]
net: stmmac: dwmac-generic: add missing compatible strings

devicetree binding for stmmac states:
- compatible: Should be "snps,dwmac-<ip_version>", "snps,dwmac"
For backwards compatibility: "st,spear600-gmac" is also supported.

Since dwmac-generic.c calls stmmac_probe_config_dt explicitly,
another alternative would have been to remove all compatible strings
other than "snps,dwmac" and "st,spear600-gmac" from dwmac-generic.c.

However, that would probably do more good than harm, since when trying
to figure out what hardware a certain driver supports, you usually look
at the compatible strings in the struct of_device_id, and not in some
function defined in a completely different file.

No functional change intended.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobindings: net: stmmac: correct note about TSO
Niklas Cassel [Wed, 7 Dec 2016 12:41:06 +0000 (13:41 +0100)]
bindings: net: stmmac: correct note about TSO

snps,tso was previously placed under AXI BUS Mode parameters,
suggesting that the property should be in the stmmac-axi-config node.

TSO (TCP Segmentation Offloading) has nothing to do with AXI BUS Mode
parameters, and the parser actually expects it to be in the root node,
not in the stmmac-axi-config.

Also added a note about snps,tso only being available on GMAC4 and newer.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Alexandre TORGUE <alexandre.torgue@st.com>
Acked-by: Rob Herring <robh@kernel.org>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ll_temac: Utilize of_get_mac_address()
Tobias Klauser [Wed, 7 Dec 2016 12:25:29 +0000 (13:25 +0100)]
net: ll_temac: Utilize of_get_mac_address()

Do not open code getting the MAC address exclusively from the
"local-mac-address" property, but instead use of_get_mac_address()
which looks up the MAC address using the 3 typical property names.

Also avoid casting away the const qualifier of the return value by
making temac_init_mac_address() take a const void* address.

Follows commit b34296a9c047 ("net: ethoc: Utilize
of_get_mac_address()").

Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: axienet: Utilize of_get_mac_address()
Tobias Klauser [Wed, 7 Dec 2016 12:24:44 +0000 (13:24 +0100)]
net: axienet: Utilize of_get_mac_address()

Do not open code getting the MAC address exclusively from the
"local-mac-address" property, but instead use of_get_mac_address()
which looks up the MAC address using the 3 typical property names.

Also avoid casting away the const qualifier of the return value by
making axienet_set_mac_address() take a const void* address.

Follows commit b34296a9c047 ("net: ethoc: Utilize
of_get_mac_address()").

Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'cls_flower-flags'
David S. Miller [Thu, 8 Dec 2016 16:32:50 +0000 (11:32 -0500)]
Merge branch 'cls_flower-flags'

Or Gerlitz says:

====================
net/sched: cls_flower: Add support for matching on dissection flags

This series add the UAPI to provide set of flags for matching, where the
flags provided from user-space are mapped to flow-dissector flags.

The 1st flag allows to match on whether the packet is an
IP fragment and corresponds to the FLOW_DIS_IS_FRAGMENT flag.

v2->v3:
  - replace BIT() with << (kbuild test robot)

v1->v2:
 - dropped the flow dissector patch (#1) as no changes are needed there (Jiri)
 - applied code review comments from Jiri to the flower patch
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx5e: Offload TC matching on packets being IP fragments
Or Gerlitz [Wed, 7 Dec 2016 12:03:11 +0000 (14:03 +0200)]
net/mlx5e: Offload TC matching on packets being IP fragments

Enable offloading of matching on packets being fragments.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_flower: Add support for matching on flags
Or Gerlitz [Wed, 7 Dec 2016 12:03:10 +0000 (14:03 +0200)]
net/sched: cls_flower: Add support for matching on flags

Add UAPI to provide set of flags for matching, where the flags
provided from user-space are mapped to flow-dissector flags.

The 1st flag allows to match on whether the packet is an
IP fragment and corresponds to the FLOW_DIS_IS_FRAGMENT flag.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvneta: Indent some statements
Dan Carpenter [Wed, 7 Dec 2016 11:32:17 +0000 (14:32 +0300)]
net: mvneta: Indent some statements

These two statements were not indented correctly so it's sort of
confusing.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: uninitialized variable in xgene_enet_free_pagepool()
Dan Carpenter [Wed, 7 Dec 2016 11:14:24 +0000 (14:14 +0300)]
drivers: net: xgene: uninitialized variable in xgene_enet_free_pagepool()

We never set "slots" in this function.

Fixes: a9380b0f7be8 ("drivers: net: xgene: Add support for Jumbo frame")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovhost: remove unnecessary smp_mb from vhost_work_queue
Peng Tao [Wed, 7 Dec 2016 09:52:19 +0000 (17:52 +0800)]
vhost: remove unnecessary smp_mb from vhost_work_queue

test_and_set_bit() already implies a memory barrier.

Signed-off-by: Peng Tao <bergwolf@gmail.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovhost-vsock: remove unused vq variable
Peng Tao [Wed, 7 Dec 2016 09:52:18 +0000 (17:52 +0800)]
vhost-vsock: remove unused vq variable

Signed-off-by: Peng Tao <bergwolf@gmail.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoicmp: correct return value of icmp_rcv()
Zhang Shengju [Wed, 7 Dec 2016 06:52:53 +0000 (14:52 +0800)]
icmp: correct return value of icmp_rcv()

Currently, icmp_rcv() always return zero on a packet delivery upcall.

To make its behavior more compliant with the way this API should be
used, this patch changes this to let it return NET_RX_SUCCESS when the
packet is proper handled, and NET_RX_DROP otherwise.

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Thu, 8 Dec 2016 00:16:46 +0000 (19:16 -0500)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter/IPVS updates for net-next

The following patchset contains a large Netfilter update for net-next,
to summarise:

1) Add support for stateful objects. This series provides a nf_tables
   native alternative to the extended accounting infrastructure for
   nf_tables. Two initial stateful objects are supported: counters and
   quotas. Objects are identified by a user-defined name, you can fetch
   and reset them anytime. You can also use a maps to allow fast lookups
   using any arbitrary key combination. More info at:

   http://marc.info/?l=netfilter-devel&m=148029128323837&w=2

2) On-demand registration of nf_conntrack and defrag hooks per netns.
   Register nf_conntrack hooks if we have a stateful ruleset, ie.
   state-based filtering or NAT. The new nf_conntrack_default_on sysctl
   enables this from newly created netnamespaces. Default behaviour is not
   modified. Patches from Florian Westphal.

