Similar to fsl_pq_mdio.c, this driver is for the 10G MDIO controller on
Freescale Frame Manager Ethernet controllers.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The network classifier cgroup initalizes each cgroups instance classid value to
0.  However, the sock_update_classid function only updates classid's in sockets
if the tasks cgroup classid is not zero, and if it differs from the current
classid.  The later check is to prevent cache line dirtying, but the former is
detrimental, as it prevents resetting a classid for a cgroup to 0.  While this
is not a common action, it has administrative usefulness (if the admin wants to
disable classification of a certain group temporarily for instance).

Easy fix, just remove the zero check.  Tested successfully by myself
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Antonio Quartulli says:

====================
Included changes:
- a set of codestyle rearrangements/fixes
- new feature to early detect new joining (mesh-unaware) clients
- a minor fix for the gw-feature
- substitution of shift operations with the BIT() macro
- reorganization of the main batman-adv structure (struct batadv_priv)
- some more (very) minor cleanups and fixes
===================
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fixes a broken build due to a missing header:
...
  CC      net/ipv4/proc.o
In file included from include/net/net_namespace.h:15,
                 from net/ipv4/proc.c:35:
include/net/netns/packet.h:11: error: field 'sklist_lock' has incomplete type
...

The lock of netns_packet has been replaced by a recent patch to be a mutex instead of a spinlock,
but we need to replace the header file to be linux/mutex.h instead of linux/spinlock.h as well.

See commit 0fa7fa98:
packet: Protect packet sk list with mutex (v2) patch,
Signed-off-by: Rami Rosen <rosenr@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Biederman pointed out that not holding RTNL while calling
call_netdevice_notifiers() was racy.

This patch is a direct transcription his feedback
against commit 0115e8e3 (net: remove delay at device dismantle)

Thanks Eric !
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Gao feng <gaofeng@cn.fujitsu.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

In order to understand  where a broadcast packet is coming from and use
this information to detect not yet announced clients, this patch modifies the
interface_rx() function by passing a new argument: the orig node
corresponding to the node that originated the received packet (if known).
This new argument if not NULL for broadcast packets only (other packets does not
have source field).
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

With the current TT mechanism a new client joining the network is not
immediately able to communicate with other hosts because its MAC address has not
been announced yet. This situation holds until the first OGM containing its
joining event will be spread over the mesh network.

This behaviour can be acceptable in networks where the originator interval is a
small value (e.g. 1sec) but if that value is set to an higher time (e.g. 5secs)
the client could suffer from several malfunctions like DHCP client timeouts,
etc.

This patch adds an early detection mechanism that makes nodes in the network
able to recognise "not yet announced clients" by means of the broadcast packets
they emitted on connection (e.g. ARP or DHCP request). The added client will
then be confirmed upon receiving the OGM claiming it or purged if such OGM
is not received within a fixed amount of time.
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

When enabling promiscuous mode, tt queries for other hosts might be
received. Before this patch, "foreign" tt queries were processed like
any other query and thus forwarded to its destination again and thereby
causing a loop.

This patch adds a check to drop foreign tt queries.
Signed-off-by: Martin Hundebøll <martin@hundeboll.net>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

batadv_check_unicast_packet() is needed in batadv_recv_tt_query(), so
move the former to before the latter.
Signed-off-by: Martin Hundebøll <martin@hundeboll.net>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

The structure batadv_priv grows everytime a new feature is introduced. It gets
hard to find the parts of the struct that belongs to a specific feature. This
becomes even harder by the fact that not every feature uses a prefix in the
member name.

The variables for bridge loop avoidence, gateway handling, translation table
and visualization server are moved into separate structs that are included in
the bat_priv main struct.
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

If this call fails, some of the orig_nodes spaces may have been
resized for the increased number of interface, and some may not.
If we would just continue with the larger number of interfaces,
this would lead to access to not allocated memory later.

We better check the return code, and don't add the interface if
no memory is available. OTOH, keeping some of the orig_nodes
with too much memory allocated should hurt no one (except for
a few too many bytes allocated).
Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

the word millisecond is misspelled in several comments. This patch fixes it.
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

The batadv_tt_orig_list_entry structure didn't have any refcounting mechanism so
far. This patch introduces it and makes the structure being usable in much more
complex context.
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

As much as I'm happy to see LWN links sprinkled through the kernel by the
dozen, this one in particular reflects a very old state of reality; the
associated comment is now incorrect.  So just delete it.
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Marek Lindner <lindner_marek@yahoo.de>
Acked-by: Martin Hundebøll <martin@hundeboll.net>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

for consistency reasons within the code and with the documentation,
we should always call it "claim" and "unclaim".
Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

This is especially useful if there are no claims yet, but we still want
to know which gateways are using bridge loop avoidance in the network.
Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Signed-off-by: Antonio Quartulli <ordex@autistici.org>

Change since v1:

* Fixed inuse counters access spotted by Eric

In patch eea68e2f (packet: Report socket mclist info via diag module) I've
introduced a "scheduling in atomic" problem in packet diag module -- the
socket list is traversed under rcu_read_lock() while performed under it sk
mclist access requires rtnl lock (i.e. -- mutex) to be taken.

