arm64: mm: use ARCH_HAS_DEBUG_WX instead of arch defined

Extract DEBUG_WX to mm/Kconfig.debug for shared use.  Change to use
ARCH_HAS_DEBUG_WX instead of DEBUG_WX defined by arch port.

Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Link: http://lkml.kernel.org/r/e19709e7576f65e303245fe520cad5f7bae72763.1587455584.git.zong.li@sifive.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86: mm: use ARCH_HAS_DEBUG_WX instead of arch defined

Extract DEBUG_WX to mm/Kconfig.debug for shared use.  Change to use
ARCH_HAS_DEBUG_WX instead of DEBUG_WX defined by arch port.

Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Link: http://lkml.kernel.org/r/430736828d149df3f5b462d291e845ec690e0141.1587455584.git.zong.li@sifive.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

riscv: support DEBUG_WX

Support DEBUG_WX to check whether there are mapping with write and execute
permission at the same time.

[akpm@linux-foundation.org: replace macros with C]
Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Link: http://lkml.kernel.org/r/282e266311bced080bc6f7c255b92f87c1eb65d6.1587455584.git.zong.li@sifive.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add DEBUG_WX support

Patch series "Extract DEBUG_WX to shared use".

Some architectures support DEBUG_WX function, it's verbatim from each
others, so extract to mm/Kconfig.debug for shared use.

PPC and ARM ports don't support generic page dumper yet, so we only
refine x86 and arm64 port in this patch series.

For RISC-V port, the DEBUG_WX support depends on other patches which
be merged already:
  - RISC-V page table dumper
  - Support strict kernel memory permissions for security

This patch (of 4):

Some architectures support DEBUG_WX function, it's verbatim from each
others.  Extract to mm/Kconfig.debug for shared use.

[akpm@linux-foundation.org: reword text, per Will Deacon & Zong Li]
Link: http://lkml.kernel.org/r/20200427194245.oxRJKj3fn%25akpm@linux-foundation.org
[zong.li@sifive.com: remove the specific name of arm64]
Link: http://lkml.kernel.org/r/3a6a92ecedc54e1d0fc941398e63d504c2cd5611.1589178399.git.zong.li@sifive.com
[zong.li@sifive.com: add MMU dependency for DEBUG_WX]
Link: http://lkml.kernel.org/r/4a674ac7863ff39ca91847b10e51209771f99416.1589178399.git.zong.li@sifive.com
Suggested-by: Palmer Dabbelt <palmer@dabbelt.com>
Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: http://lkml.kernel.org/r/cover.1587455584.git.zong.li@sifive.com
Link: http://lkml.kernel.org/r/23980cd0f0e5d79e24a92169116407c75bcc650d.1587455584.git.zong.li@sifive.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/base/memory.c: cache memory blocks in xarray to accelerate lookup

Searching for a particular memory block by id is an O(n) operation because
each memory block's underlying device is kept in an unsorted linked list
on the subsystem bus.

We can cut the lookup cost to O(log n) if we cache each memory block
in an xarray.  This time complexity improvement is significant on
systems with many memory blocks.  For example:

1. A 128GB POWER9 VM with 256MB memblocks has 512 blocks.  With this
   change  memory_dev_init() completes ~12ms faster and walk_memory_blocks()
   completes ~12ms faster.

Before:
[    0.005042] memory_dev_init: adding memory blocks
[    0.021591] memory_dev_init: added memory blocks
[    0.022699] walk_memory_blocks: walking memory blocks
[    0.038730] walk_memory_blocks: walked memory blocks 0-511

After:
[    0.005057] memory_dev_init: adding memory blocks
[    0.009415] memory_dev_init: added memory blocks
[    0.010519] walk_memory_blocks: walking memory blocks
[    0.014135] walk_memory_blocks: walked memory blocks 0-511

2. A 256GB POWER9 LPAR with 256MB memblocks has 1024 blocks.  With
   this change memory_dev_init() completes ~88ms faster and
   walk_memory_blocks() completes ~87ms faster.

Before:
[    0.252246] memory_dev_init: adding memory blocks
[    0.395469] memory_dev_init: added memory blocks
[    0.409413] walk_memory_blocks: walking memory blocks
[    0.433028] walk_memory_blocks: walked memory blocks 0-511
[    0.433094] walk_memory_blocks: walking memory blocks
[    0.500244] walk_memory_blocks: walked memory blocks 131072-131583

After:
[    0.245063] memory_dev_init: adding memory blocks
[    0.299539] memory_dev_init: added memory blocks
[    0.313609] walk_memory_blocks: walking memory blocks
[    0.315287] walk_memory_blocks: walked memory blocks 0-511
[    0.315349] walk_memory_blocks: walking memory blocks
[    0.316988] walk_memory_blocks: walked memory blocks 131072-131583

3. A 32TB POWER9 LPAR with 256MB memblocks has 131072 blocks.  With
   this change we complete memory_dev_init() ~37 minutes faster and
   walk_memory_blocks() at least ~30 minutes faster.  The exact timing
   for walk_memory_blocks() is  missing, though I observed that the
   soft lockups in walk_memory_blocks() disappeared with the change,
   suggesting that lower bound.

Before:
[   13.703907] memory_dev_init: adding blocks
[ 2287.406099] memory_dev_init: added all blocks
[ 2347.494986] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 2527.625378] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 2707.761977] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 2887.899975] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3068.028318] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3248.158764] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3428.287296] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3608.425357] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3788.554572] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 3968.695071] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160
[ 4148.823970] [c000000014c5bb60] [c000000000869af4] walk_memory_blocks+0x94/0x160

After:
[   13.696898] memory_dev_init: adding blocks
[   15.660035] memory_dev_init: added all blocks
(the walk_memory_blocks traces disappear)

There should be no significant negative impact for machines with few
memory blocks.  A sparse xarray has a small footprint and an O(log n)
lookup is negligibly slower than an O(n) lookup for only the smallest
number of memory blocks.

1. A 16GB x86 machine with 128MB memblocks has 132 blocks.  With this
   change memory_dev_init() completes ~300us faster and walk_memory_blocks()
   completes no faster or slower.  The improvement is pretty close to noise.

Before:
[    0.224752] memory_dev_init: adding memory blocks
[    0.227116] memory_dev_init: added memory blocks
[    0.227183] walk_memory_blocks: walking memory blocks
[    0.227183] walk_memory_blocks: walked memory blocks 0-131

After:
[    0.224911] memory_dev_init: adding memory blocks
[    0.226935] memory_dev_init: added memory blocks
[    0.227089] walk_memory_blocks: walking memory blocks
[    0.227089] walk_memory_blocks: walked memory blocks 0-131

[david@redhat.com: document the locking]
Link: http://lkml.kernel.org/r/bc21eec6-7251-4c91-2f57-9a0671f8d414@redhat.com
Signed-off-by: Scott Cheloha <cheloha@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Nathan Lynch <nathanl@linux.ibm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Rafael J. Wysocki <rafael@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Rick Lindsley <ricklind@linux.vnet.ibm.com>
Cc: Scott Cheloha <cheloha@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200121231028.13699-1-cheloha@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/thp: rename pmd_mknotpresent() as pmd_mkinvalid()

pmd_present() is expected to test positive after pmdp_mknotpresent() as
the PMD entry still points to a valid huge page in memory.
pmdp_mknotpresent() implies that given PMD entry is just invalidated from
MMU perspective while still holding on to pmd_page() referred valid huge
page thus also clearing pmd_present() test.  This creates the following
situation which is counter intuitive.

[pmd_present(pmd_mknotpresent(pmd)) = true]

This renames pmd_mknotpresent() as pmd_mkinvalid() reflecting the helper's
functionality more accurately while changing the above mentioned situation
as follows.  This does not create any functional change.

[pmd_present(pmd_mkinvalid(pmd)) = true]

This is not applicable for platforms that define own pmdp_invalidate() via
__HAVE_ARCH_PMDP_INVALIDATE.  Suggestion for renaming came during a
previous discussion here.

https://patchwork.kernel.org/patch/11019637/

[anshuman.khandual@arm.com: change pmd_mknotvalid() to pmd_mkinvalid() per Will]
Link: http://lkml.kernel.org/r/1587520326-10099-3-git-send-email-anshuman.khandual@arm.com
Suggested-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Will Deacon <will@kernel.org>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Link: http://lkml.kernel.org/r/1584680057-13753-3-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc/mm: drop platform defined pmd_mknotpresent()

Patch series "mm/thp: Rename pmd_mknotpresent() as pmd_mknotvalid()", v2.

This series renames pmd_mknotpresent() as pmd_mknotvalid().  Before that
it drops an existing pmd_mknotpresent() definition from powerpc platform
which was never required as it defines it's pmdp_invalidate() through
subscribing __HAVE_ARCH_PMDP_INVALIDATE.  This does not create any
functional change.

This rename was suggested by Catalin during a previous discussion while we
were trying to change the THP helpers on arm64 platform for migration.

https://patchwork.kernel.org/patch/11019637/

This patch (of 2):

Platform needs to define pmd_mknotpresent() for generic pmdp_invalidate()
only when __HAVE_ARCH_PMDP_INVALIDATE is not subscribed.  Otherwise
platform specific pmd_mknotpresent() is not required.  Hence just drop it.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1587520326-10099-1-git-send-email-anshuman.khandual@arm.com
Link: http://lkml.kernel.org/r/1584680057-13753-1-git-send-email-anshuman.khandual@arm.com
Link: http://lkml.kernel.org/r/1584680057-13753-2-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: thp: don't need to drain lru cache when splitting and mlocking THP

Since commit 8f182270dfec ("mm/swap.c: flush lru pvecs on compound page
arrival") THP would not stay in pagevec anymore.  So the optimization made
by commit d965432234db ("thp: increase split_huge_page() success rate")
doesn't make sense anymore, which tries to unpin munlocked THPs from
pagevec by draining pagevec.

