GitHub/LineageOS/G12/android_kernel_amlogic_linux-4.9.git
10 years agohyperv: Add IPv6 into the hash computation for vRSS
Haiyang Zhang [Thu, 30 Oct 2014 21:07:17 +0000 (14:07 -0700)]
hyperv: Add IPv6 into the hash computation for vRSS

This will allow the workload spreading via vRSS for IPv6.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoip6_tunnel: allow to change mode for the ip6tnl0
Alexey Andriyanov [Wed, 29 Oct 2014 07:54:52 +0000 (10:54 +0300)]
ip6_tunnel: allow to change mode for the ip6tnl0

The fallback device is in ipv6 mode by default.
The mode can not be changed in runtime, so there
is no way to decapsulate ip4in6 packets coming from
various sources without creating the specific tunnel
ifaces for each peer.

This allows to update the fallback tunnel device, but only
the mode could be changed. Usual command should work for the
fallback device: `ip -6 tun change ip6tnl0 mode any`

The fallback device can not be hidden from the packet receiver
as a regular tunnel, but there is no need for synchronization
as long as we do single assignment.

Cc: David S. Miller <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexey Andriyanov <alan@al-an.info>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: remove assignment in if condition
Fabian Frederick [Wed, 29 Oct 2014 11:57:51 +0000 (12:57 +0100)]
ipv6: remove assignment in if condition

Do assignment before if condition and test !skb like in rawv6_recvmsg()

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: remove inline on static in c file
Fabian Frederick [Wed, 29 Oct 2014 10:38:17 +0000 (11:38 +0100)]
ipv6: remove inline on static in c file

remove __inline__ / inline and let compiler decide what to do
with static functions

Inspired-by: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: spelling s/incomming/incoming
Fabian Frederick [Wed, 29 Oct 2014 09:00:26 +0000 (10:00 +0100)]
ipv6: spelling s/incomming/incoming

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipx: remove all unnecessary castings on ntohl
Fabian Frederick [Wed, 29 Oct 2014 08:31:42 +0000 (09:31 +0100)]
ipx: remove all unnecessary castings on ntohl

Apply commit e0f36310f793
("ipx: remove unnecessary casting on ntohl")
to all seq_printf/08lX

Inspired-by: "David S. Miller" <davem@davemloft.net>
Inspired-by: Joe Perches <joe@perches.com>
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'bpf-next'
David S. Miller [Thu, 30 Oct 2014 19:45:01 +0000 (15:45 -0400)]
Merge branch 'bpf-next'

Alexei Starovoitov says:

====================
bpf: reduce verifier memory consumption and add tests

Small set of cleanups:
 - reduce verifier memory consumption
 - add verifier test to check register state propagation and state equivalence
 - add JIT test reduced from recent nmap triggered crash
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotest: bpf: add a testcase reduced from nmap
Alexei Starovoitov [Tue, 28 Oct 2014 22:11:43 +0000 (15:11 -0700)]
test: bpf: add a testcase reduced from nmap

nmap generates classic BPF programs to filter ARP packets with given target MAC
which triggered a bug in eBPF x64 JIT. The bug was fixed in
commit e0ee9c12157d ("x86: bpf_jit: fix two bugs in eBPF JIT compiler")
This patch is adding a testcase in eBPF instructions (those that
were generated by classic->eBPF converter) to be processed by JIT.
The test is primarily targeting JIT compiler.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosamples: bpf: add a verifier test and summary line
Alexei Starovoitov [Tue, 28 Oct 2014 22:11:42 +0000 (15:11 -0700)]
samples: bpf: add a verifier test and summary line

- add a test specifically targeting verifier state pruning.
It checks state propagation between registers, storing that
state into stack and state pruning algorithm recognizing
equivalent stack and register states.

- add summary line to spot failures easier

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobpf: reduce verifier memory consumption
Alexei Starovoitov [Tue, 28 Oct 2014 22:11:41 +0000 (15:11 -0700)]
bpf: reduce verifier memory consumption

verifier keeps track of register state spilled to stack.
registers are 8-byte wide and always aligned, so instead of tracking them
in every byte-sized stack slot, use MAX_BPF_STACK / 8 array to track
spilled register state.
Though verifier runs in user context and its state freed immediately
after verification, it makes sense to reduce its memory usage.
This optimization reduces sizeof(struct verifier_state)
from 12464 to 1712 on 64-bit and from 6232 to 1112 on 32-bit.

Note, this patch doesn't change existing limits, which are there to bound
time and memory during verification: 4k total number of insns in a program,
1k number of jumps (states to visit) and 32k number of processed insn
(since an insn may be visited multiple times). Theoretical worst case memory
during verification is 1712 * 1k = 17Mbyte. Out-of-memory situation triggers
cleanup and rejects the program.

Suggested-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'dsa-next'
David S. Miller [Thu, 30 Oct 2014 18:54:21 +0000 (14:54 -0400)]
Merge branch 'dsa-next'

Guenter Roeck says:

====================
net: dsa: Fixes and enhancements

Patch 01/15 addresses a bug indicated by an an annoying and unhelpful
log message.

Patches 02/15 and 03/15 are minor enhancements, adding support for
known switch revisions.

Patches 04/15 and 05/15 add support for MV88E6352 and MV88E6176.

Patch 06/15 adds support for hardware monitoring, specifically for
reporting the chip temperature, to the dsa subsystem.

Patches 07/15 and 08/15 implement hardware monitoring for MV88E6352,
MV88E6176, MV88E6123, MV88E6161, and MV88E6165.

Patch 09/15 and 10/15 add support for EEPROM access to the DSA subsystem.

Patch 11/15 implements EEPROM access for MV88E6352 and MV88E6176.

Patch 12/15 adds support for reading switch registers to the DSA
subsystem.

Patches 13/15 amd 14/15 implement support for reading switch registers
to the drivers for MV88E6352, MV88E6176, MV88E6123, MV88E6161, and MV88E6165.

Patch 15/15 adds support for reading additional RMON registers to the drivers
for  MV88E6352, MV88E6176, MV88E6123, MV88E6161, and MV88E6165.

The series was tested on top of v3.18-rc2 in an x86 system with MV88E6352.
Testing in systems with 88E613188E6060 and MV88E6165 was done earlier
(I don't have access to those systems right now). The series was also build
tested using my build system at http://server.roeck-us.net:8010/builders.
Look into the 'dsa' column for build results.

The series merges cleanly into net-next as of today (10/29).

v3:
- Fix bug in eeprom patches seen if devicetree is enabled:
  eeprom-length property is attached to switch devicetree node,
  not to dsa node, and there was a compile error.
v2:
- Made reporting chip temperatures through the hwmon subsystem optional
  with new Kconfig option
- Changed the hwmon chip name to <network device name>_dsa<index>
- Made EEPROM presence and size configurable through platform and devicetree
  data
- Various minor changes and fixes (see individual patches for details)
====================

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Provide additional RMON statistics
Guenter Roeck [Wed, 29 Oct 2014 17:45:07 +0000 (10:45 -0700)]
net: dsa: Provide additional RMON statistics

Display sw_in_discards, sw_in_filtered, and sw_out_filtered for chips
supported by mv88e6123_61_65 and mv88e6352 drivers.

The variables are provided in port registers, not the normal status registers.
Mark by adding 0x100 to the register offset and add special handling code
to mv88e6xxx_get_ethtool_stats.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6352: Add support for reading switch registers
Guenter Roeck [Wed, 29 Oct 2014 17:45:06 +0000 (10:45 -0700)]
net: dsa/mv88e6352: Add support for reading switch registers

Report switch register values to ethtool.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6123_61_65: Add support for reading switch registers
Guenter Roeck [Wed, 29 Oct 2014 17:45:05 +0000 (10:45 -0700)]
net: dsa/mv88e6123_61_65: Add support for reading switch registers

The infrastructure can now report switch registers to ethtool.
Add support for it to the mv88e6123_61_65 driver.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Add support for reading switch registers with ethtool
Guenter Roeck [Wed, 29 Oct 2014 17:45:04 +0000 (10:45 -0700)]
net: dsa: Add support for reading switch registers with ethtool

Add support for reading switch registers with 'ethtool -d'.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6352: Implement EEPROM access functions
Guenter Roeck [Wed, 29 Oct 2014 17:45:03 +0000 (10:45 -0700)]
net: dsa/mv88e6352: Implement EEPROM access functions

MV88E6352 supports read and write access to its configuration eeprom.

