openwrt/staging/blogic.git
9 years agobatman-adv: Use common Jenkins Hash implementation
Sven Eckelmann [Sun, 1 Mar 2015 08:46:18 +0000 (09:46 +0100)]
batman-adv: Use common Jenkins Hash implementation

An unoptimized version of the Jenkins one-at-a-time hash function is used
and partially copied all over the code wherever an hashtable is used.
Instead the optimized version shared between the whole kernel should be
used to reduce code duplication and use better optimized code.

Only the DAT code must use the old implementation because it is used as
distributed hash function which has to be common for all nodes.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobpf: allow programs to write to certain skb fields
Alexei Starovoitov [Thu, 4 Jun 2015 17:11:54 +0000 (10:11 -0700)]
bpf: allow programs to write to certain skb fields

allow programs read/write skb->mark, tc_index fields and
((struct qdisc_skb_cb *)cb)->data.

mark and tc_index are generically useful in TC.
cb[0]-cb[4] are primarily used to pass arguments from one
program to another called via bpf_tail_call() which can
be seen in sockex3_kern.c example.

All fields of 'struct __sk_buff' are readable to socket and tc_cls_act progs.
mark, tc_index are writeable from tc_cls_act only.
cb[0]-cb[4] are writeable by both sockets and tc_cls_act.

Add verifier tests and improve sample code.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: make programs see skb->data == L2 for ingress and egress
Alexei Starovoitov [Thu, 4 Jun 2015 17:11:53 +0000 (10:11 -0700)]
bpf: make programs see skb->data == L2 for ingress and egress

eBPF programs attached to ingress and egress qdiscs see inconsistent skb->data.
For ingress L2 header is already pulled, whereas for egress it's present.
This is known to program writers which are currently forced to use
BPF_LL_OFF workaround.
Since programs don't change skb internal pointers it is safe to do
pull/push right around invocation of the program and earlier taps and
later pt->func() will not be affected.
Multiple taps via packet_rcv(), tpacket_rcv() are doing the same trick
around run_filter/BPF_PROG_RUN even if skb_shared.

This fix finally allows programs to use optimized LD_ABS/IND instructions
without BPF_LL_OFF for higher performance.
tc ingress + cls_bpf + samples/bpf/tcbpf1_kern.o
       w/o JIT   w/JIT
before  20.5     23.6 Mpps
after   21.8     26.6 Mpps

Old programs with BPF_LL_OFF will still work as-is.

We can now undo most of the earlier workaround commit:
a166151cbe33 ("bpf: fix bpf helpers to use skb->mac_header relative offsets")

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: remove redundant checks II
Eric Dumazet [Thu, 4 Jun 2015 15:01:12 +0000 (08:01 -0700)]
tcp: remove redundant checks II

For same reasons than in commit 12e25e1041d0 ("tcp: remove redundant
checks"), we can remove redundant checks done for timewait sockets.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofddi: print an address with %p format specifier rather than %x
Colin Ian King [Fri, 5 Jun 2015 18:19:11 +0000 (19:19 +0100)]
fddi: print an address with %p format specifier rather than %x

The debug is printing the struct smt_header * address using
the %x format specifier. Fix it to use %p instead.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agowan: dscc4: fix build warning Wunused-but-set-variable
Nicholas Mc Guire [Sat, 6 Jun 2015 08:41:17 +0000 (10:41 +0200)]
wan: dscc4: fix build warning Wunused-but-set-variable

Fix:
drivers/net/wan/dscc4.c: In function 'dscc4_open':
drivers/net/wan/dscc4.c:1049:25: warning: variable 'ppriv' set but not used
[-Wunused-but-set-variable]

This has been in there unused since 1da177e4c3f (Linux-2.6.12-rc2) simply
remove it.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoinet: add IP_BIND_ADDRESS_NO_PORT to overcome bind(0) limitations
Eric Dumazet [Sun, 7 Jun 2015 04:17:57 +0000 (21:17 -0700)]
inet: add IP_BIND_ADDRESS_NO_PORT to overcome bind(0) limitations

When an application needs to force a source IP on an active TCP socket
it has to use bind(IP, port=x).

As most applications do not want to deal with already used ports, x is
often set to 0, meaning the kernel is in charge to find an available
port.
But kernel does not know yet if this socket is going to be a listener or
be connected.
It has very limited choices (no full knowledge of final 4-tuple for a
connect())

With limited ephemeral port range (about 32K ports), it is very easy to
fill the space.

This patch adds a new SOL_IP socket option, asking kernel to ignore
the 0 port provided by application in bind(IP, port=0) and only
remember the given IP address.

The port will be automatically chosen at connect() time, in a way
that allows sharing a source port as long as the 4-tuples are unique.

This new feature is available for both IPv4 and IPv6 (Thanks Neal)

Tested:

Wrote a test program and checked its behavior on IPv4 and IPv6.

strace(1) shows sequences of bind(IP=127.0.0.2, port=0) followed by
connect().
Also getsockname() show that the port is still 0 right after bind()
but properly allocated after connect().

socket(PF_INET, SOCK_STREAM, IPPROTO_IP) = 5
setsockopt(5, SOL_IP, IP_BIND_ADDRESS_NO_PORT, [1], 4) = 0
bind(5, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("127.0.0.2")}, 16) = 0
getsockname(5, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("127.0.0.2")}, [16]) = 0
connect(5, {sa_family=AF_INET, sin_port=htons(53174), sin_addr=inet_addr("127.0.0.3")}, 16) = 0
getsockname(5, {sa_family=AF_INET, sin_port=htons(38050), sin_addr=inet_addr("127.0.0.2")}, [16]) = 0

IPv6 test :

socket(PF_INET6, SOCK_STREAM, IPPROTO_IP) = 7
setsockopt(7, SOL_IP, IP_BIND_ADDRESS_NO_PORT, [1], 4) = 0
bind(7, {sa_family=AF_INET6, sin6_port=htons(0), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, 28) = 0
getsockname(7, {sa_family=AF_INET6, sin6_port=htons(0), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, [28]) = 0
connect(7, {sa_family=AF_INET6, sin6_port=htons(57300), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, 28) = 0
getsockname(7, {sa_family=AF_INET6, sin6_port=htons(60964), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, [28]) = 0

I was able to bind()/connect() a million concurrent IPv4 sockets,
instead of ~32000 before patch.

