platform/kernel/linux-exynos.git
9 years agomlxsw: Use correct skb length when dumping payload
Ido Schimmel [Thu, 6 Aug 2015 14:41:57 +0000 (16:41 +0200)]
mlxsw: Use correct skb length when dumping payload

Do not use the length of the transmitted skb (which was freed), but
that of the response skb.

This issue was discovered using the Kernel Address sanitizer (KASan).

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: Simplify mlxsw_sx_port_xmit function
Ido Schimmel [Thu, 6 Aug 2015 14:41:56 +0000 (16:41 +0200)]
mlxsw: Simplify mlxsw_sx_port_xmit function

Previously we only checked if the transmission queue is not full in the
middle of the xmit function. This lead to complex logic due to the fact
that sometimes we need to reallocate the headroom for our Tx header.

Allow the switch driver to know if the transmission queue is not full
before sending the packet and remove this complex logic.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: Strip FCS from incoming packets
Jiri Pirko [Thu, 6 Aug 2015 14:41:55 +0000 (16:41 +0200)]
mlxsw: Strip FCS from incoming packets

FCS of incoming packets is already checked by HW. Just strip it out.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: Make pci module dependent on HAS_DMA and HAS_IOMEM
Jiri Pirko [Thu, 6 Aug 2015 14:41:54 +0000 (16:41 +0200)]
mlxsw: Make pci module dependent on HAS_DMA and HAS_IOMEM

This resolves compile errors on um-allyesconfig.

Note that there are many other drivers which have the same issue.

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: Make system port to local port mapping explicit
Ido Schimmel [Thu, 6 Aug 2015 14:41:53 +0000 (16:41 +0200)]
mlxsw: Make system port to local port mapping explicit

System ports are unique identifiers in a multi-ASIC environment that
represent all the available ports in the system. Local ports on the
other hand, are unique only within the local ASIC.

Since system port to local port mapping is not part of the HW-SW
contract and since only single-ASIC configurations are currently
supported, set an explicit 1:1 mapping by configuring the Switch System
Port Record (SSPR) register.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: Call free_netdev when removing port
Ido Schimmel [Thu, 6 Aug 2015 14:41:52 +0000 (16:41 +0200)]
mlxsw: Call free_netdev when removing port

When removing a port's netdevice we should also free the memory
allocated by alloc_etherdev(). Do this by calling free_netdev() at the
end of the teardown sequence.

Reported-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ethernet: Fix double word "the the" in eth.c
Masanari Iida [Thu, 6 Aug 2015 12:27:54 +0000 (21:27 +0900)]
net: ethernet: Fix double word "the the" in eth.c

This patch fix double word "the the" in
Documentation/DocBook/networking/API-eth-get-headlen.html
Documentation/DocBook/networking/netdev.html
Documentation/DocBook/networking.xml

These files are generated from comment in source,
so I have to fix comment in net/ethernet/eth.c.

Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: phy: add RealTek RTL8211DN phy id
Shaohui Xie [Thu, 6 Aug 2015 11:03:35 +0000 (19:03 +0800)]
net: phy: add RealTek RTL8211DN phy id

RTL8211DN is compatible with RTL8211E.

Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Enforce payload type of traffic sent using explicit NULL
Robert Shearman [Thu, 6 Aug 2015 10:04:56 +0000 (11:04 +0100)]
mpls: Enforce payload type of traffic sent using explicit NULL

RFC 4182 s2 states that if an IPv4 Explicit NULL label is the only
label on the stack, then after popping the resulting packet must be
treated as a IPv4 packet and forwarded based on the IPv4 header. The
same is true for IPv6 Explicit NULL with an IPv6 packet following.

Therefore, when installing the IPv4/IPv6 Explicit NULL label routes,
add an attribute that specifies the expected payload type for use at
forwarding time for determining the type of the encapsulated packet
instead of inspecting the first nibble of the packet.

Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'bpf-perf'
David S. Miller [Mon, 10 Aug 2015 05:50:06 +0000 (22:50 -0700)]
Merge branch 'bpf-perf'

Kaixu Xia says:

====================
bpf: Introduce the new ability of eBPF programs to access hardware PMU counter

This patchset is base on the net-next:
 git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git
commit 9dc20a649609c95ce7c5ac4282656ba627b67d49.

Previous patch v6 url:
https://lkml.org/lkml/2015/8/4/188

changes in V7:
 - rebase the whole patch set to net-next tree(9dc20a64);
 - split out the core perf APIs into Patch 1/5;
 - change the return value of function perf_event_attrs()
   from struct perf_event * to const struct perf_event * in
   Patch 1/5;
 - rename the function perf_event_read_internal() to perf_event_
   read_local() and rewrite it in Patch 1/5;
 - rename the function check_func_limit() to check_map_func
   _compatibility() and remove the unnecessary pass pointer to
   a pointer in Patch 4/5;

changes in V6:
 - make the Patch 1/4 commit message more meaning and readable;
 - remove the unnecessary comment in Patch 2/4 and make it clean;
 - declare the function perf_event_release_kernel() in include/
   linux/perf_event.h to fix the build error when CONFIG_PERF_EVENTS
   isn't configured in Patch 2/4;
 - add function perf_event_attrs() to get the struct perf_event_attr
   in Patch 2/4.
 - move the related code from kernel/trace/bpf_trace.c to kernel/
   events/core.c and add function perf_event_read_internal() to
   avoid poking inside of the event outside of perf code in Patch 3/4;
 - generial the func & map match-pair with an array in Patch 3/4;

changes in V5:
 - move struct fd_array_map_ops* fd_ops to bpf_map;
 - move array perf event decrement refcnt function to
   map_free;
 - fix the NULL ptr of perf_event_get();
 - move bpf_perf_event_read() to kernel/bpf/bpf_trace.c;
 - get rid of the remaining struct bpf_prog;
 - move the unnecessay cast on void *;

changes in V4:
 - make the bpf_prog_array_map more generic;
 - fix the bug of event refcnt leak;
 - use more useful errno in bpf_perf_event_read();

changes in V3:
 - collapse V2 patches 1-3 into one;
 - drop the function map->ops->map_traverse_elem() and release
   the struct perf_event in map_free;
 - only allow to access bpf_perf_event_read() from programs;
 - update the perf_event_array_map elem via xchg();
 - pass index directly to bpf_perf_event_read() instead of
   MAP_KEY;

changes in V2:
 - put atomic_long_inc_not_zero() between fdget() and fdput();
 - limit the event type to PERF_TYPE_RAW and PERF_TYPE_HARDWARE;
 - Only read the event counter on current CPU or on current
   process;
 - add new map type BPF_MAP_TYPE_PERF_EVENT_ARRAY to store the
   pointer to the struct perf_event;
 - according to the perf_event_map_fd and key, the function
   bpf_perf_event_read() can get the Hardware PMU counter value;

Patch 5/5 is a simple example and shows how to use this new eBPF
programs ability. The PMU counter data can be found in
/sys/kernel/debug/tracing/trace(trace_pipe).(the cycles PMU
value when 'kprobe/sys_write' sampling)

