netvm: allow the use of __GFP_MEMALLOC by specific sockets

Allow specific sockets to be tagged SOCK_MEMALLOC and use __GFP_MEMALLOC
for their allocations.  These sockets will be able to go below watermarks
and allocate from the emergency reserve.  Such sockets are to be used to
service the VM (iow.  to swap over).  They must be handled kernel side,
exposing such a socket to user-space is a bug.

There is a risk that the reserves be depleted so for now, the
administrator is responsible for increasing min_free_kbytes as necessary
to prevent deadlock for their workloads.

[a.p.zijlstra@chello.nl: Original patches]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

net: introduce sk_gfp_atomic() to allow addition of GFP flags depending on the individual socket

Introduce sk_gfp_atomic(), this function allows to inject sock specific
flags to each sock related allocation.  It is only used on allocation
paths that may be required for writing pages back to network storage.

[davem@davemloft.net: Use sk_gfp_atomic only when necessary]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: ignore mempolicies when using ALLOC_NO_WATERMARK

The reserve is proportionally distributed over all !highmem zones in the
system.  So we need to allow an emergency allocation access to all zones.
In order to do that we need to break out of any mempolicy boundaries we
might have.

In my opinion that does not break mempolicies as those are user oriented
and not system oriented.  That is, system allocations are not guaranteed
to be within mempolicy boundaries.  For instance IRQs do not even have a
mempolicy.

So breaking out of mempolicy boundaries for 'rare' emergency allocations,
which are always system allocations (as opposed to user) is ok.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: only set page->pfmemalloc when ALLOC_NO_WATERMARKS was used

__alloc_pages_slowpath() is called when the number of free pages is below
the low watermark.  If the caller is entitled to use ALLOC_NO_WATERMARKS
then the page will be marked page->pfmemalloc.  This protects more pages
than are strictly necessary as we only need to protect pages allocated
below the min watermark (the pfmemalloc reserves).

This patch only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was
required to allocate the page.

[rientjes@google.com: David noticed the problem during review]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: allow PF_MEMALLOC from softirq context

This is needed to allow network softirq packet processing to make use of
PF_MEMALLOC.

Currently softirq context cannot use PF_MEMALLOC due to it not being
associated with a task, and therefore not having task flags to fiddle with
- thus the gfp to alloc flag mapping ignores the task flags when in
interrupts (hard or soft) context.

Allowing softirqs to make use of PF_MEMALLOC therefore requires some
trickery.  This patch borrows the task flags from whatever process happens
to be preempted by the softirq.  It then modifies the gfp to alloc flags
mapping to not exclude task flags in softirq context, and modify the
softirq code to save, clear and restore the PF_MEMALLOC flag.

The save and clear, ensures the preempted task's PF_MEMALLOC flag doesn't
leak into the softirq.  The restore ensures a softirq's PF_MEMALLOC flag
cannot leak back into the preempted process.  This should be safe due to
the following reasons

Softirqs can run on multiple CPUs sure but the same task should not be
executing the same softirq code. Neither should the softirq
handler be preempted by any other softirq handler so the flags
should not leak to an unrelated softirq.

Softirqs re-enable hardware interrupts in __do_softirq() so can be
preempted by hardware interrupts so PF_MEMALLOC is inherited
by the hard IRQ. However, this is similar to a process in
reclaim being preempted by a hardirq. While PF_MEMALLOC is
set, gfp_to_alloc_flags() distinguishes between hard and
soft irqs and avoids giving a hardirq the ALLOC_NO_WATERMARKS
flag.

If the softirq is deferred to ksoftirq then its flags may be used
        instead of a normal tasks but as the softirq cannot be preempted,
        the PF_MEMALLOC flag does not leak to other code by accident.

[davem@davemloft.net: Document why PF_MEMALLOC is safe]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce __GFP_MEMALLOC to allow access to emergency reserves

__GFP_MEMALLOC will allow the allocation to disregard the watermarks, much
like PF_MEMALLOC.  It allows one to pass along the memalloc state in
object related allocation flags as opposed to task related flags, such as
sk->sk_allocation.  This removes the need for ALLOC_PFMEMALLOC as callers
using __GFP_MEMALLOC can get the ALLOC_NO_WATERMARK flag which is now
enough to identify allocations related to page reclaim.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: slub: optimise the SLUB fast path to avoid pfmemalloc checks

This patch removes the check for pfmemalloc from the alloc hotpath and
puts the logic after the election of a new per cpu slab.  For a pfmemalloc
page we do not use the fast path but force the use of the slow path which
is also used for the debug case.

This has the side-effect of weakening pfmemalloc processing in the
following way;

1. A process that is allocating for network swap calls __slab_alloc.
   pfmemalloc_match is true so the freelist is loaded and c->freelist is
   now pointing to a pfmemalloc page.

2. A process that is attempting normal allocations calls slab_alloc,
   finds the pfmemalloc page on the freelist and uses it because it did
   not check pfmemalloc_match()

The patch allows non-pfmemalloc allocations to use pfmemalloc pages with
the kmalloc slabs being the most vunerable caches on the grounds they
are most likely to have a mix of pfmemalloc and !pfmemalloc requests. A
later patch will still protect the system as processes will get throttled
if the pfmemalloc reserves get depleted but performance will not degrade
as smoothly.

[mgorman@suse.de: Expanded changelog]
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages

When a user or administrator requires swap for their application, they
create a swap partition and file, format it with mkswap and activate it
with swapon.  Swap over the network is considered as an option in diskless
systems.  The two likely scenarios are when blade servers are used as part
of a cluster where the form factor or maintenance costs do not allow the
use of disks and thin clients.

The Linux Terminal Server Project recommends the use of the Network Block
Device (NBD) for swap according to the manual at
https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download
There is also documentation and tutorials on how to setup swap over NBD at
places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The
nbd-client also documents the use of NBD as swap.  Despite this, the fact
is that a machine using NBD for swap can deadlock within minutes if swap
is used intensively.  This patch series addresses the problem.

The core issue is that network block devices do not use mempools like
normal block devices do.  As the host cannot control where they receive
packets from, they cannot reliably work out in advance how much memory
they might need.  Some years ago, Peter Zijlstra developed a series of
patches that supported swap over an NFS that at least one distribution is
carrying within their kernels.  This patch series borrows very heavily
from Peter's work to support swapping over NBD as a pre-requisite to
supporting swap-over-NFS.  The bulk of the complexity is concerned with
preserving memory that is allocated from the PFMEMALLOC reserves for use
by the network layer which is needed for both NBD and NFS.

Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to
preserve access to pages allocated under low memory situations
to callers that are freeing memory.

Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks

Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC
reserves without setting PFMEMALLOC.

Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves
for later use by network packet processing.

Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required

Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set.

Patches 7-12 allows network processing to use PFMEMALLOC reserves when
the socket has been marked as being used by the VM to clean pages. If
packets are received and stored in pages that were allocated under
low-memory situations and are unrelated to the VM, the packets
are dropped.

Patch 11 reintroduces __skb_alloc_page which the networking
folk may object to but is needed in some cases to propogate
pfmemalloc from a newly allocated page to an skb. If there is a
strong objection, this patch can be dropped with the impact being
that swap-over-network will be slower in some cases but it should
not fail.

Patch 13 is a micro-optimisation to avoid a function call in the
common case.

Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use
PFMEMALLOC if necessary.

Patch 15 notes that it is still possible for the PFMEMALLOC reserve
to be depleted. To prevent this, direct reclaimers get throttled on
a waitqueue if 50% of the PFMEMALLOC reserves are depleted.  It is
expected that kswapd and the direct reclaimers already running
will clean enough pages for the low watermark to be reached and
the throttled processes are woken up.

Patch 16 adds a statistic to track how often processes get throttled

Some basic performance testing was run using kernel builds, netperf on
loopback for UDP and TCP, hackbench (pipes and sockets), iozone and
sysbench.  Each of them were expected to use the sl*b allocators
reasonably heavily but there did not appear to be significant performance
variances.

For testing swap-over-NBD, a machine was booted with 2G of RAM with a
swapfile backed by NBD.  8*NUM_CPU processes were started that create
anonymous memory mappings and read them linearly in a loop.  The total
size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under
memory pressure.