lpaa23:~# ulimit -n 1000010
lpaa23:~# ./bind --connect --num-flows=1000000 &
1000000 sockets

lpaa23:~# grep TCP /proc/net/sockstat
TCP: inuse 2000063 orphan 0 tw 47 alloc 2000157 mem 66

Check that a given source port is indeed used by many different
connections :

lpaa23:~# ss -t src :40000 | head -10
State      Recv-Q Send-Q   Local Address:Port          Peer Address:Port
ESTAB      0      0           127.0.0.2:40000         127.0.202.33:44983
ESTAB      0      0           127.0.0.2:40000         127.2.27.240:44983
ESTAB      0      0           127.0.0.2:40000           127.2.98.5:44983
ESTAB      0      0           127.0.0.2:40000        127.0.124.196:44983
ESTAB      0      0           127.0.0.2:40000         127.2.139.38:44983
ESTAB      0      0           127.0.0.2:40000          127.1.59.80:44983
ESTAB      0      0           127.0.0.2:40000          127.3.6.228:44983
ESTAB      0      0           127.0.0.2:40000          127.0.38.53:44983
ESTAB      0      0           127.0.0.2:40000         127.1.197.10:44983

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Fix static checker warning
Hariprasad Shenai [Fri, 5 Jun 2015 09:06:33 +0000 (14:36 +0530)]
cxgb4: Fix static checker warning

The patch e85c9a7abfa4: ("cxgb4/cxgb4vf: Add code to calculate T5 BAR2
Offsets for SGE Queue Registers") from Dec 3, 2014, leads to the
following static checker warning:

        drivers/net/ethernet/chelsio/cxgb4/t4_hw.c:5358
t4_bar2_sge_qregs()
        warn: should '(qid >> qpp_shift) << page_shift' be a 64 bit type?

This patch fixes it

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cxgb4-next'
David S. Miller [Sat, 6 Jun 2015 04:26:07 +0000 (21:26 -0700)]
Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
Free VI, flush sge ec and some other misc. fixes

This patch series adds the following.
Free VI interface during remove, flush SGE ec routine, rename
t4_link_start to t4_link_l1cfg since it only does l1 configuration, set
mac addr from when we can't contact firmware for debug purpose, set pcie
completion timeout and use fw interface to access TP_PIO_XXX registers

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

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

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Use FW LDST cmd to access TP_PIO_{ADDR, DATA} register first
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:52 +0000 (14:24 +0530)]
cxgb4: Use FW LDST cmd to access TP_PIO_{ADDR, DATA} register first

The TP_PIO_{ADDR,DATA} registers are are in conflict with the firmware's
use of these registers. Added a routine to access it through FW LDST
cmd.
Access all TP_PIO_{ADDR,DATA} register access through new routine if FW
is alive. If firmware is dead, than fall back to indirect access.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: program pci completion timeout
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:51 +0000 (14:24 +0530)]
cxgb4: program pci completion timeout

Set pci completion timeout to 0xd.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Set mac addr from vpd, when we can't contact firmware
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:50 +0000 (14:24 +0530)]
cxgb4: Set mac addr from vpd, when we can't contact firmware

Grab the Adapter MAC Address out of the VPD and use it for the "debug"
network interface when either we can't contact the firmware

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Rename t4_link_start() to t4_link_l1cfg
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:49 +0000 (14:24 +0530)]
cxgb4: Rename t4_link_start() to t4_link_l1cfg

t4_link_start() was completely misnamed.  It does _not_ start up the
link. It merely does the L1 Configuration for the link. The Link Up
process is started automatically by the firmware when the number of
enabled Virtual Interfaces on a port goes from 0 to 1. So renaming
this routine to t4_link_l1cfg() for better documentation.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add sge ec context flush service
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:48 +0000 (14:24 +0530)]
cxgb4: Add sge ec context flush service

Add function to flush the sge ec context cache, and utilize
this new function in the driver

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Free Virtual Interfaces in remove routine
Hariprasad Shenai [Fri, 5 Jun 2015 08:54:47 +0000 (14:24 +0530)]
cxgb4: Free Virtual Interfaces in remove routine

Free VI interfaces in remove routine. If we don't do this then the
firmware will never drop the physical link to the peer.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mlx5-next'
David S. Miller [Thu, 4 Jun 2015 23:41:02 +0000 (16:41 -0700)]
Merge branch 'mlx5-next'

Or Gerlitz says:

====================
mlx5: Add Interface Step Sequence ID support

ISSI (Interface Step Sequence ID) defines the step sequence ID of the
interface between the driver to the firmware and is incremented by
steps of one. ISSI is used to enable deprecating/modifying features,
command interfaces and such, while maintaining compatibility.

As the driver serves both ConnectIB (CIB) and ConnectX4, we carefully
made sure that the IB functionality keeps running also on older CIB
firmware releases that don't support ISSI.

The Ethernet functionailty is available only on ConnectX4 where all
firmware releases support the feature since the very basic ISSI level.
So at this point no need for compatility code there.

As done prior to this series, when the Ethernet functionlity is enabled,
during the initialization flow, the core driver performs a query of the
supported ISSIs using the QUERY_ISSI command, and then, if ISSI is supported,
sets the actual issi value informing the firmware on which ISSI level to run,
using SET_ISSI command.

Previously, the IB driver wasn't ready to work on that mode, and hence
building both the IB driver and the Ethernet functionality in the core
driver were disallowed by Kconfigs, with this series, we allow users to
enable them both.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlx5: Enable mutual support for IB and Ethernet
Haggai Abramonvsky [Thu, 4 Jun 2015 16:30:48 +0000 (19:30 +0300)]
mlx5: Enable mutual support for IB and Ethernet

Ethernet functionality is only available when working in ISSI > 0 mode.

Previously, the IB driver wasn't ready to work on that mode, and hence
building both the IB driver and the Ethernet functionality in the core
driver were disallowed by Kconfigs.

Now, once we have all the pre-steps in place, we can remove this limitation.

The last steps in the IB driver for getting that setup to work are:
create dummy SRQ for the driver's use (until now we could use XRC_SRQ
as SRQ and XRC_SRQ, after moving to ISSI > 0, we separate XRC SRQs from
basic SRQs) and adapt the create QP function to be compatible with ISSI > 0.

