]> git.karo-electronics.de Git - linux-beck.git/log
linux-beck.git
9 years agonet/mlx5: Ethernet resource handling files
Amir Vadai [Thu, 28 May 2015 19:28:47 +0000 (22:28 +0300)]
net/mlx5: Ethernet resource handling files

This patch contains the resource handling files:
- flow_table.c: This file contains the code to handle the low level API
to configure hardware flow table. It is separated from
the flow_table_en.c, because it will be used in the
future by Raw Ethernet QP in mlx5_ib too.
- en_flow_table.[ch]: Ethernet flow steering handling. The flow table
object contain a mapping between flow specs and TIRs.
This mechanism will be used also to configure e-switch
in the future, when SR-IOV support will be added.
- transobj.[ch] - Low level functions to create/modify/destroy the
                  transport objects: RQ/SQ/TIR/TIS
- vport.[ch] - Handle attributes of a virtual port (vPort) in the
  embedded switch. Currently this switch is a passthrough, until SR-IOV
  support will be added.

Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5: Ethernet Datapath files
Amir Vadai [Thu, 28 May 2015 19:28:46 +0000 (22:28 +0300)]
net/mlx5: Ethernet Datapath files

en_[rt]x.c contains the data path related code specific to tx or rx.
en_txrx.c contains data path code which is common for both the rx and
tx, this is mainly napi related code.

Below are the objects that are being used by the hardware and the driver
in the data path:

Channel - one channel per IRQ. Every channel object contains:
  RQ  - describes the rx queue
  TIR - One TIR (Transport Interface Receive) object per flow type. TIR
        contains attributes for a type of rx flow (e.g IPv4, IPv6 etc).
        A flow is defined in the Flow Table.
        Currently TIR describes the RSS hash parameters if exists and LRO
        attributes.
  SQ  - describes the a tx queue. There is one SQ (Send Queue) per
        TC (traffic class).
  TIS - There is one TIS (Transport Interface Send) per TC.  It
        describes the TC and may later be extended to describe more
transport properties.

Both RQ and SQ inherit from the object WQ (work queue). This common code
to describe the layout of CQE's WQE's in memory is in the files wq.[cj]

For every channel there is one NAPI context that is used for RX and
for TX.

Driver is using netdev_alloc_skb() to allocate skb's.

Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Set/Query port MTU commands
Saeed Mahameed [Thu, 28 May 2015 19:28:45 +0000 (22:28 +0300)]
net/mlx5_core: Set/Query port MTU commands

Introduce set/Query low level functions to access MTU in hardware. To be
used by the netdev.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Modify CQ moderation parameters
Rana Shahout [Thu, 28 May 2015 19:28:44 +0000 (22:28 +0300)]
net/mlx5_core: Modify CQ moderation parameters

Introduce mlx5_core_modify_cq_moderation() to be used by the netdev, to
set hardware coalescing.

Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Implement get/set port status
Rana Shahout [Thu, 28 May 2015 19:28:43 +0000 (22:28 +0300)]
net/mlx5_core: Implement get/set port status

Implemet get/set port status low level functions to be exposed by the
netdev.

Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Implement access functions of ptys register fields
Saeed Mahameed [Thu, 28 May 2015 19:28:42 +0000 (22:28 +0300)]
net/mlx5_core: Implement access functions of ptys register fields

Those registers will be used by the ethtool to set/get settings.

Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: New device capabilities handling
Saeed Mahameed [Thu, 28 May 2015 19:28:41 +0000 (22:28 +0300)]
net/mlx5_core: New device capabilities handling

- Query all supported types of dev caps on driver load.
- Store the Cap data outbox per cap type into driver private data.
- Introduce new Macros to access/dump stored caps (using the auto
  generated data types).
- Obsolete SW representation of dev caps (no need for SW copy for each
  cap).
- Modify IB driver to use new macros for checking caps.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: HW data structs/types definitions cleanup
Saeed Mahameed [Thu, 28 May 2015 19:28:40 +0000 (22:28 +0300)]
net/mlx5_core: HW data structs/types definitions cleanup

mlx5_ifc.h was heavily modified here since it is now generated by a
script from the device specification (PRM rev 0.25). This specification
is backward compatible to existing hardware.

Some structures/fields were added here in order to enable the Ethernet
functionality of the driver.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Set irq affinity hints
Saeed Mahameed [Thu, 28 May 2015 19:28:39 +0000 (22:28 +0300)]
net/mlx5_core: Set irq affinity hints

Preparation for upcoming ethernet driver.
- Move msix array from eq_table struct to priv since its not related to
  eq_table
- Intorduce irq_info struct to hold all irq information
- Move name from mlx5_eq to irq_info struct since it is irq property.
- Set IRQ affinity hints

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core,mlx5_ib: Do not use vmap() on coherent memory
Amir Vadai [Thu, 28 May 2015 19:28:38 +0000 (22:28 +0300)]
net/mlx5_core,mlx5_ib: Do not use vmap() on coherent memory

As David Daney pointed in mlx4_core driver [1], mlx5_core is also
misusing the DMA-API.

This patch is removing the code that vmap() memory allocated by
dma_alloc_coherent().

After this patch, users of this drivers might fail allocating resources
on memory fragmeneted systems.  This will be fixed later on.

[1] - https://patchwork.ozlabs.org/patch/458531/

CC: David Daney <david.daney@cavium.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 31 May 2015 01:09:58 +0000 (18:09 -0700)]
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-05-28

This series contains updates to ethtool, ixgbe, i40e and i40evf.

John adds helper routines for ethtool to pass VF to rx_flow_spec.  Since
the ring_cookie is 64 bits wide which is much larger than what could be
used for actual queue index values, provide helper routines to pack a VF
index into the cookie.  Then John provides a ixgbe patch to allow flow
director to use the entire queue space.

Neerav provides a i40e patch to collect XOFF Rx stats, where it was not
being collected before.

Anjali provides ATR support for tunneled packets, as well as stats to
count tunnel ATR hits.  Cleaned up PF struct members which are
unnecessary, since we can use the stat index macro directly.  Cleaned
up flow director ATR/SB messages to a higher debug level since they
are not useful unless silicon validation is happening.

Greg provides a patch to disable offline diagnostics if VFs are enabled
since ethtool offline diagnostic tests are not designed (out of scope)
to disable VF functions for testing and re-enable afterward.  Also cleans
up TODO comment that is no longer needed.

Vasu provides a fix an FCoE EOF case where i40e_fcoe_ctxt_eof() maybe
called before i40e_fcoe_eof_is_supported() is called.

Jesse adds skb->xmit_more support for i40evf.  Then provides a performance
enhancement for i40evf by inlining some functions which provides a 15%
gain in small packet performance.  Also cleans up the use of time_stamp
since it is no longer used to determine if there is a tx_hang and was
a part of a previous tx_hang design which is no longer used.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotipc: unconditionally put sock refcnt when sock timer to be deleted is pending
Ying Xue [Thu, 28 May 2015 05:19:22 +0000 (13:19 +0800)]
tipc: unconditionally put sock refcnt when sock timer to be deleted is pending

As sock refcnt is taken when sock timer is started in
sk_reset_timer(), the sock refcnt should be put when sock timer
to be deleted is in pending state no matter what "probing_state"
value of tipc sock is.

Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoif_vlan: fix vlaue -> value typo
Vivien Didelot [Thu, 28 May 2015 01:07:26 +0000 (21:07 -0400)]
if_vlan: fix vlaue -> value typo

Fixes "vlaue" for "value" in include/linux/if_vlan.h.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: allow BPF programs access skb->skb_iif and skb->dev->ifindex fields
Alexei Starovoitov [Wed, 27 May 2015 22:30:39 +0000 (15:30 -0700)]
bpf: allow BPF programs access skb->skb_iif and skb->dev->ifindex fields

classic BPF already exposes skb->dev->ifindex via SKF_AD_IFINDEX extension.
Allow eBPF program to access it as well. Note that classic aborts execution
of the program if 'skb->dev == NULL' (which is inconvenient for program
writers), whereas eBPF returns zero in such case.
Also expose the 'skb_iif' field, since programs triggered by redirected
packet need to known the original interface index.
Summary:
__skb->ifindex         -> skb->dev->ifindex
__skb->ingress_ifindex -> skb->skb_iif

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agohv_netvsc: Properly size the vrss queues
KY Srinivasan [Wed, 27 May 2015 20:16:57 +0000 (13:16 -0700)]
hv_netvsc: Properly size the vrss queues

The current algorithm for deciding on the number of VRSS channels is
not optimal since we open up the min of number of CPUs online and the
number of VRSS channels the host is offering. So on a 32 VCPU guest
we could potentially open 32 VRSS subchannels. Experimentation has
shown that it is best to limit the number of VRSS channels to the number
of CPUs within a NUMA node.

Here is the new algorithm for deciding on the number of sub-channels we
would open up:
        1) Pick the minimum of what the host is offering and what the driver
           in the guest is specifying as the default value.
        2) Pick the minimum of (1) and the numbers of CPUs in the NUMA
           node the primary channel is bound to.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: limit tcp/udp rmem/wmem to SOCK_{RCV,SND}BUF_MIN
Sorin Dumitru [Wed, 27 May 2015 19:16:49 +0000 (22:16 +0300)]
net: limit tcp/udp rmem/wmem to SOCK_{RCV,SND}BUF_MIN

This is similar to b1cb59cf2efe(net: sysctl_net_core: check SNDBUF
and RCVBUF for min length). I don't think too small values can cause
crashes in the case of udp and tcp, but I've seen this set to too
small values which triggered awful performance. It also makes the
setting consistent across all the wmem/rmem sysctls.

Signed-off-by: Sorin Dumitru <sdumitru@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'stmmac-IPQ806x'
David S. Miller [Sun, 31 May 2015 00:04:44 +0000 (17:04 -0700)]
Merge branch 'stmmac-IPQ806x'

Mathieu Olivari says:

====================
Add support for QCA IPQ806x Ethernet GMAC controller

This patch set adds support for the integrated Ethernet GMAC controller
on QCA IPQ806x SoC. This controller is based on a Gigabit Synopsys
DesignWare IP, already supported in the stmmac driver located in
drivers/net/ethernet/stmicro/stmmac.

This change is done as a follow-up to the following thread:
*http://www.spinics.net/lists/netdev/msg311265.html
While previous attempt was creating a new driver to drive this controller,
this new post leverages the existing stmmac driver by implementing the
SoC specific glue to it.

Aside from the pure stmmac glue layer, we have a couple of related
patches:
*IPQ806x NSS clock addition is cherry-picked and refreshed from the
 following thread: https://lkml.org/lkml/2014/8/6/390
*phy-handle and fixed-link support are also added in this change set so the
 driver can be fully functional on platforms using device-trees as well as
 ethernet switches.

V2:
 *Fix MODULE_LICENSE to "Dual BSD/GPL" as the dwmac-ipq806x.c is using
  ISC license.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: stmmac: ipq806x: document device tree bindings
Mathieu Olivari [Wed, 27 May 2015 18:02:50 +0000 (11:02 -0700)]
net: stmmac: ipq806x: document device tree bindings

Add the device tree bindings documentation for the QCA IPQ806x
variant of the Synopsys DesignWare MAC.

Signed-off-by: Mathieu Olivari <mathieu@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agostmmac: add ipq806x glue layer
Mathieu Olivari [Wed, 27 May 2015 18:02:49 +0000 (11:02 -0700)]
stmmac: add ipq806x glue layer

The ethernet controller available in IPQ806x is a Synopsys DesignWare
Gigabit MAC IP core, already supported by the stmmac driver.

This glue layer implements some platform specific settings required to
get the controller working on an IPQ806x based platform.

Signed-off-by: Mathieu Olivari <mathieu@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agostmmac: add fixed-link device-tree support
Mathieu Olivari [Wed, 27 May 2015 18:02:48 +0000 (11:02 -0700)]
stmmac: add fixed-link device-tree support

In case DT is used, this change adds the ability to the stmmac driver to
detect a fixed-link PHY, instanciate it, and use it during
phy_connect().

Fixed link PHYs DT usage is described in:
Documentation/devicetree/bindings/net/fixed-link.txt

Signed-off-by: Mathieu Olivari <mathieu@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agostmmac: add phy-handle support to the platform layer
Mathieu Olivari [Wed, 27 May 2015 18:02:47 +0000 (11:02 -0700)]
stmmac: add phy-handle support to the platform layer

On stmmac driver, PHY specification in device-tree was done using the
non-standard property "snps,phy-addr". Specifying a PHY on a different
MDIO bus that the one within the stmmac controller doesn't seem to be
possible when device-tree is used.

This change adds support for the phy-handle property, as specified in
Documentation/devicetree/bindings/net/ethernet.txt.

Signed-off-by: Mathieu Olivari <mathieu@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoclk: qcom: Add support for NSS/GMAC clocks and resets
Stephen Boyd [Wed, 27 May 2015 18:02:46 +0000 (11:02 -0700)]
clk: qcom: Add support for NSS/GMAC clocks and resets

Add the NSS/GMAC clocks and the TCM clock and NSS resets.

Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Mathieu Olivari <mathieu@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Fri, 29 May 2015 03:23:01 +0000 (20:23 -0700)]
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
pull request (net-next): ipsec-next 2015-05-28

1) Remove xfrm_queue_purge as this is the same as skb_queue_purge.

2) Optimize policy and state walk.

3) Use a sane return code if afinfo registration fails.

4) Only check fori a acquire state if the state is not valid.

5) Remove a unnecessary NULL check before xfrm_pol_hold
   as it checks the input for NULL.

6) Return directly if the xfrm hold queue is empty, avoid
   to take a lock as it is nothing to do in this case.

7) Optimize the inexact policy search and allow for matching
   of policies with priority ~0U.

All from Li RongQing.

Please pull or let me know if there are problems.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: qlcnic: clean up sysfs error codes
Vladimir Zapolskiy [Tue, 26 May 2015 00:49:45 +0000 (03:49 +0300)]
net: qlcnic: clean up sysfs error codes

Replace confusing QL_STATUS_INVALID_PARAM == -1 == -EPERM with -EINVAL
and QLC_STATUS_UNSUPPORTED_CMD == -2 == -ENOENT with -EOPNOTSUPP, the
latter error code is arguable, but it is already used in the driver,
so let it be here as well.

Also remove always false (!buf) check on read(), the driver should
not care if userspace gets its EFAULT or not.

Signed-off-by: Vladimir Zapolskiy <vz@mleia.com>
Acked-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotreewide: Add missing vmalloc.h inclusion.
David S. Miller [Thu, 28 May 2015 18:35:41 +0000 (11:35 -0700)]
treewide: Add missing vmalloc.h inclusion.

All of these files were only building on non-x86 because of
the indirect of inclusion of vmalloc.h by, of all things,
"net/inet_hashtables.h"

None of this got caught during build testing, because on x86
there is an implicit vmalloc.h include via on of the arch asm/
headers.

This fixes all of these

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoi40e: Bump version to 1.3.4
Catherine Sullivan [Fri, 17 Apr 2015 00:06:12 +0000 (20:06 -0400)]
i40e: Bump version to 1.3.4

Bump.

Change-ID: I54ec2787a9fead5e18447078f26e5dd27f01da44
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: remove time_stamp member
Jesse Brandeburg [Fri, 17 Apr 2015 00:06:11 +0000 (20:06 -0400)]
i40e/i40evf: remove time_stamp member

The driver doesn't use the time_stamp member to determine if there is a
tx_hang any more. There really isn't any point to the variable at all
so just remove it. It was left over from a previous tx_hang design.

Change-ID: I4c814827e1bcb46e45118fe37acdcfa814fb62a0
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: force inline transmit functions
Jesse Brandeburg [Fri, 17 Apr 2015 00:06:10 +0000 (20:06 -0400)]
i40e/i40evf: force inline transmit functions

Inlining these functions gives us about 15% more 64 byte packets per
second when using pktgen. 13.3 million to 15 million with a single
queue.