  $ cat /sys/kernel/debug/tracing/trace_pipe
  $ ./tracex6
       ...
       syslog-ng-548   [000] d..1    76.905673: : CPU-0   681765271
       syslog-ng-548   [000] d..1    76.905690: : CPU-0   681787855
       syslog-ng-548   [000] d..1    76.905707: : CPU-0   681810504
       syslog-ng-548   [000] d..1    76.905725: : CPU-0   681834771
       syslog-ng-548   [000] d..1    76.905745: : CPU-0   681859519
       syslog-ng-548   [000] d..1    76.905766: : CPU-0   681890419
       syslog-ng-548   [000] d..1    76.905783: : CPU-0   681914045
       syslog-ng-548   [000] d..1    76.905800: : CPU-0   681935950
       syslog-ng-548   [000] d..1    76.905816: : CPU-0   681958299
              ls-690   [005] d..1    82.241308: : CPU-5   3138451
              sh-691   [004] d..1    82.244570: : CPU-4   7324988
           <...>-699   [007] d..1    99.961387: : CPU-7   3194027
           <...>-695   [003] d..1    99.961474: : CPU-3   288901
           <...>-695   [003] d..1    99.961541: : CPU-3   383145
           <...>-695   [003] d..1    99.961591: : CPU-3   450365
           <...>-695   [003] d..1    99.961639: : CPU-3   515751
           <...>-695   [003] d..1    99.961686: : CPU-3   579047
       ...

The detail of patches is as follow:

Patch 1/5 add the necessary core perf APIs perf_event_attrs(),
perf_event_get(),perf_event_read_local() when accessing events
counters in eBPF programs

Patch 2/5 rewrites part of the bpf_prog_array map code and make it
more generic;

Patch 3/5 introduces a new bpf map type. This map only stores the
pointer to struct perf_event;

Patch 4/5 implements function bpf_perf_event_read() that get the
selected hardware PMU conuter;

Patch 5/5 gives a simple example.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosamples/bpf: example of get selected PMU counter value
Kaixu Xia [Thu, 6 Aug 2015 07:02:36 +0000 (07:02 +0000)]
samples/bpf: example of get selected PMU counter value

This is a simple example and shows how to use the new ability
to get the selected Hardware PMU counter value.

Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: Implement function bpf_perf_event_read() that get the selected hardware PMU...
Kaixu Xia [Thu, 6 Aug 2015 07:02:35 +0000 (07:02 +0000)]
bpf: Implement function bpf_perf_event_read() that get the selected hardware PMU conuter

According to the perf_event_map_fd and index, the function
bpf_perf_event_read() can convert the corresponding map
value to the pointer to struct perf_event and return the
Hardware PMU counter value.

Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: Add new bpf map type to store the pointer to struct perf_event
Kaixu Xia [Thu, 6 Aug 2015 07:02:34 +0000 (07:02 +0000)]
bpf: Add new bpf map type to store the pointer to struct perf_event

Introduce a new bpf map type 'BPF_MAP_TYPE_PERF_EVENT_ARRAY'.
This map only stores the pointer to struct perf_event. The
user space event FDs from perf_event_open() syscall are converted
to the pointer to struct perf_event and stored in map.

Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: Make the bpf_prog_array_map more generic
Wang Nan [Thu, 6 Aug 2015 07:02:33 +0000 (07:02 +0000)]
bpf: Make the bpf_prog_array_map more generic

All the map backends are of generic nature. In order to avoid
adding much special code into the eBPF core, rewrite part of
the bpf_prog_array map code and make it more generic. So the
new perf_event_array map type can reuse most of code with
bpf_prog_array map and add fewer lines of special code.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoperf: add the necessary core perf APIs when accessing events counters in eBPF programs
Kaixu Xia [Thu, 6 Aug 2015 07:02:32 +0000 (07:02 +0000)]
perf: add the necessary core perf APIs when accessing events counters in eBPF programs

This patch add three core perf APIs:
 - perf_event_attrs(): export the struct perf_event_attr from struct
   perf_event;
 - perf_event_get(): get the struct perf_event from the given fd;
 - perf_event_read_local(): read the events counters active on the
   current CPU;
These APIs are needed when accessing events counters in eBPF programs.

The API perf_event_read_local() comes from Peter and I add the
corresponding SOB.

Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mv88e6xxx-switchdev-fdb'
David S. Miller [Mon, 10 Aug 2015 05:48:10 +0000 (22:48 -0700)]
Merge branch 'mv88e6xxx-switchdev-fdb'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: support switchdev FDB objects

This patchset refactors the DSA and mv88e6xxx code to use the switchdev FDB
objects.

The first two patches add minor but necessary changes to switchdev, the third
one implements the switchdev glue in DSA for FDB routines, and the remaining
ones refactor the FDB access functions in the mv88e6xxx code.

Below is an usage example (ports 0-2 belongs to br0, ports 3-4 belongs to br1):

    # bridge fdb add 3c:97:0e:11:30:6e dev swp2
    # bridge fdb add 3c:97:0e:11:40:78 dev swp3
    # bridge fdb add 3c:97:0e:11:50:86 dev swp4
    # bridge fdb del 3c:97:0e:11:40:78 dev swp3
    # bridge fdb
    01:00:5e:00:00:01 dev eth0 self permanent
    01:00:5e:00:00:01 dev eth1 self permanent
    00:50:d2:10:78:15 dev swp0 master br0 permanent
    3c:97:0e:11:30:6e dev swp2 self static
    00:50:d2:10:78:15 dev swp3 master br1 permanent
    3c:97:0e:11:50:86 dev swp4 self static
    # cat /sys/kernel/debug/dsa0/atu
    # DB   T/P  Vec State Addr
    # 001  Port 004   e   3c:97:0e:11:30:6e
    # 004  Port 010   e   3c:97:0e:11:50:86

For the 88E6xxx switches, FIDs 1 to num_ports will be reserved for non-bridged
ports and bridge groups, and the remaining will be later used by VLANs.

This change is necessary to welcome the support for hardware VLANs (which will
follow soon).

Changes in v2:

 - remove ndo_bridge_{get,set,del}link from switchdev/DSA glue code

 - use ether_addr_copy instead of memcpy for MAC addresses

 - constify MAC address in port_fdb_{add,del}

 - split the mv88e6xxx code refactoring into several patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rework FDB add/del operations
Vivien Didelot [Thu, 6 Aug 2015 05:44:08 +0000 (01:44 -0400)]
net: dsa: mv88e6xxx: rework FDB add/del operations

Add a low level function for the ATU Load operation, and provide FDB add
and delete wrappers functions.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rework FDB getnext operation
Vivien Didelot [Thu, 6 Aug 2015 05:44:07 +0000 (01:44 -0400)]
net: dsa: mv88e6xxx: rework FDB getnext operation

This commit adds a low level _mv88e6xxx_atu_getnext function and helpers
to rewrite the mv88e6xxx_port_fdb_getnext operation.

A mv88e6xxx_atu_entry structure is added for convenient access to the
hardware, and GLOBAL_ATU_FID is defined instead of the raw 0x01 value.

The previous implementation did not handle the eventual trunk mapping.
If the related bit is set, then the ATU data register would contain the
trunk ID, and not the port vector.

Check this in the FDB getnext operation and do not handle it (yet).

