GitHub/mt8127/android_kernel_alcatel_ttab.git
12 years agoradix_tree: take radix_tree_path off stack
Hugh Dickins [Fri, 13 Jan 2012 01:20:41 +0000 (17:20 -0800)]
radix_tree: take radix_tree_path off stack

Down, down in the deepest depths of GFP_NOIO page reclaim, we have
shrink_page_list() calling __remove_mapping() calling __delete_from_
swap_cache() or __delete_from_page_cache().

You would not expect those to need much stack, but in fact they call
radix_tree_delete(): which declares a 192-byte radix_tree_path array on
its stack (to record the node,offsets it visits when descending, in case
it needs to ascend to update them).  And if any tag is still set [1],
that calls radix_tree_tag_clear(), which declares a further such
192-byte radix_tree_path array on the stack.  (At least we have
interrupts disabled here, so won't then be pushing registers too.)

That was probably a good choice when most users were 32-bit (array of
half the size), and adding fields to radix_tree_node would have bloated
it unnecessarily.  But nowadays many are 64-bit, and each
radix_tree_node contains a struct rcu_head, which is only used when
freeing; whereas the radix_tree_path info is only used for updating the
tree (deleting, clearing tags or setting tags if tagged) when a lock
must be held, of no interest when accessing the tree locklessly.

So add a parent pointer to the radix_tree_node, in union with the
rcu_head, and remove all uses of the radix_tree_path.  There would be
space in that union to save the offset when descending as before (we can
argue that a lock must already be held to exclude other users), but
recalculating it when ascending is both easy (a constant shift and a
constant mask) and uncommon, so it seems better just to do that.

Two little optimizations: no need to decrement height when descending,
adjusting shift is enough; and once radix_tree_tag_if_tagged() has set
tag on a node and its ancestors, it need not ascend from that node
again.

perf on the radix tree test harness reports radix_tree_insert() as 2%
slower (now having to set parent), but radix_tree_delete() 24% faster.
Surely that's an exaggeration from rtth's artificially low map shift 3,
but forcing it back to 6 still rates radix_tree_delete() 8% faster.

[1] Can a pagecache tag (dirty, writeback or towrite) actually still be
set at the time of radix_tree_delete()? Perhaps not if the filesystem is
well-behaved.  But although I've not tracked any stack overflow down to
this cause, I have observed a curious case in which a dirty tag is set
and left set on tmpfs: page migration's migrate_page_copy() happens to
use __set_page_dirty_nobuffers() to set PageDirty on the newpage, and
that sets PAGECACHE_TAG_DIRTY as a side-effect - harmless to a
filesystem which doesn't use tags, except for this stack depth issue.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Nai Xia <nai.xia@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoradix_tree: remove radix_tree_indirect_to_ptr()
Xiao Guangrong [Fri, 13 Jan 2012 01:20:39 +0000 (17:20 -0800)]
radix_tree: remove radix_tree_indirect_to_ptr()

It is not used anymore, remove it

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agodio: optimize cache misses in the submission path
Andi Kleen [Fri, 13 Jan 2012 01:20:35 +0000 (17:20 -0800)]
dio: optimize cache misses in the submission path

Some investigation of a transaction processing workload showed that a
major consumer of cycles in __blockdev_direct_IO is the cache miss while
accessing the block size.  This is because it has to walk the chain from
block_dev to gendisk to queue.

The block size is needed early on to check alignment and sizes.  It's only
done if the check for the inode block size fails.  But the costly block
device state is unconditionally fetched.

- Reorganize the code to only fetch block dev state when actually
  needed.

Then do a prefetch on the block dev early on in the direct IO path.  This
is worth it, because there is substantial code run before we actually
touch the block dev now.

- I also added some unlikelies to make it clear the compiler that block
  device fetch code is not normally executed.

This gave a small, but measurable improvement on a large database
benchmark (about 0.3%)

[akpm@linux-foundation.org: coding-style fixes]
[sfr@canb.auug.org.au: using prefetch requires including prefetch.h]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agovfs: cache request_queue in struct block_device
Andi Kleen [Fri, 13 Jan 2012 01:20:34 +0000 (17:20 -0800)]
vfs: cache request_queue in struct block_device

This makes it possible to get from the inode to the request_queue with one
less cache miss.  Used in followon optimization.

The livetime of the pointer is the same as the gendisk.

This assumes that the queue will always stay the same in the gendisk while
it's visible to block_devices.  I think that's safe correct?

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agofs/direct-io.c: calculate fs_count correctly in get_more_blocks()
Tao Ma [Fri, 13 Jan 2012 01:20:33 +0000 (17:20 -0800)]
fs/direct-io.c: calculate fs_count correctly in get_more_blocks()

In get_more_blocks(), we use dio_count to calcuate fs_count and do some
tricky things to increase fs_count if dio_count isn't aligned.  But
actually it still has some corner cases that can't be coverd.  See the
following example:

dio_write foo -s 1024 -w 4096

(direct write 4096 bytes at offset 1024).  The same goes if the offset
isn't aligned to fs_blocksize.

In this case, the old calculation counts fs_count to be 1, but actually we
will write into 2 different blocks (if fs_blocksize=4096).  The old code
just works, since it will call get_block twice (and may have to allocate
and create extents twice for filesystems like ext4).  So we'd better call
get_block just once with the proper fs_count.

Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agodrivers/parport/parport_pc.c: fix warnings
Andrew Morton [Fri, 13 Jan 2012 01:20:32 +0000 (17:20 -0800)]
drivers/parport/parport_pc.c: fix warnings

drivers/parport/parport_pc.c: In function '__check_irq':
drivers/parport/parport_pc.c:3415: warning: return from incompatible pointer type
drivers/parport/parport_pc.c: In function '__check_dma':
drivers/parport/parport_pc.c:3417: warning: return from incompatible pointer type

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agopanic: don't print redundant backtraces on oops
Andi Kleen [Fri, 13 Jan 2012 01:20:30 +0000 (17:20 -0800)]
panic: don't print redundant backtraces on oops

When an oops causes a panic and panic prints another backtrace it's pretty
common to have the original oops data be scrolled away on a 80x50 screen.

The second backtrace is quite redundant and not needed anyways.

So don't print the panic backtrace when oops_in_progress is true.

[akpm@linux-foundation.org: add comment]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agosysctl: add the kernel.ns_last_pid control
Pavel Emelyanov [Fri, 13 Jan 2012 01:20:27 +0000 (17:20 -0800)]
sysctl: add the kernel.ns_last_pid control

The sysctl works on the current task's pid namespace, getting and setting
its last_pid field.

Writing is allowed for CAP_SYS_ADMIN-capable tasks thus making it possible
to create a task with desired pid value.  This ability is required badly
for the checkpoint/restore in userspace.

This approach suits all the parties for now.

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokdump: add udev events for memory online/offline
Michael Holzheu [Fri, 13 Jan 2012 01:20:23 +0000 (17:20 -0800)]
kdump: add udev events for memory online/offline

Currently no udev events for memory hotplug "online" and "offline" are
generated:

  # udevadm monitor
  # echo offline > /sys/devices/system/memory/memory4/state
  ==> No event

When kdump is loaded, kexec detects the current memory configuration and
stores it in the pre-allocated ELF core header.  Therefore, for kdump it
is necessary to reload the kdump kernel with kexec when the memory
configuration changes (e.g.  for online/offline hotplug memory).

In order to do this automatically, udev rules should be used.  This kernel
patch adds udev events for "online" and "offline".  Together with this
kernel patch, the following udev rules for online/offline have to be added
to "/etc/udev/rules.d/98-kexec.rules":

  SUBSYSTEM=="memory", ACTION=="online", PROGRAM="/etc/init.d/kdump restart"
  SUBSYSTEM=="memory", ACTION=="offline", PROGRAM="/etc/init.d/kdump restart"

[sfr@canb.auug.org.au: fixups for class to subsystem conversion]
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoinclude/linux/crash_dump.h needs elf.h
Fabio Estevam [Fri, 13 Jan 2012 01:20:20 +0000 (17:20 -0800)]
include/linux/crash_dump.h needs elf.h

Building an ARM target we get the following warnings:

  CC      arch/arm/kernel/setup.o
  In file included from arch/arm/kernel/setup.c:39:
  arch/arm/include/asm/elf.h:102:1: warning: "vmcore_elf64_check_arch" redefined
  In file included from arch/arm/kernel/setup.c:24:
  include/linux/crash_dump.h:30:1: warning: this is the location of the previous definition

Quoting Russell King:

"linux/crash_dump.h makes no attempt to include asm/elf.h, but it depends
on stuff in asm/elf.h to determine how stuff inside this file is defined
at parse time.

So, if asm/elf.h is included after linux/crash_dump.h or not at all, you
get a different result from the situation where asm/elf.h is included
before."

So add elf.h header to crash_dump.h to avoid this problem.

The original discussion about this can be found at:
http://www.spinics.net/lists/arm-kernel/msg154113.html

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: <stable@vger.kernel.org> [3.2.1]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokdump: fix crash_kexec()/smp_send_stop() race in panic()
Michael Holzheu [Fri, 13 Jan 2012 01:20:18 +0000 (17:20 -0800)]
kdump: fix crash_kexec()/smp_send_stop() race in panic()

When two CPUs call panic at the same time there is a possible race
condition that can stop kdump.  The first CPU calls crash_kexec() and the
second CPU calls smp_send_stop() in panic() before crash_kexec() finished
on the first CPU.  So the second CPU stops the first CPU and therefore
kdump fails:

1st CPU:
  panic()->crash_kexec()->mutex_trylock(&kexec_mutex)-> do kdump

2nd CPU:
  panic()->crash_kexec()->kexec_mutex already held by 1st CPU
       ->smp_send_stop()-> stop 1st CPU (stop kdump)

This patch fixes the problem by introducing a spinlock in panic that
allows only one CPU to process crash_kexec() and the subsequent panic
code.

All other CPUs call the weak function panic_smp_self_stop() that stops the
CPU itself.  This function can be overloaded by architecture code.  For
example "tile" can use their lower-power "nap" instruction for that.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokdump: crashk_res init check for /sys/kernel/kexec_crash_size
Michael Holzheu [Fri, 13 Jan 2012 01:20:15 +0000 (17:20 -0800)]
kdump: crashk_res init check for /sys/kernel/kexec_crash_size

Currently it is possible to set the crash_size via the sysfs
/sys/kernel/kexec_crash_size even if no crash kernel memory has been
defined with the "crashkernel" parameter.  In this case "crashk_res" is
not initialized and crashk_res.start = crashk_res.end = 0.  Unfortunately
resource_size(&crashk_res) returns 1 in this case.  This breaks the s390
implementation of crash_(un)map_reserved_pages().

To fix the problem the correct "old_size" is now calculated in
crash_shrink_memory().  "old_size is set to "0" if crashk_res is not
initialized.  With this change crash_shrink_memory() will do nothing, when
"crashk_res" is not initialized.  It will return "0" for "echo 0 >
/sys/kernel/kexec_crash_size" and -EINVAL for "echo [not zero] >
/sys/kernel/kexec_crash_size".

In addition to that this patch also simplifies the "ret = -EINVAL" vs.
"ret = 0" logic as suggested by Simon Horman.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Reviewed-by: Dave Young <dyoung@redhat.com>
Reviewed-by: WANG Cong <xiyou.wangcong@gmail.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokdump: add missing RAM resource in crash_shrink_memory()
Michael Holzheu [Fri, 13 Jan 2012 01:20:14 +0000 (17:20 -0800)]
kdump: add missing RAM resource in crash_shrink_memory()

When shrinking crashkernel memory using /sys/kernel/kexec_crash_size for
the newly added memory no RAM resource is created at the moment.

