GitHub/exynos8895/android_kernel_samsung_universal8895.git
10 years agoipv4: remove the useless argument from ip_tunnel_hash()
Duan Jiong [Sun, 19 Jan 2014 08:43:42 +0000 (16:43 +0800)]
ipv4: remove the useless argument from ip_tunnel_hash()

Since commit c544193214("GRE: Refactor GRE tunneling code")
introduced function ip_tunnel_hash(), the argument itn is no
longer in use, so remove it.

Signed-off-by: Duan Jiong <duanj.fnst@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobond: make slave_sysfs_ops static
stephen hemminger [Sat, 18 Jan 2014 22:54:18 +0000 (14:54 -0800)]
bond: make slave_sysfs_ops static

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: remove unnecessary initializations in net_dev_init
Sabrina Dubroca [Sat, 18 Jan 2014 18:19:27 +0000 (19:19 +0100)]
net: remove unnecessary initializations in net_dev_init

softnet_data is already set to 0, no need to use memset or initialize
specific fields to 0 or NULL afterwards.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: add vxlan description
Jesse Brandeburg [Fri, 17 Jan 2014 19:00:33 +0000 (11:00 -0800)]
net: add vxlan description

Add a description to the vxlan module, helping save the world
from the minions of destruction and confusion.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
CC: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'sfc'
David S. Miller [Tue, 21 Jan 2014 22:46:56 +0000 (14:46 -0800)]
Merge branch 'sfc'

Shradha Shah says:

====================
Bug Fixes for SFC driver

I am taking over the upstream patch submission work for the
sfc driver from Ben Hutchings.

These patches are bug fixes to the sfc driver involving
replacement of the PORT RESET MC command and fixing transposed
ptp_{under,over}size_sync_window_statistics

The PORT_RESET bug fix is needed for all versions supporting EF10
i.e all versions including and after 3.12.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Fix transposed ptp_{under, over}size_sync_windows statistics
Ben Hutchings [Fri, 17 Jan 2014 19:48:17 +0000 (19:48 +0000)]
sfc: Fix transposed ptp_{under, over}size_sync_windows statistics

Somehow I transposed these two while bringing the original statistics
support upstream.

Fixes: 99691c4ac112 ('sfc: Add PTP counters to ethtool stats')
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Change efx_mcdi_reset_port to use ENTITY_RESET MC command.
Jon Cooper [Fri, 17 Jan 2014 19:48:06 +0000 (19:48 +0000)]
sfc: Change efx_mcdi_reset_port to use ENTITY_RESET MC command.

PORT_RESET MC command was NOP in the ef10 firmware hence we are using
ENTITY_RESET to make sure all resource allocations are reset.

Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: export tcf_hash_search() instead of tcf_hash_lookup()
WANG Cong [Fri, 17 Jan 2014 19:37:03 +0000 (11:37 -0800)]
net_sched: act: export tcf_hash_search() instead of tcf_hash_lookup()

So that we will not expose struct tcf_common to modules.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: fetch hinfo from a->ops->hinfo
WANG Cong [Fri, 17 Jan 2014 19:37:02 +0000 (11:37 -0800)]
net_sched: act: fetch hinfo from a->ops->hinfo

Every action ops has a pointer to hash info, so we don't need to
hard-code it in each module.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: fix NULL pointer dereference in stmmac_get_tx_hwtstamp
damuzi000 [Fri, 17 Jan 2014 15:47:59 +0000 (23:47 +0800)]
net: stmmac: fix NULL pointer dereference in stmmac_get_tx_hwtstamp

When timestamping is enabled, stmmac_tx_clean will call
stmmac_get_tx_hwtstamp to get tx TS.
But the skb can be NULL because the last of its tx_skbuff is NULL
if this packet frame is filled in more than one descriptors.

To fix the issue, change the code:
- Store TX skb to the tx_skbuff[] of frame's last segment.
- Check skb is not NULL in stmmac_get_tx_hwtstamp.

Signed-off-by: Bruce Liu <damuzi000@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: sunxi platform extensions for GMAC in Allwinner A20 SoC's
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:47 +0000 (21:24 +0800)]
net: stmmac: sunxi platform extensions for GMAC in Allwinner A20 SoC's

The Allwinner A20 has an ethernet controller that seems to be
an early version of Synopsys DesignWare MAC 10/100/1000 Universal,
which is supported by the stmmac driver.

Allwinner's GMAC requires setting additional registers in the SoC's
clock control unit.

The exact version of the DWMAC IP that Allwinner uses is unknown,
thus the exact feature set is unknown.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Use driver data and callbacks tied with compatible strings
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:46 +0000 (21:24 +0800)]
net: stmmac: Use driver data and callbacks tied with compatible strings

The stmmac driver core allows passing feature flags and callbacks via
platform data. Add a similar stmmac_of_data to pass flags and callbacks
tied to compatible strings. This allows us to extend stmmac with glue
layers for different SoCs.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Deprecate snps, phy-addr and auto-detect PHY address
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:45 +0000 (21:24 +0800)]
net: stmmac: Deprecate snps, phy-addr and auto-detect PHY address

The snps,phy-addr device tree property is non-standard, and should be
removed in favor of proper phy node support. Remove it from the binding
documents and warn if the property is still used.

Most PHYs respond to address 0, but a few don't, so auto-detect PHY
address by default, to make up for the lack of explicit address selection.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Honor DT parameter to force DMA store and forward mode
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:44 +0000 (21:24 +0800)]
net: stmmac: Honor DT parameter to force DMA store and forward mode

"snps,force_sf_dma_mode" is documented in stmmac device tree bindings,
but is never handled by the driver.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoblackfin: Update stmmac callback signatures
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:43 +0000 (21:24 +0800)]
blackfin: Update stmmac callback signatures

stmmac callbacks have been extended for better separation.
Update them to avoid breakage.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Allocate and pass soc/board specific data to callbacks
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:42 +0000 (21:24 +0800)]
net: stmmac: Allocate and pass soc/board specific data to callbacks

The current .init and .exit callbacks requires access to driver
private data structures. This is not a good seperation and abstraction.

Instead, we add a new .setup callback for allocating private data, and
pass the returned pointer to the other callbacks.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Add support for optional reset control
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:41 +0000 (21:24 +0800)]
net: stmmac: Add support for optional reset control

The DWMAC has a reset assert line, which is used on some SoCs. Add an
optional reset control to stmmac driver core.

To support reset control deferred probing, this patch changes the driver
probe function to return the actual error, instead of just -EINVAL.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: stmmac: Enable stmmac main clock when probing hardware
Chen-Yu Tsai [Fri, 17 Jan 2014 13:24:40 +0000 (21:24 +0800)]
net: stmmac: Enable stmmac main clock when probing hardware

The stmmac driver does not enable the main clock during the probe phase.
If the clock was not enabled by the boot loader or was disabled by the
kernel, hardware features and the MDIO bus would not be probed properly.

Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoDT: net: davinci_emac: "phy-handle" property is actually optional
Sergei Shtylyov [Thu, 16 Jan 2014 22:32:13 +0000 (01:32 +0300)]
DT: net: davinci_emac: "phy-handle" property is actually optional

Though described as required, the "phy-handle" property for the DaVinci EMAC
binding is actually optional, as the driver will happily function without it,
assuming 100/FULL link; the property is not specified  either in the example
device node,  or in the actual EMAC device nodes for DA850 and AM3517 device
trees.

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: fix "queues" uevent between network namespaces
Weilong Chen [Thu, 16 Jan 2014 09:24:31 +0000 (17:24 +0800)]
net: fix "queues" uevent between network namespaces

When I create a new namespace with 'ip netns add net0', or add/remove
new links in a namespace with 'ip link add/delete type veth', rx/tx
queues events can be got in all namespaces. That is because rx/tx queue
ktypes do not have namespace support, and their kobj parents are setted to
NULL. This patch is to fix it.