3) Allocate 4k chunks and then use these for x_tables counter allocation
   requests, this improves ruleset load time and also datapath ruleset
   evaluation, patches from Florian Westphal.

4) Add support for ebpf to the existing x_tables bpf extension.
   From Willem de Bruijn.

5) Update layer 4 checksum if any of the pseudoheader fields is updated.
   This provides a limited form of 1:1 stateless NAT that make sense in
   specific scenario, eg. load balancing.

6) Add support to flush sets in nf_tables. This series comes with a new
   set->ops->deactivate_one() indirection given that we have to walk
   over the list of set elements, then deactivate them one by one.
   The existing set->ops->deactivate() performs an element lookup that
   we don't need.

7) Two patches to avoid cloning packets, thus speed up packet forwarding
   via nft_fwd from ingress. From Florian Westphal.

8) Two IPVS patches via Simon Horman: Decrement ttl in all modes to
   prevent infinite loops, patch from Dwip Banerjee. And one minor
   refactoring from Gao feng.

9) Revisit recent log support for nf_tables netdev families: One patch
   to ensure that we correctly handle non-ethernet packets. Another
   patch to add missing logger definition for netdev. Patches from
   Liping Zhang.

10) Three patches for nft_fib, one to address insufficient register
    initialization and another to solve incorrect (although harmless)
    byteswap operation. Moreover update xt_rpfilter and nft_fib to match
    lbcast packets with zeronet as source, eg. DHCP Discover packets
    (0.0.0.0 -> 255.255.255.255). Also from Liping Zhang.

11) Built-in DCCP, SCTP and UDPlite conntrack and NAT support, from
    Davide Caratti. While DCCP is rather hopeless lately, and UDPlite has
    been broken in many-cast mode for some little time, let's give them a
    chance by placing them at the same level as other existing protocols.
    Thus, users don't explicitly have to modprobe support for this and
    NAT rules work for them. Some people point to the lack of support in
    SOHO Linux-based routers that make deployment of new protocols harder.
    I guess other middleboxes outthere on the Internet are also to blame.
    Anyway, let's see if this has any impact in the midrun.

12) Skip software SCTP software checksum calculation if the NIC comes
    with SCTP checksum offload support. From Davide Caratti.

13) Initial core factoring to prepare conversion to hook array. Three
    patches from Aaron Conole.

14) Gao Feng made a wrong conversion to switch in the xt_multiport
    extension in a patch coming in the previous batch. Fix it in this
    batch.

15) Get vmalloc call in sync with kmalloc flags to avoid a warning
    and likely OOM killer intervention from x_tables. From Marcelo
    Ricardo Leitner.

16) Update Arturo Borrero's email address in all source code headers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Thu, 8 Dec 2016 00:03:23 +0000 (19:03 -0500)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2016-12-07

This series contains updates to i40e and i40evf only.

Filip modifies the i40e to log link speed change and when the link is
brought up and down.

Mitch replaces i40e_txd_use_count() with a new function which is slightly
faster and better documented so the dim witted can better follow the
code.  Fixes the locking of the service task so that it is actually
done in the service task and not in the scheduling function which calls
the service task.

Jacob, being the busy little beaver he is, provides most of the changes
starting restores a workaround that is still needed in some configurations,
specifically the Ethernet Controller XL710 for 40GbE QSFP+.  Removes
duplicate code and simplifies the i40e_vsi_add_vlan() and
i40e_vsi_kill_vlan() functions.  Removes detection of PTP frames over L4
(UDP) on the XL710 MAC, since there was a product decision to defeature
it.  Fixed a previous refactor of active filters which caused issues in
the accounting of active_filters.  Remaining work was done in the VLAN
filters to improve readability and simplify code as much as possible
to reduce inconsistencies.

Alex fixes foul budget accounting in core code by returning actual
work done, capped to budget-1.

Henry fixes the "ethtool -p" function for 1G BaseT PHYs.

Carolyn adds support for 25G devices for i40e and i40evf.

Michal adds functions to apply the correct access method for external PHYs
which could use Clause22 or Clause45 depending on the PHY.

v2: dropped last patch from previous series, since changes are needed based
    on feedback from Sergei Shtylyov
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodummy: expend mtu range for dummy device
Zhang Shengju [Wed, 7 Dec 2016 09:41:33 +0000 (17:41 +0800)]
dummy: expend mtu range for dummy device

After commit 61e84623ace3 ("net: centralize net_device min/max MTU checking"),
the mtu range for dummy device becomes [68, 1500].

This patch extends it to [0, 65535].

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonlmon: use core MTU range checking in nlmon driver
Zhang Shengju [Wed, 7 Dec 2016 09:26:05 +0000 (17:26 +0800)]
nlmon: use core MTU range checking in nlmon driver

Since commit 61e84623ace3 ("net: centralize net_device min/max MTU checking"),
mtu range is checked at dev_set_mtu().

This patch adds min_mtu for nlmon device and remove unnecessary
ndo_change_mtu() function.

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodriver: macvlan: Remove the rcu member of macvlan_port
Gao Feng [Wed, 7 Dec 2016 04:23:18 +0000 (12:23 +0800)]
driver: macvlan: Remove the rcu member of macvlan_port

When free macvlan_port in macvlan_port_destroy, it is safe to free
directly because netdev_rx_handler_unregister could enforce one
grace period.
So it is unnecessary to use kfree_rcu for macvlan_port.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodriver: ipvlan: Free ipvl_port directly with kfree instead of kfree_rcu
Gao Feng [Wed, 7 Dec 2016 00:44:47 +0000 (08:44 +0800)]
driver: ipvlan: Free ipvl_port directly with kfree instead of kfree_rcu

There are two functions which would free the ipvl_port now. The first
is ipvlan_port_create. It frees the ipvl_port in the error handler,
so it could kfree it directly. The second is ipvlan_port_destroy. It
invokes netdev_rx_handler_unregister which enforces one grace period
by synchronize_net firstly, so it also could kfree the ipvl_port
directly and safely.

So it is unnecessary to use kfree_rcu to free ipvl_port.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: fix loading of BPF_MAXINSNS sized programs
Daniel Borkmann [Wed, 7 Dec 2016 00:15:44 +0000 (01:15 +0100)]
bpf: fix loading of BPF_MAXINSNS sized programs

General assumption is that single program can hold up to BPF_MAXINSNS,
that is, 4096 number of instructions. It is the case with cBPF and
that limit was carried over to eBPF. When recently testing digest, I
noticed that it's actually not possible to feed 4096 instructions
via bpf(2).