Also fix the function names in i40evf to i40evf not i40e while we are
touching the function header.

Change-ID: I3294ae9b085cf438672b6db5f9af122490ead9d0
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40evf: skb->xmit_more support
Jesse Brandeburg [Fri, 17 Apr 2015 00:06:09 +0000 (20:06 -0400)]
i40evf: skb->xmit_more support

Eric added support for skb->xmit_more in i40e, this ports that into
i40evf as well.

Support skb->xmit_more in i40evf is straightforward; we need to move
around i40e_maybe_stop_tx() call to correctly test netif_xmit_stopped()
before taking the decision to not kick the NIC.

Change-ID: Idddda6a2e4a7ab335631c91ced51f55b25eb8468
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Move the FD ATR/SB messages to a higher debug level
Anjali Singhai Jain [Fri, 17 Apr 2015 00:06:06 +0000 (20:06 -0400)]
i40e: Move the FD ATR/SB messages to a higher debug level

These are not useful unless SV is happening as there is a FD flush counter
that tracks this.

Change-ID: If2655b5a29687247d03a51d35f69854bbeb711ce
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: fix unrecognized FCOE EOF case
Vasu Dev [Fri, 17 Apr 2015 00:06:05 +0000 (20:06 -0400)]
i40e: fix unrecognized FCOE EOF case

Because i40e_fcoe_ctxt_eof should never be called without
i40e_fcoe_eof_is_supported being called first, the EOF in fcoe_ctxt_eof
should always be valid and therefore we do not need to print an error
if it is not valid.

However, a WARN ON to easily catch any calls to i40e_fcoe_ctxt_eof that
aren't preceded with a call to i40e_fcoe_eof_is_supported is helpful.

Change-ID: I3b536b1981ec0bce80576a74440b7dea3908bdb9
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Remove unneeded TODO
Greg Rose [Fri, 17 Apr 2015 00:06:03 +0000 (20:06 -0400)]
i40e/i40evf: Remove unneeded TODO

There's no need for a counter so remove the TODO comment.

Change-ID: I3321dda04934c4f5fda9b279ab666192bda44214
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Remove unnecessary pf members
Anjali Singhai Jain [Fri, 17 Apr 2015 00:06:02 +0000 (20:06 -0400)]
i40e: Remove unnecessary pf members

We can use the stat index macro directly, a variable is not required.

Change-ID: I19f08ac16353dc0cd87a1a8248d714e15a54aa8a
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add stats to count Tunnel ATR hits
Anjali Singhai Jain [Fri, 17 Apr 2015 00:06:01 +0000 (20:06 -0400)]
i40e/i40evf: Add stats to count Tunnel ATR hits

Add a 3rd dynamic filter counter to track Tunneled ATR hits separately.
Ethtool port stat "fdir_atr_tunnel_match"

Change-ID: Idd978b6db2a462b5722397cd2ffd04ef055f8655
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add ATR support for tunneled TCP/IPv4/IPv6 packets.
Anjali Singhai Jain [Fri, 17 Apr 2015 00:06:00 +0000 (20:06 -0400)]
i40e/i40evf: Add ATR support for tunneled TCP/IPv4/IPv6 packets.

Without this, RSS would have done inner header load balancing. Now we can
get the benefits of ATR for tunneled packets to better align TX and RX
queues with the right core/interrupt.

Change-ID: I07d0e0a192faf28fdd33b2f04c32b2a82ff97ddd
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Disable offline diagnostics if VFs are enabled
Greg Rose [Fri, 17 Apr 2015 00:05:59 +0000 (20:05 -0400)]
i40e: Disable offline diagnostics if VFs are enabled

Require the user to disable virtual functions before running the device
offline diagnostics.  The offline diagnostics are intended to ensure
basic operation of the device - it is beyond the scope of the diagnostic
test to handle the additional complexity of bringing all the virtual
functions offline and then back online for each test run.

Change-ID: Ic0b854851a09fc85df0c9e82c220e45885457c30
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Collect PFC XOFF RX stats even in single TC case
Neerav Parikh [Fri, 17 Apr 2015 00:05:58 +0000 (20:05 -0400)]
i40e: Collect PFC XOFF RX stats even in single TC case

When PFC is enabled for any UP in single TC configuration the driver didn't
collect the PFC XOFF RX stats. Though a single TC with PFC enabled is not a
common scenario do not prevent the driver from collecting stats if firmware
indicates that PFC is enabled.

Change-ID: Ie20bd58b07608b528f3c6d95894c9ae56b00077a
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoixgbe: Allow flow director to use entire queue space
John Fastabend [Tue, 26 May 2015 15:23:33 +0000 (08:23 -0700)]
ixgbe: Allow flow director to use entire queue space

Flow director is exported to user space using the ethtool ntuple
support. However, currently it only supports steering traffic to a
subset of the queues in use by the hardware. This change allows
flow director to specify queues that have been assigned to virtual
functions by partitioning the ring_cookie into a 8bit VF specifier
followed by 32bit queue index. At the moment we don't have any
ethernet drivers with more than 2^32 queues on a single function
as best I can tell and nor do I expect this to happen anytime
soon. This way the ring_cookie's normal use for specifying a queue
on a specific PCI function continues to work as expected.

CC: Alex Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoethtool: Add helper routines to pass vf to rx_flow_spec
John Fastabend [Tue, 26 May 2015 15:22:49 +0000 (08:22 -0700)]
ethtool: Add helper routines to pass vf to rx_flow_spec

The ring_cookie is 64 bits wide which is much larger than can be used
for actual queue index values. So provide some helper routines to
pack a VF index into the cookie. This is useful to steer packets to
a VF ring without having to know the queue layout of the device.

CC: Alex Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agotcp/dccp: warn user for preferred ip_local_port_range
Eric Dumazet [Wed, 27 May 2015 18:34:37 +0000 (11:34 -0700)]
tcp/dccp: warn user for preferred ip_local_port_range

After commit 07f4c90062f8f ("tcp/dccp: try to not exhaust
ip_local_port_range in connect()") it is advised to have an even number
of ports described in /proc/sys/net/ipv4/ip_local_port_range

This means start/end values should have a different parity.

Let's warn sysadmins of this, so that they can update their settings
if they want to.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: connect() from bound sockets can be faster
Eric Dumazet [Wed, 27 May 2015 17:46:02 +0000 (10:46 -0700)]
tcp: connect() from bound sockets can be faster

__inet_hash_connect() does not use its third argument (port_offset)
if socket was already bound to a source port.

No need to perform useless but expensive md5 computations.

Reported-by: Crestez Dan Leonard <cdleonard@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cxgb4-next'
David S. Miller [Wed, 27 May 2015 18:26:38 +0000 (14:26 -0400)]
Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
cxgb4/cxgb4vf: Adds FL starvation support and cleanup

This patch series adds the following.
Adds debugfs entry to inject freelist starvation and some function and
argument cleanup

This patch series has been created against net-next tree and includes
patches on cxgb4 and cxgb4vf driver.

We have included all the maintainers of respective drivers. Kindly review
the change and let us know in case of any review comments.

Thanks

V2:
Skipping patch "cxgb4: Add support for loopback between VI of same port".
This needs some major code change, since module param is not recommended.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4/cxgb4vf: function and argument name cleanup
Hariprasad Shenai [Wed, 27 May 2015 17:00:24 +0000 (22:30 +0530)]
cxgb4/cxgb4vf: function and argument name cleanup

This patch changes variable name 'fn' to 'pf' of structure adapter.
A 'fn' usually stands for PCI function which could be a PF or a VF.
However, the use of this particular variable is explicitly limited to PF
only. So, be specific about it in the variable name.

Also corrects arguments passed for fn t4_ofld_eq_free, t4_ctrl_eq_free,
t4_eth_eq_free, t4_iq_free, t4_alloc_vi, t4_fw_hello, t4_wr_mbox and
t4_cfg_pfvf function.

