An inactive file list is considered low when its active counterpart is
bigger, regardless of whether it is a global zone LRU list or a memcg
zone LRU list.  The only difference is in how the LRU size is assessed.

get_lru_size() does the right thing for both global and memcg reclaim
situations.

Get rid of inactive_file_is_low_global() and
mem_cgroup_inactive_file_is_low() by using get_lru_size() and compare
the numbers in common code.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Acked-by: NMichal Hocko <mhocko@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

swap_lock is heavily contended when I test swap to 3 fast SSD (even
slightly slower than swap to 2 such SSD).  The main contention comes
from swap_info_get().  This patch tries to fix the gap with adding a new
per-partition lock.

Global data like nr_swapfiles, total_swap_pages, least_priority and
swap_list are still protected by swap_lock.

nr_swap_pages is an atomic now, it can be changed without swap_lock.  In
theory, it's possible get_swap_page() finds no swap pages but actually
there are free swap pages.  But sounds not a big problem.

Accessing partition specific data (like scan_swap_map and so on) is only
protected by swap_info_struct.lock.

Changing swap_info_struct.flags need hold swap_lock and
swap_info_struct.lock, because scan_scan_map() will check it.  read the
flags is ok with either the locks hold.

If both swap_lock and swap_info_struct.lock must be hold, we always hold
the former first to avoid deadlock.

swap_entry_free() can change swap_list.  To delete that code, we add a
new highest_priority_index.  Whenever get_swap_page() is called, we
check it.  If it's valid, we use it.

It's a pity get_swap_page() still holds swap_lock().  But in practice,
swap_lock() isn't heavily contended in my test with this patch (or I can
say there are other much more heavier bottlenecks like TLB flush).  And
BTW, looks get_swap_page() doesn't really need the lock.  We never free
swap_info[] and we check SWAP_WRITEOK flag.  The only risk without the
lock is we could swapout to some low priority swap, but we can quickly
recover after several rounds of swap, so sounds not a big deal to me.
But I'd prefer to fix this if it's a real problem.

"swap: make each swap partition have one address_space" improved the
swapout speed from 1.7G/s to 2G/s.  This patch further improves the
speed to 2.3G/s, so around 15% improvement.  It's a multi-process test,
so TLB flush isn't the biggest bottleneck before the patches.

[arnd@arndb.de: fix it for nommu]
[hughd@google.com: add missing unlock]
[minchan@kernel.org: get rid of lockdep whinge on sys_swapon]
Signed-off-by: NShaohua Li <shli@fusionio.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Cc: Dan Magenheimer <dan.magenheimer@oracle.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NHugh Dickins <hughd@google.com>
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch introduces PF_MEMALLOC_NOIO on process flag('flags' field of
'struct task_struct'), so that the flag can be set by one task to avoid
doing I/O inside memory allocation in the task's context.

The patch trys to solve one deadlock problem caused by block device, and
the problem may happen at least in the below situations:

- during block device runtime resume, if memory allocation with
  GFP_KERNEL is called inside runtime resume callback of any one of its
  ancestors(or the block device itself), the deadlock may be triggered
  inside the memory allocation since it might not complete until the block
  device becomes active and the involed page I/O finishes.  The situation
  is pointed out first by Alan Stern.  It is not a good approach to
  convert all GFP_KERNEL[1] in the path into GFP_NOIO because several
  subsystems may be involved(for example, PCI, USB and SCSI may be
  involved for usb mass stoarage device, network devices involved too in
  the iSCSI case)

- during block device runtime suspend, because runtime resume need to
  wait for completion of concurrent runtime suspend.

- during error handling of usb mass storage deivce, USB bus reset will
  be put on the device, so there shouldn't have any memory allocation with
  GFP_KERNEL during USB bus reset, otherwise the deadlock similar with
  above may be triggered.  Unfortunately, any usb device may include one
  mass storage interface in theory, so it requires all usb interface
  drivers to handle the situation.  In fact, most usb drivers don't know
  how to handle bus reset on the device and don't provide .pre_set() and
  .post_reset() callback at all, so USB core has to unbind and bind driver
  for these devices.  So it is still not practical to resort to GFP_NOIO
  for solving the problem.

Also the introduced solution can be used by block subsystem or block
drivers too, for example, set the PF_MEMALLOC_NOIO flag before doing
actual I/O transfer.

It is not a good idea to convert all these GFP_KERNEL in the affected
path into GFP_NOIO because these functions doing that may be implemented
as library and will be called in many other contexts.

In fact, memalloc_noio_flags() can convert some of current static
GFP_NOIO allocation into GFP_KERNEL back in other non-affected contexts,
at least almost all GFP_NOIO in USB subsystem can be converted into
GFP_KERNEL after applying the approach and make allocation with GFP_NOIO
only happen in runtime resume/bus reset/block I/O transfer contexts
generally.

