GitHub/exynos8895/android_kernel_samsung_universal8895.git
11 years agoMerge branch 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
Linus Torvalds [Tue, 30 Apr 2013 02:07:40 +0000 (19:07 -0700)]
Merge branch 'for-3.10' of git://git./linux/kernel/git/tj/wq

Pull workqueue updates from Tejun Heo:
 "A lot of activities on workqueue side this time.  The changes achieve
  the followings.

   - WQ_UNBOUND workqueues - the workqueues which are per-cpu - are
     updated to be able to interface with multiple backend worker pools.
     This involved a lot of churning but the end result seems actually
     neater as unbound workqueues are now a lot closer to per-cpu ones.

   - The ability to interface with multiple backend worker pools are
     used to implement unbound workqueues with custom attributes.
     Currently the supported attributes are the nice level and CPU
     affinity.  It may be expanded to include cgroup association in
     future.  The attributes can be specified either by calling
     apply_workqueue_attrs() or through /sys/bus/workqueue/WQ_NAME/* if
     the workqueue in question is exported through sysfs.

     The backend worker pools are keyed by the actual attributes and
     shared by any workqueues which share the same attributes.  When
     attributes of a workqueue are changed, the workqueue binds to the
     worker pool with the specified attributes while leaving the work
     items which are already executing in its previous worker pools
     alone.

     This allows converting custom worker pool implementations which
     want worker attribute tuning to use workqueues.  The writeback pool
     is already converted in block tree and there are a couple others
     are likely to follow including btrfs io workers.

   - WQ_UNBOUND's ability to bind to multiple worker pools is also used
     to make it NUMA-aware.  Because there's no association between work
     item issuer and the specific worker assigned to execute it, before
     this change, using unbound workqueue led to unnecessary cross-node
     bouncing and it couldn't be helped by autonuma as it requires tasks
     to have implicit node affinity and workers are assigned randomly.

     After these changes, an unbound workqueue now binds to multiple
     NUMA-affine worker pools so that queued work items are executed in
     the same node.  This is turned on by default but can be disabled
     system-wide or for individual workqueues.

     Crypto was requesting NUMA affinity as encrypting data across
     different nodes can contribute noticeable overhead and doing it
     per-cpu was too limiting for certain cases and IO throughput could
     be bottlenecked by one CPU being fully occupied while others have
     idle cycles.

  While the new features required a lot of changes including
  restructuring locking, it didn't complicate the execution paths much.
  The unbound workqueue handling is now closer to per-cpu ones and the
  new features are implemented by simply associating a workqueue with
  different sets of backend worker pools without changing queue,
  execution or flush paths.

  As such, even though the amount of change is very high, I feel
  relatively safe in that it isn't likely to cause subtle issues with
  basic correctness of work item execution and handling.  If something
  is wrong, it's likely to show up as being associated with worker pools
  with the wrong attributes or OOPS while workqueue attributes are being
  changed or during CPU hotplug.

  While this creates more backend worker pools, it doesn't add too many
  more workers unless, of course, there are many workqueues with unique
  combinations of attributes.  Assuming everything else is the same,
  NUMA awareness costs an extra worker pool per NUMA node with online
  CPUs.

  There are also a couple things which are being routed outside the
  workqueue tree.

   - block tree pulled in workqueue for-3.10 so that writeback worker
     pool can be converted to unbound workqueue with sysfs control
     exposed.  This simplifies the code, makes writeback workers
     NUMA-aware and allows tuning nice level and CPU affinity via sysfs.

   - The conversion to workqueue means that there's no 1:1 association
     between a specific worker, which makes writeback folks unhappy as
     they want to be able to tell which filesystem caused a problem from
     backtrace on systems with many filesystems mounted.  This is
     resolved by allowing work items to set debug info string which is
     printed when the task is dumped.  As this change involves unifying
     implementations of dump_stack() and friends in arch codes, it's
     being routed through Andrew's -mm tree."

* 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (84 commits)
  workqueue: use kmem_cache_free() instead of kfree()
  workqueue: avoid false negative WARN_ON() in destroy_workqueue()
  workqueue: update sysfs interface to reflect NUMA awareness and a kernel param to disable NUMA affinity
  workqueue: implement NUMA affinity for unbound workqueues
  workqueue: introduce put_pwq_unlocked()
  workqueue: introduce numa_pwq_tbl_install()
  workqueue: use NUMA-aware allocation for pool_workqueues
  workqueue: break init_and_link_pwq() into two functions and introduce alloc_unbound_pwq()
  workqueue: map an unbound workqueues to multiple per-node pool_workqueues
  workqueue: move hot fields of workqueue_struct to the end
  workqueue: make workqueue->name[] fixed len
  workqueue: add workqueue->unbound_attrs
  workqueue: determine NUMA node of workers accourding to the allowed cpumask
  workqueue: drop 'H' from kworker names of unbound worker pools
  workqueue: add wq_numa_tbl_len and wq_numa_possible_cpumask[]
  workqueue: move pwq_pool_locking outside of get/put_unbound_pool()
  workqueue: fix memory leak in apply_workqueue_attrs()
  workqueue: fix unbound workqueue attrs hashing / comparison
  workqueue: fix race condition in unbound workqueue free path
  workqueue: remove pwq_lock which is no longer used
  ...

11 years agoMerge branch 'for-3.10-async' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
Linus Torvalds [Tue, 30 Apr 2013 02:06:59 +0000 (19:06 -0700)]
Merge branch 'for-3.10-async' of git://git./linux/kernel/git/tj/wq

Pull async update from Tejun Heo:
 "This contains three cleanup patches for async from Lai.  All three
  patches are essentially cosmetic."

* 'for-3.10-async' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
  async: rename and redefine async_func_ptr
  async: remove unused @node from struct async_domain
  async: simplify lowest_in_progress()

11 years agoMerge branch 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu
Linus Torvalds [Tue, 30 Apr 2013 02:06:16 +0000 (19:06 -0700)]
Merge branch 'for-3.10' of git://git./linux/kernel/git/tj/percpu

Pull percpu patch from Tejun Heo:
 "A puny pull request for percpu.  We were expecting more cleanup
  patches but didn't happen this time, so just a single patch adding
  documentation from Christoph."

* 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu:
  percpu: add documentation on this_cpu operations

11 years agoMerge branch 'akpm' (incoming from Andrew)
Linus Torvalds [Tue, 30 Apr 2013 00:29:08 +0000 (17:29 -0700)]
Merge branch 'akpm' (incoming from Andrew)

Merge first batch of fixes from Andrew Morton:

 - A couple of kthread changes

 - A few minor audit patches

 - A number of fbdev patches.  Florian remains AWOL so I'm picking up
   some of these.

 - A few kbuild things

 - ocfs2 updates

 - Almost all of the MM queue

(And in the meantime, I already have the second big batch from Andrew
pending in my mailbox ;^)

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (149 commits)
  memcg: take reference before releasing rcu_read_lock
  mem hotunplug: fix kfree() of bootmem memory
  mmKconfig: add an option to disable bounce
  mm, nobootmem: do memset() after memblock_reserve()
  mm, nobootmem: clean-up of free_low_memory_core_early()
  fs/buffer.c: remove unnecessary init operation after allocating buffer_head.
  numa, cpu hotplug: change links of CPU and node when changing node number by onlining CPU
  mm: fix memory_hotplug.c printk format warning
  mm: swap: mark swap pages writeback before queueing for direct IO
  swap: redirty page if page write fails on swap file
  mm, memcg: give exiting processes access to memory reserves
  thp: fix huge zero page logic for page with pfn == 0
  memcg: avoid accessing memcg after releasing reference
  fs: fix fsync() error reporting
  memblock: fix missing comment of memblock_insert_region()
  mm: Remove unused parameter of pages_correctly_reserved()
  firmware, memmap: fix firmware_map_entry leak
  mm/vmstat: add note on safety of drain_zonestat
  mm: thp: add split tail pages to shrink page list in page reclaim
  mm: allow for outstanding swap writeback accounting
  ...

11 years agoMerge tag 'clk-for-linus-3.10' of git://git.linaro.org/people/mturquette/linux
Linus Torvalds [Mon, 29 Apr 2013 23:43:54 +0000 (16:43 -0700)]
Merge tag 'clk-for-linus-3.10' of git://git.linaro.org/people/mturquette/linux

Pull clock framework update from Michael Turquette:
 "The common clock framework changes for 3.10 include many fixes for
  existing platforms, as well as adoption of the framework by new
  platforms and devices.

  Some long-needed fixes to the core framework are here as well as new
  features such as improved initialization of clocks from DT as well as
  framework reentrancy for nested clock operations."