Example:

  $ cat /proc/iomem
  00000000-bfffffff : System RAM
    00000000-005b7ac3 : Kernel code
    005b7ac4-009743bf : Kernel data
    009bb000-00a85c33 : Kernel bss
  c0000000-cfffffff : Crash kernel
  d0000000-ffffffff : System RAM

  $ echo 0 > /sys/kernel/kexec_crash_size
  $ cat /proc/iomem
  00000000-bfffffff : System RAM
    00000000-005b7ac3 : Kernel code
    005b7ac4-009743bf : Kernel data
    009bb000-00a85c33 : Kernel bss
                                   <<-- here is System RAM missing
  d0000000-ffffffff : System RAM

One result of this bug is that the memory chunk can never be set offline
using memory hotplug.  With this patch I insert a new "System RAM"
resource for the released memory.  Then the upper example looks like the
following:

  $ echo 0 > /sys/kernel/kexec_crash_size
  $ cat /proc/iomem
  00000000-bfffffff : System RAM
    00000000-005b7ac3 : Kernel code
    005b7ac4-009743bf : Kernel data
    009bb000-00a85c33 : Kernel bss
  c0000000-cfffffff : System RAM   <<-- new rescoure
  d0000000-ffffffff : System RAM

And now I can set chunk c0000000-cfffffff offline.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokexec: remove KMSG_DUMP_KEXEC
WANG Cong [Fri, 13 Jan 2012 01:20:11 +0000 (17:20 -0800)]
kexec: remove KMSG_DUMP_KEXEC

KMSG_DUMP_KEXEC is useless because we already save kernel messages inside
/proc/vmcore, and it is unsafe to allow modules to do other stuffs in a
crash dump scenario.

[akpm@linux-foundation.org: fix powerpc build]
Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
Reported-by: Vivek Goyal <vgoyal@redhat.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Acked-by: Jarod Wilson <jarod@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.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>
12 years agocpumask: update setup_node_to_cpumask_map() comments
Wanlong Gao [Fri, 13 Jan 2012 01:20:09 +0000 (17:20 -0800)]
cpumask: update setup_node_to_cpumask_map() comments

node_to_cpumask() has been replaced by cpumask_of_node(), and wholly
removed since commit 29c337a0 ("cpumask: remove obsolete node_to_cpumask
now everyone uses cpumask_of_node").

So update the comments for setup_node_to_cpumask_map().

Signed-off-by: Wanlong Gao <gaowanlong@cn.fujitsu.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm/vmalloc.c: eliminate extra loop in pcpu_get_vm_areas error path
Kautuk Consul [Fri, 13 Jan 2012 01:20:08 +0000 (17:20 -0800)]
mm/vmalloc.c: eliminate extra loop in pcpu_get_vm_areas error path

If either of the vas or vms arrays are not properly kzalloced, then the
code jumps to the err_free label.

The err_free label runs a loop to check and free each of the array members
of the vas and vms arrays which is not required for this situation as none
of the array members have been allocated till this point.

Eliminate the extra loop we have to go through by introducing a new label
err_free2 and then jumping to it.

[akpm@linux-foundation.org: remove now-unneeded tests]
Signed-off-by: Kautuk Consul <consul.kautuk@gmail.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>
12 years agomm: rearrange putback_inactive_pages
Hugh Dickins [Fri, 13 Jan 2012 01:20:07 +0000 (17:20 -0800)]
mm: rearrange putback_inactive_pages

There is sometimes confusion between the global putback_lru_pages() in
migrate.c and the static putback_lru_pages() in vmscan.c: rename the
latter putback_inactive_pages(): it helps shrink_inactive_list() rather as
move_active_pages_to_lru() helps shrink_active_list().

Remove unused scan_control arg from putback_inactive_pages() and from
update_isolated_counts().  Move clear_active_flags() inside
update_isolated_counts().  Move NR_ISOLATED accounting up into
shrink_inactive_list() itself, so the balance is clearer.

Do the spin_lock_irq() before calling putback_inactive_pages() and
spin_unlock_irq() after return from it, so that it better matches
update_isolated_counts() and move_active_pages_to_lru().

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: remove isolate_pages()
Hugh Dickins [Fri, 13 Jan 2012 01:20:06 +0000 (17:20 -0800)]
mm: remove isolate_pages()

The isolate_pages() level in vmscan.c offers little but indirection: merge
it into isolate_lru_pages() as the compiler does, and use the names
nr_to_scan and nr_scanned in each case.

Signed-off-by: Hugh Dickins <hughd@google.com>
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>
12 years agomm: remove del_page_from_lru, add page_off_lru
Hugh Dickins [Fri, 13 Jan 2012 01:20:04 +0000 (17:20 -0800)]
mm: remove del_page_from_lru, add page_off_lru

del_page_from_lru() repeats del_page_from_lru_list(), also working out
which LRU the page was on, clearing the relevant bits.  Decouple those
functions: remove del_page_from_lru() and add page_off_lru().

Signed-off-by: Hugh Dickins <hughd@google.com>
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>
12 years agomm: enum lru_list lru
Hugh Dickins [Fri, 13 Jan 2012 01:20:01 +0000 (17:20 -0800)]
mm: enum lru_list lru

Mostly we use "enum lru_list lru": change those few "l"s to "lru"s.

Signed-off-by: Hugh Dickins <hughd@google.com>
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>
12 years agomm: no blank line after EXPORT_SYMBOL in swap.c
Hugh Dickins [Fri, 13 Jan 2012 01:20:00 +0000 (17:20 -0800)]
mm: no blank line after EXPORT_SYMBOL in swap.c

checkpatch rightly protests

  WARNING: EXPORT_SYMBOL(foo); should immediately follow its function/variable

so fix the five offenders in mm/swap.c.

Signed-off-by: Hugh Dickins <hughd@google.com>
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>
12 years agomm: fewer underscores in ____pagevec_lru_add
Hugh Dickins [Fri, 13 Jan 2012 01:19:58 +0000 (17:19 -0800)]
mm: fewer underscores in ____pagevec_lru_add

What's so special about ____pagevec_lru_add() that it needs four leading
underscores?  Nothing, it just helped to distinguish from
__pagevec_lru_add() in 2.6.28 development.  Cut two leading underscores.

Signed-off-by: Hugh Dickins <hughd@google.com>
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>
12 years agomm: take pagevecs off reclaim stack
Hugh Dickins [Fri, 13 Jan 2012 01:19:56 +0000 (17:19 -0800)]
mm: take pagevecs off reclaim stack

Replace pagevecs in putback_lru_pages() and move_active_pages_to_lru()
by lists of pages_to_free: then apply Konstantin Khlebnikov's
free_hot_cold_page_list() to them instead of pagevec_release().

Which simplifies the flow (no need to drop and retake lock whenever
pagevec fills up) and reduces stale addresses in stack backtraces
(which often showed through the pagevecs); but more importantly,
removes another 120 bytes from the deepest stacks in page reclaim.
Although I've not recently seen an actual stack overflow here with
a vanilla kernel, move_active_pages_to_lru() has often featured in
deep backtraces.

However, free_hot_cold_page_list() does not handle compound pages
(nor need it: a Transparent HugePage would have been split by the
time it reaches the call in shrink_page_list()), but it is possible
for putback_lru_pages() or move_active_pages_to_lru() to be left
holding the last reference on a THP, so must exclude the unlikely
compound case before putting on pages_to_free.

Remove pagevec_strip(), its work now done in move_active_pages_to_lru().
The pagevec in scan_mapping_unevictable_pages() remains in mm/vmscan.c,
but that is never on the reclaim path, and cannot be replaced by a list.

Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: fix mem_cgroup_print_bad_page
Hugh Dickins [Fri, 13 Jan 2012 01:19:54 +0000 (17:19 -0800)]
memcg: fix mem_cgroup_print_bad_page

If DEBUG_VM, mem_cgroup_print_bad_page() is called whenever bad_page()
shows a "Bad page state" message, removes page from circulation, adds a
taint and continues.  This is at a very low level, often when a spinlock
is held (sometimes when page table lock is held, for example).

We want to recover from this badness, not make it worse: we must not
kmalloc memory here, we must not do a cgroup path lookup via dubious
pointers.  No doubt that code was useful to debug a particular case at one
time, and may be again, but take it out of the mainline kernel.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: fix split_huge_page_refcounts()
Hugh Dickins [Fri, 13 Jan 2012 01:19:52 +0000 (17:19 -0800)]
memcg: fix split_huge_page_refcounts()

This patch started off as a cleanup: __split_huge_page_refcounts() has to
cope with two scenarios, when the hugepage being split is already on LRU,
and when it is not; but why does it have to split that accounting across
three different sites?  Consolidate it in lru_add_page_tail(), handling
evictable and unevictable alike, and use standard add_page_to_lru_list()
when accounting is needed (when the head is not yet on LRU).

But a recent regression in -next, I guess the removal of PageCgroupAcctLRU
test from mem_cgroup_split_huge_fixup(), makes this now a necessary fix:
under load, the MEM_CGROUP_ZSTAT count was wrapping to a huge number,
messing up reclaim calculations and causing a freeze at rmdir of cgroup.

Add a VM_BUG_ON to mem_cgroup_lru_del_list() when we're about to wrap that
count - this has not been the only such incident.  Document that
lru_add_page_tail() is for Transparent HugePages by #ifdef around it.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: check if reclaim should really abort even if compaction_ready() is true...
Mel Gorman [Fri, 13 Jan 2012 01:19:49 +0000 (17:19 -0800)]
mm: vmscan: check if reclaim should really abort even if compaction_ready() is true for one zone

If compaction can proceed for a given zone, shrink_zones() does not
reclaim any more pages from it.  After commit [e0c2327: vmscan: abort
reclaim/compaction if compaction can proceed], do_try_to_free_pages()
tries to finish as soon as possible once one zone can compact.

This was intended to prevent slabs being shrunk unnecessarily but there
are side-effects.  One is that a small zone that is ready for compaction
will abort reclaim even if the chances of successfully allocating a THP
from that zone is small.  It also means that reclaim can return too early
even though sc->nr_to_reclaim pages were not reclaimed.

This partially reverts the commit until it is proven that slabs are really
being shrunk unnecessarily but preserves the check to return 1 to avoid
OOM if reclaim was aborted prematurely.

[aarcange@redhat.com: This patch replaces a revert from Andrea]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: when reclaiming for compaction, ensure there are sufficient free pages...
Mel Gorman [Fri, 13 Jan 2012 01:19:45 +0000 (17:19 -0800)]
mm: vmscan: when reclaiming for compaction, ensure there are sufficient free pages available

In commit e0887c19 ("vmscan: limit direct reclaim for higher order
allocations"), Rik noted that reclaim was too aggressive when THP was
enabled.  In his initial patch he used the number of free pages to decide
if reclaim should abort for compaction.  My feedback was that reclaim and
compaction should be using the same logic when deciding if reclaim should
be aborted.

Unfortunately, this had the effect of reducing THP success rates when the
workload included something like streaming reads that continually
allocated pages.  The window during which compaction could run and return
a THP was too small.

This patch combines Rik's two patches together.  compaction_suitable() is
still used to decide if reclaim should be aborted to allow compaction is
used.  However, it will also ensure that there is a reasonable buffer of
free pages available.  This improves upon the THP allocation success rates
but bounds the number of pages that are freed for compaction.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel<riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: compaction: introduce sync-light migration for use by compaction
Mel Gorman [Fri, 13 Jan 2012 01:19:43 +0000 (17:19 -0800)]
mm: compaction: introduce sync-light migration for use by compaction

This patch adds a lightweight sync migrate operation MIGRATE_SYNC_LIGHT
mode that avoids writing back pages to backing storage.  Async compaction
maps to MIGRATE_ASYNC while sync compaction maps to MIGRATE_SYNC_LIGHT.
For other migrate_pages users such as memory hotplug, MIGRATE_SYNC is
used.

This avoids sync compaction stalling for an excessive length of time,
particularly when copying files to a USB stick where there might be a
large number of dirty pages backed by a filesystem that does not support
->writepages.

[aarcange@redhat.com: This patch is heavily based on Andrea's work]
[akpm@linux-foundation.org: fix fs/nfs/write.c build]
[akpm@linux-foundation.org: fix fs/btrfs/disk-io.c build]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: page allocator: do not call direct reclaim for THP allocations while compaction...
Mel Gorman [Fri, 13 Jan 2012 01:19:41 +0000 (17:19 -0800)]
mm: page allocator: do not call direct reclaim for THP allocations while compaction is deferred

If compaction is deferred, direct reclaim is used to try to free enough
pages for the allocation to succeed.  For small high-orders, this has a
reasonable chance of success.  However, if the caller has specified
__GFP_NO_KSWAPD to limit the disruption to the system, it makes more sense
to fail the allocation rather than stall the caller in direct reclaim.
This patch skips direct reclaim if compaction is deferred and the caller
specifies __GFP_NO_KSWAPD.

