softirq: reduce latencies

In various network workloads, __do_softirq() latencies can be up
to 20 ms if HZ=1000, and 200 ms if HZ=100.

This is because we iterate 10 times in the softirq dispatcher,
and some actions can consume a lot of cycles.

This patch changes the fallback to ksoftirqd condition to :

- A time limit of 2 ms.
- need_resched() being set on current task

When one of this condition is met, we wakeup ksoftirqd for further
softirq processing if we still have pending softirqs.

Using need_resched() as the only condition can trigger RCU stalls,
as we can keep BH disabled for too long.

I ran several benchmarks and got no significant difference in
throughput, but a very significant reduction of latencies (one order
of magnitude) :

In following bench, 200 antagonist "netperf -t TCP_RR" are started in
background, using all available cpus.

Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC
IRQ (hard+soft)

Before patch :

# netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k
RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
RT_LATENCY=550110.424
MIN_LATENCY=146858
MAX_LATENCY=997109
P50_LATENCY=305000
P90_LATENCY=550000
P99_LATENCY=710000
MEAN_LATENCY=376989.12
STDDEV_LATENCY=184046.92

After patch :

# netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k
RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET
to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind
RT_LATENCY=40545.492
MIN_LATENCY=9834
MAX_LATENCY=78366
P50_LATENCY=33583
P90_LATENCY=59000
P99_LATENCY=69000
MEAN_LATENCY=38364.67
STDDEV_LATENCY=12865.26

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Miller <davem@davemloft.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: more precise pkt_len computation

One long standing problem with TSO/GSO/GRO packets is that skb->len
doesn't represent a precise amount of bytes on wire.

Headers are only accounted for the first segment.
For TCP, thats typically 66 bytes per 1448 bytes segment missing,
an error of 4.5 % for normal MSS value.

As consequences :

1) TBF/CBQ/HTB/NETEM/... can send more bytes than the assigned limits.
2) Device stats are slightly under estimated as well.

Fix this by taking account of headers in qdisc_skb_cb(skb)->pkt_len
computation.

Packet schedulers should use qdisc pkt_len instead of skb->len for their
bandwidth limitations, and TSO enabled devices drivers could use pkt_len
if their statistics are not hardware assisted, and if they don't scratch
skb->cb[] first word.

Both egress and ingress paths work, thanks to commit fda55eca5a
(net: introduce skb_transport_header_was_set()) : If GRO built
a GSO packet, it also set the transport header for us.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: Paolo Valente <paolo.valente@unimore.it>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

doc: Clarify behavior when sysctl tcp_ecn = 1

Recent commit (commit 7e3a2dc52953 doc: make the description of how tcp_ecn
works more explicit and clear ) clarified the behavior of tcp_ecn sysctl
variable but description is inconsistent. When requested by incoming conections,
ECN is enabled with not just tcp_ecn = 2 but also with tcp_ecn = 1.

This patch makes it clear that with tcp_ecn = 1, ECN is enabled when requested
by incoming connections.

Also fix spelling of 'incoming'.

Signed-off-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: align define usage to satisfy static checkers

Static checkers complained that the E1H_FUNC_MAX define is used
incorrectly in bnx2x_pretend_func(). The complaint was justified,
although its not a real bug, as the first part of the conditional
protects us in this case (a real bug would happen if a VF tried to
use the pretend func, but there are no VFs in E1H chips).

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

veth: fix a NULL deref in netif_carrier_off

In commit d0e2c55e7c94 (veth: avoid a NULL deref in veth_stats_one)
we now clear the peer pointers in veth_dellink()

veth_close() must therefore make sure the peer pointer is set.

Reported-by: Tom Parkin <tom.parkin@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: Fix incorrect PFVF CMASK

With Hard-Wired firmware configuration it was incorrectly provisioning the VFs
Channel Access Rights Mask.

Signed-off-by: Jay Hernandez <jay@chelsio.com>
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DSA: Convert spaces to tabs where appropriate

Fix DSA whitespace issues reported by checkpatch.pl

Signed-off-by: Barry Grussling <barry@grussling.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DSA: Convert printk calls to netdev_info calls

Convert DSA printk calls to netdev_info calls as recommended by
checkpatch.pl.