Reported-by: Libo Chen <chenlibo@huawei.com>
Signed-off-by: Libo Chen <chenlibo@huawei.com>
Signed-off-by: Weilong Chen <chenweilong@huawei.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: remove capab from struct tc_action_ops
WANG Cong [Wed, 15 Jan 2014 23:49:30 +0000 (15:49 -0800)]
net_sched: act: remove capab from struct tc_action_ops

It is not actually implemented.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: document accel_priv parameter for __dev_queue_xmit()
Jason Wang [Mon, 20 Jan 2014 03:25:13 +0000 (11:25 +0800)]
net: document accel_priv parameter for __dev_queue_xmit()

To silent "make htmldocs" warning.

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: fix missing SCTP mailing list address update
Jean Sacren [Sun, 19 Jan 2014 22:24:53 +0000 (15:24 -0700)]
sctp: fix missing SCTP mailing list address update

The commit 91705c61b5202 ("net: sctp: trivial: update mailing list
address") updated almost all the SCTP mailing list address from

"lksctp-developers@lists.sourceforge.net"
to
"linux-sctp@vger.kernel.org"

except for the one in include/linux/sctp.h file. Fix this way trivial
one so that all is updated.

Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Cc: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: optimize link local address search
Hannes Frederic Sowa [Sun, 19 Jan 2014 20:58:19 +0000 (21:58 +0100)]
ipv6: optimize link local address search

ipv6_link_dev_addr sorts newly added addresses by scope in
ifp->addr_list. Smaller scope addresses are added to the tail of the
list. Use this fact to iterate in reverse over addr_list and break out
as soon as a higher scoped one showes up, so we can spare some cycles
on machines with lot's of addresses.

The ordering of the addresses is not relevant and we are more likely to
get the eui64 generated address with this change anyway.

Suggested-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agophy: cleanup 10g code
stephen hemminger [Sun, 19 Jan 2014 19:48:20 +0000 (11:48 -0800)]
phy: cleanup 10g code

Code should avoid needless exports, don't export something unless it used.
Make local functions static and remove unused stubs.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: fix sparse warnings
Fengguang Wu [Sun, 19 Jan 2014 19:37:12 +0000 (11:37 -0800)]
qlcnic: fix sparse warnings

Previous patch changed prototypes, but forgot functions.

Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: make IPV6_RECVPKTINFO work for ipv4 datagrams
Hannes Frederic Sowa [Mon, 20 Jan 2014 02:43:08 +0000 (03:43 +0100)]
ipv6: make IPV6_RECVPKTINFO work for ipv4 datagrams

We currently don't report IPV6_RECVPKTINFO in cmsg access ancillary data
for IPv4 datagrams on IPv6 sockets.

This patch splits the ip6_datagram_recv_ctl into two functions, one
which handles both protocol families, AF_INET and AF_INET6, while the
ip6_datagram_recv_specific_ctl only handles IPv6 cmsg data.

ip6_datagram_recv_*_ctl never reported back any errors, so we can make
them return void. Also provide a helper for protocols which don't offer dual
personality to further use ip6_datagram_recv_ctl, which is exported to
modules.

I needed to shuffle the code for ping around a bit to make it easier to
implement dual personality for ping ipv6 sockets in future.

Reported-by: Gert Doering <gert@space.net>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosch_netem: replace magic numbers with enumerate
Yang Yingliang [Sat, 18 Jan 2014 10:13:31 +0000 (18:13 +0800)]
sch_netem: replace magic numbers with enumerate

Replace some magic numbers which describe states of 4-state model
loss generator with enumerate.

Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: add flowlabel_consistency sysctl
Florent Fourcot [Fri, 17 Jan 2014 16:15:05 +0000 (17:15 +0100)]
ipv6: add flowlabel_consistency sysctl

With the introduction of IPV6_FL_F_REFLECT, there is no guarantee of
flow label unicity. This patch introduces a new sysctl to protect the old
behaviour, enable by default.

Changelog of V3:
 * rename ip6_flowlabel_consistency to flowlabel_consistency
 * use net_info_ratelimited()
 * checkpatch cleanups

Signed-off-by: Florent Fourcot <florent.fourcot@enst-bretagne.fr>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: add a flag to get the flow label used remotly
Florent Fourcot [Fri, 17 Jan 2014 16:15:04 +0000 (17:15 +0100)]
ipv6: add a flag to get the flow label used remotly

This information is already available via IPV6_FLOWINFO
of IPV6_2292PKTOPTIONS, and them a filtering to get the flow label
information. But it is probably logical and easier for users to add this
here, and to control both sent/received flow label values with the
IPV6_FLOWLABEL_MGR option.

Signed-off-by: Florent Fourcot <florent.fourcot@enst-bretagne.fr>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: add the IPV6_FL_F_REFLECT flag to IPV6_FL_A_GET
Florent Fourcot [Fri, 17 Jan 2014 16:15:03 +0000 (17:15 +0100)]
ipv6: add the IPV6_FL_F_REFLECT flag to IPV6_FL_A_GET

With this option, the socket will reply with the flow label value read
on received packets.

The goal is to have a connection with the same flow label in both
direction of the communication.

Changelog of V4:
 * Do not erase the flow label on the listening socket. Use pktopts to
 store the received value

Signed-off-by: Florent Fourcot <florent.fourcot@enst-bretagne.fr>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv4: be friend with drop monitor
Eric Dumazet [Sun, 19 Jan 2014 02:27:49 +0000 (18:27 -0800)]
ipv4: be friend with drop monitor

Replace some dev_kfree_skb() with kfree_skb() calls when
we drop one skb, this might help bug tracking.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: add build-time checks for msg->msg_name size
Steffen Hurrle [Fri, 17 Jan 2014 21:53:15 +0000 (22:53 +0100)]
net: add build-time checks for msg->msg_name size

This is a follow-up patch to f3d3342602f8bc ("net: rework recvmsg
handler msg_name and msg_namelen logic").

DECLARE_SOCKADDR validates that the structure we use for writing the
name information to is not larger than the buffer which is reserved
for msg->msg_name (which is 128 bytes). Also use DECLARE_SOCKADDR
consistently in sendmsg code paths.

Signed-off-by: Steffen Hurrle <steffen@hurrle.net>
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: introduce SO_BPF_EXTENSIONS
Michal Sekletar [Fri, 17 Jan 2014 16:09:45 +0000 (17:09 +0100)]
net: introduce SO_BPF_EXTENSIONS

For user space packet capturing libraries such as libpcap, there's
currently only one way to check which BPF extensions are supported
by the kernel, that is, commit aa1113d9f85d ("net: filter: return
-EINVAL if BPF_S_ANC* operation is not supported"). For querying all
extensions at once this might be rather inconvenient.

Therefore, this patch introduces a new option which can be used as
an argument for getsockopt(), and allows one to obtain information
about which BPF extensions are supported by the current kernel.

As David Miller suggests, we do not need to define any bits right
now and status quo can just return 0 in order to state that this
versions supports SKF_AD_PROTOCOL up to SKF_AD_PAY_OFFSET. Later
additions to BPF extensions need to add their bits to the
bpf_tell_extensions() function, as documented in the comment.

Signed-off-by: Michal Sekletar <msekleta@redhat.com>
Cc: David Miller <davem@davemloft.net>
Reviewed-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Sat, 18 Jan 2014 08:55:41 +0000 (00:55 -0800)]
Merge git://git./linux/kernel/git/davem/net

Conflicts:
drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
net/ipv4/tcp_metrics.c

Overlapping changes between the "don't create two tcp metrics objects
with the same key" race fix in net and the addition of the destination
address in the lookup key in net-next.

Minor overlapping changes in bnx2x driver.

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sat, 18 Jan 2014 06:19:28 +0000 (22:19 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) The value choosen for the new SO_MAX_PACING_RATE socket option on
    parisc was very poorly choosen, let's fix it while we still can.
    From Eric Dumazet.

 2) Our generic reciprocal divide was found to handle some edge cases
    incorrectly, part of this is encoded into the BPF as deep as the JIT
    engines themselves.  Just use a real divide throughout for now.
    From Eric Dumazet.

