Add user_reserve_kbytes knob.

Limit the growth of the memory reserved for other user processes to
min(3% current process size, user_reserve_pages).  Only about 8MB is
necessary to enable recovery in the default mode, and only a few hundred
MB are required even when overcommit is disabled.

user_reserve_pages defaults to min(3% free pages, 128MB)

I arrived at 128MB by taking the max VSZ of sshd, login, bash, and top ...
then adding the RSS of each.

This only affects OVERCOMMIT_NEVER mode.

Background

1. user reserve

__vm_enough_memory reserves a hardcoded 3% of the current process size for
other applications when overcommit is disabled.  This was done so that a
user could recover if they launched a memory hogging process.  Without the
reserve, a user would easily run into a message such as:

bash: fork: Cannot allocate memory

2. admin reserve

Additionally, a hardcoded 3% of free memory is reserved for root in both
overcommit 'guess' and 'never' modes.  This was intended to prevent a
scenario where root-cant-log-in and perform recovery operations.

Note that this reserve shrinks, and doesn't guarantee a useful reserve.

Motivation

The two hardcoded memory reserves should be updated to account for current
memory sizes.

Also, the admin reserve would be more useful if it didn't shrink too much.

When the current code was originally written, 1GB was considered
"enterprise".  Now the 3% reserve can grow to multiple GB on large memory
systems, and it only needs to be a few hundred MB at most to enable a user
or admin to recover a system with an unwanted memory hogging process.

I've found that reducing these reserves is especially beneficial for a
specific type of application load:

 * single application system
 * one or few processes (e.g. one per core)
 * allocating all available memory
 * not initializing every page immediately
 * long running

I've run scientific clusters with this sort of load.  A long running job
sometimes failed many hours (weeks of CPU time) into a calculation.  They
weren't initializing all of their memory immediately, and they weren't
using calloc, so I put systems into overcommit 'never' mode.  These
clusters run diskless and have no swap.

However, with the current reserves, a user wishing to allocate as much
memory as possible to one process may be prevented from using, for
example, almost 2GB out of 32GB.

The effect is less, but still significant when a user starts a job with
one process per core.  I have repeatedly seen a set of processes
requesting the same amount of memory fail because one of them could not
allocate the amount of memory a user would expect to be able to allocate.
For example, Message Passing Interfce (MPI) processes, one per core.  And
it is similar for other parallel programming frameworks.

Changing this reserve code will make the overcommit never mode more useful
by allowing applications to allocate nearly all of the available memory.

Also, the new admin_reserve_kbytes will be safer than the current behavior
since the hardcoded 3% of available memory reserve can shrink to something
useless in the case where applications have grabbed all available memory.

Risks

* "bash: fork: Cannot allocate memory"

  The downside of the first patch-- which creates a tunable user reserve
  that is only used in overcommit 'never' mode--is that an admin can set
  it so low that a user may not be able to kill their process, even if
  they already have a shell prompt.

  Of course, a user can get in the same predicament with the current 3%
  reserve--they just have to launch processes until 3% becomes negligible.

* root-cant-log-in problem

  The second patch, adding the tunable rootuser_reserve_pages, allows
  the admin to shoot themselves in the foot by setting it too small.  They
  can easily get the system into a state where root-can't-log-in.

  However, the new admin_reserve_kbytes will be safer than the current
  behavior since the hardcoded 3% of available memory reserve can shrink
  to something useless in the case where applications have grabbed all
  available memory.

Alternatives

 * Memory cgroups provide a more flexible way to limit application memory.

   Not everyone wants to set up cgroups or deal with their overhead.

 * We could create a fourth overcommit mode which provides smaller reserves.

   The size of useful reserves may be drastically different depending
   on the whether the system is embedded or enterprise.

 * Force users to initialize all of their memory or use calloc.

   Some users don't want/expect the system to overcommit when they malloc.
   Overcommit 'never' mode is for this scenario, and it should work well.