The check for > BPF_MAXINSNS was added back then to bpf_check() in
cbd357008604 ("bpf: verifier (add ability to receive verification log)").
However, 09756af46893 ("bpf: expand BPF syscall with program load/unload")
added yet another check that comes before that into bpf_prog_load(),
but this time bails out already in case of >= BPF_MAXINSNS.

Fix it up and perform the check early in bpf_prog_load(), so we can drop
the second one in bpf_check(). It makes sense, because also a 0 insn
program is useless and we don't want to waste any resources doing work
up to bpf_check() point. The existing bpf(2) man page documents E2BIG
as the official error for such cases, so just stick with it as well.

Fixes: 09756af46893 ("bpf: expand BPF syscall with program load/unload")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: do not call phy_ethtool_ksettings_set from atomic context
Niklas Cassel [Tue, 6 Dec 2016 13:47:15 +0000 (14:47 +0100)]
net: stmmac: do not call phy_ethtool_ksettings_set from atomic context

>From what I can tell, spin_lock(&priv->lock) is not needed, since the
phy_ethtool_ksettings_set call is not given the priv struct.

phy_start_aneg takes the phydev->lock. Calls to phy_adjust_link
from phy_state_machine also takes the phydev->lock.

[   13.718319] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
[   13.726717] in_atomic(): 1, irqs_disabled(): 0, pid: 1307, name: ethtool
[   13.742115] Hardware name: Axis ARTPEC-6 Platform
[   13.746829] [<80110568>] (unwind_backtrace) from [<8010c2bc>] (show_stack+0x18/0x1c)
[   13.754575] [<8010c2bc>] (show_stack) from [<80433484>] (dump_stack+0x80/0xa0)
[   13.761801] [<80433484>] (dump_stack) from [<80145428>] (___might_sleep+0x108/0x170)
[   13.769554] [<80145428>] (___might_sleep) from [<806c9b50>] (mutex_lock+0x24/0x44)
[   13.777128] [<806c9b50>] (mutex_lock) from [<8050cbc0>] (phy_start_aneg+0x1c/0x13c)
[   13.784783] [<8050cbc0>] (phy_start_aneg) from [<8050d338>] (phy_ethtool_ksettings_set+0x98/0xd0)
[   13.793656] [<8050d338>] (phy_ethtool_ksettings_set) from [<80517adc>] (stmmac_ethtool_set_link_ksettings+0xa0/0xb4)
[   13.804184] [<80517adc>] (stmmac_ethtool_set_link_ksettings) from [<805c5138>] (ethtool_set_settings+0xd4/0x13c)
[   13.814358] [<805c5138>] (ethtool_set_settings) from [<805c9718>] (dev_ethtool+0x13c4/0x211c)
[   13.822882] [<805c9718>] (dev_ethtool) from [<805dc7c0>] (dev_ioctl+0x480/0x8e0)
[   13.830291] [<805dc7c0>] (dev_ioctl) from [<80260e34>] (do_vfs_ioctl+0x94/0xa00)
[   13.837699] [<80260e34>] (do_vfs_ioctl) from [<802617dc>] (SyS_ioctl+0x3c/0x60)
[   13.845011] [<802617dc>] (SyS_ioctl) from [<801088bc>] (__sys_trace_return+0x0/0x10)

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'ti-cpts-update-and-fixes'
David S. Miller [Wed, 7 Dec 2016 16:13:49 +0000 (11:13 -0500)]
Merge branch 'ti-cpts-update-and-fixes'

Grygorii Strashko says:

====================
net: ethernet: ti: cpts: update and fixes

It is preparation series intended to clean up and optimize TI CPTS driver to
facilitate further integration with other TI's SoCs like Keystone 2.

Changes in v5:
- fixed copy paste error in cpts_release
- reworked cc.mult/shift and cc_mult initialization

Changes in v4:
- fixed build error in patch
  "net: ethernet: ti: cpts: clean up event list if event pool is empty"
- rebased on top of net-next

Changes in v3:
- patches reordered: fixes and small updates moved first
- added comments in code about cpts->cc_mult
- conversation range (maxsec) limited to 10sec

Changes in v2:
- patch "net: ethernet: ti: cpts: rework initialization/deinitialization"
  was split on 4 patches
- applied comments from Richard Cochran
- dropped patch
  "net: ethernet: ti: cpts: add return value to tx and rx timestamp funcitons"
- new patches added:
  "net: ethernet: ti: cpts: drop excessive writes to CTRL and INT_EN regs"
  and "clocksource: export the clocks_calc_mult_shift to use by timestamp code"

Links on prev versions:
v4: https://lkml.org/lkml/2016/12/6/496
v3: https://www.spinics.net/lists/devicetree/msg153474.html
v2: http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1282034.html
v1: http://www.spinics.net/lists/linux-omap/msg131925.html
====================

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: fix overflow check period
Grygorii Strashko [Wed, 7 Dec 2016 00:00:45 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: fix overflow check period

The CPTS drivers uses 8sec period for overflow checking with
assumption that CPTS retclk will not exceed 500MHz. But that's not
true on some TI platforms (Kesytone 2). As result, it is possible that
CPTS counter will overflow more than once between two readings.

Hence, fix it by selecting overflow check period dynamically as
max_sec_before_overflow/2, where
 max_sec_before_overflow = max_counter_val / rftclk_freq.

Cc: John Stultz <john.stultz@linaro.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: calc mult and shift from refclk freq
Grygorii Strashko [Wed, 7 Dec 2016 00:00:44 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: calc mult and shift from refclk freq

The cyclecounter mult and shift values can be calculated based on the
CPTS rfclk frequency and timekeepnig framework provides required algos
and API's.

Hence, calc mult and shift basing on CPTS rfclk frequency if both
cpts_clock_shift and cpts_clock_mult properties are not provided in DT (the
basis of calculation algorithm is borrowed from
__clocksource_update_freq_scale() commit 7d2f944a2b83 ("clocksource:
Provide a generic mult/shift factor calculation")). After this change
cpts_clock_shift and cpts_clock_mult DT properties will become optional.

Cc: John Stultz <john.stultz@linaro.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoclocksource: export the clocks_calc_mult_shift to use by timestamp code
Murali Karicheri [Wed, 7 Dec 2016 00:00:43 +0000 (18:00 -0600)]
clocksource: export the clocks_calc_mult_shift to use by timestamp code

The CPSW CPTS driver is capable of doing timestamping on tx/rx packets and
requires to know mult and shift factors for timestamp conversion from raw
value to nanoseconds (ptp clock). Now these mult and shift factors are
calculated manually and provided through DT, which makes very hard to
support of a lot number of platforms, especially if CPTS refclk is not the
same for some kind of boards and depends on efuse settings (Keystone 2
platforms). Hence, export clocks_calc_mult_shift() to allow drivers like
CPSW CPTS (and other ptp drivesr) to benefit from automaitc calculation of
mult and shift factors.

