mm: oom: let oom_reap_task and exit_mmap run concurrently

This is purely required because exit_aio() may block and exit_mmap() may
never start, if the oom_reap_task cannot start running on a mm with
mm_users == 0.

At the same time if the OOM reaper doesn't wait at all for the memory of
the current OOM candidate to be freed by exit_mmap->unmap_vmas, it would
generate a spurious OOM kill.

If it wasn't because of the exit_aio or similar blocking functions in
the last mmput, it would be enough to change the oom_reap_task() in the
case it finds mm_users == 0, to wait for a timeout or to wait for
__mmput to set MMF_OOM_SKIP itself, but it's not just exit_mmap the
problem here so the concurrency of exit_mmap and oom_reap_task is
apparently warranted.

It's a non standard runtime, exit_mmap() runs without mmap_sem, and
oom_reap_task runs with the mmap_sem for reading as usual (kind of
MADV_DONTNEED).

The race between the two is solved with a combination of
tsk_is_oom_victim() (serialized by task_lock) and MMF_OOM_SKIP
(serialized by a dummy down_write/up_write cycle on the same lines of
the ksm_exit method).

If the oom_reap_task() may be running concurrently during exit_mmap,
exit_mmap will wait it to finish in down_write (before taking down mm
structures that would make the oom_reap_task fail with use after free).

If exit_mmap comes first, oom_reap_task() will skip the mm if
MMF_OOM_SKIP is already set and in turn all memory is already freed and
furthermore the mm data structures may already have been taken down by
free_pgtables.

[aarcange@redhat.com: incremental one liner]
Link: http://lkml.kernel.org/r/20170726164319.GC29716@redhat.com
[rientjes@google.com: remove unused mmput_async]
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1708141733130.50317@chino.kir.corp.google.com
[aarcange@redhat.com: microoptimization]
Link: http://lkml.kernel.org/r/20170817171240.GB5066@redhat.com
Link: http://lkml.kernel.org/r/20170726162912.GA29716@redhat.com
Fixes: 26db62f179d1 ("oom: keep mm of the killed task available")
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Reported-by: David Rientjes <rientjes@google.com>
Tested-by: David Rientjes <rientjes@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swap: choose swap device according to numa node

If the system has more than one swap device and swap device has the node
information, we can make use of this information to decide which swap
device to use in get_swap_pages() to get better performance.

The current code uses a priority based list, swap_avail_list, to decide
which swap device to use and if multiple swap devices share the same
priority, they are used round robin.  This patch changes the previous
single global swap_avail_list into a per-numa-node list, i.e.  for each
numa node, it sees its own priority based list of available swap
devices.  Swap device's priority can be promoted on its matching node's
swap_avail_list.

The current swap device's priority is set as: user can set a >=0 value,
or the system will pick one starting from -1 then downwards.  The
priority value in the swap_avail_list is the negated value of the swap
device's due to plist being sorted from low to high.  The new policy
doesn't change the semantics for priority >=0 cases, the previous
starting from -1 then downwards now becomes starting from -2 then
downwards and -1 is reserved as the promoted value.

Take 4-node EX machine as an example, suppose 4 swap devices are
available, each sit on a different node:
swapA on node 0
swapB on node 1
swapC on node 2
swapD on node 3

After they are all swapped on in the sequence of ABCD.

Current behaviour:
their priorities will be:
swapA: -1
swapB: -2
swapC: -3
swapD: -4
And their position in the global swap_avail_list will be:
swapA   -> swapB   -> swapC   -> swapD
prio:1     prio:2     prio:3     prio:4

New behaviour:
their priorities will be(note that -1 is skipped):
swapA: -2
swapB: -3
swapC: -4
swapD: -5
And their positions in the 4 swap_avail_lists[nid] will be:
swap_avail_lists[0]: /* node 0's available swap device list */
swapA   -> swapB   -> swapC   -> swapD
prio:1     prio:3     prio:4     prio:5
swap_avali_lists[1]: /* node 1's available swap device list */
swapB   -> swapA   -> swapC   -> swapD
prio:1     prio:2     prio:4     prio:5
swap_avail_lists[2]: /* node 2's available swap device list */
swapC   -> swapA   -> swapB   -> swapD
prio:1     prio:2     prio:3     prio:5
swap_avail_lists[3]: /* node 3's available swap device list */
swapD   -> swapA   -> swapB   -> swapC
prio:1     prio:2     prio:3     prio:4

To see the effect of the patch, a test that starts N process, each mmap
a region of anonymous memory and then continually write to it at random
position to trigger both swap in and out is used.

On a 2 node Skylake EP machine with 64GiB memory, two 170GB SSD drives
are used as swap devices with each attached to a different node, the
result is:

runtime=30m/processes=32/total test size=128G/each process mmap region=4G
kernel         throughput
vanilla        13306
auto-binding   15169 +14%

runtime=30m/processes=64/total test size=128G/each process mmap region=2G
kernel         throughput
vanilla        11885
auto-binding   14879 +25%

[aaron.lu@intel.com: v2]
Link: http://lkml.kernel.org/r/20170814053130.GD2369@aaronlu.sh.intel.com
Link: http://lkml.kernel.org/r/20170816024439.GA10925@aaronlu.sh.intel.com
[akpm@linux-foundation.org: use kmalloc_array()]
Link: http://lkml.kernel.org/r/20170814053130.GD2369@aaronlu.sh.intel.com
Link: http://lkml.kernel.org/r/20170816024439.GA10925@aaronlu.sh.intel.com
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
Cc: "Chen, Tim C" <tim.c.chen@intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: replace TIF_MEMDIE checks by tsk_is_oom_victim

TIF_MEMDIE is set only to the tasks whick were either directly selected
by the OOM killer or passed through mark_oom_victim from the allocator
path.  tsk_is_oom_victim is more generic and allows to identify all
tasks (threads) which share the mm with the oom victim.

Please note that the freezer still needs to check TIF_MEMDIE because we
cannot thaw tasks which do not participage in oom_victims counting
otherwise a !TIF_MEMDIE task could interfere after oom_disbale returns.

Link: http://lkml.kernel.org/r/20170810075019.28998-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: do not rely on TIF_MEMDIE for memory reserves access

For ages we have been relying on TIF_MEMDIE thread flag to mark OOM
victims and then, among other things, to give these threads full access
to memory reserves.  There are few shortcomings of this implementation,
though.

First of all and the most serious one is that the full access to memory
reserves is quite dangerous because we leave no safety room for the
system to operate and potentially do last emergency steps to move on.

Secondly this flag is per task_struct while the OOM killer operates on
mm_struct granularity so all processes sharing the given mm are killed.
Giving the full access to all these task_structs could lead to a quick
memory reserves depletion.  We have tried to reduce this risk by giving
TIF_MEMDIE only to the main thread and the currently allocating task but
that doesn't really solve this problem while it surely opens up a room
for corner cases - e.g.  GFP_NO{FS,IO} requests might loop inside the
allocator without access to memory reserves because a particular thread
was not the group leader.

Now that we have the oom reaper and that all oom victims are reapable
after 1b51e65eab64 ("oom, oom_reaper: allow to reap mm shared by the
kthreads") we can be more conservative and grant only partial access to
memory reserves because there are reasonable chances of the parallel
memory freeing.  We still want some access to reserves because we do not
want other consumers to eat up the victim's freed memory.  oom victims
will still contend with __GFP_HIGH users but those shouldn't be so
aggressive to starve oom victims completely.

Introduce ALLOC_OOM flag and give all tsk_is_oom_victim tasks access to
the half of the reserves.  This makes the access to reserves independent
on which task has passed through mark_oom_victim.  Also drop any usage
of TIF_MEMDIE from the page allocator proper and replace it by
tsk_is_oom_victim as well which will make page_alloc.c completely
TIF_MEMDIE free finally.

CONFIG_MMU=n doesn't have oom reaper so let's stick to the original
ALLOC_NO_WATERMARKS approach.

There is a demand to make the oom killer memcg aware which will imply
many tasks killed at once.  This change will allow such a usecase
without worrying about complete memory reserves depletion.

Link: http://lkml.kernel.org/r/20170810075019.28998-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

z3fold: use per-cpu unbuddied lists

It's been noted that z3fold doesn't scale well when it's run in a large
number of threads on many cores, which can be easily reproduced with fio
'randrw' test with --numjobs=32.  E.g.  the result for 1 cluster (4 cores)
is:

Run status group 0 (all jobs):
   READ: io=244785MB, aggrb=496883KB/s, minb=15527KB/s, ...
  WRITE: io=246735MB, aggrb=500841KB/s, minb=15651KB/s, ...

While for 8 cores (2 clusters) the result is:

Run status group 0 (all jobs):
   READ: io=244785MB, aggrb=265942KB/s, minb=8310KB/s, ...
  WRITE: io=246735MB, aggrb=268060KB/s, minb=8376KB/s, ...

The bottleneck here is the pool lock which many threads become waiting
upon.  To reduce that spin lock contention, z3fold can operate only on
the lists local to the current CPU whenever possible.  Due to the nature
of z3fold unbuddied list handling (it only takes the first entry off the
list on a hot path), if the z3fold pool is big enough and balanced well
enough, limiting search to only local unbuddied list doesn't lead to a
significant compression ratio degrade (2.57x vs 2.65x in our
measurements).

This patch also introduces two worker threads: one for async in-page
object layout optimization and one for releasing freed pages.  This is
done to speed up z3fold_free() which is often on a hot path.

The fio results for 8-core case are now the following:

Run status group 0 (all jobs):
   READ: io=244785MB, aggrb=1568.3MB/s, minb=50182KB/s, ...
  WRITE: io=246735MB, aggrb=1580.8MB/s, minb=50582KB/s, ...

So we're in for almost 6x performance increase.

Link: http://lkml.kernel.org/r/20170806181443.f9b65018f8bde25ef990f9e8@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: don't use VMA based swap readahead if HDD is used as swap

VMA based swap readahead will readahead the virtual pages that is
continuous in the virtual address space.  While the original swap
readahead will readahead the swap slots that is continuous in the swap
device.  Although VMA based swap readahead is more correct for the swap
slots to be readahead, it will trigger more small random readings, which
may cause the performance of HDD (hard disk) to degrade heavily, and may
finally exceed the benefit.

To avoid the issue, in this patch, if the HDD is used as swap, the VMA
based swap readahead will be disabled, and the original swap readahead
will be used instead.

Link: http://lkml.kernel.org/r/20170807054038.1843-6-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: add sysfs interface for VMA based swap readahead

The sysfs interface to control the VMA based swap readahead is added as
follow,

/sys/kernel/mm/swap/vma_ra_enabled

Enable the VMA based swap readahead algorithm, or use the original
global swap readahead algorithm.

/sys/kernel/mm/swap/vma_ra_max_order

Set the max order of the readahead window size for the VMA based swap
readahead algorithm.