Signed-off-by: Haggai Abramovsky <hagaya@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoIB/mlx5: Don't create IB instance over Ethernet ports
Majd Dibbiny [Thu, 4 Jun 2015 16:30:47 +0000 (19:30 +0300)]
IB/mlx5: Don't create IB instance over Ethernet ports

Since we still don't have RoCE support in mlx5, avoid
creating IB driver instance over Ethernet ports.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoIB/mlx5: Avoid using the MAD_IFC command under ISSI > 0 mode
Majd Dibbiny [Thu, 4 Jun 2015 16:30:46 +0000 (19:30 +0300)]
IB/mlx5: Avoid using the MAD_IFC command under ISSI > 0 mode

In ISSI > 0 mode, most of the MAD_IFC command features are deprecated, and can't
be used. Therefore, when in that mode, we replace all of them with other commands
that provide the required functionality.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Add more query port helpers
Majd Dibbiny [Thu, 4 Jun 2015 16:30:45 +0000 (19:30 +0300)]
net/mlx5_core: Add more query port helpers

Add the following helpers:

1. mlx5_query_port_proto_oper -- queries the port speed port mask
2. mlx5_query_port_link_width_oper - queries the port link with bitmask
3. mlx5_query_port_vl_hw_cap - queries the Virtual Lanes supported on this port

These helpers will be used from the IB driver when working in ISSI > 0 mode.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Use port number when querying port ptys
Majd Dibbiny [Thu, 4 Jun 2015 16:30:44 +0000 (19:30 +0300)]
net/mlx5_core: Use port number when querying port ptys

Until now, mlx5_query_port_ptys always queried port number one.

Added new argument in the function's prototype so we can also query
the second port. This will be needed  when thr helper will be invoked
from the IB driver on non FPP (Function-Per-Port) devices.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Use port number in the query port mtu helpers
Majd Dibbiny [Thu, 4 Jun 2015 16:30:43 +0000 (19:30 +0300)]
net/mlx5_core: Use port number in the query port mtu helpers

Extend the function prototypes for max and operational mtu to take the
local port number. In the Ethernet driver is this hard coded to one,
since ConnectX4 Ethernet devices are always function-per-port.
The IB driver also serves older devices (ConnectIB) which isn't such,
and hence the part can vary.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Get vendor-id using the query adapter command
Majd Dibbiny [Thu, 4 Jun 2015 16:30:42 +0000 (19:30 +0300)]
net/mlx5_core: Get vendor-id using the query adapter command

Add two wrapper functions to the query adapter command:

1. mlx5_query_board_id -- replaces the old mlx5_cmd_query_adapter.

2. mlx5_core_query_vendor_id -- retrieves the vendor_id from the
   query_adapter command.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Add new query HCA vport commands
Majd Dibbiny [Thu, 4 Jun 2015 16:30:41 +0000 (19:30 +0300)]
net/mlx5_core: Add new query HCA vport commands

Added the implementation for the following commands:

1. QUERY_HCA_VPORT_GID
2. QUERY_HCA_VPORT_PKEY
3. QUERY_HCA_VPORT_CONTEXT

They will be needed when we move to work with ISSI > 0 in the IB driver too.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Make the vport helpers available for the IB driver too
Majd Dibbiny [Thu, 4 Jun 2015 16:30:40 +0000 (19:30 +0300)]
net/mlx5_core: Make the vport helpers available for the IB driver too

Move the vport header file to be under include/linux/mlx5, such that
the mlx5 IB can use it as well.

Also add nic_ prefix to the vport NIC commands to differeniate between
HCA vport commands and NIC vport commands.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Check the return bitmask when querying ISSI
Haggai Abramonvsky [Thu, 4 Jun 2015 16:30:39 +0000 (19:30 +0300)]
net/mlx5_core: Check the return bitmask when querying ISSI

The determination of the supported ISSI versions should be conditioned
on the returned mask, and not only on the return status of the query
ISSI command, fix that.

Signed-off-by: Haggai Abramovsky <hagaya@mellanox.com>
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Enable XRCs and SRQs when using ISSI > 0
Haggai Abramonvsky [Thu, 4 Jun 2015 16:30:38 +0000 (19:30 +0300)]
net/mlx5_core: Enable XRCs and SRQs when using ISSI > 0

When working in ISSI > 0 mode, the model exposed by the device for
XRCs and SRQs is different. XRCs use XRC SRQs and plain SRQs are based
on RPM (Receive Memory Pool).

Add helper functions to create, modify, query, and arm XRC SRQs and RMPs.

Signed-off-by: Haggai Abramovsky <hagaya@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_core: Apply proper name convention to helpers
Haggai Abramonvsky [Thu, 4 Jun 2015 16:30:37 +0000 (19:30 +0300)]
net/mlx5_core: Apply proper name convention to helpers

Some core helper functions were named with mlx5_ only prefix, fix that to
mlx5_core_ so we're aligned with the overall scheme used for core services.

Signed-off-by: Haggai Abramovsky <hagaya@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5_en: Add missing check for memory allocation failure
Amir Vadai [Thu, 4 Jun 2015 16:30:36 +0000 (19:30 +0300)]
net/mlx5_en: Add missing check for memory allocation failure

The patch afb736e9330a: "net/mlx5: Ethernet resource handling files"
from May 28, 2015, leads to the following static checker warning:

drivers/net/ethernet/mellanox/mlx5/core/en_flow_table.c:726 mlx5e_create_main_flow_table()
error: potential null dereference 'g'.  (kcalloc returns null)

Fixes: afb736e9330a ("net/mlx5: Ethernet resource handling files")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'flow_key_hashing'
David S. Miller [Thu, 4 Jun 2015 22:44:32 +0000 (15:44 -0700)]
Merge branch 'flow_key_hashing'

Tom Herbert says:

====================
net: Increase inputs to flow_keys hashing

This patch set adds new fields to the flow_keys structure and hashes
over these fields to get a better flow hash. In particular, these
patches now include hashing over the full IPv6 addresses in order
to defend against address spoofing that always results in the
same hash. The new input also includes the Ethertype, L4 protocol,
VLAN, flow label, GRE keyid, and MPLS entropy label.

In order to increase hash inputs, we switch to using jhash2
which operates an an array of u32's. jhash2 operates on multiples of
three words. The data in the hash is constructed for that, and there
are are two variants for IPv4 and Ipv6 addressing. For IPv4 addresses,
jhash is performed over six u32's and for IPv6 it is done over twelve.

flow_keys can store either IPv4 or IPv6 addresses (addr_proto field
is a selector). ipv6_addr_hash is no longer used to convert addresses
for setting in flow table. For legacy uses of flow keys outside of
flow_dissector the flow_get_u32_src and flow_get_u32_dst functions
have been added to get u32 representation representations of addresses
in flow_keys.