The new user and admin reserve tunables are simple to use, with low
overhead compared to cgroups.  The patches preserve current behavior where
3% of memory is less than 128MB, except that the admin reserve doesn't
shrink to an unusable size under pressure.  The code allows admins to tune
for embedded and enterprise usage.

FAQ

 * How is the root-cant-login problem addressed?
   What happens if admin_reserve_pages is set to 0?

   Root is free to shoot themselves in the foot by setting
   admin_reserve_kbytes too low.

   On x86_64, the minimum useful reserve is:
     8MB for overcommit 'guess'
   128MB for overcommit 'never'

   admin_reserve_pages defaults to min(3% free memory, 8MB)

   So, anyone switching to 'never' mode needs to adjust
   admin_reserve_pages.

 * How do you calculate a minimum useful reserve?

   A user or the admin needs enough memory to login and perform
   recovery operations, which includes, at a minimum:

   sshd or login + bash (or some other shell) + top (or ps, kill, etc.)

   For overcommit 'guess', we can sum resident set sizes (RSS)
   because we only need enough memory to handle what the recovery
   programs will typically use. On x86_64 this is about 8MB.

   For overcommit 'never', we can take the max of their virtual sizes (VSZ)
   and add the sum of their RSS. We use VSZ instead of RSS because mode
   forces us to ensure we can fulfill all of the requested memory allocations--
   even if the programs only use a fraction of what they ask for.
   On x86_64 this is about 128MB.

   When swap is enabled, reserves are useful even when they are as
   small as 10MB, regardless of overcommit mode.

   When both swap and overcommit are disabled, then the admin should
   tune the reserves higher to be absolutley safe. Over 230MB each
   was safest in my testing.

 * What happens if user_reserve_pages is set to 0?

   Note, this only affects overcomitt 'never' mode.

   Then a user will be able to allocate all available memory minus
   admin_reserve_kbytes.

   However, they will easily see a message such as:

   "bash: fork: Cannot allocate memory"

   And they won't be able to recover/kill their application.
   The admin should be able to recover the system if
   admin_reserve_kbytes is set appropriately.

 * What's the difference between overcommit 'guess' and 'never'?

   "Guess" allows an allocation if there are enough free + reclaimable
   pages. It has a hardcoded 3% of free pages reserved for root.

   "Never" allows an allocation if there is enough swap + a configurable
   percentage (default is 50) of physical RAM. It has a hardcoded 3% of
   free pages reserved for root, like "Guess" mode. It also has a
   hardcoded 3% of the current process size reserved for additional
   applications.

 * Why is overcommit 'guess' not suitable even when an app eventually
   writes to every page? It takes free pages, file pages, available
   swap pages, reclaimable slab pages into consideration. In other words,
   these are all pages available, then why isn't overcommit suitable?

   Because it only looks at the present state of the system. It
   does not take into account the memory that other applications have
   malloced, but haven't initialized yet. It overcommits the system.

Test Summary

There was little change in behavior in the default overcommit 'guess'
mode with swap enabled before and after the patch. This was expected.

Systems run most predictably (i.e. no oom kills) in overcommit 'never'
mode with swap enabled. This also allowed the most memory to be allocated
to a user application.

Overcommit 'guess' mode without swap is a bad idea. It is easy to
crash the system. None of the other tested combinations crashed.
This matches my experience on the Roadrunner supercomputer.

Without the tunable user reserve, a system in overcommit 'never' mode
and without swap does not allow the admin to recover, although the
admin can.

With the new tunable reserves, a system in overcommit 'never' mode
and without swap can be configured to:

1. maximize user-allocatable memory, running close to the edge of
recoverability

2. maximize recoverability, sacrificing allocatable memory to
ensure that a user cannot take down a system

Test Description

Fedora 18 VM - 4 x86_64 cores, 5725MB RAM, 4GB Swap

System is booted into multiuser console mode, with unnecessary services
turned off. Caches were dropped before each test.