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rename ATU MAC accessors
Vivien Didelot [Thu, 6 Aug 2015 05:44:06 +0000 (01:44 -0400)]
net: dsa: mv88e6xxx: rename ATU MAC accessors

Rename the __mv88e6xxx_{read,write}_addr functions to more explicit
_mv88e6xxx_atu_mac_{read,write} functions, which also respect the single
underscore convention used in the file (meaning SMI lock must be held).

In the meantime, define their MAC address parameters as an array of
ETH_ALEN bytes instead of a char pointer.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: extend fid mask
Vivien Didelot [Thu, 6 Aug 2015 05:44:05 +0000 (01:44 -0400)]
net: dsa: mv88e6xxx: extend fid mask

The driver currently manages one FID per port (or bridge group), with a
mask of DSA_MAX_PORTS bits, where 0 means that the FID is in use.

The Marvell 88E6xxx switches support up to 4094 FIDs (from 1 to 0xfff;
FID 0 means that multiple address databases are not being used).

This patch changes the fid_mask for an fid_bitmap of 4096 bits.

>From now on, FIDs 1 to num_ports are reserved for non-bridged ports and
bridge groups (a bridge group gets the FID of its first member). The
remaining bits will be reserved for VLAN entries.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: add support for switchdev FDB objects
Vivien Didelot [Thu, 6 Aug 2015 05:44:04 +0000 (01:44 -0400)]
net: dsa: add support for switchdev FDB objects

Remove the fdb_{add,del,getnext} function pointer in favor of new
port_fdb_{add,del,getnext}.

Implement the switchdev_port_obj_{add,del,dump} functions in DSA to
support the SWITCHDEV_OBJ_PORT_FDB objects.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: switchdev: support static FDB addresses
Vivien Didelot [Thu, 6 Aug 2015 05:44:03 +0000 (01:44 -0400)]
net: switchdev: support static FDB addresses

This patch adds a is_static boolean to the switchdev_obj_fdb structure,
in order to set the ndm_state to either NUD_NOARP or NUD_REACHABLE.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: switchdev: change fdb addr for a byte array
Vivien Didelot [Thu, 6 Aug 2015 05:44:02 +0000 (01:44 -0400)]
net: switchdev: change fdb addr for a byte array

The address in the switchdev_obj_fdb structure is currently represented
as a pointer. Replacing it for a 6-byte array allows switchdev to carry
addresses directly read from hardware registers, not stored by the
switch chip driver (as in Rocker).

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet:wimax: Fix doucble word "the the" in networking.xml
Masanari Iida [Thu, 6 Aug 2015 03:52:47 +0000 (12:52 +0900)]
net:wimax: Fix doucble word "the the" in networking.xml

This patch fix a double word "the the"
in Documentation/DocBook/networking.xml and
Documentation/DocBook/networking/API-Wimax-report-rfkill-sw.html.

These files are generated from comment in source, so I had to
fix the typo in net/wimax/io-rfkill.c

Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Fix race condition in store_rps_map
Tom Herbert [Wed, 5 Aug 2015 16:39:27 +0000 (09:39 -0700)]
net: Fix race condition in store_rps_map

There is a race condition in store_rps_map that allows jump label
count in rps_needed to go below zero. This can happen when
concurrently attempting to set and a clear map.

Scenario:

1. rps_needed count is zero
2. New map is assigned by setting thread, but rps_needed count _not_ yet
   incremented (rps_needed count still zero)
2. Map is cleared by second thread, old_map set to that just assigned
3. Second thread performs static_key_slow_dec, rps_needed count now goes
   negative

Fix is to increment or decrement rps_needed under the spinlock.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Make 100 percents packets sampled when sampling rate is 1.
Wenyu Zhang [Wed, 5 Aug 2015 07:30:47 +0000 (00:30 -0700)]
openvswitch: Make 100 percents packets sampled when sampling rate is 1.

When sampling rate is 1, the sampling probability is UINT32_MAX. The packet
should be sampled even the prandom32() generate the number of UINT32_MAX.
And none packet need be sampled when the probability is 0.

Signed-off-by: Wenyu Zhang <wenyuz@vmware.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovxlan: combine VXLAN_FLOWBASED into VXLAN_COLLECT_METADATA
Alexei Starovoitov [Wed, 5 Aug 2015 05:51:07 +0000 (22:51 -0700)]
vxlan: combine VXLAN_FLOWBASED into VXLAN_COLLECT_METADATA

IFLA_VXLAN_FLOWBASED is useless without IFLA_VXLAN_COLLECT_METADATA,
so combine them into single IFLA_VXLAN_COLLECT_METADATA flag.
'flowbased' doesn't convey real meaning of the vxlan tunnel mode.
This mode can be used by routing, tc+bpf and ovs.
Only ovs is strictly flow based, so 'collect metadata' is a better
name for this tunnel mode.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'rds-tcp-netns'
David S. Miller [Fri, 7 Aug 2015 18:29:58 +0000 (11:29 -0700)]
Merge branch 'rds-tcp-netns'

Sowmini Varadhan says:

====================
RDS-TCP: Network namespace support

This patch series contains the set of changes to correctly set up
the infra for PF_RDS sockets that use TCP as the transport in multiple
network namespaces.

Patch 1 in the series is the minimal set of changes to allow
a single instance of RDS-TCP to run in any (i.e init_net or other) net
namespace.  The changes in this patch set ensure that the execution of
'modprobe [-r] rds_tcp' sets up the kernel TCP sockets
relative to the current netns, so that RDS applications can send/recv
packets from that netns, and the netns can later be deleted cleanly.

Patch 2 of the series further allows multiple RDS-TCP instances,
one per network namespace. The changes in this patch allows dynamic
creation/tear-down of RDS-TCP client and server sockets  across all
current and future namespaces.

v2 changes from RFC sent out earlier:
    David Ahern comments in patch 1, net_device notifier in patch 2,
    patch 3 broken off and submitted separately.
v3: Cong Wang review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS-TCP: Support multiple RDS-TCP listen endpoints, one per netns.
Sowmini Varadhan [Wed, 5 Aug 2015 05:43:26 +0000 (01:43 -0400)]
RDS-TCP: Support multiple RDS-TCP listen endpoints, one per netns.

Register pernet subsys init/stop functions that will set up
and tear down per-net RDS-TCP listen endpoints. Unregister
pernet subusys functions on 'modprobe -r' to clean up these
end points.

Enable keepalive on both accept and connect socket endpoints.
The keepalive timer expiration will ensure that client socket
endpoints will be removed as appropriate from the netns when
an interface is removed from a namespace.

Register a device notifier callback that will clean up all
sockets (and thus avoid the need to wait for keepalive timeout)
when the loopback device is unregistered from the netns indicating
that the netns is getting deleted.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS-TCP: Make RDS-TCP work correctly when it is set up in a netns other than init_net
Sowmini Varadhan [Wed, 5 Aug 2015 05:43:25 +0000 (01:43 -0400)]
RDS-TCP: Make RDS-TCP work correctly when it is set up in a netns other than init_net

Open the sockets calling sock_create_kern() with the correct struct net
pointer, and use that struct net pointer when verifying the
address passed to rds_bind().

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Fri, 7 Aug 2015 07:19:09 +0000 (00:19 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-08-05

This series contains updates to i40e, i40evf and e1000e.