Without the patches and using SLUB, the machine locks up within minutes
and runs to completion with them applied.  With SLAB, the story is
different as an unpatched kernel run to completion.  However, the patched
kernel completed the test 45% faster.

MICRO
                                         3.5.0-rc2 3.5.0-rc2
 vanilla     swapnbd
Unrecognised test vmscan-anon-mmap-write
MMTests Statistics: duration
Sys Time Running Test (seconds)             197.80    173.07
User+Sys Time Running Test (seconds)        206.96    182.03
Total Elapsed Time (seconds)               3240.70   1762.09

This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages

Allocations of pages below the min watermark run a risk of the machine
hanging due to a lack of memory.  To prevent this, only callers who have
PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are
allowed to allocate with ALLOC_NO_WATERMARKS.  Once they are allocated to
a slab though, nothing prevents other callers consuming free objects
within those slabs.  This patch limits access to slab pages that were
alloced from the PFMEMALLOC reserves.

When this patch is applied, pages allocated from below the low watermark
are returned with page->pfmemalloc set and it is up to the caller to
determine how the page should be protected.  SLAB restricts access to any
page with page->pfmemalloc set to callers which are known to able to
access the PFMEMALLOC reserve.  If one is not available, an attempt is
made to allocate a new page rather than use a reserve.  SLUB is a bit more
relaxed in that it only records if the current per-CPU page was allocated
from PFMEMALLOC reserve and uses another partial slab if the caller does
not have the necessary GFP or process flags.  This was found to be
sufficient in tests to avoid hangs due to SLUB generally maintaining
smaller lists than SLAB.

In low-memory conditions it does mean that !PFMEMALLOC allocators can fail
a slab allocation even though free objects are available because they are
being preserved for callers that are freeing pages.

[a.p.zijlstra@chello.nl: Original implementation]
[sebastian@breakpoint.cc: Correct order of page flag clearing]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: David Miller <davem@davemloft.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: Eric B Munson <emunson@mgebm.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: fix kswapd looping forever problem

When hotplug offlining happens on zone A, it starts to mark freed page as
MIGRATE_ISOLATE type in buddy for preventing further allocation.
(MIGRATE_ISOLATE is very irony type because it's apparently on buddy but
we can't allocate them).

When the memory shortage happens during hotplug offlining, current task
starts to reclaim, then wake up kswapd.  Kswapd checks watermark, then go
sleep because current zone_watermark_ok_safe doesn't consider
MIGRATE_ISOLATE freed page count.  Current task continue to reclaim in
direct reclaim path without kswapd's helping.  The problem is that
zone->all_unreclaimable is set by only kswapd so that current task would
be looping forever like below.

__alloc_pages_slowpath
restart:
wake_all_kswapd
rebalance:
__alloc_pages_direct_reclaim
do_try_to_free_pages
if global_reclaim && !all_unreclaimable
return 1; /* It means we did did_some_progress */
skip __alloc_pages_may_oom
should_alloc_retry
goto rebalance;

If we apply KOSAKI's patch[1] which doesn't depends on kswapd about
setting zone->all_unreclaimable, we can solve this problem by killing some
task in direct reclaim path.  But it doesn't wake up kswapd, still.  It
could be a problem still if other subsystem needs GFP_ATOMIC request.  So
kswapd should consider MIGRATE_ISOLATE when it calculate free pages BEFORE
going sleep.

This patch counts the number of MIGRATE_ISOLATE page block and
zone_watermark_ok_safe will consider it if the system has such blocks
(fortunately, it's very rare so no problem in POV overhead and kswapd is
never hotpath).

Copy/modify from Mel's quote
"
Ideal solution would be "allocating" the pageblock.
It would keep the free space accounting as it is but historically,
memory hotplug didn't allocate pages because it would be difficult to
detect if a pageblock was isolated or if part of some balloon.
Allocating just full pageblocks would work around this, However,
it would play very badly with CMA.
"

[1] http://lkml.org/lkml/2012/6/14/74

[akpm@linux-foundation.org: simplify nr_zone_isolate_freepages(), rework zone_watermark_ok_safe() comment, simplify set_pageblock_isolate() and restore_pageblock_isolate()]
[akpm@linux-foundation.org: fix CONFIG_MEMORY_ISOLATION=n build]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Suggested-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Tested-by: Aaditya Kumar <aaditya.kumar.30@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix free page check in zone_watermark_ok()

__zone_watermark_ok currently compares free_pages which is a signed type
with z->lowmem_reserve[classzone_idx] which is unsigned which might lead
to sign overflow if free_pages doesn't satisfy the given order (or it came
as negative already) and then we rely on the following order loop to fix
it (which doesn't work for order-0).  Let's fix the type conversion and do
not rely on the given value of free_pages or follow up fixups.

This patch fixes it because "memory-hotplug: fix kswapd looping forever
problem" depends on this.

As benefit of this patch, it doesn't rely on the loop to exit
__zone_watermark_ok in case of high order check and make the first test
effective.(ie, if (free_pages <= min + lowmem_reserve))

Aaditya reported this problem when he test my hotplug patch.

Reported-off-by: Aaditya Kumar <aaditya.kumar@ap.sony.com>
Tested-by: Aaditya Kumar <aaditya.kumar@ap.sony.com>
Signed-off-by: Aaditya Kumar <aaditya.kumar@ap.sony.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: factor out memory isolate functions

mm/page_alloc.c has some memory isolation functions but they are used only
when we enable CONFIG_{CMA|MEMORY_HOTPLUG|MEMORY_FAILURE}.  So let's make
it configurable by new CONFIG_MEMORY_ISOLATION so that it can reduce
binary size and we can check it simple by CONFIG_MEMORY_ISOLATION, not if
defined CONFIG_{CMA|MEMORY_HOTPLUG|MEMORY_FAILURE}.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: move all oom handling to memcontrol.c

By globally defining check_panic_on_oom(), the memcg oom handler can be
moved entirely to mm/memcontrol.c.  This removes the ugly #ifdef in the
oom killer and cleans up the code.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: reduce dependency on tasklist_lock

Since exiting tasks require write_lock_irq(&tasklist_lock) several times,
try to reduce the amount of time the readside is held for oom kills.  This
makes the interface with the memcg oom handler more consistent since it
now never needs to take tasklist_lock unnecessarily.

The only time the oom killer now takes tasklist_lock is when iterating the
children of the selected task, everything else is protected by
rcu_read_lock().

This requires that a reference to the selected process, p, is grabbed
before calling oom_kill_process().  It may release it and grab a reference
on another one of p's threads if !p->mm, but it also guarantees that it
will release the reference before returning.

[hughd@google.com: fix duplicate put_task_struct()]
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: introduce own oom handler to iterate only over its own threads

The global oom killer is serialized by the per-zonelist
try_set_zonelist_oom() which is used in the page allocator.  Concurrent
oom kills are thus a rare event and only occur in systems using
mempolicies and with a large number of nodes.

Memory controller oom kills, however, can frequently be concurrent since
there is no serialization once the oom killer is called for oom conditions
in several different memcgs in parallel.

This creates a massive contention on tasklist_lock since the oom killer
requires the readside for the tasklist iteration.  If several memcgs are
calling the oom killer, this lock can be held for a substantial amount of
time, especially if threads continue to enter it as other threads are
exiting.

Since the exit path grabs the writeside of the lock with irqs disabled in
a few different places, this can cause a soft lockup on cpus as a result
of tasklist_lock starvation.

The kernel lacks unfair writelocks, and successful calls to the oom killer
usually result in at least one thread entering the exit path, so an
alternative solution is needed.

This patch introduces a seperate oom handler for memcgs so that they do
not require tasklist_lock for as much time.  Instead, it iterates only
over the threads attached to the oom memcg and grabs a reference to the
selected thread before calling oom_kill_process() to ensure it doesn't
prematurely exit.