* tag 'clk-for-linus-3.10' of git://git.linaro.org/people/mturquette/linux: (44 commits)
  clk: add clk_ignore_unused option to keep boot clocks on
  clk: ux500: fix mismatched types
  clk: vexpress: Add separate SP810 driver
  clk: si5351: make clk-si5351 depend on CONFIG_OF
  clk: export __clk_get_flags for modular clock providers
  clk: vt8500: Missing breaks in vtwm_pll_round_rate/_set_rate.
  clk: sunxi: Unify oscillator clock
  clk: composite: allow fixed rates & fixed dividers
  clk: composite: rename 'div' references to 'rate'
  clk: add si5351 i2c common clock driver
  clk: add device tree fixed-factor-clock binding support
  clk: Properly handle notifier return values
  clk: ux500: abx500: Define clock tree for ab850x
  clk: ux500: Add support for sysctrl clocks
  clk: mvebu: Fix valid value range checking for cpu_freq_select
  clk: Fixup locking issues for clk_set_parent
  clk: Fixup errorhandling for clk_set_parent
  clk: Restructure code for __clk_reparent
  clk: sunxi: drop an unnecesary kmalloc
  clk: sunxi: drop CLK_IGNORE_UNUSED
  ...

11 years agoMerge tag 'spi-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi
Linus Torvalds [Mon, 29 Apr 2013 23:38:41 +0000 (16:38 -0700)]
Merge tag 'spi-v3.10' of git://git./linux/kernel/git/broonie/spi

Pull spi updates from Mark Brown:
 "A fairly quiet release for SPI, mainly driver work.  A few highlights:

   - Supports bits per word compatibility checking in the core.
   - Allow use of the IP used in Freescale SPI controllers outside
     Freescale SoCs.
   - DMA support for the Atmel SPI driver.
   - New drivers for the BCM2835 and Tegra114"

* tag 'spi-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi: (68 commits)
  spi-topcliff-pch: fix to use list_for_each_entry_safe() when delete list items
  spi-topcliff-pch: missing platform_driver_unregister() on error in pch_spi_init()
  ARM: dts: add pinctrl property for spi node for atmel SoC
  ARM: dts: add spi nodes for the atmel boards
  ARM: dts: add spi nodes for atmel SoC
  ARM: at91: add clocks for spi dt entries
  spi/spi-atmel: add dmaengine support
  spi/spi-atmel: add flag to controller data for lock operations
  spi/spi-atmel: add physical base address
  spi/sirf: fix MODULE_DEVICE_TABLE
  MAINTAINERS: Add git repository and update my address
  spi/s3c64xx: Check for errors in dmaengine prepare_transfer()
  spi/s3c64xx: Fix non-dmaengine usage
  spi: omap2-mcspi: fix error return code in omap2_mcspi_probe()
  spi/s3c64xx: let device core setup the default pin configuration
  MAINTAINERS: Update Grant's email address and maintainership
  spi: omap2-mcspi: Fix transfers if DMADEVICES is not set
  spi: s3c64xx: move to generic dmaengine API
  spi-gpio: init CS before spi_bitbang_setup()
  spi: spi-mpc512x-psc: let transmiter/receiver enabled when in xfer loop
  ...

11 years agoMerge tag 'regulator-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Mon, 29 Apr 2013 23:32:25 +0000 (16:32 -0700)]
Merge tag 'regulator-v3.10' of git://git./linux/kernel/git/broonie/regulator

Pull regulator updates from Mark Brown:
 "The diffstat and changelog here is dominated by Lee Jones' heroic
  efforts to sync the ab8500 driver that's been maintained out of tree
  with mainline (plus Axel's cleanup work on the results) but there's a
  few other things here:

   - Axel Lin added regulator_map_voltage_ascend() optimising a common
     pattern for drivers using the core code.
   - Milo Kim tought the regulator core to handle regulators sharing an
     enable GPIO, avoiding the need to do hacks to support such systems.
   - Andrew Bresticker added code to handle missing supplies for
     regulators more sensibly for device tree systems, reducing the need
     for stubbing there.

  plus the usual batch of driver specific updates and fixes"

* tag 'regulator-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator: (152 commits)
  regulator: mc13892: Fix MC13892_SWITCHERS0_SWxHI bit in set_voltage_sel
  regulator: Remove NULL test before calling regulator_unregister()
  regulator: mc13783: Add device tree probe support
  regulator: mc13xxx: Add warning of incorrect names of regulators
  regulator: max77686: Don't update max77686->opmode if update register fails
  regulator: max8952: Add missing config.of_node setting for regulator register
  regulator: ab3100: Fix regulator register error handling
  regulator: tps6524x: Use regulator_map_voltage_ascend
  regulator: lp8788-buck: Use regulator_map_voltage_ascend
  regulator: lp872x: Use regulator_map_voltage_ascend
  regulator: mc13892: Use regulator_map_voltage_ascend for mc13892_sw_regulator_ops
  regulator: tps65023: Use regulator_map_voltage_ascend
  regulator: tps65023: Merge tps65020 ldo1 and ldo2 vsel table
  regulator: tps6507x: Use regulator_map_voltage_ascend
  regulator: mc13892: Fix MC13892_SWITCHERS0_SWxHI bit in set_voltage_sel
  regulator: ab3100: device tree support
  regulator: ab3100: refactor probe to use IDs
  regulator: max8973: Don't override control1 variable when set ramp delay bits
  regulator: tps80031: Convert tps80031_dcdc_ops to [get|set]_voltage_sel_regmap
  regulator: tps80031: Fix LDO2 track mode for TPS80031 or TPS80032-ES1.0
  ...

11 years agoMerge tag 'regmap-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Mon, 29 Apr 2013 23:31:26 +0000 (16:31 -0700)]
Merge tag 'regmap-v3.10' of git://git./linux/kernel/git/broonie/regmap

Pull regmap updates from Mark Brown:
 "In user visible terms just a couple of enhancements here, though there
  was a moderate amount of refactoring required in order to support the
  register cache sync performance improvements.

   - Support for block and asynchronous I/O during register cache
     syncing; this provides a use case dependant performance
     improvement.
   - Additional debugfs information on the memory consuption and
     register set"

* tag 'regmap-v3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap: (23 commits)
  regmap: don't corrupt work buffer in _regmap_raw_write()
  regmap: cache: Fix format specifier in dev_dbg
  regmap: cache: Make regcache_sync_block_raw static
  regmap: cache: Write consecutive registers in a single block write
  regmap: cache: Split raw and non-raw syncs
  regmap: cache: Factor out block sync
  regmap: cache: Factor out reg_present support from rbtree cache
  regmap: cache: Use raw I/O to sync rbtrees if we can
  regmap: core: Provide regmap_can_raw_write() operation
  regmap: cache: Provide a get address of value operation
  regmap: Cut down on the average # of nodes in the rbtree cache
  regmap: core: Make raw write available to regcache
  regmap: core: Warn on invalid operation combinations
  regmap: irq: Clarify error message when we fail to request primary IRQ
  regmap: rbtree Expose total memory consumption in the rbtree debugfs entry
  regmap: debugfs: Add a registers `range' file
  regmap: debugfs: Simplify calculation of `c->max_reg'
  regmap: cache: Store caches in native register format where possible
  regmap: core: Split out in place value parsing
  regmap: cache: Use regcache_get_value() to check if we updated
  ...

11 years agomemcg: take reference before releasing rcu_read_lock
Li Zefan [Mon, 29 Apr 2013 22:08:57 +0000 (15:08 -0700)]
memcg: take reference before releasing rcu_read_lock

The memcg is not referenced, so it can be destroyed at anytime right
after we exit rcu read section, so it's not safe to access it.

To fix this, we call css_tryget() to get a reference while we're still
in rcu read section.

This also removes a bogus comment above __memcg_create_cache_enqueue().

Signed-off-by: Li Zefan <lizefan@huawei.com>
Acked-by: Glauber Costa <glommer@parallels.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomem hotunplug: fix kfree() of bootmem memory
Yasuaki Ishimatsu [Mon, 29 Apr 2013 22:08:56 +0000 (15:08 -0700)]
mem hotunplug: fix kfree() of bootmem memory

When hot removing memory presented at boot time, following messages are shown:

  kernel BUG at mm/slub.c:3409!
  invalid opcode: 0000 [#1] SMP
  Modules linked in: ebtable_nat ebtables xt_CHECKSUM iptable_mangle bridge stp llc ipmi_devintf ipmi_msghandler sunrpc ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc vfat fat dm_mirror dm_region_hash dm_log dm_mod vhost_net macvtap macvlan tun uinput iTCO_wdt iTCO_vendor_support coretemp kvm_intel kvm crc32c_intel ghash_clmulni_intel microcode pcspkr sg i2c_i801 lpc_ich mfd_core igb i2c_algo_bit i2c_core e1000e ptp pps_core tpm_infineon ioatdma dca sr_mod cdrom sd_mod crc_t10dif usb_storage megaraid_sas lpfc scsi_transport_fc scsi_tgt scsi_mod
  CPU 0
  Pid: 5091, comm: kworker/0:2 Tainted: G        W    3.9.0-rc6+ #15
  RIP: kfree+0x232/0x240
  Process kworker/0:2 (pid: 5091, threadinfo ffff88084678c000, task ffff88083928ca80)
  Call Trace:
    __release_region+0xd4/0xe0
    __remove_pages+0x52/0x110
    arch_remove_memory+0x89/0xd0
    remove_memory+0xc4/0x100
    acpi_memory_device_remove+0x6d/0xb1
    acpi_device_remove+0x89/0xab
    __device_release_driver+0x7c/0xf0
    device_release_driver+0x2f/0x50
    acpi_bus_device_detach+0x6c/0x70
    acpi_ns_walk_namespace+0x11a/0x250
    acpi_walk_namespace+0xee/0x137
    acpi_bus_trim+0x33/0x7a
    acpi_bus_hot_remove_device+0xc4/0x1a1
    acpi_os_execute_deferred+0x27/0x34
    process_one_work+0x1f7/0x590
    worker_thread+0x11a/0x370
    kthread+0xee/0x100
    ret_from_fork+0x7c/0xb0
  RIP  [<ffffffff811c41d2>] kfree+0x232/0x240
   RSP <ffff88084678d968>

The reason why the messages are shown is to release a resource
structure, allocated by bootmem, by kfree().  So when we release a
resource structure, we should check whether it is allocated by bootmem
or not.

