drivers/misc: ROHM BH1780GLI ambient light sensor driver

Add support for ROHM BH1780GLI Ambient light sensor.

BH1780 supports I2C interface.  Driver supports read/update of power state
and read of lux value (through SYSFS).  Writing value 3 to power_state
enables the sensor and current lux value could be read.

Currently this driver follows the same sysfs convention as supported by
drivers/misc/isl29003.c.

Signed-off-by: Hemanth V <hemanthv@ti.com>
Reviewed-by: Daniel Mack <daniel@caiaq.de>
Acked-by: Jonathan Cameron <jic23@cam.ac.uk>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Wolfram Sang <w.sang@pengutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

scripts/mod/modpost.c: fix memory leak

sec2annotation returns malloc'ed buffer directly to printf as an argument.
 Free this buffer after printing.

Signed-off-by: Alexey Fomenko <ext-alexey.fomenko@nokia.com>
Cc: Trevor Keith <tsrk@tsrk.net>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Michal Marek <mmarek@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu: inline iommu_num_pages

A profile of a network benchmark showed iommu_num_pages rather high up:

     0.52%  iommu_num_pages

Looking at the profile, an integer divide is taking almost all of the time:

      %
         :      c000000000376ea4 <.iommu_num_pages>:
    1.93 :      c000000000376ea4:       fb e1 ff f8     std     r31,-8(r1)
    0.00 :      c000000000376ea8:       f8 21 ff c1     stdu    r1,-64(r1)
    0.00 :      c000000000376eac:       7c 3f 0b 78     mr      r31,r1
    3.86 :      c000000000376eb0:       38 84 ff ff     addi    r4,r4,-1
    0.00 :      c000000000376eb4:       38 05 ff ff     addi    r0,r5,-1
    0.00 :      c000000000376eb8:       7c 84 2a 14     add     r4,r4,r5
   46.95 :      c000000000376ebc:       7c 00 18 38     and     r0,r0,r3
   45.66 :      c000000000376ec0:       7c 84 02 14     add     r4,r4,r0
    0.00 :      c000000000376ec4:       7c 64 2b 92     divdu   r3,r4,r5
    0.00 :      c000000000376ec8:       38 3f 00 40     addi    r1,r31,64
    0.00 :      c000000000376ecc:       eb e1 ff f8     ld      r31,-8(r1)
    1.61 :      c000000000376ed0:       4e 80 00 20     blr

Since every caller of iommu_num_pages passes in a constant power of two
we can inline this such that the divide is replaced by a shift. The
entire function is only a few instructions once optimised, so it is
a good candidate for inlining overall.

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

vfs: fix warning: 'dirent' is used uninitialized in this function

Using:

gcc (GCC) 4.5.0 20100610 (prerelease)

The following warnings appear:

fs/readdir.c: In function `filldir64':
fs/readdir.c:240:15: warning: `dirent' is used uninitialized in this function
fs/readdir.c: In function `filldir':
fs/readdir.c:155:15: warning: `dirent' is used uninitialized in this function
fs/compat.c: In function `compat_filldir64':
fs/compat.c:1071:11: warning: `dirent' is used uninitialized in this function
fs/compat.c: In function `compat_filldir':
fs/compat.c:984:15: warning: `dirent' is used uninitialized in this function

The warnings are related to the use of the NAME_OFFSET() macro.  Luckily,
it appears as though the standard offsetof() macro is what is being
implemented by NAME_OFFSET(), thus we can fix the warning and use a more
standard code construct at the same time.

Signed-off-by: Kevin Winchester <kjwinchester@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

SubmittingPatches: add more about patch descriptions

Add more information about patch descriptions.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel.h: remove unused NIPQUAD and NIPQUAD_FMT

There are no more uses of NIPQUAD or NIPQUAD_FMT.  Remove the definitions.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha/h8300/m68k: remove obsolete <asm/md.h> files

Removal of these started in 2.3.43pre3, ca. 10 years ago.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/compiler-gcc.h: use __same_type() in __must_be_array()

We should use the __same_type() helper in __must_be_array().

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

asm-generic/io.h: add big endian versions of io{read,write}{16,32}

The asm-generic/iomap.h provides these functions already, but the
non-generic fallback defines do not.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch/um/kernel/ptrace.c: remove dead PROC_MM

PROC_MM doesn't exist in Kconfig.  Looking around it looks like a
left-over from 2.6.0 or even 2.4 times, last mentioned in a fedora patch
for 2.6.10.  I believe it's time to get rid of that last tiny parts here
that are still around.

Signed-off-by: Christoph Egger <siccegge@cs.fau.de>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

um: call free_irq() only on enabled channels

When I use OpenSUSE-11.2 on UML (> 2.6.25)
I get lots of such errors:

Registering fd 1 twice
Irqs : 3, 3
Ids : 0x09cb41a0, 0x09cb4120
------------[ cut here ]------------
WARNING: at kernel/irq/manage.c:896 __free_irq+0x79/0x11a()
Trying to free already-free IRQ 3
Modules linked in:
09dadc6c:  [<081b2edb>] dump_stack+0x1c/0x20
09dadc84:  [<080716da>] warn_slowpath_common+0x49/0x77
09dadc9c:  [<08071772>] warn_slowpath_fmt+0x26/0x2a
09dadcb4:  [<08094e08>] __free_irq+0x79/0x11a
09dadce4:  [<08094ed6>] free_irq+0x2d/0x49
09dadcf4:  [<0805b4bc>] close_one_chan+0x70/0x9c
09dadd0c:  [<0805b833>] close_chan+0x17/0x22
09dadd1c:  [<0805bdda>] enable_chan+0x70/0x7c
09dadd3c:  [<0805cbb7>] line_open+0x34/0x9f
09dadd54:  [<0805b21e>] con_open+0x13/0x35
09dadd6c:  [<0814dc89>] tty_open+0x285/0x384
09dadda0:  [<080b754e>] chrdev_open+0xe0/0xf9
09daddc0:  [<080b3fb2>] __dentry_open+0xf3/0x1e2
09dadde4:  [<080b4142>] nameidata_to_filp+0x35/0x49
09daddfc:  [<080bd270>] do_last+0x409/0x50e
09dade28:  [<080bea04>] do_filp_open+0x175/0x446
09dadecc:  [<080b3d89>] do_sys_open+0x4a/0x128
09dadf04:  [<080b3ea2>] sys_open+0x19/0x21
09dadf28:  [<0805ab5a>] handle_syscall+0x7a/0x98
09dadf78:  [<08068441>] userspace+0x2c9/0x370
09dadfe0:  [<08058bb3>] fork_handler+0x53/0x5b
09dadffc:  [<00766564>] 0x766564

---[ end trace 9ebc1094aaf4bded ]---

This patch fixes the issue.

Signed-off-by: Richard Weinberger <richard@nod.at>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuidle: extend cpuidle and menu governor to handle dynamic states

On some SoC chips, HW resources may be in use during any particular idle
period.  As a consequence, the cpuidle states that the SoC is safe to
enter can change from idle period to idle period.  In addition, the
latency and threshold of each cpuidle state can vary, depending on the
operating condition when the CPU becomes idle, e.g.  the current cpu
frequency, the current state of the HW blocks, etc.

cpuidle core and the menu governor, in the current form, are geared
towards cpuidle states that are static, i.e.  the availabiltiy of the
states, their latencies, their thresholds are non-changing during run
time.  cpuidle does not provide any hook that cpuidle drivers can use to
adjust those values on the fly for the current idle period before the menu
governor selects the target cpuidle state.

This patch extends cpuidle core and the menu governor to handle states
that are dynamic.  There are three additions in the patch and the patch
maintains backwards-compatibility with existing cpuidle drivers.

1) add prepare() to struct cpuidle_device.  A cpuidle driver can hook
   into the callback and cpuidle will call prepare() before calling the
   governor's select function.  The callback gives the cpuidle driver a
   chance to update the dynamic information of the cpuidle states for the
   current idle period, e.g.  state availability, latencies, thresholds,
   power values, etc.

2) add CPUIDLE_FLAG_IGNORE as one of the state flags.  In the prepare()
   function, a cpuidle driver can set/clear the flag to indicate to the
   menu governor whether a cpuidle state should be ignored, i.e.  not
   available, during the current idle period.