The corresponding ABI documentation is added too.

Link: http://lkml.kernel.org/r/20170807054038.1843-5-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: VMA based swap readahead

The swap readahead is an important mechanism to reduce the swap in
latency.  Although pure sequential memory access pattern isn't very
popular for anonymous memory, the space locality is still considered
valid.

In the original swap readahead implementation, the consecutive blocks in
swap device are readahead based on the global space locality estimation.
But the consecutive blocks in swap device just reflect the order of page
reclaiming, don't necessarily reflect the access pattern in virtual
memory.  And the different tasks in the system may have different access
patterns, which makes the global space locality estimation incorrect.

In this patch, when page fault occurs, the virtual pages near the fault
address will be readahead instead of the swap slots near the fault swap
slot in swap device.  This avoid to readahead the unrelated swap slots.
At the same time, the swap readahead is changed to work on per-VMA from
globally.  So that the different access patterns of the different VMAs
could be distinguished, and the different readahead policy could be
applied accordingly.  The original core readahead detection and scaling
algorithm is reused, because it is an effect algorithm to detect the
space locality.

The test and result is as follow,

Common test condition
=====================

Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device:
NVMe disk

Micro-benchmark with combined access pattern
============================================

vm-scalability, sequential swap test case, 4 processes to eat 50G
virtual memory space, repeat the sequential memory writing until 300
seconds.  The first round writing will trigger swap out, the following
rounds will trigger sequential swap in and out.

At the same time, run vm-scalability random swap test case in
background, 8 processes to eat 30G virtual memory space, repeat the
random memory write until 300 seconds.  This will trigger random swap-in
in the background.

This is a combined workload with sequential and random memory accessing
at the same time.  The result (for sequential workload) is as follow,

Base Optimized
---- ---------
throughput 345413 KB/s 414029 KB/s (+19.9%)
latency.average 97.14 us 61.06 us (-37.1%)
latency.50th 2 us 1 us
latency.60th 2 us 1 us
latency.70th 98 us 2 us
latency.80th 160 us 2 us
latency.90th 260 us 217 us
latency.95th 346 us 369 us
latency.99th 1.34 ms 1.09 ms
ra_hit% 52.69% 99.98%

The original swap readahead algorithm is confused by the background
random access workload, so readahead hit rate is lower.  The VMA-base
readahead algorithm works much better.

Linpack
=======

The test memory size is bigger than RAM to trigger swapping.

Base Optimized
---- ---------
elapsed_time 393.49 s 329.88 s (-16.2%)
ra_hit% 86.21% 98.82%

The score of base and optimized kernel hasn't visible changes.  But the
elapsed time reduced and readahead hit rate improved, so the optimized
kernel runs better for startup and tear down stages.  And the absolute
value of readahead hit rate is high, shows that the space locality is
still valid in some practical workloads.

Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: fix swap readahead marking

In the original implementation, it is possible that the existing pages
in the swap cache (not newly readahead) could be marked as the readahead
pages.  This will cause the statistics of swap readahead be wrong and
influence the swap readahead algorithm too.

This is fixed via marking a page as the readahead page only if it is
newly allocated and read from the disk.

When testing with linpack, after the fixing the swap readahead hit rate
increased from ~66% to ~86%.

Link: http://lkml.kernel.org/r/20170807054038.1843-3-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: add swap readahead hit statistics

Patch series "mm, swap: VMA based swap readahead", v4.

The swap readahead is an important mechanism to reduce the swap in
latency.  Although pure sequential memory access pattern isn't very
popular for anonymous memory, the space locality is still considered
valid.

In the original swap readahead implementation, the consecutive blocks in
swap device are readahead based on the global space locality estimation.
But the consecutive blocks in swap device just reflect the order of page
reclaiming, don't necessarily reflect the access pattern in virtual
memory space.  And the different tasks in the system may have different
access patterns, which makes the global space locality estimation
incorrect.

In this patchset, when page fault occurs, the virtual pages near the
fault address will be readahead instead of the swap slots near the fault
swap slot in swap device.  This avoid to readahead the unrelated swap
slots.  At the same time, the swap readahead is changed to work on
per-VMA from globally.  So that the different access patterns of the
different VMAs could be distinguished, and the different readahead
policy could be applied accordingly.  The original core readahead
detection and scaling algorithm is reused, because it is an effect
algorithm to detect the space locality.

In addition to the swap readahead changes, some new sysfs interface is
added to show the efficiency of the readahead algorithm and some other
swap statistics.

This new implementation will incur more small random read, on SSD, the
improved correctness of estimation and readahead target should beat the
potential increased overhead, this is also illustrated in the test
results below.  But on HDD, the overhead may beat the benefit, so the
original implementation will be used by default.

The test and result is as follow,

Common test condition
=====================

Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM)
Swap device: NVMe disk

Micro-benchmark with combined access pattern
============================================

vm-scalability, sequential swap test case, 4 processes to eat 50G
virtual memory space, repeat the sequential memory writing until 300
seconds.  The first round writing will trigger swap out, the following
rounds will trigger sequential swap in and out.

At the same time, run vm-scalability random swap test case in
background, 8 processes to eat 30G virtual memory space, repeat the
random memory write until 300 seconds.  This will trigger random swap-in
in the background.

This is a combined workload with sequential and random memory accessing
at the same time.  The result (for sequential workload) is as follow,

Base Optimized
---- ---------
throughput 345413 KB/s 414029 KB/s (+19.9%)
latency.average 97.14 us 61.06 us (-37.1%)
latency.50th 2 us 1 us
latency.60th 2 us 1 us
latency.70th 98 us 2 us
latency.80th 160 us 2 us
latency.90th 260 us 217 us
latency.95th 346 us 369 us
latency.99th 1.34 ms 1.09 ms
ra_hit% 52.69% 99.98%

The original swap readahead algorithm is confused by the background
random access workload, so readahead hit rate is lower.  The VMA-base
readahead algorithm works much better.

Linpack
=======

The test memory size is bigger than RAM to trigger swapping.

Base Optimized
---- ---------
elapsed_time 393.49 s 329.88 s (-16.2%)
ra_hit% 86.21% 98.82%

The score of base and optimized kernel hasn't visible changes.  But the
elapsed time reduced and readahead hit rate improved, so the optimized
kernel runs better for startup and tear down stages.  And the absolute
value of readahead hit rate is high, shows that the space locality is
still valid in some practical workloads.

This patch (of 5):

The statistics for total readahead pages and total readahead hits are
recorded and exported via the following sysfs interface.

/sys/kernel/mm/swap/ra_hits
/sys/kernel/mm/swap/ra_total

With them, the efficiency of the swap readahead could be measured, so
that the swap readahead algorithm and parameters could be tuned
accordingly.

[akpm@linux-foundation.org: don't display swap stats if CONFIG_SWAP=n]
Link: http://lkml.kernel.org/r/20170807054038.1843-2-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c: don't reinvent the wheel but use existing llist API

Although llist provides proper APIs, they are not used.  Make them used.

Link: http://lkml.kernel.org/r/1502095374-16112-1-git-send-email-byungchul.park@lge.com
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
Cc: zijun_hu <zijun_hu@htc.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Joel Fernandes <joelaf@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmstat.c: fix wrong comment

Comment for pagetypeinfo_showblockcount() is mistakenly duplicated from
pagetypeinfo_show_free()'s comment.  This commit fixes it.

Link: http://lkml.kernel.org/r/20170809185816.11244-1-sj38.park@gmail.com
Fixes: 467c996c1e19 ("Print out statistics in relation to fragmentation avoidance to /proc/pagetypeinfo")
Signed-off-by: SeongJae Park <sj38.park@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests/memfd: add memfd_create hugetlbfs selftest

With the addition of hugetlbfs support in memfd_create, the memfd
selftests should verify correct functionality with hugetlbfs.

Instead of writing a separate memfd hugetlbfs test, modify the
memfd_test program to take an optional argument 'hugetlbfs'.  If the
hugetlbfs argument is specified, basic memfd_create functionality will
be exercised on hugetlbfs.  If hugetlbfs is not specified, the current
functionality of the test is unchanged.

Note that many of the tests in memfd_test test file sealing operations.
hugetlbfs does not support file sealing, therefore for hugetlbfs all
sealing related tests are skipped.

In order to test on hugetlbfs, there needs to be preallocated huge
pages.  A new script (run_tests) is added.  This script will first run
the existing memfd_create tests.  It will then, attempt to allocate the
required number of huge pages before running the hugetlbfs test.  At the
end of testing, it will release any huge pages allocated for testing
purposes.

Link: http://lkml.kernel.org/r/1502495772-24736-3-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/shmem: add hugetlbfs support to memfd_create()

This patch came out of discussions in this e-mail thread:
  http://lkml.kernel.org/r/1499357846-7481-1-git-send-email-mike.kravetz%40oracle.com

The Oracle JVM team is developing a new garbage collection model.  This
new model requires multiple mappings of the same anonymous memory.  One
straight forward way to accomplish this is with memfd_create.  They can
use the returned fd to create multiple mappings of the same memory.

The JVM today has an option to use (static hugetlb) huge pages.  If this
option is specified, they would like to use the same garbage collection
model requiring multiple mappings to the same memory.  Using hugetlbfs,
it is possible to explicitly mount a filesystem and specify file paths
in order to get an fd that can be used for multiple mappings.  However,
this introduces additional system admin work and coordination.

Ideally they would like to get a hugetlbfs fd without requiring explicit
mounting of a filesystem.  Today, mmap and shmget can make use of
hugetlbfs without explicitly mounting a filesystem.  The patch adds this
functionality to memfd_create.

Add a new flag MFD_HUGETLB to memfd_create() that will specify the file
to be created resides in the hugetlbfs filesystem.  This is the generic
hugetlbfs filesystem not associated with any specific mount point.  As
with other system calls that request hugetlbfs backed pages, there is
the ability to encode huge page size in the flag arguments.

hugetlbfs does not support sealing operations, therefore specifying
MFD_ALLOW_SEALING with MFD_HUGETLB will result in EINVAL.

Of course, the memfd_man page would need updating if this type of
functionality moves forward.

Link: http://lkml.kernel.org/r/1502149672-7759-2-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, devm_memremap_pages: use multi-order radix for ZONE_DEVICE lookups

devm_memremap_pages() records mapped ranges in pgmap_radix with an entry
per section's worth of memory (128MB).  The key for each of those
entries is a section number.

This leads to false positives when devm_memremap_pages() is passed a
section-unaligned range as lookups in the misalignment fail to return
NULL.  We can close this hole by using the pfn as the key for entries in
the tree.  The number of entries required to describe a remapped range
is reduced by leveraging multi-order entries.