There is no means to detect if an EEPROM is connected to the switch.
Also, the switch supports EEPROMs with different sizes, but can not detect
or report the type or size of connected EEPROMs. Therefore, do not implement
the get_eeprom_len callback but depend on platform or devicetree data to
provide information about EEPROM presence and size.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodsa: Add new optional devicetree property to describe EEPROM size
Guenter Roeck [Wed, 29 Oct 2014 17:45:02 +0000 (10:45 -0700)]
dsa: Add new optional devicetree property to describe EEPROM size

The dsa core now supports reading from and writing to a switch EEPROM
if connected. Describe optional devicetree property indicating that
an EEPROM is present and its size.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Add support for switch EEPROM access
Guenter Roeck [Wed, 29 Oct 2014 17:45:01 +0000 (10:45 -0700)]
net: dsa: Add support for switch EEPROM access

On some chips it is possible to access the switch eeprom.
Add infrastructure support for it.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6123_61_65: Report chip temperature
Guenter Roeck [Wed, 29 Oct 2014 17:45:00 +0000 (10:45 -0700)]
net: dsa/mv88e6123_61_65: Report chip temperature

MV88E6123 and compatible chips support reading the chip temperature
from PHY register 6:26.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6352: Report chip temperature
Guenter Roeck [Wed, 29 Oct 2014 17:44:59 +0000 (10:44 -0700)]
net: dsa/mv88e6352: Report chip temperature

MV88E6352 supports reading the chip temperature from two PHY registers,
6:26 and 6:27. Report it using the more accurate register 6:27.
Also report temperature limit and alarm.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Add support for reporting switch chip temperatures
Guenter Roeck [Wed, 29 Oct 2014 17:44:58 +0000 (10:44 -0700)]
net: dsa: Add support for reporting switch chip temperatures

Some switches provide chip temperature data.
Add support for reporting it through the hwmon subsystem.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa/mv88e6352: Add support for MV88E6176
Guenter Roeck [Wed, 29 Oct 2014 17:44:57 +0000 (10:44 -0700)]
net: dsa/mv88e6352: Add support for MV88E6176

MV88E6176 is mostly compatible to MV88E6352 and is documented
in the same functional specification. Add support for it.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Add support for Marvell 88E6352
Guenter Roeck [Wed, 29 Oct 2014 17:44:56 +0000 (10:44 -0700)]
net: dsa: Add support for Marvell 88E6352

Marvell 88E6352 is mostly compatible to MV88E6123/61/65,
but requires indirect phy access. Also, its configuration
registers are a bit different.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Report known silicon revisions for Marvell 88E6131
Guenter Roeck [Wed, 29 Oct 2014 17:44:55 +0000 (10:44 -0700)]
net: dsa: Report known silicon revisions for Marvell 88E6131

Report known silicon revisions when probing Marvell 88E6131 switches.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Report known silicon revisions for Marvell 88E6060
Guenter Roeck [Wed, 29 Oct 2014 17:44:54 +0000 (10:44 -0700)]
net: dsa: Report known silicon revisions for Marvell 88E6060

Report known silicon revisions when probing Marvell 88E6060 switches.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: dsa: Don't set skb->protocol on outgoing tagged packets
Guenter Roeck [Wed, 29 Oct 2014 17:44:53 +0000 (10:44 -0700)]
net: dsa: Don't set skb->protocol on outgoing tagged packets

Setting skb->protocol to a private protocol type may result in warning
messages such as
e1000e 0000:00:19.0 em1: checksum_partial proto=dada!

This happens if the L3 protocol is IP or IPv6 and skb->ip_summed is set
to CHECKSUM_PARTIAL. Looking through the code, it appears that changing
skb->protocol for transmitted packets is not necessary and may actually
be harmful. For example, it prevents purposely unmodified (from a DSA
perspective) network drivers from properly setting up their transmit
checksum offload pointers since they inspect skb->protocol to set up the
IPv4 header or IPv6 header pointers. So don't unnecessarily change the
protocol field.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoneigh: optimize neigh_parms_release()
Nicolas Dichtel [Wed, 29 Oct 2014 18:29:31 +0000 (19:29 +0100)]
neigh: optimize neigh_parms_release()

In neigh_parms_release() we loop over all entries to find the entry given in
argument and being able to remove it from the list. By using a double linked
list, we can avoid this loop.

Here are some numbers with 30 000 dummy interfaces configured:

Before the patch:
$ time rmmod dummy
real 2m0.118s
user 0m0.000s
sys 1m50.048s

After the patch:
$ time rmmod dummy
real 1m9.970s
user 0m0.000s
sys 0m47.976s

Suggested-by: Thierry Herbelot <thierry.herbelot@6wind.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: introduce napi_schedule_irqoff()
Eric Dumazet [Wed, 29 Oct 2014 01:05:13 +0000 (18:05 -0700)]
net: introduce napi_schedule_irqoff()

napi_schedule() can be called from any context and has to mask hard
irqs.

Add a variant that can only be called from hard interrupts handlers
or when irqs are already masked.

Many NIC drivers can use it from their hard IRQ handler instead of
generic variant.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'xen-netback-next'
David S. Miller [Wed, 29 Oct 2014 19:59:43 +0000 (15:59 -0400)]
Merge branch 'xen-netback-next'

David Vrabel says:

====================
xen-netback: minor cleanups

Two minor xen-netback cleanups originally from Zoltan.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoxen-netback: Remove __GFP_COLD
Zoltan Kiss [Tue, 28 Oct 2014 15:29:31 +0000 (15:29 +0000)]
xen-netback: Remove __GFP_COLD

This flag is unnecessary, it came from some old code.

Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoxen-netback: Disable NAPI after disabling interrupts
Zoltan Kiss [Tue, 28 Oct 2014 15:29:30 +0000 (15:29 +0000)]
xen-netback: Disable NAPI after disabling interrupts

Otherwise the interrupt handler still calls napi_complete. Although it
won't schedule NAPI again as either NAPI_STATE_DISABLE or
NAPI_STATE_SCHED is set, it is just unnecessary, and it makes more
sense to do this way.

Signed-off-by: Zoltan Kiss <zoltan.kiss@linaro.org>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: ipv6: Add a sysctl to make optimistic addresses useful candidates
Erik Kline [Tue, 28 Oct 2014 09:11:14 +0000 (18:11 +0900)]
net: ipv6: Add a sysctl to make optimistic addresses useful candidates

Add a sysctl that causes an interface's optimistic addresses
to be considered equivalent to other non-deprecated addresses
for source address selection purposes.  Preferred addresses
will still take precedence over optimistic addresses, subject
to other ranking in the source address selection algorithm.

This is useful where different interfaces are connected to
different networks from different ISPs (e.g., a cell network
and a home wifi network).

The current behaviour complies with RFC 3484/6724, and it
makes sense if the host has only one interface, or has
multiple interfaces on the same network (same or cooperating
administrative domain(s), but not in the multiple distinct
networks case.

For example, if a mobile device has an IPv6 address on an LTE
network and then connects to IPv6-enabled wifi, while the wifi
IPv6 address is undergoing DAD, IPv6 connections will try use
the wifi default route with the LTE IPv6 address, and will get
stuck until they time out.

Also, because optimistic nodes can receive frames, issue
an RTM_NEWADDR as soon as DAD starts (with the IFA_F_OPTIMSTIC
flag appropriately set).  A second RTM_NEWADDR is sent if DAD
completes (the address flags have changed), otherwise an
RTM_DELADDR is sent.

Also: add an entry in ip-sysctl.txt for optimistic_dad.