Hogs are user memtester processes that attempt to allocate all free memory
as reported by /proc/meminfo

In overcommit 'never' mode, memory_ratio=100

Test Results

3.9.0-rc1-mm1

Overcommit | Swap | Hogs | MB Got/Wanted | OOMs | User Recovery | Admin Recovery
----------   ----   ----   -------------   ----   -------------   --------------
guess        yes    1      5432/5432       no     yes             yes
guess        yes    4      5444/5444       1      yes             yes
guess        no     1      5302/5449       no     yes             yes
guess        no     4      -               crash  no              no

never        yes    1      5460/5460       1      yes             yes
never        yes    4      5460/5460       1      yes             yes
never        no     1      5218/5432       no     no              yes
never        no     4      5203/5448       no     no              yes

3.9.0-rc1-mm1-tunablereserves

User and Admin Recovery show their respective reserves, if applicable.

Overcommit | Swap | Hogs | MB Got/Wanted | OOMs | User Recovery | Admin Recovery
----------   ----   ----   -------------   ----   -------------   --------------
guess        yes    1      5419/5419       no     - yes           8MB yes
guess        yes    4      5436/5436       1      - yes           8MB yes
guess        no     1      5440/5440       *      - yes           8MB yes
guess        no     4      -               crash  - no            8MB no

* process would successfully mlock, then the oom killer would pick it

never        yes    1      5446/5446       no     10MB yes        20MB yes
never        yes    4      5456/5456       no     10MB yes        20MB yes
never        no     1      5387/5429       no     128MB no        8MB barely
never        no     1      5323/5428       no     226MB barely    8MB barely
never        no     1      5323/5428       no     226MB barely    8MB barely

never        no     1      5359/5448       no     10MB no         10MB barely

never        no     1      5323/5428       no     0MB no          10MB barely
never        no     1      5332/5428       no     0MB no          50MB yes
never        no     1      5293/5429       no     0MB no          90MB yes

never        no     1      5001/5427       no     230MB yes       338MB yes
never        no     4*     4998/5424       no     230MB yes       338MB yes

* more memtesters were launched, able to allocate approximately another 100MB

Future Work

 - Test larger memory systems.

 - Test an embedded image.

 - Test other architectures.

 - Time malloc microbenchmarks.

 - Would it be useful to be able to set overcommit policy for
   each memory cgroup?

 - Some lines are slightly above 80 chars.
   Perhaps define a macro to convert between pages and kb?
   Other places in the kernel do this.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: make init_user_reserve() static]
Signed-off-by: NAndrew Shewmaker <agshew@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use the new interface, remove one ifdef.  No code size changes.

We could/should have been using __meminit/__meminitdata here but there's
now no point in doing that because all this code is elided at compile time.

Cc: Li Zefan <lizefan@huawei.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Saves an ifdef, no code size changes
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Squishes a warning which my change to hotplug_memory_notifier() added.

I want to keep that warning, because it is punishment for failnig to check
the hotplug_memory_notifier() return value.

Cc: Greg KH <greg@kroah.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Squishes a statement-with-no-effect warning, removes some ifdefs and
shrinks .text by 2 bytes.

Note that this code fails to check for blocking_notifier_chain_register()
failures.

Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Squishes a statement-with-no-effect warning, removes some ifdefs and
shrinks .text by one byte!

Note that this code fails to check for blocking_notifier_chain_register()
failures.
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When CONFIG_MEMORY_HOTPLUG=n, we don't want the memory-hotplug notifier
handlers to be included in the .o files, for space reasons.

The existing hotplug_memory_notifier() tries to handle this but testing
with gcc-4.4.4 shows that it doesn't work - the hotplug functions are
still present in the .o files.

So implement a new register_hotmemory_notifier() which is a copy of
register_hotcpu_notifier(), and which actually works as desired.
hotplug_memory_notifier() and register_memory_notifier() callsites
should be converted to use this new register_hotmemory_notifier().