[1], several GFP_KERNEL allocation examples in runtime resume path

- pci subsystem
acpi_os_allocate
	<-acpi_ut_allocate
		<-ACPI_ALLOCATE_ZEROED
			<-acpi_evaluate_object
				<-__acpi_bus_set_power
					<-acpi_bus_set_power
						<-acpi_pci_set_power_state
							<-platform_pci_set_power_state
								<-pci_platform_power_transition
									<-__pci_complete_power_transition
										<-pci_set_power_state
											<-pci_restore_standard_config
												<-pci_pm_runtime_resume
- usb subsystem
usb_get_status
	<-finish_port_resume
		<-usb_port_resume
			<-generic_resume
				<-usb_resume_device
					<-usb_resume_both
						<-usb_runtime_resume

- some individual usb drivers
usblp, uvc, gspca, most of dvb-usb-v2 media drivers, cpia2, az6007, ....

That is just what I have found.  Unfortunately, this allocation can only
be found by human being now, and there should be many not found since
any function in the resume path(call tree) may allocate memory with
GFP_KERNEL.
Signed-off-by: NMing Lei <ming.lei@canonical.com>
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Oliver Neukum <oneukum@suse.de>
Cc: Jiri Kosina <jiri.kosina@suse.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Decotigny <david.decotigny@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

From: Zlatko Calusic <zlatko.calusic@iskon.hr>

Commit 92df3a72 ("mm: vmscan: throttle reclaim if encountering too
many dirty pages under writeback") introduced waiting on congested zones
based on a sane algorithm in shrink_inactive_list().

What this means is that there's no more need for throttling and
additional heuristics in balance_pgdat().  So, let's remove it and tidy
up the code.
Signed-off-by: NZlatko Calusic <zlatko.calusic@iskon.hr>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now we have zone->managed_pages for "pages managed by the buddy system
in the zone", so replace zone->present_pages with zone->managed_pages if
what the user really wants is number of allocatable pages.
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Maciej Rutecki <maciej.rutecki@gmail.com>
Cc: Chris Clayton <chris2553@googlemail.com>
Cc: "Rafael J . Wysocki" <rjw@sisk.pl>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Minchan Kim <minchan@kernel.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Jianguo Wu <wujianguo@huawei.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now that balance_pgdat() is slightly tidied up, thanks to more capable
pgdat_balanced(), it's become obvious that pgdat_balanced() is called to
check the status, then break the loop if pgdat is balanced, just to be
immediately called again.  The second call is completely unnecessary, of
course.

The patch introduces pgdat_is_balanced boolean, which helps resolve the
above suboptimal behavior, with the added benefit of slightly better
documenting one other place in the function where we jump and skip lots
of code.
Signed-off-by: NZlatko Calusic <zlatko.calusic@iskon.hr>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Targeted (hard resp soft) reclaim has traditionally tried to scan one
group with decreasing priority until nr_to_reclaim (SWAP_CLUSTER_MAX
pages) is reclaimed or all priorities are exhausted.  The reclaim is
then retried until the limit is met.

This approach, however, doesn't work well with deeper hierarchies where
groups higher in the hierarchy do not have any or only very few pages
(this usually happens if those groups do not have any tasks and they
have only re-parented pages after some of their children is removed).
Those groups are reclaimed with decreasing priority pointlessly as there
is nothing to reclaim from them.

An easiest fix is to break out of the memcg iteration loop in
shrink_zone only if the whole hierarchy has been visited or sufficient
pages have been reclaimed.  This is also more natural because the
reclaimer expects that the hierarchy under the given root is reclaimed.
As a result we can simplify the soft limit reclaim which does its own
iteration.

[yinghan@google.com: break out of the hierarchy loop only if nr_reclaimed exceeded nr_to_reclaim]
[akpm@linux-foundation.org: use conventional comparison order]
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Reported-by: NYing Han <yinghan@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Glauber Costa <glommer@parallels.com>
Cc: Li Zefan <lizefan@huawei.com>
Signed-off-by: NYing Han <yinghan@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

"mm: vmscan: save work scanning (almost) empty LRU lists" made
SWAP_CLUSTER_MAX an unsigned long.

Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The restart logic for when reclaim operates back to back with compaction
is currently applied on the lruvec level.  But this does not make sense,
because the container of interest for compaction is a zone as a whole,
not the zone pages that are part of a certain memory cgroup.

Negative impact is bounded.  For one, the code checks that the lruvec
has enough reclaim candidates, so it does not risk getting stuck on a
condition that can not be fulfilled.  And the unfairness of hammering on
one particular memory cgroup to make progress in a zone will be
amortized by the round robin manner in which reclaim goes through the
memory cgroups.  Still, this can lead to unnecessary allocation
latencies when the code elects to restart on a hard to reclaim or small
group when there are other, more reclaimable groups in the zone.

Move this logic to the zone level and restart reclaim for all memory
cgroups in a zone when compaction requires more free pages from it.

