openwrt/staging/blogic.git
9 years agotipc: Add Ying Xue to TIPC maintainers list
Jon Paul Maloy [Mon, 9 Mar 2015 18:44:13 +0000 (14:44 -0400)]
tipc: Add Ying Xue to TIPC maintainers list

We remove Allan Stephens, who has moved on to other tasks, from
the TIPC maintainers list. He is replaced by Ying Xue, who has
been doing the maintenance on behalf of WindRiver since almost
three years.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Remove protocol from struct dst_ops
Eric W. Biederman [Mon, 9 Mar 2015 18:14:37 +0000 (13:14 -0500)]
net: Remove protocol from struct dst_ops

After my change to neigh_hh_init to obtain the protocol from the
neigh_table there are no more users of protocol in struct dst_ops.
Remove the protocol field from dst_ops and all of it's initializers.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.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/net...
David S. Miller [Mon, 9 Mar 2015 20:04:53 +0000 (16:04 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/net-next

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-03-09

This series contains updates to i40e and i40evf.

Greg cleans up some "hello world" strings which were left around from
early development.

Shannon modifies the drive to make sure the sizeof() calls are taking
the size of the actual struct that we care about.  Also updates the
NVMUpdate read/write so that it is less noisy when logging.  This was
because the NVMUpdate tool does not necessarily know the ReadOnly map of
the current NVM image, and must try reading and writing words that may be
protected.  This generates an error out of the Firmware request that the
driver logs.  Unfortunately, this ended up spitting out hundreds of
bogus read/write error messages.  If a user wants the noisy logging,
the change can be overridden by enabling the NVM update debugging.

Mitch fixes a possible deadlock issue where if a reset occurred when the
netdev is closed, the reset task will hang in napi_disable.  Added
ethtool RSS support as suggested by Ben Hutchings.

Jesse fixes a bug introduced in the force writeback code, where the
interrupt rate was set to 0 (maximum) by accident.  The driver must
correctly set the NOITR fields to avoid IT update as a side effect
of triggering the software interrupt.

I provided a simple cleanup to make the use of PF/VF consistent, which
was reported by Joe Perches.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Mon, 9 Mar 2015 19:58:21 +0000 (15:58 -0400)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for your net-next
tree. Basically, improvements for the packet rejection infrastructure,
deprecation of CLUSTERIP, cleanups for nf_tables and some untangling for
br_netfilter. More specifically they are:

1) Send packet to reset flow if checksum is valid, from Florian Westphal.

2) Fix nf_tables reject bridge from the input chain, also from Florian.

3) Deprecate the CLUSTERIP target, the cluster match supersedes it in
   functionality and it's known to have problems.

4) A couple of cleanups for nf_tables rule tracing infrastructure, from
   Patrick McHardy.

5) Another cleanup to place transaction declarations at the bottom of
   nf_tables.h, also from Patrick.

6) Consolidate Kconfig dependencies wrt. NF_TABLES.

7) Limit table names to 32 bytes in nf_tables.

8) mac header copying in bridge netfilter is already required when
   calling ip_fragment(), from Florian Westphal.

9) move nf_bridge_update_protocol() to br_netfilter.c, also from
   Florian.

10) Small refactor in br_netfilter in the transmission path, again from
    Florian.

11) Move br_nf_pre_routing_finish_bridge_slow() to br_netfilter.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Spelling: s/conceved/conceived/, s/as/a/
Geert Uytterhoeven [Mon, 9 Mar 2015 11:54:48 +0000 (12:54 +0100)]
mpls: Spelling: s/conceved/conceived/, s/as/a/

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotipc: fix inconsistent signal handling regression
Erik Hugne [Mon, 9 Mar 2015 09:43:42 +0000 (10:43 +0100)]
tipc: fix inconsistent signal handling regression

Commit 9bbb4ecc6819 ("tipc: standardize recvmsg routine") changed
the sleep/wakeup behaviour for sockets entering recv() or accept().
In this process the order of reporting -EAGAIN/-EINTR was reversed.
This caused problems with wrong errno being reported back if the
timeout expires. The same problem happens if the socket is
nonblocking and recv()/accept() is called when the process have
pending signals. If there is no pending data read or connections to
accept, -EINTR will be returned instead of -EAGAIN.

Signed-off-by: Erik Hugne <erik.hugne@ericsson.com>
Reviewed-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Reported-by László Benedek <laszlo.benedek@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'iwlwifi-next-for-kalle-2015-03-07' of https://git.kernel.org/pub/scm/linux...
David S. Miller [Mon, 9 Mar 2015 19:38:35 +0000 (15:38 -0400)]
Merge tag 'iwlwifi-next-for-kalle-2015-03-07' of https://git./linux/kernel/git/iwlwifi/iwlwifi-next

fix compilation when DEBUGFS isn't set

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: use gpl variant of symbol export
Jiri Pirko [Mon, 9 Mar 2015 09:26:24 +0000 (10:26 +0100)]
switchdev: use gpl variant of symbol export

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Andy Gospodarek <gospo@cumulusnetworks.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotipc: sparse: fix htons conversion warnings
Erik Hugne [Mon, 9 Mar 2015 09:19:31 +0000 (10:19 +0100)]
tipc: sparse: fix htons conversion warnings

Commit d0f91938bede ("tipc: add ip/udp media type") introduced
some new sparse warnings. Clean them up.

Signed-off-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: destroy proto tp when all filters are gone
Cong Wang [Fri, 6 Mar 2015 19:47:59 +0000 (11:47 -0800)]
net_sched: destroy proto tp when all filters are gone

Kernel automatically creates a tp for each
(kind, protocol, priority) tuple, which has handle 0,
when we add a new filter, but it still is left there
after we remove our own, unless we don't specify the
handle (literally means all the filters under
the tuple). For example this one is left:

  # tc filter show dev eth0
  filter parent 8001: protocol arp pref 49152 basic

The user-space is hard to clean up these for kernel
because filters like u32 are organized in a complex way.
So kernel is responsible to remove it after all filters
are gone.  Each type of filter has its own way to
store the filters, so each type has to provide its
way to check if all filters are gone.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim<jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoi40e: add ethtool RSS support
Mitch A Williams [Thu, 5 Mar 2015 04:14:40 +0000 (04:14 +0000)]
i40e: add ethtool RSS support

Add support for setting the RSS hash table and hash key through ethtool.
This patch incorporates suggestions from Ben Hutchings
<ben.hutchings@codethink.co.uk>.

Reported-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Bump i40e/i40evf version
Catherine Sullivan [Fri, 27 Feb 2015 09:18:37 +0000 (09:18 +0000)]
i40e/i40evf: Bump i40e/i40evf version

Bump PF version to 1.2.37 and VF version to 1.2.25

Change-ID: I0287a750408250dc055c03e1f744fd5f0caefd68
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: add MAC printing to debugfs dump VSI
Shannon Nelson [Fri, 27 Feb 2015 09:18:36 +0000 (09:18 +0000)]
i40e: add MAC printing to debugfs dump VSI

Print the LAN, SAN, and Port MACs for the VSI if debugfs command
dump VSI is used on the PF's VSI.