Signed-off-by: Barry Grussling <barry@grussling.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DSA: Convert repeated msleep calls to timeouts

Convert DSA msleep calls to timeout/usleep_range calls
as reported by checkpatch.pl.

Signed-off-by: Barry Grussling <barry@grussling.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DSA: Convert DSA comments to network-style comments

Convert DSA driver comments to network-style comments as reported by
checkpatch.pl.  Fix spelling error.

Signed-off-by: Barry Grussling <barry@grussling.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Optimize ipv6_change_dsfield().

Do not convert endian back and forth.
If the caller uses contant "mask" argument (and most callers do),
we can omit runtime endian conversion here.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Use container_of macro instead of magic number to get ipv6 header.

In ipv6_recv_error(), addr_offset points to daddr field of the ip header.
To get ipv6 header, use container_of() macro instead of substracting magic
number (24).

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: NULL dereference on error in debug code

"vfop" is NULL here.  I've changed the debugging to not use it.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Ariel Elior <ariele@broadcom.com>
Acked-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

unix: Use FIELD_SIZEOF() in af_unix_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

rxrpc: Use FIELD_SIZEOF() in af_rxrpc_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

openvswitch: Use FIELD_SIZEOF() in dp_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

netlink: Use FIELD_SIZEOF() in netlink_proto_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Use FIELD_SIZEOF() in inet6_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Use FIELD_SIZEOF() in inet_init().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

bgmac: driver for GBit MAC core on BCMA bus

BCMA is a Broadcom specific bus with devices AKA cores. All recent BCMA
based SoCs have gigabit ethernet provided by the GBit MAC core. This
patch adds driver for such a cores registering itself as a netdev. It
has been tested on a BCM4706 and BCM4718 chipsets.

In the kernel tree there is already b44 driver which has some common
things with bgmac, however there are many differences that has led to
the decision or writing a new driver:
1) GBit MAC cores appear on BCMA bus (not SSB as in case of b44)
2) There is 64bit DMA engine which differs from 32bit one
3) There is no CAM (Content Addressable Memory) in GBit MAC
4) We have 4 TX queues on GBit MAC devices (instead of 1)
5) Many registers have different addresses/values
6) RX header flags are also different

The driver in it's state is functional how, however there is of course
place for improvements:
1) Supporting more net_device_ops
2) SUpporting more ethtool_ops
3) Unaligned addressing in DMA
4) Writing separated PHY driver

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

remove init of dev->perm_addr in drivers

perm_addr is initialized correctly in register_netdevice() so to init it in
drivers is no longer needed.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: init perm_addr in register_netdevice()

Benefit from the fact that dev->addr_assign_type is set to NET_ADDR_PERM
in case the device has permanent address.

This also fixes the problem that many drivers do not set perm_addr at
all.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

netconsole: add IPv6 example in doc

Update the netconsole document as well.

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netpoll: add IPv6 support

Currently, netpoll only supports IPv4. This patch adds IPv6
support to netpoll so that we can run netconsole over IPv6 network.

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: move csum_ipv6_magic() and udp6_csum_init() into static library

As suggested by David, udp6_csum_init() is too big to be inlined,
move it to ipv6 static library, net/ipv6/ip6_checksum.c.

And the generic csum_ipv6_magic() too.

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netpoll: prepare for ipv6

This patch adjusts some struct and functions, to prepare
for supporting IPv6.

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: fec: Remove 'inline' declarations

Fix the following warning when building with W=1 option:

drivers/net/ethernet/freescale/fec.c:810:1: warning: '__inline__' is not at beginning of declaration [-Wold-style-declaration]

The inline declaration is pointless in this function, so just remove it.

While at it, also remove the other 'inline' declarations.

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: introduce skb_transport_header_was_set()

We have skb_mac_header_was_set() helper to tell if mac_header
was set on a skb. We would like the same for transport_header.