Anjali adds support for x772 devices to i40e and i40evf.  With the added
support, x772 supports offloading of the outer UDP transmit and receive
checksum for tunneled packets.  Also supports evicting ATR filters in the
hardware, so update the driver with this new feature set.

Raanan provides several fixes for e1000e, first rectifies the Energy
Efficient Ethernet in Sx code so that it only applies to parts that
actually support EEE in Sx.  Fix whitespace and moved ICH8 related define
to the proper context.  Fixed the ASPM locking which was reported by
Bjorn Helgaas.  Fix a workaround implementation for systime which could
experience a large non-linear increment of the systime value when
checking for overflow.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_dbg_ratelimited: turn into no-op when !DEBUG
Jason A. Donenfeld [Tue, 4 Aug 2015 16:26:19 +0000 (18:26 +0200)]
net_dbg_ratelimited: turn into no-op when !DEBUG

The pr_debug family of functions turns into a no-op when -DDEBUG is not
specified, opting instead to call "no_printk", which gets compiled to a
no-op (but retains gcc's nice warnings about printf-style arguments).

The problem with net_dbg_ratelimited is that it is defined to be a
variant of net_ratelimited_function, which expands to essentially:

    if (net_ratelimit())
        pr_debug(fmt, ...);

When DEBUG is not defined, then this becomes,

    if (net_ratelimit())
        ;

This seems benign, except it isn't. Firstly, there's the obvious
overhead of calling net_ratelimit needlessly, which does quite some book
keeping for the rate limiting. Given that the pr_debug and
net_dbg_ratelimited family of functions are sprinkled liberally through
performance critical code, with developers assuming they'll be compiled
out to a no-op most of the time, we certainly do not want this needless
book keeping. Secondly, and most visibly, even though no debug message
is printed when DEBUG is not defined, if there is a flood of
invocations, dmesg winds up peppered with messages such as
"net_ratelimit: 320 callbacks suppressed". This is because our
aforementioned net_ratelimit() function actually prints this text in
some circumstances. It's especially odd to see this when there isn't any
other accompanying debug message.

So, in sum, it doesn't make sense to have this function's current
behavior, and instead it should match what every other debug family of
functions in the kernel does with !DEBUG -- nothing.

This patch replaces calls to net_dbg_ratelimited when !DEBUG with
no_printk, keeping with the idiom of all the other debug print helpers.

Also, though not strictly neccessary, it guards the call with an if (0)
so that all evaluation of any arguments are sure to be compiled out.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoaf_mpls: add null dev check in find_outdev
Roopa Prabhu [Tue, 4 Aug 2015 13:36:24 +0000 (06:36 -0700)]
af_mpls: add null dev check in find_outdev

This patch adds null dev check for the 'cfg->rc_via_table ==
NEIGH_LINK_TABLE or dev_get_by_index() failed' case

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'test-bpf-next'
David S. Miller [Fri, 7 Aug 2015 05:02:32 +0000 (22:02 -0700)]
Merge branch 'test-bpf-next'

Nicolas Schichan says:

====================
test_bpf improvements

Please find below the patch series with my latest changes to test_bpf.

The first patch checks for unexpected NULL generated skbs before
running the filter.

The second patch adds the possibility for tests to generate fragmented
skbs.

The third patch tests LD_ABS and LD_IND on fragmented skbs.

The fourth patch adds the possibility to restrict the tests being run
by specifying the name/id/range of the test(s) to run via module
parameters.

The fifth patch tests LD_ABS and LD_IND on non fragmented skbs with
various sizes and alignments.

The sixth and final patch checks that the interpreter or JIT correctly
resets A and X to 0.

This serie is against today's net-next tree.

Changes in V2:

* move declaration of 'ptr' in if() block in patch 2/6.

* fix various typos in patch 4/6

* rework default init of test_range array and cleanup exclude_test()
  return condition in patch 4/6.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: add tests checking that JIT/interpreter sets A and X to 0.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:12 +0000 (15:19 +0200)]
test_bpf: add tests checking that JIT/interpreter sets A and X to 0.

It is mandatory for the JIT or interpreter to reset the A and X
registers to 0 before running the filter. Check that it is the case on
various ALU and JMP instructions.

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: add more tests for LD_ABS and LD_IND.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:11 +0000 (15:19 +0200)]
test_bpf: add more tests for LD_ABS and LD_IND.

This exerces the LD_ABS and LD_IND instructions for various sizes and
alignments. This also checks that X when used as an offset to a
BPF_IND instruction first in a filter is correctly set to 0.

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: add module parameters to filter the tests to run.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:10 +0000 (15:19 +0200)]
test_bpf: add module parameters to filter the tests to run.

When developping on the interpreter or a particular JIT, it can be
interesting to restrict the tests list to a specific test or a
particular range of tests.

This patch adds the following module parameters to the test_bpf module:

* test_name=<string>: only the specified named test will be run.

* test_id=<number>: only the test with the specified id will be run
  (see the output of test_bpf without parameters to get the test id).

* test_range=<number>,<number>: only the tests within IDs in the
  specified id range are run (see the output of test_bpf without
  parameters to get the test ids).

Any invalid range, test id or test name will result in -EINVAL being
returned and no tests being run.

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: test LD_ABS and LD_IND instructions on fragmented skbs.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:09 +0000 (15:19 +0200)]
test_bpf: test LD_ABS and LD_IND instructions on fragmented skbs.

These new tests exercise various load sizes and offsets crossing the
head/fragment boundary.

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: allow tests to specify an skb fragment.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:08 +0000 (15:19 +0200)]
test_bpf: allow tests to specify an skb fragment.

This introduce a new test->aux flag (FLAG_SKB_FRAG) to tell the
populate_skb() function to add a fragment to the test skb containing
the data specified in test->frag_data).

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotest_bpf: avoid oopsing the kernel when generate_test_data() fails.
Nicolas Schichan [Tue, 4 Aug 2015 13:19:07 +0000 (15:19 +0200)]
test_bpf: avoid oopsing the kernel when generate_test_data() fails.

Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mlx5e-next'
David S. Miller [Fri, 7 Aug 2015 05:00:59 +0000 (22:00 -0700)]
Merge branch 'mlx5e-next'

Amir Vadai says:

====================
net/mlx5e: Driver updates 04-Aug-2015

This patchset introduces two features to the ConnectX-4 driver: Patch 8/8
("Support physical port counters") exposes some hardware counters through
ethtool. Rest of the patches are preparation and usage of what we call
light-weight netdev open/close. Some flows that used to be in the ndo_open/stop
are moved to the PCI probe/remove flows - i.e. we will make the netdev
open/close operations more "light-weight".

The benefits of this change are:
1) Reduce the execution time of the stop/open operations.
2) Avoid saving SW shadows of resource configurations that must
   persist through stop/open operations (e.g flow table steering
   rules), and avoid deleting/applying them from/to the device upon
   netdev stop/open.
3) Avoid synchronizing threads that access those resources with the
   netdev stop/open threads.