Async compaction only considers a subset of pages so it is possible for
compaction to be deferred prematurely and not enter direct reclaim even in
cases where it should.  To compensate for this, this patch also defers
compaction only if sync compaction failed.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: Rik van Riel<riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: compaction: make isolate_lru_page() filter-aware again
Mel Gorman [Fri, 13 Jan 2012 01:19:38 +0000 (17:19 -0800)]
mm: compaction: make isolate_lru_page() filter-aware again

Commit 39deaf85 ("mm: compaction: make isolate_lru_page() filter-aware")
noted that compaction does not migrate dirty or writeback pages and that
is was meaningless to pick the page and re-add it to the LRU list.  This
had to be partially reverted because some dirty pages can be migrated by
compaction without blocking.

This patch updates "mm: compaction: make isolate_lru_page" by skipping
over pages that migration has no possibility of migrating to minimise LRU
disruption.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel<riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: compaction: determine if dirty pages can be migrated without blocking within...
Mel Gorman [Fri, 13 Jan 2012 01:19:34 +0000 (17:19 -0800)]
mm: compaction: determine if dirty pages can be migrated without blocking within ->migratepage

Asynchronous compaction is used when allocating transparent hugepages to
avoid blocking for long periods of time.  Due to reports of stalling,
there was a debate on disabling synchronous compaction but this severely
impacted allocation success rates.  Part of the reason was that many dirty
pages are skipped in asynchronous compaction by the following check;

if (PageDirty(page) && !sync &&
mapping->a_ops->migratepage != migrate_page)
rc = -EBUSY;

This skips over all mapping aops using buffer_migrate_page() even though
it is possible to migrate some of these pages without blocking.  This
patch updates the ->migratepage callback with a "sync" parameter.  It is
the responsibility of the callback to fail gracefully if migration would
block.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: do not OOM if aborting reclaim to start compaction
Mel Gorman [Fri, 13 Jan 2012 01:19:33 +0000 (17:19 -0800)]
mm: vmscan: do not OOM if aborting reclaim to start compaction

During direct reclaim it is possible that reclaim will be aborted so that
compaction can be attempted to satisfy a high-order allocation.  If this
decision is made before any pages are reclaimed, it is possible that 0 is
returned to the page allocator potentially triggering an OOM.  This has
not been observed but it is a possibility so this patch addresses it.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: check if we isolated a compound page during lumpy scan
Andrea Arcangeli [Fri, 13 Jan 2012 01:19:29 +0000 (17:19 -0800)]
mm: vmscan: check if we isolated a compound page during lumpy scan

Properly take into account if we isolated a compound page during the lumpy
scan in reclaim and skip over the tail pages when encountered.  This
corrects the values given to the tracepoint for number of lumpy pages
isolated and will avoid breaking the loop early if compound pages smaller
than the requested allocation size are requested.

[mgorman@suse.de: Updated changelog]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: compaction: use synchronous compaction for /proc/sys/vm/compact_memory
Mel Gorman [Fri, 13 Jan 2012 01:19:26 +0000 (17:19 -0800)]
mm: compaction: use synchronous compaction for /proc/sys/vm/compact_memory

When asynchronous compaction was introduced, the
/proc/sys/vm/compact_memory handler should have been updated to always use
synchronous compaction.  This did not happen so this patch addresses it.

The assumption is if a user writes to /proc/sys/vm/compact_memory, they
are willing for that process to stall.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: compaction: allow compaction to isolate dirty pages
Mel Gorman [Fri, 13 Jan 2012 01:19:22 +0000 (17:19 -0800)]
mm: compaction: allow compaction to isolate dirty pages

Short summary: There are severe stalls when a USB stick using VFAT is
used with THP enabled that are reduced by this series.  If you are
experiencing this problem, please test and report back and considering I
have seen complaints from openSUSE and Fedora users on this as well as a
few private mails, I'm guessing it's a widespread issue.  This is a new
type of USB-related stall because it is due to synchronous compaction
writing where as in the past the big problem was dirty pages reaching
the end of the LRU and being written by reclaim.

Am cc'ing Andrew this time and this series would replace
mm-do-not-stall-in-synchronous-compaction-for-thp-allocations.patch.
I'm also cc'ing Dave Jones as he might have merged that patch to Fedora
for wider testing and ideally it would be reverted and replaced by this
series.

That said, the later patches could really do with some review.  If this
series is not the answer then a new direction needs to be discussed
because as it is, the stalls are unacceptable as the results in this
leader show.

For testers that try backporting this to 3.1, it won't work because
there is a non-obvious dependency on not writing back pages in direct
reclaim so you need those patches too.

Changelog since V5
o Rebase to 3.2-rc5
o Tidy up the changelogs a bit

Changelog since V4
o Added reviewed-bys, credited Andrea properly for sync-light
o Allow dirty pages without mappings to be considered for migration
o Bound the number of pages freed for compaction
o Isolate PageReclaim pages on their own LRU list

This is against 3.2-rc5 and follows on from discussions on "mm: Do
not stall in synchronous compaction for THP allocations" and "[RFC
PATCH 0/5] Reduce compaction-related stalls". Initially, the proposed
patch eliminated stalls due to compaction which sometimes resulted in
user-visible interactivity problems on browsers by simply never using
sync compaction. The downside was that THP success allocation rates
were lower because dirty pages were not being migrated as reported by
Andrea. His approach at fixing this was nacked on the grounds that
it reverted fixes from Rik merged that reduced the amount of pages
reclaimed as it severely impacted his workloads performance.

This series attempts to reconcile the requirements of maximising THP
usage, without stalling in a user-visible fashion due to compaction
or cheating by reclaiming an excessive number of pages.

Patch 1 partially reverts commit 39deaf85 to allow migration to isolate
dirty pages. This is because migration can move some dirty
pages without blocking.

Patch 2 notes that the /proc/sys/vm/compact_memory handler is not using
synchronous compaction when it should be. This is unrelated
to the reported stalls but is worth fixing.

Patch 3 checks if we isolated a compound page during lumpy scan and
account for it properly. For the most part, this affects
tracing so it's unrelated to the stalls but worth fixing.

Patch 4 notes that it is possible to abort reclaim early for compaction
and return 0 to the page allocator potentially entering the
"may oom" path. This has not been observed in practice but
the rest of the series potentially makes it easier to happen.

Patch 5 adds a sync parameter to the migratepage callback and gives
the callback responsibility for migrating the page without
blocking if sync==false. For example, fallback_migrate_page
will not call writepage if sync==false. This increases the
number of pages that can be handled by asynchronous compaction
thereby reducing stalls.

Patch 6 restores filter-awareness to isolate_lru_page for migration.
In practice, it means that pages under writeback and pages
without a ->migratepage callback will not be isolated
for migration.

Patch 7 avoids calling direct reclaim if compaction is deferred but
makes sure that compaction is only deferred if sync
compaction was used.

Patch 8 introduces a sync-light migration mechanism that sync compaction
uses. The objective is to allow some stalls but to not call
->writepage which can lead to significant user-visible stalls.

Patch 9 notes that while we want to abort reclaim ASAP to allow
compation to go ahead that we leave a very small window of
opportunity for compaction to run. This patch allows more pages
to be freed by reclaim but bounds the number to a reasonable
level based on the high watermark on each zone.

Patch 10 allows slabs to be shrunk even after compaction_ready() is
true for one zone. This is to avoid a problem whereby a single
small zone can abort reclaim even though no pages have been
reclaimed and no suitably large zone is in a usable state.

Patch 11 fixes a problem with the rate of page scanning. As reclaim is
rarely stalling on pages under writeback it means that scan
rates are very high. This is particularly true for direct
reclaim which is not calling writepage. The vmstat figures
implied that much of this was busy work with PageReclaim pages
marked for immediate reclaim. This patch is a prototype that
moves these pages to their own LRU list.

This has been tested and other than 2 USB keys getting trashed,
nothing horrible fell out. That said, I am a bit unhappy with the
rescue logic in patch 11 but did not find a better way around it. It
does significantly reduce scan rates and System CPU time indicating
it is the right direction to take.

What is of critical importance is that stalls due to compaction
are massively reduced even though sync compaction was still
allowed. Testing from people complaining about stalls copying to USBs
with THP enabled are particularly welcome.

The following tests all involve THP usage and USB keys in some
way. Each test follows this type of pattern

1. Read from some fast fast storage, be it raw device or file. Each time
   the copy finishes, start again until the test ends
2. Write a large file to a filesystem on a USB stick. Each time the copy
   finishes, start again until the test ends
3. When memory is low, start an alloc process that creates a mapping
   the size of physical memory to stress THP allocation. This is the
   "real" part of the test and the part that is meant to trigger
   stalls when THP is enabled. Copying continues in the background.
4. Record the CPU usage and time to execute of the alloc process
5. Record the number of THP allocs and fallbacks as well as the number of THP
   pages in use a the end of the test just before alloc exited
6. Run the test 5 times to get an idea of variability
7. Between each run, sync is run and caches dropped and the test
   waits until nr_dirty is a small number to avoid interference
   or caching between iterations that would skew the figures.

The individual tests were then

writebackCPDeviceBasevfat
Disable THP, read from a raw device (sda), vfat on USB stick
writebackCPDeviceBaseext4
Disable THP, read from a raw device (sda), ext4 on USB stick
writebackCPDevicevfat
THP enabled, read from a raw device (sda), vfat on USB stick
writebackCPDeviceext4
THP enabled, read from a raw device (sda), ext4 on USB stick
writebackCPFilevfat
THP enabled, read from a file on fast storage and USB, both vfat
writebackCPFileext4
THP enabled, read from a file on fast storage and USB, both ext4

The kernels tested were

3.1 3.1
vanilla 3.2-rc5
freemore Patches 1-10
immediate Patches 1-11
andrea The 8 patches Andrea posted as a basis of comparison

The results are very long unfortunately. I'll start with the case
where we are not using THP at all

writebackCPDeviceBasevfat
                   3.1.0-vanilla         rc5-vanilla       freemore-v6r1        isolate-v6r1         andrea-v2r1
System Time         1.28 (    0.00%)   54.49 (-4143.46%)   48.63 (-3687.69%)    4.69 ( -265.11%)   51.88 (-3940.81%)
+/-                 0.06 (    0.00%)    2.45 (-4305.55%)    4.75 (-8430.57%)    7.46 (-13282.76%)    4.76 (-8440.70%)
User Time           0.09 (    0.00%)    0.05 (   40.91%)    0.06 (   29.55%)    0.07 (   15.91%)    0.06 (   27.27%)
+/-                 0.02 (    0.00%)    0.01 (   45.39%)    0.02 (   25.07%)    0.00 (   77.06%)    0.01 (   52.24%)
Elapsed Time      110.27 (    0.00%)   56.38 (   48.87%)   49.95 (   54.70%)   11.77 (   89.33%)   53.43 (   51.54%)
+/-                 7.33 (    0.00%)    3.77 (   48.61%)    4.94 (   32.63%)    6.71 (    8.50%)    4.76 (   35.03%)
THP Active          0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)
+/-                 0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)
Fault Alloc         0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)
+/-                 0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)
Fault Fallback      0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)
+/-                 0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)    0.00 (    0.00%)

The THP figures are obviously all 0 because THP was enabled. The
main thing to watch is the elapsed times and how they compare to
times when THP is enabled later. It's also important to note that
elapsed time is improved by this series as System CPu time is much
reduced.