But even if we know a resource structure is allocated by bootmem, we
cannot release it since SLxB cannot treat it.  So for reusing a resource
structure, this patch remembers it by using bootmem_resource as follows:

When releasing a resource structure by free_resource(), free_resource()
checks whether the resource structure is allocated by bootmem or not.
If it is allocated by bootmem, free_resource() adds it to
bootmem_resource.  If it is not allocated by bootmem, free_resource()
release it by kfree().

And when getting a new resource structure by get_resource(),
get_resource() checks whether bootmem_resource has released resource
structures or not.  If there is a released resource structure,
get_resource() returns it.  If there is not a releaed resource
structure, get_resource() returns new resource structure allocated by
kzalloc().

[akpm@linux-foundation.org: s/get_resource/alloc_resource/]
Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Reviewed-by: Toshi Kani <toshi.kani@hp.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ram Pai <linuxram@us.ibm.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>
11 years agommKconfig: add an option to disable bounce
Vinayak Menon [Mon, 29 Apr 2013 22:08:55 +0000 (15:08 -0700)]
mmKconfig: add an option to disable bounce

There are times when HIGHMEM is enabled, but we don't prefer
CONFIG_BOUNCE to be enabled.  CONFIG_BOUNCE can reduce the block device
throughput, and this is not ideal for machines where we don't gain much
by enabling it.  So provide an option to deselect CONFIG_BOUNCE.  The
observation was made while measuring eMMC throughput using iozone on an
ARM device with 1GB RAM.

Signed-off-by: Vinayak Menon <vinayakm.list@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, nobootmem: do memset() after memblock_reserve()
Joonsoo Kim [Mon, 29 Apr 2013 22:08:53 +0000 (15:08 -0700)]
mm, nobootmem: do memset() after memblock_reserve()

Currently, we do memset() before reserving the area.  This may not cause
any problem, but it is somewhat weird.  So change execution order.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jiang Liu <liuj97@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, nobootmem: clean-up of free_low_memory_core_early()
Joonsoo Kim [Mon, 29 Apr 2013 22:08:52 +0000 (15:08 -0700)]
mm, nobootmem: clean-up of free_low_memory_core_early()

Remove unused argument and make function static, because there is no user
outside of nobootmem.c

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jiang Liu <liuj97@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/buffer.c: remove unnecessary init operation after allocating buffer_head.
majianpeng [Mon, 29 Apr 2013 22:08:51 +0000 (15:08 -0700)]
fs/buffer.c: remove unnecessary init operation after allocating buffer_head.

bh allocation uses kmem_cache_zalloc() so we needn't call
'init_buffer(bh, NULL, NULL)' and perform other set-zero-operations.

Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agonuma, cpu hotplug: change links of CPU and node when changing node number by onlining CPU
Yasuaki Ishimatsu [Mon, 29 Apr 2013 22:08:50 +0000 (15:08 -0700)]
numa, cpu hotplug: change links of CPU and node when changing node number by onlining CPU

When booting x86 system contains memoryless node, node numbers of CPUs
on memoryless node were changed to nearest online node number by
init_cpu_to_node() because the node is not online.

In my system, node numbers of cpu#30-44 and 75-89 were changed from 2 to
0 as follows:

  $ numactl --hardware
  available: 2 nodes (0-1)
  node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 30 31 32 33 34 35 36 37 38 39 40
  41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 75 76 77 78 79 80 81 82
  83 84 85 86 87 88 89
  node 0 size: 32394 MB
  node 0 free: 27898 MB
  node 1 cpus: 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 60 61 62 63 64 65 66
  67 68 69 70 71 72 73 74
  node 1 size: 32768 MB
  node 1 free: 30335 MB

If we hot add memory to memoryless node and offine/online all CPUs on
the node, node numbers of these CPUs are changed to correct node numbers
by srat_detect_node() because the node become online.

In this case, node numbers of cpu#30-44 and 75-89 were changed from 0 to
2 in my system as follows:

  $ numactl --hardware
  available: 3 nodes (0-2)
  node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 45 46 47 48 49 50 51 52 53 54 55
  56 57 58 59
  node 0 size: 32394 MB
  node 0 free: 27218 MB
  node 1 cpus: 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 60 61 62 63 64 65 66
  67 68 69 70 71 72 73 74
  node 1 size: 32768 MB
  node 1 free: 30014 MB
  node 2 cpus: 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 75 76 77 78 79 80 81
  82 83 84 85 86 87 88 89
  node 2 size: 16384 MB
  node 2 free: 16384 MB

But "cpu to node" and "node to cpu" links were not changed as follows:

  $ ls /sys/devices/system/cpu/cpu30/|grep node
  node0
  $ ls /sys/devices/system/node/node0/|grep cpu30
  cpu30

"numactl --hardware" shows that cpu30 belongs to node 2.  But sysfs
links does not change.

This patch changes "cpu to node" and "node to cpu" links when node
number changed by onlining CPU.

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: fix memory_hotplug.c printk format warning
Randy Dunlap [Mon, 29 Apr 2013 22:08:49 +0000 (15:08 -0700)]
mm: fix memory_hotplug.c printk format warning

PFN_PHYS() is a phys_addr_t, which can be u32 or u64.
Fix the build warning when phys_addr_t is u32.

  mm/memory_hotplug.c: warning: format '%llx' expects argument of type 'long long unsigned int', but argument 2 has type 'unsigned int' [-Wformat]:  => 1685:3
  mm/memory_hotplug.c: warning: format '%llx' expects argument of type 'long long unsigned int', but argument 3 has type 'unsigned int' [-Wformat]:  => 1685:3

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: swap: mark swap pages writeback before queueing for direct IO
Mel Gorman [Mon, 29 Apr 2013 22:08:48 +0000 (15:08 -0700)]
mm: swap: mark swap pages writeback before queueing for direct IO

As pointed out by Andrew Morton, the swap-over-NFS writeback is not
setting PageWriteback before it is queued for direct IO.  While swap
pages do not participate in BDI or process dirty accounting and the IO
is synchronous, the writeback bit is still required and not setting it
in this case was an oversight.  swapoff depends on the page writeback to
synchronoise all pending writes on a swap page before it is reused.
Swapcache freeing and reuse depend on checking the PageWriteback under
lock to ensure the page is safe to reuse.

Direct IO handlers and the direct IO handler for NFS do not deal with
PageWriteback as they are synchronous writes.  In the case of NFS, it
schedules pages (or a page in the case of swap) for IO and then waits
synchronously for IO to complete in nfs_direct_write().  It is
recognised that this is a slowdown from normal swap handling which is
asynchronous and uses a completion handler.  Shoving PageWriteback
handling down into direct IO handlers looks like a bad fit to handle the
swap case although it may have to be dealt with some day if swap is
converted to use direct IO in general and bmap is finally done away
with.  At that point it will be necessary to refit asynchronous direct
IO with completion handlers onto the swap subsystem.

As swapcache currently depends on PageWriteback to protect against
races, this patch sets PageWriteback under the page lock before queueing
it for direct IO.  It is cleared when the direct IO handler returns.  IO
errors are treated similarly to the direct-to-bio case except PageError
is not set as in the case of swap-over-NFS, it is likely to be a
transient error.

It was asked what prevents such a page being reclaimed in parallel.
With this patch applied, such a page will now be skipped (most of the
time) or blocked until the writeback completes.  Reclaim checks
PageWriteback under the page lock before calling try_to_free_swap and
the page lock should prevent the page being requeued for IO before it is
freed.

This and Jerome's related patch should considered for -stable as far
back as 3.6 when swap-over-NFS was introduced.

[akpm@linux-foundation.org: use pr_err_ratelimited()]
[akpm@linux-foundation.org: remove hopefully-unneeded cast in printk]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [3.6+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoswap: redirty page if page write fails on swap file
Jerome Marchand [Mon, 29 Apr 2013 22:08:47 +0000 (15:08 -0700)]
swap: redirty page if page write fails on swap file

Since commit 62c230bc1790 ("mm: add support for a filesystem to activate
swap files and use direct_IO for writing swap pages"), swap_writepage()
calls direct_IO on swap files.  However, in that case the page isn't
redirtied if I/O fails, and is therefore handled afterwards as if it has
been successfully written to the swap file, leading to memory corruption
when the page is eventually swapped back in.

This patch sets the page dirty when direct_IO() fails.  It fixes a
memory corruption that happened while using swap-over-NFS.