__netif_receive_skb() doesn't reset the transport header if already
set by GRO layer.

Note that network stacks usually reset the transport header anyway,
after pulling the network header, so this change only allows
a followup patch to have more precise qdisc pkt_len computation
for GSO packets at ingress side.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'delete-8390-EISA' of git://git./linux/kernel/git/paulg/linux

Paul Gortmaker says:

====================
I'd like to propose that we get rid of these old 8390 EISA drivers.
Of the five deleted here, I wrote four -- and while that doesn't give
me any authority for deletion above anyone else, it does at least
allow me to comment on the absolute absence of anyone reaching
out to the driver author for assistance in the last dozen years.

Eventually we'll probably get rid of EISA bus support, since in
x86, the hardware is close to 20 years old and already too resource
constrained to be useful today.  However there might still be
a few DEC Alpha enthusiasts with old EISA machines kept alive,
and so I expect we'll have to wait a bit longer to get unanimous
agreement to proceed with the full EISA removal (although I'd
love to be proven wrong on that).

Most of the DEC Alpha machines shipped in a PCI configuration, and
even the few that were EISA had DEC tulip based ethernet and no
reason to be needing the inferior 8390 technology.  So the interest
here for any possible DEC enthusiasts with EISA boxes about these
old 8390 drivers should be nil.

These really were rare cards -- in fact the smc-ultra32 is the only
one that I'd ever seen in person.  Even back in the mid 90's when
the drivers were written, I would guess that the user base was less
than 10 people across all of them.

The following patch was created with --irreversible-delete for
ease of review (it skips showing the content of files that are
deleted); however the complete patch can be pulled as per below.
====================

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

ethtool: consolidate work with ethtool_ops

No need to check if ethtool_ops == NULL since it can't be.
Use local variable "ops" in functions where it is present
instead of dev->ethtool_ops
Introduce local variable "ops" in functions where dev->ethtool_ops is used
many times.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Reviewed-by: Ben Hutchings <bhutchings@solarflare.com>
Reviewed-by: Flavio Leitner <fbl@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Segregate SR-IOV code

In this patch the SR-IOV code is segregated from the main bulk of
the bnx2x code. The CONFIG_BNX2X_SRIOV define is added to Broadcom's
Kconfig, and allows the elision of the building of all the SR-IOV
support code in the driver.
The define is dependant on the kernel CONFIG_PCI_IOV configuration
define.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: fec: report correct hardware stamping info to ethtool

Report correct hardware stamping capability by ethtool interface.
The v1.0 ptp4l check it.

Signed-off-by: Frank Li <Frank.Li@freescale.com>
Acked-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

veth: avoid a NULL deref in veth_stats_one

commit 2681128f0ced8a (veth: extend device features) added a NULL deref
in veth_stats_one(), as veth_get_stats64() was not testing if the peer
device was setup or not.

At init time, we call dev_get_stats() before veth pair is fully setup.

[  178.854758]  [<ffffffffa00f5677>] veth_get_stats64+0x47/0x70 [veth]
[  178.861013]  [<ffffffff814f0a2d>] dev_get_stats+0x6d/0x130
[  178.866486]  [<ffffffff81504efc>] rtnl_fill_ifinfo+0x47c/0x930
[  178.872299]  [<ffffffff81505b93>] rtmsg_ifinfo+0x83/0x100
[  178.877678]  [<ffffffff81505cc6>] rtnl_configure_link+0x76/0xa0
[  178.883580]  [<ffffffffa00f52fa>] veth_newlink+0x16a/0x350 [veth]
[  178.889654]  [<ffffffff815061cc>] rtnl_newlink+0x4dc/0x5e0
[  178.895128]  [<ffffffff81505e1e>] ? rtnl_newlink+0x12e/0x5e0
[  178.900769]  [<ffffffff8150587d>] rtnetlink_rcv_msg+0x11d/0x310
[  178.906669]  [<ffffffff81505760>] ? __rtnl_unlock+0x20/0x20
[  178.912225]  [<ffffffff81521f89>] netlink_rcv_skb+0xa9/0xd0
[  178.917779]  [<ffffffff81502d55>] rtnetlink_rcv+0x25/0x40
[  178.923159]  [<ffffffff815218d1>] netlink_unicast+0x1b1/0x230
[  178.928887]  [<ffffffff81521c4e>] netlink_sendmsg+0x2fe/0x3b0
[  178.934615]  [<ffffffff814dbe22>] sock_sendmsg+0xd2/0xf0

So we must check if peer was setup in veth_get_stats64()

As pointed out by Ben Hutchings, priv->peer is missing proper
synchronization. Adding RCU protection is a safe and well documented
way to make sure we don't access about to be freed or already
freed data.