This still requires tasklist_lock for the tasklist dump, iterating
children of the selected process, and killing all other threads on the
system sharing the same memory as the selected victim.  So while this
isn't a complete solution to tasklist_lock starvation, it significantly
reduces the amount of time that it is held.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Sha Zhengju <handai.szj@taobao.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: introduce helper function to process threads during scan

This patch introduces a helper function to process each thread during the
iteration over the tasklist.  A new return type, enum oom_scan_t, is
defined to determine the future behavior of the iteration:

 - OOM_SCAN_OK: continue scanning the thread and find its badness,

 - OOM_SCAN_CONTINUE: do not consider this thread for oom kill, it's
   ineligible,

 - OOM_SCAN_ABORT: abort the iteration and return, or

 - OOM_SCAN_SELECT: always select this thread with the highest badness
   possible.

There is no functional change with this patch.  This new helper function
will be used in the next patch in the memory controller.

Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Sha Zhengju <handai.szj@taobao.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: move declaration for mem_cgroup_out_of_memory to oom.h

mem_cgroup_out_of_memory() is defined in mm/oom_kill.c, so declare it in
linux/oom.h rather than linux/memcontrol.h.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hotplug: mark memory hotplug code in page_alloc.c as __meminit

Mark functions used by both boot and memory hotplug as __meminit to reduce
memory footprint when memory hotplug is disabled.

Alos guard zone_pcp_update() with CONFIG_MEMORY_HOTPLUG because it's only
used by memory hotplug code.

Signed-off-by: Jiang Liu <liuj97@gmail.com>
Cc: Wei Wang <Bessel.Wang@huawei.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Keping Chen <chenkeping@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hotplug: free zone->pageset when a zone becomes empty

When a zone becomes empty after memory offlining, free zone->pageset.
Otherwise it will cause memory leak when adding memory to the empty zone
again because build_all_zonelists() will allocate zone->pageset for an
empty zone.

Signed-off-by: Jiang Liu <liuj97@gmail.com>
Signed-off-by: Wei Wang <Bessel.Wang@huawei.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Keping Chen <chenkeping@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hotplug: correctly add new zone to all other nodes' zone lists

When online_pages() is called to add new memory to an empty zone, it
rebuilds all zone lists by calling build_all_zonelists().  But there's a
bug which prevents the new zone to be added to other nodes' zone lists.

online_pages() {
build_all_zonelists()
.....
node_set_state(zone_to_nid(zone), N_HIGH_MEMORY)
}

Here the node of the zone is put into N_HIGH_MEMORY state after calling
build_all_zonelists(), but build_all_zonelists() only adds zones from
nodes in N_HIGH_MEMORY state to the fallback zone lists.
build_all_zonelists()

    ->__build_all_zonelists()
->build_zonelists()
    ->find_next_best_node()
->for_each_node_state(n, N_HIGH_MEMORY)

So memory in the new zone will never be used by other nodes, and it may
cause strange behavor when system is under memory pressure.  So put node
into N_HIGH_MEMORY state before calling build_all_zonelists().

Signed-off-by: Jianguo Wu <wujianguo@huawei.com>
Signed-off-by: Jiang Liu <liuj97@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Keping Chen <chenkeping@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hotplug: correctly setup fallback zonelists when creating new pgdat

When hotadd_new_pgdat() is called to create new pgdat for a new node, a
fallback zonelist should be created for the new node.  There's code to try
to achieve that in hotadd_new_pgdat() as below:

/*
 * The node we allocated has no zone fallback lists. For avoiding
 * to access not-initialized zonelist, build here.
 */
mutex_lock(&zonelists_mutex);
build_all_zonelists(pgdat, NULL);
mutex_unlock(&zonelists_mutex);

But it doesn't work as expected.  When hotadd_new_pgdat() is called, the
new node is still in offline state because node_set_online(nid) hasn't
been called yet.  And build_all_zonelists() only builds zonelists for
online nodes as:

        for_each_online_node(nid) {
                pg_data_t *pgdat = NODE_DATA(nid);

                build_zonelists(pgdat);
                build_zonelist_cache(pgdat);
        }

Though we hope to create zonelist for the new pgdat, but it doesn't.  So
add a new parameter "pgdat" the build_all_zonelists() to build pgdat for
the new pgdat too.

Signed-off-by: Jiang Liu <liuj97@gmail.com>
Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Keping Chen <chenkeping@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcg: replace inexistence move_lock_page_cgroup() by move_lock_mem_cgroup() in comment

Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcg: mem_cgroup_relize_xxx_limit can guarantee memcg->res.limit <= memcg->memsw.limit

Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcg: complete documentation for tcp memcg files

Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: setup pageblock_order before it's used by sparsemem

On architectures with CONFIG_HUGETLB_PAGE_SIZE_VARIABLE set, such as
Itanium, pageblock_order is a variable with default value of 0.  It's set
to the right value by set_pageblock_order() in function
free_area_init_core().

But pageblock_order may be used by sparse_init() before free_area_init_core()
is called along path:
sparse_init()
    ->sparse_early_usemaps_alloc_node()
->usemap_size()
    ->SECTION_BLOCKFLAGS_BITS
->((1UL << (PFN_SECTION_SHIFT - pageblock_order)) *
NR_PAGEBLOCK_BITS)

The uninitialized pageblock_size will cause memory wasting because
usemap_size() returns a much bigger value then it's really needed.

For example, on an Itanium platform,
sparse_init() pageblock_order=0 usemap_size=24576
free_area_init_core() before pageblock_order=0, usemap_size=24576
free_area_init_core() after pageblock_order=12, usemap_size=8

That means 24K memory has been wasted for each section, so fix it by calling
set_pageblock_order() from sparse_init().

Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Jiang Liu <liuj97@gmail.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Keping Chen <chenkeping@huawei.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory.c:print_vma_addr(): call up_read(&mm->mmap_sem) directly

Call up_read(&mm->mmap_sem) directly since we have already got mm via
current->mm at the beginning of print_vma_addr().

Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: remove obsolete shrink_control comment

09f363c7 ("vmscan: fix shrinker callback bug in fs/super.c") fixed a
shrinker callback which was returning -1 when nr_to_scan is zero, which
caused excessive slab scanning.  But 635697c6 ("vmscan: fix initial
shrinker size handling") fixed the problem, again so we can freely return
-1 although nr_to_scan is zero.  So let's revert 09f363c7 because the
comment added in 09f363c7 made an unnecessary rule.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: CONFIG_HAVE_MEMBLOCK_NODE -> CONFIG_HAVE_MEMBLOCK_NODE_MAP

0ee332c14518699 ("memblock: Kill early_node_map[]") wanted to replace
CONFIG_ARCH_POPULATES_NODE_MAP with CONFIG_HAVE_MEMBLOCK_NODE_MAP but
ended up replacing one occurence with a reference to the non-existent
symbol CONFIG_HAVE_MEMBLOCK_NODE.

The resulting omission of code would probably have been causing problems
to 32-bit machines with memory hotplug.

Signed-off-by: Rabin Vincent <rabin@rab.in>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: have order > 0 compaction start off where it left

Order > 0 compaction stops when enough free pages of the correct page
order have been coalesced.  When doing subsequent higher order
allocations, it is possible for compaction to be invoked many times.

However, the compaction code always starts out looking for things to
compact at the start of the zone, and for free pages to compact things to
at the end of the zone.

This can cause quadratic behaviour, with isolate_freepages starting at the
end of the zone each time, even though previous invocations of the
compaction code already filled up all free memory on that end of the zone.

This can cause isolate_freepages to take enormous amounts of CPU with
certain workloads on larger memory systems.

The obvious solution is to have isolate_freepages remember where it left
off last time, and continue at that point the next time it gets invoked
for an order > 0 compaction.  This could cause compaction to fail if
cc->free_pfn and cc->migrate_pfn are close together initially, in that
case we restart from the end of the zone and try once more.

Forced full (order == -1) compactions are left alone.

[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: s/laste/last/, use 80 cols]
Signed-off-by: Rik van Riel <riel@redhat.com>
Reported-by: Jim Schutt <jaschut@sandia.gov>
Tested-by: Jim Schutt <jaschut@sandia.gov>
Cc: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: rename mem_control_xxx to memcg_xxx

Replace memory_cgroup_xxx() with memcg_xxx()

Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix bad behavior in use_hierarchy file

I have an application that does the following:

* copy the state of all controllers attached to a hierarchy
* replicate it as a child of the current level.

I would expect writes to the files to mostly succeed, since they are
inheriting sane values from parents.