Signed-off-by: Jerome Marchand <jmarchan@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [3.6+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, memcg: give exiting processes access to memory reserves
David Rientjes [Mon, 29 Apr 2013 22:08:45 +0000 (15:08 -0700)]
mm, memcg: give exiting processes access to memory reserves

A memcg may livelock when oom if the process that grabs the hierarchy's
oom lock is never the first process with PF_EXITING set in the memcg's
task iteration.

The oom killer, both global and memcg, will defer if it finds an
eligible process that is in the process of exiting and it is not being
ptraced.  The idea is to allow it to exit without using memory reserves
before needlessly killing another process.

This normally works fine except in the memcg case with a large number of
threads attached to the oom memcg.  In this case, the memcg oom killer
only gets called for the process that grabs the hierarchy's oom lock;
all others end up blocked on the memcg's oom waitqueue.  Thus, if the
process that grabs the hierarchy's oom lock is never the first
PF_EXITING process in the memcg's task iteration, the oom killer is
constantly deferred without anything making progress.

The fix is to give PF_EXITING processes access to memory reserves so
that we've marked them as oom killed without any iteration.  This allows
__mem_cgroup_try_charge() to succeed so that the process may exit.  This
makes the memcg oom killer exemption for TIF_MEMDIE tasks, now
immediately granted for processes with pending SIGKILLs and those in the
exit path, to be equivalent to what is done for the global oom killer.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agothp: fix huge zero page logic for page with pfn == 0
Kirill A. Shutemov [Mon, 29 Apr 2013 22:08:44 +0000 (15:08 -0700)]
thp: fix huge zero page logic for page with pfn == 0

Current implementation of huge zero page uses pfn value 0 to indicate
that the page hasn't allocated yet.  It assumes that buddy page
allocator can't return page with pfn == 0.

Let's rework the code to store 'struct page *' of huge zero page, not
its pfn.  This way we can avoid the weak assumption.

[akpm@linux-foundation.org: fix sparse warning]
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Minchan Kim <minchan@kernel.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: avoid accessing memcg after releasing reference
Li Zefan [Mon, 29 Apr 2013 22:08:43 +0000 (15:08 -0700)]
memcg: avoid accessing memcg after releasing reference

This might cause a use-after-free bug.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs: fix fsync() error reporting
Dmitry Monakhov [Mon, 29 Apr 2013 22:08:42 +0000 (15:08 -0700)]
fs: fix fsync() error reporting

There are two convenient ways to report errors to userspace

1) retun error to original syscall for example write(2)
2) mark mapping with error flag and return it on later fsync(2)

Second one is broken if (mapping->nrpages == 0) This is real-life
situation because after error pages are likey to be truncated or
invalidated.

We have to return an error regardless to number of pages in the mapping.

#Original testcase: git@github.com:dmonakhov/xfstests.git
MOUNT_OPTIONS="-b1024"
./check shared/305

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Reviewed-by: Jan Kara <jack@suse.cz>
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>
11 years agomemblock: fix missing comment of memblock_insert_region()
Tang Chen [Mon, 29 Apr 2013 22:08:41 +0000 (15:08 -0700)]
memblock: fix missing comment of memblock_insert_region()

There is no comment for parameter nid of memblock_insert_region().
This patch adds comment for it.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: Remove unused parameter of pages_correctly_reserved()
Tang Chen [Mon, 29 Apr 2013 22:08:40 +0000 (15:08 -0700)]
mm: Remove unused parameter of pages_correctly_reserved()

nr_pages is not used in pages_correctly_reserved().
So remove it.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Reviewed-by: Wang Shilong <wangsl-fnst@cn.fujitsu.com>
Reviewed-by: Wen Congyang <wency@cn.fujitsu.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofirmware, memmap: fix firmware_map_entry leak
Yasuaki Ishimatsu [Mon, 29 Apr 2013 22:08:39 +0000 (15:08 -0700)]
firmware, memmap: fix firmware_map_entry leak

When hot removing memory, a firmware_map_entry which has memory range of
the memory is released by release_firmware_map_entry().  If the entry is
allocated by bootmem, release_firmware_map_entry() adds the entry to
map_entires_bootmem list when firmware_map_find_entry() finds the entry
from map_entries list.  But firmware_map_find_entry never find the entry
sicne map_entires list does not have the entry.  So the entry just
leaks.

Here are steps of leaking firmware_map_entry:
firmware_map_remove()
-> firmware_map_find_entry()
   Find released entry from map_entries list
-> firmware_map_remove_entry()
   Delete the entry from map_entries list
-> remove_sysfs_fw_map_entry()
   ...
   -> release_firmware_map_entry()
      -> firmware_map_find_entry()
         Find the entry from map_entries list but the entry has been
         deleted from map_entries list. So the entry is not added
         to map_entries_bootmem. Thus the entry leaks

release_firmware_map_entry() should not call firmware_map_find_entry()
since releaed entry has been deleted from map_entries list.  So the
patch delete firmware_map_find_entry() from releae_firmware_map_entry()

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Reviewed-by: Tang Chen <tangchen@cn.fujitsu.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/vmstat: add note on safety of drain_zonestat
Cody P Schafer [Mon, 29 Apr 2013 22:08:38 +0000 (15:08 -0700)]
mm/vmstat: add note on safety of drain_zonestat

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: thp: add split tail pages to shrink page list in page reclaim
Shaohua Li [Mon, 29 Apr 2013 22:08:36 +0000 (15:08 -0700)]
mm: thp: add split tail pages to shrink page list in page reclaim

In page reclaim, huge page is split.  split_huge_page() adds tail pages
to LRU list.  Since we are reclaiming a huge page, it's better we
reclaim all subpages of the huge page instead of just the head page.
This patch adds split tail pages to shrink page list so the tail pages
can be reclaimed soon.

Before this patch, run a swap workload:
  thp_fault_alloc 3492
  thp_fault_fallback 608
  thp_collapse_alloc 6
  thp_collapse_alloc_failed 0
  thp_split 916

With this patch:
  thp_fault_alloc 4085
  thp_fault_fallback 16
  thp_collapse_alloc 90
  thp_collapse_alloc_failed 0
  thp_split 1272

fallback allocation is reduced a lot.

[akpm@linux-foundation.org: fix CONFIG_SWAP=n build]
Signed-off-by: Shaohua Li <shli@fusionio.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.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>
11 years agomm: allow for outstanding swap writeback accounting
Seth Jennings [Mon, 29 Apr 2013 22:08:35 +0000 (15:08 -0700)]
mm: allow for outstanding swap writeback accounting

To prevent flooding the swap device with writebacks, frontswap backends
need to count and limit the number of outstanding writebacks.  The
incrementing of the counter can be done before the call to
__swap_writepage().  However, the caller must receive a notification
when the writeback completes in order to decrement the counter.

To achieve this functionality, this patch modifies __swap_writepage() to
take the bio completion callback function as an argument.

end_swap_bio_write(), the normal bio completion function, is also made
non-static so that code doing the accounting can call it after the
accounting is done.

There should be no behavioural change to existing code.

Signed-off-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: break up swap_writepage() for frontswap backends
Seth Jennings [Mon, 29 Apr 2013 22:08:34 +0000 (15:08 -0700)]
mm: break up swap_writepage() for frontswap backends

swap_writepage() is currently where frontswap hooks into the swap write
path to capture pages with the frontswap_store() function.  However, if
a frontswap backend wants to "resume" the writeback of a page to the
swap device, it can't call swap_writepage() as the page will simply
reenter the backend.

This patch separates swap_writepage() into a top and bottom half, the
bottom half named __swap_writepage() to allow a frontswap backend, like
zswap, to resume writeback beyond the frontswap_store() hook.

__add_to_swap_cache() is also made non-static so that the page for which
writeback is to be resumed can be added to the swap cache.

Signed-off-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/mmap: check for RLIMIT_AS before unmapping
Cyril Hrubis [Mon, 29 Apr 2013 22:08:33 +0000 (15:08 -0700)]
mm/mmap: check for RLIMIT_AS before unmapping

Fix a corner case for MAP_FIXED when requested mapping length is larger
than rlimit for virtual memory.  In such case any overlapping mappings
are unmapped before we check for the limit and return ENOMEM.

The check is moved before the loop that unmaps overlapping parts of
existing mappings.  When we are about to hit the limit (currently mapped
pages + len > limit) we scan for overlapping pages and check again
accounting for them.

This fixes situation when userspace program expects that the previous
mappings are preserved after the mmap() syscall has returned with error.
(POSIX clearly states that successfull mapping shall replace any
previous mappings.)

This corner case was found and can be tested with LTP testcase:

testcases/open_posix_testsuite/conformance/interfaces/mmap/24-2.c

In this case the mmap, which is clearly over current limit, unmaps
dynamic libraries and the testcase segfaults right after returning into
userspace.

I've also looked at the second instance of the unmapping loop in the
do_brk().  The do_brk() is called from brk() syscall and from vm_brk().
The brk() syscall checks for overlapping mappings and bails out when
there are any (so it can't be triggered from the brk syscall).  The
vm_brk() is called only from binmft handlers so it shouldn't be
triggered unless binmft handler created overlapping mappings.