Reported-by: Tom Parkin <tparkin@katalix.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

chelsio: Use netdev_<level> and pr_<level>

Use more current logging styles.

Convert printks to pr_<level> and
printks with ("%s: ...", dev->name to netdev_<level>(dev, "...

Add pr_fmt #defines where appropriate.
Coalesce formats.
Use pr_<level>_once where appropriate.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net: delete 8390 based EISA drivers.

The NS8390 chip was essentially the 1st widespread PC ethernet
chip, starting its life on 8 bit ISA cards in the late 1980s.
Even with better technologies available (bus mastering etc)
the 8390 managed to get used on a few rare EISA cards in the
early to mid 1990s.

The EISA bus in the x86 world was largely confined to systems
ranging from 486 to 586 (essentially 200MHz or lower, and less
than 100MB RAM) -- i.e. machines unlikely to be still in service,
and even less likely to be running a 3.9+ kernel.

On top of that, only one of the five really ever was considered
non-experimental; the smc-ultra32 was the one -- since it was
largely just an EISA version of the popular smc-ultra ISA card.
All the others had such a tiny user base that they simply never
could be considered anything more than experimental.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

drivers/net: remove orphaned references to micro channel

We threw away the microchannel support, but the removal wasn't
completely trivial since there was namespace overlap with the
machine check support, and hence some orphaned dependencies
survived the deletion.  This attempts to sweep those up and
send them to the bit-bucket.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

team: use strlcpy with ethtool_drvinfo fields

The fields must be null-terminated.

Signed-off-by: Flavio Leitner <fbl@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: make sysctl_tcp_ecn namespace aware

As per suggestion from Eric Dumazet this patch makes tcp_ecn sysctl
namespace aware.  The reason behind this patch is to ease the testing
of ecn problems on the internet and allows applications to tune their
own use of ecn.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ndisc: Use struct rd_msg for redirect message.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: splice: avoid high order page splitting

splice() can handle pages of any order, but network code tries hard to
split them in PAGE_SIZE units. Not quite successfully anyway, as
__splice_segment() assumed poff < PAGE_SIZE. This is true for
the skb->data part, not necessarily for the fragments.

This patch removes this logic to give the pages as they are in the skb.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willy Tarreau <w@1wt.eu>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: use ETHTOOL_FWVERS_LEN instead of ETHTOOL_BUSINFO_LEN for fw_ver strings

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

ethtool: fix drvinfo strings set in drivers

Use strlcpy where possible to ensure the string is \0 terminated.
Use always sizeof(string) instead of 32, ETHTOOL_BUSINFO_LEN
and custom defines.
Use snprintf instead of sprint.
Remove unnecessary inits of ->fw_version
Remove unnecessary inits of drvinfo struct.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

ethtool: set addr_assign_type to NET_ADDR_SET when addr is passed on create

In case user passed address via netlink during create, NET_ADDR_PERM was set.
That is not correct so fix this by setting NET_ADDR_SET.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

ethoc: fix mac address set

Function ethoc_set_mac_address() was incorrectly using passed pointer as
pointer to address, that is not correct.
Struct sockaddr have to be be used here.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Remove IS_ENABLED(CONFIG_HWMON) check

Commit de0a41484c47d783dd4d442914815076aa2caac2 added Kconfig logic to
select HWMON and removed all the IS_ENABLED(CONFIG_HWMON) checks in the
tg3.c file. It missed this one check in the header.

Update version to 3.129 and update copyright year.

Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Improve PCI function number detection.