writebackCPDevicevfat

                   3.1.0-vanilla         rc5-vanilla       freemore-v6r1        isolate-v6r1         andrea-v2r1
System Time         1.22 (    0.00%)   13.89 (-1040.72%)   46.40 (-3709.20%)    4.44 ( -264.37%)   47.37 (-3789.33%)
+/-                 0.06 (    0.00%)   22.82 (-37635.56%)    3.84 (-6249.44%)    6.48 (-10618.92%)    6.60
(-10818.53%)
User Time           0.06 (    0.00%)    0.06 (   -6.90%)    0.05 (   17.24%)    0.05 (   13.79%)    0.04 (   31.03%)
+/-                 0.01 (    0.00%)    0.01 (   33.33%)    0.01 (   33.33%)    0.01 (   39.14%)    0.01 (   25.46%)
Elapsed Time     10445.54 (    0.00%) 2249.92 (   78.46%)   70.06 (   99.33%)   16.59 (   99.84%)  472.43 (
95.48%)
+/-               643.98 (    0.00%)  811.62 (  -26.03%)   10.02 (   98.44%)    7.03 (   98.91%)   59.99 (   90.68%)
THP Active         15.60 (    0.00%)   35.20 (  225.64%)   65.00 (  416.67%)   70.80 (  453.85%)   62.20 (  398.72%)
+/-                18.48 (    0.00%)   51.29 (  277.59%)   15.99 (   86.52%)   37.91 (  205.18%)   22.02 (  119.18%)
Fault Alloc       121.80 (    0.00%)   76.60 (   62.89%)  155.40 (  127.59%)  181.20 (  148.77%)  286.60 (  235.30%)
+/-                73.51 (    0.00%)   61.11 (   83.12%)   34.89 (   47.46%)   31.88 (   43.36%)   68.13 (   92.68%)
Fault Fallback    881.20 (    0.00%)  926.60 (   -5.15%)  847.60 (    3.81%)  822.00 (    6.72%)  716.60 (   18.68%)
+/-                73.51 (    0.00%)   61.26 (   16.67%)   34.89 (   52.54%)   31.65 (   56.94%)   67.75 (    7.84%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)       3540.88   1945.37    716.04     64.97   1937.03
Total Elapsed Time (seconds)              52417.33  11425.90    501.02    230.95   2520.28

The first thing to note is the "Elapsed Time" for the vanilla kernels
of 2249 seconds versus 56 with THP disabled which might explain the
reports of USB stalls with THP enabled. Applying the patches brings
performance in line with THP-disabled performance while isolating
pages for immediate reclaim from the LRU cuts down System CPU time.

The "Fault Alloc" success rate figures are also improved. The vanilla
kernel only managed to allocate 76.6 pages on average over the course
of 5 iterations where as applying the series allocated 181.20 on
average albeit it is well within variance. It's worth noting that
applies the series at least descreases the amount of variance which
implies an improvement.

Andrea's series had a higher success rate for THP allocations but
at a severe cost to elapsed time which is still better than vanilla
but still much worse than disabling THP altogether. One can bring my
series close to Andrea's by removing this check

        /*
         * If compaction is deferred for high-order allocations, it is because
         * sync compaction recently failed. In this is the case and the caller
         * has requested the system not be heavily disrupted, fail the
         * allocation now instead of entering direct reclaim
         */
        if (deferred_compaction && (gfp_mask & __GFP_NO_KSWAPD))
                goto nopage;

I didn't include a patch that removed the above check because hurting
overall performance to improve the THP figure is not what the average
user wants. It's something to consider though if someone really wants
to maximise THP usage no matter what it does to the workload initially.

This is summary of vmstat figures from the same test.

                                       3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1
Page Ins                                  3257266139  1111844061    17263623    10901575   161423219
Page Outs                                   81054922    30364312     3626530     3657687     8753730
Swap Ins                                        3294        2851        6560        4964        4592
Swap Outs                                     390073      528094      620197      790912      698285
Direct pages scanned                      1077581700  3024951463  1764930052   115140570  5901188831
Kswapd pages scanned                        34826043     7112868     2131265     1686942     1893966
Kswapd pages reclaimed                      28950067     4911036     1246044      966475     1497726
Direct pages reclaimed                     805148398   280167837     3623473     2215044    40809360
Kswapd efficiency                                83%         69%         58%         57%         79%
Kswapd velocity                              664.399     622.521    4253.852    7304.360     751.490
Direct efficiency                                74%          9%          0%          1%          0%
Direct velocity                            20557.737  264745.137 3522673.849  498551.938 2341481.435
Percentage direct scans                          96%         99%         99%         98%         99%
Page writes by reclaim                        722646      529174      620319      791018      699198
Page writes file                              332573        1080         122         106         913
Page writes anon                              390073      528094      620197      790912      698285
Page reclaim immediate                             0  2552514720  1635858848   111281140  5478375032
Page rescued immediate                             0           0           0       87848           0
Slabs scanned                                  23552       23552        9216        8192        9216
Direct inode steals                              231           0           0           0           0
Kswapd inode steals                                0           0           0           0           0
Kswapd skipped wait                            28076         786           0          61           6
THP fault alloc                                  609         383         753         906        1433
THP collapse alloc                                12           6           0           0           6
THP splits                                       536         211         456         593        1136
THP fault fallback                              4406        4633        4263        4110        3583
THP collapse fail                                120         127           0           0           4
Compaction stalls                               1810         728         623         779        3200
Compaction success                               196          53          60          80         123
Compaction failures                             1614         675         563         699        3077
Compaction pages moved                        193158       53545      243185      333457      226688
Compaction move failure                         9952        9396       16424       23676       45070

The main things to look at are

1. Page In/out figures are much reduced by the series.

2. Direct page scanning is incredibly high (264745.137 pages scanned
   per second on the vanilla kernel) but isolating PageReclaim pages
   on their own list reduces the number of pages scanned significantly.

3. The fact that "Page rescued immediate" is a positive number implies
   that we sometimes race removing pages from the LRU_IMMEDIATE list
   that need to be put back on a normal LRU but it happens only for
   0.07% of the pages marked for immediate reclaim.

writebackCPDeviceext4
                   3.1.0-vanilla         rc5-vanilla       freemore-v6r1        isolate-v6r1         andrea-v2r1
System Time         1.51 (    0.00%)    1.77 (  -17.66%)    1.46 (    2.92%)    1.15 (   23.77%)    1.89 (  -25.63%)
+/-                 0.27 (    0.00%)    0.67 ( -148.52%)    0.33 (  -22.76%)    0.30 (  -11.15%)    0.19 (   30.16%)
User Time           0.03 (    0.00%)    0.04 (  -37.50%)    0.05 (  -62.50%)    0.07 ( -112.50%)    0.04 (  -18.75%)
+/-                 0.01 (    0.00%)    0.02 ( -146.64%)    0.02 (  -97.91%)    0.02 (  -75.59%)    0.02 (  -63.30%)
Elapsed Time      124.93 (    0.00%)  114.49 (    8.36%)   96.77 (   22.55%)   27.48 (   78.00%)  205.70 (  -64.65%)
+/-                20.20 (    0.00%)   74.39 ( -268.34%)   59.88 ( -196.48%)    7.72 (   61.79%)   25.03 (  -23.95%)
THP Active        161.80 (    0.00%)   83.60 (   51.67%)  141.20 (   87.27%)   84.60 (   52.29%)   82.60 (   51.05%)
+/-                71.95 (    0.00%)   43.80 (   60.88%)   26.91 (   37.40%)   59.02 (   82.03%)   52.13 (   72.45%)
Fault Alloc       471.40 (    0.00%)  228.60 (   48.49%)  282.20 (   59.86%)  225.20 (   47.77%)  388.40 (   82.39%)
+/-                88.07 (    0.00%)   87.42 (   99.26%)   73.79 (   83.78%)  109.62 (  124.47%)   82.62 (   93.81%)
Fault Fallback    531.60 (    0.00%)  774.60 (  -45.71%)  720.80 (  -35.59%)  777.80 (  -46.31%)  614.80 (  -15.65%)
+/-                88.07 (    0.00%)   87.26 (    0.92%)   73.79 (   16.22%)  109.62 (  -24.47%)   82.29 (    6.56%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)         50.22     33.76     30.65     24.14    128.45
Total Elapsed Time (seconds)               1113.73   1132.19   1029.45    759.49   1707.26

Similar test but the USB stick is using ext4 instead of vfat. As
ext4 does not use writepage for migration, the large stalls due to
compaction when THP is enabled are not observed. Still, isolating
PageReclaim pages on their own list helped completion time largely
by reducing the number of pages scanned by direct reclaim although
time spend in congestion_wait could also be a factor.

Again, Andrea's series had far higher success rates for THP allocation
at the cost of elapsed time. I didn't look too closely but a quick
look at the vmstat figures tells me kswapd reclaimed 8 times more pages
than the patch series and direct reclaim reclaimed roughly three times
as many pages. It follows that if memory is aggressively reclaimed,
there will be more available for THP.

writebackCPFilevfat
                   3.1.0-vanilla         rc5-vanilla       freemore-v6r1        isolate-v6r1         andrea-v2r1
System Time         1.76 (    0.00%)   29.10 (-1555.52%)   46.01 (-2517.18%)    4.79 ( -172.35%)   54.89 (-3022.53%)
+/-                 0.14 (    0.00%)   25.61 (-18185.17%)    2.15 (-1434.83%)    6.60 (-4610.03%)    9.75
(-6863.76%)
User Time           0.05 (    0.00%)    0.07 (  -45.83%)    0.05 (   -4.17%)    0.06 (  -29.17%)    0.06 (  -16.67%)
+/-                 0.02 (    0.00%)    0.02 (   20.11%)    0.02 (   -3.14%)    0.01 (   31.58%)    0.01 (   47.41%)
Elapsed Time     22520.79 (    0.00%) 1082.85 (   95.19%)   73.30 (   99.67%)   32.43 (   99.86%)  291.84 (  98.70%)
+/-              7277.23 (    0.00%)  706.29 (   90.29%)   19.05 (   99.74%)   17.05 (   99.77%)  125.55 (   98.27%)
THP Active         83.80 (    0.00%)   12.80 (   15.27%)   15.60 (   18.62%)   13.00 (   15.51%)    0.80 (    0.95%)
+/-                66.81 (    0.00%)   20.19 (   30.22%)    5.92 (    8.86%)   15.06 (   22.54%)    1.17 (    1.75%)
Fault Alloc       171.00 (    0.00%)   67.80 (   39.65%)   97.40 (   56.96%)  125.60 (   73.45%)  133.00 (   77.78%)
+/-                82.91 (    0.00%)   30.69 (   37.02%)   53.91 (   65.02%)   55.05 (   66.40%)   21.19 (   25.56%)
Fault Fallback    832.00 (    0.00%)  935.20 (  -12.40%)  906.00 (   -8.89%)  877.40 (   -5.46%)  870.20 (   -4.59%)
+/-                82.91 (    0.00%)   30.69 (   62.98%)   54.01 (   34.86%)   55.05 (   33.60%)   20.91 (   74.78%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)       7229.81    928.42    704.52     80.68   1330.76
Total Elapsed Time (seconds)             112849.04   5618.69    571.11    360.54   1664.28

In this case, the test is reading/writing only from filesystems but as
it's vfat, it's slow due to calling writepage during compaction. Little
to observe really - the time to complete the test goes way down
with the series applied and THP allocation success rates go up in
comparison to 3.2-rc5.  The success rates are lower than 3.1.0 but
the elapsed time for that kernel is abysmal so it is not really a
sensible comparison.