[152363.820563] BUG: scheduling while atomic: crtools/12517/0x10000002
[152363.820573] 4 locks held by crtools/12517:
[152363.820581]  #0:  (sock_diag_mutex){+.+.+.}, at: [<ffffffff81a2dcb5>] sock_diag_rcv+0x1f/0x3e
[152363.820613]  #1:  (sock_diag_table_mutex){+.+.+.}, at: [<ffffffff81a2de70>] sock_diag_rcv_msg+0xdb/0x11a
[152363.820644]  #2:  (nlk->cb_mutex){+.+.+.}, at: [<ffffffff81a67d01>] netlink_dump+0x23/0x1ab
[152363.820693]  #3:  (rcu_read_lock){.+.+..}, at: [<ffffffff81b6a049>] packet_diag_dump+0x0/0x1af

Similar thing was then re-introduced by further packet diag patches (fanount
mutex and pgvec mutex for rings) :(

Apart from being terribly sorry for the above, I propose to change the packet
sk list protection from spinlock to mutex. This lock currently protects two
modifications:

* sklist
* prot inuse counters

The sklist modifications can be just reprotected with mutex since they already
occur in a sleeping context. The inuse counters modifications are trickier -- the
__this_cpu_-s are used inside, thus requiring the caller to handle the potential
issues with contexts himself. Since packet sockets' counters are modified in two
places only (packet_create and packet_release) we only need to protect the context
from being preempted. BH disabling is not required in this case.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The helper functions which translate IEEE MDIO Manageable Device (MMD)
Energy-Efficient Ethernet (EEE) registers 3.20, 7.60 and 7.61 to and from
the comparable ethtool supported/advertised settings will be needed by
drivers other than those in PHYLIB (e.g. e1000e in a follow-on patch).

In the same fashion as similar translation functions in linux/mii.h, move
these functions from the PHYLIB core to the linux/mdio.h header file so the
code will not have to be duplicated in each driver needing MMD-to-ethtool
(and vice-versa) translations.  The function and some variable names have
been renamed to be more descriptive.

Not tested on the only hardware that currently calls the related functions,
stmmac, because I don't have access to any.  Has been compile tested and
the translations have been tested on a locally modified version of e1000e.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Instead of using a hard-coded value for the status variable, it would make
the code more readable to use its destined define from linux/if_packet.h.

Signed-off-by: daniel.borkmann@tik.ee.ethz.ch
Signed-off-by: David S. Miller <davem@davemloft.net>

Since we have already in BH context when *_write_space(),
*_data_ready() as well as *_state_change() are called, it's
unnecessary to disable BH.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently the "bonding" driver does not support load balancing outgoing
traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4)
are currently supported; this patch adds transmit hashing for IPv6 (and
TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the
bonding driver. In addition, bounds checking has been added to all
transmit hashing functions.

The algorithm chosen (xor'ing the bottom three quads of the source and
destination addresses together, then xor'ing each byte of that result into
the bottom byte, finally xor'ing with the last bytes of the MAC addresses)
was selected after testing almost 400,000 unique IPv6 addresses harvested
from server logs. This algorithm had the most even distribution for both
big- and little-endian architectures while still using few instructions. Its
behavior also attempts to closely match that of the IPv4 algorithm.

The IPv6 flow label was intentionally not included in the hash as it appears
to be unset in the vast majority of IPv6 traffic sampled, and the current
algorithm not using the flow label already offers a very even distribution.

Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets,
ie, they are not balanced based on layer 4 information. Additionally,
IPv6 packets with intermediate headers are not balanced based on layer
4 information. In practice these intermediate headers are not common and
this should not cause any problems, and the alternative (a packet-parsing
loop and look-up table) seemed slow and complicated for little gain.
Tested-by: John Eaglesham <linux@8192.net>
Signed-off-by: John Eaglesham <linux@8192.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6gre_err() miscomputes grehlen (sizeof(ipv6h) is 4 or 8,
not 40 as expected), and should take into account 'offset' parameter.