Example output:
[260221.871244] i40e 0000:04:00.0: MAC address: 68:05:ca:26:15:e0 SAN MAC: 00:00:00:00:02:00 Port MAC: 68:05:ca:26:15:e3

Change-ID: I0b393113dfb5ee7ff4f9e5227e4177885f0cc15e
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agonetfilter: bridge: move DNAT helper to br_netfilter
Pablo Neira Ayuso [Mon, 9 Mar 2015 11:30:12 +0000 (12:30 +0100)]
netfilter: bridge: move DNAT helper to br_netfilter

Only one caller, there is no need to keep this in a header.
Move it to br_netfilter.c where this belongs to.

Based on patch from Florian Westphal.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
9 years agonetfilter: bridge: refactor conditional in br_nf_dev_queue_xmit
Florian Westphal [Wed, 4 Mar 2015 23:52:36 +0000 (00:52 +0100)]
netfilter: bridge: refactor conditional in br_nf_dev_queue_xmit

simpilifies followup patch that re-works brnf ip_fragment handling.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
9 years agonetfilter: bridge: move nf_bridge_update_protocol to where its used
Florian Westphal [Wed, 4 Mar 2015 23:52:34 +0000 (00:52 +0100)]
netfilter: bridge: move nf_bridge_update_protocol to where its used

no need to keep it in a header file.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
9 years agobridge: move mac header copying into br_netfilter
Florian Westphal [Wed, 4 Mar 2015 23:52:33 +0000 (00:52 +0100)]
bridge: move mac header copying into br_netfilter

The mac header only has to be copied back into the skb for
fragments generated by ip_fragment(), which only happens
for bridge forwarded packets with nf-call-iptables=1 && active nf_defrag.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
9 years agoi40e: Fix inconsistent use of PF/VF vs pf/vf
Jeff Kirsher [Fri, 27 Feb 2015 09:18:34 +0000 (09:18 +0000)]
i40e: Fix inconsistent use of PF/VF vs pf/vf

Joe Perches pointed out that we were inconsistent in the use of
PF vs pf or VF vs vf in our driver code.  Since acronyms are usually
capitalized to denote that it is an acronym, changed all references to
be consistent throughout the code.

Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: tame the nvmupdate read and write complaints
Shannon Nelson [Fri, 27 Feb 2015 09:18:33 +0000 (09:18 +0000)]
i40e: tame the nvmupdate read and write complaints

The NVMUpdate tool doesn't necessarily know the ReadOnly map of the current
NVM image, and must try reading and writing words that may be protected.
This generates an error out of the Firmware request that the driver logs.
Unfortunately, this ends up spitting out hundreds of bogus read and write
error message that looks rather messy.

This patch checks the error type and under normal conditions will not print
the typical read and write errors during NVMUpdate.  This can be overridden
by enabling the NVM update debugging.  This results in a much less messy log
file, and likely many fewer customer support questions.

Change-ID: Id4ff2e9048c523b0ff503aa5ab181b025ec948ea
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: fix accidental write to ITR registers
Jesse Brandeburg [Fri, 27 Feb 2015 09:18:32 +0000 (09:18 +0000)]
i40e/i40evf: fix accidental write to ITR registers

Fix a bug introduced in the force writeback code, where the interrupt
rate was set to 0 (maximum) by accident.

The driver must correctly set the NOITR fields to avoid ITR update
as a side effect of triggering the software interrupt.

Change-ID: I290851ae04ef3811c43aab5ee33242029f26c1a3
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40vf: don't stop me now
Mitch Williams [Fri, 27 Feb 2015 09:18:31 +0000 (09:18 +0000)]
i40vf: don't stop me now

If a reset occurs when the netdev is closed, the reset task will hang in
napi_disable, causing deadlocks and general grumpiness.

Check to make sure the device is actually running before stopping
everything. This allows the reset task to complete and have a real good
time.

Change-ID: Iaaea84acbcb9b3810c216b14c3326e4287b75b58
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: future proof some sizeof calls
Shannon Nelson [Fri, 27 Feb 2015 09:18:30 +0000 (09:18 +0000)]
i40e: future proof some sizeof calls

Make sure the sizeof() calls are taking the size of the actual struct
that we care about.  By using the pointer variable, we'll always get
the right struct size, even if the variable type changes sometime in
the future.

Change-ID: Id5858f883cf42447365ea3733080d7714f975bce
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Remove "hello world" strings from i40e driver
Greg Rose [Fri, 27 Feb 2015 09:18:29 +0000 (09:18 +0000)]
i40e: Remove "hello world" strings from i40e driver

While using the Linux "strings" command I found these two strings in the
driver.  There's no need for them and they're kinda silly.

Change-ID: I4e19b02983d48b631e9a9979f49790492845f221
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agonet/macb: Update DT bindings documentation
Boris BREZILLON [Sat, 7 Mar 2015 06:37:06 +0000 (07:37 +0100)]
net/macb: Update DT bindings documentation

Add missing "cdns,at91sam9260-macb", "atmel,sama5d3-gem" and
"atmel,sama5d4-gem" compatible strings.

Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoethernet: codespell comment spelling fixes
Joe Perches [Sat, 7 Mar 2015 04:49:12 +0000 (20:49 -0800)]
ethernet: codespell comment spelling fixes

To test a checkpatch spelling patch, I ran codespell against
drivers/net/ethernet/.

$ git ls-files drivers/net/ethernet/ | \
  while read file ; do \
    codespell -w $file; \
  done

I removed a false positive in e1000_hw.h

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: stmmac: make reset control an optional requirement
Dinh Nguyen [Fri, 6 Mar 2015 23:48:28 +0000 (17:48 -0600)]
net: stmmac: make reset control an optional requirement

Not having a reset control line to the ethernet controller should not be a
hard failure. Instead, add support for deferred probing and just print out
a debug statement.

Signed-off-by: Dinh Nguyen <dinguyen@opensource.altera.com>
Cc: Vince Bridgers <vbridger@opensource.altera.com>
CC: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mpls-next'
David S. Miller [Sun, 8 Mar 2015 23:30:31 +0000 (19:30 -0400)]
Merge branch 'mpls-next'

Eric W. Biederman says:

====================
mpls: Minor fixes and cleanups

This is a bunch of small changes that have come out of the discussions
of the mpls code and the automated tests that people run against things.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoneigh: Use neigh table index for neigh_packet_xmit
Eric W. Biederman [Sat, 7 Mar 2015 22:25:56 +0000 (16:25 -0600)]
neigh: Use neigh table index for neigh_packet_xmit

Remove a little bit of unnecessary work when transmitting a packet with
neigh_packet_xmit.  Use the neighbour table index not the address family
as a parameter.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Fix the openvswitch select of NET_MPLS_GSO
Eric W. Biederman [Sat, 7 Mar 2015 22:24:23 +0000 (16:24 -0600)]
mpls: Fix the openvswitch select of NET_MPLS_GSO

Fix the OPENVSWITCH Kconfig option and old Kconfigs by having
OPENVSWITCH select both NET_MPLS_GSO and MPLSO.