For flow lables we also eliminate the short circuit in flow_dissector
for non-zero flow label. The flow label is now considered additional
input to ports.

Testing: Ran netperf TCP_RR for 200 flows using IPv4 and IPv6 comparing
before the patches and with the patches. Did not detect any performance
degradation.

v2:
  - Took out MPLS entropy label. Will add this later.
v3:
  - Ensure hash start offset is a four byte boundary. Add BUG_BUILD_ON
    to check for this.
  - Fixes sparse error in GRE to get entropy from keyid.
v4:
  - Rebase to Jiri changes to generalize flow dissection
  - Support TIPC as its own address
  - Bring back MPLS entropy label dissection
  - Remove FLOW_DISSECTOR_KEY_IPV6_HASH_ADDRS

v5:
  - Minor fixes from feedback

v6:
  - Cleanup and sparse issue with flow label
  - Change keyid to returned by flow_dissector to be __be32
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Add MPLS entropy label in flow_keys
Tom Herbert [Thu, 4 Jun 2015 16:16:46 +0000 (09:16 -0700)]
mpls: Add MPLS entropy label in flow_keys

In flow dissector if an MPLS header contains an entropy label this is
saved in the new keyid field of flow_keys. The entropy label is
then represented in the flow hash function input.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add GRE keyid in flow_keys
Tom Herbert [Thu, 4 Jun 2015 16:16:45 +0000 (09:16 -0700)]
net: Add GRE keyid in flow_keys

In flow dissector if a GRE header contains a keyid this is saved in the
new keyid field of flow_keys. The GRE keyid is then represented
in the flow hash function input.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add IPv6 flow label to flow_keys
Tom Herbert [Thu, 4 Jun 2015 16:16:44 +0000 (09:16 -0700)]
net: Add IPv6 flow label to flow_keys

In flow_dissector set the flow label in flow_keys for IPv6. This also
removes the shortcircuiting of flow dissection when a non-zero label
is present, the flow label can be considered to provide additional
entropy for a hash.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add VLAN ID to flow_keys
Tom Herbert [Thu, 4 Jun 2015 16:16:43 +0000 (09:16 -0700)]
net: Add VLAN ID to flow_keys

In flow_dissector set vlan_id in flow_keys when VLAN is found.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Get rid of IPv6 hash addresses flow keys
Tom Herbert [Thu, 4 Jun 2015 16:16:42 +0000 (09:16 -0700)]
net: Get rid of IPv6 hash addresses flow keys

We don't need to return the IPv6 address hash as part of flow keys.
In general, using the IPv6 address hash is risky in a hash value
since the underlying use of xor provides no entropy. If someone
really needs the hash value they can get it from the full IPv6
addresses in flow keys (e.g. from flow_get_u32_src).

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add keys for TIPC address
Tom Herbert [Thu, 4 Jun 2015 16:16:41 +0000 (09:16 -0700)]
net: Add keys for TIPC address

Add a new flow key for TIPC addresses.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add full IPv6 addresses to flow_keys
Tom Herbert [Thu, 4 Jun 2015 16:16:40 +0000 (09:16 -0700)]
net: Add full IPv6 addresses to flow_keys

This patch adds full IPv6 addresses into flow_keys and uses them as
input to the flow hash function. The implementation supports either
IPv4 or IPv6 addresses in a union, and selector is used to determine
how may words to input to jhash2.

We also add flow_get_u32_dst and flow_get_u32_src functions which are
used to get a u32 representation of the source and destination
addresses. For IPv6, ipv6_addr_hash is called. These functions retain
getting the legacy values of src and dst in flow_keys.

With this patch, Ethertype and IP protocol are now included in the
flow hash input.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Get skb hash over flow_keys structure
Tom Herbert [Thu, 4 Jun 2015 16:16:39 +0000 (09:16 -0700)]
net: Get skb hash over flow_keys structure

This patch changes flow hashing to use jhash2 over the flow_keys
structure instead just doing jhash_3words over src, dst, and ports.
This method will allow us take more input into the hashing function
so that we can include full IPv6 addresses, VLAN, flow labels etc.
without needing to resort to xor'ing which makes for a poor hash.

Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Remove superfluous setting of key_basic
Tom Herbert [Thu, 4 Jun 2015 16:16:38 +0000 (09:16 -0700)]
net: Remove superfluous setting of key_basic

key_basic is set twice in __skb_flow_dissect which seems unnecessary.
Remove second one.

Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agompls: Add definition for IPPROTO_MPLS
Tom Herbert [Thu, 4 Jun 2015 16:16:37 +0000 (09:16 -0700)]
mpls: Add definition for IPPROTO_MPLS

Add uapi define for MPLS over IP.

Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Simplify GRE case in flow_dissector
Tom Herbert [Thu, 4 Jun 2015 16:16:36 +0000 (09:16 -0700)]
net: Simplify GRE case in flow_dissector

Do break when we see routing flag or a non-zero version number in GRE
header.

Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoatm:he - Do not initialise statics to 0.
Shailendra Verma [Thu, 4 Jun 2015 14:41:00 +0000 (20:11 +0530)]
atm:he - Do not initialise statics to 0.

According to <stdbool.h> false is always '0' and
Static variables are initialised to 0 by GCC.

Signed-off-by: Shailendra Verma <shailendra.capricorn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: fix build due to missing tc_verd
Alexei Starovoitov [Thu, 4 Jun 2015 15:33:48 +0000 (08:33 -0700)]
bpf: fix build due to missing tc_verd

fix build error:
net/core/filter.c: In function 'bpf_clone_redirect':
net/core/filter.c:1429:18: error: 'struct sk_buff' has no member named 'tc_verd'
  if (G_TC_AT(skb2->tc_verd) & AT_INGRESS)

Fixes: 3896d655f4d4 ("bpf: introduce bpf_clone_redirect() helper")
Reported-by: Or Gerlitz <gerlitz.or@gmail.com>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: double default TSQ output bytes limit
Wei Liu [Wed, 3 Jun 2015 10:10:42 +0000 (11:10 +0100)]
tcp: double default TSQ output bytes limit

Xen virtual network driver has higher latency than a physical NIC.
Having only 128K as limit for TSQ introduced 30% regression in guest
throughput.