Signed-off-by: Erik Kline <ek@google.com>
Acked-by: Lorenzo Colitti <lorenzo@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'r8152-next'
David S. Miller [Wed, 29 Oct 2014 19:09:16 +0000 (15:09 -0400)]
Merge branch 'r8152-next'

Hayes Wang says:

====================
r8152: support nway_reset

Fix the CHECK from checkpatch.pl and support nway_reset.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agor8152: support nway_reset of ethtool
hayeswang [Tue, 28 Oct 2014 06:05:52 +0000 (14:05 +0800)]
r8152: support nway_reset of ethtool

Support the nway_reset() function for ethtool.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agor8152: rename tx_underun
hayeswang [Tue, 28 Oct 2014 06:05:51 +0000 (14:05 +0800)]
r8152: rename tx_underun

Replace tx_underun with tx_underrun for checkpatch.pl.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotcp: allow for bigger reordering level
Eric Dumazet [Tue, 28 Oct 2014 04:45:24 +0000 (21:45 -0700)]
tcp: allow for bigger reordering level

While testing upcoming Yaogong patch (converting out of order queue
into an RB tree), I hit the max reordering level of linux TCP stack.

Reordering level was limited to 127 for no good reason, and some
network setups [1] can easily reach this limit and get limited
throughput.

Allow a new max limit of 300, and add a sysctl to allow admins to even
allow bigger (or lower) values if needed.

[1] Aggregation of links, per packet load balancing, fabrics not doing
 deep packet inspections, alternative TCP congestion modules...

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yaogong Wang <wygivan@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: ethernet: realtek: atp: checkpatch errors and warnings corrected
Roberto Medina [Mon, 27 Oct 2014 23:51:56 +0000 (00:51 +0100)]
net: ethernet: realtek: atp: checkpatch errors and warnings corrected

Several warnings and errors of coding style rules corrected.
Compile tested.

Signed-off-by: Roberto Medina <robertoxmed@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: skb_segment() should preserve backpressure
Toshiaki Makita [Mon, 27 Oct 2014 17:30:51 +0000 (10:30 -0700)]
net: skb_segment() should preserve backpressure

This patch generalizes commit d6a4a1041176 ("tcp: GSO should be TSQ
friendly") to protocols using skb_set_owner_w()

TCP uses its own destructor (tcp_wfree) and needs a more complex scheme
as explained in commit 6ff50cd55545 ("tcp: gso: do not generate out of
order packets")

This allows UDP sockets using UFO to get proper backpressure,
thus avoiding qdisc drops and excessive cpu usage.

Here are performance test results (macvlan on vlan):

- Before
# netperf -t UDP_STREAM ...
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   60.00      144096 1224195    1258.56
212992           60.00          51              0.45

Average:        CPU     %user     %nice   %system   %iowait    %steal     %idle
Average:        all      0.23      0.00     25.26      0.08      0.00     74.43

- After
# netperf -t UDP_STREAM ...
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   60.00      109593      0     957.20
212992           60.00      109593            957.20

Average:        CPU     %user     %nice   %system   %iowait    %steal     %idle
Average:        all      0.18      0.00      8.38      0.02      0.00     91.43

[edumazet] Rewrote patch and changelog.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agogianfar: Implement PAUSE frame generation support
Matei Pavaluca [Mon, 27 Oct 2014 08:42:44 +0000 (10:42 +0200)]
gianfar: Implement PAUSE frame generation support

The hardware can automatically generate pause frames when the number
of free buffers drops under a certain threshold, but in order to do this,
the address of the last free buffer needs to be written to a specific
register for each RX queue.

This has to be done in 'gfar_clean_rx_ring' which is called for each
RX queue. In order not to impact performance, by adding a register write
for each incoming packet, this operation is done only when the PAUSE frame
transmission is enabled.

Whenever the link is readjusted, this capability is turned on or off.

Signed-off-by: Matei Pavaluca <matei.pavaluca@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoFix the way the local advertising flow options are determined
Pavaluca Matei-B46610 [Mon, 27 Oct 2014 08:42:43 +0000 (10:42 +0200)]
Fix the way the local advertising flow options are determined

Local flow control options needed in order to resolve the negotiation
are incorrectly calculated.

Previously 'mii_advertise_flowctrl' was called to determine the local advertising
options, but these were determined based on FLOW_CTRL_RX/TX flags which are
never set through ethtool.
The patch simply translates from ethtool flow options to mii flow options.

Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAdd flow control support flags to gianfar's capabilities
Pavaluca Matei-B46610 [Mon, 27 Oct 2014 08:42:42 +0000 (10:42 +0200)]
Add flow control support flags to gianfar's capabilities

The phy device supports 802.3x flow control, but the specific flags are not set
in the phy initialisation code. Flow control flags need to be added to the
supported capabilities of the phydev by the driver.

This is needed in order for ethtool to work ('ethtool -A' code checks for these
flags)

Signed-off-by: Pavaluca Matei <matei.pavaluca@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoopenvswitch: Export lockdep_ovsl_is_held to modules.
David S. Miller [Tue, 28 Oct 2014 21:27:23 +0000 (17:27 -0400)]
openvswitch: Export lockdep_ovsl_is_held to modules.

ERROR: "lockdep_ovsl_is_held" [net/openvswitch/vport-gre.ko] undefined!

Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'mellanox-next'
David S. Miller [Tue, 28 Oct 2014 21:18:07 +0000 (17:18 -0400)]
Merge branch 'mellanox-next'

Amir Vadai says:

====================
Mellanox ethernet driver update Oct-27-2014

This patchset introduces some small bug fixes, support in get/set of
vlan offload and get/set/capabilities of the link.

First 7 patches by Saeed, add support in setting/getting link speed and getting
cable capabilities.
Next 2 patches also by Saeed, enable the user to turn rx/tx vlan offloading on
and off.
Jenni fixed a bug in error flow during device initalization.
Ido and Jack fixed some code duplication and errors discovered by static checker.
last patch by me is a fix to make ethtool report the actual rings used by
indirection QP.

Patches were applied and tested against commit 61ed53d ("Merge tag 'ntb-3.18'
of git://github.com/jonmason/ntb")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_en: Report actual number of rings in indirection table
Amir Vadai [Mon, 27 Oct 2014 09:37:47 +0000 (11:37 +0200)]
net/mlx4_en: Report actual number of rings in indirection table

Hardware requires the number of rings in indirection table to be a power
of 2. When setting number of channels to a non power of 2 number,
indirection table is using only the closest power of 2 rings.
Report this number in 'ethtool -x' and not the total number of rx rings.

Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_en: Move spinlocks and work initalizations to beginning of init_netdev
Eugenia Emantayev [Mon, 27 Oct 2014 09:37:46 +0000 (11:37 +0200)]
net/mlx4_en: Move spinlocks and work initalizations to beginning of init_netdev

Upon failures, destroy_netdev is called, and spinlocks/works must be
initialized before calling it. Otherwise kernel panic may occur.

Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_en: Call napi_synchronize on stop_port
Ido Shamay [Mon, 27 Oct 2014 09:37:45 +0000 (11:37 +0200)]
net/mlx4_en: Call napi_synchronize on stop_port

This is instead of calling the actual implementation of
napi_synchronize, for better encapsulation.