Draining lru cache before isolating THP in mlock path is also unnecessary.
b676b293fb48 ("mm, thp: fix mapped pages avoiding unevictable list on
mlock") added it and 9a73f61bdb8a ("thp, mlock: do not mlock PTE-mapped
file huge pages") accidentally carried it over after the above
optimization went in.

Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Link: http://lkml.kernel.org/r/1585946493-7531-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: get unmapped area below TASK_UNMAPPED_BASE for hugetlbfs

In a 32-bit program, running on arm64 architecture.  When the address
space below mmap base is completely exhausted, shmat() for huge pages will
return ENOMEM, but shmat() for normal pages can still success on no-legacy
mode.  This seems not fair.

For normal pages, the calling trace of get_unmapped_area() is:

=> mm->get_unmapped_area()
if on legacy mode,
=> arch_get_unmapped_area()
=> vm_unmapped_area()
if on no-legacy mode,
=> arch_get_unmapped_area_topdown()
=> vm_unmapped_area()

For huge pages, the calling trace of get_unmapped_area() is:

=> file->f_op->get_unmapped_area()
=> hugetlb_get_unmapped_area()
=> vm_unmapped_area()

To solve this issue, we only need to make hugetlb_get_unmapped_area() take
the same way as mm->get_unmapped_area().  Add *bottomup() and *topdown()
for hugetlbfs, and check current mm->get_unmapped_area() to decide which
one to use.  If mm->get_unmapped_area is equal to
arch_get_unmapped_area_topdown(), hugetlb_get_unmapped_area() calls
topdown routine, otherwise calls bottomup routine.

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Shijie Hu <hushijie3@huawei.com>
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Will Deacon <will@kernel.org>
Cc: Xiaoming Ni <nixiaoming@huawei.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: yangerkun <yangerkun@huawei.com>
Cc: ChenGang <cg.chen@huawei.com>
Cc: Chen Jie <chenjie6@huawei.com>
Link: http://lkml.kernel.org/r/20200518065338.113664-1-hushijie3@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sparc32: register memory occupied by kernel as memblock.memory

sparc32 never registered the memory occupied by the kernel image with
memblock_add() and it only reserved this memory with meblock_reserve().

With openbios as system firmware, the memory occupied by the kernel is
reserved in openbios and removed from mem.available.  The prom setup code
in the kernel uses mem.available to set up the memory banks and
essentially there is a hole for the memory occupied by the kernel image.

Later in bootmem_init() this memory is memblock_reserve()d.

Up until recently, memmap initialization would call __init_single_page()
for the pages in that hole, the free_low_memory_core_early() would mark
them as reserved and everything would be Ok.

After the change in memmap initialization introduced by the commit "mm:
memmap_init: iterate over memblock regions rather that check each PFN",
the hole is skipped and the page structs for it are not initialized.  And
when they are passed from memblock to page allocator as reserved, the
latter gets confused.

Simply registering the memory occupied by the kernel with memblock_add()
resolves this issue.

Tested on qemu-system-sparc with Debian Etch [1] userspace.

[1] https://people.debian.org/~aurel32/qemu/sparc/debian_etch_sparc_small.qcow2

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Guenter Roeck <linux@roeck-us.net>
Link: https://lkml.kernel.org/r/20200517000050.GA87467@roeck-us.nlllllet/
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/memblock.h: fix minor typo and unclear comment

Fix a minor typo "usabe->usable" for the current discription of member
variable "memory" in struct memblock.

BTW, I think it's unclear the member variable "base" in struct
memblock_type is currently described as the physical address of memory
region, change it to base address of the region is clearer since the
variable is decorated as phys_addr_t.

Signed-off-by: chenqiwu <chenqiwu@xiaomi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Link: http://lkml.kernel.org/r/1588846952-32166-1-git-send-email-qiwuchen55@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, mempolicy: fix up gup usage in lookup_node

ba841078cd05 ("mm/mempolicy: Allow lookup_node() to handle fatal signal")
has added a special casing for 0 return value because that was a possible
gup return value when interrupted by fatal signal.  This has been fixed by
ae46d2aa6a7f ("mm/gup: Let __get_user_pages_locked() return -EINTR for
fatal signal") in the mean time so ba841078cd05 can be reverted.

This patch however doesn't go all the way to revert it because the check
for 0 is wrong and confusing here.  Firstly it is inherently unsafe to
access the page when get_user_pages_locked returns 0 (aka no page
returned).

Fortunatelly this will not happen because get_user_pages_locked will not
return 0 when nr_pages > 0 unless FOLL_NOWAIT is specified which is not
the case here.  Document this potential error code in gup code while we
are at it.

Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Xu <peterx@redhat.com>
Link: http://lkml.kernel.org/r/20200421071026.18394-1-mhocko@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page_owner_sort.c: filter out unneeded line

To see a sorted result from page_owner, there must be a tiresome
preprocessing step before running page_owner_sort.  This patch simply
filters out lines which start with "PFN" while reading the page owner
report.

Signed-off-by: Changhee Han <ch0.han@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Link: http://lkml.kernel.org/r/20200429052940.16968-1-ch0.han@lge.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: swap: memcg: fix memcg stats for huge pages

The commit 2262185c5b28 ("mm: per-cgroup memory reclaim stats") added
PGLAZYFREE, PGACTIVATE & PGDEACTIVATE stats for cgroups but missed
couple of places and PGLAZYFREE missed huge page handling. Fix that.
Also for PGLAZYFREE use the irq-unsafe function to update as the irq is
already disabled.

Fixes: 2262185c5b28 ("mm: per-cgroup memory reclaim stats")
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: http://lkml.kernel.org/r/20200527182947.251343-1-shakeelb@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: swap: fix vmstats for huge pages

Many of the callbacks called by pagevec_lru_move_fn() does not correctly
update the vmstats for huge pages. Fix that. Also __pagevec_lru_add_fn()
use the irq-unsafe alternative to update the stat as the irqs are
already disabled.

Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: http://lkml.kernel.org/r/20200527182916.249910-1-shakeelb@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: limit the range of LRU type balancing

When LRU cost only shows up on one list, we abruptly stop scanning that
list altogether.  That's an extreme reaction: by the time the other list
starts thrashing and the pendulum swings back, we may have no recent age
information on the first list anymore, and we could have significant
latencies until the scanner has caught up.

Soften this change in the feedback system by ensuring that no list
receives less than a third of overall pressure, and only distribute the
other 66% according to LRU cost.  This ensures that we maintain a minimum
rate of aging on the entire workingset while it's being pressured, while
still allowing a generous rate of convergence when the relative sizes of
the lists need to adjust.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-15-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: reclaim writepage is IO cost

The VM tries to balance reclaim pressure between anon and file so as to
reduce the amount of IO incurred due to the memory shortage.  It already
counts refaults and swapins, but in addition it should also count
writepage calls during reclaim.

For swap, this is obvious: it's IO that wouldn't have occurred if the
anonymous memory hadn't been under memory pressure.  From a relative
balancing point of view this makes sense as well: even if anon is cold and
reclaimable, a cache that isn't thrashing may have equally cold pages that
don't require IO to reclaim.

For file writeback, it's trickier: some of the reclaim writepage IO would
have likely occurred anyway due to dirty expiration.  But not all of it -
premature writeback reduces batching and generates additional writes.
Since the flushers are already woken up by the time the VM starts writing
cache pages one by one, let's assume that we'e likely causing writes that
wouldn't have happened without memory pressure.  In addition, the per-page
cost of IO would have probably been much cheaper if written in larger
batches from the flusher thread rather than the single-page-writes from
kswapd.

For our purposes - getting the trend right to accelerate convergence on a
stable state that doesn't require paging at all - this is sufficiently
accurate.  If we later wanted to optimize for sustained thrashing, we can
still refine the measurements.

Count all writepage calls from kswapd as IO cost toward the LRU that the
page belongs to.

Why do this dynamically?  Don't we know in advance that anon pages require
IO to reclaim, and so could build in a static bias?

First, scanning is not the same as reclaiming.  If all the anon pages are
referenced, we may not swap for a while just because we're scanning the
anon list.  During this time, however, it's important that we age
anonymous memory and the page cache at the same rate so that their
hot-cold gradients are comparable.  Everything else being equal, we still
want to reclaim the coldest memory overall.

Second, we keep copies in swap unless the page changes.  If there is
swap-backed data that's mostly read (tmpfs file) and has been swapped out
before, we can reclaim it without incurring additional IO.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-14-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: determine anon/file pressure balance at the reclaim root

We split the LRU lists into anon and file, and we rebalance the scan
pressure between them when one of them begins thrashing: if the file cache
experiences workingset refaults, we increase the pressure on anonymous
pages; if the workload is stalled on swapins, we increase the pressure on
the file cache instead.

With cgroups and their nested LRU lists, we currently don't do this
correctly.  While recursive cgroup reclaim establishes a relative LRU
order among the pages of all involved cgroups, LRU pressure balancing is
done on an individual cgroup LRU level.  As a result, when one cgroup is
thrashing on the filesystem cache while a sibling may have cold anonymous
pages, pressure doesn't get equalized between them.

