Merge updates from Andrew Morton:
- various misc bits
- DAX updates
- OCFS2
- most of MM
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (119 commits)
mm,fork: introduce MADV_WIPEONFORK
x86,mpx: make mpx depend on x86-64 to free up VMA flag
mm: add /proc/pid/smaps_rollup
mm: hugetlb: clear target sub-page last when clearing huge page
mm: oom: let oom_reap_task and exit_mmap run concurrently
swap: choose swap device according to numa node
mm: replace TIF_MEMDIE checks by tsk_is_oom_victim
mm, oom: do not rely on TIF_MEMDIE for memory reserves access
z3fold: use per-cpu unbuddied lists
mm, swap: don't use VMA based swap readahead if HDD is used as swap
mm, swap: add sysfs interface for VMA based swap readahead
mm, swap: VMA based swap readahead
mm, swap: fix swap readahead marking
mm, swap: add swap readahead hit statistics
mm/vmalloc.c: don't reinvent the wheel but use existing llist API
mm/vmstat.c: fix wrong comment
selftests/memfd: add memfd_create hugetlbfs selftest
mm/shmem: add hugetlbfs support to memfd_create()
mm, devm_memremap_pages: use multi-order radix for ZONE_DEVICE lookups
mm/vmalloc.c: halve the number of comparisons performed in pcpu_get_vm_areas()
...
Introduce MADV_WIPEONFORK semantics, which result in a VMA being empty
in the child process after fork. This differs from MADV_DONTFORK in one
important way.
If a child process accesses memory that was MADV_WIPEONFORK, it will get
zeroes. The address ranges are still valid, they are just empty.
If a child process accesses memory that was MADV_DONTFORK, it will get a
segmentation fault, since those address ranges are no longer valid in
the child after fork.
Since MADV_DONTFORK also seems to be used to allow very large programs
to fork in systems with strict memory overcommit restrictions, changing
the semantics of MADV_DONTFORK might break existing programs.
MADV_WIPEONFORK only works on private, anonymous VMAs.
The use case is libraries that store or cache information, and want to
know that they need to regenerate it in the child process after fork.
Examples of this would be:
- systemd/pulseaudio API checks (fail after fork) (replacing a getpid
check, which is too slow without a PID cache)
- PKCS#11 API reinitialization check (mandated by specification)
- glibc's upcoming PRNG (reseed after fork)
- OpenSSL PRNG (reseed after fork)
The security benefits of a forking server having a re-inialized PRNG in
every child process are pretty obvious. However, due to libraries
having all kinds of internal state, and programs getting compiled with
many different versions of each library, it is unreasonable to expect
calling programs to re-initialize everything manually after fork.
A further complication is the proliferation of clone flags, programs
bypassing glibc's functions to call clone directly, and programs calling
unshare, causing the glibc pthread_atfork hook to not get called.
It would be better to have the kernel take care of this automatically.
The patch also adds MADV_KEEPONFORK, to undo the effects of a prior
MADV_WIPEONFORK.
This is similar to the OpenBSD minherit syscall with MAP_INHERIT_ZERO:
https://man.openbsd.org/minherit.2
[akpm@linux-foundation.org: numerically order arch/parisc/include/uapi/asm/mman.h #defines]
Link: http://lkml.kernel.org/r/20170811212829.29186-3-riel@redhat.com
Signed-off-by: Rik van Riel <riel@redhat.com>
Reported-by: Florian Weimer <fweimer@redhat.com>
Reported-by: Colm MacCártaigh <colm@allcosts.net>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Kees Cook <keescook@chromium.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Drewry <wad@chromium.org>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Huge page helps to reduce TLB miss rate, but it has higher cache
footprint, sometimes this may cause some issue. For example, when
clearing huge page on x86_64 platform, the cache footprint is 2M. But
on a Xeon E5 v3 2699 CPU, there are 18 cores, 36 threads, and only 45M
LLC (last level cache). That is, in average, there are 2.5M LLC for
each core and 1.25M LLC for each thread.
If the cache pressure is heavy when clearing the huge page, and we clear
the huge page from the begin to the end, it is possible that the begin
of huge page is evicted from the cache after we finishing clearing the
end of the huge page. And it is possible for the application to access
the begin of the huge page after clearing the huge page.
To help the above situation, in this patch, when we clear a huge page,
the order to clear sub-pages is changed. In quite some situation, we
can get the address that the application will access after we clear the
huge page, for example, in a page fault handler. Instead of clearing
the huge page from begin to end, we will clear the sub-pages farthest
from the the sub-page to access firstly, and clear the sub-page to
access last. This will make the sub-page to access most cache-hot and
sub-pages around it more cache-hot too. If we cannot know the address
the application will access, the begin of the huge page is assumed to be
the the address the application will access.
With this patch, the throughput increases ~28.3% in vm-scalability
anon-w-seq test case with 72 processes on a 2 socket Xeon E5 v3 2699
system (36 cores, 72 threads). The test case creates 72 processes, each
process mmap a big anonymous memory area and writes to it from the begin
to the end. For each process, other processes could be seen as other
workload which generates heavy cache pressure. At the same time, the
cache miss rate reduced from ~33.4% to ~31.7%, the IPC (instruction per
cycle) increased from 0.56 to 0.74, and the time spent in user space is
reduced ~7.9%
Christopher Lameter suggests to clear bytes inside a sub-page from end
to begin too. But tests show no visible performance difference in the
tests. May because the size of page is small compared with the cache
size.
Thanks Andi Kleen to propose to use address to access to determine the
order of sub-pages to clear.
The hugetlbfs access address could be improved, will do that in another
patch.
[ying.huang@intel.com: improve readability of clear_huge_page()]
Link: http://lkml.kernel.org/r/20170830051842.1397-1-ying.huang@intel.com
Link: http://lkml.kernel.org/r/20170815014618.15842-1-ying.huang@intel.com
Suggested-by: Andi Kleen <andi.kleen@intel.com>
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Jan Kara <jack@suse.cz>
Reviewed-by: Michal Hocko <mhocko@suse.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Nadia Yvette Chambers <nyc@holomorphy.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Shaohua Li <shli@fb.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
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: 26db62f179 ("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>
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>
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 1b51e65eab ("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>
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>
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>
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>
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>
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>
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: 944d9fec8d ("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>
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>
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>
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>
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>
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 fa06235b8e ("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>
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>
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 35cd78156c ("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 0a0337e0d1 ("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>