3) add power_specified bit to struct cpuidle_device.  The menu governor
   currently assumes that the cpuidle states are arranged in the order of
   increasing latency, threshold, and power savings.  This is true or can
   be made true for static states.  Once the state parameters are dynamic,
   the latencies, thresholds, and power savings for the cpuidle states can
   increase or decrease by different amounts from idle period to idle
   period.  So the assumption of increasing latency, threshold, and power
   savings from Cn to C(n+1) can no longer be guaranteed.

It can be straightforward to calculate the power consumption of each
available state and to specify it in power_usage for the idle period.
Using the power_usage fields, the menu governor then selects the state
that has the lowest power consumption and that still satisfies all other
critieria.  The power_specified bit defaults to 0.  For existing cpuidle
drivers, cpuidle detects that power_specified is 0 and fills in a dummy
set of power_usage values.

Signed-off-by: Ai Li <aili@codeaurora.org>
Cc: Len Brown <len.brown@intel.com>
Acked-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hibernation: freeze swap at hibernation

When taking a memory snapshot in hibernate_snapshot(), all (directly
called) memory allocations use GFP_ATOMIC.  Hence swap misusage during
hibernation never occurs.

But from a pessimistic point of view, there is no guarantee that no page
allcation has __GFP_WAIT.  It is better to have a global indication "we
enter hibernation, don't use swap!".

This patch tries to freeze new-swap-allocation during hibernation.  (All
user processes are frozenm so swapin is not a concern).

This way, no updates will happen to swap_map[] between
hibernate_snapshot() and save_image().  Swap is thawed when swsusp_free()
is called.  We can be assured that swap corruption will not occur.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Hugh Dickins <hughd@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Ondrej Zary <linux@rainbow-software.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix corruption of hibernation caused by reusing swap during image saving

Since 2.6.31, swap_map[]'s refcounting was changed to show that a used
swap entry is just for swap-cache, can be reused.  Then, while scanning
free entry in swap_map[], a swap entry may be able to be reclaimed and
reused.  It was caused by commit c9e444103b5e7a5 ("mm: reuse unused swap
entry if necessary").

But this caused deta corruption at resume. The scenario is

- Assume a clean-swap cache, but mapped.

- at hibernation_snapshot[], clean-swap-cache is saved as
  clean-swap-cache and swap_map[] is marked as SWAP_HAS_CACHE.

- then, save_image() is called.  And reuse SWAP_HAS_CACHE entry to save
  image, and break the contents.

After resume:

- the memory reclaim runs and finds clean-not-referenced-swap-cache and
  discards it because it's marked as clean.  But here, the contents on
  disk and swap-cache is inconsistent.

Hance memory is corrupted.

This patch avoids the bug by not reclaiming swap-entry during hibernation.
This is a quick fix for backporting.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Reported-by: Ondreg Zary <linux@rainbow-software.org>
Tested-by: Ondreg Zary <linux@rainbow-software.org>
Tested-by: Andrea Gelmini <andrea.gelmini@gmail.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: implement HW performance events on the EV67 and later CPUs

This implements hardware performance events for the EV67 and later CPUs
within the Linux performance events subsystem.  Only using the performance
monitoring unit in HP/Compaq's so called "Aggregrate mode" is supported.

The code has been implemented in a manner that makes extension to other
older Alpha CPUs relatively straightforward should some mug wish to
indulge themselves.

Signed-off-by: Michael Cree <mcree@orcon.net.nz>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jay Estabrook <jay.estabrook@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: add wrperfmon.h header file to aid use of wrperfmon PALcall

Signed-off-by: Michael Cree <mcree@orcon.net.nz>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jay Estabrook <jay.estabrook@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: add performance monitor interrupt counter

The following patches implement hardware performance events for the Alpha
EV67 and later CPUs.  I have had this running on a Compaq XP1000 (EV67,
single CPU) for a few days now.  Pretty cool -- discovered that the glibc
exp2() library routine uses on average 985 cycles to execute 777 CPU
instructions whereas Compaq's CPML library version of exp2() uses on
average 32 cycles to execute 47 CPU instructions to achieve the same
thing!

This patch:

Add performance monitor interrupt counternd and export the count to user
space via /proc/interrupts.

Signed-off-by: Michael Cree <mcree@orcon.net.nz>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jay Estabrook <jay.estabrook@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: avoid resetting wb_start after each writeback round

WB_SYNC_NONE writeback is done in rounds of 1024 pages so that we don't
write out some huge inode for too long while starving writeout of other
inodes.  To avoid livelocks, we record time we started writeback in
wbc->wb_start and do not write out inodes which were dirtied after this
time.  But currently, writeback_inodes_wb() resets wb_start each time it
is called thus effectively invalidating this logic and making any
WB_SYNC_NONE writeback prone to livelocks.

This patch makes sure wb_start is set only once when we start writeback.

Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Acked-by: Jens Axboe <jaxboe@fusionio.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: cleanup for mm_slots_hash

Use compile-allocated memory instead of dynamic allocated memory for
mm_slots_hash.

Use hash_ptr() instead divisions for bucket calculation.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Izik Eidus <ieidus@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: raise the bar to PAGEOUT_IO_SYNC stalls

Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.

http://lkml.org/lkml/2010/4/4/86

In the above thread, Andreas Mohr described that

Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).

This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device

OOM also happens in Andreas' system.  The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path.  One in
do_fork() and the other two in unix_stream_sendmsg().  They are blocked on
this condition:

(sc->order && priority < DEF_PRIORITY - 2)

which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago.  That condition may be too
permissive.  In Andreas' case, 512MB/1024 = 512KB.  If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.

It's a severe problem in three ways.

Firstly, it can easily happen in daily desktop usage.  vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure.  Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges.  For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists.  And order-1 to order-3 allocations are more than common with
SLUB.  Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches.  For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.

Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system.  This "one slow USB device stalls the whole system"
avalanching effect is very bad.

Thirdly, once stalled, the stall time could be intolerable long for the
users.  When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.

So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size.  As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages.  We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).

The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes.  So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.

Andreas tested an older version of this patch and reported that it mostly
fixed his problem.  Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.

Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.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/vmalloc.c: check kmalloc() return value

kmalloc() may fail, if so return -ENOMEM.

Signed-off-by: Kulikov Vasiliy <segooon@gmail.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: add mm_vmscan_memcg_isolate tracepoint

Memcg also need to trace page isolation information as global reclaim.
This patch does it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: convert mm_vmscan_lru_isolate to DEFINE_EVENT

Mel Gorman recently added some vmscan tracepoints.  Unfortunately they are
covered only global reclaim.  But we want to trace memcg reclaim too.

Thus, this patch convert them to DEFINE_TRACE macro.  it help to reuse
tracepoint definition for other similar usage (i.e.  memcg).  This patch
have no functionally change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.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, vmscan: add memcg reclaim tracepoint

Memcg also need to trace reclaim progress as direct reclaim.  This patch
add it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: convert direct reclaim tracepoint to DEFINE_TRACE

Mel Gorman recently added some vmscan tracepoints.  Unfortunately they are
covered only global reclaim.  But we want to trace memcg reclaim too.

Thus, this patch convert them to DEFINE_TRACE macro.  it help to reuse
tracepoint definition for other similar usage (i.e.  memcg).  This patch
have no functionally change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: shrink_slab() requires the number of lru_pages, not the page order

Presently shrink_slab() has the following scanning equation.

                            lru_scanned        max_pass
  basic_scan_objects = 4 x -------------  x -----------------------------
                            lru_pages        shrinker->seeks (default:2)

  scan_objects = min(basic_scan_objects, max_pass * 2)

If we pass very small value as lru_pages instead real number of lru pages,
shrink_slab() drop much objects rather than necessary.  And now,
__zone_reclaim() pass 'order' as lru_pages by mistake.  That produces a
bad result.

For example, if we receive very low memory pressure (scan = 32, order =
0), shrink_slab() via zone_reclaim() always drop _all_ icache/dcache
objects.  (see above equation, very small lru_pages make very big
scan_objects result).

This patch fixes it.

[akpm@linux-foundation.org: fix layout, typos]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.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>

mmu-notifiers: remove mmu notifier calls in apply_to_page_range()

It is not appropriate for apply_to_page_range() to directly call any mmu
notifiers, because it is a general purpose function whose effect depends
on what context it is called in and what the callback function does.