In practice this approach usually yields just one entry in the tree if
the size and starting address are of the same power-of-2 alignment.
Previously we always needed nr_entries = mapping_size / 128MB.

Link: https://lists.01.org/pipermail/linux-nvdimm/2016-August/006666.html
Link: http://lkml.kernel.org/r/150215410565.39310.13767886055248249438.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reported-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c: halve the number of comparisons performed in pcpu_get_vm_areas()

In pcpu_get_vm_areas(), it checks each range is not overlapped.  To make
sure it is, only (N^2)/2 comparison is necessary, while current code
does N^2 times.  By starting from the next range, it achieves the goal
and the continue could be removed.

Also,

 - the overlap check of two ranges could be done with one clause

 - one typo in comment is fixed.

Link: http://lkml.kernel.org/r/20170803063822.48702-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmstat: fix divide error at __fragmentation_index

When order is -1 or too big, *1UL << order* will be 0, which will cause
a divide error.  Although it seems that all callers of
__fragmentation_index() will only do so with a valid order, the patch
can make it more robust.

Should prevent reoccurrences of
https://bugzilla.kernel.org/show_bug.cgi?id=196555

Link: http://lkml.kernel.org/r/1501751520-2598-1-git-send-email-wen.yang99@zte.com.cn
Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
Reviewed-by: Jiang Biao <jiang.biao2@zte.com.cn>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hugetlb: do not allocate non-migrateable gigantic pages from movable zones

alloc_gigantic_page doesn't consider movability of the gigantic hugetlb
when scanning eligible ranges for the allocation.  As 1GB hugetlb pages
are not movable currently this can break the movable zone assumption
that all allocations are migrateable and as such break memory hotplug.

Reorganize the code and use the standard zonelist allocations scheme
that we use for standard hugetbl pages.  htlb_alloc_mask will ensure
that only migratable hugetlb pages will ever see a movable zone.

Link: http://lkml.kernel.org/r/20170803083549.21407-1-mhocko@kernel.org
Fixes: 944d9fec8d7a ("hugetlb: add support for gigantic page allocation at runtime")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: provide pid in userfault msg - add feat union

No ABI change, but this will make it more explicit to software that ptid
is only available if requested by passing UFFD_FEATURE_THREAD_ID to
UFFDIO_API.  The fact it's a union will also self document it shouldn't
be taken for granted there's a tpid there.

Link: http://lkml.kernel.org/r/20170802165145.22628-7-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Alexey Perevalov <a.perevalov@samsung.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: provide pid in userfault msg

It could be useful for calculating downtime during postcopy live
migration per vCPU.  Side observer or application itself will be
informed about proper task's sleep during userfaultfd processing.

Process's thread id is being provided when user requeste it by setting
UFFD_FEATURE_THREAD_ID bit into uffdio_api.features.

Link: http://lkml.kernel.org/r/20170802165145.22628-6-aarcange@redhat.com
Signed-off-by: Alexey Perevalov <a.perevalov@samsung.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: call userfaultfd_unmap_prep only if __split_vma succeeds

A __split_vma is not a worthy event to report, and it's definitely not a
unmap so it would be incorrect to report unmap for the whole region to
the userfaultfd manager if a __split_vma fails.

So only call userfaultfd_unmap_prep after the __vma_splitting is over
and do_munmap cannot fail anymore.

Also add unlikely because it's better to optimize for the vast majority
of apps that aren't using userfaultfd in a non cooperative way.  Ideally
we should also find a way to eliminate the branch entirely if
CONFIG_USERFAULTFD=n, but it would complicate things so stick to
unlikely for now.

Link: http://lkml.kernel.org/r/20170802165145.22628-5-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Alexey Perevalov <a.perevalov@samsung.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: selftest: explicit failure if the SIGBUS test failed

Showing zero in the output isn't very self explanatory as a successful
result.  Show a more explicit error output if the test fails.

Link: http://lkml.kernel.org/r/20170802165145.22628-4-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Alexey Perevalov <a.perevalov@samsung.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: selftest: exercise UFFDIO_COPY/ZEROPAGE -EEXIST

This will retry the UFFDIO_COPY/ZEROPAGE to verify it returns -EEXIST at
the first invocation and then later every 10 seconds.

In the filebacked MAP_SHARED case this also verifies the -EEXIST
triggered in the filesystem pagecache insertion, if the offset in the
file was not a hole.

shmem MAP_SHARED tries to index the newly allocated pagecache in the
radix tree before checking the pagetable so it doesn't need any
assistance to exercise that case.

hugetlbfs checks the pmd to be not none before trying to index the
hugetlbfs page in the radix tree, so it requires to run UFFDIO_COPY into
an alias mapping (the alternative would be to use MADV_DONTNEED to only
zap the pagetables, but that doesn't work on hugetlbfs).

[akpm@linux-foundation.org: fix uffdio_zeropage(), per Mike Kravetz]
Link: http://lkml.kernel.org/r/20170802165145.22628-3-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Alexey Perevalov <a.perevalov@samsung.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: selftest: add tests for UFFD_FEATURE_SIGBUS feature

Add tests for UFFD_FEATURE_SIGBUS feature.  The tests will verify signal
delivery instead of userfault events.  Also, test use of UFFDIO_COPY to
allocate memory and retry accessing monitored area after signal
delivery.

Also fix a bug in uffd_poll_thread() where 'uffd' is leaked.

Link: http://lkml.kernel.org/r/1501552446-748335-3-git-send-email-prakash.sangappa@oracle.com
Signed-off-by: Prakash Sangappa <prakash.sangappa@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: userfaultfd: add feature to request for a signal delivery

In some cases, userfaultfd mechanism should just deliver a SIGBUS signal
to the faulting process, instead of the page-fault event.  Dealing with
page-fault event using a monitor thread can be an overhead in these
cases.  For example applications like the database could use the
signaling mechanism for robustness purpose.

Database uses hugetlbfs for performance reason.  Files on hugetlbfs
filesystem are created and huge pages allocated using fallocate() API.
Pages are deallocated/freed using fallocate() hole punching support.
These files are mmapped and accessed by many processes as shared memory.
The database keeps track of which offsets in the hugetlbfs file have
pages allocated.

Any access to mapped address over holes in the file, which can occur due
to bugs in the application, is considered invalid and expect the process
to simply receive a SIGBUS.  However, currently when a hole in the file
is accessed via the mapped address, kernel/mm attempts to automatically
allocate a page at page fault time, resulting in implicitly filling the
hole in the file.  This may not be the desired behavior for applications
like the database that want to explicitly manage page allocations of
hugetlbfs files.

Using userfaultfd mechanism with this support to get a signal, database
application can prevent pages from being allocated implicitly when
processes access mapped address over holes in the file.

This patch adds UFFD_FEATURE_SIGBUS feature to userfaultfd mechnism to
request for a SIGBUS signal.

See following for previous discussion about the database requirement
leading to this proposal as suggested by Andrea.

http://www.spinics.net/lists/linux-mm/msg129224.html

Link: http://lkml.kernel.org/r/1501552446-748335-2-git-send-email-prakash.sangappa@oracle.com
Signed-off-by: Prakash Sangappa <prakash.sangappa@oracle.com>
Reviewed-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: rename global_page_state to global_zone_page_state

global_page_state is error prone as a recent bug report pointed out [1].
It only returns proper values for zone based counters as the enum it
gets suggests.  We already have global_node_page_state so let's rename
global_page_state to global_zone_page_state to be more explicit here.
All existing users seems to be correct:

$ git grep "global_page_state(NR_" | sed 's@.*(\(NR_[A-Z_]*\)).*@\1@' | sort | uniq -c
      2 NR_BOUNCE
      2 NR_FREE_CMA_PAGES
     11 NR_FREE_PAGES
      1 NR_KERNEL_STACK_KB
      1 NR_MLOCK
      2 NR_PAGETABLE

This patch shouldn't introduce any functional change.

[1] http://lkml.kernel.org/r/201707260628.v6Q6SmaS030814@www262.sakura.ne.jp

Link: http://lkml.kernel.org/r/20170801134256.5400-2-hannes@cmpxchg.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: shm: use new hugetlb size encoding definitions

Use the common definitions from hugetlb_encode.h header file for
encoding hugetlb size definitions in shmget system call flags.

In addition, move these definitions from the internal (kernel) to user
(uapi) header file.

Link: http://lkml.kernel.org/r/1501527386-10736-4-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Suggested-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: arch: consolidate mmap hugetlb size encodings

A non-default huge page size can be encoded in the flags argument of the
mmap system call.  The definitions for these encodings are in arch
specific header files.  However, all architectures use the same values.

Consolidate all the definitions in the primary user header file
(uapi/linux/mman.h).  Include definitions for all known huge page sizes.
Use the generic encoding definitions in hugetlb_encode.h as the basis
for these definitions.

Link: http://lkml.kernel.org/r/1501527386-10736-3-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: hugetlb: define system call hugetlb size encodings in single file

Patch series "Consolidate system call hugetlb page size encodings".

These patches are the result of discussions in
https://lkml.org/lkml/2017/3/8/548.  The following changes are made in the
patch set:

1) Put all the log2 encoded huge page size definitions in a common
   header file.  The idea is have a set of definitions that can be use as
   the basis for system call specific definitions such as MAP_HUGE_* and
   SHM_HUGE_*.

2) Remove MAP_HUGE_* definitions in arch specific files.  All these
   definitions are the same.  Consolidate all definitions in the primary
   user header file (uapi/linux/mman.h).

3) Remove SHM_HUGE_* definitions intended for user space from kernel
   header file, and add to user (uapi/linux/shm.h) header file.  Add
   definitions for all known huge page size encodings as in mmap.

This patch (of 3):

If hugetlb pages are requested in mmap or shmget system calls, a huge
page size other than default can be requested.  This is accomplished by
encoding the log2 of the huge page size in the upper bits of the flag
argument.  asm-generic and arch specific headers all define the same
values for these encodings.

Put common definitions in a single header file.  The primary uapi header
files for mmap and shm will use these definitions as a basis for
definitions specific to those system calls.

Link: http://lkml.kernel.org/r/1501527386-10736-2-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@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>

include/linux/fs.h: remove unneeded forward definition of mm_struct

Link: http://lkml.kernel.org/r/20170525102927.6163-1-jlayton@redhat.com
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/sync.c: remove unnecessary NULL f_mapping check in sync_file_range

fsync codepath assumes that f_mapping can never be NULL, but
sync_file_range has a check for that.

Remove the one from sync_file_range as I don't see how you'd ever get a
NULL pointer in here.