Cc: John Stultz <john.stultz@linaro.org>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: move dt props parsing to cpts driver
Grygorii Strashko [Wed, 7 Dec 2016 00:00:42 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: move dt props parsing to cpts driver

Move DT properties parsing into CPTS driver to simplify CPSW
code and CPTS driver porting on other SoC in the future
(like Keystone 2) - with this change it will not be required
to add the same DT parsing code in Keystone 2 NETCP driver.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: rework initialization/deinitialization
Grygorii Strashko [Wed, 7 Dec 2016 00:00:41 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: rework initialization/deinitialization

The current implementation CPTS initialization and deinitialization
(represented by cpts_register/unregister()) does too many static
initialization from .ndo_open(), which is reasonable to do once at probe
time instead, and also require caller to allocate memory for struct cpts,
which is internal for CPTS driver in general.

This patch splits CPTS initialization and deinitialization on two parts:

- static initializtion cpts_create()/cpts_release() which expected to be
executed when parent driver is probed/removed;

- dynamic part cpts_register/unregister() which expected to be executed
when network device is opened/closed.

As result, current code of CPTS parent driver - CPSW - will be simplified
(and it also will allow simplify adding support for Keystone 2 devices in
the future), plus more initialization errors will be catched earlier. In
addition, this change allows to clean up cpts.h for the case when CPTS is
disabled.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: drop excessive writes to CTRL and INT_EN regs
Grygorii Strashko [Wed, 7 Dec 2016 00:00:40 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: drop excessive writes to CTRL and INT_EN regs

CPTS module and IRQs are always enabled when CPTS is registered,
before starting overflow check work, and disabled during
deregistration, when overflow check work has been canceled already.
So, It doesn't require to (re)enable CPTS module and IRQs in
cpts_overflow_check().

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: clean up event list if event pool is empty
WingMan Kwok [Wed, 7 Dec 2016 00:00:39 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: clean up event list if event pool is empty

When a CPTS user does not exit gracefully by disabling cpts
timestamping and leaving a joined multicast group, the system
continues to receive and timestamps the ptp packets which eventually
occupy all the event list entries.  When this happns, the added code
tries to remove some list entries which are expired.

Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: disable cpts when unregistered
Grygorii Strashko [Wed, 7 Dec 2016 00:00:38 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: disable cpts when unregistered

The cpts now is left enabled after unregistration.
Hence, disable it in cpts_unregister().

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: fix registration order
Grygorii Strashko [Wed, 7 Dec 2016 00:00:37 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: fix registration order

The ptp clock registered before spinlock, which is protecting it, and
before timecounter and cyclecounter initialization in cpts_register().

So, ensure that ptp clock is registered the last, after everything
else is done.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: fix unbalanced clk api usage in cpts_register/unregister
Grygorii Strashko [Wed, 7 Dec 2016 00:00:36 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: fix unbalanced clk api usage in cpts_register/unregister

There are two issues with TI CPTS code which are reproducible when TI
CPSW ethX device passes few up/down iterations:
- cpts refclk prepare counter continuously incremented after each
up/down iteration;
- devm_clk_get(dev, "cpts") is called many times.

Hence, fix these issues by using clk_disable_unprepare() in
cpts_clk_release() and skipping devm_clk_get() if cpts refclk has been
acquired already.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpsw: minimize direct access to struct cpts
Grygorii Strashko [Wed, 7 Dec 2016 00:00:35 +0000 (18:00 -0600)]
net: ethernet: ti: cpsw: minimize direct access to struct cpts

This will provide more flexibility in changing CPTS internals and also
required for further changes.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: allow cpts to be built separately
Grygorii Strashko [Wed, 7 Dec 2016 00:00:34 +0000 (18:00 -0600)]
net: ethernet: ti: allow cpts to be built separately

TI CPTS IP is used as part of TI OMAP CPSW driver, but it's also
present as part of NETCP on TI Keystone 2 SoCs. So, It's required
to enable build of CPTS for both this drivers and this can be
achieved by allowing CPTS to be built separately.

Hence, allow cpts to be built separately and convert it to be
a module as both CPSW and NETCP drives can be built as modules.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpts: switch to readl/writel_relaxed()
Grygorii Strashko [Wed, 7 Dec 2016 00:00:33 +0000 (18:00 -0600)]
net: ethernet: ti: cpts: switch to readl/writel_relaxed()

Switch to readl/writel_relaxed() APIs, because this is recommended
API and the CPTS IP is reused on Keystone 2 SoCs
where LE/BE modes are supported.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bnxt_en-RDMA'
David S. Miller [Wed, 7 Dec 2016 15:59:27 +0000 (10:59 -0500)]
Merge branch 'bnxt_en-RDMA'

Michael Chan says:

====================
bnxt_en: Add interface to support RDMA driver.

This series adds an interface to support a brand new RDMA driver bnxt_re.
The first step is to re-arrange some code so that pci_enable_msix() can
be called during pci probe.  The purpose is to allow the RDMA driver to
initialize and stay initialized whether the netdev is up or down.

Then we make some changes to VF resource allocation so that there is
enough resources to support RDMA.

Finally the last patch adds a simple interface to allow the RDMA driver to
probe and register itself with any bnxt_en devices that support RDMA.
Once registered, the RDMA driver can request MSIX, send fw messages, and
receive some notifications.

v2: Fixed kbuild test robot warnings.

David, please consider this series for net-next.  Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add interface to support RDMA driver.
Michael Chan [Wed, 7 Dec 2016 05:26:21 +0000 (00:26 -0500)]
bnxt_en: Add interface to support RDMA driver.

Since the network driver and RDMA driver operate on the same PCI function,
we need to create an interface to allow the RDMA driver to share resources
with the network driver.

1. Create a new bnxt_en_dev struct which will be returned by
bnxt_ulp_probe() upon success.  After that, all calls from the RDMA driver
to bnxt_en will pass a pointer to this struct.

2. This struct contains additional function pointers to register, request
msix, send fw messages, register for async events.

3. If the RDMA driver wants to enable RDMA on the function, it needs to
call the function pointer bnxt_register_device().  A ulp_ops structure
is passed for RCU protected upcalls from bnxt_en to the RDMA driver.

4. The RDMA driver can call firmware APIs using the bnxt_send_fw_msg()
function pointer.

5. 1 stats context is reserved when the RDMA driver registers.  MSIX
and completion rings are reserved when the RDMA driver calls
bnxt_request_msix() function pointer.