[akpm@linux-foundation.org: no need for min_t]
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Reclaim pressure balance between anon and file pages is calculated
through a tuple of numerators and a shared denominator.

Exceptional cases that want to force-scan anon or file pages configure
the numerators and denominator such that one list is preferred, which is
not necessarily the most obvious way:

    fraction[0] = 1;
    fraction[1] = 0;
    denominator = 1;
    goto out;

Make this easier by making the force-scan cases explicit and use the
fractionals only in case they are calculated from reclaim history.

[akpm@linux-foundation.org: avoid using unintialized_var()]
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix comment style and elaborate on why anonymous memory is force-scanned
when file cache runs low.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

A swappiness of 0 has a slightly different meaning for global reclaim
(may swap if file cache really low) and memory cgroup reclaim (never
swap, ever).

In addition, global reclaim at highest priority will scan all LRU lists
equal to their size and ignore other balancing heuristics.  UNLESS
swappiness forbids swapping, then the lists are balanced based on recent
reclaim effectiveness.  UNLESS file cache is running low, then anonymous
pages are force-scanned.

This (total mess of a) behaviour is implicit and not obvious from the
way the code is organized.  At least make it apparent in the code flow
and document the conditions.  It will be it easier to come up with sane
semantics later.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Reviewed-by: NSatoru Moriya <satoru.moriya@hds.com>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NMel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

In certain cases (kswapd reclaim, memcg target reclaim), a fixed minimum
amount of pages is scanned from the LRU lists on each iteration, to make
progress.

Do not make this minimum bigger than the respective LRU list size,
however, and save some busy work trying to isolate and reclaim pages
that are not there.

Empty LRU lists are quite common with memory cgroups in NUMA
environments because there exists a set of LRU lists for each zone for
each memory cgroup, while the memory of a single cgroup is expected to
stay on just one node.  The number of expected empty LRU lists is thus

  memcgs * (nodes - 1) * lru types

Each attempt to reclaim from an empty LRU list does expensive size
comparisons between lists, acquires the zone's lru lock etc.  Avoid
that.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Reviewed-by: NMichal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit e9868505 ("mm, vmscan: only evict file pages when we have
plenty") makes a point of not going for anonymous memory while there is
still enough inactive cache around.

The check was added only for global reclaim, but it is just as useful to
reduce swapping in memory cgroup reclaim:

    200M-memcg-defconfig-j2

                                     vanilla                   patched
    Real time              454.06 (  +0.00%)         453.71 (  -0.08%)
    User time              668.57 (  +0.00%)         668.73 (  +0.02%)
    System time            128.92 (  +0.00%)         129.53 (  +0.46%)
    Swap in               1246.80 (  +0.00%)         814.40 ( -34.65%)
    Swap out              1198.90 (  +0.00%)         827.00 ( -30.99%)
    Pages allocated   16431288.10 (  +0.00%)    16434035.30 (  +0.02%)
    Major faults           681.50 (  +0.00%)         593.70 ( -12.86%)
    THP faults             237.20 (  +0.00%)         242.40 (  +2.18%)
    THP collapse           241.20 (  +0.00%)         248.50 (  +3.01%)
    THP splits             157.30 (  +0.00%)         161.40 (  +2.59%)
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Acked-by: NMichal Hocko <mhocko@suse.cz>
Acked-by: NRik van Riel <riel@redhat.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Satoru Moriya <satoru.moriya@hds.com>
Cc: Simon Jeons <simon.jeons@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

CONFIG_HOTPLUG is going away as an option.  As a result, the __dev*
markings need to be removed.

This change removes the use of __devinit from the file.

Based on patches originally written by Bill Pemberton, but redone by me
in order to handle some of the coding style issues better, by hand.

Cc: Bill Pemberton <wfp5p@virginia.edu>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

An unintended consequence of commit 4ae0a48b ("mm: modify
pgdat_balanced() so that it also handles order-0") is that
wait_iff_congested() can now be called with NULL 'struct zone *'
producing kernel oops like this:

  BUG: unable to handle kernel NULL pointer dereference
  IP: [<ffffffff811542d9>] wait_iff_congested+0x59/0x140

This trivial patch fixes it.
Reported-by: NZhouping Liu <zliu@redhat.com>
Reported-and-tested-by: NSedat Dilek <sedat.dilek@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: NZlatko Calusic <zlatko.calusic@iskon.hr>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Teach pgdat_balanced() about order-0 allocations so that we can simplify
code in a few places in vmstat.c.
Suggested-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NZlatko Calusic <zlatko.calusic@iskon.hr>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

On a 4GB RAM machine, where Normal zone is much smaller than DMA32 zone,
the Normal zone gets fragmented in time.  This requires relatively more
pressure in balance_pgdat to get the zone above the required watermark.
Unfortunately, the congestion_wait() call in there slows it down for a
completely wrong reason, expecting that there's a lot of
writeback/swapout, even when there's none (much more common).  After a
few days, when fragmentation progresses, this flawed logic translates to
a very high CPU iowait times, even though there's no I/O congestion at
all.  If THP is enabled, the problem occurs sooner, but I was able to
see it even on !THP kernels, just by giving it a bit more time to occur.