Also uses pskb_may_pull() to cope with some fragged skbs
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dmitry Kozlov <xeb@mail.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch reverts commit 56892261 (xfrm: Use rcu_dereference_bh to
deference pointer protected by rcu_read_lock_bh), and fixes bugs
introduced in commit 418a99ac ( Replace rwlock on xfrm_policy_afinfo
with rcu )

1) We properly use RCU variant in this file, not a mix of RCU/RCU_BH

2) We must defer some writes after the synchronize_rcu() call or a reader
 can crash dereferencing NULL pointer.

3) Now we use the xfrm_policy_afinfo_lock spinlock only from process
context, we no longer need to block BH in xfrm_policy_register_afinfo()
and xfrm_policy_unregister_afinfo()

4) Can use RCU_INIT_POINTER() instead of rcu_assign_pointer() in
xfrm_policy_unregister_afinfo()

5) Remove a forward inline declaration (xfrm_policy_put_afinfo()),
  and also move xfrm_policy_get_afinfo() declaration.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Fan Du <fan.du@windriver.com>
Cc: Priyanka Jain <Priyanka.Jain@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

I noticed extra one second delay in device dismantle, tracked down to
a call to dst_dev_event() while some call_rcu() are still in RCU queues.

These call_rcu() were posted by rt_free(struct rtable *rt) calls.

We then wait a little (but one second) in netdev_wait_allrefs() before
kicking again NETDEV_UNREGISTER.

As the call_rcu() are now completed, dst_dev_event() can do the needed
device swap on busy dst.

To solve this problem, add a new NETDEV_UNREGISTER_FINAL, called
after a rcu_barrier(), but outside of RTNL lock.

Use NETDEV_UNREGISTER_FINAL with care !

Change dst_dev_event() handler to react to NETDEV_UNREGISTER_FINAL

Also remove NETDEV_UNREGISTER_BATCH, as its not used anymore after
IP cache removal.

With help from Gao feng
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
This is the first batch of Netfilter and IPVS updates for your
net-next tree. Mostly cleanups for the Netfilter side. They are:

* Remove unnecessary RTNL locking now that we have support
  for namespace in nf_conntrack, from Patrick McHardy.

* Cleanup to eliminate unnecessary goto in the initialization
  path of several Netfilter tables, from Jean Sacren.

* Another cleanup from Wu Fengguang, this time to PTR_RET instead
  of if IS_ERR then return PTR_ERR.

* Use list_for_each_entry_continue_rcu in nf_iterate, from
  Michael Wang.

* Add pmtu_disc sysctl option to disable PMTU in their tunneling
  transmitter, from Julian Anastasov.

* Generalize application protocol registration in IPVS and modify
  IPVS FTP helper to use it, from Julian Anastasov.

* update Kconfig. The IPVS FTP helper depends on the Netfilter FTP
  helper for NAT support, from Julian Anastasov.

* Add logic to update PMTU for IPIP packets in IPVS, again
  from Julian Anastasov.

* A couple of sparse warning fixes for IPVS and Netfilter from
  Claudiu Ghioc and Patrick McHardy respectively.

Patrick's IPv6 NAT changes will follow after this batch, I need
to flush this batch first before refreshing my tree.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
This series contains updates to ethtool.h, e1000, e1000e, and igb to
implement MDI/MDIx control.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Usually it's a good practice to use goto statement for error recovery
when initializing the module. This approach could be an overkill if:

 1) there is only one fail case;
 2) success and failure use the same return statement.

For a cleaner approach, remove the unnecessary goto statement and
directly implement error recovery.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This patch replaces list_for_each_continue_rcu() with
list_for_each_entry_continue_rcu() to allow removing
list_for_each_continue_rcu().
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Merge fixes from Andrew Morton.

Random drivers and some VM fixes.

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (17 commits)
  mm: compaction: Abort async compaction if locks are contended or taking too long
  mm: have order > 0 compaction start near a pageblock with free pages
  rapidio/tsi721: fix unused variable compiler warning
  rapidio/tsi721: fix inbound doorbell interrupt handling
  drivers/rtc/rtc-rs5c348.c: fix hour decoding in 12-hour mode
  mm: correct page->pfmemalloc to fix deactivate_slab regression
  drivers/rtc/rtc-pcf2123.c: initialize dynamic sysfs attributes
  mm/compaction.c: fix deferring compaction mistake
  drivers/misc/sgi-xp/xpc_uv.c: SGI XPC fails to load when cpu 0 is out of IRQ resources
  string: do not export memweight() to userspace
  hugetlb: update hugetlbpage.txt
  checkpatch: add control statement test to SINGLE_STATEMENT_DO_WHILE_MACRO
  mm: hugetlbfs: correctly populate shared pmd
  cciss: fix incorrect scsi status reporting
  Documentation: update mount option in filesystem/vfat.txt
  mm: change nr_ptes BUG_ON to WARN_ON
  cs5535-clockevt: typo, it's MFGPT, not MFPGT

Pull media fixes from Mauro Carvalho Chehab:
 "For bug fixes, at soc_camera, si470x, uvcvideo, iguanaworks IR driver,
  radio_shark Kbuild fixes, and at the V4L2 core (radio fixes)."