6. When the RDMA driver calls bnxt_unregister_device(), all RDMA resources
will be cleaned up.

v2: Fixed 2 uninitialized variable warnings.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Refactor the driver registration function with firmware.
Michael Chan [Wed, 7 Dec 2016 05:26:20 +0000 (00:26 -0500)]
bnxt_en: Refactor the driver registration function with firmware.

The driver register function with firmware consists of passing version
information and registering for async events.  To support the RDMA driver,
the async events that we need to register may change.  Separate the
driver register function into 2 parts so that we can just update the
async events for the RDMA driver.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Reserve RDMA resources by default.
Michael Chan [Wed, 7 Dec 2016 05:26:19 +0000 (00:26 -0500)]
bnxt_en: Reserve RDMA resources by default.

If the device supports RDMA, we'll setup network default rings so that
there are enough minimum resources for RDMA, if possible.  However, the
user can still increase network rings to the max if he wants.  The actual
RDMA resources won't be reserved until the RDMA driver registers.

v2: Fix compile warning when BNXT_CONFIG_SRIOV is not set.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Improve completion ring allocation for VFs.
Michael Chan [Wed, 7 Dec 2016 05:26:18 +0000 (00:26 -0500)]
bnxt_en: Improve completion ring allocation for VFs.

All available remaining completion rings not used by the PF should be
made available for the VFs so that there are enough rings in the VF to
support RDMA.  The earlier workaround code of capping the rings by the
statistics context is removed.

When SRIOV is disabled, call a new function bnxt_restore_pf_fw_resources()
to restore FW resources.  Later on we need to add some logic to account
for RDMA resources.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Move function reset to bnxt_init_one().
Michael Chan [Wed, 7 Dec 2016 05:26:17 +0000 (00:26 -0500)]
bnxt_en: Move function reset to bnxt_init_one().

Now that MSIX is enabled in bnxt_init_one(), resources may be allocated by
the RDMA driver before the network device is opened.  So we cannot do
function reset in bnxt_open() which will clear all the resources.

The proper place to do function reset now is in bnxt_init_one().
If we get AER, we'll do function reset as well.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Enable MSIX early in bnxt_init_one().
Michael Chan [Wed, 7 Dec 2016 05:26:16 +0000 (00:26 -0500)]
bnxt_en: Enable MSIX early in bnxt_init_one().

To better support the new RDMA driver, we need to move pci_enable_msix()
from bnxt_open() to bnxt_init_one().  This way, MSIX vectors are available
to the RDMA driver whether the network device is up or down.

Part of the existing bnxt_setup_int_mode() function is now refactored into
a new bnxt_init_int_mode().  bnxt_init_int_mode() is called during
bnxt_init_one() to enable MSIX.  The remaining logic in
bnxt_setup_int_mode() to map the IRQs to the completion rings is called
during bnxt_open().

v2: Fixed compile warning when CONFIG_BNXT_SRIOV is not set.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add bnxt_set_max_func_irqs().
Michael Chan [Wed, 7 Dec 2016 05:26:15 +0000 (00:26 -0500)]
bnxt_en: Add bnxt_set_max_func_irqs().

By refactoring existing code into this new function.  The new function
will be used in subsequent patches.

v2: Fixed compile warning when CONFIG_BNXT_SRIOV is not set.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: sock_rps_record_flow() is for connected sockets
Eric Dumazet [Wed, 7 Dec 2016 03:32:50 +0000 (19:32 -0800)]
net: sock_rps_record_flow() is for connected sockets

Paolo noticed a cache line miss in UDP recvmsg() to access
sk_rxhash, sharing a cache line with sk_drops.

sk_drops might be heavily incremented by cpus handling a flood targeting
this socket.

We might place sk_drops on a separate cache line, but lets try
to avoid wasting 64 bytes per socket just for this, since we have
other bottlenecks to take care of.

sock_rps_record_flow() should only access sk_rxhash for connected
flows.

Testing sk_state for TCP_ESTABLISHED covers most of the cases for
connected sockets, for a zero cost, since system calls using
sock_rps_record_flow() also access sk->sk_prot which is on the
same cache line.

A follow up patch will provide a static_key (Jump Label) since most
hosts do not even use RFS.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetfilter: nft_quota: allow to restore consumed quota
Pablo Neira Ayuso [Wed, 7 Dec 2016 13:40:53 +0000 (14:40 +0100)]
netfilter: nft_quota: allow to restore consumed quota

Allow to restore consumed quota, this is useful to restore the quota
state across reboots.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: xt_bpf: support ebpf
Willem de Bruijn [Tue, 6 Dec 2016 21:25:02 +0000 (16:25 -0500)]
netfilter: xt_bpf: support ebpf

Add support for attaching an eBPF object by file descriptor.

The iptables binary can be called with a path to an elf object or a
pinned bpf object. Also pass the mode and path to the kernel to be
able to return it later for iptables dump and save.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: x_tables: avoid warn and OOM killer on vmalloc call
Marcelo Ricardo Leitner [Fri, 2 Dec 2016 09:46:38 +0000 (07:46 -0200)]
netfilter: x_tables: avoid warn and OOM killer on vmalloc call

Andrey Konovalov reported that this vmalloc call is based on an
userspace request and that it's spewing traces, which may flood the logs
and cause DoS if abused.

Florian Westphal also mentioned that this call should not trigger OOM
killer.

This patch brings the vmalloc call in sync to kmalloc and disables the
warn trace on allocation failure and also disable OOM killer invocation.

Note, however, that under such stress situation, other places may
trigger OOM killer invocation.

Reported-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: support for set flushing
Pablo Neira Ayuso [Mon, 5 Dec 2016 22:35:50 +0000 (23:35 +0100)]
netfilter: nf_tables: support for set flushing

This patch adds support for set flushing, that consists of walking over
the set elements if the NFTA_SET_ELEM_LIST_ELEMENTS attribute is set.
This patch requires the following changes:

1) Add set->ops->deactivate_one() operation: This allows us to
   deactivate an element from the set element walk path, given we can
   skip the lookup that happens in ->deactivate().