Link: http://lkml.kernel.org/r/20170525110509.9434-1-jlayton@redhat.com
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: selftest: enable testing of UFFDIO_ZEROPAGE for shmem

Link: http://lkml.kernel.org/r/1497939652-16528-8-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: report UFFDIO_ZEROPAGE as available for shmem VMAs

Now when shmem VMAs can be filled with zero page via userfaultfd we can
report that UFFDIO_ZEROPAGE is available for those VMAs

Link: http://lkml.kernel.org/r/1497939652-16528-7-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: shmem: wire up shmem_mfill_zeropage_pte

For shmem VMAs we can use shmem_mfill_zeropage_pte for UFFDIO_ZEROPAGE

Link: http://lkml.kernel.org/r/1497939652-16528-6-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: mcopy_atomic: introduce mfill_atomic_pte helper

Shuffle the code a bit to improve readability.

Link: http://lkml.kernel.org/r/1497939652-16528-5-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: shmem: add shmem_mfill_zeropage_pte for userfaultfd support

shmem_mfill_zeropage_pte is the low level routine that implements the
userfaultfd UFFDIO_ZEROPAGE command.  Since for shmem mappings zero
pages are always allocated and accounted, the new method is a slight
extension of the existing shmem_mcopy_atomic_pte.

Link: http://lkml.kernel.org/r/1497939652-16528-4-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: introduce shmem_inode_acct_block

The shmem_acct_block and the update of used_blocks are following one
another in all the places they are used.  Combine these two into a
helper function.

Link: http://lkml.kernel.org/r/1497939652-16528-3-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: shmem_charge: verify max_block is not exceeded before inode update

Patch series "userfaultfd: enable zeropage support for shmem".

These patches enable support for UFFDIO_ZEROPAGE for shared memory.

The first two patches are not strictly related to userfaultfd, they are
just minor refactoring to reduce amount of code duplication.

This patch (of 7):

Currently we update inode and shmem_inode_info before verifying that
used_blocks will not exceed max_blocks.  In case it will, we undo the
update.  Let's switch the order and move the verification of the blocks
count before the inode and shmem_inode_info update.

Link: http://lkml.kernel.org/r/1497939652-16528-2-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: add THP swapping out fallback counting

When swapping out THP (Transparent Huge Page), instead of swapping out
the THP as a whole, sometimes we have to fallback to split the THP into
normal pages before swapping, because no free swap clusters are
available, or cgroup limit is exceeded, etc.  To count the number of the
fallback, a new VM event THP_SWPOUT_FALLBACK is added, and counted when
we fallback to split the THP.

Link: http://lkml.kernel.org/r/20170724051840.2309-13-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: delay splitting THP after swapped out

In this patch, splitting transparent huge page (THP) during swapping out
is delayed from after adding the THP into the swap cache to after
swapping out finishes.  After the patch, more operations for the
anonymous THP reclaiming, such as writing the THP to the swap device,
removing the THP from the swap cache could be batched.  So that the
performance of anonymous THP swapping out could be improved.

This is the second step for the THP swap support.  The plan is to delay
splitting the THP step by step and avoid splitting the THP finally.

With the patchset, the swap out throughput improves 42% (from about
5.81GB/s to about 8.25GB/s) in the vm-scalability swap-w-seq test case
with 16 processes.  At the same time, the IPI (reflect TLB flushing)
reduced about 78.9%.  The test is done on a Xeon E5 v3 system.  The swap
device used is a RAM simulated PMEM (persistent memory) device.  To test
the sequential swapping out, the test case creates 8 processes, which
sequentially allocate and write to the anonymous pages until the RAM and
part of the swap device is used up.

Link: http://lkml.kernel.org/r/20170724051840.2309-12-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg, THP, swap: make mem_cgroup_swapout() support THP

This patch makes mem_cgroup_swapout() works for the transparent huge
page (THP).  Which will move the memory cgroup charge from memory to
swap for a THP.

This will be used for the THP swap support.  Where a THP may be swapped
out as a whole to a set of (HPAGE_PMD_NR) continuous swap slots on the
swap device.

Link: http://lkml.kernel.org/r/20170724051840.2309-11-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Shaohua Li <shli@kernel.org>
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg, THP, swap: avoid to duplicated charge THP in swap cache

For a THP (Transparent Huge Page), tail_page->mem_cgroup is NULL.  So to
check whether the page is charged already, we need to check the head
page.  This is not an issue before because it is impossible for a THP to
be in the swap cache before.  But after we add delaying splitting THP
after swapped out support, it is possible now.

Link: http://lkml.kernel.org/r/20170724051840.2309-10-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Shaohua Li <shli@kernel.org>
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg, THP, swap: support move mem cgroup charge for THP swapped out

PTE mapped THP (Transparent Huge Page) will be ignored when moving
memory cgroup charge.  But for THP which is in the swap cache, the
memory cgroup charge for the swap of a tail-page may be moved in current
implementation.  That isn't correct, because the swap charge for all
sub-pages of a THP should be moved together.  Following the processing
of the PTE mapped THP, the mem cgroup charge moving for the swap entry
for a tail-page of a THP is ignored too.

Link: http://lkml.kernel.org/r/20170724051840.2309-9-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Shaohua Li <shli@kernel.org>
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: support splitting THP for THP swap out

After adding swapping out support for THP (Transparent Huge Page), it is
possible that a THP in swap cache (partly swapped out) need to be split.
To split such a THP, the swap cluster backing the THP need to be split
too, that is, the CLUSTER_FLAG_HUGE flag need to be cleared for the swap
cluster.  The patch implemented this.

And because the THP swap writing needs the THP keeps as huge page during
writing.  The PageWriteback flag is checked before splitting.

Link: http://lkml.kernel.org/r/20170724051840.2309-8-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: test code to write THP to swap device as a whole

To support delay splitting THP (Transparent Huge Page) after swapped
out, we need to enhance swap writing code to support to write a THP as a
whole.  This will improve swap write IO performance.

As Ming Lei <ming.lei@redhat.com> pointed out, this should be based on
multipage bvec support, which hasn't been merged yet.  So this patch is
only for testing the functionality of the other patches in the series.
And will be reimplemented after multipage bvec support is merged.

Link: http://lkml.kernel.org/r/20170724051840.2309-7-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Shaohua Li <shli@kernel.org>
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

block, THP: make block_device_operations.rw_page support THP

The .rw_page in struct block_device_operations is used by the swap
subsystem to read/write the page contents from/into the corresponding
swap slot in the swap device.  To support the THP (Transparent Huge
Page) swap optimization, the .rw_page is enhanced to support to
read/write THP if possible.

Link: http://lkml.kernel.org/r/20170724051840.2309-6-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: don't allocate huge cluster for file backed swap device

It's hard to write a whole transparent huge page (THP) to a file backed
swap device during swapping out and the file backed swap device isn't
very popular.  So the huge cluster allocation for the file backed swap
device is disabled.

Link: http://lkml.kernel.org/r/20170724051840.2309-5-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: make reuse_swap_page() works for THP swapped out

After supporting to delay THP (Transparent Huge Page) splitting after
swapped out, it is possible that some page table mappings of the THP are
turned into swap entries.  So reuse_swap_page() need to check the swap
count in addition to the map count as before.  This patch done that.

In the huge PMD write protect fault handler, in addition to the page map
count, the swap count need to be checked too, so the page lock need to
be acquired too when calling reuse_swap_page() in addition to the page
table lock.

[ying.huang@intel.com: silence a compiler warning]
Link: http://lkml.kernel.org/r/87bmnzizjy.fsf@yhuang-dev.intel.com
Link: http://lkml.kernel.org/r/20170724051840.2309-4-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: support to reclaim swap space for THP swapped out

The normal swap slot reclaiming can be done when the swap count reaches
SWAP_HAS_CACHE.  But for the swap slot which is backing a THP, all swap
slots backing one THP must be reclaimed together, because the swap slot
may be used again when the THP is swapped out again later.  So the swap
slots backing one THP can be reclaimed together when the swap count for
all swap slots for the THP reached SWAP_HAS_CACHE.  In the patch, the
functions to check whether the swap count for all swap slots backing one
THP reached SWAP_HAS_CACHE are implemented and used when checking
whether a swap slot can be reclaimed.

To make it easier to determine whether a swap slot is backing a THP, a
new swap cluster flag named CLUSTER_FLAG_HUGE is added to mark a swap
cluster which is backing a THP (Transparent Huge Page).  Because THP
swap in as a whole isn't supported now.  After deleting the THP from the
swap cache (for example, swapping out finished), the CLUSTER_FLAG_HUGE
flag will be cleared.  So that, the normal pages inside THP can be
swapped in individually.

[ying.huang@intel.com: fix swap_page_trans_huge_swapped on HDD]
Link: http://lkml.kernel.org/r/874ltsm0bi.fsf@yhuang-dev.intel.com
Link: http://lkml.kernel.org/r/20170724051840.2309-3-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, THP, swap: support to clear swap cache flag for THP swapped out

Patch series "mm, THP, swap: Delay splitting THP after swapped out", v3.

This is the second step of THP (Transparent Huge Page) swap
optimization.  In the first step, the splitting huge page is delayed
from almost the first step of swapping out to after allocating the swap
space for the THP and adding the THP into the swap cache.  In the second
step, the splitting is delayed further to after the swapping out
finished.  The plan is to delay splitting THP step by step, finally
avoid splitting THP for the THP swapping out and swap out/in the THP as
a whole.

In the patchset, more operations for the anonymous THP reclaiming, such
as TLB flushing, writing the THP to the swap device, removing the THP
from the swap cache are batched.  So that the performance of anonymous
THP swapping out are improved.

During the development, the following scenarios/code paths have been
checked,

 - swap out/in
 - swap off
 - write protect page fault
 - madvise_free
 - process exit
 - split huge page

With the patchset, the swap out throughput improves 42% (from about
5.81GB/s to about 8.25GB/s) in the vm-scalability swap-w-seq test case
with 16 processes.  At the same time, the IPI (reflect TLB flushing)
reduced about 78.9%.  The test is done on a Xeon E5 v3 system.  The swap
device used is a RAM simulated PMEM (persistent memory) device.  To test
the sequential swapping out, the test case creates 8 processes, which
sequentially allocate and write to the anonymous pages until the RAM and
part of the swap device is used up.

Below is the part of the cover letter for the first step patchset of THP
swap optimization which applies to all steps.

=========================

Recently, the performance of the storage devices improved so fast that
we cannot saturate the disk bandwidth with single logical CPU when do
page swap out even on a high-end server machine.  Because the
performance of the storage device improved faster than that of single
logical CPU.  And it seems that the trend will not change in the near
future.  On the other hand, the THP becomes more and more popular
because of increased memory size.  So it becomes necessary to optimize
THP swap performance.

The advantages of the THP swap support include:

 - Batch the swap operations for the THP to reduce TLB flushing and lock
   acquiring/releasing, including allocating/freeing the swap space,
   adding/deleting to/from the swap cache, and writing/reading the swap
   space, etc. This will help improve the performance of the THP swap.