Simplify the code to detect PCI function number on 5717, 5719, and 5720.
If shared memory does not have proper signature, read the function number
from register directly.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Add NVRAM support for 5762

Detect NVRAM types for 5762 and read OTP firmware version.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Add support for new 5762 ASIC

Add basic support for 5762 which is a 57765_PLUS class device.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ndisc: Remove unused space at tail of skb for ndisc messages. (TAKE 3)

Currently, the size of skb allocated for NDISC is MAX_HEADER +
LL_RESERVED_SPACE(dev) + packet length + dev->needed_tailroom,
but only LL_RESERVED_SPACE(dev) bytes is "reserved" for headers.
As a result, the skb looks like this (after construction of the
message):

head       data                   tail                       end
+--------------------------------------------------------------+
+           |                      |          |                |
+--------------------------------------------------------------+
|<-hlen---->|<---ipv6 packet------>|<--tlen-->|<--MAX_HEADER-->|
    =LL_                               = dev
     RESERVED_                           ->needed_
     SPACE(dev)                            tailroom

As the name implies, "MAX_HEADER" is used for headers, and should
be "reserved" in prior to packet construction.  Or, if some space
is really required at the tail of ther skb, it should be
explicitly documented.

We have several option after construction of NDISC message:

Option 1:

head       data                   tail       end
+---------------------------------------------+
+           |                      |          |
+---------------------------------------------+
|<-hlen---->|<---ipv6 packet------>|<--tlen-->|
   =LL_                                = dev
    RESERVED_                           ->needed_
    SPACE(dev)                            tailroom

Option 2:

head            data                   tail       end
+--------------------------------------------------+
+                |                      |          |
+--------------------------------------------------+
|<--MAX_HEADER-->|<---ipv6 packet------>|<--tlen-->|
                                            = dev
                                             ->needed_
                                               tailroom

Option 3:

head                        data                   tail       end
+--------------------------------------------------------------+
+                |           |                      |          |
+--------------------------------------------------------------+
|<--MAX_HEADER-->|<-hlen---->|<---ipv6 packet------>|<--tlen-->|
                    =LL_                                = dev
                     RESERVED_                          ->needed_
                     SPACE(dev)                           tailroom

Our tunnel drivers try expanding headroom and the space for tunnel
encapsulation was not a mandatory space -- so we are not seeing
bugs here --, but just for optimization for performance critial
situations.

Since NDISC messages are not performance critical unlike TCP,
and as we know outgoing device, LL_RESERVED_SPACE(dev) should be
just enough for the device in most (if not all) cases:
  LL_RESERVED_SPACE(dev) <= LL_MAX_HEADER <= MAX_HEADER
Note that LL_RESERVED_SPACE(dev) is also enough for NDISC over
SIT (e.g., ISATAP).

So, I think Option 1 is just fine here.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

NET: FEC: dynamtic check DMA desc buff type

MX6 and mx28 support enhanced DMA descriptor buff to support 1588
ptp. But MX25, MX3x, MX5x can't support enhanced DMA descriptor buff.
Check fec type and choose correct DMA descriptor buff type.

Remove static config CONFIG_FEC_PTP.
ptp function will be auto detected.

Signed-off-by: Frank Li <Frank.Li@freescale.com>
Acked-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

ethernet/broadcom/tg3: Fix sparse warning: constant 0x7fffffffffffffff is so big it is long long

Sparse complains that:
drivers/net/ethernet/broadcom/tg3.c:5670:55: sparse: constant
0x7fffffffffffffff is so big it is long long (on x86/32 bit)

so we suffix the constant with LL in the header file.

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

smsc95xx: enable dynamic autosuspend

This patch enables USB dynamic autosuspend for LAN9500A.  This
saves very little power in itself, but it allows power saving
in upstream hubs/hosts.

The earlier devices in this family (LAN9500/9512/9514) do not
support this feature.

Signed-off-by: Steve Glendinning <steve.glendinning@shawell.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

smsc95xx: eliminate duplicate warnings on io failure

The register read/write functions already log a warning if
an access fails, so this patch removes the additional warnings
logged by callers that don't add any more information.