This patch raises the limit to 256K. This reduces the regression to 8%.
This buys us more time to work out a proper solution in the long run.

Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: remove redundant checks
Eric Dumazet [Thu, 4 Jun 2015 06:49:21 +0000 (23:49 -0700)]
tcp: remove redundant checks

tcp_v4_rcv() checks the following before calling tcp_v4_do_rcv():

if (th->doff < sizeof(struct tcphdr) / 4)
    goto bad_packet;
if (!pskb_may_pull(skb, th->doff * 4))
    goto discard_it;

So following check in tcp_v4_do_rcv() is redundant
and "goto csum_err;" is wrong anyway.

if (skb->len < tcp_hdrlen(skb) || ...)
goto csum_err;

A second check can be removed after no_tcp_socket label for same reason.

Same tests can be removed in tcp_v6_do_rcv()

Note : short tcp frames are not properly accounted in tcpInErrs MIB,
because pskb_may_pull() failure simply drops incoming skb, we might
fix this in a separate patch.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'switchdev-next'
David S. Miller [Thu, 4 Jun 2015 06:47:24 +0000 (23:47 -0700)]
Merge branch 'switchdev-next'

Scott Feldman says:

====================
switchdev: documentation updates

Update the switchdev documentation to clarify that for IPV4 FIB object, the add
op is used for both adding and modifying the FIB entry on the device.

I found a few other white-space and grammer issues to fix, and to clarify the
static FDB ops.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: documentation: use switchdev_port_obj_xxx for IPv4 FIB add/modify/delete ops
Scott Feldman [Thu, 4 Jun 2015 03:43:43 +0000 (20:43 -0700)]
switchdev: documentation: use switchdev_port_obj_xxx for IPv4 FIB add/modify/delete ops

Clarify in documentation and code that IPV4 FIB add operation is used for
both adding a new FIB entry to the device and for modifying an existing FIB
entry on the device.

Also, remove left-over references to ipv4_fib ops and replace with details
on SWITCHDEV_PORT_IPV4_FIB object.

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: documentation: for static FDB ops, use switchdev_port_fdb_xxx ops
Scott Feldman [Thu, 4 Jun 2015 03:43:42 +0000 (20:43 -0700)]
switchdev: documentation: for static FDB ops, use switchdev_port_fdb_xxx ops

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: documentation: fix grammer error
Scott Feldman [Thu, 4 Jun 2015 03:43:41 +0000 (20:43 -0700)]
switchdev: documentation: fix grammer error

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoswitchdev: documentation: fix longer-than-80-char lines
Scott Feldman [Thu, 4 Jun 2015 03:43:40 +0000 (20:43 -0700)]
switchdev: documentation: fix longer-than-80-char lines

Signed-off-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'wireless-drivers-next-for-davem-2015-06-03' of git://git.kernel.org/pub...
David S. Miller [Thu, 4 Jun 2015 06:44:57 +0000 (23:44 -0700)]
Merge tag 'wireless-drivers-next-for-davem-2015-06-03' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
new driver mt7601u for MediaTek Wi-Fi devices MT7601U

ath10k:

* qca6174 power consumption improvements, enable ASPM etc (Michal)

wil6210:

* support Wi-Fi Simple Configuration in STA mode

iwlwifi:

* a few fixes (re-enablement of interrupts for certain new
  platforms that have special power states)
* Rework completely the RBD allocation model towards new
  multi RX hardware.
* cleanups
* scan reworks continuation (Luca)

mwifiex:

* improve firmware debug functionality

rtlwifi:

* update regulatory database

brcmfmac:

* cleanup and new feature support in PCIe code
* alternative nvram loading for router support
====================

Conflicts:
drivers/net/wireless/iwlwifi/Kconfig

Trivial conflict in iwlwifi Kconfig, two commits adding
the same two chip numbers to the help text, but order
transposed.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cxgb4-next'
David S. Miller [Thu, 4 Jun 2015 06:40:20 +0000 (23:40 -0700)]
Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
cxgb4: Adds support for various ethtool stats

This patch series adds the following:
Adds support to dump adapter specific stats in ethtool
Adds support to dump channel stats in ethtool
Adds support to dump loopback port stats in ethtool
Remove wake-on-lan get/set ethtool support

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

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

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Remove WOL get/set ethtool support
Hariprasad Shenai [Wed, 3 Jun 2015 15:34:42 +0000 (21:04 +0530)]
cxgb4: Remove WOL get/set ethtool support

Remove ethtool get/set support for wake on lan, adapter doesn't support
it.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add support to dump loopback port stats
Hariprasad Shenai [Wed, 3 Jun 2015 15:34:41 +0000 (21:04 +0530)]
cxgb4: Add support to dump loopback port stats

Add support in ethtool to dump loopback port statistics

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add support in ethtool to dump channel stats
Hariprasad Shenai [Wed, 3 Jun 2015 15:34:40 +0000 (21:04 +0530)]
cxgb4: Add support in ethtool to dump channel stats

Add support in ethtool to dump adapter channel stats

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add ethtool support to get adapter stats
Hariprasad Shenai [Wed, 3 Jun 2015 15:34:39 +0000 (21:04 +0530)]
cxgb4: Add ethtool support to get adapter stats

Add ethtool support to get adapter specific hardware statistics

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge
David S. Miller [Thu, 4 Jun 2015 03:22:46 +0000 (20:22 -0700)]
Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge

Antonio Quartulli says:

====================
pull request: batman-adv 20150603

here you have our second batch of patches intended for net-next.

In this patchset you won't find any new features, but quite some code
cleanup work, a bunch of code style fixes and also comments corrections
by Markus Pargmann.

Moreover you have a patch from Sven Eckelmann removing an unnecessary
NULL check in batadv_iv_ogm_update_seqnos().

Please pull or let me know of any problem!
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: tulip: rearrange order of searching for substrings
Rasmus Villemoes [Wed, 3 Jun 2015 11:44:03 +0000 (13:44 +0200)]
net: tulip: rearrange order of searching for substrings

Currently, two of the branches are dead code, since an earlier smaller
substring would have been found ("TP" in the "TP_NW" case and either
of "BNC" and "AUI" in the "BNC_AUI" case). Rearrange the strstr()
calls so that the longer strings are searched for first.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
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 [Thu, 4 Jun 2015 03:19:16 +0000 (20:19 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

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

This series contains updates to e1000e only.