While we're there, let's repair the existing hotplug_memory_notifier():
it simply stomps on the register_memory_notifier() return value, so
well-behaved code cannot check for errors.  Apparently non of the
existing callers were well-behaved :(

Cc: Andrew Shewmaker <agshew@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

[sfr@canb.auug.org.au: add missing semicolon]
Signed-off-by: NCody P Schafer <cody@linux.vnet.ibm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NCody P Schafer <cody@linux.vnet.ibm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

powerpc and x86 were opencoding copies of setup_nr_node_ids(), which
page_alloc provides but makes static.  Make it avaliable to the archs in
linux/mm.h.
Signed-off-by: NCody P Schafer <cody@linux.vnet.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When booting on a large memory system, the kernel spends considerable
time in memmap_init_zone() setting up memory zones.  Analysis shows
significant time spent in __early_pfn_to_nid().

The routine memmap_init_zone() checks each PFN to verify the nid is
valid.  __early_pfn_to_nid() sequentially scans the list of pfn ranges
to find the right range and returns the nid.  This does not scale well.
On a 4 TB (single rack) system there are 308 memory ranges to scan.  The
higher the PFN the more time spent sequentially spinning through memory
ranges.

Since memmap_init_zone() increments pfn, it will almost always be
looking for the same range as the previous pfn, so check that range
first.  If it is in the same range, return that nid.  If not, scan the
list as before.

A 4 TB (single rack) UV1 system takes 512 seconds to get through the
zone code.  This performance optimization reduces the time by 189
seconds, a 36% improvement.

A 2 TB (single rack) UV2 system goes from 212.7 seconds to 99.8 seconds,
a 112.9 second (53%) reduction.

[akpm@linux-foundation.org: make the statics __meminitdata]
[akpm@linux-foundation.org: fix comment formatting]
[akpm@linux-foundation.org: fix ia64, per yinghai]
[akpm@linux-foundation.org: add missing semicolon, per Tony]
Signed-off-by: NRuss Anderson <rja@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Tested-by: N"Luck, Tony" <tony.luck@intel.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Lin Feng <linfeng@cn.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NJianguo Wu <wujianguo@huawei.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The following problem was reported against a distribution kernel when
zone_reclaim was enabled but the same problem applies to the mainline
kernel.  The reproduction case was as follows

1. Run numactl -m +0 dd if=largefile of=/dev/null
   This allocates a large number of clean pages in node 0

2. numactl -N +0 memhog 0.5*Mg
   This start a memory-using application in node 0.

The expected behaviour is that the clean pages get reclaimed and the
application uses node 0 for its memory.  The observed behaviour was that
the memory for the memhog application was allocated off-node since
commits cd38b115 ("mm: page allocator: initialise ZLC for first zone
eligible for zone_reclaim") and commit 76d3fbf8 ("mm: page
allocator: reconsider zones for allocation after direct reclaim").

The assumption of those patches was that it was always preferable to
allocate quickly than stall for long periods of time and they were meant
to take care that the zone was only marked full when necessary but an
important case was missed.

In the allocator fast path, only the low watermarks are checked.  If the
zones free pages are between the low and min watermark then allocations
from the allocators slow path will succeed.  However, zone_reclaim will
only reclaim SWAP_CLUSTER_MAX or 1<<order pages.  There is no guarantee
that this will meet the low watermark causing the zone to be marked full
prematurely.

This patch will only mark the zone full after zone_reclaim if it the min
watermarks are checked or if page reclaim failed to make sufficient
progress.

[mhocko@suse.cz: fix alloc_flags test]
Signed-off-by: NMel Gorman <mgorman@suse.de>
Reported-by: NHedi Berriche <hedi@sgi.com>
Tested-by: NHedi Berriche <hedi@sgi.com>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Reviewed-by: NWanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Memory hotplug can happen on a machine under load, memory shortness
and fragmentation, so huge page allocations for the vmemmap are not
guaranteed to succeed.

Try to fall back to regular pages before failing the hotplug event
completely.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We already have generic code to allocate vmemmap with regular pages, use
it.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

No need to maintain addr_end and p_end when they are never actually read
anywhere on !pse setups.  Remove the dead code.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The sparse code, when asking the architecture to populate the vmemmap,
specifies the section range as a starting page and a number of pages.