But that is not the case for use_hierarchy.  If it is set to 0, we succeed
ok.  If we're set to 1, the value of the file is automatically set to 1 in
the children, but if userspace tries to write the very same 1, it will
fail.  That same situation happens if we set use_hierarchy, create a
child, and then try to write 1 again.

Now, there is no reason whatsoever for failing to write a value that is
already there.  It doesn't even match the comments, that states:

 /* If parent's use_hierarchy is set, we can't make any modifications
  * in the child subtrees...

since we are not changing anything.

So test the new value against the one we're storing, and automatically
return 0 if we're not proposing a change.

Signed-off-by: Glauber Costa <glommer@parallels.com>
Cc: Dhaval Giani <dhaval.giani@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unused LRU_ALL_EVICTABLE

Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: rename config variables

Sanity:

CONFIG_CGROUP_MEM_RES_CTLR -> CONFIG_MEMCG
CONFIG_CGROUP_MEM_RES_CTLR_SWAP -> CONFIG_MEMCG_SWAP
CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED -> CONFIG_MEMCG_SWAP_ENABLED
CONFIG_CGROUP_MEM_RES_CTLR_KMEM -> CONFIG_MEMCG_KMEM

[mhocko@suse.cz: fix missed bits]
Cc: Glauber Costa <glommer@parallels.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: clean up __count_immobile_pages()

The __count_immobile_pages() naming is rather awkward.  Choose a more
clear name and add a comment.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: do not use page_count() without a page pin

d179e84ba ("mm: vmscan: do not use page_count without a page pin") fixed
this problem in vmscan.c but same problem is in __count_immobile_pages().

I copy and paste d179e84ba's contents for description.

"It is unsafe to run page_count during the physical pfn scan because
compound_head could trip on a dangling pointer when reading
page->first_page if the compound page is being freed by another CPU."

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Wanpeng Li <liwp.linux@gmail.com>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: replace some information in tasklist dump

The number of ptes and swap entries are used in the oom killer's badness
heuristic, so they should be shown in the tasklist dump.

This patch adds those fields and replaces cpu and oom_adj values that are
currently emitted.  Cpu isn't interesting and oom_adj is deprecated and
will be removed later this year, the same information is already displayed
as oom_score_adj which is used internally.

At the same time, make the documentation a little more clear to state this
information is helpful to determine why the oom killer chose the task it
did to kill.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: fix potential killing of thread that is disabled from oom killing

/proc/sys/vm/oom_kill_allocating_task will immediately kill current when
the oom killer is called to avoid a potentially expensive tasklist scan
for large systems.

Currently, however, it is not checking current's oom_score_adj value which
may be OOM_SCORE_ADJ_MIN, meaning that it has been disabled from oom
killing.

This patch avoids killing current in such a condition and simply falls
back to the tasklist scan since memory still needs to be freed.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: clear pages_scanned only if draining a pcp adds pages to the buddy allocator again

commit 2ff754fa8f ("mm: clear pages_scanned only if draining a pcp adds
pages to the buddy allocator again") fixed one free_pcppages_bulk()
misuse.  But two another miuse still exist.

This patch fixes it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, fadvise: don't return -EINVAL when filesystem cannot implement fadvise()

Eric Wong reported his test suite failex when /tmp is tmpfs.

https://lkml.org/lkml/2012/2/24/479

Currentlt the input check of POSIX_FADV_WILLNEED has two problems.

- requires a_ops->readpage.  But in fact, force_page_cache_readahead()
  requires that the target filesystem has either ->readpage or ->readpages.

- returns -EINVAL when the filesystem doesn't have ->readpage.  But
  posix says that fadvise is merely a hint.  Thus fadvise() should return
  0 if filesystem has no means of implementing fadvise().  The userland
  application should not know nor care whcih type of filesystem backs the
  TMPDIR directory, as Eric pointed out.  There is nothing which userspace
  can do to solve this error.

So change the return value to 0 when filesytem doesn't support readahead.

[akpm@linux-foundation.org: checkpatch fixes]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Tested-by: Eric Wong <normalperson@yhbt.net>
Reviewed-by: Wanlong Gao <gaowanlong@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction: cleanup on compaction_deferred

When CONFIG_COMPACTION is enabled, compaction_deferred() tries to
recalculate the deferred limit again, which isn't necessary.

When CONFIG_COMPACTION is disabled, compaction_deferred() should return
"true" or "false" since it has "bool" for its return value.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: make mem_cgroup_force_empty_list() return bool

mem_cgroup_force_empty_list() just returns 0 or -EBUSY and -EBUSY
indicates 'you need to retry'.  Make mem_cgroup_force_empty_list() return
a bool to simplify the logic.

[akpm@linux-foundation.org: rework mem_cgroup_force_empty_list()'s comment]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: mem_cgroup_move_parent() doesn't need gfp_mask

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: clean up force_empty_list() return value check

After bf544fdc241da8 "memcg: move charges to root cgroup if
use_hierarchy=0 in mem_cgroup_move_hugetlb_parent()"
mem_cgroup_move_parent() returns only -EBUSY or -EINVAL.  So we can remove
the -ENOMEM and -EINTR checks.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove check for signal_pending() during rmdir()

After bf544fdc241da8 "memcg: move charges to root cgroup if
use_hierarchy=0 in mem_cgroup_move_hugetlb_parent()", no memory reclaim
will occur when removing a memory cgroup.  If -EINTR is returned here,
cgroup will show a warning.

We don't need to handle any user interruption signal.  Remove this.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memblock.c:memblock_double_array(): cosmetic cleanups

This function is an 80-column eyesore, quite unnecessarily.  Clean that
up, and use standard comment layout style.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Greg Pearson <greg.pearson@hp.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: do not schedule if current has been killed

The oom killer currently schedules away from current in an uninterruptible
sleep if it does not have access to memory reserves.  It's possible that
current was killed because it shares memory with the oom killed thread or
because it was killed by the user in the interim, however.

This patch only schedules away from current if it does not have a pending
kill, i.e.  if it does not share memory with the oom killed thread.  It's
possible that it will immediately retry its memory allocation and fail,
but it will immediately be given access to memory reserves if it calls the
oom killer again.

This prevents the delay of memory freeing when threads that share memory
with the oom killed thread get unnecessarily scheduled.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: remove exclude and wakeup rmdir calls from migrate

We already hold the hugetlb_lock.  That should prevent a parallel cgroup
rmdir from touching page's hugetlb cgroup.  So remove the exclude and
wakeup calls.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: assign the page hugetlb cgroup when we move the page to active list.

A page's hugetlb cgroup assignment and movement to the active list should
occur with hugetlb_lock held.  Otherwise when we remove the hugetlb cgroup
we will iterate the active list and find pages with NULL hugetlb cgroup
values.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: move all the in use pages to active list

When we fail to allocate pages from the reserve pool, hugetlb tries to
allocate huge pages using alloc_buddy_huge_page.  Add these to the active
list.  We also need to add the huge page we allocate when we soft offline
the oldpage to active list.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: add HugeTLB controller documentation

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: migrate hugetlb cgroup info from oldpage to new page during migration

With HugeTLB pages, hugetlb cgroup is uncharged in compound page
destructor.  Since we are holding a hugepage reference, we can be sure
that old page won't get uncharged till the last put_page().

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: add hugetlb cgroup control files

Add the control files for hugetlb controller

[akpm@linux-foundation.org: s/CONFIG_CGROUP_HUGETLB_RES_CTLR/CONFIG_MEMCG_HUGETLB/g]
[akpm@linux-foundation.org: s/CONFIG_MEMCG_HUGETLB/CONFIG_CGROUP_HUGETLB/]
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: add support for cgroup removal

Add support for cgroup removal.  If we don't have parent cgroup, the
charges are moved to root cgroup.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: add charge/uncharge routines for hugetlb cgroup

Add the charge and uncharge routines for hugetlb cgroup.  We do cgroup
charging in page alloc and uncharge in compound page destructor.
Assigning page's hugetlb cgroup is protected by hugetlb_lock.