A Kbuild test robot reported that when NET_MPLS_GSO is selected by
OPENVSWITCH the generated .config is broken because MPLS is not
selected.

Cc: Simon Horman <horms@verge.net.au>
Fixes: cec9166ca4e mpls: Refactor how the mpls module is built
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Correct the ttl decrement.
Eric W. Biederman [Sat, 7 Mar 2015 22:23:23 +0000 (16:23 -0600)]
mpls: Correct the ttl decrement.

According to RFC3032 section 2.4.2  packets with an outgoing
ttl of 0 MUST NOT be forwarded.  According to section 2.4.1
an outgoing TTL of 0 comes from an incomming TTL <= 1.

Therefore any packets that is received with a ttl <= 1 should
not have it's ttl decremented and forwarded.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Better error code for unsupported option.
Eric W. Biederman [Sat, 7 Mar 2015 22:22:40 +0000 (16:22 -0600)]
mpls: Better error code for unsupported option.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Cleanup the rcu usage in the code.
Eric W. Biederman [Sat, 7 Mar 2015 22:21:56 +0000 (16:21 -0600)]
mpls: Cleanup the rcu usage in the code.

Sparse was generating a lot of warnings mostly from missing annotations
in the code.  Add missing annotations and in a few cases tweak the code
for performance by moving work before loops.

This also fixes a problematic ommision of rcu_assign_pointer and
rcu_dereference.

Hopefully with complete rcu annotations any new rcu errors will stick
out like a sore thumb.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Fix the kzalloc argument order in mpls_rt_alloc
Eric W. Biederman [Sat, 7 Mar 2015 22:19:41 +0000 (16:19 -0600)]
mpls: Fix the kzalloc argument order in mpls_rt_alloc

*Blink* I got the argument order wrong to kzalloc and the
code was working properly when tested. *Blink*

Fix that.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoneterion: remove reference to ifconfig
stephen hemminger [Sun, 8 Mar 2015 18:48:49 +0000 (11:48 -0700)]
neterion: remove reference to ifconfig

Remove reference to obsolete ifconfig command.
MTU can be changed with ip command instead.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.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/jkirsher/net...
David S. Miller [Sun, 8 Mar 2015 03:36:12 +0000 (22:36 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/net-next

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-03-07

This series contains updates to i40e and i40evf only.

Most notably, Greg provides the patch to remove the dreaded configfs
changes in the driver.

Shannon cleans up a sparse warning by simply straighting out the code
so it is less convoluted.  Fixes an issue where the vector allocation
was trying too hard to save vectors for VMDq, to the point of not giving
the PF enough when in a tight situation, such as an NPAR partition.
Changed the driver to make sure that the PF will get all the queues and
vectors it wants to fill out its destiny.  Cleans up reporting to only
print the port and VEB stats if it is the first partition of a
multiplexed port.

Catherine cleans up some duplicated code by simply removing the duplicate
code.

Kamil cleans up the driver by removing an un-needed endian conversion
because it is already done by a register read function.

Jesse fixes a variable width of a datatype, where a u16 should have been
a u32.  Also cleans up debug_read_register() to resolve some sparse
warnings.  Updates the driver to use prefetch() to get the next Tx
descriptor, like in ixgbe, to improve performance.

Akeem moves around code to enable/disable loopback so that other non-SRIOV
supported driver functions can take advantage of the changes.

Anjali cleans up the logging for adding/deleting FD-SB filters, since
ethtool shows all the filters on an interface.  Updates the driver to
use l4_tunnel type generically to keep code flow simple.  Simplifies
the RSS code since the driver initializes the rss_size_max in sw_init.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6352: Add support for EEE
Guenter Roeck [Sat, 7 Mar 2015 06:23:52 +0000 (22:23 -0800)]
net: dsa: mv88e6352: Add support for EEE

Enable EEE support for MV88E6352.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: Add EEE support
Guenter Roeck [Sat, 7 Mar 2015 06:23:51 +0000 (22:23 -0800)]
net: dsa: mv88e6xxx: Add EEE support

EEE configuration is similar for the various MV88E6xxx chips.
Add generic support for it.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: bcmgenet: rework Rx queue init
Petri Gynther [Fri, 6 Mar 2015 21:45:00 +0000 (13:45 -0800)]
net: bcmgenet: rework Rx queue init

In preparation for supporting multiple Rx queues:
1. Move the initialization of priv->num_rx_bds, priv->rx_bds, and
   priv->rx_cbs from bcmgenet_init_rx_ring() to bcmgenet_init_dma()
   since they are not specific to a single Rx queue. Mimics the Tx
   init model where priv->num_tx_bds, priv->tx_bds, and priv->tx_cbs
   are initialized in bcmgenet_init_dma().
2. Program DMA_MBUF_DONE_THRESH = 1 so that future Rx queues Q0-Q15
   will get per-packet Rx interrupt.
3. Group DMA_START_ADDR, RDMA_READ_PTR, RDMA_WRITE_PTR, and DMA_END_ADDR
   initialization together. Mimics the Tx init model.
4. There is 1-to-1 mapping between RxCBs and RxBDs.
   Precalculate RxCB->bd_addr so that it can be used in the future.

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'macb-next'
David S. Miller [Sun, 8 Mar 2015 03:31:56 +0000 (22:31 -0500)]
Merge branch 'macb-next'

Merge branch 'macb-next'

Boris Brezillon says:

====================
net/macb: merge at91_ether driver into macb driver

The rm9200 boards use the dedicated at91_ether driver instead of the
regular macb driver.

Both the macb and at91_ether drivers can be compiled as separated
modules.
Since the at91_ether driver uses code from the macb driver, at91_ether.ko
depends on macb.ko.

However the macb.ko module always fails to load on rm9200 boards: the
macb_probe() function expects a hclk clock which doesn't exist on rm9200.
Then the at91_ether.ko can't be loaded in turn due to unresolved
dependencies.

This series of patches fix this issue by merging at91_ether into macb.

Patch 1 is fixing a problem that might happen when enabling ARM
multi-platform suppot.

Changes since v3:
- move "net: macb: remove #if defined(CONFIG_ARCH_AT91) sections" patch
  into this series to avoid dependency on other patch series.

Changes since v2:
- rebase after changed brought by commit "net: macb: remove #if
  defined(CONFIG_ARCH_AT91) sections"

Changes since v1:
- rework probe functions to share common probing logic
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/macb: merge at91_ether driver into macb driver
Cyrille Pitchen [Sat, 7 Mar 2015 06:23:32 +0000 (07:23 +0100)]
net/macb: merge at91_ether driver into macb driver

macb and at91_ether drivers can be compiled as modules, but the at91_ether
driver use some functions and variables defined in the macb one, thus
creating a dependency on the macb driver.