 - The THP swap space read/write will be 2M sequential IO. It is
   particularly helpful for the swap read, which are usually 4k random
   IO. This will improve the performance of the THP swap too.

 - It will help the memory fragmentation, especially when the THP is
   heavily used by the applications. The 2M continuous pages will be
   free up after THP swapping out.

 - It will improve the THP utilization on the system with the swap
   turned on. Because the speed for khugepaged to collapse the normal
   pages into the THP is quite slow. After the THP is split during the
   swapping out, it will take quite long time for the normal pages to
   collapse back into the THP after being swapped in. The high THP
   utilization helps the efficiency of the page based memory management
   too.

There are some concerns regarding THP swap in, mainly because possible
enlarged read/write IO size (for swap in/out) may put more overhead on
the storage device.  To deal with that, the THP swap in should be turned
on only when necessary.

For example, it can be selected via "always/never/madvise" logic, to be
turned on globally, turned off globally, or turned on only for VMA with
MADV_HUGEPAGE, etc.

This patch (of 12):

Previously, swapcache_free_cluster() is used only in the error path of
shrink_page_list() to free the swap cluster just allocated if the THP
(Transparent Huge Page) is failed to be split.  In this patch, it is
enhanced to clear the swap cache flag (SWAP_HAS_CACHE) for the swap
cluster that holds the contents of THP swapped out.

This will be used in delaying splitting THP after swapping out support.
Because there is no THP swapping in as a whole support yet, after
clearing the swap cache flag, the swap cluster backing the THP swapped
out will be split.  So that the swap slots in the swap cluster can be
swapped in as normal pages later.

Link: http://lkml.kernel.org/r/20170724051840.2309-2-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Ross Zwisler <ross.zwisler@intel.com> [for brd.c, zram_drv.c, pmem.c]
Cc: Vishal L Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: use int for event/state parameter in several functions

Several functions use an enum type as parameter for an event/state, but
are called in some locations with an argument of a different enum type.
Adjust the interface of these functions to reality by changing the
parameter to int.

This fixes a ton of enum-conversion warnings that are generated when
building the kernel with clang.

[mka@chromium.org: also change parameter type of inc/dec/mod_memcg_page_state()]
Link: http://lkml.kernel.org/r/20170728213442.93823-1-mka@chromium.org
Link: http://lkml.kernel.org/r/20170727211004.34435-1-mka@chromium.org
Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Doug Anderson <dianders@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c: constify attribute_group structures

attribute_group are not supposed to change at runtime.  All functions
working with attribute_group provided by <linux/sysfs.h> work with const
attribute_group.  So mark the non-const structs as const.

Link: http://lkml.kernel.org/r/1501157260-3922-1-git-send-email-arvind.yadav.cs@gmail.com
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/huge_memory.c: constify attribute_group structures

attribute_group are not supposed to change at runtime.  All functions
working with attribute_group provided by <linux/sysfs.h> work with const
attribute_group.  So mark the non-const structs as const.

Link: http://lkml.kernel.org/r/1501157240-3876-1-git-send-email-arvind.yadav.cs@gmail.com
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_idle.c: constify attribute_group structures

attribute_group are not supposed to change at runtime.  All functions
working with attribute_group provided by <linux/sysfs.h> work with const
attribute_group.  So mark the non-const structs as const.

Link: http://lkml.kernel.org/r/1501157221-3832-1-git-send-email-arvind.yadav.cs@gmail.com
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/slub.c: constify attribute_group structures

attribute_group are not supposed to change at runtime.  All functions
working with attribute_group provided by <linux/sysfs.h> work with const
attribute_group.  So mark the non-const structs as const.

Link: http://lkml.kernel.org/r/1501157186-3749-1-git-send-email-arvind.yadav.cs@gmail.com
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/ksm.c: constify attribute_group structures

attribute_group are not supposed to change at runtime.  All functions
working with attribute_group provided by <linux/sysfs.h> work with const
attribute_group.  So mark the non-const structs as const.

Link: http://lkml.kernel.org/r/1501157167-3706-2-git-send-email-arvind.yadav.cs@gmail.com
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup: revert fa06235b8eb0 ("cgroup: reset css on destruction")

Commit fa06235b8eb0 ("cgroup: reset css on destruction") caused
css_reset callback to be called from the offlining path.  Although it
solves the problem mentioned in the commit description ("For instance,
memory cgroup needs to reset memory.low, otherwise pages charged to a
dead cgroup might never get reclaimed."), generally speaking, it's not
correct.

An offline cgroup can still be a resource domain, and we shouldn't grant
it more resources than it had before deletion.

For instance, if an offline memory cgroup has dirty pages, we should
still imply i/o limits during writeback.

The css_reset callback is designed to return the cgroup state into the
original state, that means reset all limits and counters.  It's
spomething different from the offlining, and we shouldn't use it from
the offlining path.  Instead, we should adjust necessary settings from
the per-controller css_offline callbacks (e.g.  reset memory.low).

Link: http://lkml.kernel.org/r/20170727130428.28856-2-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: reset memory.low during memcg offlining

A removed memory cgroup with a defined memory.low and some belonging
pagecache has very low chances to be freed.

If a cgroup has been removed, there is likely no memory pressure inside
the cgroup, and the pagecache is protected from the external pressure by
the defined low limit.  The cgroup will be freed only after the reclaim
of all belonging pages.  And it will not happen until there are any
reclaimable memory in the system.  That means, there is a good chance,
that a cold pagecache will reside in the memory for an undefined amount
of time, wasting system resources.

This problem was fixed earlier by fa06235b8eb0 ("cgroup: reset css on
destruction"), but it's not a best way to do it, as we can't really
reset all limits/counters during cgroup offlining.

Link: http://lkml.kernel.org/r/20170727130428.28856-1-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.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>

mm: remove nr_pages argument from pagevec_lookup{,_range}()

All users of pagevec_lookup() and pagevec_lookup_range() now pass
PAGEVEC_SIZE as a desired number of pages.

Just drop the argument.

Link: http://lkml.kernel.org/r/20170726114704.7626-11-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use find_get_pages_range() in filemap_range_has_page()

We want only pages from given range in filemap_range_has_page(),
furthermore we want at most a single page.

So use find_get_pages_range() instead of pagevec_lookup() and remove
unnecessary code.

Link: http://lkml.kernel.org/r/20170726114704.7626-10-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: use pagevec_lookup_range() in page_cache_seek_hole_data()

We want only pages from given range in page_cache_seek_hole_data().  Use
pagevec_lookup_range() instead of pagevec_lookup() and remove
unnecessary code.

Note that the check for getting less pages than desired can be removed
because index gets updated by pagevec_lookup_range().

Link: http://lkml.kernel.org/r/20170726114704.7626-9-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: use pagevec_lookup_range() in remove_inode_hugepages()

We want only pages from given range in remove_inode_hugepages().  Use
pagevec_lookup_range() instead of pagevec_lookup().

Link: http://lkml.kernel.org/r/20170726114704.7626-8-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Nadia Yvette Chambers <nyc@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext4: use pagevec_lookup_range() in writeback code

Both occurences of pagevec_lookup() actually want only pages from a
given range.  Use pagevec_lookup_range() for the lookup.

Link: http://lkml.kernel.org/r/20170726114704.7626-7-jack@suse.cz
Signed-off-by: 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>

ext4: use pagevec_lookup_range() in ext4_find_unwritten_pgoff()

Use pagevec_lookup_range() in ext4_find_unwritten_pgoff() since we are
interested only in pages in the given range.  Simplify the logic as a
result of not getting pages out of range and index getting automatically
advanced.

Link: http://lkml.kernel.org/r/20170726114704.7626-6-jack@suse.cz
Signed-off-by: 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>

fs: fix performance regression in clean_bdev_aliases()

Commit e64855c6cfaa ("fs: Add helper to clean bdev aliases under a bh
and use it") added a wrapper for clean_bdev_aliases() that invalidates
bdev aliases underlying a single buffer head.

However this has caused a performance regression for bonnie++ benchmark
on ext4 filesystem when delayed allocation is turned off (ext3 mode) -
average of 3 runs:

  Hmean SeqOut Char  164787.55 (  0.00%) 107189.06 (-34.95%)
  Hmean SeqOut Block 219883.89 (  0.00%) 168870.32 (-23.20%)

The reason for this regression is that clean_bdev_aliases() is slower
when called for a single block because pagevec_lookup() it uses will end
up iterating through the radix tree until it finds a page (which may
take a while) but we are only interested whether there's a page at a
particular index.

Fix the problem by using pagevec_lookup_range() instead which avoids the
needless iteration.

Fixes: e64855c6cfaa ("fs: Add helper to clean bdev aliases under a bh and use it")
Link: http://lkml.kernel.org/r/20170726114704.7626-5-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: implement find_get_pages_range()

Implement a variant of find_get_pages() that stops iterating at given
index.  This may be substantial performance gain if the mapping is
sparse.  See following commit for details.  Furthermore lots of users of
this function (through pagevec_lookup()) actually want a range lookup
and all of them are currently open-coding this.

Also create corresponding pagevec_lookup_range() function.

Link: http://lkml.kernel.org/r/20170726114704.7626-4-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make pagevec_lookup() update index

Make pagevec_lookup() (and underlying find_get_pages()) update index to
the next page where iteration should continue.  Most callers want this
and also pagevec_lookup_tag() already does this.

Link: http://lkml.kernel.org/r/20170726114704.7626-3-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fscache: remove unused ->now_uncached callback

Patch series "Ranged pagevec lookup", v2.

In this series I make pagevec_lookup() update the index (to be
consistent with pagevec_lookup_tag() and also as a preparation for
ranged lookups), provide ranged variant of pagevec_lookup() and use it
in places where it makes sense.  This not only removes some common code
but is also a measurable performance win for some use cases (see patch
4/10) where radix tree is sparse and searching & grabing of a page after
the end of the range has measurable overhead.

This patch (of 10):

The callback doesn't ever get called.  Remove it.

Link: http://lkml.kernel.org/r/20170726114704.7626-2-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, vmscan: do not loop on too_many_isolated for ever

Tetsuo Handa has reported[1][2][3] that direct reclaimers might get
stuck in too_many_isolated loop basically for ever because the last few
pages on the LRU lists are isolated by the kswapd which is stuck on fs
locks when doing the pageout or slab reclaim.  This in turn means that
there is nobody to actually trigger the oom killer and the system is
basically unusable.

too_many_isolated has been introduced by commit 35cd78156c49 ("vmscan:
throttle direct reclaim when too many pages are isolated already") to
prevent from pre-mature oom killer invocations because back then no
reclaim progress could indeed trigger the OOM killer too early.

But since the oom detection rework in commit 0a0337e0d1d1 ("mm, oom:
rework oom detection") the allocation/reclaim retry loop considers all
the reclaimable pages and throttles the allocation at that layer so we
can loosen the direct reclaim throttling.