2) Add a new nft_trans_alloc_gfp() function since we need to allocate
   transactions using GFP_ATOMIC given the set walk path happens with
   held rcu_read_lock.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nft_set: introduce nft_{hash, rbtree}_deactivate_one()
Pablo Neira Ayuso [Mon, 5 Dec 2016 22:35:49 +0000 (23:35 +0100)]
netfilter: nft_set: introduce nft_{hash, rbtree}_deactivate_one()

This new function allows us to deactivate one single element, this is
required by the set flush command that comes in a follow up patch.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: constify struct nft_ctx * parameter in nft_trans_alloc()
Pablo Neira Ayuso [Mon, 5 Dec 2016 22:35:48 +0000 (23:35 +0100)]
netfilter: nf_tables: constify struct nft_ctx * parameter in nft_trans_alloc()

Context is not modified by nft_trans_alloc(), so constify it.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nat: skip checksum on offload SCTP packets
Davide Caratti [Mon, 5 Dec 2016 14:33:57 +0000 (15:33 +0100)]
netfilter: nat: skip checksum on offload SCTP packets

SCTP GSO and hardware can do CRC32c computation after netfilter processing,
so we can avoid calling sctp_compute_checksum() on skb if skb->ip_summed
is equal to CHECKSUM_PARTIAL. Moreover, set skb->ip_summed to CHECKSUM_NONE
when the NAT code computes the CRC, to prevent offloaders from computing
it again (on ixgbe this resulted in a transmission with wrong L4 checksum).

Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: rpfilter: bypass ipv4 lbcast packets with zeronet source
Liping Zhang [Sat, 3 Dec 2016 13:25:08 +0000 (21:25 +0800)]
netfilter: rpfilter: bypass ipv4 lbcast packets with zeronet source

Otherwise, DHCP Discover packets(0.0.0.0->255.255.255.255) may be
dropped incorrectly.

Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: allow to filter stateful object dumps by type
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:06:11 +0000 (00:06 +0100)]
netfilter: nf_tables: allow to filter stateful object dumps by type

This patch adds the netlink code to filter out dump of stateful objects,
through the NFTA_OBJ_TYPE netlink attribute.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nft_objref: support for stateful object maps
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:06:03 +0000 (00:06 +0100)]
netfilter: nft_objref: support for stateful object maps

This patch allows us to refer to stateful object dictionaries, the
source register indicates the key data to be used to look up for the
corresponding state object. We can refer to these maps through names or,
alternatively, the map transaction id. This allows us to refer to both
anonymous and named maps.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: add stateful object reference to set elements
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:06:00 +0000 (00:06 +0100)]
netfilter: nf_tables: add stateful object reference to set elements

This patch allows you to refer to stateful objects from set elements.
This provides the infrastructure to create maps where the right hand
side of the mapping is a stateful object.

This allows us to build dictionaries of stateful objects, that you can
use to perform fast lookups using any arbitrary key combination.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nft_quota: add depleted flag for objects
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:05:56 +0000 (00:05 +0100)]
netfilter: nft_quota: add depleted flag for objects

Notify on depleted quota objects. The NFT_QUOTA_F_DEPLETED flag
indicates we have reached overquota.

Add pointer to table from nft_object, so we can use it when sending the
depletion notification to userspace.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: notify internal updates of stateful objects
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:05:48 +0000 (00:05 +0100)]
netfilter: nf_tables: notify internal updates of stateful objects

Introduce nf_tables_obj_notify() to notify internal state changes in
stateful objects. This is used by the quota object to report depletion
in a follow up patch.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_tables: atomic dump and reset for stateful objects
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:05:44 +0000 (00:05 +0100)]
netfilter: nf_tables: atomic dump and reset for stateful objects

This patch adds a new NFT_MSG_GETOBJ_RESET command perform an atomic
dump-and-reset of the stateful object. This also comes with add support
for atomic dump and reset for counter and quota objects.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nft_quota: dump consumed quota
Pablo Neira Ayuso [Sun, 27 Nov 2016 23:05:52 +0000 (00:05 +0100)]
netfilter: nft_quota: dump consumed quota

Add a new attribute NFTA_QUOTA_CONSUMED that displays the amount of
quota that has been already consumed. This allows us to restore the
internal state of the quota object between reboots as well as to monitor
how wasted it is.

This patch changes the logic to account for the consumed bytes, instead
of the bytes that remain to be consumed.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agoi40e: move all updates for VLAN mode into i40e_sync_vsi_filters
Jacob Keller [Fri, 11 Nov 2016 20:39:31 +0000 (12:39 -0800)]
i40e: move all updates for VLAN mode into i40e_sync_vsi_filters

In a similar fashion to how we handled exiting VLAN mode, move the logic
in i40e_vsi_add_vlan into i40e_sync_vsi_filters. Extract this logic into
its own function for ease of understanding as it will become quite
complex.

The new function, i40e_correct_mac_vlan_filters() correctly updates all
filters for when we need to enter VLAN mode, exit VLAN mode, and also
enforces the PVID when assigned.

Call i40e_correct_mac_vlan_filters from i40e_sync_vsi_filters passing it
the number of active VLAN filters, and the two temporary lists.

Remove the function for updating VLAN=0 filters from i40e_vsi_add_vlan.

The end result is that the logic for entering and exiting VLAN mode is
in one location which has the most knowledge about all filters. This
ensures that we always correctly have the non-VLAN filters assigned to
VID=0 or VID=-1 regardless of how we ended up getting to this result.

Additionally this enforces the PVID at sync time so that we know for
certain that an assigned PVID results in only filters with that PVID
will be added to the firmware.

Change-ID: I895cee81e9c92d0a16baee38bd0ca51bbb14e372
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: use (add|rm)_vlan_all_mac helper functions when changing PVID
Jacob Keller [Fri, 11 Nov 2016 20:39:30 +0000 (12:39 -0800)]
i40e: use (add|rm)_vlan_all_mac helper functions when changing PVID

The current flow for adding or updating the PVID for a VF uses
i40e_vsi_add_vlan and i40e_vsi_kill_vlan which each take, then release
the hash lock. In addition the two functions also must take special care
that they do not perform VLAN mode changes as this will make the code in
i40e_ndo_set_vf_port_vlan behave incorrectly.

Fix these issues by using the new helper functions i40e_add_vlan_all_mac
and i40e_rm_vlan_all_mac which expect the hash lock to already be taken.
Additionally these functions do not perform any state updates in regards
to VLAN mode, so they are safe to use in the PVID update flow.

It should be noted that we don't need the VLAN mode update code here,
because there are only a few flows here.

(a) we're adding a new PVID
  In this case, if we already had VLAN filters the VSI is knocked
  offline so we don't need to worry about pre-existing VLAN filters

(b) we're replacing an existing PVID
  In this case, we can't have any VLAN filters except those with the old
  PVID which we already take care of manually.