Since these drivers are sharing the same logic we can easily merge
at91_ether into macb.

In order to factorize common probing logic we've added an ->init() function
to struct macb_config (the structure associated with the compatible
string), and moved macb specific init code from macb_probe to macb_init.

Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Tested-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/macb: unify clock management
Cyrille Pitchen [Sat, 7 Mar 2015 06:23:31 +0000 (07:23 +0100)]
net/macb: unify clock management

Most of the functions from the Common Clk Framework handle NULL pointer as
input argument.

Since the TX clock is optional, we now set tx_clk to NULL value
instead of ERR_PTR(-ENOENT) when this clock is not available. This simplifies
the clock management and avoid the need to test tx_clk value.

Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Acked-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: macb: remove #if defined(CONFIG_ARCH_AT91) sections
Boris BREZILLON [Sat, 7 Mar 2015 06:23:30 +0000 (07:23 +0100)]
net: macb: remove #if defined(CONFIG_ARCH_AT91) sections

With multi platform support those sections could lead to unexpected
behavior if both ARCH_AT91 and another ARM SoC using the MACB IP are
selected.
Add two new capabilities to encode the default MII mode and the presence
of a CLKEN bit in USRIO register.
Then define the appropriate config for IPs embedded in at91 SoCs.

Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoARM: at91/dt: fix macb compatible strings
Boris BREZILLON [Sat, 7 Mar 2015 06:23:29 +0000 (07:23 +0100)]
ARM: at91/dt: fix macb compatible strings

Some at91 SoCs embed a 10/100 Mbit Ethernet IP, that is based on the
at91sam9260 SoC.
Fix at91 DTs accordingly.

Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoiwlwifi: mvm: fix compilation with IWLWIFI_DEBUGFS not set
Emmanuel Grumbach [Sat, 7 Mar 2015 17:35:37 +0000 (19:35 +0200)]
iwlwifi: mvm: fix compilation with IWLWIFI_DEBUGFS not set

The commits below broke compilation when
CONFIG_IWLWIFI_DEBUGFS is not set.
FIx that.

Fixes: ddf89ab10a93 ("iwlwifi: mvm: allow to force the Rx chains from debugfs")
Fixes: 9d761fd8a583 ("iwlwifi: mvm: add trigger for firmware dump upon missed beacons")
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoi40e: Strip configfs code
Greg Rose [Fri, 6 Mar 2015 01:41:07 +0000 (01:41 +0000)]
i40e: Strip configfs code

The use of configfs is not allowed in network drivers.  Strip the code that
uses it.

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/i40evf: Bump version
Sravanthi Tangeda [Fri, 27 Feb 2015 09:15:33 +0000 (09:15 +0000)]
i40e/i40evf: Bump version

Bump i40e to 1.2.12 and i40evf to 1.2.6.

Change-ID: I641871da3a9abd396b28eda5744a4d68493c1400
Signed-off-by: Sravanthi Tangeda <sravanthi.tangeda@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: enable prefetch of Tx descriptors during cleanup
Jesse Brandeburg [Fri, 27 Feb 2015 09:15:31 +0000 (09:15 +0000)]
i40e/i40evf: enable prefetch of Tx descriptors during cleanup

Performance can be improved a bit by imitating ixgbe and using
prefetch to get us the next Tx descriptor.

Change-ID: Ice7ffd4cd0ce87c35295059bdb7972a7f53723aa
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Simplify code for rss_size_max config
Anjali Singhai Jain [Fri, 27 Feb 2015 09:15:30 +0000 (09:15 +0000)]
i40e: Simplify code for rss_size_max config

We initialize the pf->rss_size_max in sw_init now
and hence this code can be simplified.

Change-ID: I1a7abc837604a40bc65e6c6b21190b909ed6bb21
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Simplify tunnel selection logic
Anjali Singhai Jain [Fri, 27 Feb 2015 09:15:29 +0000 (09:15 +0000)]
i40e/i40evf: Simplify tunnel selection logic

Use l4_tunnel type generically to keep code flow simple.

Change-ID: Ic52287e3b1ca4204e6b6e13431890c1a6ae9c422
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: FD filters flush policy changes
Anjali Singhai Jain [Fri, 27 Feb 2015 09:15:28 +0000 (09:15 +0000)]
i40e: FD filters flush policy changes

Since GLQF_FDCNT_0 register now has the right offset, use it to simplify our
FD flush flow.
If the filter add error happens to be for SB we just auto disable SB.

If filter error happens to be for ATR, auto disable ATR and mark
the state to FD_FLUSH_REQUESTED. Which gets cleared when flush completes.

If we are entering flush too quickly (< 30 seconds) and we have quite
a few SB rules, its time to disable ATR for good. Since SB + ATR rules
is most likely making the FD table unstable.

ATR can be re-enabled by turning ntuple off (ethtool -K ntuple off)
and will remain off after turning ntuple on till it gets unstable again.

Change-ID: I2154a2e0a5d44851a2f0eb8731e2f1d4a4d1acbc
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Avoid logs while adding/deleting FD-SB filters
Anjali Singhai Jain [Fri, 27 Feb 2015 09:15:27 +0000 (09:15 +0000)]
i40e: Avoid logs while adding/deleting FD-SB filters

It is not necessary to print FD filter add/delete log with
normal debug settings because ethtool -n ethx shows all the FD-SB
filters on an interface. The log can still be turned on through higher
debug levels and it will continue to print a log if there was an error
in the add/delete process.

Change-ID: I67db2baf49e2075d2f537de40f7895e5b02cd610
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: print port stats only on partition 1
Shannon Nelson [Fri, 27 Feb 2015 09:15:26 +0000 (09:15 +0000)]
i40e: print port stats only on partition 1

Only print the port and veb stats if this is the first partition
of a multiplexed port.

Change-ID: I7ce0c323cdee5cfd2e54d8bea5b0b9102987e671
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Move code to enable/disable Loopback to the main file
Akeem G Abodunrin [Fri, 27 Feb 2015 09:15:25 +0000 (09:15 +0000)]
i40e: Move code to enable/disable Loopback to the main file

Since changes made to enable or disable loopback for all VSIs, not only SR-IOV
or PCIOV, then it became necessary to move the associated functions to main
file - so that other non-SRIOV supported driver can take advantage of the
changes.

Change-ID: I59a49fd23a6136acda5e16f8d1e5ac7fd9c5fc05
Signed-off-by: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: rework vector reservation
Shannon Nelson [Fri, 27 Feb 2015 09:15:24 +0000 (09:15 +0000)]
i40e: rework vector reservation

The initial problem solved here is that the vector allocation was trying
too hard to save vectors for VMDq, to the point of not giving the PF enough
when in a tight situation such as an NPAR partition.  This change makes
sure that the PF will get all the queues and vectors it wants to fill
out its destiny.  Essentially, nothing is specially reserved for VMDq,
it simply gets whatever is left after the PF, FCoE, and FD sideband get
what they want.