Make shrink_inactive_list loop over too_many_isolated bounded and
returns immediately when the situation hasn't resolved after the first
sleep.

Replace congestion_wait by a simple schedule_timeout_interruptible
because we are not really waiting on the IO congestion in this path.

Please note that this patch can theoretically cause the OOM killer to
trigger earlier while there are many pages isolated for the reclaim
which makes progress only very slowly.  This would be obvious from the
oom report as the number of isolated pages are printed there.  If we
ever hit this should_reclaim_retry should consider those numbers in the
evaluation in one way or another.

[1] http://lkml.kernel.org/r/201602092349.ACG81273.OSVtMJQHLOFOFF@I-love.SAKURA.ne.jp
[2] http://lkml.kernel.org/r/201702212335.DJB30777.JOFMHSFtVLQOOF@I-love.SAKURA.ne.jp
[3] http://lkml.kernel.org/r/201706300914.CEH95859.FMQOLVFHJFtOOS@I-love.SAKURA.ne.jp

[mhocko@suse.com: switch to uninterruptible sleep]
Link: http://lkml.kernel.org/r/20170724065048.GB25221@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/20170710074842.23175-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Tested-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Rik van Riel <riel@redhat.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>

zsmalloc: zs_page_migrate: skip unnecessary loops but not return -EBUSY if zspage is not inuse

Getting -EBUSY from zs_page_migrate will make migration slow (retry) or
fail (zs_page_putback will schedule_work free_work, but it cannot ensure
the success).

I noticed this issue because my Kernel patched
(https://lkml.org/lkml/2014/5/28/113) that will remove retry in
__alloc_contig_migrate_range.

This retry will handle the -EBUSY because it will re-isolate the page
and re-call migrate_pages.  Without it will make cma_alloc fail at once
with -EBUSY.

According to the review from Minchan Kim in
https://lkml.org/lkml/2014/5/28/113, I update the patch to skip
unnecessary loops but not return -EBUSY if zspage is not inuse.

Following is what I got with highalloc-performance in a vbox with 2 cpu
1G memory 512 zram as swap.  And the swappiness is set to 100.

                                   ori          ne
                                  orig         new
Minor Faults                  50805113    50830235
Major Faults                     43918       56530
Swap Ins                         42087       55680
Swap Outs                        89718      104700
Allocation stalls                    0           0
DMA allocs                       57787       52364
DMA32 allocs                  47964599    48043563
Normal allocs                        0           0
Movable allocs                       0           0
Direct pages scanned             45493       23167
Kswapd pages scanned           1565222     1725078
Kswapd pages reclaimed         1342222     1503037
Direct pages reclaimed           45615       25186
Kswapd efficiency                  85%         87%
Kswapd velocity               1897.101    1949.042
Direct efficiency                 100%        108%
Direct velocity                 55.139      26.175
Percentage direct scans             2%          1%
Zone normal velocity          1952.240    1975.217
Zone dma32 velocity              0.000       0.000
Zone dma velocity                0.000       0.000
Page writes by reclaim       89764.000  105233.000
Page writes file                    46         533
Page writes anon                 89718      104700
Page reclaim immediate           21457        3699
Sector Reads                   3259688     3441368
Sector Writes                  3667252     3754836
Page rescued immediate               0           0
Slabs scanned                  1042872     1160855
Direct inode steals               8042       10089
Kswapd inode steals              54295       29170
Kswapd skipped wait                  0           0
THP fault alloc                    175         154
THP collapse alloc                 226         289
THP splits                           0           0
THP fault fallback                  11          14
THP collapse fail                    3           2
Compaction stalls                  536         646
Compaction success                 322         358
Compaction failures                214         288
Page migrate success            119608      111063
Page migrate failure              2723        2593
Compaction pages isolated       250179      232652
Compaction migrate scanned     9131832     9942306
Compaction free scanned        2093272     2613998
Compaction cost                    192         189
NUMA alloc hit                47124555    47193990
NUMA alloc miss                      0           0
NUMA interleave hit                  0           0
NUMA alloc local              47124555    47193990
NUMA base PTE updates                0           0
NUMA huge PMD updates                0           0
NUMA page range updates              0           0
NUMA hint faults                     0           0
NUMA hint local faults               0           0
NUMA hint local percent            100         100
NUMA pages migrated                  0           0
AutoNUMA cost                       0%          0%

[akpm@linux-foundation.org: remove newline, per Minchan]
Link: http://lkml.kernel.org/r/1500889535-19648-1-git-send-email-zhuhui@xiaomi.com
Signed-off-by: Hui Zhu <zhuhui@xiaomi.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: always flush VMA ranges affected by zap_page_range

Nadav Amit report zap_page_range only specifies that the caller protect
the VMA list but does not specify whether it is held for read or write
with callers using either.  madvise holds mmap_sem for read meaning that
a parallel zap operation can unmap PTEs which are then potentially
skipped by madvise which potentially returns with stale TLB entries
present.  While the API could be extended, it would be a difficult API
to use.  This patch causes zap_page_range() to always consider flushing
the full affected range.  For small ranges or sparsely populated
mappings, this may result in one additional spurious TLB flush.  For
larger ranges, it is possible that the TLB has already been flushed and
the overhead is negligible.  Either way, this approach is safer overall
and avoids stale entries being present when madvise returns.

This can be illustrated with the following program provided by Nadav
Amit and slightly modified.  With the patch applied, it has an exit code
of 0 indicating a stale TLB entry did not leak to userspace.

---8<---

volatile int sync_step = 0;
volatile char *p;

static inline unsigned long rdtsc()
{
unsigned long hi, lo;
__asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
 return lo | (hi << 32);
}

static inline void wait_rdtsc(unsigned long cycles)
{
unsigned long tsc = rdtsc();

while (rdtsc() - tsc < cycles);
}

void *big_madvise_thread(void *ign)
{
sync_step = 1;
while (sync_step != 2);
madvise((void*)p, PAGE_SIZE * N_PAGES, MADV_DONTNEED);
}

int main(void)
{
pthread_t aux_thread;

p = mmap(0, PAGE_SIZE * N_PAGES, PROT_READ|PROT_WRITE,
 MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);

memset((void*)p, 8, PAGE_SIZE * N_PAGES);

pthread_create(&aux_thread, NULL, big_madvise_thread, NULL);
while (sync_step != 1);

*p = 8; // Cache in TLB
sync_step = 2;
wait_rdtsc(100000);
madvise((void*)p, PAGE_SIZE, MADV_DONTNEED);
printf("data: %d (%s)\n", *p, (*p == 8 ? "stale, broken" : "cleared, fine"));
return *p == 8 ? -1 : 0;
}
---8<---

Link: http://lkml.kernel.org/r/20170725101230.5v7gvnjmcnkzzql3@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Nadav Amit <nadav.amit@gmail.com>
Cc: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c: make huge_pte_offset() consistent and document behaviour

When walking the page tables to resolve an address that points to
!p*d_present() entry, huge_pte_offset() returns inconsistent values
depending on the level of page table (PUD or PMD).

It returns NULL in the case of a PUD entry while in the case of a PMD
entry, it returns a pointer to the page table entry.

A similar inconsitency exists when handling swap entries - returns NULL
for a PUD entry while a pointer to the pte_t is retured for the PMD
entry.

Update huge_pte_offset() to make the behaviour consistent - return a
pointer to the pte_t for hugepage or swap entries.  Only return NULL in
instances where we have a p*d_none() entry and the size parameter
doesn't match the hugepage size at this level of the page table.

Document the behaviour to clarify the expected behaviour of this
function.  This is to set clear semantics for architecture specific
implementations of huge_pte_offset().

Discussions on the arm64 implementation of huge_pte_offset()
(http://www.spinics.net/lists/linux-mm/msg133699.html) showed that there
is benefit from returning a pte_t* in the case of p*d_none().

The fault handling code in hugetlb_fault() can handle p*d_none() entries
and saves an extra round trip to huge_pte_alloc().  Other callers of
huge_pte_offset() should be ok as well.

[punit.agrawal@arm.com: v2]
Link: http://lkml.kernel.org/r/20170725154114.24131-2-punit.agrawal@arm.com
Signed-off-by: Punit Agrawal <punit.agrawal@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: make __gup_device_* require THP

These functions are the only bits of generic code that use
{pud,pmd}_pfn() without checking for CONFIG_TRANSPARENT_HUGEPAGE.  This
works fine on x86, the only arch with devmap support, since the *_pfn()
functions are always defined there, but this isn't true for every
architecture.

Link: http://lkml.kernel.org/r/20170626063833.11094-1-oohall@gmail.com
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mremap: fail map duplication attempts for private mappings

mremap will attempt to create a 'duplicate' mapping if old_size == 0 is
specified.  In the case of private mappings, mremap will actually create
a fresh separate private mapping unrelated to the original.  This does
not fit with the design semantics of mremap as the intention is to
create a new mapping based on the original.

Therefore, return EINVAL in the case where an attempt is made to
duplicate a private mapping.  Also, print a warning message (once) if
such an attempt is made.

Link: http://lkml.kernel.org/r/cb9d9f6a-7095-582f-15a5-62643d65c736@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Aaron Lu <aaron.lu@intel.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_owner: don't grab zone->lock for init_pages_in_zone()

init_pages_in_zone() is run under zone->lock, which means a long lock
time and disabled interrupts on large machines.  This is currently not
an issue since it runs early in boot, but a later patch will change
that.

However, like other pfn scanners, we don't actually need zone->lock even
when other cpus are running.  The only potentially dangerous operation
here is reading bogus buddy page owner due to race, and we already know
how to handle that.  The worst that can happen is that we skip some
early allocated pages, which should not affect the debugging power of
page_owner noticeably.

Link: http://lkml.kernel.org/r/20170720134029.25268-4-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_ext: periodically reschedule during page_ext_init()

page_ext_init() can take long on large machines, so add a cond_resched()
point after each section is processed.  This will allow moving the init
to a later point at boot without triggering lockup reports.

Link: http://lkml.kernel.org/r/20170720134029.25268-3-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_owner: make init_pages_in_zone() faster

In init_pages_in_zone() we currently use the generic set_page_owner()
function to initialize page_owner info for early allocated pages.  This
means we needlessly do lookup_page_ext() twice for each page, and more
importantly save_stack(), which has to unwind the stack and find the
corresponding stack depot handle.  Because the stack is always the same
for the initialization, unwind it once in init_pages_in_zone() and reuse
the handle.  Also avoid the repeated lookup_page_ext().

This can significantly reduce boot times with page_owner=on on large
machines, especially for kernels built without frame pointer, where the
stack unwinding is noticeably slower.