This patch makes the resulting driver smaller by not containing
as many warning strings.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Steve Glendinning <steve.glendinning@shawell.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: introduce upper device lists and remove dev->master

Jiri Pirko says:

====================
This is a V6 of a repost of my previous patchset:
"[patch net-next v2 00/15] net: introduce upper device lists and remove dev->master" from Aug 14

The discussion around
"[net-next] bonding: don't allow the master to become its slave"
forced me to think about upper<->lower device connections.

This patchset adds a possibility to record upper device linkage.
All upper<->lower devices are converted to use this mechanism right after.
That leads to dev->master removal because this info becomes redundant since
"master links" have the same value.

After all changes, there is no longer possible to do things as:
"bond->someotherdevice->samebond"

Also I think that drivers like cxgb3, qlcnic, qeth would benefit by this
in future by being able to get more appropriate info about l3 addresses.

v5->v6:
- netdev_has_upper_dev() - added statement to comment that this is looking at
  the immediate upper devices only.
- renamed "RTNL semaphore" -> "RTNL lock" in all comments
- renamed __netdev_has_upper_dev() to __netdev_search_upper_dev() to emhasize
  the difference to netdev_has_upper_dev()

v4->v5:
- fixed missed typo in drivers/infiniband/hw/nes/nes_cm.c

v3->v4:
- comments in __netdev_upper_dev_link() squashed into one line
- kfree_rcu used instead of call_rcu in netdev_upper_dev_unlink()

v2->v3:
- removed recursion in __netdev_has_upper_dev()
- refreshed bits to be applicable on current net-next

v1->v2:
- s/unique/master/ better naming + stays closer to the past
- fixed vlan err goto
- original patch 15 (WARN_ON change) is squashed into the first patch
====================

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

net: kill dev->master

Nobody uses this now. Remove it.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: remove no longer used netdev_set_bond_master() and netdev_set_master()

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: remove usage of dev->master

Benefit from new upper dev list and free bonding from dev->master usage.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

nes: remove usage of dev->master

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

vlan: remove usage of dev->master in __vlan_find_dev_deep()

Also, since all users call __vlan_find_dev_deep() with rcu_read_lock,
make no possibility to call this with rtnl mutex held only.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

qeth: ensure that __vlan_find_dev_deep() is called with rcu_read_lock

Also benefit from rcu_read_lock held and use __in_dev_get_rcu() in ipv4 case.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: guard __vlan_find_dev_deep() by rcu_read_lock

rcu_read_lock was missing here

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb3: remove usage of dev->master

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

netpoll: remove usage of dev->master

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: remove usage of netdev_set_master()

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

team: remove usage of netdev_set_master()

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

rtnetlink: remove usage of dev->master

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

vlan: add link to upper device

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

macvlan: add link to upper device

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: introduce upper device lists

This lists are supposed to serve for storing pointers to all upper devices.
Eventually it will replace dev->master pointer which is used for
bonding, bridge, team but it cannot be used for vlan, macvlan where
there might be multiple upper present. In case the upper link is
replacement for dev->master, it is marked with "master" flag.

New upper device list resolves this limitation. Also, the information
stored in lists is used for preventing looping setups like
"bond->somethingelse->samebond"

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

ll_temac: fix mac address setting

Previously, when invalid address was passed to ndo_set_mac_address,
random mac was generated and set. Fix this by returning -EADDRNOTAVAIL
in this situation.

Also polish the code around a bit.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

um: net: use eth_hw_addr_random() to generate random mac

Also remove unused "mac" from uml_net struct.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: remove unnecessary NET_ADDR_RANDOM "bitclean"

NET_ADDR_SET is set in dev_set_mac_address() no need to alter
dev->addr_assign_type value in drivers.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: add address assign type "SET"

This is the way to indicate that mac address of a device has been set by
dev_set_mac_address()

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: set dev->addr_assign_type correctly

Not a bitfield, but a plain value.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: call add_device_randomness() only after successful mac change

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

rtnl: use dev_set_mac_address() instead of plain ndo_

Benefit from existence of dev_set_mac_address() and remove duplicate
code.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

team: update master carrier state

Update master's carrier state when there is any
change with its ports.