Signed-off-by: Ido Shamay <idos@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_en: Cleanups suggested by clang static checker
Jack Morgenstein [Mon, 27 Oct 2014 09:37:44 +0000 (11:37 +0200)]
net/mlx4_en: Cleanups suggested by clang static checker

clang flagged the following. All are actually cosmetic cleanups, not really bugs:

drivers/net/ethernet/mellanox/mlx4/en_main.c:233:3: warning: Value stored to 'err' is never read
                err = -ENOMEM;
                ^     ~~~~~~~
drivers/net/ethernet/mellanox/mlx4/en_main.c:293:3: warning: Value stored to 'err' is never read
                err = -ENOMEM;

drivers/net/ethernet/mellanox/mlx4/en_netdev.c:648:16: warning: Assigned value is garbage or undefined
        entry->reg_id = reg_id;
                      ^ ~~~~~~
drivers/net/ethernet/mellanox/mlx4/en_netdev.c:659:2: warning: Function call argument is an uninitialized value
        mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
(NOTE: reg_id is only used in the device-managed flow steering path, in which is it always initialized.
 This is not a bug. Cleanup here is therefore cosmetic only).

drivers/net/ethernet/mellanox/mlx4/en_rx.c:122:3: warning: Value stored to 'frag_info' is never read
                frag_info = &priv->frag_info[i];
                ^           ~~~~~~~~~~~~~~~~~~~

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_en: Add ethtool support for [rx|tx]vlan offload set to OFF/ON
Saeed Mahameed [Mon, 27 Oct 2014 09:37:43 +0000 (11:37 +0200)]
net/mlx4_en: Add ethtool support for [rx|tx]vlan offload set to OFF/ON

Move mlx4_en_reset_config to en_netdev.c as it now serves more general purpose.
Add support for turning OFF/ON the rx/tx vlan offlad.

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>
10 years agonet/mlx4_en: Add support for setting rxvlan offload OFF/ON
Saeed Mahameed [Mon, 27 Oct 2014 09:37:42 +0000 (11:37 +0200)]
net/mlx4_en: Add support for setting rxvlan offload OFF/ON

Rename mlx4_en_timestamp_config to mlx4_en_reset_config and extend it to support
choosing RX vlan offload configuration.

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>
10 years agonet/mlx4_en: Use PTYS register to set ethtool settings (Speed)
Saeed Mahameed [Mon, 27 Oct 2014 09:37:41 +0000 (11:37 +0200)]
net/mlx4_en: Use PTYS register to set ethtool settings (Speed)

Added Support to set speed or advertised link modes via ethtool:
ethtool -s <ifname> [speed <speed>] [advertise <link modes>]

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>
10 years agonet/mlx4_en: Use PTYS register to query ethtool settings
Saeed Mahameed [Mon, 27 Oct 2014 09:37:40 +0000 (11:37 +0200)]
net/mlx4_en: Use PTYS register to query ethtool settings

- If dev cap MLX4_DEV_CAP_FLAG2_ETH_PROT_CTRL is ON, query PTYS register to fill ethtool settings.
else use default values.
- Use autoneg port cap and dev backplane autoneg cap to reprort autoneg interface capbilities.
- Fix typo in mlx4_en_port_state struct field (transciver to transceiver).

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>
10 years agoethtool, net/mlx4_en: Add 100M, 20G, 56G speeds ethtool reporting support
Saeed Mahameed [Mon, 27 Oct 2014 09:37:39 +0000 (11:37 +0200)]
ethtool, net/mlx4_en: Add 100M, 20G, 56G speeds ethtool reporting support

Added 100M, 20G and 56G ethtool speed reporting support.
Update mlx4_en_test_speed self test with the new speeds.

Defined new link speeds in include/uapi/linux/ethtool.h:
+#define SPEED_20000 20000
+#define SPEED_40000 40000
+#define SPEED_56000 56000

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>
10 years agonet/mlx4_core: Add ethernet backplane autoneg device capability
Saeed Mahameed [Mon, 27 Oct 2014 09:37:38 +0000 (11:37 +0200)]
net/mlx4_core: Add ethernet backplane autoneg device capability

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>
10 years agonet/mlx4_core: Introduce ACCESS_REG CMD and eth_prot_ctrl dev cap
Saeed Mahameed [Mon, 27 Oct 2014 09:37:37 +0000 (11:37 +0200)]
net/mlx4_core: Introduce ACCESS_REG CMD and eth_prot_ctrl dev cap

Adding ACCESS REG mlx4 command and use it to implement Query method for
PTYS (Port Type and Speed Register).
Query and store eth_prot_ctrl dev cap.

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>
10 years agoethtool, net/mlx4_en: Cable info, get_module_info/eeprom ethtool support
Saeed Mahameed [Mon, 27 Oct 2014 09:37:36 +0000 (11:37 +0200)]
ethtool, net/mlx4_en: Cable info, get_module_info/eeprom ethtool support

Added support for get_module_info/get_module_eeprom ethtool support for cable info reading.

Added new cable types enum in include/uapi/linux/ethtool.h for ethtool use.
+#define ETH_MODULE_SFF_8636            0x3
+#define ETH_MODULE_SFF_8636_LEN        256
+#define ETH_MODULE_SFF_8436            0x4
+#define ETH_MODULE_SFF_8436_LEN        256

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>
10 years agonet/mlx4_core: Introduce mlx4_get_module_info for cable module info reading
Saeed Mahameed [Mon, 27 Oct 2014 09:37:35 +0000 (11:37 +0200)]
net/mlx4_core: Introduce mlx4_get_module_info for cable module info reading

Added new MAD_IFC command to read cable module info with attribute id (0xFF60).
Update include/linux/mlx4/device.h with function declaration (mlx4_get_module_info)
and the needed defines/enums for future use.

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>
10 years agodatapath: Rename last_action() as nla_is_last() and move to netlink.h
Simon Horman [Mon, 27 Oct 2014 07:12:16 +0000 (16:12 +0900)]
datapath: Rename last_action() as nla_is_last() and move to netlink.h

The original motivation for this change was to allow the helper to be used
in files other than actions.c as part of work on an odp select group
action.

It was as pointed out by Thomas Graf that this helper would be best off
living in netlink.h. Furthermore, I think that the generic nature of this
helper means it is best off in netlink.h regardless of if it is used more
than one .c file or not. Thus, I would like it considered independent of
the work on an odp select group action.

Cc: Thomas Graf <tgraf@suug.ch>
Cc: Pravin Shelar <pshelar@nicira.com>
Cc: Andy Zhou <azhou@nicira.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Thomas Graf <tgraf@noironetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: pxa168_eth: Fix providing of phy_interface mode on platform_data
Sebastian Hesselbarth [Sat, 25 Oct 2014 10:08:59 +0000 (12:08 +0200)]
net: pxa168_eth: Fix providing of phy_interface mode on platform_data

Do not add phy include to the board file but platform_data include
instead.

Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: Adding SGMII support for Marvell 88ee1145 driver
Viet Nga Dao [Fri, 24 Oct 2014 02:41:53 +0000 (19:41 -0700)]
net: phy: Adding SGMII support for Marvell 88ee1145 driver

Additional code to m88e1145_config_init function to allow the driver to
support SGMII mode.

Signed-off-by: Viet Nga Dao <vndao@altera.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoovs: Turn vports with dependencies into separate modules
Thomas Graf [Wed, 22 Oct 2014 15:29:06 +0000 (17:29 +0200)]
ovs: Turn vports with dependencies into separate modules

The internal and netdev vport remain part of openvswitch.ko. Encap
vports including vxlan, gre, and geneve can be built as separate
modules and are loaded on demand. Modules can be unloaded after use.
Datapath ports keep a reference to the vport module during their
lifetime.