This patch moves LRU balancing decision to the root of reclaim - the same
level where the LRU order is established.

It does this by tracking LRU cost recursively, so that every level of the
cgroup tree knows the aggregate LRU cost of all memory within its domain.
When the page scanner calculates the scan balance for any given individual
cgroup's LRU list, it uses the values from the ancestor cgroup that
initiated the reclaim cycle.

If one sibling is then thrashing on the cache, it will tip the pressure
balance inside its ancestors, and the next hierarchical reclaim iteration
will go more after the anon pages in the tree.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-13-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: balance LRU lists based on relative thrashing

Since the LRUs were split into anon and file lists, the VM has been
balancing between page cache and anonymous pages based on per-list ratios
of scanned vs.  rotated pages.  In most cases that tips page reclaim
towards the list that is easier to reclaim and has the fewest actively
used pages, but there are a few problems with it:

1. Refaults and LRU rotations are weighted the same way, even though
   one costs IO and the other costs a bit of CPU.

2. The less we scan an LRU list based on already observed rotations,
   the more we increase the sampling interval for new references, and
   rotations become even more likely on that list. This can enter a
   death spiral in which we stop looking at one list completely until
   the other one is all but annihilated by page reclaim.

Since commit a528910e12ec ("mm: thrash detection-based file cache sizing")
we have refault detection for the page cache.  Along with swapin events,
they are good indicators of when the file or anon list, respectively, is
too small for its workingset and needs to grow.

For example, if the page cache is thrashing, the cache pages need more
time in memory, while there may be colder pages on the anonymous list.
Likewise, if swapped pages are faulting back in, it indicates that we
reclaim anonymous pages too aggressively and should back off.

Replace LRU rotations with refaults and swapins as the basis for relative
reclaim cost of the two LRUs.  This will have the VM target list balances
that incur the least amount of IO on aggregate.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-12-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: only count actual rotations as LRU reclaim cost

When shrinking the active file list we rotate referenced pages only when
they're in an executable mapping.  The others get deactivated.  When it
comes to balancing scan pressure, though, we count all referenced pages as
rotated, even the deactivated ones.  Yet they do not carry the same cost
to the system: the deactivated page *might* refault later on, but the
deactivation is tangible progress toward freeing pages; rotations on the
other hand cost time and effort without getting any closer to freeing
memory.

Don't treat both events as equal.  The following patch will hook up LRU
balancing to cache and anon refaults, which are a much more concrete cost
signal for reclaiming one list over the other.  Thus, remove the maybe-IO
cost bias from page references, and only note the CPU cost for actual
rotations that prevent the pages from getting reclaimed.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-11-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: deactivations shouldn't bias the LRU balance

Operations like MADV_FREE, FADV_DONTNEED etc.  currently move any affected
active pages to the inactive list to accelerate their reclaim (good) but
also steer page reclaim toward that LRU type, or away from the other
(bad).

The reason why this is undesirable is that such operations are not part of
the regular page aging cycle, and rather a fluke that doesn't say much
about the remaining pages on that list; they might all be in heavy use,
and once the chunk of easy victims has been purged, the VM continues to
apply elevated pressure on those remaining hot pages.  The other LRU,
meanwhile, might have easily reclaimable pages, and there was never a need
to steer away from it in the first place.

As the previous patch outlined, we should focus on recording actually
observed cost to steer the balance rather than speculating about the
potential value of one LRU list over the other.  In that spirit, leave
explicitely deactivated pages to the LRU algorithm to pick up, and let
rotations decide which list is the easiest to reclaim.

[cai@lca.pw: fix set-but-not-used warning]
Link: http://lkml.kernel.org/r/20200522133335.GA624@Qians-MacBook-Air.local
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Qian Cai <cai@lca.pw>
Link: http://lkml.kernel.org/r/20200520232525.798933-10-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: base LRU balancing on an explicit cost model

Currently, scan pressure between the anon and file LRU lists is balanced
based on a mixture of reclaim efficiency and a somewhat vague notion of
"value" of having certain pages in memory over others.  That concept of
value is problematic, because it has caused us to count any event that
remotely makes one LRU list more or less preferrable for reclaim, even
when these events are not directly comparable and impose very different
costs on the system.  One example is referenced file pages that we still
deactivate and referenced anonymous pages that we actually rotate back to
the head of the list.

There is also conceptual overlap with the LRU algorithm itself.  By
rotating recently used pages instead of reclaiming them, the algorithm
already biases the applied scan pressure based on page value.  Thus, when
rebalancing scan pressure due to rotations, we should think of reclaim
cost, and leave assessing the page value to the LRU algorithm.

Lastly, considering both value-increasing as well as value-decreasing
events can sometimes cause the same type of event to be counted twice,
i.e.  how rotating a page increases the LRU value, while reclaiming it
succesfully decreases the value.  In itself this will balance out fine,
but it quietly skews the impact of events that are only recorded once.

The abstract metric of "value", the murky relationship with the LRU
algorithm, and accounting both negative and positive events make the
current pressure balancing model hard to reason about and modify.

This patch switches to a balancing model of accounting the concrete,
actually observed cost of reclaiming one LRU over another.  For now, that
cost includes pages that are scanned but rotated back to the list head.
Subsequent patches will add consideration for IO caused by refaulting of
recently evicted pages.

Replace struct zone_reclaim_stat with two cost counters in the lruvec, and
make everything that affects cost go through a new lru_note_cost()
function.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-9-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: drop unnecessary div0 avoidance rounding in get_scan_count()

When we calculate the relative scan pressure between the anon and file LRU
lists, we have to assume that reclaim_stat can contain zeroes.  To avoid
div0 crashes, we add 1 to all denominators like so:

        anon_prio = swappiness;
        file_prio = 200 - anon_prio;

[...]

        /*
         * The amount of pressure on anon vs file pages is inversely
         * proportional to the fraction of recently scanned pages on
         * each list that were recently referenced and in active use.
         */
        ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1);
        ap /= reclaim_stat->recent_rotated[0] + 1;

        fp = file_prio * (reclaim_stat->recent_scanned[1] + 1);
        fp /= reclaim_stat->recent_rotated[1] + 1;
        spin_unlock_irq(&pgdat->lru_lock);

        fraction[0] = ap;
        fraction[1] = fp;
        denominator = ap + fp + 1;

While reclaim_stat can contain 0, it's not actually possible for ap + fp
to be 0.  One of anon_prio or file_prio could be zero, but they must still
add up to 200.  And the reclaim_stat fraction, due to the +1 in there, is
always at least 1.  So if one of the two numerators is 0, the other one
can't be.  ap + fp is always at least 1.  Drop the + 1.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-8-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove use-once cache bias from LRU balancing

When the splitlru patches divided page cache and swap-backed pages into
separate LRU lists, the pressure balance between the lists was biased to
account for the fact that streaming IO can cause memory pressure with a
flood of pages that are used only once.  New page cache additions would
tip the balance toward the file LRU, and repeat access would neutralize
that bias again.  This ensured that page reclaim would always go for
used-once cache first.

Since e9868505987a ("mm,vmscan: only evict file pages when we have
plenty"), page reclaim generally skips over swap-backed memory entirely as
long as there is used-once cache present, and will apply the LRU balancing
when only repeatedly accessed cache pages are left - at which point the
previous use-once bias will have been neutralized.  This makes the
use-once cache balancing bias unnecessary.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-7-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: workingset: let cache workingset challenge anon

We activate cache refaults with reuse distances in pages smaller than the
size of the total cache.  This allows new pages with competitive access
frequencies to establish themselves, as well as challenge and potentially
displace pages on the active list that have gone cold.

However, that assumes that active cache can only replace other active
cache in a competition for the hottest memory.  This is not a great
default assumption.  The page cache might be thrashing while there are
enough completely cold and unused anonymous pages sitting around that we'd
only have to write to swap once to stop all IO from the cache.

Activate cache refaults when their reuse distance in pages is smaller than
the total userspace workingset, including anonymous pages.

Reclaim can still decide how to balance pressure among the two LRUs
depending on the IO situation.  Rotational drives will prefer avoiding
random IO from swap and go harder after cache.  But fundamentally, hot
cache should be able to compete with anon pages for a place in RAM.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-6-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fold and remove lru_cache_add_anon() and lru_cache_add_file()

They're the same function, and for the purpose of all callers they are
equivalent to lru_cache_add().

[akpm@linux-foundation.org: fix it for local_lock changes]
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-5-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: allow swappiness that prefers reclaiming anon over the file workingset

With the advent of fast random IO devices (SSDs, PMEM) and in-memory swap
devices such as zswap, it's possible for swap to be much faster than
filesystems, and for swapping to be preferable over thrashing filesystem
caches.

Allow setting swappiness - which defines the rough relative IO cost of
cache misses between page cache and swap-backed pages - to reflect such
situations by making the swap-preferred range configurable.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-4-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: keep separate anon and file statistics on page reclaim activity

Having statistics on pages scanned and pages reclaimed for both anon and
file pages makes it easier to evaluate changes to LRU balancing.

While at it, clean up the stat-keeping mess for isolation, putback,
reclaim stats etc.  a bit: first the physical LRU operation (isolation and
putback), followed by vmstats, reclaim_stats, and then vm events.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-3-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix LRU balancing effect of new transparent huge pages

The reclaim code that balances between swapping and cache reclaim tries to
predict likely reuse based on in-memory reference patterns alone.  This
works in many cases, but when it fails it cannot detect when the cache is
thrashing pathologically, or when we're in the middle of a swap storm.