Yanir provides 8 fixes and 1 version bump for e1000e.  First fix resolves
a possible unit hang if multiple descriptors are available in the rings
during reset or runtime suspend.  Fixed a warning on SPARC compile based
on a suggested solution from Alex Duyck.  Fixed a logical error, where
a "if" condition under which a flush should occur, was revered.  Fixed
a hardware issue that prevented i219 from working in legacy interrupts
mode.  Fixed the hardware clock configuration and suprious non-linear
increment.  Lastly, fixed a system hang which occurred during execution
of "ethtool -t <NIC>", by disabling MULR for the loopback test to avoid
the hand state.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: introduce bpf_clone_redirect() helper
Alexei Starovoitov [Tue, 2 Jun 2015 23:03:14 +0000 (16:03 -0700)]
bpf: introduce bpf_clone_redirect() helper

Allow eBPF programs attached to classifier/actions to call
bpf_clone_redirect(skb, ifindex, flags) helper which will
mirror or redirect the packet by dynamic ifindex selection
from within the program to a target device either at ingress
or at egress. Can be used for various scenarios, for example,
to load balance skbs into veths, split parts of the traffic
to local taps, etc.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_core: fix typo in mlx4_set_vf_mac
Carol Soto [Tue, 2 Jun 2015 21:07:25 +0000 (16:07 -0500)]
net/mlx4_core: fix typo in mlx4_set_vf_mac

fix typo in mlx4_set_vf_mac

Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Carol L Soto <clsoto@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_core: need to call close fw if alloc icm is called twice
Carol Soto [Tue, 2 Jun 2015 21:07:24 +0000 (16:07 -0500)]
net/mlx4_core: need to call close fw if alloc icm is called twice

If mlx4_enable_sriov is called by adapter without this
feature MLX4_DEV_CAP_FLAG2_SYS_EQS then during this path the function alloc
icm is called twice without freeing the structures from the first time.

Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Carol L Soto <clsoto@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_core: double free of dev_vfs
Carol L Soto [Tue, 2 Jun 2015 21:07:23 +0000 (16:07 -0500)]
net/mlx4_core: double free of dev_vfs

If user loads mlx4_core with num_vfs greater than
supported then variable dev->dev_vfs is freed 2 times after unloading the
driver.

Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Carol L Soto <clsoto@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodrivers/net/ethernet/dec/tulip/uli526x.c: fix misleading indentation in uli526x_timer
David Malcolm [Tue, 2 Jun 2015 19:31:17 +0000 (15:31 -0400)]
drivers/net/ethernet/dec/tulip/uli526x.c: fix misleading indentation in uli526x_timer

This code in drivers/net/ethernet/dec/tulip/uli526x.c
function "uli526x_timer":

  1086          } else
  1087                  if ((tmp_cr12 & 0x3) && db->link_failed) {
  [...snip...]
  1109                  }
  1110                  else if(!(tmp_cr12 & 0x3) && db->link_failed)
  1111                  {
  [...snip...]
  1117                  }
  1118                  db->init=0;

is misleadingly indented: the
  db->init=0
is indented as if part of the else clause at line 1086, but it is
independent of it (no braces before the "if" at line 1087).

This patch fixes the indentation to reflect the actual meaning of the code,
though is it actually meant to be part of the "else" clause?  (I'm a
compiler developer, not a kernel person).  It also adds spaces around
the assignment, to placate checkpatch.pl.

Seen via an experimental new gcc warning I'm working on for gcc 6,
-Wmisleading-indentation, using gcc r223098 adding
-Werror=misleading-indentation to KBUILD_CFLAGS in Makefile.
The experimental GCC emits this warning (as an error), rightly IMHO:

drivers/net/ethernet/dec/tulip/uli526x.c: In function ‘uli526x_timer’:
drivers/net/ethernet/dec/tulip/uli526x.c:1118:3: error: statement is
indented as if it were guarded by... [-Werror=misleading-indentation]
   db->init=0;
    ^
drivers/net/ethernet/dec/tulip/uli526x.c:1086:4: note: ...this ‘else’
clause, but it is not
  } else
     ^

Hope this is helpful
Dave

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: phy: dp83867: Add TI dp83867 phy
Dan Murphy [Tue, 2 Jun 2015 14:34:37 +0000 (09:34 -0500)]
net: phy: dp83867: Add TI dp83867 phy

Add support for the TI dp83867 Gigabit ethernet phy
device.

The DP83867 is a robust, low power, fully featured
Physical Layer transceiver with integrated PMD
sublayers to support 10BASE-T, 100BASE-TX and
1000BASE-T Ethernet protocols.

Signed-off-by: Dan Murphy <dmurphy@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoethernet: micrel: use time_after_eq
Antonio Murdaca [Tue, 2 Jun 2015 12:35:52 +0000 (14:35 +0200)]
ethernet: micrel: use time_after_eq

use the time_after_eq macro for jiffies comparison operation

Signed-off-by: Antonio Murdaca <antonio.murdaca@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx4_core: Fix build failure introduced by the EQ pool changes
Or Gerlitz [Tue, 2 Jun 2015 07:29:48 +0000 (10:29 +0300)]
net/mlx4_core: Fix build failure introduced by the EQ pool changes

When CONFIG_RFS_ACCEL or SMP aren't set, we fail to build, fix it.

Also, avoid build warning as of unused function on that setup.

Fixes: c66fa19c405a ('net/mlx4: Add EQ pool')
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobatman-adv: Remove unnecessary ret variable in algo_register
Markus Pargmann [Fri, 26 Dec 2014 11:41:41 +0000 (12:41 +0100)]
batman-adv: Remove unnecessary ret variable in algo_register

Remove ret variable and all jumps.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: Remove unnecessary ret variable
Markus Pargmann [Fri, 26 Dec 2014 11:41:40 +0000 (12:41 +0100)]
batman-adv: Remove unnecessary ret variable

We can avoid this indirect return variable by directly returning the
error values.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: main, batadv_compare_eth return bool
Markus Pargmann [Fri, 26 Dec 2014 11:41:39 +0000 (12:41 +0100)]
batman-adv: main, batadv_compare_eth return bool

Declare the returntype of batadv_compare_eth as bool.
The function called inside this helper function
(ether_addr_equal_unaligned) also uses bool as return value, so there is
no need to return int.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: main, Convert is_my_mac() to bool
Markus Pargmann [Fri, 26 Dec 2014 11:41:38 +0000 (12:41 +0100)]
batman-adv: main, Convert is_my_mac() to bool