Also renames cxgb4_t4_bar2_sge_qregs to t4_bar2_sge_qregs and renames
the latter function name in cxgb4vf driver to t4vf_bar2_sge_qregs to
avoid conflicts. Also fixes alignment for these function.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add debugfs facility to inject FL starvation
Hariprasad Shenai [Wed, 27 May 2015 17:00:23 +0000 (22:30 +0530)]
cxgb4: Add debugfs facility to inject FL starvation

Add debugfs entry to inject Freelist starvation, used only for debugging
purpose.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqla4xxx: add a missing include
Eric Dumazet [Wed, 27 May 2015 04:43:44 +0000 (21:43 -0700)]
qla4xxx: add a missing include

vmalloc.h used to be included from include/net/inet_hashtables.h
but it is no longer the case.

Fixes: 095dc8e0c368 ("tcp: fix/cleanup inet_ehash_locks_alloc()")
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumzet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'thunderx'
David S. Miller [Wed, 27 May 2015 18:19:44 +0000 (14:19 -0400)]
Merge branch 'thunderx'

Aleksey Makarov says:

====================
Adding support for Cavium ThunderX network controller

This patchset adds support for the Cavium ThunderX network controller.

changes in v6:
 * unused preprocessor symbols were removed
 * reduce no of atomic operations in SQ maintenance
 * support for TCP segmentation at driver level
 * reset RBDR if fifo state is FAIL
 * fixed an issue with link state mailbox message

changes in v5:
 * __packed were removed.  now we rely on C language ABI
 * nic_dbg() -> netdev_dbg()
 * fixes for a typo, constant spelling and using BIT_ULL
 * use print_hex_dump()
 * unnecessary conditions in a long if() chain were removed

changes in v4:
 * the patch "pci: Add Cavium PCI vendor id" was attributed correctly
 * a note that Cavium id is used in many drivers was added
 * the license comments now match MODULE_LICENSE
 * a comment explaining usage of writeq_relaxed()/readq_relaxed() was added

changes in v3:
 * code cleanup
 * issues discovered by reviewers were addressed

changes in v2:
 * non-generic module parameters removed
 * ethtool support added (nicvf_set_rxnfc())

v5: https://lkml.kernel.org/g/<1432344498-17131-1-git-send-email-aleksey.makarov@caviumnetworks.com>
v4: https://lkml.kernel.org/g/<1432000757-28700-1-git-send-email-aleksey.makarov@auriga.com>
v3: https://lkml.kernel.org/g/<1431747401-20847-1-git-send-email-aleksey.makarov@auriga.com>
v2: https://lkml.kernel.org/g/<1415596445-10061-1-git-send-email-rric@kernel.org>
v1: https://lkml.kernel.org/g/<20141030165434.GW20170@rric.localhost>
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Adding support for Cavium ThunderX network controller
Sunil Goutham [Wed, 27 May 2015 02:20:15 +0000 (19:20 -0700)]
net: Adding support for Cavium ThunderX network controller

This patch adds support for the Cavium ThunderX network controller.
The driver is on the pci bus and thus requires the Thunder PCIe host
controller driver to be enabled.

Signed-off-by: Maciej Czekaj <mjc@semihalf.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: Ganapatrao Kulkarni <ganapatrao.kulkarni@caviumnetworks.com>
Signed-off-by: Aleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Kamil Rytarowski <kamil@semihalf.com>
Signed-off-by: Thanneeru Srinivasulu <tsrinivasulu@caviumnetworks.com>
Signed-off-by: Sruthi Vangala <svangala@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agopci: Add Cavium PCI vendor id
Sunil Goutham [Wed, 27 May 2015 02:20:14 +0000 (19:20 -0700)]
pci: Add Cavium PCI vendor id

This vendor id will be used by network (vNIC), USB (xHCI),
SATA (AHCI), GPIO, I2C, MMC and maybe other drivers
for ThunderX SoC.

Acked-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: Aleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: add similarly conflicting jump test case only for classic
Daniel Borkmann [Tue, 26 May 2015 20:35:43 +0000 (22:35 +0200)]
test_bpf: add similarly conflicting jump test case only for classic

While 3b52960266a3 ("test_bpf: add more eBPF jump torture cases")
added the int3 bug test case only for eBPF, which needs exactly 11
passes to converge, here's a version for classic BPF with 11 passes,
and one that would need 70 passes on x86_64 to actually converge for
being successfully JITed. Effectively, all jumps are being optimized
out resulting in a JIT image of just 89 bytes (from originally max
BPF insns), only returning K.

Might be useful as a receipe for folks wanting to craft a test case
when backporting the fix in commit 3f7352bf21f8 ("x86: bpf_jit: fix
compilation of large bpf programs") while not having eBPF. The 2nd
one is delegated to the interpreter as the last pass still results
in shrinking, in other words, this one won't be JITed on x86_64.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'sfc-next'
David S. Miller [Wed, 27 May 2015 17:54:52 +0000 (13:54 -0400)]
Merge branch 'sfc-next'

Edward Cree says:

====================
sfc: add MCDI tracing

This patchset adds support for logging MCDI (Management-Controller-to-
 Driver Interface) interactions between the sfc driver and a bound device,
 to aid in debugging.
Solarflare has a tool to decode the resulting traces and will look to
 open-source this if there is any external interest, but the protocol is
 already detailed in drivers/net/ethernet/sfc/mcdi_pcol.h.
The logging buffer we allocate per MCDI context is a work area for
 constructing each individual message before logging it with netif_info.
The reason the buffer is long-lived is simply to avoid the overhead of
 allocating and freeing it every MCDI call, since MCDIs are already known
 to be serialised for other reasons.

--
v4: remove patch #4, which has already been applied via sshah
v3: add some explanations to cover letter and patch #4
v2: avoid long lines in cover letter; fix multiline comment style
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosfc: add module parameter to enable MCDI logging on new functions
Edward Cree [Wed, 27 May 2015 12:14:26 +0000 (13:14 +0100)]
sfc: add module parameter to enable MCDI logging on new functions

As many issues are encountered at probe time, where MCDI logging can't be
 enabled through the sysfs node, this change adds a module parameter
 'mcdi_logging_default', which defaults to false.  When set to true, newly-
 probed functions will have MCDI logging enabled.  The setting can
 subsequently be changed as normal through the sysfs node.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosfc: add sysfs entry to control MCDI tracing
Edward Cree [Wed, 27 May 2015 12:14:01 +0000 (13:14 +0100)]
sfc: add sysfs entry to control MCDI tracing

MCDI tracing is enabled per-function with a sysfs file
    /sys/class/net/<NET_DEV>/device/mcdi_logging

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosfc: add tracing of MCDI commands
Edward Cree [Wed, 27 May 2015 12:13:54 +0000 (13:13 +0100)]
sfc: add tracing of MCDI commands

MCDI tracing is conditional on CONFIG_SFC_MCDI_LOGGING, which is enabled
 by default.

Each MCDI command will produce a console line like
    sfc dom:bus:dev:fn ifname: MCDI RPC REQ: xxxxxxxx [yyyyyyyy...]
where xxxxxxxx etc. are the raw MCDI payload in 32-bit hex chunks.
The response will then produce a similar line with "RESP" instead of "REQ",
 and containing the MCDI response payload (if any).

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovxlan: release lock after each bucket in vxlan_cleanup
Sorin Dumitru [Tue, 26 May 2015 07:42:04 +0000 (10:42 +0300)]
vxlan: release lock after each bucket in vxlan_cleanup

We're seeing some softlockups from this function when there
are a lot fdb entries on a vxlan device. Taking the lock for
each bucket instead of the whole table is enough to fix that.

Signed-off-by: Sorin Dumitru <sdumitru@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp/dccp: try to not exhaust ip_local_port_range in connect()
Eric Dumazet [Sun, 24 May 2015 21:49:35 +0000 (14:49 -0700)]
tcp/dccp: try to not exhaust ip_local_port_range in connect()

A long standing problem on busy servers is the tiny available TCP port
range (/proc/sys/net/ipv4/ip_local_port_range) and the default
sequential allocation of source ports in connect() system call.