[vbabka@suse.cz: don't duplicate code of __set_page_owner(), per Michal Hocko]
[akpm@linux-foundation.org: coding-style fixes]
[vbabka@suse.cz: create statically allocated fake stack trace for early allocated pages, per Michal]
Link: http://lkml.kernel.org/r/45813564-2342-fc8d-d31a-f4b68a724325@suse.cz
Link: http://lkml.kernel.org/r/20170720134029.25268-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, sparse, page_ext: drop ugly N_HIGH_MEMORY branches for allocations

Commit f52407ce2dea ("memory hotplug: alloc page from other node in
memory online") has introduced N_HIGH_MEMORY checks to only use NUMA
aware allocations when there is some memory present because the
respective node might not have any memory yet at the time and so it
could fail or even OOM.

Things have changed since then though.  Zonelists are now always
initialized before we do any allocations even for hotplug (see
959ecc48fc75 ("mm/memory_hotplug.c: fix building of node hotplug
zonelist")).

Therefore these checks are not really needed.  In fact caller of the
allocator should never care about whether the node is populated because
that might change at any time.

Link: http://lkml.kernel.org/r/20170721143915.14161-10-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: get rid of zonelists_mutex

zonelists_mutex was introduced by commit 4eaf3f64397c ("mem-hotplug: fix
potential race while building zonelist for new populated zone") to
protect zonelist building from races.  This is no longer needed though
because both memory online and offline are fully serialized.  New users
have grown since then.

Notably setup_per_zone_wmarks wants to prevent from races between memory
hotplug, khugepaged setup and manual min_free_kbytes update via sysctl
(see cfd3da1e49bb ("mm: Serialize access to min_free_kbytes").  Let's
add a private lock for that purpose.  This will not prevent from seeing
halfway through memory hotplug operation but that shouldn't be a big
deal becuse memory hotplug will update watermarks explicitly so we will
eventually get a full picture.  The lock just makes sure we won't race
when updating watermarks leading to weird results.

Also __build_all_zonelists manipulates global data so add a private lock
for it as well.  This doesn't seem to be necessary today but it is more
robust to have a lock there.

While we are at it make sure we document that memory online/offline
depends on a full serialization either via mem_hotplug_begin() or
device_lock.

Link: http://lkml.kernel.org/r/20170721143915.14161-9-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Haicheng Li <haicheng.li@linux.intel.com>
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>

mm, page_alloc: remove stop_machine from build_all_zonelists

build_all_zonelists has been (ab)using stop_machine to make sure that
zonelists do not change while somebody is looking at them.  This is is
just a gross hack because a) it complicates the context from which we
can call build_all_zonelists (see 3f906ba23689 ("mm/memory-hotplug:
switch locking to a percpu rwsem")) and b) is is not really necessary
especially after "mm, page_alloc: simplify zonelist initialization" and
c) it doesn't really provide the protection it claims (see below).

Updates of the zonelists happen very seldom, basically only when a zone
becomes populated during memory online or when it loses all the memory
during offline.  A racing iteration over zonelists could either miss a
zone or try to work on one zone twice.  Both of these are something we
can live with occasionally because there will always be at least one
zone visible so we are not likely to fail allocation too easily for
example.

Please note that the original stop_machine approach doesn't really
provide a better exclusion because the iteration might be interrupted
half way (unless the whole iteration is preempt disabled which is not
the case in most cases) so the some zones could still be seen twice or a
zone missed.

I have run the pathological online/offline of the single memblock in the
movable zone while stressing the same small node with some memory
pressure.

Node 1, zone      DMA
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 943, 943, 943)
Node 1, zone    DMA32
  pages free     227310
        min      8294
        low      10367
        high     12440
        spanned  262112
        present  262112
        managed  241436
        protection: (0, 0, 0, 0)
Node 1, zone   Normal
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 0, 0, 1024)
Node 1, zone  Movable
  pages free     32722
        min      85
        low      117
        high     149
        spanned  32768
        present  32768
        managed  32768
        protection: (0, 0, 0, 0)

root@test1:/sys/devices/system/node/node1# while true
do
echo offline > memory34/state
echo online_movable > memory34/state
done

root@test1:/mnt/data/test/linux-3.7-rc5# numactl --preferred=1 make -j4

and it survived without any unexpected behavior.  While this is not
really a great testing coverage it should exercise the allocation path
quite a lot.

Link: http://lkml.kernel.org/r/20170721143915.14161-8-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: simplify zonelist initialization

build_zonelists gradually builds zonelists from the nearest to the most
distant node.  As we do not know how many populated zones we will have
in each node we rely on the _zoneref to terminate initialized part of
the zonelist by a NULL zone.  While this is functionally correct it is
quite suboptimal because we cannot allow updaters to race with zonelists
users because they could see an empty zonelist and fail the allocation
or hit the OOM killer in the worst case.

We can do much better, though.  We can store the node ordering into an
already existing node_order array and then give this array to
build_zonelists_in_node_order and do the whole initialization at once.
zonelists consumers still might see halfway initialized state but that
should be much more tolerateable because the list will not be empty and
they would either see some zone twice or skip over some zone(s) in the
worst case which shouldn't lead to immediate failures.

While at it let's simplify build_zonelists_node which is rather
confusing now.  It gets an index into the zoneref array and returns the
updated index for the next iteration.  Let's rename the function to
build_zonerefs_node to better reflect its purpose and give it zoneref
array to update.  The function doesn't the index anymore.  It just
returns the number of added zones so that the caller can advance the
zonered array start for the next update.

This patch alone doesn't introduce any functional change yet, though, it
is merely a preparatory work for later changes.

Link: http://lkml.kernel.org/r/20170721143915.14161-7-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: remove explicit build_all_zonelists from try_online_node

try_online_node calls hotadd_new_pgdat which already calls
build_all_zonelists.  So the additional call is redundant.  Even though
hotadd_new_pgdat will only initialize zonelists of the new node this is
the right thing to do because such a node doesn't have any memory so
other zonelists would ignore all the zones from this node anyway.

Link: http://lkml.kernel.org/r/20170721143915.14161-6-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: drop zone from build_all_zonelists

build_all_zonelists gets a zone parameter to initialize zone's pagesets.
There is only a single user which gives a non-NULL zone parameter and
that one doesn't really need the rest of the build_all_zonelists (see
commit 6dcd73d7011b ("memory-hotplug: allocate zone's pcp before
onlining pages")).

Therefore remove setup_zone_pageset from build_all_zonelists and call it
from its only user directly.  This will also remove a pointless zonlists
rebuilding which is always good.

Link: http://lkml.kernel.org/r/20170721143915.14161-5-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: do not set_cpu_numa_mem on empty nodes initialization

__build_all_zonelists reinitializes each online cpu local node for
CONFIG_HAVE_MEMORYLESS_NODES.  This makes sense because previously
memory less nodes could gain some memory during memory hotplug and so
the local node should be changed for CPUs close to such a node.  It
makes less sense to do that unconditionally for a newly creaded NUMA
node which is still offline and without any memory.

Let's also simplify the cpu loop and use for_each_online_cpu instead of
an explicit cpu_online check for all possible cpus.

Link: http://lkml.kernel.org/r/20170721143915.14161-4-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: remove boot pageset initialization from memory hotplug

boot_pageset is a boot time hack which gets superseded by normal
pagesets later in the boot process.  It makes zero sense to reinitialize
it again and again during memory hotplug.

Link: http://lkml.kernel.org/r/20170721143915.14161-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: rip out ZONELIST_ORDER_ZONE

Patch series "cleanup zonelists initialization", v1.

This is aimed at cleaning up the zonelists initialization code we have
but the primary motivation was bug report [2] which got resolved but the
usage of stop_machine is just too ugly to live.  Most patches are
straightforward but 3 of them need a special consideration.

Patch 1 removes zone ordered zonelists completely.  I am CCing linux-api
because this is a user visible change.  As I argue in the patch
description I do not think we have a strong usecase for it these days.
I have kept sysctl in place and warn into the log if somebody tries to
configure zone lists ordering.  If somebody has a real usecase for it we
can revert this patch but I do not expect anybody will actually notice
runtime differences.  This patch is not strictly needed for the rest but
it made patch 6 easier to implement.

Patch 7 removes stop_machine from build_all_zonelists without adding any
special synchronization between iterators and updater which I _believe_
is acceptable as explained in the changelog.  I hope I am not missing
anything.

Patch 8 then removes zonelists_mutex which is kind of ugly as well and
not really needed AFAICS but a care should be taken when double checking
my thinking.

This patch (of 9):

Supporting zone ordered zonelists costs us just a lot of code while the
usefulness is arguable if existent at all.  Mel has already made node
ordering default on 64b systems.  32b systems are still using
ZONELIST_ORDER_ZONE because it is considered better to fallback to a
different NUMA node rather than consume precious lowmem zones.

This argument is, however, weaken by the fact that the memory reclaim
has been reworked to be node rather than zone oriented.  This means that
lowmem requests have to skip over all highmem pages on LRUs already and
so zone ordering doesn't save the reclaim time much.  So the only
advantage of the zone ordering is under a light memory pressure when
highmem requests do not ever hit into lowmem zones and the lowmem
pressure doesn't need to reclaim.

Considering that 32b NUMA systems are rather suboptimal already and it
is generally advisable to use 64b kernel on such a HW I believe we
should rather care about the code maintainability and just get rid of
ZONELIST_ORDER_ZONE altogether.  Keep systcl in place and warn if
somebody tries to set zone ordering either from kernel command line or
the sysctl.

[mhocko@suse.com: reading vm.numa_zonelist_order will never terminate]
Link: http://lkml.kernel.org/r/20170721143915.14161-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: add config and doc file for writeback feature

This patch adds document and kconfig for using of writeback feature.

Link: http://lkml.kernel.org/r/1498459987-24562-10-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: read page from backing device

This patch enables read IO from backing device.  For the feature, it
implements two IO read functions to transfer data from backing storage.

One is asynchronous IO function and other is synchronous one.

A reason I need synchrnous IO is due to partial write which need to
complete read IO before the overwriting partial data.

We can make the partial IO's case asynchronous, too but at the moment, I
don't feel adding more complexity to support such rare use cases so want
to go with simple.

[xieyisheng1@huawei.com: read_from_bdev_async(): return 1 to avoid call page_endio() in zram_rw_page()]
Link: http://lkml.kernel.org/r/1502707447-6944-1-git-send-email-xieyisheng1@huawei.com
Link: http://lkml.kernel.org/r/1498459987-24562-9-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: write incompressible pages to backing device

This patch enables write IO to transfer data to backing device.  For
that, it implements write_to_bdev function which creates new bio and
chaining with parent bio to make the parent bio asynchrnous.

For rw_page which don't have parent bio, it submit owned bio and handle
IO completion by zram_page_end_io.

Also, this patch defines new flag ZRAM_WB to mark written page for later
read IO.