Additionally, the calculations for the reservations were harder to follow
than necessary, so I've made it more straight forward.

Change-ID: I99b384f104535b686c690b8ef0a787559485c8d4
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: clean up debug_read_register
Jesse Brandeburg [Fri, 27 Feb 2015 09:15:23 +0000 (09:15 +0000)]
i40e: clean up debug_read_register

There were some additional spaces and strange (double swapping) logic
in this function that I started looking at because sparse was warning.

This fixes the sparse warning and fixes up the other issues.

Change-ID: I72a91a4197cd45921602649040e6bd25e5f17c0a
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: store msg_enable in the right size
Jesse Brandeburg [Fri, 27 Feb 2015 09:15:22 +0000 (09:15 +0000)]
i40e: store msg_enable in the right size

The kernel returns a u32 for netif_msg_init, and we were storing
it in a u16.  Fix the width of the datatype.

Change-ID: I4b23326e5707c91cd59325c5a1ccb2ba7a3974fc
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Remove unneeded conversion
Kamil Krawczyk [Fri, 27 Feb 2015 09:15:21 +0000 (09:15 +0000)]
i40e: Remove unneeded conversion

Remove LE16 to CPU endianes conversion from i40e_read_nvm_word_srctl
function, as it's already done by register read function.

Change-ID: I739f0f20a9b8e18223e54c0ca5443e63d75da878
Signed-off-by: Kamil Krawczyk <kamil.krawczyk@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Remove duplicate code
Catherine Sullivan [Fri, 27 Feb 2015 09:15:20 +0000 (09:15 +0000)]
i40e: Remove duplicate code

This series of code was repeated twice, remove one of them.

Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Refactor i40e_debug_aq and make some functions static
Shannon Nelson [Fri, 27 Feb 2015 09:15:19 +0000 (09:15 +0000)]
i40e/i40evf: Refactor i40e_debug_aq and make some functions static

A sparse complaint in i40e_debug_aq in a funky buffer write goes away by
straightening out the code out to something less convoluted.

Also fix some other sparse warnings while we are at it, making some
functions static and using NULL instead of 0.

Change-ID: I93907534fe1f1f675830774b3d14ecf1c6ffc9a0
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agorocker: sparse: fix dynamic allocation on stack warning
Scott Feldman [Fri, 6 Mar 2015 23:54:52 +0000 (15:54 -0800)]
rocker: sparse: fix dynamic allocation on stack warning

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: quiet sparce endianess warnings
Scott Feldman [Fri, 6 Mar 2015 23:54:51 +0000 (15:54 -0800)]
rocker: quiet sparce endianess warnings

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Reviewed-by: Jonathan Toppins <jtoppins@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoebpf: bpf_map_*: fix linker error on avr32 and openrisc arch
Daniel Borkmann [Thu, 5 Mar 2015 22:27:51 +0000 (23:27 +0100)]
ebpf: bpf_map_*: fix linker error on avr32 and openrisc arch

Fengguang reported, that on openrisc and avr32 architectures, we
get the following linker errors on *_defconfig builds that have
no bpf syscall support:

  net/built-in.o:(.rodata+0x1cd0): undefined reference to `bpf_map_lookup_elem_proto'
  net/built-in.o:(.rodata+0x1cd4): undefined reference to `bpf_map_update_elem_proto'
  net/built-in.o:(.rodata+0x1cd8): undefined reference to `bpf_map_delete_elem_proto'

Fix it up by providing built-in weak definitions of the symbols,
so they can be overridden when the syscall is enabled. I think
the issue might be that gcc is not able to optimize all that away.
This patch fixes the linker errors for me, tested with Fengguang's
make.cross [1] script.

  [1] https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Fixes: d4052c4aea0c ("ebpf: remove CONFIG_BPF_SYSCALL ifdefs in socket filter code")
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 agonet: gro: remove obsolete code from skb_gro_receive()
Eric Dumazet [Thu, 5 Mar 2015 21:47:48 +0000 (13:47 -0800)]
net: gro: remove obsolete code from skb_gro_receive()

Some drivers use copybreak to copy tiny frames into smaller skb,
and this smaller skb might not have skb->head_frag set for various
reasons.

skb_gro_receive() currently doesn't allow to aggregate the smaller skb
into the previous GRO packet if this GRO packet has at least 2 MSS in
it.

Following workload easily demonstrates the problem.

netperf -t TCP_RR -H target -- -r 3000,3000

(tcpdump shows one GRO packet with 2 MSS, plus one additional packet of
104 bytes that should have been appended.)

It turns out that we can remove code from skb_gro_receive(), because
commit 8a29111c7ca6 ("net: gro: allow to build full sized skb") and its
followups removed the assumption that a GRO packet with a frag_list had
to have an empty head.

Removing this code allows the aggregation of the last (incomplete) frame
in some RPC workloads. Note that tcp_gro_receive() already takes care of
forcing a flush if necessary, including this case.

If we want to avoid using frag_list in the first place (in forwarding
workloads for example, as the outgoing NIC is generally not able to cope
with skbs having a frag_list), we need to address this separately.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoax25: remove unneeded NULL test in ax_xmit()
Dan Carpenter [Thu, 5 Mar 2015 17:48:46 +0000 (20:48 +0300)]
ax25: remove unneeded NULL test in ax_xmit()

We get a static checker warning here on devel kernels:

drivers/net/hamradio/mkiss.c:560 ax_xmit()
warn: variable dereferenced before check 'skb' (see line 532)

It turns out that the NULL check can be deleted.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mlx4-qcn'
David S. Miller [Sat, 7 Mar 2015 02:50:07 +0000 (21:50 -0500)]
Merge branch 'mlx4-qcn'

Or Gerlitz says:

====================
Add QCN support to the DCB NL layer

This series from Shani Michaeli adds support for the IEEE QCN attribute
to the kernel DCB NL stack, and implementation in the mlx4 driver which
programs the firmware according to the admin directives.

changes from V0:

 - applied feedback from John and added his acked-by to patch #1
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_en: Add QCN parameters and statistics handling
Shani Michaeli [Thu, 5 Mar 2015 18:16:13 +0000 (20:16 +0200)]
net/mlx4_en: Add QCN parameters and statistics handling

Implement the IEEE DCB handlers for set/get QCN parameters and
statistics reading per TC.

Signed-off-by: Shani Michaeli <shanim@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_core: Add basic elements for QCN
Shani Michaeli [Thu, 5 Mar 2015 18:16:12 +0000 (20:16 +0200)]
net/mlx4_core: Add basic elements for QCN

Add device capability, firmware command opcode and etc prior elements
needed for QCN suppprt. Disable SRIOV VF view/access for QCN is disabled.

While here, remove a redundant offset definition into the
QUERY_DEV_CAP mailbox.