[liwp@linux.vnet.ibm.com: add huge_page_order check to avoid incorrect uncharge]
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Wanpeng Li <liwp.linux@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb/cgroup: add the cgroup pointer to page lru

Add the hugetlb cgroup pointer to 3rd page lru.next.  This limit the usage
to hugetlb cgroup to only hugepages with 3 or more normal pages.  I guess
that is an acceptable limitation.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: add new HugeTLB cgroup

Implement a new controller that allows us to control HugeTLB allocations.
The extension allows to limit the HugeTLB usage per control group and
enforces the controller limit during page fault.  Since HugeTLB doesn't
support page reclaim, enforcing the limit at page fault time implies that,
the application will get SIGBUS signal if it tries to access HugeTLB pages
beyond its limit.  This requires the application to know beforehand how
much HugeTLB pages it would require for its use.

The charge/uncharge calls will be added to HugeTLB code in later patch.
Support for cgroup removal will be added in later patches.

[akpm@linux-foundation.org: s/CONFIG_CGROUP_HUGETLB_RES_CTLR/CONFIG_MEMCG_HUGETLB/g]
[akpm@linux-foundation.org: s/CONFIG_MEMCG_HUGETLB/CONFIG_CGROUP_HUGETLB/g]
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: make some static variables global

We will use them later in hugetlb_cgroup.c

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add a list for tracking in-use HugeTLB pages

hugepage_activelist will be used to track currently used HugeTLB pages.
We need to find the in-use HugeTLB pages to support HugeTLB cgroup removal.
On cgroup removal we update the page's HugeTLB cgroup to point to parent
cgroup.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: simplify migrate_huge_page()

Since we migrate only one hugepage, don't use linked list for passing the
page around.  Directly pass the page that need to be migrated as argument.
This also removes the usage of page->lru in the migrate path.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: use mmu_gather instead of a temporary linked list for accumulating pages

Use a mmu_gather instead of a temporary linked list for accumulating pages
when we unmap a hugepage range

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add an inline helper for finding hstate index

Add an inline helper and use it in the code.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: don't use ERR_PTR with VM_FAULT* values

The current use of VM_FAULT_* codes with ERR_PTR requires us to ensure
VM_FAULT_* values will not exceed MAX_ERRNO value.  Decouple the
VM_FAULT_* values from MAX_ERRNO.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Hillf Danton <dhillf@gmail.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: rename max_hstate to hugetlb_max_hstate

This patchset implements a cgroup resource controller for HugeTLB pages.
The controller allows to limit the HugeTLB usage per control group and
enforces the controller limit during page fault.  Since HugeTLB doesn't
support page reclaim, enforcing the limit at page fault time implies that,
the application will get SIGBUS signal if it tries to access HugeTLB pages
beyond its limit.  This requires the application to know beforehand how
much HugeTLB pages it would require for its use.

The goal is to control how many HugeTLB pages a group of task can
allocate.  It can be looked at as an extension of the existing quota
interface which limits the number of HugeTLB pages per hugetlbfs
superblock.  HPC job scheduler requires jobs to specify their resource
requirements in the job file.  Once their requirements can be met, job
schedulers like (SLURM) will schedule the job.  We need to make sure that
the jobs won't consume more resources than requested.  If they do we
should either error out or kill the application.

This patch:

Rename max_hstate to hugetlb_max_hstate.  We will be using this from other
subsystems like hugetlb controller in later patches.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Hillf Danton <dhillf@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: prepare for removal of obsolete /proc/sys/vm/nr_pdflush_threads

Since per-BDI flusher threads were introduced in 2.6, the pdflush
mechanism is not used any more.  But the old interface exported through
/proc/sys/vm/nr_pdflush_threads still exists and is obviously useless.

For back-compatibility, printk warning information and return 2 to notify
the users that the interface is removed.

Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/buddy: cleanup on should_fail_alloc_page

Currently, function should_fail() has "bool" for its return value, so it's
reasonable to change the return value of function should_fail_alloc_page()
into "bool" as well.

The patch does cleanup on function should_fail_alloc_page() to have "bool"
for its return value.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: account the total_vm in the vm_stat_account()

vm_stat_account() accounts the shared_vm, stack_vm and reserved_vm now.
But we can also account for total_vm in the vm_stat_account() which makes
the code tidy.

Even for mprotect_fixup(), we can get the right result in the end.

Signed-off-by: Huang Shijie <shijie8@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

documentation: update how page-cluster affects swap I/O

Fix of the documentation of /proc/sys/vm/page-cluster to match the
behavior of the code and add some comments about what the tunable will
change in that behavior.

Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Acked-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swap: allow swap readahead to be merged

Swap readahead works fine, but the I/O to disk is almost always done in
page size requests, despite the fact that readahead submits
1<<page-cluster pages at a time.

On older kernels the old per device plugging behavior might have captured
this and merged the requests, but currently all comes down to much more
I/Os than required.

On a single device this might not be an issue, but as soon as a server
runs on shared san resources savin I/Os not only improves swapin
throughput but also provides a lower resource utilization.

With a load running KVM in a lot of memory overcommitment (the hot memory
is 1.5 times the host memory) swapping throughput improves significantly
and the lead feels more responsive as well as achieves more throughput.

In a test setup with 16 swap disks running blocktrace on one of those disks
shows the improved merging:
Prior:
Reads Queued:     560,888,    2,243MiB  Writes Queued:     226,242,  904,968KiB
Read Dispatches:  544,701,    2,243MiB  Write Dispatches:  159,318,  904,968KiB
Reads Requeued:         0               Writes Requeued:         0
Reads Completed:  544,716,    2,243MiB  Writes Completed:  159,321,  904,980KiB
Read Merges:       16,187,   64,748KiB  Write Merges:       61,744,  246,976KiB
IO unplugs:       149,614               Timer unplugs:       2,940

With the patch:
Reads Queued:     734,315,    2,937MiB  Writes Queued:     300,188,    1,200MiB
Read Dispatches:  214,972,    2,937MiB  Write Dispatches:  215,176,    1,200MiB
Reads Requeued:         0               Writes Requeued:         0
Reads Completed:  214,971,    2,937MiB  Writes Completed:  215,177,    1,200MiB
Read Merges:      519,343,    2,077MiB  Write Merges:       73,325,  293,300KiB
IO unplugs:       337,130               Timer unplugs:      11,184

I got ~10% to ~40% more throughput in my cases and at the same time much
lower cpu consumption when broken down per transferred kilobyte (the
majority of that due to saved interrupts and better cache handling).  In a
shared SAN others might get an additional benefit as well, because this
now causes less protocol overhead.

Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove MEM_CGROUP_CHARGE_TYPE_FORCE

There are no users since commit b24028572fb69 ("memcg: remove PCG_CACHE").

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Hugh Dickins <hughd@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: rename MEM_CGROUP_CHARGE_TYPE_MAPPED as MEM_CGROUP_CHARGE_TYPE_ANON

Now, in memcg, 2 "MAPPED" enum/macro are found
 MEM_CGROUP_CHARGE_TYPE_MAPPED
 MEM_CGROUP_STAT_FILE_MAPPED

Thier names looks similar to each other but the former is used for
accounting anonymous memory. rename it as TYPE_ANON.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: rename MEM_CGROUP_STAT_SWAPOUT as MEM_CGROUP_STAT_SWAP

MEM_CGROUP_STAT_SWAPOUT represents the usage of swap rather than
the number of swap-out events. Rename it to be MEM_CGROUP_STAT_SWAP.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Hugh Dickins <hughd@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make vb_alloc() more foolproof

If someone calls vb_alloc() (or vm_map_ram() for that matter) to allocate
0 bytes (0 pages), get_order() returns BITS_PER_LONG - PAGE_CACHE_SHIFT
and interesting stuff happens.  So make debugging such problems easier and
warn about 0-size allocation.

[akpm@linux-foundation.org: use WARN_ON-return-value feature]
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc: walk vmap_areas by sorted list instead of rb_next()

There's a walk by repeating rb_next to find a suitable hole.  Could be
simply replaced by walk on the sorted vmap_area_list.  More simpler and
efficient.

Mutation of the list and tree only happens in pair within
__insert_vmap_area and __free_vmap_area, under protection of
vmap_area_lock.  The patch code is also under vmap_area_lock, so the list
walk is safe, and consistent with the tree walk.

Tested on SMP by repeating batch of vmalloc anf vfree for random sizes and
rounds for hours.