The high seek cost of rotational drives under which the algorithm evolved
also meant that mistakes could quickly result in lockups from too
aggressive swapping (which is predominantly random IO).  As a result, the
balancing code has been tuned over time to a point where it mostly goes
for page cache and defers swapping until the VM is under significant
memory pressure.

The resulting strategy doesn't make optimal caching decisions - where
optimal is the least amount of IO required to execute the workload.

The proliferation of fast random IO devices such as SSDs, in-memory
compression such as zswap, and persistent memory technologies on the
horizon, has made this undesirable behavior very noticable: Even in the
presence of large amounts of cold anonymous memory and a capable swap
device, the VM refuses to even seriously scan these pages, and can leave
the page cache thrashing needlessly.

This series sets out to address this.  Since commit ("a528910e12ec mm:
thrash detection-based file cache sizing") we have exact tracking of
refault IO - the ultimate cost of reclaiming the wrong pages.  This allows
us to use an IO cost based balancing model that is more aggressive about
scanning anonymous memory when the cache is thrashing, while being able to
avoid unnecessary swap storms.

These patches base the LRU balance on the rate of refaults on each list,
times the relative IO cost between swap device and filesystem
(swappiness), in order to optimize reclaim for least IO cost incurred.

History

I floated these changes in 2016.  At the time they were incomplete and
full of workarounds due to a lack of infrastructure in the reclaim code:
We didn't have PageWorkingset, we didn't have hierarchical cgroup
statistics, and problems with the cgroup swap controller.  As swapping
wasn't too high a priority then, the patches stalled out.  With all
dependencies in place now, here we are again with much cleaner,
feature-complete patches.

I kept the acks for patches that stayed materially the same :-)

Below is a series of test results that demonstrate certain problematic
behavior of the current code, as well as showcase the new code's more
predictable and appropriate balancing decisions.

Test #1: No convergence

This test shows an edge case where the VM currently doesn't converge at
all on a new file workingset with a stale anon/tmpfs set.

The test sets up a cold anon set the size of 3/4 RAM, then tries to
establish a new file set half the size of RAM (flat access pattern).

The vanilla kernel refuses to even scan anon pages and never converges.
The file set is perpetually served from the filesystem.

The first test kernel is with the series up to the workingset patch
applied.  This allows thrashing page cache to challenge the anonymous
workingset.  The VM then scans the lists based on the current
scanned/rotated balancing algorithm.  It converges on a stable state where
all cold anon pages are pushed out and the fileset is served entirely from
cache:

    noconverge/5.7-rc5-mm noconverge/5.7-rc5-mm-workingset
Scanned 417719308.00 (    +0.00%) 64091155.00 (   -84.66%)
Reclaimed 417711094.00 (    +0.00%) 61640308.00 (   -85.24%)
Reclaim efficiency %       100.00 (    +0.00%)       96.18 (    -3.78%)
Scanned file 417719308.00 (    +0.00%) 59211118.00 (   -85.83%)
Scanned anon 0.00 (    +0.00%)          4880037.00 (          )
Swapouts 0.00 (    +0.00%)          2439957.00 (          )
Swapins 0.00 (    +0.00%)      257.00 (          )
Refaults 415246605.00 (    +0.00%) 59183722.00 (   -85.75%)
Restore refaults 0.00 (    +0.00%)         54988252.00 (          )

The second test kernel is with the full patch series applied, which
replaces the scanned/rotated ratios with refault/swapin rate-based
balancing.  It evicts the cold anon pages more aggressively in the
presence of a thrashing cache and the absence of swapins, and so converges
with about 60% of the IO and reclaim activity:

noconverge/5.7-rc5-mm-workingset noconverge/5.7-rc5-mm-lrubalance
Scanned 64091155.00 (    +0.00%) 37579741.00 (   -41.37%)
Reclaimed 61640308.00 (    +0.00%) 35129293.00 (   -43.01%)
Reclaim efficiency %       96.18 (    +0.00%)       93.48 (    -2.78%)
Scanned file 59211118.00 (    +0.00%) 32708385.00 (   -44.76%)
Scanned anon  4880037.00 (    +0.00%)  4871356.00 (    -0.18%)
Swapouts  2439957.00 (    +0.00%)  2435565.00 (    -0.18%)
Swapins      257.00 (    +0.00%)      262.00 (    +1.94%)
Refaults 59183722.00 (    +0.00%) 32675667.00 (   -44.79%)
Restore refaults 54988252.00 (    +0.00%) 28480430.00 (   -48.21%)

We're triggering this case in host sideloading scenarios: When a host's
primary workload is not saturating the machine (primary load is usually
driven by user activity), we can optimistically sideload a batch job; if
user activity picks up and the primary workload needs the whole host
during this time, we freeze the sideload and rely on it getting pushed to
swap.  Frequently that swapping doesn't happen and the completely inactive
sideload simply stays resident while the expanding primary worklad is
struggling to gain ground.

Test #2: Kernel build

This test is a a kernel build that is slightly memory-restricted (make -j4
inside a 400M cgroup).

Despite the very aggressive swapping of cold anon pages in test #1, this
test shows that the new kernel carefully balances swap against cache
refaults when both the file and the cache set are pressured.

It shows the patched kernel to be slightly better at finding the coldest
memory from the combined anon and file set to evict under pressure.  The
result is lower aggregate reclaim and paging activity:

z     5.7-rc5-mm 5.7-rc5-mm-lrubalance
Real time    210.60 (    +0.00%)    210.97 (    +0.18%)
User time    745.42 (    +0.00%)    746.48 (    +0.14%)
System time     69.78 (    +0.00%)     69.79 (    +0.02%)
Scanned file 354682.00 (    +0.00%) 293661.00 (   -17.20%)
Scanned anon 465381.00 (    +0.00%) 378144.00 (   -18.75%)
Swapouts 185920.00 (    +0.00%) 147801.00 (   -20.50%)
Swapins  34583.00 (    +0.00%)  32491.00 (    -6.05%)
Refaults 212664.00 (    +0.00%) 172409.00 (   -18.93%)
Restore refaults  48861.00 (    +0.00%)  80091.00 (   +63.91%)
Total paging IO 433167.00 (    +0.00%) 352701.00 (   -18.58%)

Test #3: Overload

This next test is not about performance, but rather about the
predictability of the algorithm.  The current balancing behavior doesn't
always lead to comprehensible results, which makes performance analysis
and parameter tuning (swappiness e.g.) very difficult.

The test shows the balancing behavior under equivalent anon and file
input.  Anon and file sets are created of equal size (3/4 RAM), have the
same access patterns (a hot-cold gradient), and synchronized access rates.
Swappiness is raised from the default of 60 to 100 to indicate equal IO
cost between swap and cache.

With the vanilla balancing code, anon scans make up around 9% of the total
pages scanned, or a ~1:10 ratio.  This is a surprisingly skewed ratio, and
it's an outcome that is hard to explain given the input parameters to the
VM.

The new balancing model targets a 1:2 balance: All else being equal,
reclaiming a file page costs one page IO - the refault; reclaiming an anon
page costs two IOs - the swapout and the swapin.  In the test we observe a
~1:3 balance.

The scanned and paging IO numbers indicate that the anon LRU algorithm we
have in place right now does a slightly worse job at picking the coldest
pages compared to the file algorithm.  There is ongoing work to improve
this, like Joonsoo's anon workingset patches; however, it's difficult to
compare the two aging strategies when the balancing between them is
behaving unintuitively.

The slightly less efficient anon reclaim results in a deviation from the
optimal 1:2 scan ratio we would like to see here - however, 1:3 is much
closer to what we'd want to see in this test than the vanilla kernel's
aging of 10+ cache pages for every anonymous one:

overload-100/5.7-rc5-mm-workingset overload-100/5.7-rc5-mm-lrubalance-realfile
Scanned  533633725.00 (    +0.00%)   595687785.00 (   +11.63%)
Reclaimed  494325440.00 (    +0.00%)   518154380.00 (    +4.82%)
Reclaim efficiency % 92.63 (    +0.00%)  86.98 (    -6.03%)
Scanned file  484532894.00 (    +0.00%)   456937722.00 (    -5.70%)
Scanned anon   49100831.00 (    +0.00%)   138750063.00 (  +182.58%)
Swapouts    8096423.00 (    +0.00%)    48982142.00 (  +504.98%)
Swapins   10027384.00 (    +0.00%)    62325044.00 (  +521.55%)
Refaults  479819973.00 (    +0.00%)   451309483.00 (    -5.94%)
Restore refaults  426422087.00 (    +0.00%)   399914067.00 (    -6.22%)
Total paging IO  497943780.00 (    +0.00%)   562616669.00 (   +12.99%)

Test #4: Parallel IO

It's important to note that these patches only affect the situation where
the kernel has to reclaim workingset memory, which is usually a
transitionary period.  The vast majority of page reclaim occuring in a
system is from trimming the ever-expanding page cache.

These patches don't affect cache trimming behavior.  We never swap as long
as we only have use-once cache moving through the file LRU, we only
consider swapping when the cache is actively thrashing.

The following test demonstrates this.  It has an anon workingset that
takes up half of RAM and then writes a file that is twice the size of RAM
out to disk.