Signed-off-by: Shani Michaeli <shanim@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/dcb: Add IEEE QCN attribute
Shani Michaeli [Thu, 5 Mar 2015 18:16:11 +0000 (20:16 +0200)]
net/dcb: Add IEEE QCN attribute

As specified in 802.1Qau spec. Add this optional attribute to the
DCB netlink layer. To allow for application to use the new attribute,
NIC drivers should implement and register the  callbacks ieee_getqcn,
ieee_setqcn and ieee_getqcnstats.

The QCN attribute holds a set of parameters for management, and
a set of statistics to provide informative data on Congestion-Control
defined by this spec.

Signed-off-by: Shani Michaeli <shanim@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'fib_trie-next'
David S. Miller [Fri, 6 Mar 2015 20:49:35 +0000 (15:49 -0500)]
Merge branch 'fib_trie-next'

Alexander Duyck says:

====================
The rest of the FIB patches (add key_vector to fib_table)

This patch series is the rest of what I had originally planned for this kernel
release.  It adds a structure called key_vector which is embedded within
every tnode, leaf, and the trie root itself.  By doing this we can navigate
from any point within the trie to any other point fairly quickly and
avoiding NULL pointer checks in the case of a backtrace.  As a result we
can pipeline things a bit further since we don't have to worry about
dereferencing NULL in a backtrace.  This can amount to significant savings
on a long backtrace.

I decided to drop the up-level code as that conflicts with combining the
main and local tries.  I have one patch as an RFC that currently combines
the tries however it still needs some work as we have to split the local
and main tries in the event of custom rules being defined.  As such we are
probably going to be doing some more hacking on fib_table_flush_external as
that will also need to flush the local entries from the main trie and place
them back in the local trie.

v2: Rebased on the switchdev FIB offload work
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Add key vector to root, return parent key_vector in resize
Alexander Duyck [Fri, 6 Mar 2015 17:54:52 +0000 (09:54 -0800)]
fib_trie: Add key vector to root, return parent key_vector in resize

This change makes it so that the root of the trie contains a key_vector, by
doing this we make room to essentially collapse the entire trie by at least
one cache line as we can store the information about the tnode or leaf that
is pointed to in the root.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Move parent from key_vector to tnode
Alexander Duyck [Fri, 6 Mar 2015 17:54:46 +0000 (09:54 -0800)]
fib_trie: Move parent from key_vector to tnode

This change pulls the parent pointer from the key_vector and places it in
the tnode structure.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Pull empty_children and full_children into tnode
Alexander Duyck [Fri, 6 Mar 2015 17:54:39 +0000 (09:54 -0800)]
fib_trie: Pull empty_children and full_children into tnode

This pulls the information about the child array out of the key_vector and
places it in the tnode since that is where it is needed.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Move rcu from key_vector to tnode, add accessors.
Alexander Duyck [Fri, 6 Mar 2015 17:54:33 +0000 (09:54 -0800)]
fib_trie: Move rcu from key_vector to tnode, add accessors.

RCU is only needed once for the entire node, not once per key_vector so we
can pull that out and move it to the tnode structure.

In addition add accessors to be used inside the RCU functions so that we
can more easily get from the key vector to either the tnode or the trie
pointers.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Add tnode struct as a container for fields not needed in key_vector
Alexander Duyck [Fri, 6 Mar 2015 17:54:27 +0000 (09:54 -0800)]
fib_trie: Add tnode struct as a container for fields not needed in key_vector

This change pulls the fields not explicitly needed in the key_vector and
placed them in the new tnode structure.  By doing this we will eventually
be able to reduce the key_vector down to 16 bytes on 64 bit systems, and
12 bytes on 32 bit systems.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Rename tnode_child_length to child_length
Alexander Duyck [Fri, 6 Mar 2015 17:54:21 +0000 (09:54 -0800)]
fib_trie: Rename tnode_child_length to child_length

We are now checking the length of a key_vector instead of a tnode so it
makes sense to probably just rename this to child_length since it would
probably even be applicable to a leaf.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: replace tnode_get_child functions with get_child macros
Alexander Duyck [Fri, 6 Mar 2015 17:54:14 +0000 (09:54 -0800)]
fib_trie: replace tnode_get_child functions with get_child macros

I am replacing the tnode_get_child call with get_child since we are
techically pulling the child out of a key_vector now and not a tnode.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Rename tnode to key_vector
Alexander Duyck [Fri, 6 Mar 2015 17:54:08 +0000 (09:54 -0800)]
fib_trie: Rename tnode to key_vector

Rename the tnode to key_vector.  The key_vector will be the eventual
container for all of the information needed by either a leaf or a tnode.
The final result should be much smaller than the 40 bytes currently needed
for either one.

This also updates the trie struct so that it contains an array of size 1 of
tnode pointers.  This is to bring the structure more inline with how an
actual tnode itself is configured.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Return pointer to tnode pointer in resize/inflate/halve
Alexander Duyck [Fri, 6 Mar 2015 17:54:02 +0000 (09:54 -0800)]
fib_trie: Return pointer to tnode pointer in resize/inflate/halve

Resize related functions now all return a pointer to the pointer that
references the object that was resized.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Minor cleanups to fib_table_flush_external
Alexander Duyck [Fri, 6 Mar 2015 17:53:56 +0000 (09:53 -0800)]
fib_trie: Minor cleanups to fib_table_flush_external

This change just does a couple of minor cleanups on
fib_table_flush_external.  Specifically it addresses the fact that resize
was being called even though nothing was being removed from the table, and
it drops an unecessary indent since we could just call continue on the
inverse of the fi && flag check.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: macb: Fix multi queue support for xilinx ZynqMP
Punnaiah Choudary Kalluri [Fri, 6 Mar 2015 17:29:12 +0000 (18:29 +0100)]
net: macb: Fix multi queue support for xilinx ZynqMP

ZynqMP soc has single interrupt for all the queue events. So,
passing the IRQF_SHARED flag for interrupt registration call.

Signed-off-by: Punnaiah Choudary Kalluri <punnaia@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: macb: Include multi queue support for xilinx ZynqMP ethernet version
Punnaiah Choudary Kalluri [Fri, 6 Mar 2015 17:29:11 +0000 (18:29 +0100)]
net: macb: Include multi queue support for xilinx ZynqMP ethernet version

Include multi queue support for the ethernet IP version in xilinx ZynqMP
SoC.

Signed-off-by: Punnaiah Choudary Kalluri <punnaia@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'wireless-drivers-next-for-davem-2015-03-06' of git://git.kernel.org/pub...
David S. Miller [Fri, 6 Mar 2015 20:46:08 +0000 (15:46 -0500)]
Merge tag 'wireless-drivers-next-for-davem-2015-03-06' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Major changes:

brcmfmac:

* sdio improvements
* add a debugfs file so users can provide us all the revinfo we could
  ask for

iwlwifi:

* add triggers for firmware dump collection
* remove support for -9.ucode
* new statitics API
* rate control improvements

ath9k:

* add per-vif TX power capability
* BT coexistance fixes

ath10k:

* qca6174: enable STA transmit beamforming (TxBF) support
* disable multi-vif power save by default

bcma:

* enable support for PCIe Gen 2 host devices

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib: make netdev_switch_fib_ipv4_abort in header file static inline
Willem de Bruijn [Fri, 6 Mar 2015 16:44:28 +0000 (11:44 -0500)]
fib: make netdev_switch_fib_ipv4_abort in header file static inline

When building without CONFIG_NET_SWITCHDEV,
netdev_switch_fib_ipv4_abort is defined in the header file. It must
be static inline to avoid build failure at link time.