Allows to remove the error prone maintenance of the global list
vport_ops_list.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'unnecessary_resource_check'
David S. Miller [Mon, 27 Oct 2014 23:16:14 +0000 (19:16 -0400)]
Merge branch 'unnecessary_resource_check'

Varka Bhadram says:

====================
cleanup on resource check

This series removes the duplication of sanity check for
platform_get_resource() return resource. It will be checked
with devm_ioremap_resource()

changes since v2:
- Merge #1 and #2 patches into single patch
- remove the comment

changes since v1:
- remove NULL dereference on resource_size()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethernet: samsung: sxgbe: remove unnecessary check
Varka Bhadram [Fri, 24 Oct 2014 02:12:10 +0000 (07:42 +0530)]
ethernet: samsung: sxgbe: remove unnecessary check

devm_ioremap_resource checks platform_get_resource() return value.
We can remove the duplicate check here.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethernet: renesas: remove unnecessary check
Varka Bhadram [Fri, 24 Oct 2014 02:12:09 +0000 (07:42 +0530)]
ethernet: renesas: remove unnecessary check

devm_ioremap_resource checks platform_get_resource() return value.
We can remove the duplicate check here.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethernet: marvell: remove unnecessary check
Varka Bhadram [Fri, 24 Oct 2014 02:12:08 +0000 (07:42 +0530)]
ethernet: marvell: remove unnecessary check

devm_ioremap_resource checks platform_get_resource() return value.
We can remove the duplicate check here.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethernet: apm: xgene: remove unnecessary check
Varka Bhadram [Fri, 24 Oct 2014 02:12:07 +0000 (07:42 +0530)]
ethernet: apm: xgene: remove unnecessary check

devm_ioremap_resource checks platform_get_resource() return value.
We can remove the duplicate check here.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethernet: wiznet: remove unnecessary check
Varka Bhadram [Fri, 24 Oct 2014 02:12:06 +0000 (07:42 +0530)]
ethernet: wiznet: remove unnecessary check

devm_ioremap_resource checks platform_get_resource() return value.
We can remove the duplicate check here.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobridge: Add support for IEEE 802.11 Proxy ARP
Kyeyoon Park [Thu, 23 Oct 2014 21:49:17 +0000 (14:49 -0700)]
bridge: Add support for IEEE 802.11 Proxy ARP

This feature is defined in IEEE Std 802.11-2012, 10.23.13. It allows
the AP devices to keep track of the hardware-address-to-IP-address
mapping of the mobile devices within the WLAN network.

The AP will learn this mapping via observing DHCP, ARP, and NS/NA
frames. When a request for such information is made (i.e. ARP request,
Neighbor Solicitation), the AP will respond on behalf of the
associated mobile device. In the process of doing so, the AP will drop
the multicast request frame that was intended to go out to the wireless
medium.

It was recommended at the LKS workshop to do this implementation in
the bridge layer. vxlan.c is already doing something very similar.
The DHCP snooping code will be added to the userspace application
(hostapd) per the recommendation.

This RFC commit is only for IPv4. A similar approach in the bridge
layer will be taken for IPv6 as well.

Signed-off-by: Kyeyoon Park <kyeyoonp@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipx: remove __inline__ in c file on static
Fabian Frederick [Mon, 27 Oct 2014 20:12:08 +0000 (21:12 +0100)]
ipx: remove __inline__ in c file on static

Let compiler decide what to do with static void __ipxitf_put()

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipx: remove unnecessary casting on ntohl
Fabian Frederick [Mon, 27 Oct 2014 19:55:08 +0000 (20:55 +0100)]
ipx: remove unnecessary casting on ntohl

use %08X instead of %08lX and remove casting.

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipx: move extern sysctl_ipx_pprop_broadcasting to header file
Fabian Frederick [Mon, 27 Oct 2014 19:00:41 +0000 (20:00 +0100)]
ipx: move extern sysctl_ipx_pprop_broadcasting to header file

include ipx.h from sysctl_net_ipx.c

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: include linux/uaccess.h instead of asm/uaccess.h
Fabian Frederick [Mon, 27 Oct 2014 18:12:58 +0000 (19:12 +0100)]
ipv6: include linux/uaccess.h instead of asm/uaccess.h

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: replace min/casting by min_t
Fabian Frederick [Mon, 27 Oct 2014 18:11:56 +0000 (19:11 +0100)]
ipv6: replace min/casting by min_t

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv4: remove set but unused variable sha
Fabian Frederick [Mon, 27 Oct 2014 18:03:22 +0000 (19:03 +0100)]
ipv4: remove set but unused variable sha

unsigned char *sha (source) was already in original git version
 but was never used.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 's390-next'
David S. Miller [Mon, 27 Oct 2014 02:21:45 +0000 (22:21 -0400)]
Merge branch 's390-next'

Frank Blaschka says:

====================
s390: network patches for net-next

looks like there was a problem with my previous posting. Hope this time
it will work. Sorry for any inconvenience. The patches are mostly
cleanups and small enhancements for net-next
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoctcm: replace sscanf by kstrto function
Thomas Richter [Wed, 22 Oct 2014 10:18:07 +0000 (12:18 +0200)]
ctcm: replace sscanf by kstrto function

Since a single integer value is read from the supplied buffer
use the kstrto functions instead of sscanf.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agolcs: replace sscanf by kstrto function
Thomas Richter [Wed, 22 Oct 2014 10:18:06 +0000 (12:18 +0200)]
lcs: replace sscanf by kstrto function

Since a single integer value is read from the supplied buffer
use the kstrto functions instead of sscanf.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: s390 ethernet device driver dependency
Thomas Richter [Wed, 22 Oct 2014 10:18:05 +0000 (12:18 +0200)]
qeth: s390 ethernet device driver dependency

Compile the s390 10GB ethernet device driver only when
ETHERNET has been defined in the kernel configuration file.
Right now the qeth device driver is always built regardless
of which network connectivity is active.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: make local functions static in qeth_l3 module
Thomas Richter [Wed, 22 Oct 2014 10:18:04 +0000 (12:18 +0200)]
qeth: make local functions static in qeth_l3 module

This patch makes 4 local functions static and removes
the prototypes from the header file.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: fix some trace formating issues
Thomas Richter [Wed, 22 Oct 2014 10:18:03 +0000 (12:18 +0200)]
qeth: fix some trace formating issues

This patch fixes trace formatting issues using the
QETH_CARD_TEXT_ macro. The total size of each trace entry
is 8 bytes. Some of the sprintf formats exceed these 8
bytes (for example using abcd:%d and the converted value
needs more than 3 bytes). The solution is to shorten the
text prepending the value or use a different format (%x).

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: qeth_core_main make local functions static
Thomas Richter [Wed, 22 Oct 2014 10:18:02 +0000 (12:18 +0200)]
qeth: qeth_core_main make local functions static

This patch makes some global functions static and removes
the prototypes from the header file.
Also function qeth_query_card_info is not exported anymore,
there is no external user for it, this function should never
have been exported in the first place.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <blaschka@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoxen-netfront: always keep the Rx ring full of requests
David Vrabel [Wed, 22 Oct 2014 10:17:06 +0000 (11:17 +0100)]
xen-netfront: always keep the Rx ring full of requests

A full Rx ring only requires 1 MiB of memory.  This is not enough
memory that it is useful to dynamically scale the number of Rx
requests in the ring based on traffic rates, because:

a) Even the full 1 MiB is a tiny fraction of a typically modern Linux
   VM (for example, the AWS micro instance still has 1 GiB of memory).

b) Netfront would have used up to 1 MiB already even with moderate
   data rates (there was no adjustment of target based on memory
   pressure).

c) Small VMs are going to typically have one VCPU and hence only one
   queue.

Keeping the ring full of Rx requests handles bursty traffic better
than trying to converge on an optimal number of requests to keep
filled.

On a 4 core host, an iperf -P 64 -t 60 run from dom0 to a 4 VCPU guest
improved from 5.1 Gbit/s to 5.6 Gbit/s.  Gains with more bursty
traffic are expected to be higher.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'sunvnet-napi'
David S. Miller [Sat, 25 Oct 2014 20:20:20 +0000 (16:20 -0400)]
Merge branch 'sunvnet-napi'

Sowmini Varadhan says:

====================
sunvnet: NAPIfy sunvnet

This patchset converts the sunvnet driver to use the NAPI framework.
Changes since v4 to Patch1:
  vnet_event accumulates LDC_EVENT_* bits into rx_event.
  vnet_event_napi() unrolls send_events() logic to process all rx_event bits.
Changes since v5:
  Patch 1: use net_device.h definition for NAPI_POLL_WEIGHT.
  Drop sparclinux changes (patch3) per David Miller feedback

Patch 1 in the series addresses the packet-receive path- all
the vnet_event() processing is moved into NAPI context.
This patch is dependant on the sparc-next commit:
  "sparc64: Add vio_set_intr() to enable/disable Rx interrupts"
  (sparc commit id ca605b7dd740c8909408d67911d8ddd272c2b320)

Patch 2 uses RCU to fix race conditions between vnet_port_remove and
paths that access/modify port-related state, such as vnet_start_xmit.