As before, Andrea's series allocates more THPs at the cost of overall
performance.

writebackCPFileext4
                   3.1.0-vanilla         rc5-vanilla       freemore-v6r1        isolate-v6r1         andrea-v2r1
System Time         1.51 (    0.00%)    1.77 (  -17.66%)    1.46 (    2.92%)    1.15 (   23.77%)    1.89 (  -25.63%)
+/-                 0.27 (    0.00%)    0.67 ( -148.52%)    0.33 (  -22.76%)    0.30 (  -11.15%)    0.19 (   30.16%)
User Time           0.03 (    0.00%)    0.04 (  -37.50%)    0.05 (  -62.50%)    0.07 ( -112.50%)    0.04 (  -18.75%)
+/-                 0.01 (    0.00%)    0.02 ( -146.64%)    0.02 (  -97.91%)    0.02 (  -75.59%)    0.02 (  -63.30%)
Elapsed Time      124.93 (    0.00%)  114.49 (    8.36%)   96.77 (   22.55%)   27.48 (   78.00%)  205.70 (  -64.65%)
+/-                20.20 (    0.00%)   74.39 ( -268.34%)   59.88 ( -196.48%)    7.72 (   61.79%)   25.03 (  -23.95%)
THP Active        161.80 (    0.00%)   83.60 (   51.67%)  141.20 (   87.27%)   84.60 (   52.29%)   82.60 (   51.05%)
+/-                71.95 (    0.00%)   43.80 (   60.88%)   26.91 (   37.40%)   59.02 (   82.03%)   52.13 (   72.45%)
Fault Alloc       471.40 (    0.00%)  228.60 (   48.49%)  282.20 (   59.86%)  225.20 (   47.77%)  388.40 (   82.39%)
+/-                88.07 (    0.00%)   87.42 (   99.26%)   73.79 (   83.78%)  109.62 (  124.47%)   82.62 (   93.81%)
Fault Fallback    531.60 (    0.00%)  774.60 (  -45.71%)  720.80 (  -35.59%)  777.80 (  -46.31%)  614.80 (  -15.65%)
+/-                88.07 (    0.00%)   87.26 (    0.92%)   73.79 (   16.22%)  109.62 (  -24.47%)   82.29 (    6.56%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)         50.22     33.76     30.65     24.14    128.45
Total Elapsed Time (seconds)               1113.73   1132.19   1029.45    759.49   1707.26

Same type of story - elapsed times go down. In this case, allocation
success rates are roughtly the same. As before, Andrea's has higher
success rates but takes a lot longer.

Overall the series does reduce latencies and while the tests are
inherency racy as alloc competes with the cp processes, the variability
was included. The THP allocation rates are not as high as they could
be but that is because we would have to be more aggressive about
reclaim and compaction impacting overall performance.

This patch:

Commit 39deaf85 ("mm: compaction: make isolate_lru_page() filter-aware")
noted that compaction does not migrate dirty or writeback pages and that
is was meaningless to pick the page and re-add it to the LRU list.

What was missed during review is that asynchronous migration moves dirty
pages if their ->migratepage callback is migrate_page() because these can
be moved without blocking.  This potentially impacted hugepage allocation
success rates by a factor depending on how many dirty pages are in the
system.

This patch partially reverts 39deaf85 to allow migration to isolate dirty
pages again.  This increases how much compaction disrupts the LRU but that
is addressed later in the series.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Andy Isaacson <adi@hexapodia.org>
Cc: Nai Xia <nai.xia@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agovmscan/trace: Add 'file' info to trace_mm_vmscan_lru_isolate()
Tao Ma [Fri, 13 Jan 2012 01:19:20 +0000 (17:19 -0800)]
vmscan/trace: Add 'file' info to trace_mm_vmscan_lru_isolate()

In trace_mm_vmscan_lru_isolate(), we don't output 'file' information to
the trace event and it is a bit inconvenient for the user to get the
real information(like pasted below).  mm_vmscan_lru_isolate:
isolate_mode=2 order=0 nr_requested=32 nr_scanned=32 nr_taken=32
contig_taken=0 contig_dirty=0 contig_failed=0

'active' can be obtained by analyzing mode(Thanks go to Minchan and
Mel), So this patch adds 'file' to the trace event and it now looks
like: mm_vmscan_lru_isolate: isolate_mode=2 order=0 nr_requested=32
nr_scanned=32 nr_taken=32 contig_taken=0 contig_dirty=0 contig_failed=0
file=0

Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan@kernel.org>
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>
12 years agothp: improve order in lru list for split huge page
Shaohua Li [Fri, 13 Jan 2012 01:19:18 +0000 (17:19 -0800)]
thp: improve order in lru list for split huge page

Put the tail subpages of an isolated hugepage under splitting in the lru
reclaim head as they supposedly should be isolated too next.

Queues the subpages in physical order in the lru for non isolated
hugepages under splitting.  That might provide some theoretical cache
benefit to the buddy allocator later.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agothp: add tlb_remove_pmd_tlb_entry
Shaohua Li [Fri, 13 Jan 2012 01:19:16 +0000 (17:19 -0800)]
thp: add tlb_remove_pmd_tlb_entry

We have tlb_remove_tlb_entry to indicate a pte tlb flush entry should be
flushed, but not a corresponding API for pmd entry.  This isn't a
problem so far because THP is only for x86 currently and tlb_flush()
under x86 will flush entire TLB.  But this is confusion and could be
missed if thp is ported to other arch.

Also convert tlb->need_flush = 1 to a VM_BUG_ON(!tlb->need_flush) in
__tlb_remove_page() as suggested by Andrea Arcangeli.  The
__tlb_remove_page() function is supposed to be called after
tlb_remove_xxx_tlb_entry() and we can catch any misuse.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agothp: remove unnecessary tlb flush for mprotect
Shaohua Li [Fri, 13 Jan 2012 01:19:13 +0000 (17:19 -0800)]
thp: remove unnecessary tlb flush for mprotect

change_protection() will do TLB flush later, don't need duplicate tlb
flush.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agothp: improve the error code path
Shaohua Li [Fri, 13 Jan 2012 01:19:11 +0000 (17:19 -0800)]
thp: improve the error code path

Improve the error code path.  Delete unnecessary sysfs file for example.
Also remove the #ifdef xxx to make code better.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agopage_cgroup: drop multi CONFIG_MEMORY_HOTPLUG
Bob Liu [Fri, 13 Jan 2012 01:19:08 +0000 (17:19 -0800)]
page_cgroup: drop multi CONFIG_MEMORY_HOTPLUG

No need for two CONFIG_MEMORY_HOTPLUG blocks.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agopage_alloc: break early in check_for_regular_memory()
Bob Liu [Fri, 13 Jan 2012 01:19:07 +0000 (17:19 -0800)]
page_alloc: break early in check_for_regular_memory()

If there is a zone below ZONE_NORMAL has present_pages, we can set node
state to N_NORMAL_MEMORY, no need to loop to end.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: cleanup for_each_node_state()
Bob Liu [Fri, 13 Jan 2012 01:19:04 +0000 (17:19 -0800)]
memcg: cleanup for_each_node_state()

We already have for_each_node(node) define in nodemask.h, better to use it.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: simplify LRU handling by new rule
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:19:01 +0000 (17:19 -0800)]
memcg: simplify LRU handling by new rule

Now, at LRU handling, memory cgroup needs to do complicated works to see
valid pc->mem_cgroup, which may be overwritten.

This patch is for relaxing the protocol. This patch guarantees
   - when pc->mem_cgroup is overwritten, page must not be on LRU.

By this, LRU routine can believe pc->mem_cgroup and don't need to check
bits on pc->flags.  This new rule may adds small overheads to swapin.  But
in most case, lru handling gets faster.

After this patch, PCG_ACCT_LRU bit is obsolete and removed.

[akpm@linux-foundation.org: remove unneeded VM_BUG_ON(), restore hannes's christmas tree]
[akpm@linux-foundation.org: clean up code comment]
[hughd@google.com: fix NULL mem_cgroup_try_charge]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: clear pc->mem_cgroup if necessary.
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:18:58 +0000 (17:18 -0800)]
memcg: clear pc->mem_cgroup if necessary.

This is a preparation before removing a flag PCG_ACCT_LRU in page_cgroup
and reducing atomic ops/complexity in memcg LRU handling.

In some cases, pages are added to lru before charge to memcg and pages
are not classfied to memory cgroup at lru addtion.  Now, the lru where
the page should be added is determined a bit in page_cgroup->flags and
pc->mem_cgroup.  I'd like to remove the check of flag.

To handle the case pc->mem_cgroup may contain stale pointers if pages
are added to LRU before classification.  This patch resets
pc->mem_cgroup to root_mem_cgroup before lru additions.

[akpm@linux-foundation.org: fix CONFIG_CGROUP_MEM_CONT=n build]
[hughd@google.com: fix CONFIG_CGROUP_MEM_RES_CTLR=y CONFIG_CGROUP_MEM_RES_CTLR_SWAP=n build]
[akpm@linux-foundation.org: ksm.c needs memcontrol.h, per Michal]
[hughd@google.com: stop oops in mem_cgroup_reset_owner()]
[hughd@google.com: fix page migration to reset_owner]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: simplify corner case handling of LRU.
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:18:57 +0000 (17:18 -0800)]
memcg: simplify corner case handling of LRU.

This patch simplifies LRU handling of racy case (memcg+SwapCache).  At
charging, SwapCache tend to be on LRU already.  So, before overwriting
pc->mem_cgroup, the page must be removed from LRU and added to LRU
later.

This patch does
        spin_lock(zone->lru_lock);
        if (PageLRU(page))
                remove from LRU
        overwrite pc->mem_cgroup
        if (PageLRU(page))
                add to new LRU.
        spin_unlock(zone->lru_lock);

And guarantee all pages are not on LRU at modifying pc->mem_cgroup.
This patch also unfies lru handling of replace_page_cache() and
swapin.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.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>
12 years agomemcg: simplify page cache charging
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:18:55 +0000 (17:18 -0800)]
memcg: simplify page cache charging

This patch is a clean up. No functional/logical changes.

Because of commit ef6a3c6311 ("mm: add replace_page_cache_page()
function") , FUSE uses replace_page_cache() instead of
add_to_page_cache().  Then, mem_cgroup_cache_charge() is not called
against FUSE's pages from splice.

So now, mem_cgroup_cache_charge() gets pages that are not on the LRU
with the exception of PageSwapCache pages.  For checking,
WARN_ON_ONCE(PageLRU(page)) is added.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.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>
12 years agooom, memcg: fix exclusion of memcg threads after they have detached their mm
David Rientjes [Fri, 13 Jan 2012 01:18:52 +0000 (17:18 -0800)]
oom, memcg: fix exclusion of memcg threads after they have detached their mm

The oom killer relies on logic that identifies threads that have already
been oom killed when scanning the tasklist and, if found, deferring
until such threads have exited.  This is done by checking for any
candidate threads that have the TIF_MEMDIE bit set.

For memcg ooms, candidate threads are first found by calling
task_in_mem_cgroup() since the oom killer should not defer if there's an
oom killed thread in another memcg.

Unfortunately, task_in_mem_cgroup() excludes threads if they have
detached their mm in the process of exiting so TIF_MEMDIE is never
detected for such conditions.  This is different for global, mempolicy,
and cpuset oom conditions where a detached mm is only excluded after
checking for TIF_MEMDIE and deferring, if necessary, in
select_bad_process().

The fix is to return true if a task has a detached mm but is still in
the memcg or its hierarchy that is currently oom.  This will allow the
oom killer to appropriately defer rather than kill unnecessarily or, in
the worst case, panic the machine if nothing else is available to kill.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: free entries in soft_limit_tree if allocation fails
Michal Hocko [Fri, 13 Jan 2012 01:18:50 +0000 (17:18 -0800)]
memcg: free entries in soft_limit_tree if allocation fails

If we are not able to allocate tree nodes for all NUMA nodes then we
should release those that were allocated.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.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>
12 years agopage_cgroup: add helper function to get swap_cgroup
Bob Liu [Fri, 13 Jan 2012 01:18:48 +0000 (17:18 -0800)]
page_cgroup: add helper function to get swap_cgroup

There are multiple places which need to get the swap_cgroup address, so
add a helper function:

  static struct swap_cgroup *swap_cgroup_getsc(swp_entry_t ent,
                                struct swap_cgroup_ctrl **ctrl);

to simplify the code.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: remove unneeded checks from uncharge_page()
Johannes Weiner [Fri, 13 Jan 2012 01:18:45 +0000 (17:18 -0800)]
mm: memcg: remove unneeded checks from uncharge_page()

mem_cgroup_uncharge_page() is only called on either freshly allocated
pages without page->mapping or on rmapped PageAnon() pages.  There is no
need to check for a page->mapping that is not an anon_vma.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.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>
12 years agomm: memcg: remove unneeded checks from newpage_charge()
Johannes Weiner [Fri, 13 Jan 2012 01:18:43 +0000 (17:18 -0800)]
mm: memcg: remove unneeded checks from newpage_charge()

All callsites pass in freshly allocated pages and a valid mm.  As a
result, all checks pertaining to the page's mapcount, page->mapping or the
fallback to init_mm are unneeded.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Balbir Singh <bsingharora@gmail.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>
12 years agomm: page_cgroup: check page_cgroup arrays in lookup_page_cgroup() only when necessary
Johannes Weiner [Fri, 13 Jan 2012 01:18:40 +0000 (17:18 -0800)]
mm: page_cgroup: check page_cgroup arrays in lookup_page_cgroup() only when necessary

lookup_page_cgroup() is usually used only against pages that are used in
userspace.