 3) Because the initial lookup is lockless, the TCP metrics engine can
    end up creating two entries for the same lookup key.  Fix this by
    doing a second lookup under the lock before we actually create the
    new entry.  From Christoph Paasch.

 4) Fix scatter-gather list init in usbnet driver, from Bjørn Mork.

 5) Fix unintended 32-bit truncation in cxgb4 driver's bit shifting.
    From Dan Carpenter.

 6) Netlink socket dumping uses the wrong socket state for timewait
    sockets.  Fix from Neal Cardwell.

 7) Fix netlink memory leak in ieee802154_add_iface(), from Christian
    Engelmayer.

 8) Multicast forwarding in ipv4 can overflow the per-rule reference
    counts, causing all multicast traffic to cease.  Fix from Hannes
    Frederic Sowa.

 9) via-rhine needs to stop all TX queues when it resets the device,
    from Richard Weinberger.

10) Fix RDS per-cpu accesses broken by the this_cpu_* conversions.  From
    Gerald Schaefer.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  s390/bpf,jit: fix 32 bit divisions, use unsigned divide instructions
  parisc: fix SO_MAX_PACING_RATE typo
  ipv6: simplify detection of first operational link-local address on interface
  tcp: metrics: Avoid duplicate entries with the same destination-IP
  net: rds: fix per-cpu helper usage
  e1000e: Fix compilation warning when !CONFIG_PM_SLEEP
  bpf: do not use reciprocal divide
  be2net: add dma_mapping_error() check for dma_map_page()
  bnx2x: Don't release PCI bars on shutdown
  net,via-rhine: Fix tx_timeout handling
  batman-adv: fix batman-adv header overhead calculation
  qlge: Fix vlan netdev features.
  net: avoid reference counter overflows on fib_rules in multicast forwarding
  dm9601: add USB IDs for new dm96xx variants
  MAINTAINERS: add virtio-dev ML for virtio
  ieee802154: Fix memory leak in ieee802154_add_iface()
  net: usbnet: fix SG initialisation
  inet_diag: fix inet_diag_dump_icsk() to use correct state for timewait sockets
  cxgb4: silence shift wrapping static checker warning

10 years agoMerge branch 'ixgbevf'
David S. Miller [Sat, 18 Jan 2014 03:16:02 +0000 (19:16 -0800)]
Merge branch 'ixgbevf'

Aaron Brown says:

====================
Intel Wired LAN Driver Updates

This series contains updates from Emil to ixgbevf.

He cleans up the code by removing the adapter structure as a
parameter from multiple functions in favor of using the ixgbevf_ring
structure and moves hot-path specific statistic int the ring
structure for anticipated performance gains.

He also removes the Tx/Rx counters for checksum offload and adds
counters for tx_restart_queue and tx_timeout_count.

Next he makes it so that the first tx_buffer structure acts as a
central storage location for most the skb info we are about to
transmit, then takes advantage of the dma buffer always being
present in the first descriptor and mapped as single allowing a
call to dma_unmap_single which alleviates the need to check for
DMA mapping in ixgbevf_clean_tx_irq().

Finally he merges the ixgbevf_tx_map call and the ixgbevf_tx_queue
call into a single function.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: merge ixgbevf_tx_map and ixgbevf_tx_queue into a single function
Emil Tantilov [Sat, 18 Jan 2014 02:30:05 +0000 (18:30 -0800)]
ixgbevf: merge ixgbevf_tx_map and ixgbevf_tx_queue into a single function

This change merges the ixgbevf_tx_map call and the ixgbevf_tx_queue call
into a single function.  In order to make room for this setting of cmd_type
and olinfo flags is done in separate functions.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: redo dma mapping using the tx buffer info
Emil Tantilov [Sat, 18 Jan 2014 02:30:04 +0000 (18:30 -0800)]
ixgbevf: redo dma mapping using the tx buffer info

This patch takes advantage of the dma buffer always being present in the
first descriptor and mapped as single. As such we can call dma_unmap_single
and don't need to check for DMA mapping in ixgbevf_clean_tx_irq().

In addition this patch makes use of the DMA API.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: make the first tx_buffer a repository for most of the skb info
Emil Tantilov [Sat, 18 Jan 2014 02:30:03 +0000 (18:30 -0800)]
ixgbevf: make the first tx_buffer a repository for most of the skb info

This change makes it so that the first tx_buffer structure acts as a
central storage location for most of the info about the skb we are about
to transmit.

In addition this patch makes tx_flags part of the ixgbevf_tx_buffer struct.
This allows us to use the flags directly from the stucture and as result
removes the tx_flags parameter from some functions. Also as a cleanup
mapped_as_page is folded into tx_flags and some unused flags were removed.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: add tx counters
Emil Tantilov [Sat, 18 Jan 2014 02:30:02 +0000 (18:30 -0800)]
ixgbevf: add tx counters

This patch adds counters for tx_restart_queue and tx_timeout_count.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: remove counters for Tx/Rx checksum offload
Emil Tantilov [Sat, 18 Jan 2014 02:30:01 +0000 (18:30 -0800)]
ixgbevf: remove counters for Tx/Rx checksum offload

This patch removes the Tx/Rx counters for checksum offload.

The Tx counter was never updated and the Rx counter is of limited use.
This is in effort to clean up the counters and make them consistent
with the counters shown by ixgbe.

Also this patch removes some members of the adapter structure that were
never used and shuffles others to reduce number of holes.

before:
/* size: 1568, cachelines: 25, members: 48 */
/* sum members: 1519, holes: 10, sum holes: 43 */
/* padding: 6 */
/* last cacheline: 32 bytes */

after:
/* size: 1480, cachelines: 24, members: 43 */
/* sum members: 1479, holes: 1, sum holes: 1 */
/* last cacheline: 8 bytes */

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: move ring specific stats into ring specific structure
Emil Tantilov [Sat, 18 Jan 2014 02:30:00 +0000 (18:30 -0800)]
ixgbevf: move ring specific stats into ring specific structure

This patch moves hot-path specific statistics into the ring structure.
This allows us to drop the adapter structure in some functions and should
help with performance.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: make use of the dev pointer in the ixgbevf_ring struct
Emil Tantilov [Sat, 18 Jan 2014 02:29:59 +0000 (18:29 -0800)]
ixgbevf: make use of the dev pointer in the ixgbevf_ring struct

This patch cleans up the code by removing the adapter structure as
parameter from multiple functions. The adapter structure was previously
being used to access the dev pointer, but this can also be done via the
ixgbevf_ring structure. This way we can drop the adapter as parameter from
these functions.

This patch also includes small cleanups in some error code paths.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'i40e'
David S. Miller [Sat, 18 Jan 2014 03:14:05 +0000 (19:14 -0800)]
Merge branch 'i40e'

Aaron Brown says:

====================
Intel Wired LAN Driver Updates

This series contains updates to i40e.

Neerav implements DCB and DCBNL support and adds DCB options
to Kconfig.  DCB is disabled by default.

Anjali refactors flow control director to fix inconsistencies
that were preventing clean unloads of the driver, move the
queues for handling flow director error into their own hardware
VSI and implement a corrected version of the basic ethtool add
ntuple rule.

Jesse provides fixes for a compiler warning, firmware workaround,
white space fixes and renames some defines.