As the cache is funneled through the inactive file list, no anon pages are
scanned (aside from apparently some background noise of 10 pages):

  5.7-rc5-mm           5.7-rc5-mm-lrubalance
Scanned     10714722.00 (    +0.00%)        10723445.00 (    +0.08%)
Reclaimed     10703596.00 (    +0.00%)        10712166.00 (    +0.08%)
Reclaim efficiency %   99.90 (    +0.00%)      99.89 (    -0.00%)
Scanned file     10714722.00 (    +0.00%)        10723435.00 (    +0.08%)
Scanned anon    0.00 (    +0.00%)      10.00 (          )
Swapouts    0.00 (    +0.00%)       7.00 (          )
Swapins    0.00 (    +0.00%)       0.00 (    +0.00%)
Refaults   92.00 (    +0.00%)      41.00 (   -54.84%)
Restore refaults    0.00 (    +0.00%)       0.00 (    +0.00%)
Total paging IO   92.00 (    +0.00%)      48.00 (   -47.31%)

This patch (of 14):

Currently, THP are counted as single pages until they are split right
before being swapped out.  However, at that point the VM is already in the
middle of reclaim, and adjusting the LRU balance then is useless.

Always account THP by the number of basepages, and remove the fixup from
the splitting path.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Link: http://lkml.kernel.org/r/20200520232525.798933-1-hannes@cmpxchg.org
Link: http://lkml.kernel.org/r/20200520232525.798933-2-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: update page->mem_cgroup stability rules

The previous patches have simplified the access rules around
page->mem_cgroup somewhat:

1. We never change page->mem_cgroup while the page is isolated by
   somebody else.  This was by far the biggest exception to our rules and
   it didn't stop at lock_page() or lock_page_memcg().

2. We charge pages before they get put into page tables now, so the
   somewhat fishy rule about "can be in page table as long as it's still
   locked" is now gone and boiled down to having an exclusive reference to
   the page.

Document the new rules.  Any of the following will stabilize the
page->mem_cgroup association:

- the page lock
- LRU isolation
- lock_page_memcg()
- exclusive access to the page

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-20-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: delete unused lrucare handling

Swapin faults were the last event to charge pages after they had already
been put on the LRU list.  Now that we charge directly on swapin, the
lrucare portion of the charge code is unused.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Shakeel Butt <shakeelb@google.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-19-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: document the new swap control behavior

Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-18-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: charge swapin pages on instantiation

Right now, users that are otherwise memory controlled can easily escape
their containment and allocate significant amounts of memory that they're
not being charged for.  That's because swap readahead pages are not being
charged until somebody actually faults them into their page table.  This
can be exploited with MADV_WILLNEED, which triggers arbitrary readahead
allocations without charging the pages.

There are additional problems with the delayed charging of swap pages:

1. To implement refault/workingset detection for anonymous pages, we
   need to have a target LRU available at swapin time, but the LRU is not
   determinable until the page has been charged.

2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be
   stable when the page is isolated from the LRU; otherwise, the locks
   change under us.  But swapcache gets charged after it's already on the
   LRU, and even if we cannot isolate it ourselves (since charging is not
   exactly optional).

The previous patch ensured we always maintain cgroup ownership records for
swap pages.  This patch moves the swapcache charging point from the fault
handler to swapin time to fix all of the above problems.

v2: simplify swapin error checking (Joonsoo)

[hughd@google.com: fix livelock in __read_swap_cache_async()]
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Rafael Aquini <aquini@redhat.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: make swap tracking an integral part of memory control

Without swap page tracking, users that are otherwise memory controlled can
easily escape their containment and allocate significant amounts of memory
that they're not being charged for.  That's because swap does readahead,
but without the cgroup records of who owned the page at swapout, readahead
pages don't get charged until somebody actually faults them into their
page table and we can identify an owner task.  This can be maliciously
exploited with MADV_WILLNEED, which triggers arbitrary readahead
allocations without charging the pages.

Make swap swap page tracking an integral part of memcg and remove the
Kconfig options.  In the first place, it was only made configurable to
allow users to save some memory.  But the overhead of tracking cgroup
ownership per swap page is minimal - 2 byte per page, or 512k per 1G of
swap, or 0.04%.  Saving that at the expense of broken containment
semantics is not something we should present as a coequal option.

The swapaccount=0 boot option will continue to exist, and it will
eliminate the page_counter overhead and hide the swap control files, but
it won't disable swap slot ownership tracking.

This patch makes sure we always have the cgroup records at swapin time;
the next patch will fix the actual bug by charging readahead swap pages at
swapin time rather than at fault time.

v2: fix double swap charge bug in cgroup1/cgroup2 code gating

[hannes@cmpxchg.org: fix crash with cgroup_disable=memory]
Link: http://lkml.kernel.org/r/20200521215855.GB815153@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Naresh Kamboju <naresh.kamboju@linaro.org>
Link: http://lkml.kernel.org/r/20200508183105.225460-16-hannes@cmpxchg.org
Debugged-by: Hugh Dickins <hughd@google.com>
Debugged-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: prepare swap controller setup for integration

A few cleanups to streamline the swap controller setup:

- Replace the do_swap_account flag with cgroup_memory_noswap. This
  brings it in line with other functionality that is usually available
  unless explicitly opted out of - nosocket, nokmem.

- Remove the really_do_swap_account flag that stores the boot option
  and is later used to switch the do_swap_account. It's not clear why
  this indirection is/was necessary. Use do_swap_account directly.

- Minor coding style polishing

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-15-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: drop unused try/commit/cancel charge API

There are no more users. RIP in peace.

[arnd@arndb.de: fix an unused-function warning]
Link: http://lkml.kernel.org/r/20200528095640.151454-1-arnd@arndb.de
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-14-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: convert anon and file-thp to new mem_cgroup_charge() API

With the page->mapping requirement gone from memcg, we can charge anon and
file-thp pages in one single step, right after they're allocated.

This removes two out of three API calls - especially the tricky commit
step that needed to happen at just the right time between when the page is
"set up" and when it's "published" - somewhat vague and fluid concepts
that varied by page type.  All we need is a freshly allocated page and a
memcg context to charge.

v2: prevent double charges on pre-allocated hugepages in khugepaged

[hannes@cmpxchg.org: Fix crash - *hpage could be ERR_PTR instead of NULL]
Link: http://lkml.kernel.org/r/20200512215813.GA487759@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Qian Cai <cai@lca.pw>
Link: http://lkml.kernel.org/r/20200508183105.225460-13-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: switch to native NR_ANON_THPS counter

With rmap memcg locking already in place for NR_ANON_MAPPED, it's just a
small step to remove the MEMCG_RSS_HUGE wart and switch memcg to the
native NR_ANON_THPS accounting sites.

[hannes@cmpxchg.org: fixes]
Link: http://lkml.kernel.org/r/20200512121750.GA397968@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org> [build-tested]
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-12-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: switch to native NR_ANON_MAPPED counter

Memcg maintains a private MEMCG_RSS counter.  This divergence from the
generic VM accounting means unnecessary code overhead, and creates a
dependency for memcg that page->mapping is set up at the time of charging,
so that page types can be told apart.

Convert the generic accounting sites to mod_lruvec_page_state and friends
to maintain the per-cgroup vmstat counter of NR_ANON_MAPPED.  We use
lock_page_memcg() to stabilize page->mem_cgroup during rmap changes, the
same way we do for NR_FILE_MAPPED.

With the previous patch removing MEMCG_CACHE and the private NR_SHMEM
counter, this patch finally eliminates the need to have page->mapping set
up at charge time.  However, we need to have page->mem_cgroup set up by
the time rmap runs and does the accounting, so switch the commit and the
rmap callbacks around.

v2: fix temporary accounting bug by switching rmap<->commit (Joonsoo)

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-11-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: switch to native NR_FILE_PAGES and NR_SHMEM counters

Memcg maintains private MEMCG_CACHE and NR_SHMEM counters.  This
divergence from the generic VM accounting means unnecessary code overhead,
and creates a dependency for memcg that page->mapping is set up at the
time of charging, so that page types can be told apart.

Convert the generic accounting sites to mod_lruvec_page_state and friends
to maintain the per-cgroup vmstat counters of NR_FILE_PAGES and NR_SHMEM.
The page is already locked in these places, so page->mem_cgroup is stable;
we only need minimal tweaks of two mem_cgroup_migrate() calls to ensure
it's set up in time.

Then replace MEMCG_CACHE with NR_FILE_PAGES and delete the private
NR_SHMEM accounting sites.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-10-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: prepare cgroup vmstat infrastructure for native anon counters

Anonymous compound pages can be mapped by ptes, which means that if we
want to track NR_MAPPED_ANON, NR_ANON_THPS on a per-cgroup basis, we have
to be prepared to see tail pages in our accounting functions.

Make mod_lruvec_page_state() and lock_page_memcg() deal with tail pages
correctly, namely by redirecting to the head page which has the
page->mem_cgroup set up.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-9-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: prepare move_account for removal of private page type counters

When memcg uses the generic vmstat counters, it doesn't need to do
anything at charging and uncharging time.  It does, however, need to
migrate counts when pages move to a different cgroup in move_account.

Prepare the move_account function for the arrival of NR_FILE_PAGES,
NR_ANON_MAPPED, NR_ANON_THPS etc.  by having a branch for files and a
branch for anon, which can then divided into sub-branches.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-8-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: prepare uncharging for removal of private page type counters

The uncharge batching code adds up the anon, file, kmem counts to
determine the total number of pages to uncharge and references to drop.
But the next patches will remove the anon and file counters.