Signed-off-by: Cyril Hrubis <chrubis@suse.cz>
Reviewed-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: add memory.pressure_level events
Anton Vorontsov [Mon, 29 Apr 2013 22:08:31 +0000 (15:08 -0700)]
memcg: add memory.pressure_level events

With this patch userland applications that want to maintain the
interactivity/memory allocation cost can use the pressure level
notifications.  The levels are defined like this:

The "low" level means that the system is reclaiming memory for new
allocations.  Monitoring this reclaiming activity might be useful for
maintaining cache level.  Upon notification, the program (typically
"Activity Manager") might analyze vmstat and act in advance (i.e.
prematurely shutdown unimportant services).

The "medium" level means that the system is experiencing medium memory
pressure, the system might be making swap, paging out active file
caches, etc.  Upon this event applications may decide to further analyze
vmstat/zoneinfo/memcg or internal memory usage statistics and free any
resources that can be easily reconstructed or re-read from a disk.

The "critical" level means that the system is actively thrashing, it is
about to out of memory (OOM) or even the in-kernel OOM killer is on its
way to trigger.  Applications should do whatever they can to help the
system.  It might be too late to consult with vmstat or any other
statistics, so it's advisable to take an immediate action.

The events are propagated upward until the event is handled, i.e.  the
events are not pass-through.  Here is what this means: for example you
have three cgroups: A->B->C.  Now you set up an event listener on
cgroups A, B and C, and suppose group C experiences some pressure.  In
this situation, only group C will receive the notification, i.e.  groups
A and B will not receive it.  This is done to avoid excessive
"broadcasting" of messages, which disturbs the system and which is
especially bad if we are low on memory or thrashing.  So, organize the
cgroups wisely, or propagate the events manually (or, ask us to
implement the pass-through events, explaining why would you need them.)

Performance wise, the memory pressure notifications feature itself is
lightweight and does not require much of bookkeeping, in contrast to the
rest of memcg features.  Unfortunately, as of current memcg
implementation, pages accounting is an inseparable part and cannot be
turned off.  The good news is that there are some efforts[1] to improve
the situation; plus, implementing the same, fully API-compatible[2]
interface for CONFIG_MEMCG=n case (e.g.  embedded) is also a viable
option, so it will not require any changes on the userland side.

[1] http://permalink.gmane.org/gmane.linux.kernel.cgroups/6291
[2] http://lkml.org/lkml/2013/2/21/454

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix CONFIG_CGROPUPS=n warnings]
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Acked-by: Kirill A. Shutemov <kirill@shutemov.name>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Glauber Costa <glommer@parallels.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Leonid Moiseichuk <leonid.moiseichuk@nokia.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Cc: John Stultz <john.stultz@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: madvise: complete input validation before taking lock
Rasmus Villemoes [Mon, 29 Apr 2013 22:08:23 +0000 (15:08 -0700)]
mm: madvise: complete input validation before taking lock

In madvise(), there doesn't seem to be any reason for taking the
&current->mm->mmap_sem before start and len_in have been validated.
Incidentally, this removes the need for the out: label.

[akpm@linux-foundation.org: s/out_plug/out/, per David]
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.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>
11 years agomm, hotplug: avoid compiling memory hotremove functions when disabled
David Rientjes [Mon, 29 Apr 2013 22:08:22 +0000 (15:08 -0700)]
mm, hotplug: avoid compiling memory hotremove functions when disabled

__remove_pages() is only necessary for CONFIG_MEMORY_HOTREMOVE.  PowerPC
pseries will return -EOPNOTSUPP if unsupported.

Adding an #ifdef causes several other functions it depends on to also
become unnecessary, which saves in .text when disabled (it's disabled in
most defconfigs besides powerpc, including x86).  remove_memory_block()
becomes static since it is not referenced outside of
drivers/base/memory.c.

Build tested on x86 and powerpc with CONFIG_MEMORY_HOTREMOVE both enabled
and disabled.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: change __remove_pages() to call release_mem_region_adjustable()
Toshi Kani [Mon, 29 Apr 2013 22:08:20 +0000 (15:08 -0700)]
mm: change __remove_pages() to call release_mem_region_adjustable()

Change __remove_pages() to call release_mem_region_adjustable().  This
allows a requested memory range to be released from the iomem_resource
table even if it does not match exactly to an resource entry but still
fits into.  The resource entries initialized at bootup usually cover the
whole contiguous memory ranges and may not necessarily match with the
size of memory hot-delete requests.

If release_mem_region_adjustable() failed, __remove_pages() emits a
warning message and continues to proceed as it was the case with
release_mem_region().  release_mem_region(), which is defined to
__release_region(), emits a warning message and returns no error since a
void function.

Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by : Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: T Makphaibulchoke <tmac@hp.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Jiang Liu <jiang.liu@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoresource: add release_mem_region_adjustable()
Toshi Kani [Mon, 29 Apr 2013 22:08:19 +0000 (15:08 -0700)]
resource: add release_mem_region_adjustable()

Add release_mem_region_adjustable(), which releases a requested region
from a currently busy memory resource.  This interface adjusts the
matched memory resource accordingly even if the requested region does
not match exactly but still fits into.

This new interface is intended for memory hot-delete.  During bootup,
memory resources are inserted from the boot descriptor table, such as
EFI Memory Table and e820.  Each memory resource entry usually covers
the whole contigous memory range.  Memory hot-delete request, on the
other hand, may target to a particular range of memory resource, and its
size can be much smaller than the whole contiguous memory.  Since the
existing release interfaces like __release_region() require a requested
region to be exactly matched to a resource entry, they do not allow a
partial resource to be released.

This new interface is restrictive (i.e.  release under certain
conditions), which is consistent with other release interfaces,
__release_region() and __release_resource().  Additional release
conditions, such as an overlapping region to a resource entry, can be
supported after they are confirmed as valid cases.

There is no change to the existing interfaces since their restriction is
valid for I/O resources.

[akpm@linux-foundation.org: use GFP_ATOMIC under write_lock()]
[akpm@linux-foundation.org: switch back to GFP_KERNEL, less buggily]
[akpm@linux-foundation.org: remove unneeded and wrong kfree(), per Toshi]
Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by : Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Cc: T Makphaibulchoke <tmac@hp.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Jiang Liu <jiang.liu@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoresource: add __adjust_resource() for internal use
Toshi Kani [Mon, 29 Apr 2013 22:08:17 +0000 (15:08 -0700)]
resource: add __adjust_resource() for internal use

Add __adjust_resource(), which is called by adjust_resource() internally
after the resource_lock is held.  There is no interface change to
adjust_resource().  This change allows other functions to call
__adjust_resource() internally while the resource_lock is held.

Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: T Makphaibulchoke <tmac@hp.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Jiang Liu <jiang.liu@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: rewrite the comment over migrate_pages() more comprehensibly
Srivatsa S. Bhat [Mon, 29 Apr 2013 22:08:16 +0000 (15:08 -0700)]
mm: rewrite the comment over migrate_pages() more comprehensibly

The comment over migrate_pages() looks quite weird, and makes it hard to
grasp what it is trying to say.  Rewrite it more comprehensibly.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoTHP: fix comment about memory barrier
Minchan Kim [Mon, 29 Apr 2013 22:08:15 +0000 (15:08 -0700)]
THP: fix comment about memory barrier

Currently the memory barrier in __do_huge_pmd_anonymous_page doesn't
work.  Because lru_cache_add_lru uses pagevec so it could miss spinlock
easily so above rule was broken so user might see inconsistent data.

I was not first person who pointed out the problem.  Mel and Peter
pointed out a few months ago and Peter pointed out further that even
spin_lock/unlock can't make sure of it:

  http://marc.info/?t=134333512700004

In particular:

         *A = a;
         LOCK
         UNLOCK
         *B = b;

may occur as:

         LOCK, STORE *B, STORE *A, UNLOCK

At last, Hugh pointed out that even we don't need memory barrier in
there because __SetPageUpdate already have done it from Nick's commit
0ed361dec369 ("mm: fix PageUptodate data race") explicitly.

So this patch fixes comment on THP and adds same comment for
do_anonymous_page, too because everybody except Hugh was missing that.
It means we need a comment about that.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: remove CONFIG_HOTPLUG ifdefs
Yijing Wang [Mon, 29 Apr 2013 22:08:14 +0000 (15:08 -0700)]
mm: remove CONFIG_HOTPLUG ifdefs

CONFIG_HOTPLUG is going away as an option, cleanup CONFIG_HOTPLUG
ifdefs in mm files.

Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/memcontrol.c: remove unnecessary ;
Michel Lespinasse [Mon, 29 Apr 2013 22:08:13 +0000 (15:08 -0700)]
mm/memcontrol.c: remove unnecessary ;

Just a trivial issue I stumbled on while doing something else...

Signed-off-by: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: reinititalise user and admin reserves if memory is added or removed
Andrew Shewmaker [Mon, 29 Apr 2013 22:08:12 +0000 (15:08 -0700)]
mm: reinititalise user and admin reserves if memory is added or removed

Alter the admin and user reserves of the previous patches in this series
when memory is added or removed.