Fixes: 8e05fd7166c6 ("fib: hook IPv4 fib for hardware offload")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Properly validate RTA_VIA payload length
Robert Shearman [Fri, 6 Mar 2015 10:47:00 +0000 (10:47 +0000)]
mpls: Properly validate RTA_VIA payload length

If the nla length is less than 2 then the nla data could be accessed
beyond the accessible bounds. So ensure that the nla is big enough to
at least read the via_family before doing so. Replace magic value of
2.

Fixes: 03c0566542f4 ("mpls: Basic support for adding and removing routes")
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'bcmgenet-next'
David S. Miller [Fri, 6 Mar 2015 20:13:47 +0000 (15:13 -0500)]
Merge branch 'bcmgenet-next'

Petri Gynther says:

====================
net: bcmgenet: preparation for multiple Rx queues

Three small patches in preparation for supporting multiple Rx queues:
1. set hw_params->rx_queues = 0
2. adjust the call to alloc_etherdev_mqs()
3. add GENET_Q16_RX_BD_CNT and hw_params->rx_bds_per_q
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: bcmgenet: add GENET_Q16_RX_BD_CNT and hw_params->rx_bds_per_q
Petri Gynther [Fri, 6 Mar 2015 01:40:14 +0000 (17:40 -0800)]
net: bcmgenet: add GENET_Q16_RX_BD_CNT and hw_params->rx_bds_per_q

In preparation for supporting multiple Rx queues, add GENET_Q16_RX_BD_CNT
and hw_params->rx_bds_per_q.

Signed-off-by: Petri Gynther <pgynther@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: bcmgenet: adjust the call to alloc_etherdev_mqs()
Petri Gynther [Fri, 6 Mar 2015 01:40:12 +0000 (17:40 -0800)]
net: bcmgenet: adjust the call to alloc_etherdev_mqs()

In preparation for supporting multiple Rx queues, adjust the call to
alloc_etherdev_mqs() to allow max GENET_MAX_MQ_CNT + 1 Rx queues.

The actual number of Rx queues in use is correctly adjusted with:
netif_set_real_num_rx_queues(priv->dev, priv->hw_params->rx_queues + 1);

Signed-off-by: Petri Gynther <pgynther@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: bcmgenet: set hw_params->rx_queues = 0
Petri Gynther [Fri, 6 Mar 2015 01:40:10 +0000 (17:40 -0800)]
net: bcmgenet: set hw_params->rx_queues = 0

bcmgenet driver doesn't yet support multiple Rx queues.
Set hw_params->rx_queues = 0 accordingly.
The default Rx queue (Q16) is still created and operational.

Signed-off-by: Petri Gynther <pgynther@google.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/net...
David S. Miller [Fri, 6 Mar 2015 20:01:07 +0000 (15:01 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/net-next

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-03-06

This series contains updates to e1000, e1000e and igb.

Yanir provides updates to e1000e based on the patches provided by John
Linville.  First updates the code comment to better describe the changes
and the impact on the driver.  Second removed calls to ioremap/unmap for
i219 since this is only relevant to older hardware only.  Starting with
i219, the NVM will not be mapped to its one BAR but to a address region
in another bar.

Alex Duyck provides two fixes for igb, first fixes a compile warning
where a variable may be used uninitialized, so Alex initializes it.
Second fixes an issue where all of the pin register values were having
to be pushed onto the stack each time the function was called, so to
avoid this, Alex made them static const so that they should only need
to be allocated once and we can avoid all the instructions to get them
onto the stack.

Eliezer found an issue in e1000 where we needed to be calling
netif_carrier_off earlier in the down() to prevent the stack from
queuing more packets to the interface.

Sabrina Dubroca resolved a potential race condition by adding a
dummy allocator.  There was a race condition between e1000_change_mtu()
cleanups and netpoll, when changing the MTU across jumbo sizes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'pmtu-probe'
David S. Miller [Fri, 6 Mar 2015 19:57:46 +0000 (14:57 -0500)]
Merge branch 'pmtu-probe'

Fan Du says:

====================
Improvements for TCP PMTU

This patchset performs some improvements and enhancement
for current TCP PMTU as per RFC4821 with the aim to find
optimal mms size quickly, and also be adaptive to route
changes like enlarged path MTU. Then TCP PMTU could be
used to probe a effective pmtu in absence of ICMP message
for tunnels(e.g. vxlan) across different networking stack.

Patch1/4: Set probe mss base to 1024 Bytes per RFC4821
Patch2/4: Do not double probe_size for each probing,
          use a simple binary search to gain maximum performance.
  mss for next probing.
Patch3/4: Create a probe timer to detect enlarged path MTU.
Patch4/4: Update ip-sysctl.txt for new sysctl knobs.

Changelog:
v5:
  - Zero probe_size before resetting search range.
  - Update ip-sysctl.txt for new sysctl knobs.
v4:
  - Convert probe_size to mss, not directly from search_low/high
  - Clamp probe_threshold
  - Don't adjust search_high in blackhole probe, so drop orignal patch3
v3:
  - Update commit message for patch2
  - Fix pseudo timer delta calculation in patch4
v2:
  - Introduce sysctl_tcp_probe_threshold to control when
    probing will stop, as suggested by John Heffner.
  - Add patch3 to shrink current mss value for search low boundary.
  - Drop cannonical timer usages, implements pseudo timer based on
    32bits jiffies tcp_time_stamp, as suggested by Eric Dumazet.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: Documenting two sysctls for tcp PMTU probe
Fan Du [Fri, 6 Mar 2015 03:18:25 +0000 (11:18 +0800)]
ipv4: Documenting two sysctls for tcp PMTU probe

Namely tcp_probe_interval to control how often to restart
a probe. And tcp_probe_threshold to control when stop the
probing in respect to the width of search range in bytes

Signed-off-by: Fan Du <fan.du@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: Create probe timer for tcp PMTU as per RFC4821
Fan Du [Fri, 6 Mar 2015 03:18:24 +0000 (11:18 +0800)]
ipv4: Create probe timer for tcp PMTU as per RFC4821

As per RFC4821 7.3.  Selecting Probe Size, a probe timer should
be armed once probing has converged. Once this timer expired,
probing again to take advantage of any path PMTU change. The
recommended probing interval is 10 minutes per RFC1981. Probing
interval could be sysctled by sysctl_tcp_probe_interval.