The proper way to deal with this is to not wait, unless there's
congestion.  Thanks to Mel Gorman, we already have the function that
perfectly fits the job.  The patch was tested on a machine which nicely
revealed the problem after only 1 day of uptime, and it's been working
great.
Signed-off-by: NZlatko Calusic <zlatko.calusic@iskon.hr>
Acked-by: NMel Gorman <mgorman@suse.de>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Neil found that if too_many_isolated() returns true while performing
direct reclaim we can end up waiting for other threads to complete their
direct reclaim.  If those threads are allowed to enter the FS or IO to
free memory, but this thread is not, then it is possible that those
threads will be waiting on this thread and so we get a circular deadlock.

some task enters direct reclaim with GFP_KERNEL
  => too_many_isolated() false
    => vmscan and run into dirty pages
      => pageout()
        => take some FS lock
          => fs/block code does GFP_NOIO allocation
            => enter direct reclaim again
              => too_many_isolated() true
                => waiting for others to progress, however the other
                   tasks may be circular waiting for the FS lock..

The fix is to let !__GFP_IO and !__GFP_FS direct reclaims enjoy higher
priority than normal ones, by lowering the throttle threshold for the
latter.

Allowing ~1/8 isolated pages in normal is large enough.  For example, for
a 1GB LRU list, that's ~128MB isolated pages, or 1k blocked tasks (each
isolates 32 4KB pages), or 64 blocked tasks per logical CPU (assuming 16
logical CPUs per NUMA node).  So it's not likely some CPU goes idle
waiting (when it could make progress) because of this limit: there are
much more sleeping reclaim tasks than the number of CPU, so the task may
well be blocked by some low level queue/lock anyway.

Now !GFP_IOFS reclaims won't be waiting for GFP_IOFS reclaims to progress.
 They will be blocked only when there are too many concurrent !GFP_IOFS
reclaims, however that's very unlikely because the IO-less direct reclaims
is able to progress much more faster, and they won't deadlock each other.
The threshold is raised high enough for them, so that there can be
sufficient parallel progress of !GFP_IOFS reclaims.

[akpm@linux-foundation.org: tweak comment]
Signed-off-by: NWu Fengguang <fengguang.wu@intel.com>
Cc: Torsten Kaiser <just.for.lkml@googlemail.com>
Tested-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NMinchan Kim <minchan.kim@gmail.com>
Acked-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: NRik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Comment "Why it's doing so" rather than "What it does" as proposed by
Andrew Morton.
Signed-off-by: NWu Fengguang <fengguang.wu@intel.com>
Reviewed-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: NMinchan Kim <minchan.kim@gmail.com>
Reviewed-by: NRik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

N_HIGH_MEMORY stands for the nodes that has normal or high memory.
N_MEMORY stands for the nodes that has any memory.

The code here need to handle with the nodes which have memory, we should
use N_MEMORY instead.
Signed-off-by: NLai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: NHillf Danton <dhillf@gmail.com>
Signed-off-by: NWen Congyang <wency@cn.fujitsu.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Hillf Danton <dhillf@gmail.com>
Cc: Lin Feng <linfeng@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

kswapd()->try_to_freeze() is defined to return a boolean, so it's better
to use a bool to hold its return value.
Signed-off-by: NJie Liu <jeff.liu@oracle.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If we have more inactive file pages than active file pages, we skip
scanning the active file pages altogether, with the idea that we do not
want to evict the working set when there is plenty of streaming IO in the
cache.

However, the code forgot to also skip scanning anonymous pages in that
situation.  That leads to the curious situation of keeping the active file
pages protected from being paged out when there are lots of inactive file
pages, while still scanning and evicting anonymous pages.

This patch fixes that situation, by only evicting file pages when we have
plenty of them and most are inactive.

[akpm@linux-foundation.org: adjust comment layout]
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We don't need custom COMPACTION_BUILD anymore, since we have handy
IS_ENABLED().
Signed-off-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: NMinchan Kim <minchan@kernel.org>
Acked-by: NDavid Rientjes <rientjes@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

commit c702418f ("mm: vmscan: do not keep kswapd looping forever due
to individual uncompactable zones") removed zone watermark checks from
the compaction code in kswapd but left in the zone congestion clearing,
which now happens unconditionally on higher order reclaim.

This messes up the reclaim throttling logic for zones with
dirty/writeback pages, where zones should only lose their congestion
status when their watermarks have been restored.