Patch 3 leverages from the NAPIfied Rx path,
dropping superfluous usage of the irqsave/irqrestores on the vio.lock
where possible.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosunvnet: Remove irqsave/irqrestore on vio.lock
Sowmini Varadhan [Sat, 25 Oct 2014 19:12:31 +0000 (15:12 -0400)]
sunvnet: Remove irqsave/irqrestore on vio.lock

After the  NAPIfication of sunvnet, we no longer need to
synchronize by doing irqsave/restore on vio.lock in the
I/O fastpath.

NAPI ->poll() is non-reentrant, so all RX processing occurs
strictly in a serialized environment. TX reclaim is done in NAPI
context, so the netif_tx_lock can be used to serialize
critical sections between Tx and Rx paths.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosunvnet: Use RCU to synchronize port usage with vnet_port_remove()
Sowmini Varadhan [Sat, 25 Oct 2014 19:12:20 +0000 (15:12 -0400)]
sunvnet: Use RCU to synchronize port usage with vnet_port_remove()

A vnet_port_remove could be triggered as a result of an ldm-unbind
operation by the peer, module unload, or other changes to the
inter-vnet-link configuration.  When this is concurrent with
vnet_start_xmit(), there are several race sequences possible,
such as

thread 1                                    thread 2
vnet_start_xmit
-> tx_port_find
   spin_lock_irqsave(&vp->lock..)
   ret = __tx_port_find(..)
   spin_lock_irqrestore(&vp->lock..)
                                           vio_remove -> ..
                                               ->vnet_port_remove
                                           spin_lock_irqsave(&vp->lock..)
                                           cleanup
                                           spin_lock_irqrestore(&vp->lock..)
                                           kfree(port)
/* attempt to use ret will bomb */

This patch adds RCU locking for port access so that vnet_port_remove
will correctly clean up port-related state.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Dwight Engen <dwight.engen@oracle.com>
Acked-by: Bob Picco <bob.picco@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosunvnet: NAPIfy sunvnet
Sowmini Varadhan [Sat, 25 Oct 2014 19:12:12 +0000 (15:12 -0400)]
sunvnet: NAPIfy sunvnet

Move Rx packet procssing to the NAPI poll callback.
Disable VIO interrupt and unconditioanlly go into NAPI
context from vnet_event.

Note that we want to minimize the number of LDC
STOP/START messages sent. Specifically, do not send a STOP
message if vnet_walk_rx does not read all the available descriptors
because of the NAPI budget limitation. Instead, note the end index
as part of port state, and resume from this index when the
next poll callback is triggered.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Raghuram Kothakota <raghuram.kothakota@oracle.com>
Acked-by: Dwight Engen <dwight.engen@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Fri, 24 Oct 2014 20:41:02 +0000 (16:41 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/net-next

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2014-10-23

This series contains updates to i40e and i40evf.

Jesse modifies the i40e driver to only notify the firmware on link up/down
and qualified module events.  Also simplified the job of managing link
state by using the admin queue receive event for link events as a signal
to tell the driver to update link state.

Jeff (me) cleans up the inconsistent use of tabs for indentation in the admin
queue command header file.

Neerav converts the use of udelay() to usleep_range().

Anjali fixes a bug where receive would stop after some stress by adding
a sleep and restart as well as moving the setting of flow control because
it should be done at a PF level and not a VSI level.

Mitch adds code to handle link events when updating the PF switch, which
allows link information to be properly provided to VFS in all cases.

Catherine adds driver support for 10GBaseT and bumps driver version.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: llc: include linux/errno.h instead of asm/errno.h
Fabian Frederick [Wed, 22 Oct 2014 19:06:26 +0000 (21:06 +0200)]
net: llc: include linux/errno.h instead of asm/errno.h

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agolapb: move EXPORT_SYMBOL after functions.
Fabian Frederick [Wed, 22 Oct 2014 19:01:41 +0000 (21:01 +0200)]
lapb: move EXPORT_SYMBOL after functions.

See Documentation/CodingStyle Chapter 6

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'berlin_ethernet'
David S. Miller [Fri, 24 Oct 2014 19:49:25 +0000 (15:49 -0400)]
Merge branch 'berlin_ethernet'

Sebastian Hesselbarth says:

====================
Marvell PXA168 libphy handling and Berlin Ethernet

This patch series deals with a removing a IP feature that can be found
on all currently supported Marvell Ethernet IP (pxa168_eth, mv643xx_eth,
mvneta). The MAC IP allows to automatically perform PHY auto-negotiation
without software interaction.

However, this feature (a) fundamentally clashes with the way libphy works
and (b) is unable to deal with quirky PHYs that require special treatment.
In this series, pxa168_eth driver is rewritten to completely disable that
feature and properly deal with libphy provided PHYs.

As usual, a branch on top of v3.18-rc1 can be found at

git://git.infradead.org/users/hesselba/linux-berlin.git devel/bg2-bg2cd-eth-v2

Patches 1-5 should go through David's net tree, I'll pick up the DT patches
6-9.

There have been some changes,
compared to the RFT
- added phy-connection-type property to BG2Q PHY DT node
- bail out from pxa168_eth_adjust_link when there is no change in
  PHY parameters. Also, add a call to phy_print_status.
compared to v1
- move phy-connection-type to ethernet node instead of PHY node

Patch 1 adds support for Marvell 88E3016 FastEthernet PHY that is also
integrated in Marvell Berlin BG2/BG2CD SoCs.

Patch 2 allows to pass phy_interface_t on pxa168_eth platform_data that
is only used by mach-mmp/gplug. From the board setup, I guessed gplug's
PHY is connected via RMII. The patch still isn't even compile tested.

Patches 3-5 prepare proper libphy handling and finally remove all in-driver
PHY mangling related to the feature explained above.

Patches 6-9 add corresponding ethernet DT nodes to BG2, BG2CD, add a
phy-connection-type property to BG2Q and enable ethernet on BG2-based Sony
NSZ-GS7. I have tested all this on GS7 successfully with ip=dhcp on 100M FD.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: pxa168_eth: Remove in-driver PHY mangling
Sebastian Hesselbarth [Wed, 22 Oct 2014 18:26:48 +0000 (20:26 +0200)]
net: pxa168_eth: Remove in-driver PHY mangling

With properly using libphy PHYs now, remove the in-driver PHY
mangling.

Tested-by: Antoine Ténart <antoine.tenart@free-electrons.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: pxa168_eth: Remove HW auto-negotiaion
Sebastian Hesselbarth [Wed, 22 Oct 2014 18:26:47 +0000 (20:26 +0200)]
net: pxa168_eth: Remove HW auto-negotiaion

Marvell Ethernet IP supports PHY negotiation driven by HW. This
fundamentally clashes with libphy (software) driven negotiation and
also cannot cope with quirky PHYs. Therefore, always disable any HW
negotiation features and properly use libphy's phy_device.

Tested-by: Antoine Ténart <antoine.tenart@free-electrons.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: pxa168_eth: Prepare proper libphy handling
Sebastian Hesselbarth [Wed, 22 Oct 2014 18:26:46 +0000 (20:26 +0200)]
net: pxa168_eth: Prepare proper libphy handling

Current libphy handling in pxa168_eth lacks proper phy_connect. Prepare
to fix this by first moving phy properties from platform_data to private
driver data.

Tested-by: Antoine Ténart <antoine.tenart@free-electrons.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: pxa168_eth: Provide phy_interface mode on platform_data
Sebastian Hesselbarth [Wed, 22 Oct 2014 18:26:45 +0000 (20:26 +0200)]
net: pxa168_eth: Provide phy_interface mode on platform_data

The PXA168 Ethernet IP support MII and RMII connection to its PHY.
Currently, pxa168 platform_data does not provide a way to pass that
and there is one user of pxa168 platform_data (mach-mmp/gplug).
Given the pinctrl settings of gplug it uses RMII, so add and pass
a corresponding phy_interface_t.