Instead of create/destroy the resource during netdev open/stop, This patchset
changes the behavior such that upon netdev stop, traffic is redirected to a
"Drop RQ" (a RQ that silently drops, at the NIC HW level all incoming traffic).
After redirecting the traffic, RX/TX software resources could be destroyed.
During netdev open, the RX/TX rings are created and traffic is redirected to
the RX rings.

Patchset was applied and tested over commit ba7591d ("ebpf: add skb->hash to
offset map for usage in {cls, act}_bpf or filters")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Support physical port counters
Gal Pressman [Tue, 4 Aug 2015 11:05:47 +0000 (14:05 +0300)]
net/mlx5_core: Support physical port counters

Added physical port counters in the following standard formats to
ethtool statistics:
  - IEEE 802.3
  - RFC2863
  - RFC2819

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Take advantage of the light-weight netdev open/stop
Achiad Shochat [Tue, 4 Aug 2015 11:05:46 +0000 (14:05 +0300)]
net/mlx5e: Take advantage of the light-weight netdev open/stop

Now that TIRs, TISs and flow tables are kept alive while the netdev is
stopped (after executing ndo_stop()) we can do the following
improvements:

- Obsolete the active_vlans SW shadow.
- Do not delete/add flow table rules upon ndo_stop/open.
  In addition to simplifying the flow, this change also fastens
  the ndo_open/close operations.
- Obsolete synchronization of threads accessing the flow tables
  with the netdev stop/open threads.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Disable async events before unregister_netdev()
Achiad Shochat [Tue, 4 Aug 2015 11:05:45 +0000 (14:05 +0300)]
net/mlx5e: Disable async events before unregister_netdev()

It does not make sense to allow events while the netdev is
unregistered.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Rename/move functions following the ndo_stop flow change
Achiad Shochat [Tue, 4 Aug 2015 11:05:44 +0000 (14:05 +0300)]
net/mlx5e: Rename/move functions following the ndo_stop flow change

Rename some functions that used to be invoked upon ndo_open/stop and
are now invoked upon create/destroy_netdev() in order to better hint
their place in the flow.

Change some functions location in the file so that functions involved
in ndo_open/stop flow will not be interleaved with other functions.

This is a cosmetic change, no logical change here.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Light-weight netdev open/stop
Achiad Shochat [Tue, 4 Aug 2015 11:05:43 +0000 (14:05 +0300)]
net/mlx5e: Light-weight netdev open/stop

Create/destroy TIRs, TISs and flow tables upon PCI probe/remove rather
than upon the netdev ndo_open/stop.

Upon ndo_stop(), redirect all RX traffic to the (lately introduced)
"Drop RQ" and then close only the RX/TX rings, leaving the TIRs,
TISs and flow tables alive.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Introduce access function to modify RSS/LRO params
Achiad Shochat [Tue, 4 Aug 2015 11:05:42 +0000 (14:05 +0300)]
net/mlx5_core: Introduce access function to modify RSS/LRO params

To be used by the mlx5 Eth driver in following commit.

This is in preparation for netdev "light-weight" open/stop flow
change described in previous commit.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Introduce the "Drop RQ"
Achiad Shochat [Tue, 4 Aug 2015 11:05:41 +0000 (14:05 +0300)]
net/mlx5e: Introduce the "Drop RQ"

RX traffic routed to this RQ will be silently dropped, at the NIC HW
level.

This is in preparation for netdev "light-weight" open/stop flow
change described in previous commit.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Unify the RX flow
Achiad Shochat [Tue, 4 Aug 2015 11:05:40 +0000 (14:05 +0300)]
net/mlx5e: Unify the RX flow

Generally an RX packet flows through the following objects:
Flow table --> TIR --> RQT --> RQ

Where:
- TIR stands for "Transport Interface Receive", defining the RSS and
  LRO paramaters.
- RQT stands for "RQ Table", implementing the RSS indirection table.
- RQ stands for "Receive Queue"

For flows that do not need LRO, nor RSS, the driver made a shortcut to
the above RX flow by pointing to the RQ directly from the TIR, yielding
this flow:
Flow table --> TIR --> RQ

In this commit we remove this shortcut by "inserting" a single-RQ RQT
between the TIR and the RQ, i.e RX packets will reach the same RQ but
will go through an RQT of size 1, pointing to just a single RQ.

This way the RX traffic re-direction to/from the "Drop RQ" will be more
uniform (AKA "one flow"), as it will involve only RQTs re-direction and
no TIRs re-direction.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cpsw-next'
David S. Miller [Fri, 7 Aug 2015 04:59:27 +0000 (21:59 -0700)]
Merge branch 'cpsw-next'

Mugunthan V N says:

====================
CPSW interrupt handling cleanup and performance improvement

This patch series removes the irq controller disable interrupt and
adding a napi for tx event handling which improves the performance by
~180Mbps on dra7-evm

[  5] local 192.168.10.116 port 5001 connected with 192.168.10.165 port 44176
[  5]  0.0-60.0 sec  1.48 GBytes   210 Mbits/sec
[  4] local 192.168.10.116 port 5001 connected with 192.168.10.165 port 33257
[  4]  0.0-60.0 sec  2.71 GBytes   386 Mbits/sec

Changes from initial version:
* Added a patch to have napi only for first interface as there is
  no use of having seperate napis for each interface as the
  interrupt is shared by both interface and only one napi is
  scheduled for each interrupt.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodrivers: net: cpsw: add separate napi for tx
Mugunthan V N [Tue, 4 Aug 2015 10:36:20 +0000 (16:06 +0530)]
drivers: net: cpsw: add separate napi for tx

Instead of processing tx events in isr adding separate napi for
tx which improves performance by ~180Mbps with
omap2plus_defconfig on DRA74x platform. Also cleaning up rx napis
by renaming to napi_rx for better understanding the code.

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodrivers: net: cpsw: dual_emac: simplify napi usage
Mugunthan V N [Tue, 4 Aug 2015 10:36:19 +0000 (16:06 +0530)]
drivers: net: cpsw: dual_emac: simplify napi usage

Since interrupt is shared between the two ethernet interface and
in isr only one napi is scheduled at an instance so having two
napis doesn't make any difference. So making napi also as a
common resource for the dual ethernet interfaces.

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodrivers: net: cpsw: remove disable_irq/enable_irq as irq can be masked from cpsw...
Mugunthan V N [Tue, 4 Aug 2015 10:36:18 +0000 (16:06 +0530)]
drivers: net: cpsw: remove disable_irq/enable_irq as irq can be masked from cpsw itself

CPSW interrupts can be disabled by masking CPSW interrupts and
clearing interrupt by writing appropriate EOI. So removing all
disable_irq/enable_irq as discussed in [1]

[1] http://patchwork.ozlabs.org/patch/492741/

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: small cleanup in inet/inet6_fib_lookup_dev()
Dan Carpenter [Tue, 4 Aug 2015 07:44:22 +0000 (10:44 +0300)]
mpls: small cleanup in inet/inet6_fib_lookup_dev()

We recently changed this code from returning NULL to returning ERR_PTR.
There are some left over NULL assignments which we can remove.  We can
preserve the error code from ip_route_output() instead of always
returning -ENODEV.  Also these functions use a mix of gotos and direct
returns.  There is no cleanup necessary so I changed the gotos to
direct returns.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'bnx2x-cnic-bnx2fc-bd-support'
David S. Miller [Fri, 7 Aug 2015 04:54:13 +0000 (21:54 -0700)]
Merge branch 'bnx2x-cnic-bnx2fc-bd-support'

Yuval Mintz says:

====================
bnx2x, cnic, bnx2fc: add support for BD

Commit 230d00eb4bfe ("bnx2x: new Multi-function mode - BD") added support
for a new multi-function mode, but it added only the support required by
bnx2x for L2 interfaces.