Remove the clearing from the zone compaction section entirely.  The
preliminary zone check and the reclaim loop in kswapd will clear it if
the zone is considered balanced.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When a zone meets its high watermark and is compactable in case of
higher order allocations, it contributes to the percentage of the node's
memory that is considered balanced.

This requirement, that a node be only partially balanced, came about
when kswapd was desparately trying to balance tiny zones when all bigger
zones in the node had plenty of free memory.  Arguably, the same should
apply to compaction: if a significant part of the node is balanced
enough to run compaction, do not get hung up on that tiny zone that
might never get in shape.

When the compaction logic in kswapd is reached, we know that at least
25% of the node's memory is balanced properly for compaction (see
zone_balanced and pgdat_balanced).  Remove the individual zone checks
that restart the kswapd cycle.

Otherwise, we may observe more endless looping in kswapd where the
compaction code loops back to reclaim because of a single zone and
reclaim does nothing because the node is considered balanced overall.

See for example

  https://bugzilla.redhat.com/show_bug.cgi?id=866988Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reported-and-tested-by: NThorsten Leemhuis <fedora@leemhuis.info>
Reported-by: NJiri Slaby <jslaby@suse.cz>
Tested-by: NJohn Ellson <john.ellson@comcast.net>
Tested-by: NZdenek Kabelac <zkabelac@redhat.com>
Tested-by: NBruno Wolff III <bruno@wolff.to>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Kswapd does not in all places have the same criteria for a balanced
zone.  Zones are only being reclaimed when their high watermark is
breached, but compaction checks loop over the zonelist again when the
zone does not meet the low watermark plus two times the size of the
allocation.  This gets kswapd stuck in an endless loop over a small
zone, like the DMA zone, where the high watermark is smaller than the
compaction requirement.

Add a function, zone_balanced(), that checks the watermark, and, for
higher order allocations, if compaction has enough free memory.  Then
use it uniformly to check for balanced zones.

This makes sure that when the compaction watermark is not met, at least
reclaim happens and progress is made - or the zone is declared
unreclaimable at some point and skipped entirely.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reported-by: NGeorge Spelvin <linux@horizon.com>
Reported-by: NJohannes Hirte <johannes.hirte@fem.tu-ilmenau.de>
Reported-by: NTomas Racek <tracek@redhat.com>
Tested-by: NJohannes Hirte <johannes.hirte@fem.tu-ilmenau.de>
Reviewed-by: NRik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 5515061d ("mm: throttle direct reclaimers if PF_MEMALLOC
reserves are low and swap is backed by network storage") introduced a
check for fatal signals after a process gets throttled for network
storage.  The intention was that if a process was throttled and got
killed that it should not trigger the OOM killer.  As pointed out by
Minchan Kim and David Rientjes, this check is in the wrong place and too
broad.  If a system is in am OOM situation and a process is exiting, it
can loop in __alloc_pages_slowpath() and calling direct reclaim in a
loop.  As the fatal signal is pending it returns 1 as if it is making
forward progress and can effectively deadlock.

This patch moves the fatal_signal_pending() check after throttling to
throttle_direct_reclaim() where it belongs.  If the process is killed
while throttled, it will return immediately without direct reclaim
except now it will have TIF_MEMDIE set and will use the PFMEMALLOC
reserves.

Minchan pointed out that it may be better to direct reclaim before
returning to avoid using the reserves because there may be pages that
can easily reclaim that would avoid using the reserves.  However, we do
no such targetted reclaim and there is no guarantee that suitable pages
are available.  As it is expected that this throttling happens when
swap-over-NFS is used there is a possibility that the process will
instead swap which may allocate network buffers from the PFMEMALLOC
reserves.  Hence, in the swap-over-nfs case where a process can be
throtted and be killed it can use the reserves to exit or it can
potentially use reserves to swap a few pages and then exit.  This patch
takes the option of using the reserves if necessary to allow the process
exit quickly.

If this patch passes review it should be considered a -stable candidate
for 3.6.
Signed-off-by: NMel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Magenheimer <dan.magenheimer@oracle.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Sonny Rao <sonnyrao@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Jiri Slaby reported the following:

	(It's an effective revert of "mm: vmscan: scale number of pages
	reclaimed by reclaim/compaction based on failures".) Given kswapd
	had hours of runtime in ps/top output yesterday in the morning
	and after the revert it's now 2 minutes in sum for the last 24h,
	I would say, it's gone.

The intention of the patch in question was to compensate for the loss of
lumpy reclaim.  Part of the reason lumpy reclaim worked is because it
aggressively reclaimed pages and this patch was meant to be a sane
compromise.

When compaction fails, it gets deferred and both compaction and
reclaim/compaction is deferred avoid excessive reclaim.  However, since
commit c6543459 ("mm: remove __GFP_NO_KSWAPD"), kswapd is woken up
each time and continues reclaiming which was not taken into account when
the patch was developed.

Attempts to address the problem ended up just changing the shape of the
problem instead of fixing it.  The release window gets closer and while
a THP allocation failing is not a major problem, kswapd chewing up a lot
of CPU is.