This is an awkward interface, because none of the arch-specific code
actually thinks of the range in terms of 'struct page' units and always
translates it to bytes first.

In addition, later patches mix huge page and regular page backing for
the vmemmap.  For this, they need to call vmemmap_populate_basepages()
on sub-section ranges with PAGE_SIZE and PMD_SIZE in mind.  But these
are not necessarily multiples of the 'struct page' size and so this unit
is too coarse.

Just translate the section range into bytes once in the generic sparse
code, then pass byte ranges down the stack.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Tested-by: NDavid S. Miller <davem@davemloft.net>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Hot-adding memory on x86_64 normally requires huge page allocation.
When this is done to a VM guest, it's usually because the system is
already tight on memory, so the request tends to fail.  Try to avoid
this by adding __GFP_REPEAT to the allocation flags.

Addresses http://bugs.debian.org/699913Signed-off-by: NBen Hutchings <ben@decadent.org.uk>
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reported-by: NBernhard Schmidt <Bernhard.Schmidt@lrz.de>
Tested-by: NBernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Particularly in oom conditions, it's troublesome that hugetlb memory is
not displayed.  All other meminfo that is emitted will not add up to
what is expected, and there is no artifact left in the kernel log to
show that a potentially significant amount of memory is actually
allocated as hugepages which are not available to be reclaimed.

Booting with hugepages=8192 on the command line, this memory is now
shown in oom conditions.  For example, with echo m >
/proc/sysrq-trigger:

  Node 0 hugepages_total=2048 hugepages_free=2048 hugepages_surp=0 hugepages_size=2048kB
  Node 1 hugepages_total=2048 hugepages_free=2048 hugepages_surp=0 hugepages_size=2048kB
  Node 2 hugepages_total=2048 hugepages_free=2048 hugepages_surp=0 hugepages_size=2048kB
  Node 3 hugepages_total=2048 hugepages_free=2048 hugepages_surp=0 hugepages_size=2048kB

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: NDavid Rientjes <rientjes@google.com>
Acked-by: NMichal Hocko <mhocko@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Using mbind to change the mempolicy to MPOL_BIND on several adjacent
mmapped blocks may result in a reset of the mempolicy to MPOL_DEFAULT in
vma_adjust.

Test code.  Correct result is three lines containing "OK".

#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
#include <numaif.h>
#include <errno.h>

/* gcc mbind_test.c -lnuma -o mbind_test -Wall */
#define MAXNODE 4096

void allocate()
{
	int ret;
	int len;
	int policy = -1;
	unsigned char *p;
	unsigned long mask[MAXNODE] = { 0 };
	unsigned long retmask[MAXNODE] = { 0 };

	len = getpagesize() * 0x2fc00;
	p = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS,
		 -1, 0);
	if (p == MAP_FAILED)
		printf("mbind err: %d\n", errno);

	mask[0] = 1;
	ret = mbind(p, len, MPOL_BIND, mask, MAXNODE, 0);
	if (ret < 0)
		printf("mbind err: %d %d\n", ret, errno);
	ret = get_mempolicy(&policy, retmask, MAXNODE, p, MPOL_F_ADDR);
	if (ret < 0)
		printf("get_mempolicy err: %d %d\n", ret, errno);

	if (policy == MPOL_BIND)
		printf("OK\n");
	else
		printf("ERROR: policy is %d\n", policy);
}

int main()
{
	allocate();
	allocate();
	allocate();
	return 0;
}
Signed-off-by: NSteven T Hampson <steven.t.hampson@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

ARM processors with LPAE enabled use 3 levels of page tables, with an
entry in the top level (pgd) covering 1GB of virtual space.  Because of
the branch relocation limitations on ARM, the loadable modules are
mapped 16MB below PAGE_OFFSET, making the corresponding 1GB pgd shared
between kernel modules and user space.