In particular, if it is being used as part of an mmu notifier
implementation, the recursive calls can be particularly problematic.

It is up to apply_to_page_range's caller to do any notifier calls if
necessary.  It does not affect any in-tree users because they all operate
on init_mm, and mmu notifiers only pertain to usermode mappings.

[stefano.stabellini@eu.citrix.com: remove unused local `start']
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Cc: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: protect reading of reclaim_stat with lru_lock

Rik van Riel pointed out reading reclaim_stat should be protected
lru_lock, otherwise vmscan might sweep 2x much pages.

This fault was introduced by

  commit 4f98a2fee8acdb4ac84545df98cccecfd130f8db
  Author: Rik van Riel <riel@redhat.com>
  Date:   Sat Oct 18 20:26:32 2008 -0700

    vmscan: split LRU lists into anon & file sets

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: avoid subtraction of unsigned types

'slab_reclaimable' and 'nr_pages' are unsigned.  Subtraction is unsafe
because negative results would be misinterpreted.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.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>

drivers/base/node.c: reduce stack usage of node_read_meminfo()

drivers/base/node.c: In function 'node_read_meminfo':
drivers/base/node.c:139: warning: the frame size of 848 bytes is
larger than 512 bytes

Fix it by splitting the sprintf() into three parts.  It has no functional
change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: set VM_FAULT_WRITE in do_swap_page()

Set the flag if do_swap_page is decowing the page the same way do_wp_page
would too.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: add exclusive page to private anon_vma on swapin

On swapin it is fairly common for a page to be owned exclusively by one
process.  In that case we want to add the page to the anon_vma of that
process's VMA, instead of to the root anon_vma.

This will reduce the amount of rmap searching that the swapout code needs
to do.

Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@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>

oom: deprecate oom_adj tunable

/proc/pid/oom_adj is now deprecated so that that it may eventually be
removed.  The target date for removal is August 2012.

A warning will be printed to the kernel log if a task attempts to use this
interface.  Future warning will be suppressed until the kernel is rebooted
to prevent spamming the kernel log.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: badness heuristic rewrite

This a complete rewrite of the oom killer's badness() heuristic which is
used to determine which task to kill in oom conditions.  The goal is to
make it as simple and predictable as possible so the results are better
understood and we end up killing the task which will lead to the most
memory freeing while still respecting the fine-tuning from userspace.

Instead of basing the heuristic on mm->total_vm for each task, the task's
rss and swap space is used instead.  This is a better indication of the
amount of memory that will be freeable if the oom killed task is chosen
and subsequently exits.  This helps specifically in cases where KDE or
GNOME is chosen for oom kill on desktop systems instead of a memory
hogging task.

The baseline for the heuristic is a proportion of memory that each task is
currently using in memory plus swap compared to the amount of "allowable"
memory.  "Allowable," in this sense, means the system-wide resources for
unconstrained oom conditions, the set of mempolicy nodes, the mems
attached to current's cpuset, or a memory controller's limit.  The
proportion is given on a scale of 0 (never kill) to 1000 (always kill),
roughly meaning that if a task has a badness() score of 500 that the task
consumes approximately 50% of allowable memory resident in RAM or in swap
space.

The proportion is always relative to the amount of "allowable" memory and
not the total amount of RAM systemwide so that mempolicies and cpusets may
operate in isolation; they shall not need to know the true size of the
machine on which they are running if they are bound to a specific set of
nodes or mems, respectively.

Root tasks are given 3% extra memory just like __vm_enough_memory()
provides in LSMs.  In the event of two tasks consuming similar amounts of
memory, it is generally better to save root's task.

Because of the change in the badness() heuristic's baseline, it is also
necessary to introduce a new user interface to tune it.  It's not possible
to redefine the meaning of /proc/pid/oom_adj with a new scale since the
ABI cannot be changed for backward compatability.  Instead, a new tunable,
/proc/pid/oom_score_adj, is added that ranges from -1000 to +1000.  It may
be used to polarize the heuristic such that certain tasks are never
considered for oom kill while others may always be considered.  The value
is added directly into the badness() score so a value of -500, for
example, means to discount 50% of its memory consumption in comparison to
other tasks either on the system, bound to the mempolicy, in the cpuset,
or sharing the same memory controller.

/proc/pid/oom_adj is changed so that its meaning is rescaled into the
units used by /proc/pid/oom_score_adj, and vice versa.  Changing one of
these per-task tunables will rescale the value of the other to an
equivalent meaning.  Although /proc/pid/oom_adj was originally defined as
a bitshift on the badness score, it now shares the same linear growth as
/proc/pid/oom_score_adj but with different granularity.  This is required
so the ABI is not broken with userspace applications and allows oom_adj to
be deprecated for future removal.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: move badness() declaration into oom.h

Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: multi threaded process coredump don't make deadlock

Oleg pointed out current PF_EXITING check is wrong. Because PF_EXITING
is per-thread flag, not per-process flag. He said,

   Two threads, group-leader L and its sub-thread T. T dumps the code.
   In this case both threads have ->mm != NULL, L has PF_EXITING.

   The first problem is, select_bad_process() always return -1 in this
   case (even if the caller is T, this doesn't matter).

   The second problem is that we should add TIF_MEMDIE to T, not L.

I think we can remove this dubious PF_EXITING check. but as first step,
This patch add the protection of multi threaded issue.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: 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>

oom: give the dying task a higher priority

In a system under heavy load it was observed that even after the
oom-killer selects a task to die, the task may take a long time to die.

Right after sending a SIGKILL to the task selected by the oom-killer this
task has its priority increased so that it can exit() soon, freeing
memory.  That is accomplished by:

        /*
         * We give our sacrificial lamb high priority and access to
         * all the memory it needs. That way it should be able to
         * exit() and clear out its resources quickly...
         */
  p->rt.time_slice = HZ;
  set_tsk_thread_flag(p, TIF_MEMDIE);

It sounds plausible giving the dying task an even higher priority to be
sure it will be scheduled sooner and free the desired memory.  It was
suggested on LKML using SCHED_FIFO:1, the lowest RT priority so that this
task won't interfere with any running RT task.

If the dying task is already an RT task, leave it untouched.  Another good
suggestion, implemented here, was to avoid boosting the dying task
priority in case of mem_cgroup OOM.

Signed-off-by: Luis Claudio R. Goncalves <lclaudio@uudg.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: remove child->mm check from oom_kill_process()

The current "child->mm == p->mm" check prevents selection of vfork()ed
task.  But we don't have any reason to don't consider vfork().

Removed.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: cleanup has_intersects_mems_allowed()

presently has_intersects_mems_allowed() has own thread iterate logic, but
it should use while_each_thread().

It slightly improve the code readability.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: move OOM_DISABLE check from oom_kill_task to out_of_memory()

Presently if oom_kill_allocating_task is enabled and current have
OOM_DISABLED, following printk in oom_kill_process is called twice.

    pr_err("%s: Kill process %d (%s) score %lu or sacrifice child\n",
            message, task_pid_nr(p), p->comm, points);

So, OOM_DISABLE check should be more early.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: kill duplicate OOM_DISABLE check

select_bad_process() and badness() have the same OOM_DISABLE check.  This
patch kills one.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: /proc/<pid>/oom_score treat kernel thread honestly

If a kernel thread is using use_mm(), badness() returns a positive value.
This is not a big issue because caller take care of it correctly.  But
there is one exception, /proc/<pid>/oom_score calls badness() directly and
doesn't care that the task is a regular process.

Another example, /proc/1/oom_score return !0 value.  But it's unkillable.
This incorrectness makes administration a little confusing.

This patch fixes it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: oom_kill_process() needs to check that p is unkillable

When oom_kill_allocating_task is enabled, an argument task of
oom_kill_process is not selected by select_bad_process(), It's just
out_of_memory() caller task.  It mean the task can be unkillable.  check
it first.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: make oom_unkillable_task() helper function

Presently we have the same task check in two places. Unify it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: oom_kill_process() doesn't select kthread child

Presently select_bad_process() has a PF_KTHREAD check, but
oom_kill_process doesn't.  It mean oom_kill_process() may choose wrong
task, especially, when the child are using use_mm().

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

oom: don't try to kill oom_unkillable child

Presently, badness() doesn't care about either CPUSET nor mempolicy.  Then
if the victim child process have disjoint nodemask, OOM Killer might kill
innocent process.