Shannon reworks the device ID #defines to follow the
DEV_ID_ convention followed by our other drivers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoBluetooth: remove direct compilation of 6lowpan_iphc.c
Stephen Warren [Fri, 17 Jan 2014 19:29:24 +0000 (12:29 -0700)]
Bluetooth: remove direct compilation of 6lowpan_iphc.c

It's now built as a separate utility module, and enabling BT selects
that module in Kconfig. This fixes:

net/ieee802154/built-in.o:(___ksymtab_gpl+lowpan_process_data+0x0): multiple definition of `__ksymtab_lowpan_process_data'
net/bluetooth/built-in.o:(___ksymtab_gpl+lowpan_process_data+0x0): first defined here
net/ieee802154/built-in.o:(___ksymtab_gpl+lowpan_header_compress+0x0): multiple definition of `__ksymtab_lowpan_header_compress'
net/bluetooth/built-in.o:(___ksymtab_gpl+lowpan_header_compress+0x0): first defined here
net/ieee802154/built-in.o: In function `lowpan_header_compress':
net/ieee802154/6lowpan_iphc.c:606: multiple definition of `lowpan_header_compress'
net/bluetooth/built-in.o:/home/swarren/shared/git_wa/kernel/kernel.git/net/bluetooth/../ieee802154/6lowpan_iphc.c:606: first defined here
net/ieee802154/built-in.o: In function `lowpan_process_data':
net/ieee802154/6lowpan_iphc.c:344: multiple definition of `lowpan_process_data'
net/bluetooth/built-in.o:/home/swarren/shared/git_wa/kernel/kernel.git/net/bluetooth/../ieee802154/6lowpan_iphc.c:344: first defined here
make[1]: *** [net/built-in.o] Error 1

(this change probably simply wasn't "git add"d to a53d34c3465b)

Fixes: a53d34c3465b ("net: move 6lowpan compression code to separate module")
Fixes: 18722c247023 ("Bluetooth: Enable 6LoWPAN support for BT LE devices")
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: Fix device ID define names to align to standard
Shannon Nelson [Fri, 17 Jan 2014 23:36:39 +0000 (15:36 -0800)]
i40e: Fix device ID define names to align to standard

Rework the device ID #defines to follow the _DEV_ID convention
already established in the other Intel drivers.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: add DCB option to Kconfig
Neerav Parikh [Fri, 17 Jan 2014 23:36:38 +0000 (15:36 -0800)]
i40e: add DCB option to Kconfig

Allow compiling DCB related files if I40E_DCB option
is supported in the kernel configuration.

DCB is disabled by default.

Signed-off-by: Neerav Parikh <Neerav.Parikh@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Tested-By: Jack Morgan<jack.morgan@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: add DCB and DCBNL support
Neerav Parikh [Fri, 17 Jan 2014 23:36:37 +0000 (15:36 -0800)]
i40e: add DCB and DCBNL support

This patch adds capability to configure DCB on i40e network
interfaces using Intel XL710 adapter firmware APIs.

By default all VSIs are only enabled for the default traffic
class enabled by firmware for any given PF. The driver would
query the firmware for the traffic classes that are enabled for
the port and reconfigure the LAN VSI to match to the port traffic
class settings. All other VSIs are only enabled for the default
traffic class settings for now.

The driver registers and listens to firmware events that may
require change in the DCB settings. It may reconfigure the VSI
settings based on these events.

This patch exposes IEEE DCBNL interfaces for the i40e driver to
allow any application to query the DCB settings on the adapter.

Signed-off-by: Neerav Parikh <Neerav.Parikh@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-By: Jack Morgan<jack.morgan@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: implement DCB support infastructure
Neerav Parikh [Fri, 17 Jan 2014 23:36:36 +0000 (15:36 -0800)]
i40e: implement DCB support infastructure

Intel XL710 series of adapters support QoS as per the
IEEE 802.1 DCB (Data Center Bridging) standard.
This is supported in conjuction with:
- Enhanced Transmission Selection (ETS) - IEEE 802.1Qaz
- Priority Flow Control (PFC) - IEEE 802.1Qbb
- DCB eXchange Protocol (DCBX) - IEEE 802.1Qaz

On Intel XL710 adapters DCBX is performed by the adapter
firmware. The firmware runs DCBX in willing mode and configures
the port as per the DCB settings recommended by it's link
partner.

By default in absence of any DCBX; firmware would configure the
port with a single traffic class and all of the port bandwith
will be allocated to that traffic class.

This patch adds functions and calls to support querying and
configuring DCB using firmware APIs.

Signed-off-by: Neerav Parikh <Neerav.Parikh@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-By: Jack Morgan<jack.morgan@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: refactor flow director
Anjali Singhai Jain [Fri, 17 Jan 2014 23:36:35 +0000 (15:36 -0800)]
i40e: refactor flow director

The i40e hardware was generating some inconsistent results
when using current programming methods.  This refactor
fixes the inconsistencies that were preventing clean
unloads of the driver, and moves the queues for handling
flow director errors into their own hardware VSI.

This patch also implements a corrected version of the
basic ethtool add ntuple rule, which will disable
the driver's automatic flow programming.  A future patch
adds remove/replay/list support for ntuple.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: rename defines
Jesse Brandeburg [Fri, 17 Jan 2014 23:36:34 +0000 (15:36 -0800)]
i40e: rename defines

The FLAG_FDIR_* defines can be renamed to be more descriptive.
This patch is in preparation for the following where the fdir
code is refactored.

Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: whitespace fixes
Jesse Brandeburg [Fri, 17 Jan 2014 23:36:33 +0000 (15:36 -0800)]
i40e: whitespace fixes

Fix more whitespace issues, including making some locals declared
in a nicer order.

Also update Copyright string printed when the driver loads.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: Change firmware workaround
Jesse Brandeburg [Fri, 17 Jan 2014 23:36:32 +0000 (15:36 -0800)]
i40e: Change firmware workaround

Remove a workaround that is no longer necessary and implement
a better understanding of what firmware is returning in the MSI-X
vector count.  This makes it so that the driver ends up with the
right amount of queues when using all available MSI-X vectors.

Change-ID: I34e60cc71dcfb1b5412f37df956fedcc49ade187
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: fix compile warning on checksum_local
Jesse Brandeburg [Fri, 17 Jan 2014 23:36:31 +0000 (15:36 -0800)]
i40e: fix compile warning on checksum_local

Compile testing with higher warning levels found this complaint:
i40e_nvm.c: warning: 'checksum_local' may be used uninitialized in
this function

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoBluetooth: remove direct compilation of 6lowpan_iphc.c
Stephen Warren [Fri, 17 Jan 2014 19:29:24 +0000 (12:29 -0700)]
Bluetooth: remove direct compilation of 6lowpan_iphc.c

It's now built as a separate utility module, and enabling BT selects
that module in Kconfig. This fixes:

net/ieee802154/built-in.o:(___ksymtab_gpl+lowpan_process_data+0x0): multiple definition of `__ksymtab_lowpan_process_data'
net/bluetooth/built-in.o:(___ksymtab_gpl+lowpan_process_data+0x0): first defined here
net/ieee802154/built-in.o:(___ksymtab_gpl+lowpan_header_compress+0x0): multiple definition of `__ksymtab_lowpan_header_compress'
net/bluetooth/built-in.o:(___ksymtab_gpl+lowpan_header_compress+0x0): first defined here
net/ieee802154/built-in.o: In function `lowpan_header_compress':
net/ieee802154/6lowpan_iphc.c:606: multiple definition of `lowpan_header_compress'
net/bluetooth/built-in.o:/home/swarren/shared/git_wa/kernel/kernel.git/net/bluetooth/../ieee802154/6lowpan_iphc.c:606: first defined here
net/ieee802154/built-in.o: In function `lowpan_process_data':
net/ieee802154/6lowpan_iphc.c:344: multiple definition of `lowpan_process_data'
net/bluetooth/built-in.o:/home/swarren/shared/git_wa/kernel/kernel.git/net/bluetooth/../ieee802154/6lowpan_iphc.c:344: first defined here
make[1]: *** [net/built-in.o] Error 1

(this change probably simply wasn't "git add"d to a53d34c3465b)

Fixes: a53d34c3465b ("net: move 6lowpan compression code to separate module")
Fixes: 18722c247023 ("Bluetooth: Enable 6LoWPAN support for BT LE devices")
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovirtio-net: fix build error when CONFIG_AVERAGE is not enabled
Michael Dalton [Fri, 17 Jan 2014 09:27:08 +0000 (01:27 -0800)]
virtio-net: fix build error when CONFIG_AVERAGE is not enabled

Commit ab7db91705e9 ("virtio-net: auto-tune mergeable rx buffer size for
improved performance") introduced a virtio-net dependency on EWMA.
The inclusion of EWMA is controlled by CONFIG_AVERAGE. Fix build error
when CONFIG_AVERAGE is not enabled by adding select AVERAGE to
virtio-net's Kconfig entry.