If free_pgtables() is called with the default ceiling 0,
free_pgd_range() (and subsequently called functions) also frees the page
table shared between user space and kernel modules (which is normally
handled by the ARM-specific pgd_free() function).  This patch changes
defines the ARM USER_PGTABLES_CEILING to TASK_SIZE when CONFIG_ARM_LPAE
is enabled.

Note that the pgd_free() function already checks the presence of the
shared pmd page allocated by pgd_alloc() and frees it, though with
ceiling 0 this wasn't necessary.
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org>	[3.3+]
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

On architectures where a pgd entry may be shared between user and kernel
(e.g.  ARM+LPAE), freeing page tables needs a ceiling other than 0.
This patch introduces a generic USER_PGTABLES_CEILING that arch code can
override.  It is the responsibility of the arch code setting the ceiling
to ensure the complete freeing of the page tables (usually in
pgd_free()).

[catalin.marinas@arm.com: commit log; shift_arg_pages(), asm-generic/pgtables.h changes]
Signed-off-by: NHugh Dickins <hughd@google.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: <stable@vger.kernel.org>	[3.3+]
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Since commit 2d11085e ("memcg: do not create memsw files if swap
accounting is disabled") memsw files are created only if memcg swap
accounting is enabled so it doesn't make any sense to check for it
explicitly in mem_cgroup_read(), mem_cgroup_write() and
mem_cgroup_reset().
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Tejun Heo <tj@kernel.org>
Acked-by: NJohannes Weiner <hannes@cmpxchg.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove the WARN_ON_ONCE(!mm) check as the comment suggested.  Kernel
code calls find_vma only when it is absolutely sure that the mm_struct
arg to it is non-NULL.
Signed-off-by: NZhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: k80c <k80ck80c@gmail.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now, vmap_area_list is exported as VMCOREINFO for makedumpfile to get
the start address of vmalloc region (vmalloc_start).  The address which
contains vmalloc_start value is represented as below:

  vmap_area_list.next - OFFSET(vmap_area.list) + OFFSET(vmap_area.va_start)

However, both OFFSET(vmap_area.va_start) and OFFSET(vmap_area.list)
aren't exported as VMCOREINFO.

So this patch exports them externally with small cleanup.

[akpm@linux-foundation.org: vmalloc.h should include list.h for list_head]
Signed-off-by: NAtsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now, there is no need to maintain vmlist after initializing vmalloc.  So
remove related code and data structure.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Although our intention is to unexport internal structure entirely, but
there is one exception for kexec.  kexec dumps address of vmlist and
makedumpfile uses this information.

We are about to remove vmlist, then another way to retrieve information
of vmalloc layer is needed for makedumpfile.  For this purpose, we
export vmap_area_list, instead of vmlist.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch is a preparatory step for removing vmlist entirely.  For
above purpose, we change iterating a vmap_list codes to iterating a
vmap_area_list.  It is somewhat trivial change, but just one thing
should be noticed.

Using vmap_area_list in vmallocinfo() introduce ordering problem in SMP
system.  In s_show(), we retrieve some values from vm_struct.
vm_struct's values is not fully setup when va->vm is assigned.  Full
setup is notified by removing VM_UNLIST flag without holding a lock.
When we see that VM_UNLIST is removed, it is not ensured that vm_struct
has proper values in view of other CPUs.  So we need smp_[rw]mb for
ensuring that proper values is assigned when we see that VM_UNLIST is
removed.

Therefore, this patch not only change a iteration list, but also add a
appropriate smp_[rw]mb to right places.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch is a preparatory step for removing vmlist entirely.  For
above purpose, we change iterating a vmap_list codes to iterating a
vmap_area_list.  It is somewhat trivial change, but just one thing
should be noticed.

vmlist is lack of information about some areas in vmalloc address space.
For example, vm_map_ram() allocate area in vmalloc address space, but it
doesn't make a link with vmlist.  To provide full information about
vmalloc address space is better idea, so we don't use va->vm and use
vmap_area directly.  This makes get_vmalloc_info() more precise.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now, when we hold a vmap_area_lock, va->vm can't be discarded.  So we can
safely access to va->vm when iterating a vmap_area_list with holding a
vmap_area_lock.  With this property, change iterating vmlist codes in
vread/vwrite() to iterating vmap_area_list.