It is much clearer to see a bool type as return value than 'int' for
functions that are supposed to return true or false.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: Remove unnecessary check for orig_ifinfo not NULL
Sven Eckelmann [Sun, 1 Mar 2015 15:56:26 +0000 (16:56 +0100)]
batman-adv: Remove unnecessary check for orig_ifinfo not NULL

orig_ifinfo is dereferenced multiple times in batadv_iv_ogm_update_seqnos
before the check for NULL is done. The function also exists at the
beginning when orig_ifinfo would have been NULL. This makes the check at
the end unnecessary and only confuses the reader/code analyzers.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: types, Fix comment on bcast_own
Markus Pargmann [Fri, 26 Dec 2014 11:41:37 +0000 (12:41 +0100)]
batman-adv: types, Fix comment on bcast_own

batadv_orig_bat_iv->bcast_own is actually not a bitfield, it is an
array. Adjust the comment accordingly.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Antonio Quartulli <antonio@meshcoding.com>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agoe1000e: Bump the version to 3.2.5
Yanir Lubetkin [Tue, 2 Jun 2015 14:05:50 +0000 (17:05 +0300)]
e1000e: Bump the version to 3.2.5

Bump the version to reflect the driver changes and bug fixes for i219.
Also update the copyright, while we are at it.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@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 unit hang during loopback test
Yanir Lubetkin [Tue, 2 Jun 2015 14:05:47 +0000 (17:05 +0300)]
e1000e: fix unit hang during loopback test

System would hang during execution of "ethtool -t <NIC>" for the same
reason that required flushing the descriptor rings. This fix disables
MULR for the loopback test to avoid the hang state.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@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 systim issues
Yanir Lubetkin [Tue, 2 Jun 2015 14:05:45 +0000 (17:05 +0300)]
e1000e: fix systim issues

Two issues involving systim were reported.
1. Clock is not running in the correct frequency
2. In some situations, systim values were not incremented linearly
This patch fixes the hardware clock configuration and the spurious
non-linear increment.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@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 legacy interrupt handling in i219
Yanir Lubetkin [Tue, 2 Jun 2015 14:05:42 +0000 (17:05 +0300)]
e1000e: fix legacy interrupt handling in i219

This fix handles a hardware issue that prevented i219 from
working in legacy interrupts mode (IntMode=0)

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@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 flush_desc_ring implementation
Yanir Lubetkin [Tue, 2 Jun 2015 14:05:38 +0000 (17:05 +0300)]
e1000e: fix flush_desc_ring implementation

The indication that a descriptor ring flush is required was read from
FEXTNVM7 by mistake. It should be read from the PCI config space.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@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 logical error in flush_desc_rings
Yanir Lubetkin [Wed, 22 Apr 2015 16:25:17 +0000 (19:25 +0300)]
e1000e: fix logical error in flush_desc_rings

The condition under which the flush should occur was reversed.  The fix
should be applied before any HW reset (unless followed by bus reset)
and before any power state transition from D0.

If E1000_FEXTNVM7_NEED_DESCRING_FLUSH bit is set in FEXTNVM7 and TDLEN > 0
the Tx ring should be flushed. (fixes ~95% of the hang states).
If the E1000_FEXTNVM7_NEED_DESCRING_FLUSH did not clear, we should also
flush the RX ring. Bug was caught by Alexander Duyck during a code review
when examining this fix.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: remove call to do_div and sign mismatch warning
Yanir Lubetkin [Wed, 22 Apr 2015 02:55:43 +0000 (05:55 +0300)]
e1000e: remove call to do_div and sign mismatch warning

Fixes a warning that was reported by Yanjiang Jin
<yanjiang.jin@windriver.com> by implementing the solution suggested by
Alexander Duyck <alexander.h.duyck@redhat.com>.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: i219 execute unit hang fix on every reset or power state transition
Yanir Lubetkin [Wed, 22 Apr 2015 01:15:01 +0000 (04:15 +0300)]
e1000e: i219 execute unit hang fix on every reset or power state transition

After testing various cases, the conclusion is that the fix MUST be
executed BEFORE any event that the HW is reset or transition to D3.
To fix that I moved the execution to the relevant places but per
Alexander Duyck's review, ensure now that the DMA is valid and was not
freed before manipulating the ring.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoe1000e: i219 fix unit hang on reset and runtime D3
Yanir Lubetkin [Mon, 13 Apr 2015 23:20:21 +0000 (02:20 +0300)]
e1000e: i219 fix unit hang on reset and runtime D3

Unit hang may occur if multiple descriptors are available in the rings
during reset or runtime suspend. This state can be detected by testing
bit 8 in the FEXTNVM7 register. If this bit is set and there are pending
descriptors in one of the rings, we must flush them prior to reset. Same
applies entering runtime suspend.

Signed-off-by: Yanir Lubetkin <yanirx.lubetkin@intel.com>
Reviewed-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoMerge tag 'iwlwifi-next-for-kalle-2015-06-03' of https://git.kernel.org/pub/scm/linux...
Kalle Valo [Wed, 3 Jun 2015 09:15:51 +0000 (12:15 +0300)]
Merge tag 'iwlwifi-next-for-kalle-2015-06-03' of https://git./linux/kernel/git/iwlwifi/iwlwifi-next

* a few fixes (re-enablement of interrupts for certain new
  platforms that have special power states)
* Rework completely the RBD allocation model towards new
  multi RX hardware.
* cleanups
* scan reworks continuation (Luca)

9 years agobatman-adv: iv_ogm, fix comment function name
Markus Pargmann [Fri, 26 Dec 2014 11:41:36 +0000 (12:41 +0100)]
batman-adv: iv_ogm, fix comment function name

This is a small copy paste fix for batadv_ing_buffer_avg.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: iv_ogm, fix coding style
Markus Pargmann [Fri, 26 Dec 2014 11:41:35 +0000 (12:41 +0100)]
batman-adv: iv_ogm, fix coding style

The kernel coding style says, that there should not be multiple
assignments in one row.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: iv_ogm, Fix dup_status comment
Markus Pargmann [Fri, 26 Dec 2014 11:41:34 +0000 (12:41 +0100)]
batman-adv: iv_ogm, Fix dup_status comment

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: iv_ogm_orig_update, style, add missing brackets
Markus Pargmann [Fri, 26 Dec 2014 11:41:33 +0000 (12:41 +0100)]
batman-adv: iv_ogm_orig_update, style, add missing brackets