If memory is added and the reserves have been eliminated or increased
above the default max, then we'll trust the admin.

If memory is removed and there isn't enough free memory, then we need to
reset the reserves.

Otherwise keep the reserve set by the admin.

The reserve reset code is the same as the reserve initialization code.

I tested hot addition and removal by triggering it via sysfs.  The
reserves shrunk when they were set high and memory was removed.  They
were reset higher when memory was added again.

[akpm@linux-foundation.org: use register_hotmemory_notifier()]
[akpm@linux-foundation.org: init_user_reserve() and init_admin_reserve can no longer be __meminit]
[fengguang.wu@intel.com: make init_reserve_notifier() static]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Andrew Shewmaker <agshew@gmail.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: replace hardcoded 3% with admin_reserve_pages knob
Andrew Shewmaker [Mon, 29 Apr 2013 22:08:11 +0000 (15:08 -0700)]
mm: replace hardcoded 3% with admin_reserve_pages knob

Add an admin_reserve_kbytes knob to allow admins to change the hardcoded
memory reserve to something other than 3%, which may be multiple
gigabytes on large memory systems.  Only about 8MB is necessary to
enable recovery in the default mode, and only a few hundred MB are
required even when overcommit is disabled.

This affects OVERCOMMIT_GUESS and OVERCOMMIT_NEVER.

admin_reserve_kbytes is initialized to min(3% free pages, 8MB)

I arrived at 8MB by summing the RSS of sshd or login, bash, and top.

Please see first patch in this series for full background, motivation,
testing, and full changelog.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: make init_admin_reserve() static]
Signed-off-by: Andrew Shewmaker <agshew@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: limit growth of 3% hardcoded other user reserve
Andrew Shewmaker [Mon, 29 Apr 2013 22:08:10 +0000 (15:08 -0700)]
mm: limit growth of 3% hardcoded other user reserve

Add user_reserve_kbytes knob.

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

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

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

This only affects OVERCOMMIT_NEVER mode.

Background

1. user reserve

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

bash: fork: Cannot allocate memory

2. admin reserve

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

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

Motivation

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

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

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

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

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

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

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

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

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

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

Risks

* "bash: fork: Cannot allocate memory"

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

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

* root-cant-log-in problem

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

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

Alternatives

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

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

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

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

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

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

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

FAQ

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

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

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

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

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

 * How do you calculate a minimum useful reserve?

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

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

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

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

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

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

 * What happens if user_reserve_pages is set to 0?

   Note, this only affects overcomitt 'never' mode.

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

   However, they will easily see a message such as:

   "bash: fork: Cannot allocate memory"

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

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

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

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

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

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

Test Summary

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

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

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

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

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

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

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

Test Description

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

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

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

In overcommit 'never' mode, memory_ratio=100

Test Results

3.9.0-rc1-mm1

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

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

3.9.0-rc1-mm1-tunablereserves

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

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

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

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

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

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

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

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

Future Work

 - Test larger memory systems.

 - Test an embedded image.

 - Test other architectures.

 - Time malloc microbenchmarks.

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

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

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: make init_user_reserve() static]
Signed-off-by: Andrew Shewmaker <agshew@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agokernel/cpuset.c: use register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:08 +0000 (15:08 -0700)]
kernel/cpuset.c: use register_hotmemory_notifier()

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

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

Cc: Li Zefan <lizefan@huawei.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/proc/kcore.c: use register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:08 +0000 (15:08 -0700)]
fs/proc/kcore.c: use register_hotmemory_notifier()

Saves an ifdef, no code size changes

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/base/node.c: switch to register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:07 +0000 (15:08 -0700)]
drivers/base/node.c: switch to register_hotmemory_notifier()

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

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

Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/slub.c: use register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:06 +0000 (15:08 -0700)]
mm/slub.c: use register_hotmemory_notifier()

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

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

Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc/util.c: use register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:05 +0000 (15:08 -0700)]
ipc/util.c: use register_hotmemory_notifier()

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

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

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinclude/linux/memory.h: implement register_hotmemory_notifier()
Andrew Morton [Mon, 29 Apr 2013 22:08:04 +0000 (15:08 -0700)]
include/linux/memory.h: implement register_hotmemory_notifier()

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

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

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

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

Cc: Andrew Shewmaker <agshew@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopowerpc/mm/numa: use setup_nr_node_ids() instead of opencoding.
Cody P Schafer [Mon, 29 Apr 2013 22:08:03 +0000 (15:08 -0700)]
powerpc/mm/numa: use setup_nr_node_ids() instead of opencoding.

[sfr@canb.auug.org.au: add missing semicolon]
Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: 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>
11 years agox86/mm/numa: use setup_nr_node_ids() instead of opencoding.
Cody P Schafer [Mon, 29 Apr 2013 22:08:02 +0000 (15:08 -0700)]
x86/mm/numa: use setup_nr_node_ids() instead of opencoding.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopage_alloc: make setup_nr_node_ids() usable for arch init code
Cody P Schafer [Mon, 29 Apr 2013 22:08:01 +0000 (15:08 -0700)]
page_alloc: make setup_nr_node_ids() usable for arch init code

powerpc and x86 were opencoding copies of setup_nr_node_ids(), which
page_alloc provides but makes static.  Make it avaliable to the archs in
linux/mm.h.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: speedup in __early_pfn_to_nid
Russ Anderson [Mon, 29 Apr 2013 22:07:59 +0000 (15:07 -0700)]
mm: speedup in __early_pfn_to_nid

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

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

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

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

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

[akpm@linux-foundation.org: make the statics __meminitdata]
[akpm@linux-foundation.org: fix comment formatting]
[akpm@linux-foundation.org: fix ia64, per yinghai]
[akpm@linux-foundation.org: add missing semicolon, per Tony]
Signed-off-by: Russ Anderson <rja@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Tested-by: "Luck, Tony" <tony.luck@intel.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Lin Feng <linfeng@cn.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/migrate: fix comment typo syncronous->synchronous
Jianguo Wu [Mon, 29 Apr 2013 22:07:58 +0000 (15:07 -0700)]
mm/migrate: fix comment typo syncronous->synchronous

Signed-off-by: Jianguo Wu <wujianguo@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: page_alloc: avoid marking zones full prematurely after zone_reclaim()
Mel Gorman [Mon, 29 Apr 2013 22:07:57 +0000 (15:07 -0700)]
mm: page_alloc: avoid marking zones full prematurely after zone_reclaim()

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

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

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

The expected behaviour is that the clean pages get reclaimed and the
application uses node 0 for its memory.  The observed behaviour was that
the memory for the memhog application was allocated off-node since
commits cd38b115d5ad ("mm: page allocator: initialise ZLC for first zone
eligible for zone_reclaim") and commit 76d3fbf8fbf6 ("mm: page
allocator: reconsider zones for allocation after direct reclaim").

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

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

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

[mhocko@suse.cz: fix alloc_flags test]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Hedi Berriche <hedi@sgi.com>
Tested-by: Hedi Berriche <hedi@sgi.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agox86-64: fall back to regular page vmemmap on allocation failure
Johannes Weiner [Mon, 29 Apr 2013 22:07:56 +0000 (15:07 -0700)]
x86-64: fall back to regular page vmemmap on allocation failure

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

Try to fall back to regular pages before failing the hotplug event
completely.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agox86-64: use vmemmap_populate_basepages() for !pse setups
Johannes Weiner [Mon, 29 Apr 2013 22:07:54 +0000 (15:07 -0700)]
x86-64: use vmemmap_populate_basepages() for !pse setups

We already have generic code to allocate vmemmap with regular pages, use
it.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agox86-64: remove dead debugging code for !pse setups
Johannes Weiner [Mon, 29 Apr 2013 22:07:52 +0000 (15:07 -0700)]
x86-64: remove dead debugging code for !pse setups

No need to maintain addr_end and p_end when they are never actually read
anywhere on !pse setups.  Remove the dead code.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agosparse-vmemmap: specify vmemmap population range in bytes
Johannes Weiner [Mon, 29 Apr 2013 22:07:50 +0000 (15:07 -0700)]
sparse-vmemmap: specify vmemmap population range in bytes

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

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

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

Just translate the section range into bytes once in the generic sparse
code, then pass byte ranges down the stack.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: David S. Miller <davem@davemloft.net>
Tested-by: David S. Miller <davem@davemloft.net>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: try harder to allocate vmemmap blocks
Ben Hutchings [Mon, 29 Apr 2013 22:07:49 +0000 (15:07 -0700)]
mm: try harder to allocate vmemmap blocks

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

Addresses http://bugs.debian.org/699913

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Tested-by: Bernhard Schmidt <Bernhard.Schmidt@lrz.de>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, hugetlb: include hugepages in meminfo
David Rientjes [Mon, 29 Apr 2013 22:07:48 +0000 (15:07 -0700)]
mm, hugetlb: include hugepages in meminfo

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

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

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

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: David Rientjes <rientjes@google.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>
11 years agomm: merging memory blocks resets mempolicy
Hampson, Steven T [Mon, 29 Apr 2013 22:07:47 +0000 (15:07 -0700)]
mm: merging memory blocks resets mempolicy