This patch fixes it.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.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>

vmscan: update isolated page counters outside of main path in shrink_inactive_list()

When shrink_inactive_list() isolates pages, it updates a number of
counters using temporary variables to gather them.  These consume stack
and it's in the main path that calls ->writepage().  This patch moves the
accounting updates outside of the main path to reduce stack usage.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: set up pagevec as late as possible in shrink_page_list()

shrink_page_list() sets up a pagevec to release pages as according as they
are free.  It uses significant amounts of stack on the pagevec.  This
patch adds pages to be freed via pagevec to a linked list which is then
freed en-masse at the end.  This avoids using stack in the main path that
potentially calls writepage().

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: set up pagevec as late as possible in shrink_inactive_list()

shrink_inactive_list() sets up a pagevec to release unfreeable pages.  It
uses significant amounts of stack doing this.  This patch splits
shrink_inactive_list() to take the stack usage out of the main path so
that callers to writepage() do not contain an unused pagevec on the stack.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: remove unnecessary temporary vars in do_try_to_free_pages

Remove temporary variable that is only used once and does not help clarify
code.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: simplify shrink_inactive_list()

Now, max_scan of shrink_inactive_list() is always passed less than
SWAP_CLUSTER_MAX.  then, we can remove scanning pages loop in it.  This
patch also help stack diet.

detail
 - remove "while (nr_scanned < max_scan)" loop
 - remove nr_freed (now, we use nr_reclaimed directly)
 - remove nr_scan (now, we use nr_scanned directly)
 - rename max_scan to nr_to_scan
 - pass nr_to_scan into isolate_pages() directly instead
   using SWAP_CLUSTER_MAX

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: kill prev_priority completely

Since 2.6.28 zone->prev_priority is unused. Then it can be removed
safely. It reduce stack usage slightly.

Now I have to say that I'm sorry. 2 years ago, I thought prev_priority
can be integrate again, it's useful. but four (or more) times trying
haven't got good performance number. Thus I give up such approach.

The rest of this changelog is notes on prev_priority and why it existed in
the first place and why it might be not necessary any more. This information
is based heavily on discussions between Andrew Morton, Rik van Riel and
Kosaki Motohiro who is heavily quotes from.

Historically prev_priority was important because it determined when the VM
would start unmapping PTE pages. i.e. there are no balances of note within
the VM, Anon vs File and Mapped vs Unmapped. Without prev_priority, there
is a potential risk of unnecessarily increasing minor faults as a large
amount of read activity of use-once pages could push mapped pages to the
end of the LRU and get unmapped.

There is no proof this is still a problem but currently it is not considered
to be. Active files are not deactivated if the active file list is smaller
than the inactive list reducing the liklihood that file-mapped pages are
being pushed off the LRU and referenced executable pages are kept on the
active list to avoid them getting pushed out by read activity.

Even if it is a problem, prev_priority prev_priority wouldn't works
nowadays. First of all, current vmscan still a lot of UP centric code. it
expose some weakness on some dozens CPUs machine. I think we need more and
more improvement.

The problem is, current vmscan mix up per-system-pressure, per-zone-pressure
and per-task-pressure a bit. example, prev_priority try to boost priority to
other concurrent priority. but if the another task have mempolicy restriction,
it is unnecessary, but also makes wrong big latency and exceeding reclaim.
per-task based priority + prev_priority adjustment make the emulation of
per-system pressure. but it have two issue 1) too rough and brutal emulation
2) we need per-zone pressure, not per-system.

Another example, currently DEF_PRIORITY is 12. it mean the lru rotate about
2 cycle (1/4096 + 1/2048 + 1/1024 + .. + 1) before invoking OOM-Killer.
but if 10,0000 thrreads enter DEF_PRIORITY reclaim at the same time, the
system have higher memory pressure than priority==0 (1/4096*10,000 > 2).
prev_priority can't solve such multithreads workload issue. In other word,
prev_priority concept assume the sysmtem don't have lots threads."

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: tracing: add a postprocessing script for reclaim-related ftrace events

Add a simple post-processing script for the reclaim-related trace events.
It can be used to give an indication of how much traffic there is on the
LRU lists and how severe latencies due to reclaim are.  Example output
looks like the following