If a host is having a lot of active TCP sessions, chances are
very high that all ports are in use by at least one flow,
and subsequent bind(0) attempts fail, or have to scan a big portion of
space to find a slot.

In this patch, I changed the starting point in __inet_hash_connect()
so that we try to favor even [1] ports, leaving odd ports for bind()
users.

We still perform a sequential search, so there is no guarantee, but
if connect() targets are very different, end result is we leave
more ports available to bind(), and we spread them all over the range,
lowering time for both connect() and bind() to find a slot.

This strategy only works well if /proc/sys/net/ipv4/ip_local_port_range
is even, ie if start/end values have different parity.

Therefore, default /proc/sys/net/ipv4/ip_local_port_range was changed to
32768 - 60999 (instead of 32768 - 61000)

There is no change on security aspects here, only some poor hashing
schemes could be eventually impacted by this change.

[1] : The odd/even property depends on ip_local_port_range values parity

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'ip_frag_next'
David S. Miller [Wed, 27 May 2015 17:03:31 +0000 (13:03 -0400)]
Merge branch 'ip_frag_next'

Florian Westphal says:

====================
net: force refragmentation for DF reassembed skbs

output path tests:

    if (skb->len > mtu) ip_fragment()

This breaks connectivity in one corner case:
 If the skb was reassembled, but has the DF bit set and ..
 .. its reassembled size is <= outdev mtu ..
 .. we will forward a DF packet larger than what the sender
    transmitted on wire.

If a router later in the path can't forward this packet, it will send an
icmp error in response to an mtu that the original sender never exceeded.

This changes ipv4 defrag/output path to

a) force refragmentation for DF reassembled skbs and
b) set DF bit on all fragments when refragmenting if it was set on original
frags.

tested via:
from scapy.all import *
dip="10.23.42.2"
payload="A"*1400
packet=IP(dst=dip,id=12345,flags='DF')/UDP(sport=42,dport=42)/payload
frags=fragment(packet,fragsize=1200)
for fragment in frags:
    send(fragment)

Without this patch, we generate fragments without df bit set based
on the outgoing device mtu when fragmenting after forwarding, ie.

IP (ttl 64, id 12345, offset 0, flags [+, DF], proto UDP (17), length 1204)
    192.168.7.1.42 > 10.23.42.2.42: UDP, length 1400
IP (ttl 64, id 12345, offset 1184, flags [DF], proto UDP (17), length 244)
    192.168.7.1 > 10.23.42.2: ip-proto-17

on ingress will either turn into

IP (ttl 63, id 12345, offset 0, flags [+], proto UDP (17), length 1396)
    192.168.7.1.42 > 10.23.42.2.42: UDP, length 1400
IP (ttl 63, id 12345, offset 1376, flags [none], proto UDP (17), length 52)

(mtu 1400: We strip df and send larger fragment), or

IP (ttl 63, id 12345, offset 0, flags [DF], proto UDP (17), length 1428)
    192.168.7.1.42 > 10.23.42.2.42: [udp sum ok] UDP, length 1400

if mtu is 1500.  And in this case things break; router with a smaller mtu
will send icmp error, but original sender only sent packets <= 1204 byte.

With patch, we keep intent of such fragments and will emit DF-fragments
that won't exceed 1204 byte in size.

Joint work with Hannes Frederic Sowa.

Changes since v2:
 - split unrelated patches from series
 - rework changelog of patch #2 to better illustrate breakage
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoip_fragment: don't forward defragmented DF packet
Florian Westphal [Fri, 22 May 2015 14:32:51 +0000 (16:32 +0200)]
ip_fragment: don't forward defragmented DF packet

We currently always send fragments without DF bit set.

Thus, given following setup:

mtu1500 - mtu1500:1400 - mtu1400:1280 - mtu1280
   A           R1              R2         B

Where R1 and R2 run linux with netfilter defragmentation/conntrack
enabled, then if Host A sent a fragmented packet _with_ DF set to B, R1
will respond with icmp too big error if one of these fragments exceeded
1400 bytes.

However, if R1 receives fragment sizes 1200 and 100, it would
forward the reassembled packet without refragmenting, i.e.
R2 will send an icmp error in response to a packet that was never sent,
citing mtu that the original sender never exceeded.

The other minor issue is that a refragmentation on R1 will conceal the
MTU of R2-B since refragmentation does not set DF bit on the fragments.

This modifies ip_fragment so that we track largest fragment size seen
both for DF and non-DF packets, and set frag_max_size to the largest
value.

If the DF fragment size is larger or equal to the non-df one, we will
consider the packet a path mtu probe:
We set DF bit on the reassembled skb and also tag it with a new IPCB flag
to force refragmentation even if skb fits outdev mtu.

We will also set DF bit on each fragment in this case.

Joint work with Hannes Frederic Sowa.

Reported-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ipv4: avoid repeated calls to ip_skb_dst_mtu helper
Florian Westphal [Fri, 22 May 2015 14:32:50 +0000 (16:32 +0200)]
net: ipv4: avoid repeated calls to ip_skb_dst_mtu helper

ip_skb_dst_mtu is small inline helper, but its called in several places.

before: 17061      44       0   17105    42d1 net/ipv4/ip_output.o
after:  16805      44       0   16849    41d1 net/ipv4/ip_output.o

Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'phy_rgmii'
David S. Miller [Wed, 27 May 2015 04:27:36 +0000 (00:27 -0400)]
Merge branch 'phy_rgmii'

Florian Fainelli says:

====================
net: phy: phy_interface_is_rgmii helper

As you suggested, here is the helper function to avoid missing some RGMII
interface checks. Had to wait for net to be merged in net-next to avoid
submitting the same patch/commit.

Dan, you might want to rebase your dp83867 submission to use that helper
when you this patchset gets merged into net-next, thanks!
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: phy: Utilize phy_interface_is_rgmii
Florian Fainelli [Tue, 26 May 2015 19:19:59 +0000 (12:19 -0700)]
net: phy: Utilize phy_interface_is_rgmii

Update all open-coded tests for all 4 PHY_INTERFACE_MODE_RGMII* values
to use the newly introduced helper: phy_interface_is_rgmii.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: phy: Add phy_interface_is_rgmii helper
Florian Fainelli [Tue, 26 May 2015 19:19:58 +0000 (12:19 -0700)]
net: phy: Add phy_interface_is_rgmii helper

RGMII interfaces come in 4 different flavors that the PHY library needs
to care about: regular RGMII (no delays), RGMII with either RX or TX
delay, and both. In order to avoid errors of checking only for one type
of RGMII interface and miss the 3 others, introduce a convenience
function which tests for all values.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: Fix fib_trie.c build, missing linux/vmalloc.h include.
David S. Miller [Wed, 27 May 2015 04:19:03 +0000 (00:19 -0400)]
ipv4: Fix fib_trie.c build, missing linux/vmalloc.h include.

We used to get this indirectly I supposed, but no longer do.

Either way, an explicit include should have been done in the
first place.

   net/ipv4/fib_trie.c: In function '__node_free_rcu':
>> net/ipv4/fib_trie.c:293:3: error: implicit declaration of function 'vfree' [-Werror=implicit-function-declaration]
      vfree(n);
      ^
   net/ipv4/fib_trie.c: In function 'tnode_alloc':
>> net/ipv4/fib_trie.c:312:3: error: implicit declaration of function 'vzalloc' [-Werror=implicit-function-declaration]
      return vzalloc(size);
      ^
>> net/ipv4/fib_trie.c:312:3: warning: return makes pointer from integer without a cast
   cc1: some warnings being treated as errors

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: tcp_tso_autosize() minimum is one packet
Eric Dumazet [Tue, 26 May 2015 15:55:28 +0000 (08:55 -0700)]
tcp: tcp_tso_autosize() minimum is one packet

By making sure sk->sk_gso_max_segs minimal value is one,
and sysctl_tcp_min_tso_segs minimal value is one as well,
tcp_tso_autosize() will return a non zero value.