Build failure reported using config make ARCH=s390 defconfig.

Signed-off-by: Michael Dalton <mwdalton@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'ixgbe'
David S. Miller [Sat, 18 Jan 2014 02:57:35 +0000 (18:57 -0800)]
Merge branch 'ixgbe'

Aaron Brown says:

====================
Intel Wired LAN Driver Updates

This series contains an updates to ixgbe and ixgbevf.

Jacob add braces around some ixgbe_qv_lock_* calls lto better adhere
to Kernel style guidelines.  Don bumps the versions on ixgbe and
ixgbevf to match internal driver functionality better.
====================

Reviewed-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: bump version
Don Skidmore [Fri, 17 Jan 2014 09:21:38 +0000 (01:21 -0800)]
ixgbevf: bump version

Bump the version number to better match functionality provided with out of
tree driver of the same version.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbe: bump version number
Don Skidmore [Fri, 17 Jan 2014 09:21:37 +0000 (01:21 -0800)]
ixgbe: bump version number

Bump the version number to better match functionality provided with out
of tree driver of the same version.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbe: add braces around else condition in ixgbe_qv_lock_* calls
Jacob Keller [Fri, 17 Jan 2014 09:21:36 +0000 (01:21 -0800)]
ixgbe: add braces around else condition in ixgbe_qv_lock_* calls

This patch adds braces around the ixgbe_qv_lock_* calls which previously only
had braces around the if portion. Kernel style guidelines for this require
parenthesis around all conditions if they are required around one. In addition
the comment while not illegal C syntax makes the code look wrong at a cursory
glance. This patch corrects the style and adds braces so that the full if-else
block is uniform.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agos390/bpf,jit: fix 32 bit divisions, use unsigned divide instructions
Heiko Carstens [Fri, 17 Jan 2014 08:37:15 +0000 (09:37 +0100)]
s390/bpf,jit: fix 32 bit divisions, use unsigned divide instructions

The s390 bpf jit compiler emits the signed divide instructions "dr" and "d"
for unsigned divisions.
This can cause problems: the dividend will be zero extended to a 64 bit value
and the divisor is the 32 bit signed value as specified A or X accumulator,
even though A and X are supposed to be treated as unsigned values.

The divide instrunctions will generate an exception if the result cannot be
expressed with a 32 bit signed value.
This is the case if e.g. the dividend is 0xffffffff and the divisor either 1
or also 0xffffffff (signed: -1).

To avoid all these issues simply use unsigned divide instructions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: ftgmac100: use kfree_skb() where appropriate
Eric Dumazet [Fri, 17 Jan 2014 07:38:24 +0000 (23:38 -0800)]
net: ftgmac100: use kfree_skb() where appropriate

In order to get correct drop monitor notifications for dropped
packets, we should call kfree_skb() instead of dev_kfree_skb()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'bonding_slave_sysfs'
David S. Miller [Sat, 18 Jan 2014 02:52:58 +0000 (18:52 -0800)]
Merge branch 'bonding_slave_sysfs'

Scott Feldman says:

====================
bonding: add slave netlink and sysfs support

v2:

  - Address review comment from Ding (and Veacesiav): handle kobj cleanup
    if sysfs_create_file() fails when adding slave attribute nodes.

v1:

  The following series adds bonding slave netlink and sysfs interfaces.
  Slave interfaces get a new IFLA_SLAVE set of netlink attributes, along
  with RTM_NEWLINK notification when slave's active status changes.  The
  sysfs interface adds read-only nodes for slave attributes under a /slave
  dir, simliar to how bond interfaces get a /bonding dir for bonding
  attributes.
====================

Reviewed-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: add netlink attributes to slave link dev
sfeldma@cumulusnetworks.com [Fri, 17 Jan 2014 06:57:56 +0000 (22:57 -0800)]
bonding: add netlink attributes to slave link dev

If link is IFF_SLAVE, extend link dev netlink attributes to include
slave attributes with new IFLA_SLAVE nest.  Add netlink notification
(RTM_NEWLINK) when slave status changes from backup to active, or
visa-versa.

Adds new ndo_get_slave op to net_device_ops to fill skb with IFLA_SLAVE
attributes.  Currently only used by bonding driver, but could be
used by other aggregating devices with slaves.

Signed-off-by: Scott Feldman <sfeldma@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: add sysfs /slave dir for bond slave devices.
sfeldma@cumulusnetworks.com [Fri, 17 Jan 2014 06:57:49 +0000 (22:57 -0800)]
bonding: add sysfs /slave dir for bond slave devices.

Add sub-directory under /sys/class/net/<interface>/slave with
read-only attributes for slave.  Directory only appears when
<interface> is a slave.

$ tree /sys/class/net/eth2/slave/
/sys/class/net/eth2/slave/
├── ad_aggregator_id
├── link_failure_count
├── mii_status
├── perm_hwaddr
├── queue_id
└── state

$ cat /sys/class/net/eth2/slave/*
2
0
up
40:02:10:ef:06:01
0
active

Signed-off-by: Scott Feldman <sfeldma@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: vxlan: do not use vxlan_net before checking event type
Daniel Borkmann [Fri, 17 Jan 2014 11:55:06 +0000 (12:55 +0100)]
net: vxlan: do not use vxlan_net before checking event type

Jesse Brandeburg reported that commit acaf4e70997f caused a panic
when adding a network namespace while vxlan module was present in
the system:

[<ffffffff814d0865>] vxlan_lowerdev_event+0xf5/0x100
[<ffffffff816e9e5d>] notifier_call_chain+0x4d/0x70
[<ffffffff810912be>] __raw_notifier_call_chain+0xe/0x10
[<ffffffff810912d6>] raw_notifier_call_chain+0x16/0x20
[<ffffffff815d9610>] call_netdevice_notifiers_info+0x40/0x70
[<ffffffff815d9656>] call_netdevice_notifiers+0x16/0x20
[<ffffffff815e1bce>] register_netdevice+0x1be/0x3a0
[<ffffffff815e1dce>] register_netdev+0x1e/0x30
[<ffffffff814cb94a>] loopback_net_init+0x4a/0xb0
[<ffffffffa016ed6e>] ? lockd_init_net+0x6e/0xb0 [lockd]
[<ffffffff815d6bac>] ops_init+0x4c/0x150
[<ffffffff815d6d23>] setup_net+0x73/0x110
[<ffffffff815d725b>] copy_net_ns+0x7b/0x100
[<ffffffff81090e11>] create_new_namespaces+0x101/0x1b0
[<ffffffff81090f45>] copy_namespaces+0x85/0xb0
[<ffffffff810693d5>] copy_process.part.26+0x935/0x1500
[<ffffffff811d5186>] ? mntput+0x26/0x40
[<ffffffff8106a15c>] do_fork+0xbc/0x2e0
[<ffffffff811b7f2e>] ? ____fput+0xe/0x10
[<ffffffff81089c5c>] ? task_work_run+0xac/0xe0
[<ffffffff8106a406>] SyS_clone+0x16/0x20
[<ffffffff816ee689>] stub_clone+0x69/0x90
[<ffffffff816ee329>] ? system_call_fastpath+0x16/0x1b

Apparently loopback device is being registered first and thus we
receive an event notification when vxlan_net is not ready. Hence,
when we call net_generic() and request vxlan_net_id, we seem to
access garbage at that point in time. In setup_net() where we set
up a newly allocated network namespace, we traverse the list of
pernet ops ...

list_for_each_entry(ops, &pernet_list, list) {
error = ops_init(ops, net);
if (error < 0)
goto out_undo;
}

... and loopback_net_init() is invoked first here, so in the middle
of setup_net() we get this notification in vxlan. As currently we
only care about devices that unregister, move access through
net_generic() there. Fix is based on Cong Wang's proposal, but
only changes what is needed here. It sucks a bit as we only work
around the actual cure: right now it seems the only way to check if
a netns actually finished traversing all init ops would be to check
if it's part of net_namespace_list. But that I find quite expensive
each time we go through a notifier callback. Anyway, did a couple
of tests and it seems good for now.