Reclaim latencies expressed as order-latency_in_ms
uname-3942             9-200.179000000004 9-98.7900000000373 9-99.8330000001006
kswapd0-311            0-662.097999999998 0-2.79700000002049 \
0-149.100000000035 0-3295.73600000003 0-9806.31799999997 0-35528.833 \
0-10043.197 0-129740.979 0-3.50500000000466 0-3.54899999999907 \
0-9297.78999999992 0-3.48499999998603 0-3596.97999999998 0-3.92799999995623 \
0-3.35000000009313 0-16729.017 0-3.57799999997951 0-47435.0630000001 \
0-3.7819999998901 0-5864.06999999995 0-18635.334 0-10541.289 9-186011.565 \
9-3680.86300000001 9-1379.06499999994 9-958571.115 9-66215.474 \
9-6721.14699999988 9-1962.15299999993 9-1094806.125 9-2267.83199999994 \
9-47120.9029999999 9-427653.886 9-2.6359999999404 9-632.148999999976 \
9-476.753000000026 9-495.577000000048 9-8.45900000003166 9-6.6820000000298 \
9-1.30500000016764 9-251.746000000043 9-383.905000000028 9-80.1419999999925 \
9-281.160000000149 9-14.8780000000261 9-381.45299999998 9-512.07799999998 \
9-49.5519999999087 9-167.439000000013 9-183.820999999996 9-239.527999999933 \
9-19.9479999998584 9-148.747999999905 9-164.583000000101 9-16.9480000000913 \
9-192.376000000164 9-64.1010000000242 9-1.40800000005402 9-3.60800000000745 \
9-17.1359999999404 9-4.69500000006519 9-2.06400000001304 9-1582488.554 \
9-6244.19499999983 9-348153.812 9-2.0999999998603 9-0.987999999895692 \
0-32218.473 0-1.6140000000596 0-1.28100000019185 0-1.41300000017509 \
0-1.32299999985844 0-602.584000000032 0-1.34400000004098 0-1.6929999999702 \
1-22101.8190000001 9-174876.724 9-16.2420000000857 9-175.165999999736 \
9-15.8589999997057 9-0.604999999981374 9-3061.09000000032 9-479.277000000235 \
9-1.54499999992549 9-771.985000000335 9-4.88700000010431 9-15.0649999999441 \
9-0.879999999888241 9-252.01500000013 9-1381.03600000031 9-545.689999999944 \
9-3438.0129999998 9-3343.70099999988
bench-stresshig-3942   9-7063.33900000004 9-129960.482 9-2062.27500000002 \
9-3845.59399999992 9-171.82799999998 9-16493.821 9-7615.23900000006 \
9-10217.848 9-983.138000000035 9-2698.39999999991 9-4016.1540000001 \
9-5522.37700000009 9-21630.429 \
9-15061.048 9-10327.953 9-542.69700000016 9-317.652000000002 \
9-8554.71699999995 9-1786.61599999992 9-1899.31499999994 9-2093.41899999999 \
9-4992.62400000007 9-942.648999999976 9-1923.98300000001 9-3.7980000001844 \
9-5.99899999983609 9-0.912000000011176 9-1603.67700000014 9-1.98300000000745 \
9-3.96500000008382 9-0.902999999932945 9-2802.72199999983 9-1078.24799999991 \
9-2155.82900000014 9-10.058999999892 9-1984.723 9-1687.97999999998 \
9-1136.05300000007 9-3183.61699999985 9-458.731000000145 9-6.48600000003353 \
9-1013.25200000009 9-8415.22799999989 9-10065.584 9-2076.79600000009 \
9-3792.65699999989 9-71.2010000001173 9-2560.96999999997 9-2260.68400000012 \
9-2862.65799999982 9-1255.81500000018 9-15.7440000001807 9-4.33499999996275 \
9-1446.63800000004 9-238.635000000009 9-60.1790000000037 9-4.38800000003539 \
9-639.567000000039 9-306.698000000091 9-31.4070000001229 9-74.997999999905 \
9-632.725999999791 9-1625.93200000003 9-931.266000000061 9-98.7749999999069 \
9-984.606999999844 9-225.638999999966 9-421.316000000108 9-653.744999999879 \
9-572.804000000004 9-769.158999999985 9-603.918000000063 9-4.28499999991618 \
9-626.21399999992 9-1721.25 9-0.854999999981374 9-572.39599999995 \
9-681.881999999983 9-1345.12599999993 9-363.666999999899 9-3823.31099999999 \
9-2991.28200000012 9-4.27099999994971 9-309.76500000013 9-3068.35700000008 \
9-788.25 9-3515.73999999999 9-2065.96100000013 9-286.719999999972 \
9-316.076000000117 9-344.151000000071 9-2.51000000000931 9-306.688000000082 \
9-1515.00099999993 9-336.528999999864 9-793.491999999853 9-457.348999999929 \
9-13620.155 9-119.933999999892 9-35.0670000000391 9-918.266999999993 \
9-828.569000000134 9-4863.81099999999 9-105.222000000067 9-894.23900000006 \
9-110.964999999851 9-0.662999999942258 9-12753.3150000002 9-12.6129999998957 \
9-13368.0899999999 9-12.4199999999255 9-1.00300000002608 9-1.41100000008009 \
9-10300.5290000001 9-16.502000000095 9-30.7949999999255 9-6283.0140000002 \
9-4320.53799999994 9-6826.27300000004 9-3.07299999985844 9-1497.26799999992 \
9-13.4040000000969 9-3.12999999988824 9-3.86100000003353 9-11.3539999998175 \
9-0.10799999977462 9-21.780999999959 9-209.695999999996 9-299.647000000114 \
9-6.01699999999255 9-20.8349999999627 9-22.5470000000205 9-5470.16800000006 \
9-7.60499999998137 9-0.821000000229105 9-1.56600000010803 9-14.1669999998994 \
9-0.209000000031665 9-1.82300000009127 9-1.70000000018626 9-19.9429999999702 \
9-124.266999999993 9-0.0389999998733401 9-6.71400000015274 9-16.7710000001825 \
9-31.0409999999683 9-0.516999999992549 9-115.888000000035 9-5.19900000002235 \
9-222.389999999898 9-11.2739999999758 9-80.9050000000279 9-8.14500000001863 \
9-4.44599999999627 9-0.218999999808148 9-0.715000000083819 9-0.233000000007451
\
9-48.2630000000354 9-248.560999999987 9-374.96800000011 9-644.179000000004 \
9-0.835999999893829 9-79.0060000000522 9-128.447999999858 9-0.692000000039116 \
9-5.26500000013039 9-128.449000000022 9-2.04799999995157 9-12.0990000001621 \
9-8.39899999997579 9-10.3860000001732 9-11.9310000000987 9-53.4450000000652 \
9-0.46999999997206 9-2.96299999998882 9-17.9699999999721 9-0.776000000070781 \
9-25.2919999998994 9-33.1110000000335 9-0.434000000124797 9-0.641000000061467 \
9-0.505000000121072 9-1.12800000002608 9-149.222000000067 9-1.17599999997765 \
9-3247.33100000001 9-10.7439999999478 9-153.523000000045 9-1.38300000014715 \
9-794.762000000104 9-3.36199999996461 9-128.765999999829 9-181.543999999994 \
9-78149.8229999999 9-176.496999999974 9-89.9940000001807 9-9.12700000009499 \
9-250.827000000048 9-0.224999999860302 9-0.388999999966472 9-1.16700000036508 \
9-32.1740000001155 9-12.6800000001676 9-0.0720000001601875 9-0.274999999906868
\
9-0.724000000394881 9-266.866000000387 9-45.5709999999963 9-4.54399999976158 \
9-8.27199999988079 9-4.38099999958649 9-0.512000000104308 9-0.0640000002458692
\
9-5.20000000018626 9-0.0839999997988343 9-12.816000000108 9-0.503000000026077 \
9-0.507999999914318 9-6.23999999975786 9-3.35100000025705 9-18.8530000001192 \
9-25.2220000000671 9-68.2309999996796 9-98.9939999999478 9-0.441000000108033 \
9-4.24599999981001 9-261.702000000048 9-3.01599999982864 9-0.0749999997206032 \
9-0.0370000000111759 9-4.375 9-3.21800000034273 9-11.3960000001825 \
9-0.0540000000037253 9-0.286000000312924 9-0.865999999921769 \
9-0.294999999925494 9-6.45999999996275 9-4.31099999975413 9-128.248999999836 \
9-0.282999999821186 9-102.155000000261 9-0.0860000001266599 \
9-0.0540000000037253 9-0.935000000055879 9-0.0670000002719462 \
9-5.8640000000596 9-19.9860000000335 9-4.18699999991804 9-0.566000000108033 \
9-2.55099999997765 9-0.702000000048429 9-131.653999999631 9-0.638999999966472 \
9-14.3229999998584 9-183.398000000045 9-178.095999999903 9-3.22899999981746 \
9-7.31399999978021 9-22.2400000002235 9-11.7979999999516 9-108.10599999968 \
9-99.0159999998286 9-102.640999999829 9-38.414000000339
Process                  Direct     Wokeup      Pages      Pages    Pages
details                   Rclms     Kswapd    Scanned    Sync-IO ASync-IO
cc1-30800                     0          1          0          0        0      wakeup-0=1
cc1-24260                     0          1          0          0        0      wakeup-0=1
cc1-24152                     0         12          0          0        0      wakeup-0=12
cc1-8139                      0          1          0          0        0      wakeup-0=1
cc1-4390                      0          1          0          0        0      wakeup-0=1
cc1-4648                      0          7          0          0        0      wakeup-0=7
cc1-4552                      0          3          0          0        0      wakeup-0=3
dd-4550                       0         31          0          0        0      wakeup-0=31
date-4898                     0          1          0          0        0      wakeup-0=1
cc1-6549                      0          7          0          0        0      wakeup-0=7
as-22202                      0         17          0          0        0      wakeup-0=17
cc1-6495                      0          9          0          0        0      wakeup-0=9
cc1-8299                      0          1          0          0        0      wakeup-0=1
cc1-6009                      0          1          0          0        0      wakeup-0=1
cc1-2574                      0          2          0          0        0      wakeup-0=2
cc1-30568                     0          1          0          0        0      wakeup-0=1
cc1-2679                      0          6          0          0        0      wakeup-0=6
sh-13747                      0         12          0          0        0      wakeup-0=12
cc1-22193                     0         18          0          0        0      wakeup-0=18
cc1-30725                     0          2          0          0        0      wakeup-0=2
as-4392                       0          2          0          0        0      wakeup-0=2
cc1-28180                     0         14          0          0        0      wakeup-0=14
cc1-13697                     0          2          0          0        0      wakeup-0=2
cc1-22207                     0          8          0          0        0      wakeup-0=8
cc1-15270                     0        179          0          0        0      wakeup-0=179
cc1-22011                     0         82          0          0        0      wakeup-0=82
cp-14682                      0          1          0          0        0      wakeup-0=1
as-11926                      0          2          0          0        0      wakeup-0=2
cc1-6016                      0          5          0          0        0      wakeup-0=5
make-18554                    0         13          0          0        0      wakeup-0=13
cc1-8292                      0         12          0          0        0      wakeup-0=12
make-24381                    0          1          0          0        0      wakeup-1=1
date-18681                    0         33          0          0        0      wakeup-0=33
cc1-32276                     0          1          0          0        0      wakeup-0=1
timestamp-outpu-2809          0        253          0          0        0      wakeup-0=240 wakeup-1=13
date-18624                    0          7          0          0        0      wakeup-0=7
cc1-30960                     0          9          0          0        0      wakeup-0=9
cc1-4014                      0          1          0          0        0      wakeup-0=1
cc1-30706                     0         22          0          0        0      wakeup-0=22
uname-3942                    4          1        306          0       17      direct-9=4       wakeup-9=1
cc1-28207                     0          1          0          0        0      wakeup-0=1
cc1-30563                     0          9          0          0        0      wakeup-0=9
cc1-22214                     0         10          0          0        0      wakeup-0=10
cc1-28221                     0         11          0          0        0      wakeup-0=11
cc1-28123                     0          6          0          0        0      wakeup-0=6
kswapd0-311                   0          7     357302          0    34233      wakeup-0=7
cc1-5988                      0          7          0          0        0      wakeup-0=7
as-30734                      0        161          0          0        0      wakeup-0=161
cc1-22004                     0         45          0          0        0      wakeup-0=45
date-4590                     0          4          0          0        0      wakeup-0=4
cc1-15279                     0        213          0          0        0      wakeup-0=213
date-30735                    0          1          0          0        0      wakeup-0=1
cc1-30583                     0          4          0          0        0      wakeup-0=4
cc1-32324                     0          2          0          0        0      wakeup-0=2
cc1-23933                     0          3          0          0        0      wakeup-0=3
cc1-22001                     0         36          0          0        0      wakeup-0=36
bench-stresshig-3942        287        287      80186       6295    12196      direct-9=287       wakeup-9=287
cc1-28170                     0          7          0          0        0      wakeup-0=7
date-7932                     0         92          0          0        0      wakeup-0=92
cc1-22222                     0          6          0          0        0      wakeup-0=6
cc1-32334                     0         16          0          0        0      wakeup-0=16
cc1-2690                      0          6          0          0        0      wakeup-0=6
cc1-30733                     0          9          0          0        0      wakeup-0=9
cc1-32298                     0          2          0          0        0      wakeup-0=2
cc1-13743                     0         18          0          0        0      wakeup-0=18
cc1-22186                     0          4          0          0        0      wakeup-0=4
cc1-28214                     0         11          0          0        0      wakeup-0=11
cc1-13735                     0          1          0          0        0      wakeup-0=1
updatedb-8173                 0         18          0          0        0      wakeup-0=18
cc1-13750                     0          3          0          0        0      wakeup-0=3
cat-2808                      0          2          0          0        0      wakeup-0=2
cc1-15277                     0        169          0          0        0      wakeup-0=169
date-18317                    0          1          0          0        0      wakeup-0=1
cc1-15274                     0        197          0          0        0      wakeup-0=197
cc1-30732                     0          1          0          0        0      wakeup-0=1