Signed-off-by: Hong Zhiguo <honkiko@gmail.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/media/video/v4l2-ioctl.c: fix build

Fix zillions of these:

drivers/media/video/v4l2-ioctl.c:1848: error: unknown field 'func' specified in initializer
drivers/media/video/v4l2-ioctl.c:1848: warning: missing braces around initializer
drivers/media/video/v4l2-ioctl.c:1848: warning: (near initialization for 'v4l2_ioctls[0].<anonymous>')
drivers/media/video/v4l2-ioctl.c:1848: warning: initialization makes integer from pointer without a cast
drivers/media/video/v4l2-ioctl.c:1848: error: initializer element is not computable at load time
drivers/media/video/v4l2-ioctl.c:1848: error: (near initialization for 'v4l2_ioctls[0].<anonymous>.offset')

Cc: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

xtensa: select generic atomic64_t support

This will fix build errors:

block/blk-cgroup.c:609:2: error: unknown type name 'atomic64_t'
block/blk-cgroup.c:609:2: error: implicit declaration of function 'ATOMIC64_INIT' [-Werror=implicit-function-declaration]

Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fault-injection: fix failcmd.sh warning

"fault-injection: add tool to run command with failslab or
fail_page_alloc" added tools/testing/fault-injection/failcmd.sh to make it
easier to inject slab/page allocation failures by fault injection.

failcmd.sh prints the following warning when running with arguments
for command.

# ./failcmd.sh echo aaa
failcmd.sh: line 209: [: echo: binary operator expected
aaa

This warning is caused by an improper check whether at least one
parameter is left after parsing command options.

Fix it by testing the length of $1 instead of $@

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge tag 'writeback-proportions' of git://git./linux/kernel/git/wfg/linux

Pull writeback updates from Wu Fengguang:
 "Use time based periods to age the writeback proportions, which can
  adapt equally well to fast/slow devices."

Fix up trivial conflict in comment in fs/sync.c

* tag 'writeback-proportions' of git://git.kernel.org/pub/scm/linux/kernel/git/wfg/linux:
  writeback: Fix some comment errors
  block: Convert BDI proportion calculations to flexible proportions
  lib: Fix possible deadlock in flexible proportion code
  lib: Proportions with flexible period

Merge tag 'nfs-for-3.6-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client updates from Trond Myklebust:
 "Features include:
   - More preparatory patches for modularising NFSv2/v3/v4.  Split out
     the various NFSv2/v3/v4-specific code into separate files
   - More preparation for the NFSv4 migration code
   - Ensure that OPEN(O_CREATE) observes the pNFS mds threshold
     parameters
   - pNFS fast failover when the data servers are down
   - Various cleanups and debugging patches"

* tag 'nfs-for-3.6-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (67 commits)
  nfs: fix fl_type tests in NFSv4 code
  NFS: fix pnfs regression with directio writes
  NFS: fix pnfs regression with directio reads
  sunrpc: clnt: Add missing braces
  nfs: fix stub return type warnings
  NFS: exit_nfs_v4() shouldn't be an __exit function
  SUNRPC: Add a missing spin_unlock to gss_mech_list_pseudoflavors
  NFS: Split out NFS v4 client functions
  NFS: Split out the NFS v4 filesystem types
  NFS: Create a single nfs_clone_super() function
  NFS: Split out NFS v4 server creating code
  NFS: Initialize the NFS v4 client from init_nfs_v4()
  NFS: Move the v4 getroot code to nfs4getroot.c
  NFS: Split out NFS v4 file operations
  NFS: Initialize v4 sysctls from nfs_init_v4()
  NFS: Create an init_nfs_v4() function
  NFS: Split out NFS v4 inode operations
  NFS: Split out NFS v3 inode operations
  NFS: Split out NFS v2 inode operations
  NFS: Clean up nfs4_proc_setclientid() and friends
  ...

Merge tag 'mfd-for-linus-3.6-1' of git://git./linux/kernel/git/sameo/mfd-2.6

Pull MFD fix from Samuel Ortiz:
 "This one fixes an s5m8767 regulator build breakage due to a merge
  conflict caused by the MFD s5m API changes."

* tag 'mfd-for-linus-3.6-1' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/mfd-2.6:
  regulator: Fix an s5m8767 build failure

Merge branch 'v4l_for_linus' of git://git./linux/kernel/git/mchehab/linux-media

Pull media updates from Mauro Carvalho Chehab:
 "This is the first part of the media patches for v3.6.

  This patch series contain:
   - new DVB frontend: rtl2832
   - new video drivers: adv7393
   - some unused files got removed
   - a selection API cleanup between V4L2 and V4L2 subdev API's
   - a major redesign at v4l-ioctl2, in order to clean it up
   - several driver fixes and improvements."

* 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media: (174 commits)
  v4l: Export v4l2-common.h in include/linux/Kbuild
  media: Revert "[media] Terratec Cinergy S2 USB HD Rev.2"
  [media] media: Use pr_info not homegrown pr_reg macro
  [media] Terratec Cinergy S2 USB HD Rev.2
  [media] v4l: Correct conflicting V4L2 subdev selection API documentation
  [media] Feature removal: V4L2 selections API target and flag definitions
  [media] v4l: Unify selection flags documentation
  [media] v4l: Unify selection flags
  [media] v4l: Common documentation for selection targets
  [media] v4l: Unify selection targets across V4L2 and V4L2 subdev interfaces
  [media] v4l: Remove "_ACTUAL" from subdev selection API target definition names
  [media] V4L: Remove "_ACTIVE" from the selection target name definitions
  [media] media: dvb-usb: print mac address via native %pM
  [media] s5p-tv: Use module_i2c_driver in sii9234_drv.c file
  [media] media: gpio-ir-recv: add allowed_protos for platform data
  [media] s5p-jpeg: Use module_platform_driver in jpeg-core.c file
  [media] saa7134: fix spelling of detach in label
  [media] cx88-blackbird: replace ioctl by unlocked_ioctl
  [media] cx88: don't use current_norm
  [media] cx88: fix a number of v4l2-compliance violations
  ...

Merge branch 'akpm' (Andrew's patch-bomb)

Merge Andrew's first set of patches:
 "Non-MM patches:

   - lots of misc bits

   - tree-wide have_clk() cleanups

   - quite a lot of printk tweaks.  I draw your attention to "printk:
     convert the format for KERN_<LEVEL> to a 2 byte pattern" which
     looks a bit scary.  But afaict it's solid.

   - backlight updates

   - lib/ feature work (notably the addition and use of memweight())

   - checkpatch updates

   - rtc updates

   - nilfs updates

   - fatfs updates (partial, still waiting for acks)

   - kdump, proc, fork, IPC, sysctl, taskstats, pps, etc

   - new fault-injection feature work"

* Merge emailed patches from Andrew Morton <akpm@linux-foundation.org>: (128 commits)
  drivers/misc/lkdtm.c: fix missing allocation failure check
  lib/scatterlist: do not re-write gfp_flags in __sg_alloc_table()
  fault-injection: add tool to run command with failslab or fail_page_alloc
  fault-injection: add selftests for cpu and memory hotplug
  powerpc: pSeries reconfig notifier error injection module
  memory: memory notifier error injection module
  PM: PM notifier error injection module
  cpu: rewrite cpu-notifier-error-inject module
  fault-injection: notifier error injection
  c/r: fcntl: add F_GETOWNER_UIDS option
  resource: make sure requested range is included in the root range
  include/linux/aio.h: cpp->C conversions
  fs: cachefiles: add support for large files in filesystem caching
  pps: return PTR_ERR on error in device_create
  taskstats: check nla_reserve() return
  sysctl: suppress kmemleak messages
  ipc: use Kconfig options for __ARCH_WANT_[COMPAT_]IPC_PARSE_VERSION
  ipc: compat: use signed size_t types for msgsnd and msgrcv
  ipc: allow compat IPC version field parsing if !ARCH_WANT_OLD_COMPAT_IPC
  ipc: add COMPAT_SHMLBA support
  ...

drivers/misc/lkdtm.c: fix missing allocation failure check

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=44691

Reported-by: <rucsoftsec@gmail.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/scatterlist: do not re-write gfp_flags in __sg_alloc_table()

We are seeing a lot of sg_alloc_table allocation failures using the new
drm prime infrastructure.  We isolated the cause to code in
__sg_alloc_table that was re-writing the gfp_flags.