* 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
  [media] media: soc_camera: don't clear pix->sizeimage in JPEG mode
  [media] media: mx2_camera: Fix clock handling for i.MX27
  [media] video: mx2_camera: Use clk_prepare_enable/clk_disable_unprepare
  [media] video: mx1_camera: Use clk_prepare_enable/clk_disable_unprepare
  [media] media: mx3_camera: buf_init() add buffer state check
  [media] radio-shark2: Only compile led support when CONFIG_LED_CLASS is set
  [media] radio-shark: Only compile led support when CONFIG_LED_CLASS is set
  [media] radio-shark*: Call cancel_work_sync from disconnect rather then release
  [media] radio-shark*: Remove work-around for dangling pointer in usb intfdata
  [media] Add USB dependency for IguanaWorks USB IR Transceiver
  [media] Add missing logging for rangelow/high of hwseek
  [media] VIDIOC_ENUM_FREQ_BANDS fix
  [media] mem2mem_testdev: fix querycap regression
  [media] si470x: v4l2-compliance fixes
  [media] DocBook: Remove a spurious character
  [media] uvcvideo: Reset the bytesused field when recycling an erroneous buffer

Pull networking update from David Miller:
 "A couple weeks of bug fixing in there.  The largest chunk is all the
  broken crap Amerigo Wang found in the netpoll layer."

 1) netpoll and it's users has several serious bugs:
    a) uses GFP_KERNEL with locks held
    b) interfaces requiring interrupts disabled are called with them
       enabled
    c) and vice versa
    d) VLAN tag demuxing, as per all other RX packet input paths, is not
       applied

    All from Amerigo Wang.

 2) Hopefully cure the ipv4 mapped ipv6 address TCP early demux bugs for
    good, from Neal Cardwell.

 3) Unlike AF_UNIX, AF_PACKET sockets don't set a default credentials
    when the user doesn't specify one explicitly during sendmsg().
    Instead we attach an empty (zero) SCM credential block which is
    definitely not what we want.  Fix from Eric Dumazet.

 4) IPv6 illegally invokes netdevice notifiers with RCU lock held, fix
    from Ben Hutchings.

 5) inet_csk_route_child_sock() checks wrong inet options pointer, fix
    from Christoph Paasch.

 6) When AF_PACKET is used for transmit, packet loopback doesn't behave
    properly when a socket fanout is enabled, from Eric Leblond.

 7) On bluetooth l2cap channel create failure, we leak the socket, from
    Jaganath Kanakkassery.

 8) Fix all the netprio file handling bugs found by Al Viro, from John
    Fastabend.

 9) Several error return and NULL deref bug fixes in networking drivers
    from Julia Lawall.

10) A large smattering of struct padding et al.  kernel memory leaks to
    userspace found of Mathias Krause.

11) Conntrack expections in netfilter can access an uninitialized timer,
    fix from Pablo Neira Ayuso.

12) Several netfilter SIP tracker bug fixes from Patrick McHardy.

13) IPSEC ipv6 routes are not initialized correctly all the time,
    resulting in an OOPS in inet_putpeer().  Also from Patrick McHardy.

14) Bridging does rcu_dereference() outside of RCU protected area, from
    Stephen Hemminger.

15) Fix routing cache removal performance regression when looking up
    output routes that have a local destination.  From Zheng Yan.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (87 commits)
  af_netlink: force credentials passing [CVE-2012-3520]
  ipv4: fix ip header ident selection in __ip_make_skb()
  ipv4: Use newinet->inet_opt in inet_csk_route_child_sock()
  tcp: fix possible socket refcount problem
  net: tcp: move sk_rx_dst_set call after tcp_create_openreq_child()
  net/core/dev.c: fix kernel-doc warning
  netconsole: remove a redundant netconsole_target_put()
  net: ipv6: fix oops in inet_putpeer()
  net/stmmac: fix issue of clk_get for Loongson1B.
  caif: Do not dereference NULL in chnl_recv_cb()
  af_packet: don't emit packet on orig fanout group
  drivers/net/irda: fix error return code
  drivers/net/wan/dscc4.c: fix error return code
  drivers/net/wimax/i2400m/fw.c: fix error return code
  smsc75xx: add missing entry to MAINTAINERS
  net: qmi_wwan: new devices: UML290 and K5006-Z
  net: sh_eth: Add eth support for R8A7779 device
  netdev/phy: skip disabled mdio-mux nodes
  dt: introduce for_each_available_child_of_node, of_get_next_available_child
  net: netprio: fix cgrp create and write priomap race
  ...