Fixes: acaf4e70997f ("net: vxlan: when lower dev unregisters remove vxlan dev as well")
Reported-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Tested-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'ixgbe'
David S. Miller [Sat, 18 Jan 2014 02:38:13 +0000 (18:38 -0800)]
Merge branch 'ixgbe'

Aaron Brown says:

====================
Intel Wired LAN Driver Updates

This series contains updates to ixgbe Ethan Zhao.  The first one replaces
the magic number "63" with a macro, IXGBE_MAX_VFS_DRV_LIMIT, the second
moves the call to set driver_max_VFS to before SRIOV is enabled.

The code of these patches match the v3 (1/2) and v2 (2/2) versions sent
to the e1000-devel and netdev mailing lists.  The intermediate versions
(v4, v5) are from sorting out style issues, mostly tabs to spaces and
split lines probably introduced via mailer.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbe: set driver_max_VFs should be done before enabling SRIOV
ethan.zhao [Fri, 17 Jan 2014 03:41:05 +0000 (19:41 -0800)]
ixgbe: set driver_max_VFs should be done before enabling SRIOV

commit 43dc4e01 Limit number of reported VFs to device
 specific value It doesn't work and always returns -EBUSY because VFs are
 already enabled.

ixgbe_enable_sriov()
        pci_enable_sriov()
                sriov_enable()
                {
                ... ..
                iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
                pci_cfg_access_lock(dev);
                ... ...
                }

pci_sriov_set_totalvfs()
{
... ...
if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
                return -EBUSY;
...
}

So should set driver_max_VFs with pci_sriov_set_totalvfs() before
enable VFs with ixgbe_enable_sriov().

V2: revised for net-next tree.

Signed-off-by: Ethan Zhao <ethan.kernel@gmail.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbe: define IXGBE_MAX_VFS_DRV_LIMIT macro and cleanup const 63
ethan.zhao [Fri, 17 Jan 2014 03:41:04 +0000 (19:41 -0800)]
ixgbe: define IXGBE_MAX_VFS_DRV_LIMIT macro and cleanup const 63

Because ixgbe driver limit the max number of VF
 functions could be enabled to 63, so define one macro IXGBE_MAX_VFS_DRV_LIMIT
 and cleanup the const 63 in code.

v3: revised for net-next tree.

Signed-off-by: Ethan Zhao <ethan.kernel@gmail.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv4: fix a dst leak in tunnels
Eric Dumazet [Fri, 17 Jan 2014 00:41:19 +0000 (16:41 -0800)]
ipv4: fix a dst leak in tunnels

This patch :

1) Remove a dst leak if DST_NOCACHE was set on dst
   Fix this by holding a reference only if dst really cached.

2) Remove a lockdep warning in __tunnel_dst_set()
    This was reported by Cong Wang.

3) Remove usage of a spinlock where xchg() is enough

4) Remove some spurious inline keywords.
   Let compiler decide for us.

Fixes: 7d442fab0a67 ("ipv4: Cache dst in tunnels")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Cong Wang <cwang@twopensource.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: Add support for r7s72100
Simon Horman [Fri, 17 Jan 2014 00:22:28 +0000 (09:22 +0900)]
sh_eth: Add support for r7s72100

The r7s72100 SoC includes a fast ethernet controller.

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Acked-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: Use bool as return type of sh_eth_is_gether()
Simon Horman [Fri, 17 Jan 2014 00:22:27 +0000 (09:22 +0900)]
sh_eth: Use bool as return type of sh_eth_is_gether()

Return a boolean from sh_eth_is_gether() and refactor it as a one-liner.

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Acked-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: send Change Status Report after DAD is completed
Flavio Leitner [Thu, 16 Jan 2014 21:27:59 +0000 (19:27 -0200)]
ipv6: send Change Status Report after DAD is completed

The RFC 3810 defines two type of messages for multicast
listeners. The "Current State Report" message, as the name
implies, refreshes the *current* state to the querier.
Since the querier sends Query messages periodically, there
is no need to retransmit the report.

On the other hand, any change should be reported immediately
using "State Change Report" messages. Since it's an event
triggered by a change and that it can be affected by packet
loss, the rfc states it should be retransmitted [RobVar] times
to make sure routers will receive timely.

Currently, we are sending "Current State Reports" after
DAD is completed.  Before that, we send messages using
unspecified address (::) which should be silently discarded
by routers.

This patch changes to send "State Change Report" messages
after DAD is completed fixing the behavior to be RFC compliant
and also to pass TAHI IPv6 testsuite.

Signed-off-by: Flavio Leitner <fbl@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoparisc: fix SO_MAX_PACING_RATE typo
Eric Dumazet [Thu, 16 Jan 2014 19:15:12 +0000 (11:15 -0800)]
parisc: fix SO_MAX_PACING_RATE typo

SO_MAX_PACING_RATE definition on parisc got a typo.
Its not too late to fix it, before 3.13 is official.

Fixes: 62748f32d501 ("net: introduce SO_MAX_PACING_RATE")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: simplify detection of first operational link-local address on interface
Hannes Frederic Sowa [Thu, 16 Jan 2014 19:13:04 +0000 (20:13 +0100)]
ipv6: simplify detection of first operational link-local address on interface

In commit 1ec047eb4751e3 ("ipv6: introduce per-interface counter for
dad-completed ipv6 addresses") I build the detection of the first
operational link-local address much to complex. Additionally this code
now has a race condition.

Replace it with a much simpler variant, which just scans the address
list when duplicate address detection completes, to check if this is
the first valid link local address and send RS and MLD reports then.

Fixes: 1ec047eb4751e3 ("ipv6: introduce per-interface counter for dad-completed ipv6 addresses")
Reported-by: Jiri Pirko <jiri@resnulli.us>
Cc: Flavio Leitner <fbl@redhat.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Flavio Leitner <fbl@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotcp: metrics: Avoid duplicate entries with the same destination-IP
Christoph Paasch [Thu, 16 Jan 2014 19:01:21 +0000 (20:01 +0100)]
tcp: metrics: Avoid duplicate entries with the same destination-IP

Because the tcp-metrics is an RCU-list, it may be that two
soft-interrupts are inside __tcp_get_metrics() for the same
destination-IP at the same time. If this destination-IP is not yet part of
the tcp-metrics, both soft-interrupts will end up in tcpm_new and create
a new entry for this IP.
So, we will have two tcp-metrics with the same destination-IP in the list.

This patch checks twice __tcp_get_metrics(). First without holding the
lock, then while holding the lock. The second one is there to confirm
that the entry has not been added by another soft-irq while waiting for
the spin-lock.

Fixes: 51c5d0c4b169b (tcp: Maintain dynamic metrics in local cache.)
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: remove unused code
stephen hemminger [Thu, 16 Jan 2014 18:32:21 +0000 (10:32 -0800)]
qlcnic: remove unused code

Remove function  qlcnic_enable_eswitch which was defined
but never used in current code.

Compile tested only.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: make local functions static
stephen hemminger [Thu, 16 Jan 2014 18:31:27 +0000 (10:31 -0800)]
qlcnic: make local functions static

Functions only used in one file should be static.
Found by running make namespacecheck

Compile tested only.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: tcp: fix flowlabel value in ACK messages send from TIME_WAIT
Florent Fourcot [Thu, 16 Jan 2014 16:21:22 +0000 (17:21 +0100)]
ipv6: tcp: fix flowlabel value in ACK messages send from TIME_WAIT

This patch is following the commit b903d324bee262 (ipv6: tcp: fix TCLASS
value in ACK messages sent from TIME_WAIT).