Kswapd                   Kswapd      Order      Pages      Pages    Pages
Instance                Wakeups  Re-wakeup    Scanned    Sync-IO ASync-IO
kswapd0-311                  91         24     357302          0    34233      wake-0=31 wake-1=1 wake-9=59       rewake-0=10 rewake-1=1 rewake-9=13

Summary
Direct reclaims:      291
Direct reclaim pages scanned: 437794
Direct reclaim write sync I/O: 6295
Direct reclaim write async I/O: 46446
Wake kswapd requests: 2152
Time stalled direct reclaim:  519.163009000002 ms

Kswapd wakeups: 91
Kswapd pages scanned: 357302
Kswapd reclaim write sync I/O: 0
Kswapd reclaim write async I/O: 34233
Time kswapd awake: 5282.749757 ms

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: tracing: add trace event when a page is written

Add a trace event for when page reclaim queues a page for IO and records
whether it is synchronous or asynchronous.  Excessive synchronous IO for a
process can result in noticeable stalls during direct reclaim.  Excessive
IO from page reclaim may indicate that the system is seriously under
provisioned for the amount of dirty pages that exist.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: tracing: add trace events for LRU page isolation

Add an event for when pages are isolated en-masse from the LRU lists.
This event augments the information available on LRU traffic and can be
used to evaluate lumpy reclaim.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: tracing: add trace events for kswapd wakeup, sleeping and direct reclaim

Add two trace events for kswapd waking up and going asleep for the
purposes of tracking kswapd activity and two trace events for direct
reclaim beginning and ending.  The information can be used to work out how
much time a process or the system is spending on the reclamation of pages
and in the case of direct reclaim, how many pages were reclaimed for that
process.  High frequency triggering of these events could point to memory
pressure problems.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: recalculate lru_pages on each priority

shrink_zones() need relatively long time and lru_pages can change
dramatically during shrink_zones().  So lru_pages should be recalculated
for each priority.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.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>

vmscan: zone_reclaim don't call disable_swap_token()

Swap token don't works when zone reclaim is enabled since it was born.
Because __zone_reclaim() always call disable_swap_token() unconditionally.

This kill swap token feature completely.  As far as I know, nobody want to
that.  Remove it.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: implement writeback livelock avoidance using page tagging

We try to avoid livelocks of writeback when some steadily creates dirty
pages in a mapping we are writing out.  For memory-cleaning writeback,
using nr_to_write works reasonably well but we cannot really use it for
data integrity writeback.  This patch tries to solve the problem.

The idea is simple: Tag all pages that should be written back with a
special tag (TOWRITE) in the radix tree.  This can be done rather quickly
and thus livelocks should not happen in practice.  Then we start doing the
hard work of locking pages and sending them to disk only for those pages
that have TOWRITE tag set.

Note: Adding new radix tree tag grows radix tree node from 288 to 296
bytes for 32-bit archs and from 552 to 560 bytes for 64-bit archs.
However, the number of slab/slub items per page remains the same (13 and 7
respectively).

Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

radix-tree: omplement function radix_tree_range_tag_if_tagged

Implement function for setting one tag if another tag is set for each item
in given range.

Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: add anon_vma bug checks

Verify the refcounting doesn't go wrong, and resurrect the check in
__page_check_anon_rmap as in old anon-vma code.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: resurrect page_address_in_vma anon_vma check

With root anon-vma it's trivial to keep doing the usual check as in
old-anon-vma code.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: always use anon_vma root pointer

Always use anon_vma->root pointer instead of anon_vma_chain.prev.

Also optimize the map-paths, if a mapping is already established no need
to overwrite it with root anon-vma list, we can keep the more finegrined
anon-vma and skip the overwrite: see the PageAnon check in !exclusive
case.  This is also the optimization that hidden the ksm bug as this tends
to make ksm_might_need_to_copy skip the copy, but only the proper fix to
ksm_might_need_to_copy guarantees not triggering the ksm bug unless ksm is
in use.  this is an optimization only...

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
[kamezawa.hiroyu@jp.fujitsu.com: fix false positive BUG_ON in __page_set_anon_rmap]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: fix ksm swapin time optimization

The new anon-vma code, was suboptimal and it lead to erratic invocation of
ksm_does_need_to_copy.  That leads to host hangs or guest vnc lockup, or
weird behavior.  It's unclear why ksm_does_need_to_copy is unstable but
the point is that when KSM is not in use, ksm_does_need_to_copy must never
run or we bounce pages for no good reason.  I suspect the same hangs will
happen with KVM swaps.  But this at least fixes the regression in the
new-anon-vma code and it only let KSM bugs triggers when KSM is in use.

The code in do_swap_page likely doesn't cope well with a not-swapcache,
especially the memcg code.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: Izik Eidus <ieidus@yahoo.com>
Cc: Avi Kivity <avi@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: always add new vmas at the end

Make sure to always add new VMAs at the end of the list.  This is
important so rmap_walk does not miss a VMA that was created during the
rmap_walk.

The old code got this right most of the time due to luck, but was buggy
when anon_vma_prepare reused a mergeable anon_vma.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmap: remove unnecessary lock from __vma_link

There's no anon-vma related mangling happening inside __vma_link anymore
so no need of anon_vma locking there.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: reduce pagefault lock contention

I'm running a shmem pagefault test case (see attached file) under a 64 CPU
system.  Profile shows shmem_inode_info->lock is heavily contented and
100% CPUs time are trying to get the lock.  In the pagefault (no swap)
case, shmem_getpage gets the lock twice, the last one is avoidable if we
prealloc a page so we could reduce one time of locking.  This is what
below patch does.

The result of the test case:
2.6.35-rc3: ~20s
2.6.35-rc3 + patch: ~12s
so this is 40% improvement.

One might argue if we could have better locking for shmem.  But even shmem
is lockless, the pagefault will soon have pagecache lock heavily contented
because shmem must add new page to pagecache.  So before we have better
locking for pagecache, improving shmem locking doesn't have too much
improvement.  I did a similar pagefault test against a ramfs file, the
test result is ~10.5s.

[akpm@linux-foundation.org: fix comment, clean up code layout, elimintate code duplication]
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Zhang, Yanmin" <yanmin.zhang@intel.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: make tmpfs scalable with percpu_counter for used blocks

The current implementation of tmpfs is not scalable.  We found that
stat_lock is contended by multiple threads when we need to get a new page,
leading to useless spinning inside this spin lock.