This adds the required changes to support the new multi-function mode in
the offloaded storage protocols.

Dave,

Please consider applying this series to `net-next'.

Do notice that this involves non-networking driver changes -
but sending this as a single series seemed like the best approach as
we had to have bnx2x changes to support the new functionality.
If this is problematic, please tell us what's the preferred solution here.

Changes from previous versions
------------------------------

 - From v1 - no actual changes; v1 failed to reach netdev so in order to
   keep things in line I've termed this one v2.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2fc: Read npiv table from nvram and create vports.
Joe Carnuccio [Tue, 4 Aug 2015 06:37:30 +0000 (09:37 +0300)]
bnx2fc: Read npiv table from nvram and create vports.

Signed-off-by: Joe Carnuccio <joe.carnuccio@qlogic.com>
Signed-off-by: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2x: Add BD support for storage
Yuval Mintz [Tue, 4 Aug 2015 06:37:29 +0000 (09:37 +0300)]
bnx2x: Add BD support for storage

Commit 230d00eb4bfe ("bnx2x: new Multi-function mode - BD") adds support
for the new mode in bnx2x. This expands this support by implementing
APIs required by our storage drivers to support that mode.

Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocnic: Add the interfaces to get FC-NPIV table.
Adheer Chandravanshi [Tue, 4 Aug 2015 06:37:28 +0000 (09:37 +0300)]
cnic: Add the interfaces to get FC-NPIV table.

Signed-off-by: Adheer Chandravanshi <adheer.chandravanshi@qlogic.com>
Signed-off-by: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocnic: Populate upper layer driver state in MFW
Tej Parkash [Tue, 4 Aug 2015 06:37:27 +0000 (09:37 +0300)]
cnic: Populate upper layer driver state in MFW

Signed-off-by: Tej Parkash <tej.parkash@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: use netdev_err after register_netdev
Scott Feldman [Tue, 4 Aug 2015 05:31:18 +0000 (22:31 -0700)]
rocker: use netdev_err after register_netdev

After successful register_netdev, we can use netdev_err rather the more
generic dev_err.

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: NULL port if port probe fails
Scott Feldman [Tue, 4 Aug 2015 05:31:17 +0000 (22:31 -0700)]
rocker: NULL port if port probe fails

Set port to NULL if port probe fails so we don't try to remove partially
initialized port on port probe err cleanup path.

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoe1000e: Increase driver version number
Raanan Avargil [Sun, 19 Jul 2015 13:33:21 +0000 (16:33 +0300)]
e1000e: Increase driver version number

Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: Fix tight loop implementation of systime read algorithm
Raanan Avargil [Sun, 19 Jul 2015 13:33:20 +0000 (16:33 +0300)]
e1000e: Fix tight loop implementation of systime read algorithm

Change the algorithm. Read systimel twice and check for overflow.
If there was no overflow, use the first value.
If there was an overflow, read systimeh again and use the second
systimel value.

Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: Fix incorrect ASPM locking
Raanan Avargil [Mon, 6 Jul 2015 14:57:36 +0000 (17:57 +0300)]
e1000e: Fix incorrect ASPM locking

This patch fixes wrong locking usage.
In the context of slot reset, we should use lock.
And during resume, there is no need of lock.

Reported-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: Cosmetic changes
Raanan Avargil [Mon, 6 Jul 2015 13:58:54 +0000 (16:58 +0300)]
e1000e: Cosmetic changes

1) Replace spaces with tab.
2) Move ich8lan related define to the proper context.

Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: Fix EEE in Sx implementation
Raanan Avargil [Mon, 6 Jul 2015 13:48:00 +0000 (16:48 +0300)]
e1000e: Fix EEE in Sx implementation

This patch implements the EEE in Sx code so that it only applies to parts
that support EEE in Sx (as opposed to all parts that support EEE).
It also uses the existing eee_advert and eee_lp_abiliity to set just the
bits (100/1000) that should be set.

Signed-off-by: Raanan Avargil <raanan.avargil@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Add AQ commands for NVM Update for X722
Shannon Nelson [Fri, 5 Jun 2015 16:20:34 +0000 (12:20 -0400)]
i40e: Add AQ commands for NVM Update for X722

X722 does NVM update via the adminq queue, so we need to add support for
that.

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add ATR HW eviction support for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:33 +0000 (12:20 -0400)]
i40e/i40evf: Add ATR HW eviction support for X722

X722 supports evicting ATR filters in the HW. With this patch, we enable
the feature in the driver and avoid filter deletion by the driver.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Add IWARP support for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:32 +0000 (12:20 -0400)]
i40e: Add IWARP support for X722

X722 supports IWARP, this patch handles checking for PE critical errors.
Since the driver doesn't support the IWARP interface for now, this patch
just does bare minimum to log a message oif a PE critical error
happens.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add TX/RX outer UDP checksum support for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:31 +0000 (12:20 -0400)]
i40e/i40evf: Add TX/RX outer UDP checksum support for X722

X722 supports offloading of outer UDP TX and RX checksum for tunneled
packets. This patch exposes the support and leaves it enabled by
default.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add support for writeback on ITR feature for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:30 +0000 (12:20 -0400)]
i40e/i40evf: Add support for writeback on ITR feature for X722

X722 fixes an issue from X710 where TX descriptor WB would not happen if
the interrupts were disabled. In order for the write backs to happen a
bit needs to be set in the dynamic interrupt control register called
WB_ON_ITR. With this feature, the SW driver need not arm SW interrupts to
work around the issue in X710.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: RSS changes for X722
Anjali Singhai Jain [Tue, 23 Jun 2015 23:00:04 +0000 (19:00 -0400)]
i40e/i40evf: RSS changes for X722

X722 uses the admin queue to configure RSS. This patch adds the necessary
flow changes to configure RSS through AQ. It also adds the separate VMDQ2
lookup tables and hash key programming for X722.

X722 also exposes a different set of PCTYPES for RSS, this patch
accommodates those changes.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@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: Update register.h file for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:28 +0000 (12:20 -0400)]
i40e/i40evf: Update register.h file for X722

Update the i40e and i40evf register.h file with the registers for X722.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Update FW API with X722 support
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:27 +0000 (12:20 -0400)]
i40e/i40evf: Update FW API with X722 support

This patch does the firmware API update to support the new X722 device.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add flags for X722 capabilities
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:26 +0000 (12:20 -0400)]
i40e/i40evf: Add flags for X722 capabilities

Add capabilities flags specific to X722.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add device ids for X722
Anjali Singhai Jain [Fri, 5 Jun 2015 16:20:25 +0000 (12:20 -0400)]
i40e/i40evf: Add device ids for X722

Adding device ids for new hardware X722

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Jim Young <james.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Wed, 5 Aug 2015 06:57:45 +0000 (23:57 -0700)]
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 net-next, they are:

1) A couple of cleanups for the netfilter core hook from Eric Biederman.