Eric Dumazet suggested to implement pseudo timer based on 32bits
jiffies tcp_time_stamp instead of using classic timer for such
rare event.

Signed-off-by: Fan Du <fan.du@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: Use binary search to choose tcp PMTU probe_size
Fan Du [Fri, 6 Mar 2015 03:18:23 +0000 (11:18 +0800)]
ipv4: Use binary search to choose tcp PMTU probe_size

Current probe_size is chosen by doubling mss_cache,
the probing process will end shortly with a sub-optimal
mss size, and the link mtu will not be taken full
advantage of, in return, this will make user to tweak
tcp_base_mss with care.

Use binary search to choose probe_size in a fine
granularity manner, an optimal mss will be found
to boost performance as its maxmium.

In addition, introduce a sysctl_tcp_probe_threshold
to control when probing will stop in respect to
the width of search range.

Test env:
Docker instance with vxlan encapuslation(82599EB)
iperf -c 10.0.0.24  -t 60

before this patch:
1.26 Gbits/sec

After this patch: increase 26%
1.59 Gbits/sec

Signed-off-by: Fan Du <fan.du@intel.com>
Acked-by: John Heffner <johnwheffner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: Raise tcp PMTU probe mss base size
Fan Du [Fri, 6 Mar 2015 03:18:22 +0000 (11:18 +0800)]
ipv4: Raise tcp PMTU probe mss base size

Quotes from RFC4821 7.2.  Selecting Initial Values

   It is RECOMMENDED that search_low be initially set to an MTU size
   that is likely to work over a very wide range of environments.  Given
   today's technologies, a value of 1024 bytes is probably safe enough.
   The initial value for search_low SHOULD be configurable.

Moreover, set a small value will introduce extra time for the search
to converge. So set the initial probe base mss size to 1024 Bytes.

Signed-off-by: Fan Du <fan.du@intel.com>
Acked-by: John Heffner <johnwheffner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoDECnet: Only use neigh_ops for adding the link layer header
Eric W. Biederman [Wed, 4 Mar 2015 16:16:43 +0000 (10:16 -0600)]
DECnet: Only use neigh_ops for adding the link layer header

Other users users of the neighbour table use neigh->output as the method
to decided when and which link-layer header to place on a packet.
DECnet has been using neigh->output to decide which DECnet headers to
place on a packet depending which neighbour the packet is destined for.

The DECnet usage isn't totally wrong but it can run into problems if the
neighbour output function is run for a second time as the teql driver
and the bridge netfilter code can do.

Therefore to avoid pathologic problems later down the line and make the
neighbour code easier to understand by refactoring the decnet output
code to only use a neighbour method to add a link layer header to a
packet.

This is done by moving the neigbhour operations lookup from
dn_to_neigh_output to dn_neigh_output_packet.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobonding: implement bond_poll_controller()
Mahesh Bandewar [Thu, 5 Mar 2015 05:57:52 +0000 (21:57 -0800)]
bonding: implement bond_poll_controller()

This patches implements the poll_controller support for all
bonding driver. If the slaves have poll_controller net_op defined,
this implementation calls them. This is mode agnostic implementation
and iterates through all slaves (based on mode) and calls respective
handler.

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: fix some sparse warnings
Scott Feldman [Fri, 6 Mar 2015 09:14:37 +0000 (01:14 -0800)]
rocker: fix some sparse warnings

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: fix CONFIG_IP_MULTIPLE_TABLES compile issue
Scott Feldman [Fri, 6 Mar 2015 09:14:36 +0000 (01:14 -0800)]
switchdev: fix CONFIG_IP_MULTIPLE_TABLES compile issue

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoe1000: add dummy allocator to fix race condition between mtu change and netpoll
Sabrina Dubroca [Thu, 26 Feb 2015 05:35:41 +0000 (05:35 +0000)]
e1000: add dummy allocator to fix race condition between mtu change and netpoll

There is a race condition between e1000_change_mtu's cleanups and
netpoll, when we change the MTU across jumbo size:

Changing MTU frees all the rx buffers:
    e1000_change_mtu -> e1000_down -> e1000_clean_all_rx_rings ->
        e1000_clean_rx_ring

Then, close to the end of e1000_change_mtu:
    pr_info -> ... -> netpoll_poll_dev -> e1000_clean ->
        e1000_clean_rx_irq -> e1000_alloc_rx_buffers -> e1000_alloc_frag

And when we come back to do the rest of the MTU change:
    e1000_up -> e1000_configure -> e1000_configure_rx ->
        e1000_alloc_jumbo_rx_buffers

alloc_jumbo finds the buffers already != NULL, since data (shared with
page in e1000_rx_buffer->rxbuf) has been re-alloc'd, but it's garbage,
or at least not what is expected when in jumbo state.

This results in an unusable adapter (packets don't get through), and a
NULL pointer dereference on the next call to e1000_clean_rx_ring
(other mtu change, link down, shutdown):

BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff81194d6e>] put_compound_page+0x7e/0x330

    [...]

Call Trace:
 [<ffffffff81195445>] put_page+0x55/0x60
 [<ffffffff815d9f44>] e1000_clean_rx_ring+0x134/0x200
 [<ffffffff815da055>] e1000_clean_all_rx_rings+0x45/0x60
 [<ffffffff815df5e0>] e1000_down+0x1c0/0x1d0
 [<ffffffff811e2260>] ? deactivate_slab+0x7f0/0x840
 [<ffffffff815e21bc>] e1000_change_mtu+0xdc/0x170
 [<ffffffff81647050>] dev_set_mtu+0xa0/0x140
 [<ffffffff81664218>] do_setlink+0x218/0xac0
 [<ffffffff814459e9>] ? nla_parse+0xb9/0x120
 [<ffffffff816652d0>] rtnl_newlink+0x6d0/0x890
 [<ffffffff8104f000>] ? kvm_clock_read+0x20/0x40
 [<ffffffff810a2068>] ? sched_clock_cpu+0xa8/0x100
 [<ffffffff81663802>] rtnetlink_rcv_msg+0x92/0x260

By setting the allocator to a dummy version, netpoll can't mess up our
rx buffers.  The allocator is set back to a sane value in
e1000_configure_rx.

Fixes: edbbb3ca1077 ("e1000: implement jumbo receive with partial descriptors")
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000: call netif_carrier_off early on down
Eliezer Tamir [Wed, 25 Feb 2015 15:52:49 +0000 (15:52 +0000)]
e1000: call netif_carrier_off early on down

When bringing down an interface netif_carrier_off() should be
one the first things we do, since this will prevent the stack
from queuing more packets to this interface.
This operation is very fast, and should make the device behave
much nicer when trying to bring down an interface under load.

Also, this would Do The Right Thing (TM) if this device has some
sort of fail-over teaming and redirect traffic to the other IF.

Move netif_carrier_off as early as possible.

Signed-off-by: Eliezer Tamir <eliezer.tamir@linux.intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>