This patch makes use of the percpu_counter library to maintain local count
of used blocks to speed up getting and returning of pages.  So the
acquisition of stat_lock is unnecessary for getting and returning blocks,
improving the performance of tmpfs on system with large number of cpus.
On a 4 socket 32 core NHM-EX system, we saw improvement of 270%.

The implementation below has a slight chance of race between threads
causing a slight overshoot of the maximum configured blocks.  However, any
overshoot is small, and is bounded by the number of cpus.  This happens
when the number of used blocks is slightly below the maximum configured
blocks when a thread checks the used block count, and another thread
allocates the last block before the current thread does.  This should not
be a problem for tmpfs, as the overshoot is most likely to be a few blocks
and bounded.  If a strict limit is really desired, then configured the max
blocks to be the limit less the number of cpus in system.

Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: add accurate compare function to percpu_counter library

Add percpu_counter_compare that allows for a quick but accurate comparison
of percpu_counter with a given value.

A rough count is provided by the count field in percpu_counter structure,
without accounting for the other values stored in individual cpu counters.

The actual count is a sum of count and the cpu counters.  However, count
field is never different from the actual value by a factor of
batch*num_online_cpu.  We do not need to get actual count for comparison
if count is different from the given value by this factor and allows for
quick comparison without summing up all the per cpu counters.

Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gcc-4.6: mm: fix unused but set warnings

No real bugs, just some dead code and some fixups.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gcc-4.6: pagemap: avoid unused-but-set variable

Avoid quite a lot of warnings in header files in a gcc 4.6 -Wall builds

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

avr32: invoke oom-killer from page fault

As explained in commit 1c0fe6e3bd ("mm: invoke oom-killer from page
fault") , we want to call the architecture independent oom killer when
getting an unexplained OOM from handle_mm_fault, rather than simply
killing current.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.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>

mempolicy: reduce stack size of migrate_pages()

migrate_pages() is using >500 bytes stack. Reduce it.

   mm/mempolicy.c: In function 'sys_migrate_pages':
   mm/mempolicy.c:1344: warning: the frame size of 528 bytes is larger than 512 bytes

[akpm@linux-foundation.org: don't play with a might-be-NULL pointer]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

topology: alternate fix for ia64 tiger_defconfig build breakage

Define stubs for the numa_*_id() generic percpu related functions for
non-NUMA configurations in <asm-generic/topology.h> where the other
non-numa stubs live.

Fixes ia64 !NUMA build breakage -- e.g., tiger_defconfig

Back out now unneeded '#ifndef CONFIG_NUMA' guards from ia64 smpboot.c

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Acked-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmzone.h: remove dead prototype

get_zone_counts() was dropped from kernel tree, see:
http://www.mail-archive.com/mm-commits@vger.kernel.org/msg07313.html but
its prototype remains.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use for_each_online_cpu() in vmstat

The sum_vm_events passes cpumask for for_each_cpu().  But it's useless
since we have for_each_online_cpu.  Althougth it's tirival overhead, it's
not good about coding consistency.

Let's use for_each_online_cpu instead of for_each_cpu with cpumask
argument.

Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Reviewed-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>

oom: fold __out_of_memory into out_of_memory

__out_of_memory() only has a single caller, so fold it into
out_of_memory() and add a comment about locking for its call to
oom_kill_process().

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>

oom: remove constraint argument from select_bad_process and __out_of_memory

select_bad_process() and __out_of_memory() doe not need their enum
oom_constraint arguments: it's possible to pass a NULL nodemask if
constraint == CONSTRAINT_MEMORY_POLICY in the caller, out_of_memory().

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: rename try_set_zone_oom() to try_set_zonelist_oom()

We have been used naming try_set_zone_oom and clear_zonelist_oom.
The role of functions is to lock of zonelist for preventing parallel
OOM. So clear_zonelist_oom makes sense but try_set_zone_oome is rather
awkward and unmatched with clear_zonelist_oom.

Let's change it with try_set_zonelist_oom.

Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: 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>

oom: remove unnecessary code and cleanup

Remove the redundancy in __oom_kill_task() since:

 - init can never be passed to this function: it will never be PF_EXITING
   or selectable from select_bad_process(), and

 - it will never be passed a task from oom_kill_task() without an ->mm
   and we're unconcerned about detachment from exiting tasks, there's no
   reason to protect them against SIGKILL or access to memory reserves.

Also moves the kernel log message to a higher level since the verbosity is
not always emitted here; we need not print an error message if an exiting
task is given a longer timeslice.

__oom_kill_task() only has a single caller, so it can be merged into that
function at the same time.

Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: 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>

oom: move sysctl declarations to oom.h

The three oom killer sysctl variables (sysctl_oom_dump_tasks,
sysctl_oom_kill_allocating_task, and sysctl_panic_on_oom) are better
declared in include/linux/oom.h rather than kernel/sysctl.c.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: 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>

oom: remove special handling for pagefault ooms

It is possible to remove the special pagefault oom handler by simply oom
locking all system zones and then calling directly into out_of_memory().

All populated zones must have ZONE_OOM_LOCKED set, otherwise there is a
parallel oom killing in progress that will lead to eventual memory freeing
so it's not necessary to needlessly kill another task.  The context in
which the pagefault is allocating memory is unknown to the oom killer, so
this is done on a system-wide level.

If a task has already been oom killed and hasn't fully exited yet, this
will be a no-op since select_bad_process() recognizes tasks across the
system with TIF_MEMDIE set.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: 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>

oom: extract panic helper function

There are various points in the oom killer where the kernel must determine
whether to panic or not.  It's better to extract this to a helper function
to remove all the confusion as to its semantics.

Also fix a call to dump_header() where tasklist_lock is not read- locked,
as required.

There's no functional change with this patch.

Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: 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>

oom: avoid oom killer for lowmem allocations

If memory has been depleted in lowmem zones even with the protection
afforded to it by /proc/sys/vm/lowmem_reserve_ratio, it is unlikely that
killing current users will help.  The memory is either reclaimable (or
migratable) already, in which case we should not invoke the oom killer at
all, or it is pinned by an application for I/O.  Killing such an
application may leave the hardware in an unspecified state and there is no
guarantee that it will be able to make a timely exit.

Lowmem allocations are now failed in oom conditions when __GFP_NOFAIL is
not used so that the task can perhaps recover or try again later.

Previously, the heuristic provided some protection for those tasks with
CAP_SYS_RAWIO, but this is no longer necessary since we will not be
killing tasks for the purposes of ISA allocations.

high_zoneidx is gfp_zone(gfp_flags), meaning that ZONE_NORMAL will be the
default for all allocations that are not __GFP_DMA, __GFP_DMA32,
__GFP_HIGHMEM, and __GFP_MOVABLE on kernels configured to support those
flags.  Testing for high_zoneidx being less than ZONE_NORMAL will only
return true for allocations that have either __GFP_DMA or __GFP_DMA32.

Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: 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>

oom: enable oom tasklist dump by default

The oom killer tasklist dump, enabled with the oom_dump_tasks sysctl, is
very helpful information in diagnosing why a user's task has been killed.
It emits useful information such as each eligible thread's memory usage
that can determine why the system is oom, so it should be enabled by
default.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: 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>

oom: select task from tasklist for mempolicy ooms

The oom killer presently kills current whenever there is no more memory
free or reclaimable on its mempolicy's nodes.  There is no guarantee that
current is a memory-hogging task or that killing it will free any
substantial amount of memory, however.

In such situations, it is better to scan the tasklist for nodes that are
allowed to allocate on current's set of nodes and kill the task with the
highest badness() score.  This ensures that the most memory-hogging task,
or the one configured by the user with /proc/pid/oom_adj, is always
selected in such scenarios.

Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: 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>

oom: sacrifice child with highest badness score for parent

When a task is chosen for oom kill, the oom killer first attempts to
sacrifice a child not sharing its parent's memory instead.  Unfortunately,
this often kills in a seemingly random fashion based on the ordering of
the selected task's child list.  Additionally, it is not guaranteed at all
to free a large amount of memory that we need to prevent additional oom
killing in the very near future.

Instead, we now only attempt to sacrifice the worst child not sharing its
parent's memory, if one exists.  The worst child is indicated with the
highest badness() score.  This serves two advantages: we kill a
memory-hogging task more often, and we allow the configurable
/proc/pid/oom_adj value to be considered as a factor in which child to
kill.