Maintain an aggregate nr_pages in the uncharge_gather struct.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-7-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: convert page cache to a new mem_cgroup_charge() API

The try/commit/cancel protocol that memcg uses dates back to when pages
used to be uncharged upon removal from the page cache, and thus couldn't
be committed before the insertion had succeeded.  Nowadays, pages are
uncharged when they are physically freed; it doesn't matter whether the
insertion was successful or not.  For the page cache, the transaction
dance has become unnecessary.

Introduce a mem_cgroup_charge() function that simply charges a newly
allocated page to a cgroup and sets up page->mem_cgroup in one single
step.  If the insertion fails, the caller doesn't have to do anything but
free/put the page.

Then switch the page cache over to this new API.

Subsequent patches will also convert anon pages, but it needs a bit more
prep work.  Right now, memcg depends on page->mapping being already set up
at the time of charging, so that it can maintain its own MEMCG_CACHE and
MEMCG_RSS counters.  For anon, page->mapping is set under the same pte
lock under which the page is publishd, so a single charge point that can
block doesn't work there just yet.

The following prep patches will replace the private memcg counters with
the generic vmstat counters, thus removing the page->mapping dependency,
then complete the transition to the new single-point charge API and delete
the old transactional scheme.

v2: leave shmem swapcache when charging fails to avoid double IO (Joonsoo)
v3: rebase on preceeding shmem simplification patch

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-6-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: move out cgroup swaprate throttling

The cgroup swaprate throttling is about matching new anon allocations to
the rate of available IO when that is being throttled.  It's the io
controller hooking into the VM, rather than a memory controller thing.

Rename mem_cgroup_throttle_swaprate() to cgroup_throttle_swaprate(), and
drop the @memcg argument which is only used to check whether the preceding
page charge has succeeded and the fault is proceeding.

We could decouple the call from mem_cgroup_try_charge() here as well, but
that would cause unnecessary churn: the following patches convert all
callsites to a new charge API and we'll decouple as we go along.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-5-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: shmem: remove rare optimization when swapin races with hole punching

Commit 215c02bc33bb ("tmpfs: fix shmem_getpage_gfp() VM_BUG_ON")
recognized that hole punching can race with swapin and removed the
BUG_ON() for a truncated entry from the swapin path.

The patch also added a swapcache deletion to optimize this rare case:
Since swapin has the page locked, and free_swap_and_cache() merely
trylocks, this situation can leave the page stranded in swapcache.
Usually, page reclaim picks up stale swapcache pages, and the race can
happen at any other time when the page is locked.  (The same happens for
non-shmem swapin racing with page table zapping.) The thinking here was:
we already observed the race and we have the page locked, we may as well
do the cleanup instead of waiting for reclaim.

However, this optimization complicates the next patch which moves the
cgroup charging code around.  As this is just a minor speedup for a race
condition that is so rare that it required a fuzzer to trigger the
original BUG_ON(), it's no longer worth the complications.

Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200511181056.GA339505@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: drop @compound parameter from memcg charging API

The memcg charging API carries a boolean @compound parameter that tells
whether the page we're dealing with is a hugepage.
mem_cgroup_commit_charge() has another boolean @lrucare that indicates
whether the page needs LRU locking or not while charging.  The majority of
callsites know those parameters at compile time, which results in a lot of
naked "false, false" argument lists.  This makes for cryptic code and is a
breeding ground for subtle mistakes.

Thankfully, the huge page state can be inferred from the page itself and
doesn't need to be passed along.  This is safe because charging completes
before the page is published and somebody may split it.

Simplify the callsites by removing @compound, and let memcg infer the
state by using hpage_nr_pages() unconditionally.  That function does
PageTransHuge() to identify huge pages, which also helpfully asserts that
nobody passes in tail pages by accident.

The following patches will introduce a new charging API, best not to carry
over unnecessary weight.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-4-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: fix stat-corrupting race in charge moving

The move_lock is a per-memcg lock, but the VM accounting code that needs
to acquire it comes from the page and follows page->mem_cgroup under RCU
protection.  That means that the page becomes unlocked not when we drop
the move_lock, but when we update page->mem_cgroup.  And that assignment
doesn't imply any memory ordering.  If that pointer write gets reordered
against the reads of the page state - page_mapped, PageDirty etc.  the
state may change while we rely on it being stable and we can end up
corrupting the counters.

Place an SMP memory barrier to make sure we're done with all page state by
the time the new page->mem_cgroup becomes visible.

Also replace the open-coded move_lock with a lock_page_memcg() to make it
more obvious what we're serializing against.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-3-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix NUMA node file count error in replace_page_cache()

Patch series "mm: memcontrol: charge swapin pages on instantiation", v2.

This patch series reworks memcg to charge swapin pages directly at
swapin time, rather than at fault time, which may be much later, or
not happen at all.

Changes in version 2:
- prevent double charges on pre-allocated hugepages in khugepaged
- leave shmem swapcache when charging fails to avoid double IO (Joonsoo)
- fix temporary accounting bug by switching rmap<->commit (Joonsoo)
- fix double swap charge bug in cgroup1/cgroup2 code gating
- simplify swapin error checking (Joonsoo)
- mm: memcontrol: document the new swap control behavior (Alex)
- review tags

The delayed swapin charging scheme we have right now causes problems:

- Alex's per-cgroup lru_lock patches rely on pages that have been
  isolated from the LRU to have a stable page->mem_cgroup; otherwise
  the lock may change underneath him. Swapcache pages are charged only
  after they are added to the LRU, and charging doesn't follow the LRU
  isolation protocol.

- Joonsoo's anon workingset patches need a suitable LRU at the time
  the page enters the swap cache and displaces the non-resident
  info. But the correct LRU is only available after charging.

- It's a containment hole / DoS vector. Users can trigger arbitrarily
  large swap readahead using MADV_WILLNEED. The memory is never
  charged unless somebody actually touches it.

- It complicates the page->mem_cgroup stabilization rules

In order to charge pages directly at swapin time, the memcg code base
needs to be prepared, and several overdue cleanups become a necessity:

To charge pages at swapin time, we need to always have cgroup
ownership tracking of swap records. We also cannot rely on
page->mapping to tell apart page types at charge time, because that's
only set up during a page fault.

To eliminate the page->mapping dependency, memcg needs to ditch its
private page type counters (MEMCG_CACHE, MEMCG_RSS, NR_SHMEM) in favor
of the generic vmstat counters and accounting sites, such as
NR_FILE_PAGES, NR_ANON_MAPPED etc.

To switch to generic vmstat counters, the charge sequence must be
adjusted such that page->mem_cgroup is set up by the time these
counters are modified.

The series is structured as follows:

1. Bug fixes
2. Decoupling charging from rmap
3. Swap controller integration into memcg
4. Direct swapin charging

This patch (of 19):

When replacing one page with another one in the cache, we have to decrease
the file count of the old page's NUMA node and increase the one of the new
NUMA node, otherwise the old node leaks the count and the new node
eventually underflows its counter.

Fixes: 74d609585d8b ("page cache: Add and replace pages using the XArray")
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/20200508183105.225460-1-hannes@cmpxchg.org
Link: http://lkml.kernel.org/r/20200508183105.225460-2-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: update the comment of should_continue_reclaim()

try_to_compact_zone() has been replaced by try_to_compact_pages(), which
is necessary to be updated in the comment of should_continue_reclaim().

Signed-off-by: Qiwu Chen <chenqiwu@xiaomi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200501034907.22991-1-chenqiwu@xiaomi.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: change prototype for shrink_page_list

commit 3c710c1ad11b ("mm, vmscan extract shrink_page_list reclaim counters
into a struct") changed data type for the function, so changing return
type for funciton and its caller.

Signed-off-by: Vaneet Narang <v.narang@samsung.com>
Signed-off-by: Maninder Singh <maninder1.s@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Amit Sahrawat <a.sahrawat@samsung.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/1588168259-25604-1-git-send-email-maninder1.s@samsung.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: count layzfree pages and fix nr_isolated_* mismatch

Fix an nr_isolate_* mismatch problem between cma and dirty lazyfree pages.

If try_to_unmap_one is used for reclaim and it detects a dirty lazyfree
page, then the lazyfree page is changed to a normal anon page having
SwapBacked by commit 802a3a92ad7a ("mm: reclaim MADV_FREE pages").  Even
with the change, reclaim context correctly counts isolated files because
it uses is_file_lru to distinguish file.  And the change to anon is not
happened if try_to_unmap_one is used for migration.  So migration context
like compaction also correctly counts isolated files even though it uses
page_is_file_lru insted of is_file_lru.  Recently page_is_file_cache was
renamed to page_is_file_lru by commit 9de4f22a60f7 ("mm: code cleanup for
MADV_FREE").

But the nr_isolate_* mismatch problem happens on cma alloc.  There is
reclaim_clean_pages_from_list which is being used only by cma.  It was
introduced by commit 02c6de8d757c ("mm: cma: discard clean pages during
contiguous allocation instead of migration") to reclaim clean file pages
without migration.  The cma alloc uses both reclaim_clean_pages_from_list
and migrate_pages, and it uses page_is_file_lru to count isolated files.
If there are dirty lazyfree pages allocated from cma memory region, the
pages are counted as isolated file at the beginging but are counted as
isolated anon after finished.