Tested-by: Antoine Ténart <antoine.tenart@free-electrons.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agophy: marvell: Add support for 88E3016 FastEthernet PHY
Sebastian Hesselbarth [Wed, 22 Oct 2014 18:26:44 +0000 (20:26 +0200)]
phy: marvell: Add support for 88E3016 FastEthernet PHY

Marvell 88E3016 is a FastEthernet PHY that also can be found in Marvell
Berlin SoCs as integrated PHY.

Tested-by: Antoine Ténart <antoine.tenart@free-electrons.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonatsemi/macsonic: Remove superfluous interrupt disable/restore
Geert Uytterhoeven [Tue, 21 Oct 2014 17:53:57 +0000 (19:53 +0200)]
natsemi/macsonic: Remove superfluous interrupt disable/restore

As of commit e4dc601bf99ccd1c ("m68k: Disable/restore interrupts in
hwreg_present()/hwreg_write()"), this is no longer needed.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agocirrus/mac89x0: Remove superfluous interrupt disable/restore
Geert Uytterhoeven [Tue, 21 Oct 2014 17:53:11 +0000 (19:53 +0200)]
cirrus/mac89x0: Remove superfluous interrupt disable/restore

As of commit e4dc601bf99ccd1c ("m68k: Disable/restore interrupts in
hwreg_present()/hwreg_write()"), this is no longer needed.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: typhoon: Remove redundant casts
Rasmus Villemoes [Tue, 21 Oct 2014 14:51:43 +0000 (16:51 +0200)]
net: typhoon: Remove redundant casts

Both image_data and typhoon_fw->data are const u8*, so the cast to u8*
is unnecessary and confusing.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: David Dillow <dave@thedillows.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoRemoved unused function sctp_addr_is_valid()
Sébastien Barré [Tue, 21 Oct 2014 13:26:15 +0000 (15:26 +0200)]
Removed unused function sctp_addr_is_valid()

sctp_addr_is_valid() only appeared in its definition.

Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Sébastien Barré <sebastien.barre@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'ipv6_route'
David S. Miller [Fri, 24 Oct 2014 04:14:52 +0000 (00:14 -0400)]
Merge branch 'ipv6_route'

Martin KaFai Lau says:

====================
ipv6: Reduce the number of fib6_lookup() calls from ip6_pol_route()

This patch set is trying to reduce the number of fib6_lookup()
calls from ip6_pol_route().

I have adapted davem's udpflooda and kbench_mod test
(https://git.kernel.org/pub/scm/linux/kernel/git/davem/net_test_tools.git) to
support IPv6 and here is the result:

Before:
[root]# for i in $(seq 1 3); do time ./udpflood -l 20000000 -c 250 2401:face:face:face::2; done

real    0m34.190s
user    0m3.047s
sys     0m31.108s

real    0m34.635s
user    0m3.125s
sys     0m31.475s

real    0m34.517s
user    0m3.034s
sys     0m31.449s

[root]# insmod ip6_route_kbench.ko oif=2 src=2401:face:face:face::1 dst=2401:face:face:face::2
[  660.160976] ip6_route_kbench: ip6_route_output tdiff: 933
[  660.207261] ip6_route_kbench: ip6_route_output tdiff: 988
[  660.253492] ip6_route_kbench: ip6_route_output tdiff: 896
[  660.298862] ip6_route_kbench: ip6_route_output tdiff: 898

After:
[root]# for i in $(seq 1 3); do time ./udpflood -l 20000000 -c 250 2401:face:face:face::2; done

real    0m32.695s
user    0m2.925s
sys     0m29.737s

real    0m32.636s
user    0m3.007s
sys     0m29.596s

real    0m32.797s
user    0m2.866s
sys     0m29.898s

[root]# insmod ip6_route_kbench.ko oif=2 src=2401:face:face:face::1 dst=2401:face:face:face::2
[  881.220793] ip6_route_kbench: ip6_route_output tdiff: 684
[  881.253477] ip6_route_kbench: ip6_route_output tdiff: 640
[  881.286867] ip6_route_kbench: ip6_route_output tdiff: 630
[  881.320749] ip6_route_kbench: ip6_route_output tdiff: 653

/****************************** udpflood.c ******************************/
/* It is an adaptation of the Eric Dumazet's and David Miller's
 * udpflood tool, by adding IPv6 support.
 */

typedef uint32_t u32;

static int debug =3D 0;

/* Allow -fstrict-aliasing */
typedef union sa_u {
struct sockaddr_storage a46;
struct sockaddr_in a4;
struct sockaddr_in6 a6;
} sa_u;

static int usage(void)
{
printf("usage: udpflood [ -l count ] [ -m message_size ] [ -c num_ip_addrs=
 ] IP_ADDRESS\n");
return -1;
}

static u32 get_last32h(const sa_u *sa)
{
if (sa->a46.ss_family =3D=3D PF_INET)
return ntohl(sa->a4.sin_addr.s_addr);
else
return ntohl(sa->a6.sin6_addr.s6_addr32[3]);
}

static void set_last32h(sa_u *sa, u32 last32h)
{
if (sa->a46.ss_family =3D=3D PF_INET)
sa->a4.sin_addr.s_addr =3D htonl(last32h);
else
sa->a6.sin6_addr.s6_addr32[3] =3D htonl(last32h);
}

static void print_saddr(const sa_u *sa, const char *msg)
{
char buf[64];

if (!debug)
return;

switch (sa->a46.ss_family) {
case PF_INET:
inet_ntop(PF_INET, &(sa->a4.sin_addr.s_addr), buf,
  sizeof(buf));
break;
case PF_INET6:
inet_ntop(PF_INET6, &(sa->a6.sin6_addr), buf, sizeof(buf));
break;
}

printf("%s: %s\n", msg, buf);
}

static int send_packets(const sa_u *sa, size_t num_addrs, int count, int ms=
g_sz)
{
char *msg =3D malloc(msg_sz);
sa_u saddr;
u32 start_addr32h, end_addr32h, cur_addr32h;
int fd, i, err;

if (!msg)
return -ENOMEM;

memset(msg, 0, msg_sz);

memcpy(&saddr, sa, sizeof(saddr));
cur_addr32h =3D start_addr32h =3D get_last32h(&saddr);
end_addr32h =3D start_addr32h + num_addrs;

fd =3D socket(saddr.a46.ss_family, SOCK_DGRAM, 0);
if (fd < 0) {
perror("socket");
err =3D fd;
goto out_nofd;
}

/* connect to avoid the kernel spending time in figuring
 * out the source address (i.e pin the src address)
 */
err =3D connect(fd, (struct sockaddr *) &saddr, sizeof(saddr));
if (err < 0) {
perror("connect");
goto out;
}

print_saddr(&saddr, "start_addr");
for (i =3D 0; i < count; i++) {
print_saddr(&saddr, "sendto");
err =3D sendto(fd, msg, msg_sz, 0, (struct sockaddr *)&saddr,
     sizeof(saddr));
if (err < 0) {
perror("sendto");
goto out;
}

if (++cur_addr32h >=3D end_addr32h)
cur_addr32h =3D start_addr32h;
set_last32h(&saddr, cur_addr32h);
}

err =3D 0;
out:
close(fd);
out_nofd:
free(msg);
return err;
}

int main(int argc, char **argv, char **envp)
{
int port, msg_sz, count, num_addrs, ret;

sa_u start_addr;

port =3D 6000;
msg_sz =3D 32;
count =3D 10000000;
num_addrs =3D 1;

while ((ret =3D getopt(argc, argv, "dl:s:p:c:")) >=3D 0) {
switch (ret) {
case 'l':
sscanf(optarg, "%d", &count);
break;
case 's':
sscanf(optarg, "%d", &msg_sz);
break;
case 'p':
sscanf(optarg, "%d", &port);
break;
case 'c':
sscanf(optarg, "%d", &num_addrs);
break;
case 'd':
debug =3D 1;
break;
case '?':
return usage();
}
}

if (num_addrs < 1)
return usage();

if (!argv[optind])
return usage();

start_addr.a4.sin_port =3D htons(port);
if (inet_pton(PF_INET, argv[optind], &start_addr.a4.sin_addr))
start_addr.a46.ss_family =3D PF_INET;
else if (inet_pton(PF_INET6, argv[optind], &start_addr.a6.sin6_addr.s6_add=
r))
start_addr.a46.ss_family =3D PF_INET6;
else
return usage();

return send_packets(&start_addr, num_addrs, count, msg_sz);
}

/****************** ip6_route_kbench_mod.c ******************/

/* We can't just use "get_cycles()" as on some platforms, such
 * as sparc64, that gives system cycles rather than cpu clock
 * cycles.
 */

static inline unsigned long long get_tick(void)
{
unsigned long long t;