There is a comment in the code that suggest that there is an assumption
about the allocation coming from a memory pool.  This was likely true
when sg lists were primarily used for disk I/O.

Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Cong Wang <amwang@redhat.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Rob Clark <rob.clark@linaro.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Inki Dae <inki.dae@samsung.com>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Sonny Rao <sonnyrao@chromium.org>
Cc: Olof Johansson <olofj@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fault-injection: add tool to run command with failslab or fail_page_alloc

This adds tools/testing/fault-injection/failcmd.sh to run a command while
injecting slab/page allocation failures via fault injection.

Example:

Run a command "make -C tools/testing/selftests/ run_tests" with
injecting slab allocation failure.

# ./tools/testing/fault-injection/failcmd.sh \
-- make -C tools/testing/selftests/ run_tests

Same as above except to specify 100 times failures at most instead of
one time at most by default.

# ./tools/testing/fault-injection/failcmd.sh --times=100 \
-- make -C tools/testing/selftests/ run_tests

Same as above except to inject page allocation failure instead of slab
allocation failure.

# env FAILCMD_TYPE=fail_page_alloc \
./tools/testing/fault-injection/failcmd.sh --times=100 \
-- make -C tools/testing/selftests/ run_tests

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fault-injection: add selftests for cpu and memory hotplug

This adds two selftests

* tools/testing/selftests/cpu-hotplug/on-off-test.sh is testing script
for CPU hotplug

1. Online all hot-pluggable CPUs
2. Offline all hot-pluggable CPUs
3. Online all hot-pluggable CPUs again
4. Exit if cpu-notifier-error-inject.ko is not available
5. Offline all hot-pluggable CPUs in preparation for testing
6. Test CPU hot-add error handling by injecting notifier errors
7. Online all hot-pluggable CPUs in preparation for testing
8. Test CPU hot-remove error handling by injecting notifier errors

* tools/testing/selftests/memory-hotplug/on-off-test.sh is doing the
similar thing for memory hotplug.

1. Online all hot-pluggable memory
2. Offline 10% of hot-pluggable memory
3. Online all hot-pluggable memory again
4. Exit if memory-notifier-error-inject.ko is not available
5. Offline 10% of hot-pluggable memory in preparation for testing
6. Test memory hot-add error handling by injecting notifier errors
7. Online all hot-pluggable memory in preparation for testing
8. Test memory hot-remove error handling by injecting notifier errors

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc: pSeries reconfig notifier error injection module

This provides the ability to inject artifical errors to pSeries reconfig
notifier chain callbacks.  It is controlled through debugfs interface
under /sys/kernel/debug/notifier-error-inject/pSeries-reconfig

If the notifier call chain should be failed with some events
notified, write the error code to "actions/<notifier event>/error".

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory: memory notifier error injection module

This provides the ability to inject artifical errors to memory hotplug
notifier chain callbacks.  It is controlled through debugfs interface
under /sys/kernel/debug/notifier-error-inject/memory

If the notifier call chain should be failed with some events notified,
write the error code to "actions/<notifier event>/error".

Example: Inject memory hotplug offline error (-12 == -ENOMEM)

# cd /sys/kernel/debug/notifier-error-inject/memory
# echo -12 > actions/MEM_GOING_OFFLINE/error
# echo offline > /sys/devices/system/memory/memoryXXX/state
bash: echo: write error: Cannot allocate memory

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

PM: PM notifier error injection module

This provides the ability to inject artifical errors to PM notifier chain
callbacks.  It is controlled through debugfs interface under
/sys/kernel/debug/notifier-error-inject/pm

Each of the files in "error" directory represents an event which can be
failed and contains the error code.  If the notifier call chain should be
failed with some events notified, write the error code to the files.

If the notifier call chain should be failed with some events notified,
write the error code to "actions/<notifier event>/error".

Example: Inject PM suspend error (-12 = -ENOMEM)

# cd /sys/kernel/debug/notifier-error-inject/pm
# echo -12 > actions/PM_SUSPEND_PREPARE/error
# echo mem > /sys/power/state
bash: echo: write error: Cannot allocate memory

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpu: rewrite cpu-notifier-error-inject module

Rewrite existing cpu-notifier-error-inject module to use debugfs based new
framework.

This change removes cpu_up_prepare_error and cpu_down_prepare_error module
parameters which were used to specify error code to be injected.  We could
keep these module parameters for backward compatibility by module_param_cb
but it seems overkill for this module.

This provides the ability to inject artifical errors to CPU notifier chain
callbacks.  It is controlled through debugfs interface under
/sys/kernel/debug/notifier-error-inject/cpu

If the notifier call chain should be failed with some events notified,
write the error code to "actions/<notifier event>/error".

Example1: inject CPU offline error (-1 == -EPERM)

# cd /sys/kernel/debug/notifier-error-inject/cpu
# echo -1 > actions/CPU_DOWN_PREPARE/error
# echo 0 > /sys/devices/system/cpu/cpu1/online
bash: echo: write error: Operation not permitted

Example2: inject CPU online error (-2 == -ENOENT)

# cd /sys/kernel/debug/notifier-error-inject/cpu
# echo -2 > actions/CPU_UP_PREPARE/error
# echo 1 > /sys/devices/system/cpu/cpu1/online
bash: echo: write error: No such file or directory

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fault-injection: notifier error injection

This patchset provides kernel modules that can be used to test the error
handling of notifier call chain failures by injecting artifical errors to
the following notifier chain callbacks.

 * CPU notifier
 * PM notifier
 * memory hotplug notifier
 * powerpc pSeries reconfig notifier

Example: Inject CPU offline error (-1 == -EPERM)

  # cd /sys/kernel/debug/notifier-error-inject/cpu
  # echo -1 > actions/CPU_DOWN_PREPARE/error
  # echo 0 > /sys/devices/system/cpu/cpu1/online
  bash: echo: write error: Operation not permitted

The patchset also adds cpu and memory hotplug tests to
tools/testing/selftests These tests first do simple online and offline
test and then do fault injection tests if notifier error injection
module is available.

This patch:

The notifier error injection provides the ability to inject artifical
errors to specified notifier chain callbacks.  It is useful to test the
error handling of notifier call chain failures.

This adds common basic functions to define which type of events can be
fail and to initialize the debugfs interface to control what error code
should be returned and which event should be failed.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

c/r: fcntl: add F_GETOWNER_UIDS option

When we restore file descriptors we would like them to look exactly as
they were at dumping time.

With help of fcntl it's almost possible, the missing snippet is file
owners UIDs.

To be able to read their values the F_GETOWNER_UIDS is introduced.

This option is valid iif CONFIG_CHECKPOINT_RESTORE is turned on, otherwise
returning -EINVAL.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "Serge E. Hallyn" <serge@hallyn.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

resource: make sure requested range is included in the root range

When the requested range is outside of the root range the logic in
__reserve_region_with_split will cause an infinite recursion which will
overflow the stack as seen in the warning bellow.

This particular stack overflow was caused by requesting the
(100000000-107ffffff) range while the root range was (0-ffffffff).  In
this case __request_resource would return the whole root range as
conflict range (i.e.  0-ffffffff).  Then, the logic in
__reserve_region_with_split would continue the recursion requesting the
new range as (conflict->end+1, end) which incidentally in this case
equals the originally requested range.

This patch aborts looking for an usable range when the request does not
intersect with the root range.  When the request partially overlaps with
the root range, it ajust the request to fall in the root range and then
continues with the new request.

When the request is modified or aborted errors and a stack trace are
logged to allow catching the errors in the upper layers.