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

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

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

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

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

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

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

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

int main()
{
allocate();
allocate();
allocate();
return 0;
}

Signed-off-by: Steven T Hampson <steven.t.hampson@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoarm: set the page table freeing ceiling to TASK_SIZE
Catalin Marinas [Mon, 29 Apr 2013 22:07:45 +0000 (15:07 -0700)]
arm: set the page table freeing ceiling to TASK_SIZE

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

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

Note that the pgd_free() function already checks the presence of the
shared pmd page allocated by pgd_alloc() and frees it, though with
ceiling 0 this wasn't necessary.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [3.3+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: allow arch code to control the user page table ceiling
Hugh Dickins [Mon, 29 Apr 2013 22:07:44 +0000 (15:07 -0700)]
mm: allow arch code to control the user page table ceiling

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

[catalin.marinas@arm.com: commit log; shift_arg_pages(), asm-generic/pgtables.h changes]
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: <stable@vger.kernel.org> [3.3+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: do not check for do_swap_account in mem_cgroup_{read,write,reset}
Michal Hocko [Mon, 29 Apr 2013 22:07:43 +0000 (15:07 -0700)]
memcg: do not check for do_swap_account in mem_cgroup_{read,write,reset}

Since commit 2d11085e404f ("memcg: do not create memsw files if swap
accounting is disabled") memsw files are created only if memcg swap
accounting is enabled so it doesn't make any sense to check for it
explicitly in mem_cgroup_read(), mem_cgroup_write() and
mem_cgroup_reset().

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agommap: find_vma: remove the WARN_ON_ONCE(!mm) check
Zhang Yanfei [Mon, 29 Apr 2013 22:07:42 +0000 (15:07 -0700)]
mmap: find_vma: remove the WARN_ON_ONCE(!mm) check

Remove the WARN_ON_ONCE(!mm) check as the comment suggested.  Kernel
code calls find_vma only when it is absolutely sure that the mm_struct
arg to it is non-NULL.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: k80c <k80ck80c@gmail.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agokexec, vmalloc: export additional vmalloc layer information
Atsushi Kumagai [Mon, 29 Apr 2013 22:07:40 +0000 (15:07 -0700)]
kexec, vmalloc: export additional vmalloc layer information

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

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

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

So this patch exports them externally with small cleanup.

[akpm@linux-foundation.org: vmalloc.h should include list.h for list_head]
Signed-off-by: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: remove list management of vmlist after initializing vmalloc
Joonsoo Kim [Mon, 29 Apr 2013 22:07:39 +0000 (15:07 -0700)]
mm, vmalloc: remove list management of vmlist after initializing vmalloc

Now, there is no need to maintain vmlist after initializing vmalloc.  So
remove related code and data structure.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: export vmap_area_list, instead of vmlist
Joonsoo Kim [Mon, 29 Apr 2013 22:07:37 +0000 (15:07 -0700)]
mm, vmalloc: export vmap_area_list, instead of vmlist

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

We are about to remove vmlist, then another way to retrieve information
of vmalloc layer is needed for makedumpfile.  For this purpose, we
export vmap_area_list, instead of vmlist.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: iterate vmap_area_list, instead of vmlist, in vmallocinfo()
Joonsoo Kim [Mon, 29 Apr 2013 22:07:35 +0000 (15:07 -0700)]
mm, vmalloc: iterate vmap_area_list, instead of vmlist, in vmallocinfo()

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

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

Therefore, this patch not only change a iteration list, but also add a
appropriate smp_[rw]mb to right places.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: iterate vmap_area_list in get_vmalloc_info()
Joonsoo Kim [Mon, 29 Apr 2013 22:07:34 +0000 (15:07 -0700)]
mm, vmalloc: iterate vmap_area_list in get_vmalloc_info()

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

vmlist is lack of information about some areas in vmalloc address space.
For example, vm_map_ram() allocate area in vmalloc address space, but it
doesn't make a link with vmlist.  To provide full information about
vmalloc address space is better idea, so we don't use va->vm and use
vmap_area directly.  This makes get_vmalloc_info() more precise.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: iterate vmap_area_list, instead of vmlist in vread/vwrite()
Joonsoo Kim [Mon, 29 Apr 2013 22:07:32 +0000 (15:07 -0700)]
mm, vmalloc: iterate vmap_area_list, instead of vmlist in vread/vwrite()

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

There is a little difference relate to lock, because vmlist_lock is mutex,
but, vmap_area_lock is spin_lock.  It may introduce a spinning overhead
during vread/vwrite() is executing.  But, these are debug-oriented
functions, so this overhead is not real problem for common case.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: protect va->vm by vmap_area_lock
Joonsoo Kim [Mon, 29 Apr 2013 22:07:30 +0000 (15:07 -0700)]
mm, vmalloc: protect va->vm by vmap_area_lock

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

For removing vmlist, iterating vmlist codes should be changed to
iterating a vmap_area_list.  Before implementing that, we should make
sure that when we iterate a vmap_area_list, accessing to va->vm doesn't
cause a race condition.  This patch ensure that when iterating a
vmap_area_list, there is no race condition for accessing to vm_struct.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: move get_vmalloc_info() to vmalloc.c
Joonsoo Kim [Mon, 29 Apr 2013 22:07:28 +0000 (15:07 -0700)]
mm, vmalloc: move get_vmalloc_info() to vmalloc.c

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

It is preferable that vmlist_lock and vmlist is only used in vmalloc.c
for maintainability. So move the code to vmalloc.c

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm, vmalloc: change iterating a vmlist to find_vm_area()
Joonsoo Kim [Mon, 29 Apr 2013 22:07:27 +0000 (15:07 -0700)]
mm, vmalloc: change iterating a vmlist to find_vm_area()

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

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

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

This patch:

The purpose of iterating a vmlist is finding vm area with specific virtual
address.  find_vm_area() is provided for this purpose and more efficient,
because it uses a rbtree.  So change it.

Signed-off-by: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Guan Xuetao <gxt@mprc.pku.edu.cn>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Chris Metcalf <cmetcalf@tilera.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Atsushi Kumagai <kumagai-atsushi@mxc.nes.nec.co.jp>
Cc: Dave Anderson <anderson@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: make snapshotting pages for stable writes a per-bio operation
Darrick J. Wong [Mon, 29 Apr 2013 22:07:25 +0000 (15:07 -0700)]
mm: make snapshotting pages for stable writes a per-bio operation

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

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

[akpm@linux-foundation.org: rename _submit_bh()'s `flags' to `bio_flags', delobotomize the _submit_bh declaration]
[akpm@linux-foundation.org: teeny cleanup]
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/hugetlb: add more arch-defined huge_pte functions
Gerald Schaefer [Mon, 29 Apr 2013 22:07:23 +0000 (15:07 -0700)]
mm/hugetlb: add more arch-defined huge_pte functions

Commit abf09bed3cce ("s390/mm: implement software dirty bits")
introduced another difference in the pte layout vs.  the pmd layout on
s390, thoroughly breaking the s390 support for hugetlbfs.  This requires
replacing some more pte_xxx functions in mm/hugetlbfs.c with a
huge_pte_xxx version.

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

[akpm@linux-foundation.org: fix warning]
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Hillf Danton <dhillf@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz> [for !s390 parts]
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs: don't compile in drop_caches.c when CONFIG_SYSCTL=n
Josh Triplett [Mon, 29 Apr 2013 22:07:22 +0000 (15:07 -0700)]
fs: don't compile in drop_caches.c when CONFIG_SYSCTL=n

drop_caches.c provides code only invokable via sysctl, so don't compile it
in when CONFIG_SYSCTL=n.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agocgroup: remove css_get_next
Michal Hocko [Mon, 29 Apr 2013 22:07:21 +0000 (15:07 -0700)]
cgroup: remove css_get_next

Now that we have generic and well ordered cgroup tree walkers there is
no need to keep css_get_next in the place.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: further simplify mem_cgroup_iter
Michal Hocko [Mon, 29 Apr 2013 22:07:19 +0000 (15:07 -0700)]
memcg: further simplify mem_cgroup_iter

mem_cgroup_iter basically does two things currently.  It takes care of
the house keeping (reference counting, raclaim cookie) and it iterates
through a hierarchy tree (by using cgroup generic tree walk).  The code
would be much more easier to follow if we move the iteration outside of
the function (to __mem_cgrou_iter_next) so the distinction is more
clear.  This patch doesn't introduce any functional changes.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: simplify mem_cgroup_iter
Michal Hocko [Mon, 29 Apr 2013 22:07:18 +0000 (15:07 -0700)]
memcg: simplify mem_cgroup_iter

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

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

We could get rid of the surrounding while loop but keep it in for now to
make review easier.  It will go away in the following patch.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: relax memcg iter caching
Michal Hocko [Mon, 29 Apr 2013 22:07:17 +0000 (15:07 -0700)]
memcg: relax memcg iter caching

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

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

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

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

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

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

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

Spotted by Ying Han.