(c) we're removing an existing PVID
  Similarly to above, we can't have any existing VLAN filters except
  those with the old PVID which we already take care of correctly.

Because of this, we do not need (or even want) the special accounting
done in i40e_vsi_add_vlan, so use of the helpers is a saner alternative.
It also opens the door for a future patch which will refactor the flow
of i40e_vsi_add_vlan now that it is not needed in this function.

Change-ID: Ia841f63da94e12b106f41cf7d28ce8ce92f2ad99
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: factor out addition/deletion of VLAN per each MAC address
Jacob Keller [Fri, 11 Nov 2016 20:39:29 +0000 (12:39 -0800)]
i40e: factor out addition/deletion of VLAN per each MAC address

A future refactor of how the PF assigns a PVID to a VF will want to be
able to add and remove a block of filters by VLAN without worrying about
accidentally triggering the accounting for I40E_VLAN_ANY. Additionally
the PVID assignment would like to be able to batch several changes under
one use of the mac_filter_hash_lock.

Factor out the addition and deletion of a VLAN on all MACs into their
own function which i40e_vsi_(add|kill)_vlan can use. These new functions
expect the caller to take the hash lock, as well as perform any
necessary accounting for updating I40E_VLAN_ANY filters if we are now
operating under VLAN mode.

Change-ID: If79e5b60b770433275350a74b3f1880333a185d5
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: delete filter after adding its replacement when converting
Jacob Keller [Fri, 11 Nov 2016 20:39:28 +0000 (12:39 -0800)]
i40e: delete filter after adding its replacement when converting

Fix a subtle issue with the code for converting VID=-1 filters into VID=0
filters when adding a new VLAN. Previously the code deleted the VID=-1
filter, and then added a new VID=0 filter. In the rare case that the
addition fails due to -ENOMEM, we end up completely deleting the filter
which prevents recovery if memory pressure subsides. While it is not
strictly an issue because it is likely that memory issues would result
in many other problems, we shouldn't delete the filter until after the
addition succeeds.

Change-ID: Icba07ddd04ecc6a3b27c2e29f2c1c8673d266826
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: refactor i40e_update_filter_state to avoid passing aq_err
Jacob Keller [Fri, 11 Nov 2016 20:39:27 +0000 (12:39 -0800)]
i40e: refactor i40e_update_filter_state to avoid passing aq_err

The current caller of i40e_update_filter_state incorrectly passes
aq_ret, an i40e_status variable, instead of the expected aq_err. This
happens to work because i40e_status is actually just a typedef integer,
and 0 is still the successful return. However i40e_update_filter_state
has special handling for ENOSPC which is currently being ignored.

Also notice that firmware does not update the per-filter response for
many types of errors, such as EINVAL. Thus, modify the filter setup so
that the firmware response memory is pre-set with I40E_AQC_MM_ERR_NO_RES.

This enables us to refactor i40e_update_filter_state, removing the need
to pass aq_err and avoiding a need for having 3 different flows for
checking the filter state.

The resulting code for i40e_update_filter_state is much simpler, only
a single loop and we always check each filter response value every time.
Since we pre-set the response value to match our expected error this
correctly works for all success and error flows.

Change-ID: Ie292c9511f34ee18c6ef40f955ad13e28b7aea7d
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: recalculate vsi->active_filters from hash contents
Jacob Keller [Fri, 11 Nov 2016 20:39:26 +0000 (12:39 -0800)]
i40e: recalculate vsi->active_filters from hash contents

Previous code refactors have accidentally caused issues with the
counting of active_filters. Avoid similar issues in the future by simply
re-counting the active filters every time after we handle add and delete
of all the filters. Additionally this allows us to simplify the check
for when we exit promiscuous mode since we can combine the check for
failed filters at the same time.

Additionally since we recount filters at the end we need to set
vsi->promisc_threshold as well.

The resulting code takes a bit longer since we do have to loop over
filters again. However, the result is more readable and less likely to
become incorrect due to failed accounting of filters in the future.
Finally, this ensures that it is not possible for vsi->active_filters to
ever underflow since we never decrement it.

Change-ID: Ib4f3a377e60eb1fa6c91ea86cc02238c08edd102
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: defeature support for PTP L4 frame detection on XL710
Jacob Keller [Fri, 11 Nov 2016 20:39:25 +0000 (12:39 -0800)]
i40e: defeature support for PTP L4 frame detection on XL710

A product decision has been made to defeature detection of PTP frames
over L4 (UDP) on the XL710 MAC. Do not advertise support for L4
timestamping.

Change-ID: I41fbb0f84ebb27c43e23098c08156f2625c6ee06
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: lock service task correctly
Mitch Williams [Mon, 21 Nov 2016 21:03:51 +0000 (13:03 -0800)]
i40e: lock service task correctly

The service task lock was being set in the scheduling function, not the
actual service task. This would potentially leave the bit set for a long
time before the task actually ran. Furthermore, if the service task
takes too long, it calls the schedule function to reschedule itself -
which would fail to take the lock and do nothing.

Instead, set and clear the lock bit in the service task itself. In the
process, get rid of the i40e_service_event_complete() function, which is
really just two lines of code that can be put right in the service task
itself.

Change-ID: I83155e682b686121e2897f4429eb7d3f7c669168
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Add functions which apply correct PHY access method for read and write operation
Michal Kosiarz [Mon, 21 Nov 2016 21:03:50 +0000 (13:03 -0800)]
i40e: Add functions which apply correct PHY access method for read and write operation

Depending on external PHY type, register access method should be
different. Clause22 or Clause45 can be chosen for different PHYs.
Implemented functions apply correct access method for used device.

Change-ID: If39d5f0da9c0b905a8cbdc1ab89885535e7d0426
Signed-off-by: Michal Kosiarz <michal.kosiarz@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Add FEC for 25g
Carolyn Wyborny [Mon, 21 Nov 2016 21:03:49 +0000 (13:03 -0800)]
i40e: Add FEC for 25g

This patch adds adminq support for Forward Error
Correction ("FEC")for 25g products.

Change-ID: Iaff4910737c239d2c730e5c22a313ce9c37d3964
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jacek Naczyk <jacek.naczyk@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Add support for 25G devices
Carolyn Wyborny [Mon, 21 Nov 2016 21:03:48 +0000 (13:03 -0800)]
i40e: Add support for 25G devices

Add support for 25G devices - defines and data structures.

One tricky part here is that the firmware support for these
Devices introduces a mismatch between the PHY type enum and
the bitfields for the phy types.

This change creates a macro and uses it to increment the 25G
PHY values when creating 25G bitfields.