This patch reverts commit 83fde0f2 ("mm: vmscan: scale number of
pages reclaimed by reclaim/compaction based on failures") and will be
revisited in the future.
Signed-off-by: NMel Gorman <mgorman@suse.de>
Cc: Zdenek Kabelac <zkabelac@redhat.com>
Tested-by: NValdis Kletnieks <Valdis.Kletnieks@vt.edu>
Cc: Jiri Slaby <jirislaby@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Johannes Hirte <johannes.hirte@fem.tu-ilmenau.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

In kswapd(), set current->reclaim_state to NULL before returning, as
current->reclaim_state holds reference to variable on kswapd()'s stack.

In rare cases, while returning from kswapd() during memory offlining,
__free_slab() and freepages() can access the dangling pointer of
current->reclaim_state.
Signed-off-by: NTakamori Yamaguchi <takamori.yamaguchi@jp.sony.com>
Signed-off-by: NAaditya Kumar <aaditya.kumar@ap.sony.com>
Acked-by: NDavid Rientjes <rientjes@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Presently CMA cannot migrate mlocked pages so it ends up failing to allocate
contiguous memory space.

This patch makes mlocked pages be migrated out.  Of course, it can affect
realtime processes but in CMA usecase, contiguous memory allocation failing
is far worse than access latency to an mlocked page being variable while
CMA is running.  If someone wants to make the system realtime, he shouldn't
enable CMA because stalls can still happen at random times.

[akpm@linux-foundation.org: tweak comment text, per Mel]
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Acked-by: NMel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

page_evictable(page, vma) is an irritant: almost all its callers pass
NULL for vma.  Remove the vma arg and use mlocked_vma_newpage(vma, page)
explicitly in the couple of places it's needed.  But in those places we
don't even need page_evictable() itself!  They're dealing with a freshly
allocated anonymous page, which has no "mapping" and cannot be mlocked yet.
Signed-off-by: NHugh Dickins <hughd@google.com>
Acked-by: NMel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: NJohannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Compaction caches if a pageblock was scanned and no pages were isolated so
that the pageblocks can be skipped in the future to reduce scanning.  This
information is not cleared by the page allocator based on activity due to
the impact it would have to the page allocator fast paths.  Hence there is
a requirement that something clear the cache or pageblocks will be skipped
forever.  Currently the cache is cleared if there were a number of recent
allocation failures and it has not been cleared within the last 5 seconds.
Time-based decisions like this are terrible as they have no relationship
to VM activity and is basically a big hammer.

Unfortunately, accurate heuristics would add cost to some hot paths so
this patch implements a rough heuristic.  There are two cases where the
cache is cleared.

1. If a !kswapd process completes a compaction cycle (migrate and free
   scanner meet), the zone is marked compact_blockskip_flush. When kswapd
   goes to sleep, it will clear the cache. This is expected to be the
   common case where the cache is cleared. It does not really matter if
   kswapd happens to be asleep or going to sleep when the flag is set as
   it will be woken on the next allocation request.

2. If there have been multiple failures recently and compaction just
   finished being deferred then a process will clear the cache and start a
   full scan.  This situation happens if there are multiple high-order
   allocation requests under heavy memory pressure.

The clearing of the PG_migrate_skip bits and other scans is inherently
racy but the race is harmless.  For allocations that can fail such as THP,
they will simply fail.  For requests that cannot fail, they will retry the
allocation.  Tests indicated that scanning rates were roughly similar to
when the time-based heuristic was used and the allocation success rates
were similar.
Signed-off-by: NMel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Richard Davies <richard@arachsys.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Avi Kivity <avi@redhat.com>
Cc: Rafael Aquini <aquini@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Drop clean cache pages instead of migration during alloc_contig_range() to
minimise allocation latency by reducing the amount of migration that is
necessary.  It's useful for CMA because latency of migration is more
important than evicting the background process's working set.  In
addition, as pages are reclaimed then fewer free pages for migration
targets are required so it avoids memory reclaiming to get free pages,
which is a contributory factor to increased latency.

I measured elapsed time of __alloc_contig_migrate_range() which migrates
10M in 40M movable zone in QEMU machine.

Before - 146ms, After - 7ms

[akpm@linux-foundation.org: fix nommu build]
Signed-off-by: NMel Gorman <mgorman@suse.de>
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Reviewed-by: NMel Gorman <mgorman@suse.de>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: NMichal Nazarewicz <mina86@mina86.com>
Cc: Rik van Riel <riel@redhat.com>
Tested-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix the return value while failing to create the kswapd kernel thread.
Also, the error message is prioritized as KERN_ERR.
Signed-off-by: NGavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: NWanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If allocation fails after compaction then compaction may be deferred for
a number of allocation attempts.  If there are subsequent failures,
compact_defer_shift is increased to defer for longer periods.  This
patch uses that information to scale the number of pages reclaimed with
compact_defer_shift until allocations succeed again.  The rationale is
that reclaiming the normal number of pages still allowed compaction to
fail and its success depends on the number of pages.  If it's failing,
reclaim more pages until it succeeds again.