CodingStyle describes that either none or both branches of a conditional
have to have brackets.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: iv_ogm_queue_add, Simplify expressions
Markus Pargmann [Fri, 26 Dec 2014 11:41:32 +0000 (12:41 +0100)]
batman-adv: iv_ogm_queue_add, Simplify expressions

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agobatman-adv: iv_ogm_aggregate_new, simplify error handling
Markus Pargmann [Fri, 26 Dec 2014 11:41:31 +0000 (12:41 +0100)]
batman-adv: iv_ogm_aggregate_new, simplify error handling

It is just a bit easier to put the error handling at one place and let
multiple error paths use the same calls.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
9 years agoiwlwifi: mvm: advertise only HW-supported ciphers
Johannes Berg [Wed, 20 May 2015 14:51:28 +0000 (16:51 +0200)]
iwlwifi: mvm: advertise only HW-supported ciphers

After the new ciphers CCMP-256 and GCMP-128/256 were implemented,
wpa_supplicant could start negotiating them and use the software
implementation. This, however, breaks D3 behaviour in the driver
since it means that WoWLAN will not be possible.

To avoid breaking that feature, advertise only ciphers that the
hardware supports.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: bump the iwlmvm API number to 15
Emmanuel Grumbach [Wed, 3 Jun 2015 06:54:18 +0000 (09:54 +0300)]
iwlwifi: bump the iwlmvm API number to 15

The driver is now ready to handle the -15.ucode.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: mvm: Add DC2DC_CONFIG_CMD (0x83) cmd & TLV
Matti Gottlieb [Sun, 31 May 2015 06:18:30 +0000 (09:18 +0300)]
iwlwifi: mvm: Add DC2DC_CONFIG_CMD (0x83) cmd & TLV

Add DC2DC_CONFIG_CMD (0x83) cmd.

Add IWL_UCODE_TLV_CAPA_DC2DC_CONFIG_SUPPORT tlv.

The command allows the driver get & set the DCDC's frequency tune.
(freq_tune is the divider that is used to calculate the actual DCDC's
clock rate)
The command always returns the current/updated frequency tune values of
the DCDC.

Signed-off-by: Matti Gottlieb <matti.gottlieb@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: mvm: Remove old scan commands
Matti Gottlieb [Sun, 31 May 2015 08:29:52 +0000 (11:29 +0300)]
iwlwifi: mvm: Remove old scan commands

The firmwares that used these commands is not supported
anymore. Remove them.

Signed-off-by: Matti Gottlieb <matti.gottlieb@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: prepare for higher API/CAPA bits
Johannes Berg [Mon, 1 Jun 2015 15:11:11 +0000 (17:11 +0200)]
iwlwifi: prepare for higher API/CAPA bits

Currently, loading the firmware fails when it has higher API or CAPA
bits than the driver supports. That's an issue with integration.

At the same time, actually using api[0] and capa[0] will become
confusing when we also have api[1] and capa[1], and it's almost
certain that we'll mix up the bits and use the bits for api[1] with
api[0] by accident.

Avoid all this by translating the API/CAPA bits to the regular kernel
test_bit() format, and also providing wrapper functions. Also use the
__bitwise__ facility of sparse to check that we're testing the right
one.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: mvm: simplify iwl_mvm_stop_roc()
Johannes Berg [Mon, 1 Jun 2015 21:38:23 +0000 (23:38 +0200)]
iwlwifi: mvm: simplify iwl_mvm_stop_roc()

As pointed out by smatch, there's no need for a loop that always
immediately terminates. Use an if statement instead and while at
it clean up the mvmvif initialization.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: dvm: enable IEEE80211_HW_SUPPORT_FAST_XMIT
Johannes Berg [Fri, 17 Apr 2015 13:50:43 +0000 (15:50 +0200)]
iwlwifi: dvm: enable IEEE80211_HW_SUPPORT_FAST_XMIT

Since the firmware is responsible for duration calculation, the
driver can easily support fast-xmit.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: mvm: enable IEEE80211_HW_SUPPORT_FAST_XMIT
Johannes Berg [Fri, 17 Apr 2015 13:50:43 +0000 (15:50 +0200)]
iwlwifi: mvm: enable IEEE80211_HW_SUPPORT_FAST_XMIT

Since the firmware is responsible for duration calculation, the
driver can easily support fast-xmit.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: pcie: New RBD allocation model
Sara Sharon [Tue, 28 Apr 2015 09:56:54 +0000 (12:56 +0300)]
iwlwifi: pcie: New RBD allocation model

As a preperation for multiple RX queues change the RBD
allocation model.

The new model includes a background allocator. The allocator is
called by the interrupt handler when there are two released
buffers by the queue, and the allocator starts allocating eight
pages per request.
When the queue has released 8 pages it tries claiming the
request. If the pages are not ready - it keeps claiming.
This new model should make sure that RBDs are always available
across the multiple queues.

The RBDs are transferred between the allocator and the queue.
The queue moves the free RBDs upon freeing them to the allocator.
The allocator moves them back to the queue's possession when the
request is claimed.
The allocator has an initial pool to make sure there are always RBDs
available for the request completion.
Release of the buffers at exit is done per pools - the allocator
frees its own initial pool and the queue frees its own pool.

Existing code refactor -
-Queue's initial pool is the size of the queue only as the allocation
of the new buffers no longer uses this pool.
-Removal of replenish background work, and replenish calls in the
interrupt handler and restock().
-The replenish() and the rxq used_list are used only during
initialization.
-Moved page allocation to a new function for code reuse.

New code -
Allocator code - new structure and functions.
Interrupt handler uses the allocator functions for replenishing buffers.
Reuse of the restock() method.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: pcie: re-enable interrupts on resume
Eliad Peller [Thu, 28 May 2015 13:50:12 +0000 (16:50 +0300)]
iwlwifi: pcie: re-enable interrupts on resume

On resume, all the interrupts are masked (CSR_INT_MASK is 0),
and ict is disabled.

Re-configure them both.

Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
9 years agoiwlwifi: mvm: don't use EBS for P2P find
David Spinadel [Tue, 26 May 2015 07:32:19 +0000 (10:32 +0300)]
iwlwifi: mvm: don't use EBS for P2P find

Don't use EBS for P2P find to make sure we find all GOs in
our only attempt.

Signed-off-by: David Spinadel <david.spinadel@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>