We can then revert 843925f33fcc293d80acf2c5c8a78adf3344d49b
("tcp: Do not apply TSO segment limit to non-TSO packets")
and save few cpu cycles in fast path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: fix/cleanup inet_ehash_locks_alloc()
Eric Dumazet [Tue, 26 May 2015 14:55:34 +0000 (07:55 -0700)]
tcp: fix/cleanup inet_ehash_locks_alloc()

If tcp ehash table is constrained to a very small number of buckets
(eg boot parameter thash_entries=128), then we can crash if spinlock
array has more entries.

While we are at it, un-inline inet_ehash_locks_alloc() and make
following changes :

- Budget 2 cache lines per cpu worth of 'spinlocks'
- Try to kmalloc() the array to avoid extra TLB pressure.
  (Most servers at Google allocate 8192 bytes for this hash table)
- Get rid of various #ifdef

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotipc: fix bug in link protocol message create function
Jon Paul Maloy [Tue, 26 May 2015 09:40:19 +0000 (05:40 -0400)]
tipc: fix bug in link protocol message create function

In commit dd3f9e70f59f43a5712eba9cf3ee4f1e6999540c
("tipc: add packet sequence number at instant of transmission") we
made a change with the consequence that packets in the link backlog
queue don't contain valid sequence numbers.

However, when we create a link protocol message, we still use the
sequence number of the first packet in the backlog, if there is any,
as "next_sent" indicator in the message. This may entail unnecessary
retransissions or stale packet transmission when there is very low
traffic on the link.

This commit fixes this issue by only using the current value of
tipc_link::snd_nxt as indicator.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fix inet_proto_csum_replace4() sparse errors
Eric Dumazet [Tue, 26 May 2015 01:50:01 +0000 (18:50 -0700)]
net: fix inet_proto_csum_replace4() sparse errors

make C=2 CF=-D__CHECK_ENDIAN__ net/core/utils.o
...
net/core/utils.c:307:72: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:307:72:    expected restricted __wsum [usertype] addend
net/core/utils.c:307:72:    got restricted __be32 [usertype] from
net/core/utils.c:308:34: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:308:34:    expected restricted __wsum [usertype] addend
net/core/utils.c:308:34:    got restricted __be32 [usertype] to
net/core/utils.c:310:70: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:310:70:    expected restricted __wsum [usertype] addend
net/core/utils.c:310:70:    got restricted __be32 [usertype] from
net/core/utils.c:310:77: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:310:77:    expected restricted __wsum [usertype] addend
net/core/utils.c:310:77:    got restricted __be32 [usertype] to
net/core/utils.c:312:72: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:312:72:    expected restricted __wsum [usertype] addend
net/core/utils.c:312:72:    got restricted __be32 [usertype] from
net/core/utils.c:313:35: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:313:35:    expected restricted __wsum [usertype] addend
net/core/utils.c:313:35:    got restricted __be32 [usertype] to

Note we can use csum_replace4() helper

Fixes: 58e3cac5613aa ("net: optimise inet_proto_csum_replace4()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: remove a sparse error in secure_dccpv6_sequence_number()
Eric Dumazet [Tue, 26 May 2015 01:55:48 +0000 (18:55 -0700)]
net: remove a sparse error in secure_dccpv6_sequence_number()

make C=2 CF=-D__CHECK_ENDIAN__ net/core/secure_seq.o
net/core/secure_seq.c:157:50: warning: restricted __be32 degrades to
integer

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: skip fdb add if the port shouldn't learn
Wilson Kok [Mon, 25 May 2015 13:39:31 +0000 (06:39 -0700)]
bridge: skip fdb add if the port shouldn't learn

Check in fdb_add_entry() if the source port should learn, similar
check is used in br_fdb_update.
Note that new fdb entries which are added manually or
as local ones are still permitted.
This patch has been tested by running traffic via a bridge port and
switching the port's state, also by manually adding/removing entries
from the bridge's fdb.

Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agopktgen: remove one sparse error
Eric Dumazet [Mon, 25 May 2015 23:06:37 +0000 (16:06 -0700)]
pktgen: remove one sparse error

net/core/pktgen.c:2672:43: warning: incorrect type in assignment (different base types)
net/core/pktgen.c:2672:43:    expected unsigned short [unsigned] [short] [usertype] <noident>
net/core/pktgen.c:2672:43:    got restricted __be16 [usertype] protocol

Let's use proper struct ethhdr instead of hard coding everything.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: ipv6_select_ident() returns a __be32
Eric Dumazet [Mon, 25 May 2015 23:02:21 +0000 (16:02 -0700)]
ipv6: ipv6_select_ident() returns a __be32

ipv6_select_ident() returns a 32bit value in network order.

Fixes: 286c2349f666 ("ipv6: Clean up ipv6_select_ident() and ip6_fragment()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cpsw-cleanups'
David S. Miller [Mon, 25 May 2015 22:19:10 +0000 (18:19 -0400)]
Merge branch 'cpsw-cleanups'

Richard Cochran says:

====================
cpsw cleanups

While working on an out-of-tree customization, I noticed a few minor
problems in the cpsw code.  This series cleans up the issues I found.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: cpsw: remove redundant calls disabling dma interrupts.
Richard Cochran [Mon, 25 May 2015 09:02:16 +0000 (11:02 +0200)]
net: cpsw: remove redundant calls disabling dma interrupts.

The function, cpsw_intr_disable, already calls cpdma_ctlr_int_ctrl.  There
is no need to disable the dma interrupts twice.  This patch removes the
extra calls.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: cpsw: remove redundant calls enabling dma interrupts.
Richard Cochran [Mon, 25 May 2015 09:02:15 +0000 (11:02 +0200)]
net: cpsw: remove redundant calls enabling dma interrupts.

The function, cpsw_intr_enable, already calls cpdma_ctlr_int_ctrl.  There
is no need to enable the dma interrupts twice.  This patch removes the
extra call.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: cpsw: remove two unused global functions
Richard Cochran [Mon, 25 May 2015 09:02:14 +0000 (11:02 +0200)]
net: cpsw: remove two unused global functions

The funtions, cpsw_ale_flush and cpsw_ale_set_ageout, have never been used
since they were first introduced.  This patch removes the dead code.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: cpsw: fix misplaced break statements.
Richard Cochran [Mon, 25 May 2015 09:02:13 +0000 (11:02 +0200)]
net: cpsw: fix misplaced break statements.

Having the breaks too far to the left makes parsing the dense switch/case
block unnecessarily harder.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'rocker-cleanups'
David S. Miller [Mon, 25 May 2015 22:17:09 +0000 (18:17 -0400)]
Merge branch 'rocker-cleanups'

Simon Horman says:

====================
rocker: unused parameter and const cleanups

This series provides some minor though verbose cleanup of rocker.

The second patch depends on the first though it could be rebased.

I had previously asked for v2 to be put on hold while some bugs I had found
in the rocker driver were shaken out. That has now happened and the bugs
turned out to be unrelated.  Accordingly I am reposting the series.

* Changes v2 -> v3
  - Rebase and update for new variables and parameters that may be const

* Changes v1 -> v2
  - Found quite a few more variables and parameters to make const
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: mark parameters and local variables as const
Simon Horman [Mon, 25 May 2015 05:28:36 +0000 (14:28 +0900)]
rocker: mark parameters and local variables as const

Mark parameters and local variables as const where possible.

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: remove unused rocker_port parameter from rocker_port_kfree
Simon Horman [Mon, 25 May 2015 05:28:35 +0000 (14:28 +0900)]
rocker: remove unused rocker_port parameter from rocker_port_kfree

Remove unused rocker_port parameter from rocker_port_kfree.
Also remove the rocker_port parameter from callers of rocker_port_kfree
where the parameter it is now unused.