Note that this is not implying that VM reclaim is not reclaiming enough
pages or that its logic is broken.  try_to_free_pages() always asks for
SWAP_CLUSTER_MAX pages to be reclaimed regardless of order and that is
what it does.  Direct reclaim stops normally with this check.

	if (sc->nr_reclaimed >= sc->nr_to_reclaim)
		goto out;

should_continue_reclaim delays when that check is made until a minimum
number of pages for reclaim/compaction are reclaimed.  It is possible
that this patch could instead set nr_to_reclaim in try_to_free_pages()
and drive it from there but that's behaves differently and not
necessarily for the better.  If driven from do_try_to_free_pages(), it
is also possible that priorities will rise.

When they reach DEF_PRIORITY-2, it will also start stalling and setting
pages for immediate reclaim which is more disruptive than not desirable
in this case.  That is a more wide-reaching change that could cause
another regression related to THP requests causing interactive jitter.

[akpm@linux-foundation.org: fix build]
Signed-off-by: NMel Gorman <mgorman@suse.de>
Acked-by: NRik van Riel <riel@redhat.com>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If kthread_run() fails, pgdat->kswapd contains errno.  When we stop this
thread, we only check whether pgdat->kswapd is NULL and access it.  If
it contains errno, it will cause page fault.  Reset pgdat->kswapd to
NULL when creating kernel thread fails can avoid this problem.
Signed-off-by: NWen Congyang <wency@cn.fujitsu.com>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

I noticed in a multi-process parallel files reading benchmark I ran on a 8
socket machine, throughput slowed down by a factor of 8 when I ran the
benchmark within a cgroup container.  I traced the problem to the
following code path (see below) when we are trying to reclaim memory from
file cache.  The res_counter_uncharge function is called on every page
that's reclaimed and created heavy lock contention.  The patch below
allows the reclaimed pages to be uncharged from the resource counter in
batch and recovered the regression.

Tim

     40.67%           usemem  [kernel.kallsyms]                   [k] _raw_spin_lock
                      |
                      --- _raw_spin_lock
                         |
                         |--92.61%-- res_counter_uncharge
                         |          |
                         |          |--100.00%-- __mem_cgroup_uncharge_common
                         |          |          |
                         |          |          |--100.00%-- mem_cgroup_uncharge_cache_page
                         |          |          |          __remove_mapping
                         |          |          |          shrink_page_list
                         |          |          |          shrink_inactive_list
                         |          |          |          shrink_mem_cgroup_zone
                         |          |          |          shrink_zone
                         |          |          |          do_try_to_free_pages
                         |          |          |          try_to_free_pages
                         |          |          |          __alloc_pages_nodemask
                         |          |          |          alloc_pages_current
Signed-off-by: NTim Chen <tim.c.chen@linux.intel.com>
Acked-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: NJohannes Weiner <hannes@cmpxchg.org>
Acked-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The may_enter_fs test turns out to be too restrictive: though I saw no
problem with it when testing on 3.5-rc6, it very soon OOMed when I tested
on 3.5-rc6-mm1.  I don't know what the difference there is, perhaps I just
slightly changed the way I started off the testing: dd if=/dev/zero
of=/mnt/temp bs=1M count=1024; rm -f /mnt/temp; sync repeatedly, in 20M
memory.limit_in_bytes cgroup to ext4 on USB stick.

ext4 (and gfs2 and xfs) turn out to allocate new pages for writing with
AOP_FLAG_NOFS: that seems a little worrying, and it's unclear to me why
the transaction needs to be started even before allocating pagecache
memory.  But it may not be worth worrying about these days: if direct
reclaim avoids FS writeback, does __GFP_FS now mean anything?

Anyway, we insisted on the may_enter_fs test to avoid hangs with the loop
device; but since that also masks off __GFP_IO, we can test for __GFP_IO
directly, ignoring may_enter_fs and __GFP_FS.

But even so, the test still OOMs sometimes: when originally testing on
3.5-rc6, it OOMed about one time in five or ten; when testing just now on
3.5-rc6-mm1, it OOMed on the first iteration.

This residual problem comes from an accumulation of pages under ordinary
writeback, not marked PageReclaim, so rightly not causing the memcg check
to wait on their writeback: these too can prevent shrink_page_list() from
freeing any pages, so many times that memcg reclaim fails and OOMs.

Deal with these in the same way as direct reclaim now deals with dirty FS
pages: mark them PageReclaim.  It is appropriate to rotate these to tail
of list when writepage completes, but more importantly, the PageReclaim
flag makes memcg reclaim wait on them if encountered again.  Increment
NR_VMSCAN_IMMEDIATE?  That's arguable: I chose not.