Jim Schutt reported a problem that pointed at compaction contending
heavily on locks.  The workload is straight-forward and in his own words;

	The systems in question have 24 SAS drives spread across 3 HBAs,
	running 24 Ceph OSD instances, one per drive.  FWIW these servers
	are dual-socket Intel 5675 Xeons w/48 GB memory.  I've got ~160
	Ceph Linux clients doing dd simultaneously to a Ceph file system
	backed by 12 of these servers.

Early in the test everything looks fine

  procs -------------------memory------------------ ---swap-- -----io---- --system-- -----cpu-------
   r  b       swpd       free       buff      cache   si   so    bi    bo   in   cs  us sy  id wa st
  31 15          0     287216        576   38606628    0    0     2  1158    2   14   1  3  95  0  0
  27 15          0     225288        576   38583384    0    0    18 2222016 203357 134876  11 56  17 15  0
  28 17          0     219256        576   38544736    0    0    11 2305932 203141 146296  11 49  23 17  0
   6 18          0     215596        576   38552872    0    0     7 2363207 215264 166502  12 45  22 20  0
  22 18          0     226984        576   38596404    0    0     3 2445741 223114 179527  12 43  23 22  0

and then it goes to pot

  procs -------------------memory------------------ ---swap-- -----io---- --system-- -----cpu-------
   r  b       swpd       free       buff      cache   si   so    bi    bo   in   cs  us sy  id wa st
  163  8          0     464308        576   36791368    0    0    11 22210  866  536   3 13  79  4  0
  207 14          0     917752        576   36181928    0    0   712 1345376 134598 47367   7 90   1  2  0
  123 12          0     685516        576   36296148    0    0   429 1386615 158494 60077   8 84   5  3  0
  123 12          0     598572        576   36333728    0    0  1107 1233281 147542 62351   7 84   5  4  0
  622  7          0     660768        576   36118264    0    0   557 1345548 151394 59353   7 85   4  3  0
  223 11          0     283960        576   36463868    0    0    46 1107160 121846 33006   6 93   1  1  0

Note that system CPU usage is very high blocks being written out has
dropped by 42%. He analysed this with perf and found

  perf record -g -a sleep 10
  perf report --sort symbol --call-graph fractal,5
    34.63%  [k] _raw_spin_lock_irqsave
            |
            |--97.30%-- isolate_freepages
            |          compaction_alloc
            |          unmap_and_move
            |          migrate_pages
            |          compact_zone
            |          compact_zone_order
            |          try_to_compact_pages
            |          __alloc_pages_direct_compact
            |          __alloc_pages_slowpath
            |          __alloc_pages_nodemask
            |          alloc_pages_vma
            |          do_huge_pmd_anonymous_page
            |          handle_mm_fault
            |          do_page_fault
            |          page_fault
            |          |
            |          |--87.39%-- skb_copy_datagram_iovec
            |          |          tcp_recvmsg
            |          |          inet_recvmsg
            |          |          sock_recvmsg
            |          |          sys_recvfrom
            |          |          system_call
            |          |          __recv
            |          |          |
            |          |           --100.00%-- (nil)
            |          |
            |           --12.61%-- memcpy
             --2.70%-- [...]

There was other data but primarily it is all showing that compaction is
contended heavily on the zone->lock and zone->lru_lock.

commit [b2eef8c0: mm: compaction: minimise the time IRQs are disabled
while isolating pages for migration] noted that it was possible for
migration to hold the lru_lock for an excessive amount of time. Very
broadly speaking this patch expands the concept.

This patch introduces compact_checklock_irqsave() to check if a lock
is contended or the process needs to be scheduled. If either condition
is true then async compaction is aborted and the caller is informed.
The page allocator will fail a THP allocation if compaction failed due
to contention. This patch also introduces compact_trylock_irqsave()
which will acquire the lock only if it is not contended and the process
does not need to schedule.
Reported-by: Jim Schutt <jaschut@sandia.gov>
Tested-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>