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoirda: use msecs_to_jiffies for conversion to jiffies
Nicholas Mc Guire [Mon, 25 May 2015 06:16:50 +0000 (08:16 +0200)]
irda: use msecs_to_jiffies for conversion to jiffies

API compliance scanning with coccinelle flagged:
./net/irda/timer.c:63:35-37: use of msecs_to_jiffies probably perferable

Converting milliseconds to jiffies by "val * HZ / 1000" technically
is not a clean solution as it does not handle all corner cases correctly.
By changing the conversion to use msecs_to_jiffies(val) conversion is
correct in all cases. Further the () around the arithmetic expression
was dropped.

Patch was compile tested for x86_64_defconfig + CONFIG_IRDA=m

Patch is against 4.1-rc4 (localversion-next is -next-20150522)

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoneterion: s2io: Fix kernel doc formatting
Joe Perches [Sat, 23 May 2015 17:32:55 +0000 (10:32 -0700)]
neterion: s2io: Fix kernel doc formatting

These two uses seem to have had carriage returns removed.
Make these entries like all the others in this file.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoirda: irda-usb: use msecs_to_jiffies for conversions
Nicholas Mc Guire [Sat, 23 May 2015 12:46:30 +0000 (14:46 +0200)]
irda: irda-usb: use msecs_to_jiffies for conversions

API compliance scanning with coccinelle flagged:

Converting milliseconds to jiffies by "val * HZ / 1000" is technically
is not a clean solution as it does not handle all corner cases correctly.
By changing the conversion to use msecs_to_jiffies(val) conversion is
correct in all cases.

in the current code:
  mod_timer(&self->rx_defer_timer, jiffies + (10 * HZ / 1000));
for HZ < 100 (e.g. CONFIG_HZ == 64|32 in alpha) this effectively results
in no delay at all.

Patch was compile tested for x86_64_defconfig (implies CONFIG_USB_IRDA=m)

Patch is against 4.1-rc4 (localversion-next is -next-20150522)

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: allow setting hash_max + multicast_router if interface is down
Linus Lüssing [Sat, 23 May 2015 01:12:34 +0000 (03:12 +0200)]
bridge: allow setting hash_max + multicast_router if interface is down

Network managers like netifd (used in OpenWRT for instance) try to
configure interface options after creation but before setting the
interface up.

Unfortunately the sysfs / bridge currently only allows to configure the
hash_max and multicast_router options when the bridge interface is up.
But since br_multicast_init() doesn't start any timers and only sets
default values and initializes timers it should be save to reconfigure
the default values after that, before things actually get active after
the bridge is set up.

Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: don't increase size when refragmenting forwarded ipv6 skbs
Florian Westphal [Thu, 21 May 2015 22:44:16 +0000 (00:44 +0200)]
ipv6: don't increase size when refragmenting forwarded ipv6 skbs

since commit 6aafeef03b9d ("netfilter: push reasm skb through instead of
original frag skbs") we will end up sometimes re-fragmenting skbs
that we've reassembled.

ipv6 defrag preserves the original skbs using the skb frag list, i.e. as long
as the skb frag list is preserved there is no problem since we keep
original geometry of fragments intact.

However, in the rare case where the frag list is munged or skb
is linearized, we might send larger fragments than what we originally
received.

A router in the path might then send packet-too-big errors even if
sender never sent fragments exceeding the reported mtu:

mtu 1500 - 1500:1400 - 1400:1280 - 1280
     A         R1         R2        B

1 - A sends to B, fragment size 1400
2 - R2 sends pkttoobig error for 1280
3 - A sends to B, fragment size 1280
4 - R2 sends pkttoobig error for 1280 again because it sees fragments of size 1400.

make sure ip6_fragment always caps MTU at largest packet size seen
when defragmented skb is forwarded.

Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoatm:he - Change 1 to true for bool type variable.
Shailendra Verma [Mon, 25 May 2015 19:47:23 +0000 (01:17 +0530)]
atm:he - Change 1 to true for bool type variable.

The variable irq_coalesce is bool type.
So assign the value true instead of 1.

Signed-off-by: Shailendra Verma <shailendra.capricorn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet:xen-netback - Change 1 to true for bool type variable.
Shailendra Verma [Mon, 25 May 2015 17:49:31 +0000 (23:19 +0530)]
net:xen-netback - Change 1 to true for bool type variable.

The variable separate_tx_rx_irq is bool type so assigning true
instead of 1.

Signed-off-by: Shailendra Verma <shailendra.capricorn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'ipv6_route_sharing'
David S. Miller [Mon, 25 May 2015 17:25:35 +0000 (13:25 -0400)]
Merge branch 'ipv6_route_sharing'

Martin KaFai Lau says:

====================
ipv6: Only create RTF_CACHE route after encountering pmtu exception

v4 -> v5:
- Patch 1 is new. Clean up the ipv6_select_ident() and ip6_fragment().

- Further simplify the newly added rt6_get_pcpu_route().  If there is a
  'prev' after cmpxchg, return prev instead of the newly created percpu
  clone.

v3 -> v4:
- Patch 8 is new. It keeps track of the DST_NOCACHE routes in a list to handle
  the iface down/unregister event.

- Remove rcu from the newly added rt6i_pcpu variable.  It is not needed
  because it has already been protected by the existing reader/writer lock.

- Thanks to 'Julian Anastasov <ja@ssi.bg>' for testing the FLOWI_FLAG_KNOWN_NH
  patches.

v2 -> v3:
- Patch 5 to 7 are new.  They take care of cases where the daddr in
  skb is not the one used to do the route look-up.  There is also
  related changes to rt6_nexthop() since v2 which is in patch 2/9.
  Thanks to 'Julian Anastasov <ja@ssi.bg>' for pointing it out.

- Fix a few problems in __ip6_rt_update_pmtu(), like setting the expire
  and mtu before inserting to the tree and don't do dst_destroy() after
  tree insertion failure.  Also update the rt6i_pmtu in fib6_add_rt2node().
  Thanks to 'Steffen Klassert <steffen.klassert@secunet.com>' for pointing
  it out.

- Merge ip6_pmtu_rt_cache_alloc() into ip6_rt_cache_alloc().

v1 -> v2:
- Move the /128 route bug fixes to another series (accepted).
- Create a function for checking (rt6i_flags & (RTF_NONEXTHOP | RTF_GATEWAY)).
- Avoid shuffling the skb network_header.  Instead, change the function
  signature to take iph instead of skb.

- Many Thanks to 'Hannes Frederic Sowa <hannes@stressinduktion.org>' on
  reviewing v1 and v2 and giving advice.

--Martin

~~~ start: v1 compose message (with the out-dated parts removed) ~~~

This series is to avoid creating a RTF_CACHE route whenever we are consulting
the fib6 tree with a new destination.  Instead, only create RTF_CACHE route
when we see a pmtu exception.

Out of all ipv6 RTF_CACHE routes that are created, the percentage that has a
different mtu is very small. In one of our end-user facing proxy server,
only 1k out of 80k RTF_CACHE routes have a smaller MTU.  For our DC
traffic, there is no mtu exception.

A large fib6 tree has problems like, 'ip -6 r show' takes a long time.
gc may kick in too often.  Also, when a service has restarted and a lot
of new TCP conn requests come in, it creates pressure on the tree by inserting
a lot of RTF_CACHE in a short time and it currently requires a write lock
to do that.

The first few patches are prep works to remove assumption that the
returned rt is always RTF_CACHE.

The patch 'ipv6: Only create RTF_CACHE routes after encountering pmtu exception'
do the lazy RTF_CACHE route creation.

The following patches added percpu rt to compensate the performance loss after
doing the RTF_CACHE lazy creation.

Here is some numbers of the udpflood test.  The udpflood has been
slightly modified to have a time limit instead of count limit.

A /64 via gateway route is used for the test. Each udpflood uses 10000 dst
addresses.  The dst addresses of different udpflood processes do not overlap
with each other.

1                    16M                          15M
10                   61M                          61M
20                   65M                          62M
40                   88M                          83M

~~~ end: v1 compose message ~~~
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Create percpu rt6_info
Martin KaFai Lau [Sat, 23 May 2015 03:56:06 +0000 (20:56 -0700)]
ipv6: Create percpu rt6_info

After the patch
'ipv6: Only create RTF_CACHE routes after encountering pmtu exception',
we need to compensate the performance hit (bouncing dst->__refcnt).