For the same reason than tclass, we have to store the flow label in the
inet_timewait_sock to provide consistency of flow label on the last ACK.

Signed-off-by: Florent Fourcot <florent.fourcot@enst-bretagne.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: rds: fix per-cpu helper usage
Gerald Schaefer [Thu, 16 Jan 2014 15:54:48 +0000 (16:54 +0100)]
net: rds: fix per-cpu helper usage

commit ae4b46e9d "net: rds: use this_cpu_* per-cpu helper" broke per-cpu
handling for rds. chpfirst is the result of __this_cpu_read(), so it is
an absolute pointer and not __percpu. Therefore, __this_cpu_write()
should not operate on chpfirst, but rather on cache->percpu->first, just
like __this_cpu_read() did before.

Cc: <stable@vger.kernel.org> # 3.8+
Signed-off-byd Gerald Schaefer <gerald.schaefer@de.ibm.com>

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'for-davem' of git://git.kernel.org/pub/scm/linux/kernel/git/linville...
David S. Miller [Sat, 18 Jan 2014 01:30:55 +0000 (17:30 -0800)]
Merge branch 'for-davem' of git://git./linux/kernel/git/linville/wireless-next

John W. Linville says:

====================
Please pull this batch of updates for the 3.14 stream!

For the mac80211 bits, Johannes says:

"This time I have uAPSD fixes since I was working on that, hwsim
improvements to make dynamic radios possible for the test suite, the
evidently long-overdue channel_change_time removal and a few other small
collected fix and improvements."

For the iwlwifi bits, Emmanuel says:

"Besides a few trivial patches, I have an important workaround for a HW
issue that has kept me busy for a long time. Along with it, a fix that
prevents an error from being printed.
Eyal fixes our behavior against SISO APs and Ilan fixes an issue with
multiple interface scenarios.
Eliad fixes an error path in our init flow.
We also have a few 'static analyzers' fix."

For the NFC bits, Samuel says:

"It includes:

* A new NFC driver for Marvell's 8897, and a few NCI fixes and
  improvements needed to support this chipset.

* An LLCP fix for how we were setting the default MIU on a p2p link. If
  there is no explicit MIU extension announced at connection time, we
  must use the default one and not the one announced at LLCP link
  establishement time.

* A pn544 EEPROM config update. Some of the currently EEPROM configured
  values are overwriting the firmware ones while other should not be set
  by the driver itself.

* Some NFC digital stack fixes and improvements. Asynchronous functions
  are better documented, RF technologies and CRC functions are set upon
  PSL_REQ reception, and a few minor bugs are fixed.

* Minor and miscelaneous pn533, mei_phy and port100 fixes."

For the ath bits, Kalle says:

"Janusz added Kconfig option for DFS. The DFS code was there already, but
after fixes to mac80211 we can now enable it.

Bartosz added a runtime firmware feature flag to disable P2P. Our 10.1
firmware branch doesn't support P2P and ath10k can now disable that. He
also added a limit for how many clients can connect to ath10k AP.

Michal fixed WEP shared authentication, in case someone still uses it.
And I added firmware debug log to help the firmware engineers."

Along with that is a small batch of ath9k updates and a few other bits
here and there.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Sat, 18 Jan 2014 01:29:36 +0000 (17:29 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull namespace fixes from Eric Biederman:
 "This is a set of 3 regression fixes.

  This fixes /proc/mounts when using "ip netns add <netns>" to display
  the actual mount point.

  This fixes a regression in clone that broke lxc-attach.

  This fixes a regression in the permission checks for mounting /proc
  that made proc unmountable if binfmt_misc was in use.  Oops.

  My apologies for sending this pull request so late.  Al Viro gave
  interesting review comments about the d_path fix that I wanted to
  address in detail before I sent this pull request.  Unfortunately a
  bad round of colds kept from addressing that in detail until today.
  The executive summary of the review was:

  Al: Is patching d_path really sufficient?
      The prepend_path, d_path, d_absolute_path, and __d_path family of
      functions is a really mess.

  Me: Yes, patching d_path is really sufficient.  Yes, the code is mess.
      No it is not appropriate to rewrite all of d_path for a regression
      that has existed for entirely too long already, when a two line
      change will do"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  vfs: Fix a regression in mounting proc
  fork:  Allow CLONE_PARENT after setns(CLONE_NEWPID)
  vfs: In d_path don't call d_dname on a mount point

10 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sat, 18 Jan 2014 00:40:27 +0000 (16:40 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fix from Paolo Bonzini:
 "Fix for a brown paper bag bug.  Thanks to Drew Jones for noticing"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  kvm: x86: fix apic_base enable check

10 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
John W. Linville [Fri, 17 Jan 2014 19:43:17 +0000 (14:43 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-next into for-davem

10 years agoMerge branch 'virtio_rx_merging'
David S. Miller [Fri, 17 Jan 2014 07:46:17 +0000 (23:46 -0800)]
Merge branch 'virtio_rx_merging'

Michael Dalton says:

====================
virtio-net: mergeable rx buffer size auto-tuning

The virtio-net device currently uses aligned MTU-sized mergeable receive
packet buffers. Network throughput for workloads with large average
packet size can be improved by posting larger receive packet buffers.
However, due to SKB truesize effects, posting large (e.g, PAGE_SIZE)
buffers reduces the throughput of workloads that do not benefit from GRO
and have no large inbound packets.

This patchset introduces virtio-net mergeable buffer size auto-tuning,
with buffer sizes ranging from aligned MTU-size to PAGE_SIZE. Packet
buffer size is chosen based on a per-receive queue EWMA of incoming
packet size.

To unify mergeable receive buffer memory allocation and improve
SKB frag coalescing, all mergeable buffer memory allocation is
migrated to per-receive queue page frag allocators.

The per-receive queue mergeable packet buffer size is exported via
sysfs, and the network device sysfs layer has been extended to add
support for device-specific per-receive queue sysfs attribute groups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovirtio-net: initial rx sysfs support, export mergeable rx buffer size
Michael Dalton [Fri, 17 Jan 2014 06:23:30 +0000 (22:23 -0800)]
virtio-net: initial rx sysfs support, export mergeable rx buffer size

Add initial support for per-rx queue sysfs attributes to virtio-net. If
mergeable packet buffers are enabled, adds a read-only mergeable packet
buffer size sysfs attribute for each RX queue.

Suggested-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agolib: Ensure EWMA does not store wrong intermediate values
Michael Dalton [Fri, 17 Jan 2014 06:23:29 +0000 (22:23 -0800)]
lib: Ensure EWMA does not store wrong intermediate values

To ensure ewma_read() without a lock returns a valid but possibly
out of date average, modify ewma_add() by using ACCESS_ONCE to prevent
intermediate wrong values from being written to avg->internal.

Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet-sysfs: add support for device-specific rx queue sysfs attributes
Michael Dalton [Fri, 17 Jan 2014 06:23:28 +0000 (22:23 -0800)]
net-sysfs: add support for device-specific rx queue sysfs attributes

Extend existing support for netdevice receive queue sysfs attributes to
permit a device-specific attribute group. Initial use case for this
support will be to allow the virtio-net device to export per-receive
queue mergeable receive buffer size.

Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovirtio-net: auto-tune mergeable rx buffer size for improved performance
Michael Dalton [Fri, 17 Jan 2014 06:23:27 +0000 (22:23 -0800)]
virtio-net: auto-tune mergeable rx buffer size for improved performance

Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag
allocators") changed the mergeable receive buffer size from PAGE_SIZE to
MTU-size, introducing a single-stream regression for benchmarks with large
average packet size. There is no single optimal buffer size for all
workloads.  For workloads with packet size <= MTU bytes, MTU + virtio-net
header-sized buffers are preferred as larger buffers reduce the TCP window
due to SKB truesize. However, single-stream workloads with large average
packet sizes have higher throughput if larger (e.g., PAGE_SIZE) buffers
are used.