Signed-off-by: Flavio Leitner <fbl@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: fix build error

Fix the following error reported by kbuild test robot.
static declaration of 'qlcnic_restore_indev_addr' follows
non-static declaration.

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: do not duplicate infrastructure functions

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: update driver version

Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: remove unused definitions from header file

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx ethtool interface routines

83xx ethtool interface routines

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx register dump routines

Add 83xx register dump routines
Update 82xx register dump routines

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: enable 83xx virtual NIC mode

Enable 83xx virtual NIC mode

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: flash template based firmware reset recovery

Flash template provides instructions to stop, restart and initalize the
firmware. These instructions are abstracted as a series of read, write and
poll operations on hardware registers. Register information and operation
specifics are not exposed to the driver. Driver reads the template from
flash and executes the instructions located at pre-defined offsets.

Template based firmware reset recovery and initialization mechanism minimize
driver changes as firmware evolves.

Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx CNA inter driver communication mechanism

Inter Driver Communication (IDC) module.
CNA function drivers(ISCSI, FCOE and NIC) which shares the adapter
relies on IDC mechanism for gracefull shut down, restart and
firmware error recovery.

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx adpater flash interface routines

83xx adapter flash memory map, data structures and interface routines

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx sysfs routines

Add 83xx sysfs interface routines
Update 82xx sysfs interface routines

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx base driver

Enable base 83xx adapter driver.

Common driver interface routines like probe,
interface up/down routines, irq and resource
allocation routines are modified to add support for 83xx
adapter.

Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx data path routines

Add 83xx adapter data path routines
Update few 82xx adapter data path routines
Modify datapath resource allocation routines

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: 83xx memory map and HW access routines

83xx adapter register map.
83xx hardware interface routines.

Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: macros for common register access

Refactor 82xx driver to support new adapter - Qlogic 83XX CNA
Use QLC_SHARED_REG_RD32 and QLC__SHARED_REG_WR32 macros
for 82xx and 83xx common register access.

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: change driver hardware interface mechanism

Refactor 82xx driver to support new adapter - Qlogic 83XX CNA

Create adapter abstraction layer and seperate 82xx hardware access routines.
Create mailbox based HW interface mechanism

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Add VF device ids and enable feature

Add the various VF device ids (of all supported hardware)
Add the calls to enable_sriov and disable_sriov to enable the
SR-IOV feature. This patch also advances the version and release
date of the bnx2x module.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support PF <-> VF Bulletin Board

The PF <-> VF Bulletin Board is a simple interface between the
PF and the VF. The main reason for the Bulletin Board is to allow
the PF to be the initiator. The VF publishes at 'acquire' stage
the GPA of a Bulletin Board structure it has allocated. The PF notes
this GPA in the VF database. The VF samples the Bulletin Board
periodically for new messages. The latest version of the BB is always
used.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support VF FLR

The FLR indication arrives as an attention from the management processor.
Upon VF flr all FLRed function in the indication have already been
released by Firmware and now we basically need to free the resources
allocated to those VFs, and clean any remainders from the device
(FLR final cleanup).

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support of PF driver of a VF release request

The 'release' request is the opposite of the 'acquire' request.
At release, all the resources allocated to the VF are reclaimed.
The release flow applies the close flow if applicable.
Note that there are actually two types of release:
1. The VF has been removed, and so issued a 'release' request
over the VF <-> PF Channel.
2. The PF is going down and so has to release all of it's VFs.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support of PF driver of a VF close request

The 'close' command is the opposite of an init request. Here the
queues of the VF are closed (if any are opened) and released.
This flow applies the 'q_teardown' flow on all the queues.
The VF state is changed by this request.
Interrupts are disabled for the VF when closed.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support of PF driver of a VF q_teardown request

The 'q_teardown' request is basically the opposite of the 'q_setup'.
Here the PF driver removes from the device the queue it opened against
the VF fastpath ring at 'setup_q' stage, along with all related
rx_mode info.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Support of PF driver of a VF q_filters request

The VF driver uses the 'q_filters' message on the VF <-> PF channel
for configuring an open queue, for example when the rxmode changes.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>