There is a little difference relate to lock, because vmlist_lock is mutex,
but, vmap_area_lock is spin_lock.  It may introduce a spinning overhead
during vread/vwrite() is executing.  But, these are debug-oriented
functions, so this overhead is not real problem for common case.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Inserting and removing an entry to vmlist is linear time complexity, so
it is inefficient.  Following patches will try to remove vmlist
entirely.  This patch is preparing step for it.

For removing vmlist, iterating vmlist codes should be changed to
iterating a vmap_area_list.  Before implementing that, we should make
sure that when we iterate a vmap_area_list, accessing to va->vm doesn't
cause a race condition.  This patch ensure that when iterating a
vmap_area_list, there is no race condition for accessing to vm_struct.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now get_vmalloc_info() is in fs/proc/mmu.c.  There is no reason that this
code must be here and it's implementation needs vmlist_lock and it iterate
a vmlist which may be internal data structure for vmalloc.

It is preferable that vmlist_lock and vmlist is only used in vmalloc.c
for maintainability. So move the code to vmalloc.c
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patchset removes vm_struct list management after initializing
vmalloc.  Adding and removing an entry to vmlist is linear time
complexity, so it is inefficient.  If we maintain this list, overall
time complexity of adding and removing area to vmalloc space is O(N),
although we use rbtree for finding vacant place and it's time complexity
is just O(logN).

And vmlist and vmlist_lock is used many places of outside of vmalloc.c.
It is preferable that we hide this raw data structure and provide
well-defined function for supporting them, because it makes that they
cannot mistake when manipulating theses structure and it makes us easily
maintain vmalloc layer.

For kexec and makedumpfile, I export vmap_area_list, instead of vmlist.
This comes from Atsushi's recommendation.  For more information, please
refer below link.  https://lkml.org/lkml/2012/12/6/184

This patch:

The purpose of iterating a vmlist is finding vm area with specific virtual
address.  find_vm_area() is provided for this purpose and more efficient,
because it uses a rbtree.  So change it.
Signed-off-by: NJoonsoo Kim <js1304@gmail.com>
Signed-off-by: NJoonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: NGuan Xuetao <gxt@mprc.pku.edu.cn>
Acked-by: NIngo Molnar <mingo@kernel.org>
Acked-by: NChris Metcalf <cmetcalf@tilera.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Walking a bio's page mappings has proved problematic, so create a new
bio flag to indicate that a bio's data needs to be snapshotted in order
to guarantee stable pages during writeback.  Next, for the one user
(ext3/jbd) of snapshotting, hook all the places where writes can be
initiated without PG_writeback set, and set BIO_SNAP_STABLE there.

We must also flag journal "metadata" bios for stable writeout, since
file data can be written through the journal.  Finally, the
MS_SNAP_STABLE mount flag (only used by ext3) is now superfluous, so get
rid of it.

[akpm@linux-foundation.org: rename _submit_bh()'s `flags' to `bio_flags', delobotomize the _submit_bh declaration]
[akpm@linux-foundation.org: teeny cleanup]
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit abf09bed ("s390/mm: implement software dirty bits")
introduced another difference in the pte layout vs.  the pmd layout on
s390, thoroughly breaking the s390 support for hugetlbfs.  This requires
replacing some more pte_xxx functions in mm/hugetlbfs.c with a
huge_pte_xxx version.

This patch introduces those huge_pte_xxx functions and their generic
implementation in asm-generic/hugetlb.h, which will now be included on
all architectures supporting hugetlbfs apart from s390.  This change
will be a no-op for those architectures.