__asm__ __volatile__("rd %%tick, %0" : "=r" (t));
return t;
}
static inline unsigned long long get_tick(void)
{
unsigned long long t;

rdtscll(t);

return t;
}
static inline unsigned long long get_tick(void)
{
return get_cycles();
}

static int flow_oif = DEFAULT_OIF;
static int flow_iif = DEFAULT_IIF;
static u32 flow_mark = DEFAULT_MARK;
static struct in6_addr flow_dst_ip_addr;
static struct in6_addr flow_src_ip_addr;
static int flow_tos = DEFAULT_TOS;

static char dst_string[64];
static char src_string[64];

module_param_string(dst, dst_string, sizeof(dst_string), 0);
module_param_string(src, src_string, sizeof(src_string), 0);

static int __init flow_setup(void)
{
if (dst_string[0] &&
    !in6_pton(dst_string, -1, &flow_dst_ip_addr.s6_addr[0], -1, NULL)) {
pr_info("cannot parse \"%s\"\n", dst_string);
return -1;
}

if (src_string[0] &&
    !in6_pton(src_string, -1, &flow_src_ip_addr.s6_addr[0], -1, NULL)) {
pr_info("cannot parse \"%s\"\n", dst_string);
return -1;
}

return 0;
}

module_param_named(oif, flow_oif, int, 0);
module_param_named(iif, flow_iif, int, 0);
module_param_named(mark, flow_mark, uint, 0);
module_param_named(tos, flow_tos, int, 0);

static int warmup_count = DEFAULT_WARMUP_COUNT;
module_param_named(count, warmup_count, int, 0);

static void flow_init(struct flowi6 *fl6)
{
memset(fl6, 0, sizeof(*fl6));
fl6->flowi6_proto = IPPROTO_ICMPV6;
fl6->flowi6_oif = flow_oif;
fl6->flowi6_iif = flow_iif;
fl6->flowi6_mark = flow_mark;
fl6->flowi6_tos = flow_tos;
fl6->daddr = flow_dst_ip_addr;
fl6->saddr = flow_src_ip_addr;
}

static struct sk_buff * fake_skb_get(void)
{
struct ipv6hdr *hdr;
struct sk_buff *skb;

skb = alloc_skb(4096, GFP_KERNEL);
if (!skb) {
pr_info("Cannot alloc SKB for test\n");
return NULL;
}
skb->dev = __dev_get_by_index(&init_net, flow_iif);
if (skb->dev == NULL) {
pr_info("Input device (%d) does not exist\n", flow_iif);
goto err;
}

skb_reset_mac_header(skb);
skb_reset_network_header(skb);
skb_reserve(skb, MAX_HEADER + sizeof(struct ipv6hdr));
hdr = ipv6_hdr(skb);

hdr->priority = 0;
hdr->version = 6;
memset(hdr->flow_lbl, 0, sizeof(hdr->flow_lbl));
hdr->payload_len = htons(sizeof(struct icmp6hdr));
hdr->nexthdr = IPPROTO_ICMPV6;
hdr->saddr = flow_src_ip_addr;
hdr->daddr = flow_dst_ip_addr;
skb->protocol = htons(ETH_P_IPV6);
skb->mark = flow_mark;

return skb;
err:
kfree_skb(skb);
return NULL;
}

static void do_full_output_lookup_bench(void)
{
unsigned long long t1, t2, tdiff;
struct rt6_info *rt;
struct flowi6 fl6;
int i;

rt = NULL;

for (i = 0; i < warmup_count; i++) {
flow_init(&fl6);

rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl6);
if (IS_ERR(rt))
break;
ip6_rt_put(rt);
}
if (IS_ERR(rt)) {
pr_info("ip_route_output_key: err=%ld\n", PTR_ERR(rt));
return;
}

flow_init(&fl6);

t1 = get_tick();
rt = (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl6);
t2 = get_tick();
if (!IS_ERR(rt))
ip6_rt_put(rt);

tdiff = t2 - t1;
pr_info("ip6_route_output tdiff: %llu\n", tdiff);
}

static void do_full_input_lookup_bench(void)
{
unsigned long long t1, t2, tdiff;
struct sk_buff *skb;
struct rt6_info *rt;
int err, i;

skb = fake_skb_get();
if (skb == NULL)
goto out_free;

err = 0;
local_bh_disable();
for (i = 0; i < warmup_count; i++) {
ip6_route_input(skb);
rt = (struct rt6_info *)skb_dst(skb);
err = (!rt || rt == init_net.ipv6.ip6_null_entry);
skb_dst_drop(skb);
if (err)
break;
}
local_bh_enable();

if (err) {
pr_info("Input route lookup fails\n");
goto out_free;
}

local_bh_disable();
t1 = get_tick();
ip6_route_input(skb);
t2 = get_tick();
local_bh_enable();

rt = (struct rt6_info *)skb_dst(skb);
err = (!rt || rt == init_net.ipv6.ip6_null_entry);
skb_dst_drop(skb);
if (err) {
pr_info("Input route lookup fails\n");
goto out_free;
}

tdiff = t2 - t1;
pr_info("ip6_route_input tdiff: %llu\n", tdiff);

out_free:
kfree_skb(skb);
}

static void do_full_lookup_bench(void)
{
if (!flow_iif)
do_full_output_lookup_bench();
else
do_full_input_lookup_bench();
}

static void do_bench(void)
{
do_full_lookup_bench();
do_full_lookup_bench();
do_full_lookup_bench();
do_full_lookup_bench();
}

static int __init kbench_init(void)
{
if (flow_setup())
return -EINVAL;

pr_info("flow [IIF(%d),OIF(%d),MARK(0x%08x),D("IP6_FMT"),"
"S("IP6_FMT"),TOS(0x%02x)]\n",
flow_iif, flow_oif, flow_mark,
IP6_PRT(flow_dst_ip_addr),
IP6_PRT(flow_src_ip_addr),
flow_tos);

if (!cpu_has_tsc) {
pr_err("X86 TSC is required, but is unavailable.\n");
return -EINVAL;
}

pr_info("sizeof(struct rt6_info)==%zu\n", sizeof(struct rt6_info));

do_bench();

return -ENODEV;
}

static void __exit kbench_exit(void)
{
}

module_init(kbench_init);
module_exit(kbench_exit);
MODULE_LICENSE("GPL");
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: Avoid redoing fib6_lookup() with reachable = 0 by saving fn
Martin KaFai Lau [Mon, 20 Oct 2014 20:42:45 +0000 (13:42 -0700)]
ipv6: Avoid redoing fib6_lookup() with reachable = 0 by saving fn

This patch save the fn before doing rt6_backtrack.
Hence, without redo-ing the fib6_lookup(), saved_fn can be used
to redo rt6_select() with RT6_LOOKUP_F_REACHABLE off.

Some minor changes I think make sense to review as a single patch:
* Remove the 'out:' goto label.
* Remove the 'reachable' variable. Only use the 'strict' variable instead.

After this patch, "failing ip6_ins_rt()" should be the only case that
requires a redo of fib6_lookup().

Cc: David Miller <davem@davemloft.net>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>