[xieyisheng1@huawei.com: fix typo in comment]
Link: http://lkml.kernel.org/r/1502707447-6944-2-git-send-email-xieyisheng1@huawei.com
Link: http://lkml.kernel.org/r/1498459987-24562-8-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: identify asynchronous IO's return value

For upcoming asynchronous IO like writeback, zram_rw_page should be
aware of that whether requested IO was completed or submitted
successfully, otherwise error.

For the goal, zram_bvec_rw has three return values.

-errno: returns error number
     0: IO request is done synchronously
     1: IO request is issued successfully.

Link: http://lkml.kernel.org/r/1498459987-24562-7-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: add free space management in backing device

With backing device, zram needs management of free space of backing
device.

This patch adds bitmap logic to manage free space which is very naive.
However, it would be simple enough as considering uncompressible pages's
frequenty in zram.

Link: http://lkml.kernel.org/r/1498459987-24562-6-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: add interface to specif backing device

For writeback feature, user should set up backing device before the zram
working.

This patch enables the interface via /sys/block/zramX/backing_dev.

Currently, it supports block device only but it could be enhanced for
file as well.

Link: http://lkml.kernel.org/r/1498459987-24562-5-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: rename zram_decompress_page to __zram_bvec_read

zram_decompress_page naming is not proper because it doesn't decompress
if page was dedup hit or stored with compression.

Use more abstract term and consistent with write path function
__zram_bvec_write.

Link: http://lkml.kernel.org/r/1498459987-24562-4-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: inline zram_compress

zram_compress does several things, compress, entry alloc and check
limitation.  I did for just readbility but it hurts modulization.:(

So this patch removes zram_compress functions and inline it in
__zram_bvec_write for upcoming patches.

Link: http://lkml.kernel.org/r/1498459987-24562-3-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: clean up duplicated codes in __zram_bvec_write

Patch series "writeback incompressible pages to storage", v1.

zRam is useful for memory saving with compressible pages but sometime,
workload can be changed and system has lots of incompressible pages
which is very harmful for zram.

This patch supports writeback feature of zram so admin can set up a
block device and with it, zram can save the memory via writing out the
incompressile pages once it found it's incompressible pages (1/4 comp
ratio) instead of keeping the page in memory.

[1-3] is just clean up and [4-8] is step by step feature enablement.
[4-8] is logically not bisectable(ie, logical unit separation)
although I tried to compiled out without breaking but I think it would
be better to review.

This patch (of 9):

__zram_bvec_write has some of duplicated logic for zram meta data
handling of same_page|compressed_page.  This patch aims to clean it up
without behavior change.

[xieyisheng1@huawei.com: fix compr_data_size stat]
Link: http://lkml.kernel.org/r/1502707447-6944-1-git-send-email-xieyisheng1@huawei.com
Link: http://lkml.kernel.org/r/1496019048-27016-1-git-send-email-minchan@kernel.org
Link: http://lkml.kernel.org/r/1498459987-24562-2-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Juneho Choi <juno.choi@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: remove zone restrictions

Historically we have enforced that any kernel zone (e.g ZONE_NORMAL) has
to precede the Movable zone in the physical memory range.  The purpose
of the movable zone is, however, not bound to any physical memory
restriction.  It merely defines a class of migrateable and reclaimable
memory.

There are users (e.g.  CMA) who might want to reserve specific physical
memory ranges for their own purpose.  Moreover our pfn walkers have to
be prepared for zones overlapping in the physical range already because
we do support interleaving NUMA nodes and therefore zones can interleave
as well.  This means we can allow each memory block to be associated
with a different zone.

Loosen the current onlining semantic and allow explicit onlining type on
any memblock.  That means that online_{kernel,movable} will be allowed
regardless of the physical address of the memblock as long as it is
offline of course.  This might result in moveble zone overlapping with
other kernel zones.  Default onlining then becomes a bit tricky but
still sensible.  echo online > memoryXY/state will online the given
block to

1) the default zone if the given range is outside of any zone
2) the enclosing zone if such a zone doesn't interleave with
   any other zone
        3) the default zone if more zones interleave for this range

where default zone is movable zone only if movable_node is enabled
otherwise it is a kernel zone.

Here is an example of the semantic with (movable_node is not present but
it work in an analogous way). We start with following memblocks, all of
them offline:

  memory34/valid_zones:Normal Movable
  memory35/valid_zones:Normal Movable
  memory36/valid_zones:Normal Movable
  memory37/valid_zones:Normal Movable
  memory38/valid_zones:Normal Movable
  memory39/valid_zones:Normal Movable
  memory40/valid_zones:Normal Movable
  memory41/valid_zones:Normal Movable

Now, we online block 34 in default mode and block 37 as movable

  root@test1:/sys/devices/system/node/node1# echo online > memory34/state
  root@test1:/sys/devices/system/node/node1# echo online_movable > memory37/state
  memory34/valid_zones:Normal
  memory35/valid_zones:Normal Movable
  memory36/valid_zones:Normal Movable
  memory37/valid_zones:Movable
  memory38/valid_zones:Normal Movable
  memory39/valid_zones:Normal Movable
  memory40/valid_zones:Normal Movable
  memory41/valid_zones:Normal Movable

As we can see all other blocks can still be onlined both into Normal and
Movable zones and the Normal is default because the Movable zone spans
only block37 now.

  root@test1:/sys/devices/system/node/node1# echo online_movable > memory41/state
  memory34/valid_zones:Normal
  memory35/valid_zones:Normal Movable
  memory36/valid_zones:Normal Movable
  memory37/valid_zones:Movable
  memory38/valid_zones:Movable Normal
  memory39/valid_zones:Movable Normal
  memory40/valid_zones:Movable Normal
  memory41/valid_zones:Movable

Now the default zone for blocks 37-41 has changed because movable zone
spans that range.

  root@test1:/sys/devices/system/node/node1# echo online_kernel > memory39/state
  memory34/valid_zones:Normal
  memory35/valid_zones:Normal Movable
  memory36/valid_zones:Normal Movable
  memory37/valid_zones:Movable
  memory38/valid_zones:Normal Movable
  memory39/valid_zones:Normal
  memory40/valid_zones:Movable Normal
  memory41/valid_zones:Movable

Note that the block 39 now belongs to the zone Normal and so block38
falls into Normal by default as well.

For completness

  root@test1:/sys/devices/system/node/node1# for i in memory[34]?
  do
echo online > $i/state 2>/dev/null
  done

  memory34/valid_zones:Normal
  memory35/valid_zones:Normal
  memory36/valid_zones:Normal
  memory37/valid_zones:Movable
  memory38/valid_zones:Normal
  memory39/valid_zones:Normal
  memory40/valid_zones:Movable
  memory41/valid_zones:Movable

Implementation wise the change is quite straightforward.  We can get rid
of allow_online_pfn_range altogether.  online_pages allows only offline
nodes already.  The original default_zone_for_pfn will become
default_kernel_zone_for_pfn.  New default_zone_for_pfn implements the
above semantic.  zone_for_pfn_range is slightly reorganized to implement
kernel and movable online type explicitly and MMOP_ONLINE_KEEP becomes a
catch all default behavior.

Link: http://lkml.kernel.org/r/20170714121233.16861-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Yasuaki Ishimatsu <yasu.isimatu@gmail.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: Kani Toshimitsu <toshi.kani@hpe.com>
Cc: <slaoub@gmail.com>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: display allowed zones in the preferred ordering

Prior to commit f1dd2cd13c4b ("mm, memory_hotplug: do not associate
hotadded memory to zones until online") we used to allow to change the
valid zone types of a memory block if it is adjacent to a different zone
type.

This fact was reflected in memoryNN/valid_zones by the ordering of
printed zones.  The first one was default (echo online > memoryNN/state)
and the other one could be onlined explicitly by online_{movable,kernel}.

This behavior was removed by the said patch and as such the ordering was
not all that important.  In most cases a kernel zone would be default
anyway.  The only exception is movable_node handled by "mm,
memory_hotplug: support movable_node for hotpluggable nodes".

Let's reintroduce this behavior again because later patch will remove
the zone overlap restriction and so user will be allowed to online
kernel resp.  movable block regardless of its placement.  Original
behavior will then become significant again because it would be
non-trivial for users to see what is the default zone to online into.

Implementation is really simple.  Pull out zone selection out of
move_pfn_range into zone_for_pfn_range helper and use it in
show_valid_zones to display the zone for default onlining and then both
kernel and movable if they are allowed.  Default online zone is not
duplicated.

Link: http://lkml.kernel.org/r/20170714121233.16861-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Yasuaki Ishimatsu <yasu.isimatu@gmail.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: Kani Toshimitsu <toshi.kani@hpe.com>
Cc: <slaoub@gmail.com>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: just build zonelist for newly added node

Commit 9adb62a5df9c ("mm/hotplug: correctly setup fallback zonelists
when creating new pgdat") tries to build the correct zonelist for a
newly added node, while it is not necessary to rebuild it for already
exist nodes.

In build_zonelists(), it will iterate on nodes with memory.  For a newly
added node, it will have memory until node_states_set_node() is called
in online_pages().

This patch avoids rebuilding the zonelists for already existing nodes.

build_zonelists_node() uses managed_zone(zone) checks, so it should not
include empty zones anyway.  So effectively we avoid some pointless work
under stop_machine().

[akpm@linux-foundation.org: tweak comment text]
[akpm@linux-foundation.org: coding-style tweak, per Vlastimil]
Link: http://lkml.kernel.org/r/20170626035822.50155-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jiang Liu <liuj97@gmail.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drm/i915: wire up shrinkctl->nr_scanned

shrink_slab() allows us to report back the number of objects we
successfully scanned (out of the target shrinkctl->nr_to_scan).  As
report the number of pages owned by each GEM object as a separate item
to the shrinker, we cannot precisely control the number of shrinker
objects we scan on each pass; and indeed may free more than requested.
If we fail to tell the shrinker about the number of objects we process,
it will continue to hold a grudge against us as any objects left
unscanned are added to the next reclaim -- and so we will keep on
"unfairly" shrinking our own slab in comparison to other slabs.

Link: http://lkml.kernel.org/r/20170822135325.9191-2-chris@chris-wilson.co.uk
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Shaohua Li <shli@fb.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: track actual nr_scanned during shrink_slab()

Some shrinkers may only be able to free a bunch of objects at a time,
and so free more than the requested nr_to_scan in one pass.

Whilst other shrinkers may find themselves even unable to scan as many
objects as they counted, and so underreport.  Account for the extra
freed/scanned objects against the total number of objects we intend to
scan, otherwise we may end up penalising the slab far more than
intended.  Similarly, we want to add the underperforming scan to the
deferred pass so that we try harder and harder in future passes.

Link: http://lkml.kernel.org/r/20170822135325.9191-1-chris@chris-wilson.co.uk
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Shaohua Li <shli@fb.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>