Original idea from Johannes Weiner.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ying Han <yinghan@google.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: rework mem_cgroup_iter to use cgroup iterators
Michal Hocko [Mon, 29 Apr 2013 22:07:15 +0000 (15:07 -0700)]
memcg: rework mem_cgroup_iter to use cgroup iterators

mem_cgroup_iter curently relies on css->id when walking down a group
hierarchy tree.  This is really awkward because the tree walk depends on
the groups creation ordering.  The only guarantee is that a parent node is
visited before its children.

Example:

 1) mkdir -p a a/d a/b/c
 2) mkdir -a a/b/c a/d

Will create the same trees but the tree walks will be different:

 1) a, d, b, c
 2) a, b, c, d

Commit 574bd9f7c7c1 ("cgroup: implement generic child / descendant walk
macros") has introduced generic cgroup tree walkers which provide either
pre-order or post-order tree walk.  This patch converts css->id based
iteration to pre-order tree walk to keep the semantic with the original
iterator where parent is always visited before its subtree.

cgroup_for_each_descendant_pre suggests using post_create and
pre_destroy for proper synchronization with groups addidition resp.
removal.  This implementation doesn't use those because a new memory
cgroup is initialized sufficiently for iteration in mem_cgroup_css_alloc
already and css reference counting enforces that the group is alive for
both the last seen cgroup and the found one resp.  it signals that the
group is dead and it should be skipped.

If the reclaim cookie is used we need to store the last visited group
into the iterator so we have to be careful that it doesn't disappear in
the mean time.  Elevated reference count on the css keeps it alive even
though the group have been removed (parked waiting for the last dput so
that it can be freed).

Per node-zone-prio iter_lock has been introduced to ensure that
css_tryget and iter->last_visited is set atomically.  Otherwise two
racing walkers could both take a references and only one release it
leading to a css leak (which pins cgroup dentry).

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: keep prev's css alive for the whole mem_cgroup_iter
Michal Hocko [Mon, 29 Apr 2013 22:07:14 +0000 (15:07 -0700)]
memcg: keep prev's css alive for the whole mem_cgroup_iter

The patchset tries to make mem_cgroup_iter saner in the way how it walks
hierarchies.  css->id based traversal is far from being ideal as it is not
deterministic because it depends on the creation ordering.  Additional to
that css_id is considered a burden for cgroup maintainers because it is
quite some code and memcg is the last user of it.  After this series only
the swap accounting uses css_id but that one will follow up later.

Diffstat (if we exclude removed/added comments) looks quite
promising. We got rid of some code:

  $ git diff mmotm... | grep -v "^[+-][[:space:]]*[/ ]\*" | diffstat
   b/include/linux/cgroup.h |    3 ---
   kernel/cgroup.c          |   33 ---------------------------------
   mm/memcontrol.c          |    4 +++-
   3 files changed, 3 insertions(+), 37 deletions(-)

The first patch is just preparatory and it changes when we release css of
the previously returned memcg.  Nothing controlversial.

The second patch is the core of the patchset and it replaces css_get_next
based on css_id by the generic cgroup pre-order.  This brings some
chalanges for the last visited group caching during the reclaim
(mem_cgroup_per_zone::reclaim_iter).  We have to use memcg pointers
directly now which means that we have to keep a reference to those groups'
css to keep them alive.

I also folded iter_lock introduced by https://lkml.org/lkml/2013/1/3/295
in the previous version into this patch.  Johannes felt the race I was
describing should be mostly harmless and I haven't been able to trigger it
so the lock doesn't deserve its own patch.  It is still needed
temporarily, though, because the reference counting on iter->last_visited
depends on it.  It will go away with the next patch.

The next patch fixups an unbounded cgroup removal holdoff caused by the
elevated css refcount.  The issue has been observed by Ying Han.  Johannes
wasn't impressed by the previous version of the fix
(https://lkml.org/lkml/2013/2/8/379) which cleaned up pending references
during mem_cgroup_css_offline when a group is removed.  He has suggested a
different way when the iterator checks whether a cached memcg is still
valid or no.  More on that in the patch but the basic idea is that every
memcg tracks the number removed subgroups and iterator records this number
when a group is cached.  These numbers are checked before
iter->last_visited is about to be used and the iteration is restarted if
it is invalid.

The fourth and fifth patches are an attempt for simplification of the
mem_cgroup_iter.  css juggling is removed and the iteration logic is moved
to a helper so that the reference counting and iteration are separated.

The last patch just removes css_get_next as there is no user for it any
longer.

My testing looked as follows:
        A (use_hierarchy=1, limit_in_bytes=150M)
       /|\
      1 2 3

Children groups were created so that the number is never higher than 3 and
their limits were random between 50-100M.  Each group hosts a kernel build
(starting with tar -xf so the tree is not shared and make -jNUM_CPUs/3)
and terminated after random time - up to 5 minutes) and then it is
removed.

This should exercise both leaf and hierarchical reclaim as well as races
with cgroup removals and debugging messages I added on top proved that.
100 groups were created during the test.

This patch:

css reference counting keeps the cgroup alive even though it has been
already removed.  mem_cgroup_iter relies on this fact and takes a
reference to the returned group.  The reference is then released on the
next iteration or mem_cgroup_iter_break.  mem_cgroup_iter currently
releases the reference right after it gets the last css_id.

This is correct because neither prev's memcg nor cgroup are accessed after
then.  This will change in the next patch so we need to hold the group
alive a bit longer so let's move the css_put at the end of the function.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Ying Han <yinghan@google.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/x86: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:12 +0000 (15:07 -0700)]
mm/x86: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Cong Wang <amwang@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Attilio Rao <attilio.rao@citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/um: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:11 +0000 (15:07 -0700)]
mm/um: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Richard Weinberger <richard@nod.at>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/SPARC: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:10 +0000 (15:07 -0700)]
mm/SPARC: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: "David S. Miller" <davem@davemloft.net>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/PPC: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:08 +0000 (15:07 -0700)]
mm/PPC: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Alexander Graf <agraf@suse.de>
Cc: "Suzuki K. Poulose" <suzuki@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/MIPS: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:07 +0000 (15:07 -0700)]
mm/MIPS: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Daney <david.daney@cavium.com>
Cc: Cong Wang <amwang@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/microblaze: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:06 +0000 (15:07 -0700)]
mm/microblaze: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Michal Simek <monstr@monstr.eu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/metag: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:05 +0000 (15:07 -0700)]
mm/metag: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/FRV: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:04 +0000 (15:07 -0700)]
mm/FRV: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Also fix a bug that totalhigh_pages should be increased when freeing
a highmem page into the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/ARM: use free_highmem_page() to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:03 +0000 (15:07 -0700)]
mm/ARM: use free_highmem_page() to free highmem pages into buddy system

Use helper function free_highmem_page() to free highmem pages into
the buddy system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: introduce free_highmem_page() helper to free highmem pages into buddy system
Jiang Liu [Mon, 29 Apr 2013 22:07:00 +0000 (15:07 -0700)]
mm: introduce free_highmem_page() helper to free highmem pages into buddy system

The original goal of this patchset is to fix the bug reported by

  https://bugzilla.kernel.org/show_bug.cgi?id=53501

Now it has also been expanded to reduce common code used by memory
initializion.

This is the second part, which applies to the previous part at:
  http://marc.info/?l=linux-mm&m=136289696323825&w=2

It introduces a helper function free_highmem_page() to free highmem
pages into the buddy system when initializing mm subsystem.
Introduction of free_highmem_page() is one step forward to clean up
accesses and modificaitons of totalhigh_pages, totalram_pages and
zone->managed_pages etc. I hope we could remove all references to
totalhigh_pages from the arch/ subdirectory.

We have only tested these patchset on x86 platforms, and have done basic
compliation tests using cross-compilers from ftp.kernel.org. That means
some code may not pass compilation on some architectures. So any help
to test this patchset are welcomed!

There are several other parts still under development:
Part3: refine code to manage totalram_pages, totalhigh_pages and
zone->managed_pages
Part4: introduce helper functions to simplify mem_init() and remove the
global variable num_physpages.

This patch:

Introduce helper function free_highmem_page(), which will be used by
architectures with HIGHMEM enabled to free highmem pages into the buddy
system.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Suzuki K. Poulose" <suzuki@in.ibm.com>
Cc: Alexander Graf <agraf@suse.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Attilio Rao <attilio.rao@citrix.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Cong Wang <amwang@redhat.com>
Cc: David Daney <david.daney@cavium.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Jiang Liu <liuj97@gmail.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Michal Simek <monstr@monstr.eu>
Cc: Michel Lespinasse <walken@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rik van Riel <riel@redhat.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm,kexec: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:58 +0000 (15:06 -0700)]
mm,kexec: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/metag: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:57 +0000 (15:06 -0700)]
mm/metag: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/arc: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:56 +0000 (15:06 -0700)]
mm/arc: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/xtensa: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:55 +0000 (15:06 -0700)]
mm/xtensa: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/x86: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:53 +0000 (15:06 -0700)]
mm/x86: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
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>
11 years agomm/unicore32: use common help functions to free reserved pages
Jiang Liu [Mon, 29 Apr 2013 22:06:52 +0000 (15:06 -0700)]
mm/unicore32: use common help functions to free reserved pages

Use common help functions to free reserved pages.

Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>