This commit auto-tunes the mergeable receiver buffer packet size by
choosing the packet buffer size based on an EWMA of the recent packet
sizes for the receive queue. Packet buffer sizes range from MTU_SIZE +
virtio-net header len to PAGE_SIZE. This improves throughput for
large packet workloads, as any workload with average packet size >=
PAGE_SIZE will use PAGE_SIZE buffers.

These optimizations interact positively with recent commit
ba275241030c ("virtio-net: coalesce rx frags when possible during rx"),
which coalesces adjacent RX SKB fragments in virtio_net. The coalescing
optimizations benefit buffers of any size.

Benchmarks taken from an average of 5 netperf 30-second TCP_STREAM runs
between two QEMU VMs on a single physical machine. Each VM has two VCPUs
with all offloads & vhost enabled. All VMs and vhost threads run in a
single 4 CPU cgroup cpuset, using cgroups to ensure that other processes
in the system will not be scheduled on the benchmark CPUs. Trunk includes
SKB rx frag coalescing.

net-next w/ virtio_net before 2613af0ed18a (PAGE_SIZE bufs): 14642.85Gb/s
net-next (MTU-size bufs):  13170.01Gb/s
net-next + auto-tune: 14555.94Gb/s

Jason Wang also reported a throughput increase on mlx4 from 22Gb/s
using MTU-sized buffers to about 26Gb/s using auto-tuning.

Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovirtio-net: use per-receive queue page frag alloc for mergeable bufs
Michael Dalton [Fri, 17 Jan 2014 06:23:26 +0000 (22:23 -0800)]
virtio-net: use per-receive queue page frag alloc for mergeable bufs

The virtio-net driver currently uses netdev_alloc_frag() for GFP_ATOMIC
mergeable rx buffer allocations. This commit migrates virtio-net to use
per-receive queue page frags for GFP_ATOMIC allocation. This change unifies
mergeable rx buffer memory allocation, which now will use skb_refill_frag()
for both atomic and GFP-WAIT buffer allocations.

To address fragmentation concerns, if after buffer allocation there
is too little space left in the page frag to allocate a subsequent
buffer, the remaining space is added to the current allocated buffer
so that the remaining space can be used to store packet data.

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: allow > 0 order atomic page alloc in skb_page_frag_refill
Michael Dalton [Fri, 17 Jan 2014 06:23:25 +0000 (22:23 -0800)]
net: allow > 0 order atomic page alloc in skb_page_frag_refill

skb_page_frag_refill currently permits only order-0 page allocs
unless GFP_WAIT is used. Change skb_page_frag_refill to attempt
higher-order page allocations whether or not GFP_WAIT is used. If
memory cannot be allocated, the allocator will fall back to
successively smaller page allocs (down to order-0 page allocs).

This change brings skb_page_frag_refill in line with the existing
page allocation strategy employed by netdev_alloc_frag, which attempts
higher-order page allocations whether or not GFP_WAIT is set, falling
back to successively lower-order page allocations on failure. Part
of migration of virtio-net to per-receive queue page frag allocators.

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Michael Dalton <mwdalton@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: fix error return code in fw_change_attrs()
Wei Yongjun [Fri, 17 Jan 2014 01:53:20 +0000 (09:53 +0800)]
net_sched: fix error return code in fw_change_attrs()

The error code was not set if change indev fail, so the error
condition wasn't reflected in the return value. Fix to return a
negative error code from this error handling case instead of 0.

Fixes: 2519a602c273 ('net_sched: optimize tcf_match_indev()')
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'tipc'
David S. Miller [Fri, 17 Jan 2014 03:11:22 +0000 (19:11 -0800)]
Merge branch 'tipc'

Ying Xue says:

====================
tipc: align TIPC behaviours of waiting for events with other stacks

Comparing the current implementations of waiting for events in TIPC
socket layer with other stacks, TIPC's behaviour is very different
because wait_event_interruptible_timeout()/wait_event_interruptible()
are always used by TIPC to wait for events while relevant socket or
port variables are fed to them as their arguments. As socket lock has
to be released temporarily before the two routines of waiting for
events are called, their arguments associated with socket or port
structures are out of socket lock protection. This might cause
serious issues where the process of calling socket syscall such as
sendsmg(), connect(), accept(), and recvmsg(), cannot be waken up
at all even if proper event arrives or improperly be woken up
although the condition of waking up the process is not satisfied
in practice.

Therefore, aligning its behaviours with similar functions implemented
in other stacks, for instance, sk_stream_wait_connect() and
inet_csk_wait_for_connect() etc, can avoid above risks for us.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: standardize recvmsg routine
Ying Xue [Fri, 17 Jan 2014 01:50:07 +0000 (09:50 +0800)]
tipc: standardize recvmsg routine

Standardize the behaviour of waiting for events in TIPC recvmsg()
so that all variables of socket or port structures are protected
within socket lock, allowing the process of calling recvmsg() to
be woken up at appropriate time.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: standardize sendmsg routine of connected socket
Ying Xue [Fri, 17 Jan 2014 01:50:06 +0000 (09:50 +0800)]
tipc: standardize sendmsg routine of connected socket

Standardize the behaviour of waiting for events in TIPC send_packet()
so that all variables of socket or port structures are protected within
socket lock, allowing the process of calling sendmsg() to be woken up
at appropriate time.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: standardize sendmsg routine of connectionless socket
Ying Xue [Fri, 17 Jan 2014 01:50:05 +0000 (09:50 +0800)]
tipc: standardize sendmsg routine of connectionless socket

Comparing the behaviour of how to wait for events in TIPC sendmsg()
with other stacks, the TIPC implementation might be perceived as
different, and sometimes even incorrect. For instance, sk_sleep()
and tport->congested variables associated with socket are exposed
without socket lock protection while wait_event_interruptible_timeout()
accesses them. So standardizing it with similar implementation
in other stacks can help us correct these errors which the process
of calling sendmsg() cannot be woken up event if an expected event
arrive at socket or improperly woken up although the wake condition
doesn't match.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: standardize accept routine
Ying Xue [Fri, 17 Jan 2014 01:50:04 +0000 (09:50 +0800)]
tipc: standardize accept routine

Comparing the behaviour of how to wait for events in TIPC accept()
with other stacks, the TIPC implementation might be perceived as
different, and sometimes even incorrect. As sk_sleep() and
sk->sk_receive_queue variables associated with socket are not
protected by socket lock, the process of calling accept() may be
woken up improperly or sometimes cannot be woken up at all. After
standardizing it with inet_csk_wait_for_connect routine, we can
get benefits including: avoiding 'thundering herd' phenomenon,
adding a timeout mechanism for accept(), coping with a pending
signal, and having sk_sleep() and sk->sk_receive_queue being
always protected within socket lock scope and so on.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: standardize connect routine
Ying Xue [Fri, 17 Jan 2014 01:50:03 +0000 (09:50 +0800)]
tipc: standardize connect routine

Comparing the behaviour of how to wait for events in TIPC connect()
with other stacks, the TIPC implementation might be perceived as
different, and sometimes even incorrect. For instance, as both
sock->state and sk_sleep() are directly fed to
wait_event_interruptible_timeout() as its arguments, and socket lock
has to be released before we call wait_event_interruptible_timeout(),
the two variables associated with socket are exposed out of socket
lock protection, thereby probably getting stale values so that the
process of calling connect() cannot be woken up exactly even if
correct event arrives or it is woken up improperly even if the wake
condition is not satisfied in practice. Therefore, standardizing its
behaviour with sk_stream_wait_connect routine can avoid these risks.

Additionally the implementation of connect routine is simplified as a
whole, allowing it to return correct values in all different cases.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove the unnecessary assignment
wangweidong [Thu, 16 Jan 2014 08:25:19 +0000 (16:25 +0800)]
sctp: remove the unnecessary assignment

When go the right path, the status is 0, no need to assign it again.
So just remove the assignment.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>