Mem-Info:
Node 0 active_anon:3045904kB inactive_anon:611448kB active_file:14892kB inactive_file:205636kB unevictable:10416kB isolated(anon):0kB isolated(file):37664kB mapped:630216kB dirty:384kB writeback:0kB shmem:42576kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no

Like log above, there were too much isolated files, 37664kB, which
triggers too_many_isolated in reclaim even when there is no actually
isolated file in system wide.  It could be reproducible by running two
programs, writing on MADV_FREE page and doing cma alloc, respectively.
Although isolated anon is 0, I found that the internal value of isolated
anon was the negative value of isolated file.

Fix this by compensating the isolated count for both LRU lists.  Count
non-discarded lazyfree pages in shrink_page_list, then compensate the
counted number in reclaim_clean_pages_from_list.

Reported-by: Yong-Taek Lee <ytk.lee@samsung.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Shaohua Li <shli@fb.com>
Link: http://lkml.kernel.org/r/20200426011718.30246-1-jaewon31.kim@samsung.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: use update_lru_size() in update_lru_sizes()

We already defined the helper update_lru_size().

Let's use this to reduce code duplication.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200331221550.1011-1-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: simplify calling a compound page destructor

None of the three callers of get_compound_page_dtor() want to know the
value; they just want to call the function.  Replace it with
destroy_compound_page() which calls the dtor for them.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Link: http://lkml.kernel.org/r/20200517105051.9352-1-willy@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: define a generic fallback for arch_clear_hugepage_flags()

There are multiple similar definitions for arch_clear_hugepage_flags() on
various platforms.  Lets just add it's generic fallback definition for
platforms that do not override.  This help reduce code duplication.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Rich Felker <dalias@libc.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Link: http://lkml.kernel.org/r/1588907271-11920-4-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: define a generic fallback for is_hugepage_only_range()

There are multiple similar definitions for is_hugepage_only_range() on
various platforms.  Lets just add it's generic fallback definition for
platforms that do not override.  This help reduce code duplication.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Rich Felker <dalias@libc.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Link: http://lkml.kernel.org/r/1588907271-11920-3-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arm64/mm: drop __HAVE_ARCH_HUGE_PTEP_GET

Patch series "mm/hugetlb: Add some new generic fallbacks", v3.

This series adds the following new generic fallbacks.  Before that it
drops __HAVE_ARCH_HUGE_PTEP_GET from arm64 platform.

1. is_hugepage_only_range()
2. arch_clear_hugepage_flags()

After this arm (32 bit) remains the sole platform defining it's own
huge_ptep_get() via __HAVE_ARCH_HUGE_PTEP_GET.

This patch (of 3):

Platform specific huge_ptep_get() is required only when fetching the huge
PTE involves more than just dereferencing the page table pointer.  This is
not the case on arm64 platform.  Hence huge_ptep_pte() can be dropped
along with it's __HAVE_ARCH_HUGE_PTEP_GET subscription.  Before that, it
updates the generic huge_ptep_get() with READ_ONCE() which will prevent
known page table issues with THP on arm64.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Link: http://lkml.kernel.org/r/1588907271-11920-1-git-send-email-anshuman.khandual@arm.com
Link: http://lkml.kernel.org/r//1506527369-19535-1-git-send-email-will.deacon@arm.com/
Link: http://lkml.kernel.org/r/1588907271-11920-2-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: avoid unnecessary check on pud and pmd entry in huge_pte_offset

When huge_pte_offset() is called, the parameter sz can only be PUD_SIZE or
PMD_SIZE.  If sz is PUD_SIZE and code can reach pud, then *pud must be
none, or normal hugetlb entry, or non-present (migration or hwpoisoned)
hugetlb entry, and we can directly return pud.  When sz is PMD_SIZE, pud
must be none or present, and if code can reach pmd, we can directly return
pmd.

So after this patch the code is simplified by first check on the parameter
sz, and avoid unnecessary checks in current code.  Same semantics of
existing code is maintained.

More details about relevant commits:
commit 9b19df292c66 ("mm/hugetlb.c: make huge_pte_offset() consistent
and document behaviour") changed the code path for pud and pmd handling,
see comments about why this patch intends to change it.
...
pud = pud_offset(p4d, addr);
if (sz != PUD_SIZE && pud_none(*pud)) // [1]
return NULL;
/* hugepage or swap? */
if (pud_huge(*pud) || !pud_present(*pud)) // [2]
return (pte_t *)pud;

pmd = pmd_offset(pud, addr);
if (sz != PMD_SIZE && pmd_none(*pmd)) // [3]
return NULL;
/* hugepage or swap? */
if (pmd_huge(*pmd) || !pmd_present(*pmd)) // [4]
return (pte_t *)pmd;

return NULL; // [5]
...
[1]: this is necessary, return NULL for sz == PMD_SIZE;
[2]: if sz == PUD_SIZE, all valid values of pud entry will cause return;
[3]: dead code, sz != PMD_SIZE never true;
[4]: all valid values of pmd entry will cause return;
[5]: dead code, because of check in [4].

Now, this patch combines [1] and [2] for pud, and combines [3], [4] and
[5] for pmd, so avoid unnecessary checks.

I don't try to catch any invalid values in page table entry, as that will
be checked by caller and avoid extra branch in this function.  Also no
assert on sz must equal PUD_SIZE or PMD_SIZE, since this function only
call for hugetlb mapping.

For commit 3c1d7e6ccb64 ("mm/hugetlb: fix a addressing exception caused by
huge_pte_offset"), since we don't read the entry more than once now,
variable pud_entry and pmd_entry are not needed.

Signed-off-by: Li Xinhai <lixinhai.lxh@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Punit Agrawal <punit.agrawal@arm.com>
Cc: Longpeng <longpeng2@huawei.com>
Link: http://lkml.kernel.org/r/1587794313-16849-1-git-send-email-lixinhai.lxh@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: fix changes to command line processing

Previously, a check for hugepages_supported was added before processing
hugetlb command line parameters.  On some architectures such as powerpc,
hugepages_supported() is not set to true until after command line
processing.  Therefore, no hugetlb command line parameters would be
accepted.

Remove the additional checks for hugepages_supported.  In hugetlb_init,
print a warning if !hugepages_supported and command line parameters were
specified.

Reported-by: Sandipan Das <sandipan.osd@gmail.com>
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/b1f04f9f-fa46-c2a0-7693-4a0679d2a1ee@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: clean up command line processing

With all hugetlb page processing done in a single file clean up code.

- Make code match desired semantics
  - Update documentation with semantics
- Make all warnings and errors messages start with 'HugeTLB:'.
- Consistently name command line parsing routines.
- Warn if !hugepages_supported() and command line parameters have
  been specified.
- Add comments to code
  - Describe some of the subtle interactions
  - Describe semantics of command line arguments

This patch also fixes issues with implicitly setting the number of
gigantic huge pages to preallocate.  Previously on X86 command line,

        hugepages=2 default_hugepagesz=1G

would result in zero 1G pages being preallocated and,

        # grep HugePages_Total /proc/meminfo
        HugePages_Total:       0
        # sysctl -a | grep nr_hugepages
        vm.nr_hugepages = 2
        vm.nr_hugepages_mempolicy = 2
        # cat /proc/sys/vm/nr_hugepages
        2

After this patch 2 gigantic pages will be preallocated and all the proc,
sysfs, sysctl and meminfo files will accurately reflect this.

To address the issue with gigantic pages, a small change in behavior was
made to command line processing.  Previously the command line,

        hugepages=128 default_hugepagesz=2M hugepagesz=2M hugepages=256

would result in the allocation of 256 2M huge pages.  The value 128 would
be ignored without any warning.  After this patch, 128 2M pages will be
allocated and a warning message will be displayed indicating the value of
256 is ignored.  This change in behavior is required because allocation of
implicitly specified gigantic pages must be done when the
default_hugepagesz= is encountered for gigantic pages.  Previously the
code waited until later in the boot process (hugetlb_init), to allocate
pages of default size.  However the bootmem allocator required for
gigantic allocations is not available at this time.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Sandipan Das <sandipan@linux.ibm.com>
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Will Deacon <will@kernel.org>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Longpeng <longpeng2@huawei.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Nitesh Narayan Lal <nitesh@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Anders Roxell <anders.roxell@linaro.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/20200417185049.275845-5-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: remove hugetlb_add_hstate() warning for existing hstate

hugetlb_add_hstate() prints a warning if the hstate already exists.  This
was originally done as part of kernel command line parsing.  If
'hugepagesz=' was specified more than once, the warning

pr_warn("hugepagesz= specified twice, ignoring\n");

would be printed.

Some architectures want to enable all huge page sizes.  They would call
hugetlb_add_hstate for all supported sizes.  However, this was done after
command line processing and as a result hstates could have already been
created for some sizes.  To make sure no warning were printed, there would
often be code like:

if (!size_to_hstate(size)
hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT)

The only time we want to print the warning is as the result of command
line processing.  So, remove the warning from hugetlb_add_hstate and add
it to the single arch independent routine processing "hugepagesz=".  After
this, calls to size_to_hstate() in arch specific code can be removed and
hugetlb_add_hstate can be called without worrying about warning messages.