2) Net namespace hook registration, also from Eric. This adds a dependency with
   the rtnl_lock. This should be fine by now but we have to keep an eye on this
   because if we ever get the per-subsys nfnl_lock before rtnl we have may
   problems in the future. But we have room to remove this in the future by
   propagating the complexity to the clients, by registering hooks for the init
   netns functions.

3) Update nf_tables to use the new net namespace hook infrastructure, also from
   Eric.

4) Three patches to refine and to address problems from the new net namespace
   hook infrastructure.

5) Switch to alternate jumpstack in xtables iff the packet is reentering. This
   only applies to a very special case, the TEE target, but Eric Dumazet
   reports that this is slowing down things for everyone else. So let's only
   switch to the alternate jumpstack if the tee target is in used through a
   static key. This batch also comes with offline precalculation of the
   jumpstack based on the callchain depth. From Florian Westphal.

6) Minimal SCTP multihoming support for our conntrack helper, from Michal
   Kubecek.

7) Reduce nf_bridge_info per skbuff scratchpad area to 32 bytes, from Florian
   Westphal.

8) Fix several checkpatch errors in bridge netfilter, from Bernhard Thaler.

9) Get rid of useless debug message in ip6t_REJECT, from Subash Abhinov.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonetfilter: ip6t_REJECT: Remove debug messages from reject_tg6()
Subash Abhinov Kasiviswanathan [Thu, 30 Jul 2015 16:53:45 +0000 (16:53 +0000)]
netfilter: ip6t_REJECT: Remove debug messages from reject_tg6()

Make it similar to reject_tg() in ipt_REJECT.

Suggested-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
9 years agoMerge branch 'cxgb4-next'
David S. Miller [Tue, 4 Aug 2015 08:24:35 +0000 (01:24 -0700)]
Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
add meminfo, bist status and misc. fixes

This patch series adds the following.
Add support to dump memory address range of various hw modules
Add support to dump edc bist status during ecc error
Read correct bits of who am i register for T6 adapter
and update T6 register range

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

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

V2: PATCH 3/4 ("cxgb4/cxgb4vf: read the correct bits of PL Who Am I
    register") Fix switch statement in get_chip_type() and some more style
    fixes based on review comment by Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Update T6 register ranges
Hariprasad Shenai [Tue, 4 Aug 2015 09:06:20 +0000 (14:36 +0530)]
cxgb4: Update T6 register ranges

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4/cxgb4vf: read the correct bits of PL Who Am I register
Hariprasad Shenai [Tue, 4 Aug 2015 09:06:19 +0000 (14:36 +0530)]
cxgb4/cxgb4vf: read the correct bits of PL Who Am I register

Read the correct bits of PL Who Am I for the Source PF field which has
changed in T6

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add support to dump edc bist status
Hariprasad Shenai [Tue, 4 Aug 2015 09:06:18 +0000 (14:36 +0530)]
cxgb4: Add support to dump edc bist status

Add support to dump edc bist status for ECC data errors

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add debugfs support to dump meminfo
Hariprasad Shenai [Tue, 4 Aug 2015 09:06:17 +0000 (14:36 +0530)]
cxgb4: Add debugfs support to dump meminfo

Add debug support to dump memory address ranges of various hardware
modules of the adapter.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Use definition for reserved label checks
Robert Shearman [Mon, 3 Aug 2015 16:50:04 +0000 (17:50 +0100)]
mpls: Use definition for reserved label checks

In multiple locations there are checks for whether the label in hand
is a reserved label or not using the arbritray value of 16. Factor
this out into a #define for better maintainability and for
documentation.

Signed-off-by: Robert Shearman <rshearma@brocade.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'lwtunnel-encap-local'
David S. Miller [Tue, 4 Aug 2015 05:26:14 +0000 (22:26 -0700)]
Merge branch 'lwtunnel-encap-local'

Robert Shearman says:

====================
lwtunnel: encap locally-generated ipv4 packets

Locally-generated IPv4 packets, such as from applications running on
the host or traceroute/ping currently don't have lwtunnel output
redirected encap applied. However, they should do in the same way as
for forwarded packets and this patch series addresses that.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv4: apply lwtunnel encap for locally-generated packets
Robert Shearman [Mon, 3 Aug 2015 16:39:21 +0000 (17:39 +0100)]
ipv4: apply lwtunnel encap for locally-generated packets

lwtunnel encap is applied for forwarded packets, but not for
locally-generated packets. This is because the output function is not
overridden in __mkroute_output, unlike it is in __mkroute_input.

The lwtunnel state is correctly set on the rth through the call to
rt_set_nexthop, so all that needs to be done is to override the dst
output function to be lwtunnel_output if there is lwtunnel state
present and it requires output redirection.

Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agolwtunnel: set skb protocol and dev
Robert Shearman [Mon, 3 Aug 2015 16:39:20 +0000 (17:39 +0100)]
lwtunnel: set skb protocol and dev

In the locally-generated packet path skb->protocol may not be set and
this is required for the lwtunnel encap in order to get the lwtstate.

This would otherwise have been set by ip_output or ip6_output so set
skb->protocol prior to calling the lwtunnel encap
function. Additionally set skb->dev in case it is needed further down
the transmit path.

Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: mdb: fix vlan_enabled access when vlans are not configured
Nikolay Aleksandrov [Mon, 3 Aug 2015 23:19:58 +0000 (01:19 +0200)]
bridge: mdb: fix vlan_enabled access when vlans are not configured

Instead of trying to access br->vlan_enabled directly use the provided
helper br_vlan_enabled().

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoact_bpf: properly support late binding of bpf action to a classifier
Daniel Borkmann [Mon, 3 Aug 2015 14:21:57 +0000 (16:21 +0200)]
act_bpf: properly support late binding of bpf action to a classifier

Since the introduction of the BPF action in d23b8ad8ab23 ("tc: add BPF
based action"), late binding was not working as expected. I.e. setting
the action part for a classifier only via 'bpf index <num>', where <num>
is the index of an existing action, is being rejected by the kernel due
to other missing parameters.

It doesn't make sense to require these parameters such as BPF opcodes
etc, as they are not going to be used anyway: in this case, they're just
allocated/parsed and then freed again w/o doing anything meaningful.

Instead, parse and verify the remaining parameters *after* the test on
tcf_hash_check(), when we really know that we're dealing with creation
of a new action or replacement of an existing one and where late binding
is thus irrelevant.

After patch, test case is now working:

  FOO="1,6 0 0 4294967295,"
  tc actions add action bpf bytecode "$FOO"
  tc filter add dev foo parent 1: bpf bytecode "$FOO" flowid 1:1 action bpf index 1
  tc actions show action bpf
    action order 0: bpf bytecode '1,6 0 0 4294967295' default-action pipe
    index 1 ref 2 bind 1
  tc filter show dev foo
    filter protocol all pref 49152 bpf
    filter protocol all pref 49152 bpf handle 0x1 flowid 1:1 bytecode '1,6 0 0 4294967295'
    action order 1: bpf bytecode '1,6 0 0 4294967295' default-action pipe
    index 1 ref 2 bind 1

Late binding of a BPF action can be useful for preloading maps (e.g. before
they hit traffic) in case of eBPF programs, or to share a single eBPF action
with multiple classifiers.