The exception is the CONFIG_DEBUG_VM-only memcg check from the page
allocator: it can run on pages without page_cgroup descriptors allocated
when the pages are fed into the page allocator for the first time during
boot or memory hotplug.

Include the array check only when CONFIG_DEBUG_VM is set and save the
unnecessary check in production kernels.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.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>
12 years agomm: memcg: lookup_page_cgroup (almost) never returns NULL
Johannes Weiner [Fri, 13 Jan 2012 01:18:38 +0000 (17:18 -0800)]
mm: memcg: lookup_page_cgroup (almost) never returns NULL

Pages have their corresponding page_cgroup descriptors set up before
they are used in userspace, and thus managed by a memory cgroup.

The only time where lookup_page_cgroup() can return NULL is in the
CONFIG_DEBUG_VM-only page sanity checking code that executes while
feeding pages into the page allocator for the first time.

Remove the NULL checks against lookup_page_cgroup() results from all
callsites where we know that corresponding page_cgroup descriptors must
be allocated, and add a comment to the callsite that actually does have
to check the return value.

[hughd@google.com: stop oops in mem_cgroup_update_page_stat()]
Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: clean up fault accounting
Johannes Weiner [Fri, 13 Jan 2012 01:18:35 +0000 (17:18 -0800)]
mm: memcg: clean up fault accounting

The fault accounting functions have a single, memcg-internal user, so they
don't need to be global.  In fact, their one-line bodies can be directly
folded into the caller.  And since faults happen one at a time, use
this_cpu_inc() directly instead of this_cpu_add(foo, 1).

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.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>
12 years agomm: unify remaining mem_cont, mem, etc. variable names to memcg
Johannes Weiner [Fri, 13 Jan 2012 01:18:32 +0000 (17:18 -0800)]
mm: unify remaining mem_cont, mem, etc. variable names to memcg

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.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>
12 years agomm: oom_kill: remove memcg argument from oom_kill_task()
Johannes Weiner [Fri, 13 Jan 2012 01:18:29 +0000 (17:18 -0800)]
mm: oom_kill: remove memcg argument from oom_kill_task()

The memcg argument of oom_kill_task() hasn't been used since 341aea2
'oom-kill: remove boost_dying_task_prio()'.  Kill it.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.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>
12 years agomemcg: fix pgpgin/pgpgout documentation
Ying Han [Fri, 13 Jan 2012 01:18:27 +0000 (17:18 -0800)]
memcg: fix pgpgin/pgpgout documentation

The two memcg stats pgpgin/pgpgout have different meaning than the ones
in vmstat, which indicates that we picked a bad naming for them.

It might be late to change the stat name, but better documentation is
always helpful.

Signed-off-by: Ying Han <yinghan@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoDocumentation/cgroups/memory.txt: fix typo
Zhu Yanhai [Fri, 13 Jan 2012 01:18:24 +0000 (17:18 -0800)]
Documentation/cgroups/memory.txt: fix typo

It should be memsw.max_usage_in_bytes. This typo has been there for
a really long time.

Signed-off-by: Zhu Yanhai <gaoyang.zyh@taobao.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: shorten preempt-disabled section around event checks
Johannes Weiner [Fri, 13 Jan 2012 01:18:23 +0000 (17:18 -0800)]
mm: memcg: shorten preempt-disabled section around event checks

Only the ratelimit checks themselves have to run with preemption
disabled, the resulting actions - checking for usage thresholds,
updating the soft limit tree - can and should run with preemption
enabled.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reported-by: Yong Zhang <yong.zhang0@gmail.com>
Tested-by: Yong Zhang <yong.zhang0@gmail.com>
Reported-by: Luis Henriques <henrix@camandro.org>
Tested-by: Luis Henriques <henrix@camandro.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: make mem_cgroup_split_huge_fixup() more efficient
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:18:20 +0000 (17:18 -0800)]
memcg: make mem_cgroup_split_huge_fixup() more efficient

In split_huge_page(), mem_cgroup_split_huge_fixup() is called to handle
page_cgroup modifcations.  It takes move_lock_page_cgroup() and modifies
page_cgroup and LRU accounting jobs and called HPAGE_PMD_SIZE - 1 times.

But thinking again,
  - compound_lock() is held at move_accout...then, it's not necessary
    to take move_lock_page_cgroup().
  - LRU is locked and all tail pages will go into the same LRU as
    head is now on.
  - page_cgroup is contiguous in huge page range.

This patch fixes mem_cgroup_split_huge_fixup() as to be called once per
hugepage and reduce costs for spliting.

[akpm@linux-foundation.org: fix typo, per Michal]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: remove unused node/section info from pc->flags
Johannes Weiner [Fri, 13 Jan 2012 01:18:18 +0000 (17:18 -0800)]
mm: memcg: remove unused node/section info from pc->flags

To find the page corresponding to a certain page_cgroup, the pc->flags
encoded the node or section ID with the base array to compare the pc
pointer to.

Now that the per-memory cgroup LRU lists link page descriptors directly,
there is no longer any code that knows the struct page_cgroup of a PFN
but not the struct page.

[hughd@google.com: remove unused node/section info from pc->flags fix]
Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: make per-memcg LRU lists exclusive
Johannes Weiner [Fri, 13 Jan 2012 01:18:15 +0000 (17:18 -0800)]
mm: make per-memcg LRU lists exclusive

Now that all code that operated on global per-zone LRU lists is
converted to operate on per-memory cgroup LRU lists instead, there is no
reason to keep the double-LRU scheme around any longer.

The pc->lru member is removed and page->lru is linked directly to the
per-memory cgroup LRU lists, which removes two pointers from a
descriptor that exists for every page frame in the system.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Ying Han <yinghan@google.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: collect LRU list heads into struct lruvec
Johannes Weiner [Fri, 13 Jan 2012 01:18:10 +0000 (17:18 -0800)]
mm: collect LRU list heads into struct lruvec

Having a unified structure with a LRU list set for both global zones and
per-memcg zones allows to keep that code simple which deals with LRU
lists and does not care about the container itself.

Once the per-memcg LRU lists directly link struct pages, the isolation
function and all other list manipulations are shared between the memcg
case and the global LRU case.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: convert global reclaim to per-memcg LRU lists
Johannes Weiner [Fri, 13 Jan 2012 01:18:06 +0000 (17:18 -0800)]
mm: vmscan: convert global reclaim to per-memcg LRU lists

The global per-zone LRU lists are about to go away on memcg-enabled
kernels, global reclaim must be able to find its pages on the per-memcg
LRU lists.

Since the LRU pages of a zone are distributed over all existing memory
cgroups, a scan target for a zone is complete when all memory cgroups
are scanned for their proportional share of a zone's memory.

The forced scanning of small scan targets from kswapd is limited to
zones marked unreclaimable, otherwise kswapd can quickly overreclaim by
force-scanning the LRU lists of multiple memory cgroups.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: remove optimization of keeping the root_mem_cgroup LRU lists empty
Johannes Weiner [Fri, 13 Jan 2012 01:18:02 +0000 (17:18 -0800)]
mm: memcg: remove optimization of keeping the root_mem_cgroup LRU lists empty

root_mem_cgroup, lacking a configurable limit, was never subject to
limit reclaim, so the pages charged to it could be kept off its LRU
lists.  They would be found on the global per-zone LRU lists upon
physical memory pressure and it made sense to avoid uselessly linking
them to both lists.

The global per-zone LRU lists are about to go away on memcg-enabled
kernels, with all pages being exclusively linked to their respective
per-memcg LRU lists.  As a result, pages of the root_mem_cgroup must
also be linked to its LRU lists again.  This is purely about the LRU
list, root_mem_cgroup is still not charged.

The overhead is temporary until the double-LRU scheme is going away
completely.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: move memcg hierarchy reclaim to generic reclaim code
Johannes Weiner [Fri, 13 Jan 2012 01:17:59 +0000 (17:17 -0800)]
mm: move memcg hierarchy reclaim to generic reclaim code

Memory cgroup limit reclaim and traditional global pressure reclaim will
soon share the same code to reclaim from a hierarchical tree of memory
cgroups.

In preparation of this, move the two right next to each other in
shrink_zone().

The mem_cgroup_hierarchical_reclaim() polymath is split into a soft
limit reclaim function, which still does hierarchy walking on its own,
and a limit (shrinking) reclaim function, which relies on generic
reclaim code to walk the hierarchy.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: per-priority per-zone hierarchy scan generations
Johannes Weiner [Fri, 13 Jan 2012 01:17:55 +0000 (17:17 -0800)]
mm: memcg: per-priority per-zone hierarchy scan generations

Memory cgroup limit reclaim currently picks one memory cgroup out of the
target hierarchy, remembers it as the last scanned child, and reclaims
all zones in it with decreasing priority levels.

The new hierarchy reclaim code will pick memory cgroups from the same
hierarchy concurrently from different zones and priority levels, it
becomes necessary that hierarchy roots not only remember the last
scanned child, but do so for each zone and priority level.

Until now, we reclaimed memcgs like this:

    mem = mem_cgroup_iter(root)
    for each priority level:
      for each zone in zonelist:
        reclaim(mem, zone)

But subsequent patches will move the memcg iteration inside the loop
over the zones:

    for each priority level:
      for each zone in zonelist:
        mem = mem_cgroup_iter(root)
        reclaim(mem, zone)

And to keep with the original scan order - memcg -> priority -> zone -
the last scanned memcg has to be remembered per zone and per priority
level.

Furthermore, global reclaim will be switched to the hierarchy walk as
well.  Different from limit reclaim, which can just recheck the limit
after some reclaim progress, its target is to scan all memcgs for the
desired zone pages, proportional to the memcg size, and so reliably
detecting a full hierarchy round-trip will become crucial.

Currently, the code relies on one reclaimer encountering the same memcg
twice, but that is error-prone with concurrent reclaimers.  Instead, use
a generation counter that is increased every time the child with the
highest ID has been visited, so that reclaimers can stop when the
generation changes.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: distinguish between memcg triggering reclaim and memcg being scanned
Johannes Weiner [Fri, 13 Jan 2012 01:17:52 +0000 (17:17 -0800)]
mm: vmscan: distinguish between memcg triggering reclaim and memcg being scanned

Memory cgroup hierarchies are currently handled completely outside of
the traditional reclaim code, which is invoked with a single memory
cgroup as an argument for the whole call stack.

Subsequent patches will switch this code to do hierarchical reclaim, so
there needs to be a distinction between a) the memory cgroup that is
triggering reclaim due to hitting its limit and b) the memory cgroup
that is being scanned as a child of a).

This patch introduces a struct mem_cgroup_zone that contains the
combination of the memory cgroup and the zone being scanned, which is
then passed down the stack instead of the zone argument.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: vmscan: distinguish global reclaim from global LRU scanning
Johannes Weiner [Fri, 13 Jan 2012 01:17:50 +0000 (17:17 -0800)]
mm: vmscan: distinguish global reclaim from global LRU scanning

The traditional zone reclaim code is scanning the per-zone LRU lists
during direct reclaim and kswapd, and the per-zone per-memory cgroup LRU
lists when reclaiming on behalf of a memory cgroup limit.

Subsequent patches will convert the traditional reclaim code to reclaim
exclusively from the per-memory cgroup LRU lists.  As a result, using
the predicate for which LRU list is scanned will no longer be
appropriate to tell global reclaim from limit reclaim.