Change-ID: I69b24d837d44cf9220bf5cb8dd46c5be89ce490b
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: use unsigned printf format specifier for active_filters count
Jacob Keller [Mon, 21 Nov 2016 21:03:47 +0000 (13:03 -0800)]
i40e: use unsigned printf format specifier for active_filters count

Replace the %d specifier used for printing vsi->active_filters and
vsi->promisc_threshold with an unsigned %u format specifier. While it is
unlikely in practice that these values will ever reach such a large
number they are unsigned values and thus should not be interpreted as
negative numbers.

Change-ID: Iff050fad5a1c8537c4c57fcd527441cd95cfc0d4
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoChanged version from 1.6.21 to 1.6.25
Bimmy Pujari [Tue, 8 Nov 2016 21:05:19 +0000 (13:05 -0800)]
Changed version from 1.6.21 to 1.6.25

Signed-off-by: Bimmy Pujari <bimmy.pujari@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Blink LED on 1G BaseT boards
Henry Tieman [Tue, 8 Nov 2016 21:05:18 +0000 (13:05 -0800)]
i40e: Blink LED on 1G BaseT boards

Before this patch "ethtool -p" was not blinking the LEDs on boards
with 1G BaseT PHYs.

This commit identifies 1G BaseT boards as having the LEDs connected
to the MAC. Also, renamed the flag to be more descriptive of usage.
The flag is now I40E_FLAG_PHY_CONTROLS_LEDS.

Change-ID: I4eb741da9780da7849ddf2dc4c0cb27ffa42a801
Signed-off-by: Henry Tieman <henry.w.tieman@intel.com>
Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: remove code to handle dev_addr specially
Jacob Keller [Tue, 8 Nov 2016 21:05:17 +0000 (13:05 -0800)]
i40e: remove code to handle dev_addr specially

The netdev->dev_addr MAC filter already exists in the
MAC/VLAN hash table, as it is added when we configure
the netdev in i40e_configure_netdev. Because we already
know that this address will be updated in the
hash_for_each loops, we do not need to handle it
specially. This removes duplicate code and simplifies
the i40e_vsi_add_vlan and i40e_vsi_kill_vlan functions.
Because we know these filters must be part of the
MAC/VLAN hash table, this should not have any functional
impact on what filters are included and is merely a code
simplification.

Change-ID: I5e648302dbdd7cc29efc6d203b7019c11f0b5705
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: napi_poll must return the work done
Alexander Duyck [Tue, 8 Nov 2016 21:05:16 +0000 (13:05 -0800)]
i40e/i40evf: napi_poll must return the work done

Currently the function i40e_napi-poll() returns 0 when it clean completely
the Rx rings, but this foul budget accounting in core code.

Fix this by returning the actual work done, capped to budget - 1, since
the core doesn't allow to return the full budget when the driver modifies
the NAPI status

This is based on a similar change that was made for the ixgbe driver by
Paolo Abeni.

Change-ID: Ic3d93ad2fa2fc8ce3164bc461e69367da0f9173b
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: restore workaround for removing default MAC filter
Jacob Keller [Tue, 8 Nov 2016 21:05:15 +0000 (13:05 -0800)]
i40e: restore workaround for removing default MAC filter

A previous commit 53cb6e9e8949 ("i40e: Removal of workaround for simple
MAC address filter deletion") removed a workaround for some
firmware versions which was reported to not be necessary in production
NICs. Unfortunately this workaround is necessary in some configurations,
specifically the Ethernet Controller XL710 for 40GbE QSFP+ (8086:1583).

Without this patch, the mentioned NICs with current firmware exhibit
issues when adding VLANs, as outlined by the following reproduction:

  $modprobe i40e
  $ip link set <device> up
  $ip link add link <device> vlan100 type vlan id 100
  $dmesg | tail
  <snip>
  kernel: i40e 0000:82:00.0: Error I40E_AQ_RC_EINVAL adding RX
filters on PF, promiscuous mode forced on

This results in filters being marked as FAILED and setting the device in
promiscuous mode.

The root cause of receiving the -EINVAL error response appears to be due
to a conflict with the default MAC filter which still exists on the
default firmware for this device. Attempting to add a new VLAN filter on
the default MAC address conflicts with the IGNORE_VLAN setting on the
default rule.

Change-ID: I4d8f6d48ac5f60cfe981b3baad30eb4d7c170d61
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: simplify txd use count calculation
Mitch Williams [Tue, 8 Nov 2016 21:05:14 +0000 (13:05 -0800)]
i40e: simplify txd use count calculation

The i40e_txd_use_count function was fast but confusing. In the comments,
it even admits that it's ugly. So replace it with a new function that is
(very) slightly faster and has extensive commenting to help the thicker
among us (including the author, who will forget in a week) understand
how it works.

Change-ID: Ifb533f13786a0bf39cb29f77969a5be2c83d9a87
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Driver prints log message on link speed change
Filip Sadowski [Tue, 8 Nov 2016 21:05:13 +0000 (13:05 -0800)]
i40e: Driver prints log message on link speed change

This patch makes the driver log link speed change. Before applying the
patch link messages were printed only on state change. Now message is
printed when link is brought up or down and when speed changes.

Change-ID: Ifbee14b4b16c24967450b3cecac6e8351dcc8f74
Signed-off-by: Filip Sadowski <filip.sadowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agotun: Use netif_receive_skb instead of netif_rx
Andrey Konovalov [Thu, 1 Dec 2016 09:34:40 +0000 (10:34 +0100)]
tun: Use netif_receive_skb instead of netif_rx

This patch changes tun.c to call netif_receive_skb instead of netif_rx
when a packet is received (if CONFIG_4KSTACKS is not enabled to avoid
stack exhaustion). The difference between the two is that netif_rx queues
the packet into the backlog, and netif_receive_skb proccesses the packet
in the current context.

This patch is required for syzkaller [1] to collect coverage from packet
receive paths, when a packet being received through tun (syzkaller collects
coverage per process in the process context).

As mentioned by Eric this change also speeds up tun/tap. As measured by
Peter it speeds up his closed-loop single-stream tap/OVS benchmark by
about 23%, from 700k packets/second to 867k packets/second.

A similar patch was introduced back in 2010 [2, 3], but the author found
out that the patch doesn't help with the task he had in mind (for cgroups
to shape network traffic based on the original process) and decided not to
go further with it. The main concern back then was about possible stack
exhaustion with 4K stacks.

[1] https://github.com/google/syzkaller

[2] https://www.spinics.net/lists/netdev/thrd440.html#130570

[3] https://www.spinics.net/lists/netdev/msg130570.html

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>