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: dsa: mv88e6xxx: call _mv88e6xxx_stats_wait with SMI lock held
Vivien Didelot [Mon, 3 Aug 2015 13:17:44 +0000 (09:17 -0400)]
net: dsa: mv88e6xxx: call _mv88e6xxx_stats_wait with SMI lock held

At switch setup, _mv88e6xxx_stats_wait was called without holding the
SMI mutex. Fix this by requesting the lock for this call.

Also, return the _mv88e6xxx_stats_wait code, since it may fail.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: mdb: add/del entry on all vlans if vlan_filter is enabled and vid is 0
Satish Ashok [Mon, 3 Aug 2015 11:29:16 +0000 (13:29 +0200)]
bridge: mdb: add/del entry on all vlans if vlan_filter is enabled and vid is 0

Before this patch when a vid was not specified, the entry was added with
vid 0 which is useless when vlan_filtering is enabled. This patch makes
the entry to be added on all configured vlans when vlan filtering is
enabled and respectively deleted from all, if the entry vid is 0.
This is also closer to the way fdb works with regard to vid 0 and vlan
filtering.

Example:
Setup:
$ bridge vlan add vid 256 dev eth4
$ bridge vlan add vid 1024 dev eth4
$ bridge vlan add vid 64 dev eth3
$ bridge vlan add vid 128 dev eth3
$ bridge vlan
port vlan ids
eth3  1 PVID Egress Untagged
 64
 128

eth4  1 PVID Egress Untagged
 256
 1024
$ echo 1 > /sys/class/net/br0/bridge/vlan_filtering

Before:
$ bridge mdb add dev br0 port eth3 grp 239.0.0.1
$ bridge mdb
dev br0 port eth3 grp 239.0.0.1 temp

After:
$ bridge mdb add dev br0 port eth3 grp 239.0.0.1
$ bridge mdb
dev br0 port eth3 grp 239.0.0.1 temp vid 1
dev br0 port eth3 grp 239.0.0.1 temp vid 128
dev br0 port eth3 grp 239.0.0.1 temp vid 64

Signed-off-by: Satish Ashok <sashok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'stacked-vlan-TSO'
David S. Miller [Mon, 3 Aug 2015 21:24:50 +0000 (14:24 -0700)]
Merge branch 'stacked-vlan-TSO'

Toshiaki Makita says:

====================
Stacked vlan TSO for virtual devices

Basically virtual devices do not need to segment double tagged packets.
This patch set adds TSO feature for double tagged packets to several
virtual devices, which can be realized by simply setting
.ndo_features_check to passthru_features_check.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotuntap: Don't segment multiple tagged packets on tap device
Toshiaki Makita [Fri, 31 Jul 2015 06:03:27 +0000 (15:03 +0900)]
tuntap: Don't segment multiple tagged packets on tap device

Tap devices don't need to segment multiple tagged packets.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: Don't segment multiple tagged packets on bridge device
Toshiaki Makita [Fri, 31 Jul 2015 06:03:26 +0000 (15:03 +0900)]
bridge: Don't segment multiple tagged packets on bridge device

Bridge devices don't need to segment multiple tagged packets since thier
ports can segment them.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoveth: Don't segment multiple tagged packets on veth device
Toshiaki Makita [Fri, 31 Jul 2015 06:03:25 +0000 (15:03 +0900)]
veth: Don't segment multiple tagged packets on veth device

Veth devices don't need to segment multiple tagged packets.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomacvlan: Don't segment multiple tagged packets on macvlan device
Toshiaki Makita [Fri, 31 Jul 2015 06:03:24 +0000 (15:03 +0900)]
macvlan: Don't segment multiple tagged packets on macvlan device

Macvlan/macvtap devices don't need to segment multiple tagged packets
since the lower devices can segment them.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovirtio_net: add gro capability
Eric Dumazet [Fri, 31 Jul 2015 16:25:17 +0000 (18:25 +0200)]
virtio_net: add gro capability

Straightforward patch to add GRO processing to virtio_net.

napi_complete_done() usage allows more aggressive aggregation,
opted-in by setting /sys/class/net/xxx/gro_flush_timeout

Tested:

Setting /sys/class/net/xxx/gro_flush_timeout to 1000 nsec,
Rick Jones reported following results.

One VM of each on a pair of OpenStack compute nodes with E5-2650Lv3 CPUs
and Intel 82599ES-based NICs. So, two "before" and two "after" VMs.
The OpenStack compute nodes were running OpenStack Kilo, with VxLAN
encapsulation being used through OVS so no GRO coming-up the host
stack.  The compute nodes themselves were running a 3.14-based kernel.

Single-stream netperf, CPU utilizations and thus service demands are
based on intra-guest reported CPU.

Throughput Mbit/s, bigger is better
        Min     Median  Average Max
4.2.0-rc3+      1364    1686    1678    1938
4.2.0-rc3+flush1k       1824    2269    2275    2647

Send Service Demand, smaller is better
        Min     Median  Average Max
4.2.0-rc3+      0.236   0.558   0.524   0.802
4.2.0-rc3+flush1k       0.176   0.503   0.471   0.738

Receive Service Demand, smaller is better.
        Min     Median  Average Max
4.2.0-rc3+      1.906   2.188   2.191   2.531
4.2.0-rc3+flush1k       0.448   0.529   0.533   0.692

Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Rick Jones <rick.jones2@hp.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: linearize skb in case frags would not fit into tx descriptor
Jiri Pirko [Sun, 2 Aug 2015 18:56:38 +0000 (20:56 +0200)]
rocker: linearize skb in case frags would not fit into tx descriptor

Suggested-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorocker: enable support for scattered packets
Ido Schimmel [Sun, 2 Aug 2015 18:56:37 +0000 (20:56 +0200)]
rocker: enable support for scattered packets

rocker supports the transmission of scattered packets, so let the kernel
know about it by setting the NETIF_F_SG bit in the device's features.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoebpf: add skb->hash to offset map for usage in {cls, act}_bpf or filters
Daniel Borkmann [Fri, 31 Jul 2015 22:46:29 +0000 (00:46 +0200)]
ebpf: add skb->hash to offset map for usage in {cls, act}_bpf or filters

Add skb->hash to the __sk_buff offset map, so it can be accessed from
an eBPF program. We currently already do this for classic BPF filters,
but not yet on eBPF, it might be useful as a demuxer in combination with
helpers like bpf_clone_redirect(), toy example:

  __section("cls-lb") int ingress_main(struct __sk_buff *skb)
  {
    unsigned int which = 3 + (skb->hash & 7);
    /* bpf_skb_store_bytes(skb, ...); */
    /* bpf_l{3,4}_csum_replace(skb, ...); */
    bpf_clone_redirect(skb, which, 0);
    return -1;
  }

I was thinking whether to add skb_get_hash(), but then concluded the
raw skb->hash seems fine in this case: we can directly access the hash
w/o extra eBPF helper function call, it's filled out by many NICs on
ingress, and in case the entropy level would not be sufficient, people
can still implement their own specific sw fallback hash mix anyway.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>