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Break up ip6_rt_copy()
Martin KaFai Lau [Sat, 23 May 2015 03:56:05 +0000 (20:56 -0700)]
ipv6: Break up ip6_rt_copy()

This patch breaks up ip6_rt_copy() into ip6_rt_copy_init() and
ip6_rt_cache_alloc().

In the later patch, we need to create a percpu rt6_info copy. Hence,
refactor the common rt6_info init codes to ip6_rt_copy_init().

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Keep track of DST_NOCACHE routes in case of iface down/unregister
Martin KaFai Lau [Sat, 23 May 2015 03:56:04 +0000 (20:56 -0700)]
ipv6: Keep track of DST_NOCACHE routes in case of iface down/unregister

This patch keeps track of the DST_NOCACHE routes in a list and replaces its
dev with loopback during the iface down/unregister event.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Create RTF_CACHE clone when FLOWI_FLAG_KNOWN_NH is set
Martin KaFai Lau [Sat, 23 May 2015 03:56:03 +0000 (20:56 -0700)]
ipv6: Create RTF_CACHE clone when FLOWI_FLAG_KNOWN_NH is set

This patch always creates RTF_CACHE clone with DST_NOCACHE
when FLOWI_FLAG_KNOWN_NH is set so that the rt6i_dst is set to
the fl6->daddr.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Julian Anastasov <ja@ssi.bg>
Tested-by: Julian Anastasov <ja@ssi.bg>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Set FLOWI_FLAG_KNOWN_NH at flowi6_flags
Martin KaFai Lau [Sat, 23 May 2015 03:56:02 +0000 (20:56 -0700)]
ipv6: Set FLOWI_FLAG_KNOWN_NH at flowi6_flags

The neighbor look-up used to depend on the rt6i_gateway (if
there is a gateway) or the rt6i_dst (if it is a RTF_CACHE clone)
as the nexthop address.  Note that rt6i_dst is set to fl6->daddr
for the RTF_CACHE clone where fl6->daddr is the one used to do
the route look-up.

Now, we only create RTF_CACHE clone after encountering exception.
When doing the neighbor look-up with a route that is neither a gateway
nor a RTF_CACHE clone, the daddr in skb will be used as the nexthop.

In some cases, the daddr in skb is not the one used to do
the route look-up.  One example is in ip_vs_dr_xmit_v6() where the
real nexthop server address is different from the one in the skb.

This patch is going to follow the IPv4 approach and ask the
ip6_pol_route() callers to set the FLOWI_FLAG_KNOWN_NH properly.

In the next patch, ip6_pol_route() will honor the FLOWI_FLAG_KNOWN_NH
and create a RTF_CACHE clone.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Julian Anastasov <ja@ssi.bg>
Tested-by: Julian Anastasov <ja@ssi.bg>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Add rt6_get_cookie() function
Martin KaFai Lau [Sat, 23 May 2015 03:56:01 +0000 (20:56 -0700)]
ipv6: Add rt6_get_cookie() function

Instead of doing the rt6->rt6i_node check whenever we need
to get the route's cookie.  Refactor it into rt6_get_cookie().
It is a prep work to handle FLOWI_FLAG_KNOWN_NH and also
percpu rt6_info later.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Only create RTF_CACHE routes after encountering pmtu exception
Martin KaFai Lau [Sat, 23 May 2015 03:56:00 +0000 (20:56 -0700)]
ipv6: Only create RTF_CACHE routes after encountering pmtu exception

This patch creates a RTF_CACHE routes only after encountering a pmtu
exception.

After ip6_rt_update_pmtu() has inserted the RTF_CACHE route to the fib6
tree, the rt->rt6i_node->fn_sernum is bumped which will fail the
ip6_dst_check() and trigger a relookup.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Combine rt6_alloc_cow and rt6_alloc_clone
Martin KaFai Lau [Sat, 23 May 2015 03:55:59 +0000 (20:55 -0700)]
ipv6: Combine rt6_alloc_cow and rt6_alloc_clone

A prep work for creating RTF_CACHE on exception only.  After this
patch, the same condition (rt->rt6i_flags & (RTF_NONEXTHOP | RTF_GATEWAY))
is checked twice. This redundancy will be removed in the later patch.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Remove external dependency on rt6i_gateway and RTF_ANYCAST
Martin KaFai Lau [Sat, 23 May 2015 03:55:58 +0000 (20:55 -0700)]
ipv6: Remove external dependency on rt6i_gateway and RTF_ANYCAST

When creating a RTF_CACHE route, RTF_ANYCAST is set based on rt6i_dst.
Also, rt6i_gateway is always set to the nexthop while the nexthop
could be a gateway or the rt6i_dst.addr.

After removing the rt6i_dst and rt6i_src dependency in the last patch,
we also need to stop the caller from depending on rt6i_gateway and
RTF_ANYCAST.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Remove external dependency on rt6i_dst and rt6i_src
Martin KaFai Lau [Sat, 23 May 2015 03:55:57 +0000 (20:55 -0700)]
ipv6: Remove external dependency on rt6i_dst and rt6i_src

This patch removes the assumptions that the returned rt is always
a RTF_CACHE entry with the rt6i_dst and rt6i_src containing the
destination and source address.  The dst and src can be recovered from
the calling site.

We may consider to rename (rt6i_dst, rt6i_src) to
(rt6i_key_dst, rt6i_key_src) later.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Reviewed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Clean up ipv6_select_ident() and ip6_fragment()
Martin KaFai Lau [Sat, 23 May 2015 03:55:56 +0000 (20:55 -0700)]
ipv6: Clean up ipv6_select_ident() and ip6_fragment()

This patch changes the ipv6_select_ident() signature to return a
fragment id instead of taking a whole frag_hdr as a param to
only set the frag_hdr->identification.

It also cleans up ip6_fragment() to obtain the fragment id at the
beginning instead of using multiple "if" later to check fragment id
has been generated or not.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add PHY firmware support for T420-BT cards
Hariprasad Shenai [Fri, 22 May 2015 16:28:21 +0000 (21:58 +0530)]
cxgb4: Add PHY firmware support for T420-BT cards

Add support for flashing 10GBaseT adapter with BCM 84834 PHY and
Aquantia AQ1202 PHY.

Updating of the PHY firmware must happen before the INITIALIZE_CMD.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: add more eBPF jump torture cases
Daniel Borkmann [Fri, 22 May 2015 23:10:07 +0000 (01:10 +0200)]
test_bpf: add more eBPF jump torture cases

Add two more eBPF test cases for JITs, i.e. the second one revealed a
bug in the x86_64 JIT compiler, where only an int3 filled image from
the allocator was emitted and later wrongly set by the compiler as the
bpf_func program code since optimization pass boundary was surpassed
w/o actually emitting opcodes.

Interpreter:

  [   45.782892] test_bpf: #242 BPF_MAXINSNS: Very long jump backwards jited:0 11 PASS
  [   45.783062] test_bpf: #243 BPF_MAXINSNS: Edge hopping nuthouse jited:0 14705 PASS

After x86_64 JIT (fixed):

  [   80.495638] test_bpf: #242 BPF_MAXINSNS: Very long jump backwards jited:1 6 PASS
  [   80.495957] test_bpf: #243 BPF_MAXINSNS: Edge hopping nuthouse jited:1 17157 PASS

Reference: http://thread.gmane.org/gmane.linux.network/364729
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'amd-xgbe-next'
David S. Miller [Mon, 25 May 2015 04:13:58 +0000 (00:13 -0400)]
Merge branch 'amd-xgbe-next'

Tom Lendacky says:

====================
amd-xgbe: AMD XGBE driver updates 2015-05-22

The following patches are included in this driver update series:

- Retrieve and set an additional hardware feature setting
- Fix the initial mode/speed determination when auto-negotiation is
  disabled
- Add additional netif_dbg support to the driver

This patch series is based on net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>