This patch adds a global_reclaim() predicate to tell direct/kswapd
reclaim from memory cgroup limit reclaim and substitutes it in all
places where currently scanning_global_lru() is used for that.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm: memcg: consolidate hierarchy iteration primitives
Johannes Weiner [Fri, 13 Jan 2012 01:17:48 +0000 (17:17 -0800)]
mm: memcg: consolidate hierarchy iteration primitives

The memcg naturalization series:

Memory control groups are currently bolted onto the side of
traditional memory management in places where better integration would
be preferrable.  To reclaim memory, for example, memory control groups
maintain their own LRU list and reclaim strategy aside from the global
per-zone LRU list reclaim.  But an extra list head for each existing
page frame is expensive and maintaining it requires additional code.

This patchset disables the global per-zone LRU lists on memory cgroup
configurations and converts all its users to operate on the per-memory
cgroup lists instead.  As LRU pages are then exclusively on one list,
this saves two list pointers for each page frame in the system:

page_cgroup array size with 4G physical memory

  vanilla: allocated 31457280 bytes of page_cgroup
  patched: allocated 15728640 bytes of page_cgroup

At the same time, system performance for various workloads is
unaffected:

100G sparse file cat, 4G physical memory, 10 runs, to test for code
bloat in the traditional LRU handling and kswapd & direct reclaim
paths, without/with the memory controller configured in

  vanilla: 71.603(0.207) seconds
  patched: 71.640(0.156) seconds

  vanilla: 79.558(0.288) seconds
  patched: 77.233(0.147) seconds

100G sparse file cat in 1G memory cgroup, 10 runs, to test for code
bloat in the traditional memory cgroup LRU handling and reclaim path

  vanilla: 96.844(0.281) seconds
  patched: 94.454(0.311) seconds

4 unlimited memcgs running kbuild -j32 each, 4G physical memory, 500M
swap on SSD, 10 runs, to test for regressions in kswapd & direct
reclaim using per-memcg LRU lists with multiple memcgs and multiple
allocators within each memcg

  vanilla: 717.722(1.440) seconds [ 69720.100(11600.835) majfaults ]
  patched: 714.106(2.313) seconds [ 71109.300(14886.186) majfaults ]

16 unlimited memcgs running kbuild, 1900M hierarchical limit, 500M
swap on SSD, 10 runs, to test for regressions in hierarchical memcg
setups

  vanilla: 2742.058(1.992) seconds [ 26479.600(1736.737) majfaults ]
  patched: 2743.267(1.214) seconds [ 27240.700(1076.063) majfaults ]

This patch:

There are currently two different implementations of iterating over a
memory cgroup hierarchy tree.

Consolidate them into one worker function and base the convenience
looping-macros on top of it.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: add mem_cgroup_replace_page_cache() to fix LRU issue
KAMEZAWA Hiroyuki [Fri, 13 Jan 2012 01:17:44 +0000 (17:17 -0800)]
memcg: add mem_cgroup_replace_page_cache() to fix LRU issue

Commit ef6a3c6311 ("mm: add replace_page_cache_page() function") added a
function replace_page_cache_page().  This function replaces a page in the
radix-tree with a new page.  WHen doing this, memory cgroup needs to fix
up the accounting information.  memcg need to check PCG_USED bit etc.

In some(many?) cases, 'newpage' is on LRU before calling
replace_page_cache().  So, memcg's LRU accounting information should be
fixed, too.

This patch adds mem_cgroup_replace_page_cache() and removes the old hooks.
 In that function, old pages will be unaccounted without touching
res_counter and new page will be accounted to the memcg (of old page).
WHen overwriting pc->mem_cgroup of newpage, take zone->lru_lock and avoid
races with LRU handling.

Background:
  replace_page_cache_page() is called by FUSE code in its splice() handling.
  Here, 'newpage' is replacing oldpage but this newpage is not a newly allocated
  page and may be on LRU. LRU mis-accounting will be critical for memory cgroup
  because rmdir() checks the whole LRU is empty and there is no account leak.
  If a page is on the other LRU than it should be, rmdir() will fail.

This bug was added in March 2011, but no bug report yet.  I guess there
are not many people who use memcg and FUSE at the same time with upstream
kernels.

The result of this bug is that admin cannot destroy a memcg because of
account leak.  So, no panic, no deadlock.  And, even if an active cgroup
exist, umount can succseed.  So no problem at shutdown.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoepoll: limit paths
Jason Baron [Fri, 13 Jan 2012 01:17:43 +0000 (17:17 -0800)]
epoll: limit paths

The current epoll code can be tickled to run basically indefinitely in
both loop detection path check (on ep_insert()), and in the wakeup paths.
The programs that tickle this behavior set up deeply linked networks of
epoll file descriptors that cause the epoll algorithms to traverse them
indefinitely.  A couple of these sample programs have been previously
posted in this thread: https://lkml.org/lkml/2011/2/25/297.

To fix the loop detection path check algorithms, I simply keep track of
the epoll nodes that have been already visited.  Thus, the loop detection
becomes proportional to the number of epoll file descriptor and links.
This dramatically decreases the run-time of the loop check algorithm.  In
one diabolical case I tried it reduced the run-time from 15 mintues (all
in kernel time) to .3 seconds.

Fixing the wakeup paths could be done at wakeup time in a similar manner
by keeping track of nodes that have already been visited, but the
complexity is harder, since there can be multiple wakeups on different
cpus...Thus, I've opted to limit the number of possible wakeup paths when
the paths are created.

This is accomplished, by noting that the end file descriptor points that
are found during the loop detection pass (from the newly added link), are
actually the sources for wakeup events.  I keep a list of these file
descriptors and limit the number and length of these paths that emanate
from these 'source file descriptors'.  In the current implemetation I
allow 1000 paths of length 1, 500 of length 2, 100 of length 3, 50 of
length 4 and 10 of length 5.  Note that it is sufficient to check the
'source file descriptors' reachable from the newly added link, since no
other 'source file descriptors' will have newly added links.  This allows
us to check only the wakeup paths that may have gotten too long, and not
re-check all possible wakeup paths on the system.

In terms of the path limit selection, I think its first worth noting that
the most common case for epoll, is probably the model where you have 1
epoll file descriptor that is monitoring n number of 'source file
descriptors'.  In this case, each 'source file descriptor' has a 1 path of
length 1.  Thus, I believe that the limits I'm proposing are quite
reasonable and in fact may be too generous.  Thus, I'm hoping that the
proposed limits will not prevent any workloads that currently work to
fail.

In terms of locking, I have extended the use of the 'epmutex' to all
epoll_ctl add and remove operations.  Currently its only used in a subset
of the add paths.  I need to hold the epmutex, so that we can correctly
traverse a coherent graph, to check the number of paths.  I believe that
this additional locking is probably ok, since its in the setup/teardown
paths, and doesn't affect the running paths, but it certainly is going to
add some extra overhead.  Also, worth noting is that the epmuex was
recently added to the ep_ctl add operations in the initial path loop
detection code using the argument that it was not on a critical path.

Another thing to note here, is the length of epoll chains that is allowed.
Currently, eventpoll.c defines:

/* Maximum number of nesting allowed inside epoll sets */
#define EP_MAX_NESTS 4

This basically means that I am limited to a graph depth of 5 (EP_MAX_NESTS
+ 1).  However, this limit is currently only enforced during the loop
check detection code, and only when the epoll file descriptors are added
in a certain order.  Thus, this limit is currently easily bypassed.  The
newly added check for wakeup paths, stricly limits the wakeup paths to a
length of 5, regardless of the order in which ep's are linked together.
Thus, a side-effect of the new code is a more consistent enforcement of
the graph depth.

Thus far, I've tested this, using the sample programs previously
mentioned, which now either return quickly or return -EINVAL.  I've also
testing using the piptest.c epoll tester, which showed no difference in
performance.  I've also created a number of different epoll networks and
tested that they behave as expectded.

I believe this solves the original diabolical test cases, while still
preserving the sane epoll nesting.

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Nelson Elhage <nelhage@ksplice.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agopipe: fail cleanly when root tries F_SETPIPE_SZ with big size
Sasha Levin [Fri, 13 Jan 2012 01:17:40 +0000 (17:17 -0800)]
pipe: fail cleanly when root tries F_SETPIPE_SZ with big size

When a user with the CAP_SYS_RESOURCE cap tries to F_SETPIPE_SZ a pipe
with size bigger than kmalloc() can alloc it spits out an ugly warning:

  ------------[ cut here ]------------
  WARNING: at mm/page_alloc.c:2095 __alloc_pages_nodemask+0x5d3/0x7a0()
  Pid: 733, comm: a.out Not tainted 3.2.0-rc1+ #4
  Call Trace:
     warn_slowpath_common+0x75/0xb0
     warn_slowpath_null+0x15/0x20
     __alloc_pages_nodemask+0x5d3/0x7a0
     __get_free_pages+0x12/0x50
     __kmalloc+0x12b/0x150
     pipe_set_size+0x75/0x120
     pipe_fcntl+0xf8/0x140
     do_fcntl+0x2d4/0x410
     sys_fcntl+0x66/0xa0
     system_call_fastpath+0x16/0x1b
  ---[ end trace 432f702e6db7b5ee ]---

Instead, make kcalloc() handle the overflow case and fail quietly.

[akpm@linux-foundation.org: switch to sizeof(*bufs) for 80-column niceness]
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Acked-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoslub: document setting min order with debug_guardpage_minorder > 0
Stanislaw Gruszka [Fri, 13 Jan 2012 01:17:39 +0000 (17:17 -0800)]
slub: document setting min order with debug_guardpage_minorder > 0

Acked-by: David Rientjes <rientjes@google.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoparisc, exec: remove redundant set_fs(USER_DS)
Mathias Krause [Fri, 13 Jan 2012 01:17:38 +0000 (17:17 -0800)]
parisc, exec: remove redundant set_fs(USER_DS)

The address limit is already set in flush_old_exec() so those calls to
set_fs(USER_DS) are redundant.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoia64, exec: remove redundant set_fs(USER_DS)
Mathias Krause [Fri, 13 Jan 2012 01:17:36 +0000 (17:17 -0800)]
ia64, exec: remove redundant set_fs(USER_DS)

The address limit is already set in flush_old_exec() so this
set_fs(USER_DS) is redundant.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agodrivers/video/nvidia/nvidia.c: fix warning
Andrew Morton [Fri, 13 Jan 2012 01:17:34 +0000 (17:17 -0800)]
drivers/video/nvidia/nvidia.c: fix warning

Fix the int/bool confusion in there.

  drivers/video/nvidia/nvidia.c:1602: warning: return from incompatible pointer type

Cc: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm,x86,um: move CMPXCHG_DOUBLE config option
Heiko Carstens [Fri, 13 Jan 2012 01:17:33 +0000 (17:17 -0800)]
mm,x86,um: move CMPXCHG_DOUBLE config option

Move CMPXCHG_DOUBLE and rename it to HAVE_CMPXCHG_DOUBLE so architectures
can simply select the option if it is supported.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm,x86,um: move CMPXCHG_LOCAL config option
Heiko Carstens [Fri, 13 Jan 2012 01:17:30 +0000 (17:17 -0800)]
mm,x86,um: move CMPXCHG_LOCAL config option

Move CMPXCHG_LOCAL and rename it to HAVE_CMPXCHG_LOCAL so architectures
can simply select the option if it is supported.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomm,slub,x86: decouple size of struct page from CONFIG_CMPXCHG_LOCAL
Heiko Carstens [Fri, 13 Jan 2012 01:17:27 +0000 (17:17 -0800)]
mm,slub,x86: decouple size of struct page from CONFIG_CMPXCHG_LOCAL

While implementing cmpxchg_double() on s390 I realized that we don't set
CONFIG_CMPXCHG_LOCAL despite the fact that we have support for it.

However setting that option will increase the size of struct page by
eight bytes on 64 bit, which we certainly do not want.  Also, it doesn't
make sense that a present cpu feature should increase the size of struct
page.

Besides that it looks like the dependency to CMPXCHG_LOCAL is wrong and
that it should depend on CMPXCHG_DOUBLE instead.