[    5.968374] WARNING: at kernel/sched.c:4129 sub_preempt_count+0x63/0x89()
[    5.975150] Modules linked in:
[    5.978184] Pid: 1, comm: swapper Not tainted 3.0.22-mid27-00004-gb72c817 #46
[    5.985324] Call Trace:
[    5.987759]  [<c1039dfc>] ? console_unlock+0x17b/0x18d
[    5.992891]  [<c1039620>] warn_slowpath_common+0x48/0x5d
[    5.998194]  [<c1031758>] ? sub_preempt_count+0x63/0x89
[    6.003412]  [<c1039644>] warn_slowpath_null+0xf/0x13
[    6.008453]  [<c1031758>] sub_preempt_count+0x63/0x89
[    6.013499]  [<c14d60c4>] _raw_spin_unlock+0x27/0x3f
[    6.018453]  [<c10c6349>] add_partial+0x36/0x3b
[    6.022973]  [<c10c7c0a>] deactivate_slab+0x96/0xb4
[    6.027842]  [<c14cf9d9>] __slab_alloc.isra.54.constprop.63+0x204/0x241
[    6.034456]  [<c103f78f>] ? kzalloc.constprop.5+0x29/0x38
[    6.039842]  [<c103f78f>] ? kzalloc.constprop.5+0x29/0x38
[    6.045232]  [<c10c7dc9>] kmem_cache_alloc_trace+0x51/0xb0
[    6.050710]  [<c103f78f>] ? kzalloc.constprop.5+0x29/0x38
[    6.056100]  [<c103f78f>] kzalloc.constprop.5+0x29/0x38
[    6.061320]  [<c17b45e9>] __reserve_region_with_split+0x1c/0xd1
[    6.067230]  [<c17b4693>] __reserve_region_with_split+0xc6/0xd1
...
[    7.179057]  [<c17b4693>] __reserve_region_with_split+0xc6/0xd1
[    7.184970]  [<c17b4779>] reserve_region_with_split+0x30/0x42
[    7.190709]  [<c17a8ebf>] e820_reserve_resources_late+0xd1/0xe9
[    7.196623]  [<c17c9526>] pcibios_resource_survey+0x23/0x2a
[    7.202184]  [<c17cad8a>] pcibios_init+0x23/0x35
[    7.206789]  [<c17ca574>] pci_subsys_init+0x3f/0x44
[    7.211659]  [<c1002088>] do_one_initcall+0x72/0x122
[    7.216615]  [<c17ca535>] ? pci_legacy_init+0x3d/0x3d
[    7.221659]  [<c17a27ff>] kernel_init+0xa6/0x118
[    7.226265]  [<c17a2759>] ? start_kernel+0x334/0x334
[    7.231223]  [<c14d7482>] kernel_thread_helper+0x6/0x10

Signed-off-by: Octavian Purdila <octavian.purdila@intel.com>
Signed-off-by: Ram Pai <linuxram@us.ibm.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/aio.h: cpp->C conversions

Convert init_sync_kiocb() from a nasty macro into a nice C function.  The
struct assignment trick takes care of zeroing all unmentioned fields.
Shrinks fs/read_write.o's .text from 9857 bytes to 9714.

Also demacroize is_sync_kiocb() and aio_ring_avail().  The latter fixes an
arg-referenced-multiple-times hand grenade.

Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: cachefiles: add support for large files in filesystem caching

Support the caching of large files.

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=31182

Signed-off-by: Justin Lecher <jlec@gentoo.org>
Signed-off-by: Suresh Jayaraman <sjayaraman@suse.com>
Tested-by: Suresh Jayaraman <sjayaraman@suse.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pps: return PTR_ERR on error in device_create

We should return PTR_ERR if the call to the device_create function fails.
Without this patch we instead return the value from a successful call to
cdev_add if the call to device_create fails.

Signed-off-by: Emil Goode <emilgoode@gmail.com>
Acked-by: Devendra Naga <devendra.aaru@gmail.com>
Cc: Alexander Gordeev <lasaine@lvk.cs.msu.su>
Cc: Rodolfo Giometti <giometti@enneenne.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

taskstats: check nla_reserve() return

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=44621

Reported-by: <rucsoftsec@gmail.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sysctl: suppress kmemleak messages

register_sysctl_table() is a strange function, as it makes internal
allocations (a header) to register a sysctl_table.  This header is a
handle to the table that is created, and can be used to unregister the
table.  But if the table is permanent and never unregistered, the header
acts the same as a static variable.

Unfortunately, this allocation of memory that is never expected to be
freed fools kmemleak in thinking that we have leaked memory.  For those
sysctl tables that are never unregistered, and have no pointer referencing
them, kmemleak will think that these are memory leaks:

unreferenced object 0xffff880079fb9d40 (size 192):
  comm "swapper/0", pid 0, jiffies 4294667316 (age 12614.152s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<ffffffff8146b590>] kmemleak_alloc+0x73/0x98
    [<ffffffff8110a935>] kmemleak_alloc_recursive.constprop.42+0x16/0x18
    [<ffffffff8110b852>] __kmalloc+0x107/0x153
    [<ffffffff8116fa72>] kzalloc.constprop.8+0xe/0x10
    [<ffffffff811703c9>] __register_sysctl_paths+0xe1/0x160
    [<ffffffff81170463>] register_sysctl_paths+0x1b/0x1d
    [<ffffffff8117047d>] register_sysctl_table+0x18/0x1a
    [<ffffffff81afb0a1>] sysctl_init+0x10/0x14
    [<ffffffff81b05a6f>] proc_sys_init+0x2f/0x31
    [<ffffffff81b0584c>] proc_root_init+0xa5/0xa7
    [<ffffffff81ae5b7e>] start_kernel+0x3d0/0x40a
    [<ffffffff81ae52a7>] x86_64_start_reservations+0xae/0xb2
    [<ffffffff81ae53ad>] x86_64_start_kernel+0x102/0x111
    [<ffffffffffffffff>] 0xffffffffffffffff

The sysctl_base_table used by sysctl itself is one such instance that
registers the table to never be unregistered.

Use kmemleak_not_leak() to suppress the kmemleak false positive.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: use Kconfig options for __ARCH_WANT_[COMPAT_]IPC_PARSE_VERSION

Rather than #define the options manually in the architecture code, add
Kconfig options for them and select them there instead.  This also allows
us to select the compat IPC version parsing automatically for platforms
using the old compat IPC interface.

Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: compat: use signed size_t types for msgsnd and msgrcv

The msgsnd and msgrcv system calls use size_t to represent the size of the
message being transferred.  POSIX states that values of msgsz greater than
SSIZE_MAX cause the result to be implementation-defined.  On Linux, this
equates to returning -EINVAL if (long) msgsz < 0.

For compat tasks where !CONFIG_ARCH_WANT_OLD_COMPAT_IPC and compat_size_t
is smaller than size_t, negative size values passed from userspace will be
interpreted as positive values by do_msg{rcv,snd} and will fail to exit
early with -EINVAL.

This patch changes the compat prototypes for msg{rcv,snd} so that the
message size is represented as a compat_ssize_t, which we cast to the
native ssize_t type for the core IPC code.

Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Chris Metcalf <cmetcalf@tilera.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: allow compat IPC version field parsing if !ARCH_WANT_OLD_COMPAT_IPC

Commit 48b25c43e6ee ("ipc: provide generic compat versions of IPC
syscalls") added a new ARCH_WANT_OLD_COMPAT_IPC config option for
architectures to select if their compat target requires the old IPC
syscall interface.

For architectures (such as AArch64) that do not require the internal
calling conventions provided by this option, but have a compat target
where the C library passes the IPC_64 flag explicitly,
compat_ipc_parse_version no longer strips out the flag before calling
the native system call implementation, resulting in unknown SHM/IPC
commands and -EINVAL being returned to userspace.

This patch separates the selection of the internal calling conventions
for the IPC syscalls from the version parsing, allowing architectures to
select __ARCH_WANT_COMPAT_IPC_PARSE_VERSION if they want to use version
parsing whilst retaining the newer syscall calling conventions.

Acked-by: Chris Metcalf <cmetcalf@tilera.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: add COMPAT_SHMLBA support

If the SHMLBA definition for a native task differs from the definition for
a compat task, the do_shmat() function would need to handle both.

This patch introduces COMPAT_SHMLBA, which is used by the compat shmat
syscall when calling the ipc code and allows architectures such as AArch64
(where the native SHMLBA is 64k but the compat (AArch32) definition is
16k) to provide the correct semantics for compat IPC system calls.

Cc: David S. Miller <davem@davemloft.net>
Cc: Chris Zankel <chris@zankel.net>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>