[akpm@linux-foundation.org: fix warning]
Signed-off-by: NGerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>	[for !s390 parts]
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

drop_caches.c provides code only invokable via sysctl, so don't compile it
in when CONFIG_SYSCTL=n.
Signed-off-by: NJosh Triplett <josh@joshtriplett.org>
Acked-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now that we have generic and well ordered cgroup tree walkers there is
no need to keep css_get_next in the place.
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: NLi Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

mem_cgroup_iter basically does two things currently.  It takes care of
the house keeping (reference counting, raclaim cookie) and it iterates
through a hierarchy tree (by using cgroup generic tree walk).  The code
would be much more easier to follow if we move the iteration outside of
the function (to __mem_cgrou_iter_next) so the distinction is more
clear.  This patch doesn't introduce any functional changes.
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The current implementation of mem_cgroup_iter has to consider both css
and memcg to find out whether no group has been found (css==NULL - aka
the loop is completed) and that no memcg is associated with the found
node (!memcg - aka css_tryget failed because the group is no longer
alive).  This leads to awkward tweaks like tests for css && !memcg to
skip the current node.

It will be much easier if we got rid off css variable altogether and
only rely on memcg.  In order to do that the iteration part has to skip
dead nodes.  This sounds natural to me and as a nice side effect we will
get a simple invariant that memcg is always alive when non-NULL and all
nodes have been visited otherwise.

We could get rid of the surrounding while loop but keep it in for now to
make review easier.  It will go away in the following patch.
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now that the per-node-zone-priority iterator caches memory cgroups
rather than their css ids we have to be careful and remove them from the
iterator when they are on the way out otherwise they might live for
unbounded amount of time even though their group is already gone (until
the global/targeted reclaim triggers the zone under priority to find out
the group is dead and let it to find the final rest).

We can fix this issue by relaxing rules for the last_visited memcg.
Instead of taking a reference to the css before it is stored into
iter->last_visited we can just store its pointer and track the number of
removed groups from each memcg's subhierarchy.

This number would be stored into iterator everytime when a memcg is
cached.  If the iter count doesn't match the curent walker root's one we
will start from the root again.  The group counter is incremented
upwards the hierarchy every time a group is removed.

The iter_lock can be dropped because racing iterators cannot leak the
reference anymore as the reference count is not elevated for
last_visited when it is cached.

Locking rules got a bit complicated by this change though.  The iterator
primarily relies on rcu read lock which makes sure that once we see a
valid last_visited pointer then it will be valid for the whole RCU walk.
smp_rmb makes sure that dead_count is read before last_visited and
last_dead_count while smp_wmb makes sure that last_visited is updated
before last_dead_count so the up-to-date last_dead_count cannot point to
an outdated last_visited.  css_tryget then makes sure that the
last_visited is still alive in case the iteration races with the cached
group removal (css is invalidated before mem_cgroup_css_offline
increments dead_count).

In short:
mem_cgroup_iter
 rcu_read_lock()
 dead_count = atomic_read(parent->dead_count)
 smp_rmb()
 if (dead_count != iter->last_dead_count)
 	last_visited POSSIBLY INVALID -> last_visited = NULL
 if (!css_tryget(iter->last_visited))
 	last_visited DEAD -> last_visited = NULL
 next = find_next(last_visited)
 css_tryget(next)
 css_put(last_visited) 	// css would be invalidated and parent->dead_count
 			// incremented if this was the last reference
 iter->last_visited = next
 smp_wmb()
 iter->last_dead_count = dead_count
 rcu_read_unlock()

cgroup_rmdir
 cgroup_destroy_locked
  atomic_add(CSS_DEACT_BIAS, &css->refcnt) // subsequent css_tryget fail
   mem_cgroup_css_offline
    mem_cgroup_invalidate_reclaim_iterators
     while(parent = parent_mem_cgroup)
     	atomic_inc(parent->dead_count)
  css_put(css) // last reference held by cgroup core

Spotted by Ying Han.

Original idea from Johannes Weiner.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Acked-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ying Han <yinghan@google.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>