Reviewers may observe that the previous implementation would iterate
through the children and attempt to kill each until one was successful and
then the parent if none were found while the new code simply kills the
most memory-hogging task or the parent.  Note that the only time
oom_kill_task() fails, however, is when a child does not have an mm or has
a /proc/pid/oom_adj of OOM_DISABLE.  badness() returns 0 for both cases,
so the final oom_kill_task() will always succeed.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-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>

oom: filter tasks not sharing the same cpuset

Tasks that do not share the same set of allowed nodes with the task that
triggered the oom should not be considered as candidates for oom kill.

Tasks in other cpusets with a disjoint set of mems would be unfairly
penalized otherwise because of oom conditions elsewhere; an extreme
example could unfairly kill all other applications on the system if a
single task in a user's cpuset sets itself to OOM_DISABLE and then uses
more memory than allowed.

Killing tasks outside of current's cpuset rarely would free memory for
current anyway.  To use a sane heuristic, we must ensure that killing a
task would likely free memory for current and avoid needlessly killing
others at all costs just because their potential memory freeing is
unknown.  It is better to kill current than another task needlessly.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-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>

oom: avoid sending exiting tasks a SIGKILL

It's unnecessary to SIGKILL a task that is already PF_EXITING and can
actually cause a NULL pointer dereference of the sighand if it has already
been detached.  Instead, simply set TIF_MEMDIE so it has access to memory
reserves and can quickly exit as the comment implies.

Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: 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>

oom: give current access to memory reserves if it has been killed

It's possible to livelock the page allocator if a thread has mm->mmap_sem
and fails to make forward progress because the oom killer selects another
thread sharing the same ->mm to kill that cannot exit until the semaphore
is dropped.

The oom killer will not kill multiple tasks at the same time; each oom
killed task must exit before another task may be killed.  Thus, if one
thread is holding mm->mmap_sem and cannot allocate memory, all threads
sharing the same ->mm are blocked from exiting as well.  In the oom kill
case, that means the thread holding mm->mmap_sem will never free
additional memory since it cannot get access to memory reserves and the
thread that depends on it with access to memory reserves cannot exit
because it cannot acquire the semaphore.  Thus, the page allocators
livelocks.

When the oom killer is called and current happens to have a pending
SIGKILL, this patch automatically gives it access to memory reserves and
returns.  Upon returning to the page allocator, its allocation will
hopefully succeed so it can quickly exit and free its memory.  If not, the
page allocator will fail the allocation if it is not __GFP_NOFAIL.

Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: 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>

oom: dump_tasks use find_lock_task_mm too fix

When find_lock_task_mm() returns a thread other than p in dump_tasks(),
its name should be displayed instead.  This is the thread that will be
targeted by the oom killer, not its mm-less parent.

This also allows us to safely dereference task->comm without needing
get_task_comm().

While we're here, remove the cast on task_cpu(task) as Andrew suggested.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.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>

oom: improve commentary in dump_tasks()

The comments in dump_tasks() should be updated to be more clear about why
tasks are filtered and how they are filtered by its argument.

An unnecessary comment concerning a check for is_global_init() is removed
since it isn't of importance.

Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.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>

oom: dump_tasks use find_lock_task_mm too

dump_task() should use find_lock_task_mm() too. It is necessary for
protecting task-exiting race.

dump_tasks() currently filters any task that does not have an attached
->mm since it incorrectly assumes that it must either be in the process of
exiting and has detached its memory or that it's a kernel thread;
multithreaded tasks may actually have subthreads that have a valid ->mm
pointer and thus those threads should actually be displayed.  This change
finds those threads, if they exist, and emit their information along with
the rest of the candidate tasks for kill.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Balbir Singh <balbir@in.ibm.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>

oom: introduce find_lock_task_mm() to fix !mm false positives

Almost all ->mm == NULL checks in oom_kill.c are wrong.

The current code assumes that the task without ->mm has already released
its memory and ignores the process.  However this is not necessarily true
when this process is multithreaded, other live sub-threads can use this
->mm.

- Remove the "if (!p->mm)" check in select_bad_process(), it is
  just wrong.

- Add the new helper, find_lock_task_mm(), which finds the live
  thread which uses the memory and takes task_lock() to pin ->mm

- change oom_badness() to use this helper instead of just checking
  ->mm != NULL.

- As David pointed out, select_bad_process() must never choose the
  task without ->mm, but no matter what oom_badness() returns the
  task can be chosen if nothing else has been found yet.

  Change oom_badness() to return int, change it to return -1 if
  find_lock_task_mm() fails, and change select_bad_process() to
  check points >= 0.

Note! This patch is not enough, we need more changes.

- oom_badness() was fixed, but oom_kill_task() still ignores
  the task without ->mm

- oom_forkbomb_penalty() should use find_lock_task_mm() too,
  and it also needs other changes to actually find the first
  first-descendant children

This will be addressed later.

[kosaki.motohiro@jp.fujitsu.com: use in badness(), __oom_kill_task()]
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-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>

oom: PF_EXITING check should take mm into account

select_bad_process() checks PF_EXITING to detect the task which is going
to release its memory, but the logic is very wrong.

- a single process P with the dead group leader disables
  select_bad_process() completely, it will always return
  ERR_PTR() while P can live forever

- if the PF_EXITING task has already released its ->mm
  it doesn't make sense to expect it is goiing to free
  more memory (except task_struct/etc)

Change the code to ignore the PF_EXITING tasks without ->mm.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Balbir Singh <balbir@in.ibm.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>

oom: check PF_KTHREAD instead of !mm to skip kthreads

select_bad_process() thinks a kernel thread can't have ->mm != NULL, this
is not true due to use_mm().

Change the code to check PF_KTHREAD.

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

buffer_head: remove redundant test from wait_on_buffer

The comment suggests that when b_count equals zero it is calling
__wait_no_buffer to trigger some debug, but as there is no debug in
__wait_on_buffer the whole thing is redundant.

AFAICT from the git log this has been the case for at least 5 years, so
it seems safe just to remove this.

Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: extend KSM refcounts to the anon_vma root

KSM reference counts can cause an anon_vma to exist after the processe it
belongs to have already exited.  Because the anon_vma lock now lives in
the root anon_vma, we need to ensure that the root anon_vma stays around
until after all the "child" anon_vmas have been freed.

The obvious way to do this is to have a "child" anon_vma take a reference
to the root in anon_vma_fork.  When the anon_vma is freed at munmap or
process exit, we drop the refcount in anon_vma_unlink and possibly free
the root anon_vma.

The KSM anon_vma reference count function also needs to be modified to
deal with the possibility of freeing 2 levels of anon_vma.  The easiest
way to do this is to break out the KSM magic and make it generic.

When compiling without CONFIG_KSM, this code is compiled out.

Signed-off-by: Rik van Riel <riel@redhat.com>
Tested-by: Larry Woodman <lwoodman@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Tested-by: Dave Young <hidave.darkstar@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: always lock the root (oldest) anon_vma

Always (and only) lock the root (oldest) anon_vma whenever we do something
in an anon_vma.  The recently introduced anon_vma scalability is due to
the rmap code scanning only the VMAs that need to be scanned.  Many common
operations still took the anon_vma lock on the root anon_vma, so always
taking that lock is not expected to introduce any scalability issues.

However, always taking the same lock does mean we only need to take one
lock, which means rmap_walk on pages from any anon_vma in the vma is
excluded from occurring during an munmap, expand_stack or other operation
that needs to exclude rmap_walk and similar functions.

Also add the proper locking to vma_adjust.

Signed-off-by: Rik van Riel <riel@redhat.com>
Tested-by: Larry Woodman <lwoodman@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: track the root (oldest) anon_vma

Track the root (oldest) anon_vma in each anon_vma tree.  Because we only
take the lock on the root anon_vma, we cannot use the lock on higher-up
anon_vmas to lock anything.  This makes it impossible to do an indirect
lookup of the root anon_vma, since the data structures could go away from
under us.

However, a direct pointer is safe because the root anon_vma is always the
last one that gets freed on munmap or exit, by virtue of the same_vma list
order and unlink_anon_vmas walking the list forward.

[akpm@linux-foundation.org: fix typo]
Signed-off-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Tested-by: Larry Woodman <lwoodman@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>