Setting PageReclaim here may occasionally race with end_page_writeback()
clearing it: lru_deactivate_fn() already faced the same race, and
correctly concluded that the window is small and the issue non-critical.

With these changes, the test runs indefinitely without OOMing on ext4,
ext3 and ext2: I'll move on to test with other filesystems later.

Trivia: invert conditions for a clearer block without an else, and goto
keep_locked to do the unlock_page.
Signed-off-by: NHugh Dickins <hughd@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujtisu.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Acked-by: NMichal Hocko <mhocko@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The current implementation of dirty pages throttling is not memcg aware
which makes it easy to have memcg LRUs full of dirty pages.  Without
throttling, these LRUs can be scanned faster than the rate of writeback,
leading to memcg OOM conditions when the hard limit is small.

This patch fixes the problem by throttling the allocating process
(possibly a writer) during the hard limit reclaim by waiting on
PageReclaim pages.  We are waiting only for PageReclaim pages because
those are the pages that made one full round over LRU and that means that
the writeback is much slower than scanning.

The solution is far from being ideal - long term solution is memcg aware
dirty throttling - but it is meant to be a band aid until we have a real
fix.  We are seeing this happening during nightly backups which are placed
into containers to prevent from eviction of the real working set.

The change affects only memcg reclaim and only when we encounter
PageReclaim pages which is a signal that the reclaim doesn't catch up on
with the writers so somebody should be throttled.  This could be
potentially unfair because it could be somebody else from the group who
gets throttled on behalf of the writer but as writers need to allocate as
well and they allocate in higher rate the probability that only innocent
processes would be penalized is not that high.

I have tested this change by a simple dd copying /dev/zero to tmpfs or
ext3 running under small memcg (1G copy under 5M, 60M, 300M and 2G
containers) and dd got killed by OOM killer every time.  With the patch I
could run the dd with the same size under 5M controller without any OOM.
The issue is more visible with slower devices for output.

* With the patch
================
* tmpfs size=2G
---------------
$ vim cgroup_cache_oom_test.sh
$ ./cgroup_cache_oom_test.sh 5M
using Limit 5M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 30.4049 s, 34.5 MB/s
$ ./cgroup_cache_oom_test.sh 60M
using Limit 60M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 31.4561 s, 33.3 MB/s
$ ./cgroup_cache_oom_test.sh 300M
using Limit 300M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 20.4618 s, 51.2 MB/s
$ ./cgroup_cache_oom_test.sh 2G
using Limit 2G for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 1.42172 s, 738 MB/s

* ext3
------
$ ./cgroup_cache_oom_test.sh 5M
using Limit 5M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 27.9547 s, 37.5 MB/s
$ ./cgroup_cache_oom_test.sh 60M
using Limit 60M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 30.3221 s, 34.6 MB/s
$ ./cgroup_cache_oom_test.sh 300M
using Limit 300M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 24.5764 s, 42.7 MB/s
$ ./cgroup_cache_oom_test.sh 2G
using Limit 2G for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 3.35828 s, 312 MB/s

* Without the patch
===================
* tmpfs size=2G
---------------
$ ./cgroup_cache_oom_test.sh 5M
using Limit 5M for group
./cgroup_cache_oom_test.sh: line 46:  4668 Killed                  dd if=/dev/zero of=$OUT/zero bs=1M count=$count
$ ./cgroup_cache_oom_test.sh 60M
using Limit 60M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 25.4989 s, 41.1 MB/s
$ ./cgroup_cache_oom_test.sh 300M
using Limit 300M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 24.3928 s, 43.0 MB/s
$ ./cgroup_cache_oom_test.sh 2G
using Limit 2G for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 1.49797 s, 700 MB/s

* ext3
------
$ ./cgroup_cache_oom_test.sh 5M
using Limit 5M for group
./cgroup_cache_oom_test.sh: line 46:  4689 Killed                  dd if=/dev/zero of=$OUT/zero bs=1M count=$count
$ ./cgroup_cache_oom_test.sh 60M
using Limit 60M for group
./cgroup_cache_oom_test.sh: line 46:  4692 Killed                  dd if=/dev/zero of=$OUT/zero bs=1M count=$count
$ ./cgroup_cache_oom_test.sh 300M
using Limit 300M for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 20.248 s, 51.8 MB/s
$ ./cgroup_cache_oom_test.sh 2G
using Limit 2G for group
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB) copied, 2.85201 s, 368 MB/s

[akpm@linux-foundation.org: tweak changelog, reordered the test to optimize for CONFIG_CGROUP_MEM_RES_CTLR=n]
[hughd@google.com: fix deadlock with loop driver]
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujtisu.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ying Han <yinghan@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Reviewed-by: NMel Gorman <mgorman@suse.de>
Acked-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NMichal Hocko <mhocko@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>