This patch:

If an architecture supports CMPXCHG_LOCAL this shouldn't result
automatically in larger struct pages if the SLUB allocator is used.
Instead introduce a new config option "HAVE_ALIGNED_STRUCT_PAGE" which
can be selected if a double word aligned struct page is required.  Also
update x86 Kconfig so that it should work as before.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoinclude/linux/linkage.h: remove unused ATTRIB_NORET macro
Joe Perches [Fri, 13 Jan 2012 01:17:25 +0000 (17:17 -0800)]
include/linux/linkage.h: remove unused ATTRIB_NORET macro

The uses have been renamed so delete the unused macro.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agotreewide: convert uses of ATTRIB_NORETURN to __noreturn
Joe Perches [Fri, 13 Jan 2012 01:17:21 +0000 (17:17 -0800)]
treewide: convert uses of ATTRIB_NORETURN to __noreturn

Use the more commonly used __noreturn instead of ATTRIB_NORETURN.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Joe Perches <joe@perches.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Haavard Skinnemoen <hskinnemoen@gmail.com>
Cc: Hans-Christian Egtvedt <egtvedt@samfundet.no>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agotreewide: remove useless NORET_TYPE macro and uses
Joe Perches [Fri, 13 Jan 2012 01:17:17 +0000 (17:17 -0800)]
treewide: remove useless NORET_TYPE macro and uses

It's a very old and now unused prototype marking so just delete it.

Neaten panic pointer argument style to keep checkpatch quiet.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Haavard Skinnemoen <hskinnemoen@gmail.com>
Cc: Hans-Christian Egtvedt <egtvedt@samfundet.no>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoinclude/linux/linkage.h: remove unused NORET_AND macro
Joe Perches [Fri, 13 Jan 2012 01:17:14 +0000 (17:17 -0800)]
include/linux/linkage.h: remove unused NORET_AND macro

The only use in kernel.h is gone so remove the macro.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokernel.h: neaten panic prototype
Joe Perches [Fri, 13 Jan 2012 01:17:13 +0000 (17:17 -0800)]
kernel.h: neaten panic prototype

Use __printf macro.
Convert NORET_AND to ATTRIB_NORET.
Use the normal kernel style for pointer arguments.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokprobes: silence DEBUG_STRICT_USER_COPY_CHECKS=y warning
Stephen Boyd [Fri, 13 Jan 2012 01:17:11 +0000 (17:17 -0800)]
kprobes: silence DEBUG_STRICT_USER_COPY_CHECKS=y warning

Enabling DEBUG_STRICT_USER_COPY_CHECKS causes the following warning:

  In file included from arch/x86/include/asm/uaccess.h:573,
                   from kernel/kprobes.c:55:
  In function 'copy_from_user',
      inlined from 'write_enabled_file_bool' at
      kernel/kprobes.c:2191:
  arch/x86/include/asm/uaccess_64.h:65:
  warning: call to 'copy_from_user_overflow' declared with attribute warning: copy_from_user() buffer size is not provably correct

presumably due to buf_size being signed causing GCC to fail to see that
buf_size can't become negative.

Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoproc: fix null pointer deref in proc_pid_permission()
Xiaotian Feng [Fri, 13 Jan 2012 01:17:08 +0000 (17:17 -0800)]
proc: fix null pointer deref in proc_pid_permission()

get_proc_task() can fail to search the task and return NULL,
put_task_struct() will then bomb the kernel with following oops:

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000010
  IP: [<ffffffff81217d34>] proc_pid_permission+0x64/0xe0
  PGD 112075067 PUD 112814067 PMD 0
  Oops: 0002 [#1] PREEMPT SMP

This is a regression introduced by commit 0499680a ("procfs: add hidepid=
and gid= mount options").  The kernel should return -ESRCH if
get_proc_task() failed.

Signed-off-by: Xiaotian Feng <dannyfeng@tencent.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Vasiliy Kulikov <segoon@openwall.com>
Cc: Stephen Wilson <wilsons@start.ca>
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>
12 years agox86: Get rid of 'dubious one-bit signed bitfield' sprase warning
Anton Vorontsov [Wed, 11 Jan 2012 01:11:46 +0000 (05:11 +0400)]
x86: Get rid of 'dubious one-bit signed bitfield' sprase warning

This very noisy sparse warning appears on almost every file in the
kernel:

  CHECK   init/main.c
  arch/x86/include/asm/thread_info.h:43:55: error: dubious one-bit signed bitfield
  arch/x86/include/asm/thread_info.h:44:46: error: dubious one-bit signed bitfield

This patch changes sig_on_uaccess_error and uaccess_err flags to unsigned
type and thus fixes the warning.

Signed-off-by: Anton Vorontsov <cbouatmailru@gmail.com>
Acked-by: Andy Lutomirski <luto@mit.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
Linus Torvalds [Thu, 12 Jan 2012 16:00:30 +0000 (08:00 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound: (526 commits)
  ASoC: twl6040 - Add method to query optimum PDM_DL1 gain
  ALSA: hda - Fix the lost power-setup of seconary pins after PM resume
  ALSA: usb-audio: add Yamaha MOX6/MOX8 support
  ALSA: virtuoso: add S/PDIF input support for all Xonars
  ALSA: ice1724 - Support for ooAoo SQ210a
  ALSA: ice1724 - Allow card info based on model only
  ALSA: ice1724 - Create capture pcm only for ADC-enabled configurations
  ALSA: hdspm - Provide unique driver id based on card serial
  ASoC: Dynamically allocate the rtd device for a non-empty release()
  ASoC: Fix recursive dependency due to select ATMEL_SSC in SND_ATMEL_SOC_SSC
  ALSA: hda - Fix the detection of "Loopback Mixing" control for VIA codecs
  ALSA: hda - Return the error from get_wcaps_type() for invalid NIDs
  ALSA: hda - Use auto-parser for HP laptops with cx20459 codec
  ALSA: asihpi - Fix potential Oops in snd_asihpi_cmode_info()
  ALSA: hdsp - Fix potential Oops in snd_hdsp_info_pref_sync_ref()
  ALSA: hda/cirrus - support for iMac12,2 model
  ASoC: cx20442: add bias control over a platform provided regulator
  ALSA: usb-audio - Avoid flood of frame-active debug messages
  ALSA: snd-usb-us122l: Delete calls to preempt_disable
  mfd: Put WM8994 into cache only mode when suspending
  ...

Fix up trivial conflicts in:
 - arch/arm/mach-s3c64xx/mach-crag6410.c:
renamed speyside_wm8962 to tobermory, added littlemill right
next to it
 - drivers/base/regmap/{regcache.c,regmap.c}:
duplicate diff that had already come in with other changes in
the regmap tree

12 years agox86/PCI: build amd_bus.o only when CONFIG_AMD_NB=y
Bjorn Helgaas [Thu, 12 Jan 2012 15:01:40 +0000 (08:01 -0700)]
x86/PCI: build amd_bus.o only when CONFIG_AMD_NB=y

We only need amd_bus.o for AMD systems with PCI.  arch/x86/pci/Makefile
already depends on CONFIG_PCI=y, so this patch just adds the dependency
on CONFIG_AMD_NB.

Cc: Yinghai Lu <yinghai@kernel.org>
Cc: stable@kernel.org # 2.6.34+ (needs adjustment for k8 -> amd rename)
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge branch 'topic/hda' into for-linus
Takashi Iwai [Thu, 12 Jan 2012 08:59:18 +0000 (09:59 +0100)]
Merge branch 'topic/hda' into for-linus

12 years agoMerge branch 'topic/misc' into for-linus
Takashi Iwai [Thu, 12 Jan 2012 08:59:14 +0000 (09:59 +0100)]
Merge branch 'topic/misc' into for-linus

12 years agoMerge branch 'for-3.3' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound...
Takashi Iwai [Thu, 12 Jan 2012 08:48:20 +0000 (09:48 +0100)]
Merge branch 'for-3.3' of git://git./linux/kernel/git/broonie/sound into topic/asoc

12 years agoMerge tag 'rmobile-for-linus' of git://github.com/pmundt/linux-sh
Linus Torvalds [Thu, 12 Jan 2012 07:29:20 +0000 (23:29 -0800)]
Merge tag 'rmobile-for-linus' of git://github.com/pmundt/linux-sh

SH/R-Mobile updates for 3.3 merge window.

* tag 'rmobile-for-linus' of git://github.com/pmundt/linux-sh: (32 commits)
  arm: mach-shmobile: add a resource name for shdma
  ARM: mach-shmobile: r8a7779 SMP support V3
  ARM: mach-shmobile: Add kota2 defconfig.
  ARM: mach-shmobile: Add marzen defconfig.
  ARM: mach-shmobile: r8a7779 power domain support V2
  ARM: mach-shmobile: Fix up marzen build for recent GIC changes.
  ARM: mach-shmobile: r8a7779 PFC function support
  ARM: mach-shmobile: Flush caches in platform_cpu_die()
  ARM: mach-shmobile: Allow SoC specific CPU kill code
  ARM: mach-shmobile: Fix headsmp.S code to use CPUINIT
  ARM: mach-shmobile: clock-r8a7779: clkz/clkzs support
  ARM: mach-shmobile: clock-r8a7779: add DIV4 clock support
  ARM: mach-shmobile: Marzen LAN89218 support
  ARM: mach-shmobile: Marzen SCIF2/SCIF4 support
  ARM: mach-shmobile: r8a7779 PFC GPIO-only support V2
  ARM: mach-shmobile: r8a7779 and Marzen base support V2
  sh: pfc: Unlock register support
  sh: pfc: Variable bitfield width config register support
  sh: pfc: Add config_reg_helper() function
  sh: pfc: Convert index to field and value pair
  ...

12 years agoMerge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh
Linus Torvalds [Thu, 12 Jan 2012 07:22:52 +0000 (23:22 -0800)]
Merge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh

SuperH updates for 3.3 merge window.

* tag 'sh-for-linus' of git://github.com/pmundt/linux-sh: (38 commits)
  sh: magicpanelr2: Update for parse_mtd_partitions() fallout.
  sh: mach-rsk: Update for parse_mtd_partitions() fallout.
  sh: sh2a: Improve cache flush/invalidate functions
  sh: also without PM_RUNTIME pm_runtime.o must be built
  sh: add a resource name for shdma
  sh: Remove redundant try_to_freeze() invocations.
  sh: Ensure IRQs are enabled across do_notify_resume().
  sh: Fix up store queue code for subsys_interface changes.
  sh: clkfwk: sh_clk_init_parent() should be called after clk_register()
  sh: add platform_device for renesas_usbhs in board-sh7757lcr
  sh: modify clock-sh7757 for renesas_usbhs
  sh: pfc: ioremap() support
  sh: use ioread32/iowrite32 and mapped_reg for div6
  sh: use ioread32/iowrite32 and mapped_reg for div4
  sh: use ioread32/iowrite32 and mapped_reg for mstp32
  sh: extend clock struct with mapped_reg member
  sh: clkfwk: clock-sh73a0: all div6_clks use SH_CLK_DIV6_EXT()
  sh: clkfwk: clock-sh7724: all div6_clks use SH_CLK_DIV6_EXT()
  sh: clock-sh7723: add CLKDEV_ICK_ID for cleanup
  serial: sh-sci: Handle GPIO function requests.
  ...

12 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 12 Jan 2012 06:52:48 +0000 (22:52 -0800)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: Fix lockup by limiting load-balance retries on lock-break
  sched: Fix CONFIG_CGROUP_SCHED dependency
  sched: Remove empty #ifdefs

12 years agosh: magicpanelr2: Update for parse_mtd_partitions() fallout.
Paul Mundt [Thu, 12 Jan 2012 04:49:05 +0000 (13:49 +0900)]
sh: magicpanelr2: Update for parse_mtd_partitions() fallout.

Follows the RSK+ change for the same rationale.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
12 years agosh: mach-rsk: Update for parse_mtd_partitions() fallout.
Paul Mundt [Thu, 12 Jan 2012 04:47:42 +0000 (13:47 +0900)]
sh: mach-rsk: Update for parse_mtd_partitions() fallout.

The RSK+ setup code was doing some pretty dubious things with
parse_mtd_partitions() in order to populate the physmap-flash map
platform data. The physmap-flash driver contains all of the functionality
that we require already, so simply drop the special casing and pad out
the platform data accordingly.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
12 years agoMerge branch 'sh/nommu' into sh-latest
Paul Mundt [Thu, 12 Jan 2012 04:11:43 +0000 (13:11 +0900)]
Merge branch 'sh/nommu' into sh-latest