[mike.kravetz@oracle.com: fix hugetlb initialization]
Link: http://lkml.kernel.org/r/4c36c6ce-3774-78fa-abc4-b7346bf24348@oracle.com
Link: http://lkml.kernel.org/r/20200428205614.246260-5-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Anders Roxell <anders.roxell@linaro.org>
Acked-by: Mina Almasry <almasrymina@google.com>
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Will Deacon <will@kernel.org>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Longpeng <longpeng2@huawei.com>
Cc: Nitesh Narayan Lal <nitesh@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/20200417185049.275845-4-mike.kravetz@oracle.com
Link: http://lkml.kernel.org/r/20200428205614.246260-4-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: move hugepagesz= parsing to arch independent code

Now that architectures provide arch_hugetlb_valid_size(), parsing of
"hugepagesz=" can be done in architecture independent code.  Create a
single routine to handle hugepagesz= parsing and remove all arch specific
routines.  We can also remove the interface hugetlb_bad_size() as this is
no longer used outside arch independent code.

This also provides consistent behavior of hugetlbfs command line options.
The hugepagesz= option should only be specified once for a specific size,
but some architectures allow multiple instances.  This appears to be more
of an oversight when code was added by some architectures to set up ALL
huge pages sizes.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Sandipan Das <sandipan@linux.ibm.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Acked-by: Mina Almasry <almasrymina@google.com>
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Will Deacon <will@kernel.org>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Longpeng <longpeng2@huawei.com>
Cc: Nitesh Narayan Lal <nitesh@redhat.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Anders Roxell <anders.roxell@linaro.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/20200417185049.275845-3-mike.kravetz@oracle.com
Link: http://lkml.kernel.org/r/20200428205614.246260-3-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: add arch_hugetlb_valid_size

Patch series "Clean up hugetlb boot command line processing", v4.

Longpeng(Mike) reported a weird message from hugetlb command line
processing and proposed a solution [1].  While the proposed patch does
address the specific issue, there are other related issues in command line
processing.  As hugetlbfs evolved, updates to command line processing have
been made to meet immediate needs and not necessarily in a coordinated
manner.  The result is that some processing is done in arch specific code,
some is done in arch independent code and coordination is problematic.
Semantics can vary between architectures.

The patch series does the following:
- Define arch specific arch_hugetlb_valid_size routine used to validate
  passed huge page sizes.
- Move hugepagesz= command line parsing out of arch specific code and into
  an arch independent routine.
- Clean up command line processing to follow desired semantics and
  document those semantics.

[1] https://lore.kernel.org/linux-mm/20200305033014.1152-1-longpeng2@huawei.com

This patch (of 3):

The architecture independent routine hugetlb_default_setup sets up the
default huge pages size.  It has no way to verify if the passed value is
valid, so it accepts it and attempts to validate at a later time.  This
requires undocumented cooperation between the arch specific and arch
independent code.

For architectures that support more than one huge page size, provide a
routine arch_hugetlb_valid_size to validate a huge page size.
hugetlb_default_setup can use this to validate passed values.

arch_hugetlb_valid_size will also be used in a subsequent patch to move
processing of the "hugepagesz=" in arch specific code to a common routine
in arch independent code.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Will Deacon <will@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Longpeng <longpeng2@huawei.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Nitesh Narayan Lal <nitesh@redhat.com>
Cc: Anders Roxell <anders.roxell@linaro.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/20200428205614.246260-1-mike.kravetz@oracle.com
Link: http://lkml.kernel.org/r/20200428205614.246260-2-mike.kravetz@oracle.com
Link: http://lkml.kernel.org/r/20200417185049.275845-1-mike.kravetz@oracle.com
Link: http://lkml.kernel.org/r/20200417185049.275845-2-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: introduce 'max_ptes_shared' tunable

'max_ptes_shared' specifies how many pages can be shared across multiple
processes.  Exceeding the number would block the collapse::

/sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_shared

A higher value may increase memory footprint for some workloads.

By default, at least half of pages has to be not shared.

[colin.king@canonical.com: fix several spelling mistakes]
Link: http://lkml.kernel.org/r/20200420084241.65433-1-colin.king@canonical.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-9-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: change CoW semantics for anon-THP

Currently we have different copy-on-write semantics for anon- and
file-THP.  For anon-THP we try to allocate huge page on the write fault,
but on file-THP we split PMD and allocate 4k page.

Arguably, file-THP semantics is more desirable: we don't necessary want to
unshare full PMD range from the parent on the first access.  This is the
primary reason THP is unusable for some workloads, like Redis.

The original THP refcounting didn't allow to have PTE-mapped compound
pages, so we had no options, but to allocate huge page on CoW (with
fallback to 512 4k pages).

The current refcounting doesn't have such limitations and we can cut a lot
of complex code out of fault path.

khugepaged is now able to recover THP from such ranges if the
configuration allows.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-8-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: allow to collapse PTE-mapped compound pages

We can collapse PTE-mapped compound pages.  We only need to avoid handling
them more than once: lock/unlock page only once if it's present in the PMD
range multiple times as it handled on compound level.  The same goes for
LRU isolation and putback.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-7-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: allow to collapse a page shared across fork

The page can be included into collapse as long as it doesn't have extra
pins (from GUP or otherwise).

Logic to check the refcount is moved to a separate function.  For pages in
swap cache, add compound_nr(page) to the expected refcount, in order to
handle the compound page case.  This is in preparation for the following
patch.

VM_BUG_ON_PAGE() was removed from __collapse_huge_page_copy() as the
invariant it checks is no longer valid: the source can be mapped multiple
times now.

[yang.shi@linux.alibaba.com: remove error message when checking external pins]
Link: http://lkml.kernel.org/r/1589317383-9595-1-git-send-email-yang.shi@linux.alibaba.com
[cai@lca.pw: fix set-but-not-used warning]
Link: http://lkml.kernel.org/r/20200521145644.GA6367@ovpn-112-192.phx2.redhat.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-6-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: drain LRU add pagevec after swapin

collapse_huge_page() tries to swap in pages that are part of the PMD
range.  Just swapped in page goes though LRU add cache.  The cache gets
extra reference on the page.

The extra reference can lead to the collapse fail: the following
__collapse_huge_page_isolate() would check refcount and abort collapse
seeing unexpected refcount.

The fix is to drain local LRU add cache in
__collapse_huge_page_swapin() if we successfully swapped in any pages.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-5-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: drain all LRU caches before scanning pages

Having a page in LRU add cache offsets page refcount and gives
false-negative on PageLRU().  It reduces collapse success rate.

Drain all LRU add caches before scanning.  It happens relatively rare and
should not disturb the system too much.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-4-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: do not stop collapse if less than half PTEs are referenced

__collapse_huge_page_swapin() checks the number of referenced PTE to
decide if the memory range is hot enough to justify swapin.

We have few problems with the approach:

 - It is way too late: we can do the check much earlier and safe time.
   khugepaged_scan_pmd() already knows if we have any pages to swap in
   and number of referenced page.

 - It stops collapse altogether if there's not enough referenced pages,
   not only swappingin.

Fix it by making the right check early. We also can avoid additional
page table scanning if khugepaged_scan_pmd() haven't found any swap
entries.

Fixes: 0db501f7a34c ("mm, thp: convert from optimistic swapin collapsing to conservative")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-3-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: add self test

Patch series "thp/khugepaged improvements and CoW semantics", v4.

The patchset adds khugepaged selftest (anon-THP only for now), expands
cases khugepaged can handle and switches anon-THP copy-on-write handling
to 4k.

This patch (of 8):

The test checks if khugepaged is able to recover huge page where we expect
to do so.  It only covers anon-THP for now.

Currently the test shows few failures.  They are going to be addressed by
the following patches.

[colin.king@canonical.com: fix several spelling mistakes]
Link: http://lkml.kernel.org/r/20200420084241.65433-1-colin.king@canonical.com
[aneesh.kumar@linux.ibm.com: replace the usage of system(3) in the test]
Link: http://lkml.kernel.org/r/20200429110727.89388-1-aneesh.kumar@linux.ibm.com
[kirill@shutemov.name: fixup for issues I've noticed]
Link: http://lkml.kernel.org/r/20200429124816.jp272trghrzxx5j5@box
[jhubbard@nvidia.com: add khugepaged to .gitignore]
Link: http://lkml.kernel.org/r/20200517002509.362401-1-jhubbard@nvidia.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Link: http://lkml.kernel.org/r/20200416160026.16538-1-kirill.shutemov@linux.intel.com
Link: http://lkml.kernel.org/r/20200416160026.16538-2-kirill.shutemov@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: add missing newline

Add missing line breaks on pr_warn().

Signed-off-by: Chen Tao <chentao107@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200603063547.235825-1-chentao107@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

padata: document multithreaded jobs

Add Documentation for multithreaded jobs.

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-9-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make deferred init's max threads arch-specific

Using padata during deferred init has only been tested on x86, so for now
limit it to this architecture.

If another arch wants this, it can find the max thread limit that's best
for it and override deferred_page_init_max_threads().

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-8-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: parallelize deferred_init_memmap()

Deferred struct page init is a significant bottleneck in kernel boot.
Optimizing it maximizes availability for large-memory systems and allows
spinning up short-lived VMs as needed without having to leave them
running.  It also benefits bare metal machines hosting VMs that are
sensitive to downtime.  In projects such as VMM Fast Restart[1], where
guest state is preserved across kexec reboot, it helps prevent application
and network timeouts in the guests.

Multithread to take full advantage of system memory bandwidth.

The maximum number of threads is capped at the number of CPUs on the node
because speedups always improve with additional threads on every system
tested, and at this phase of boot, the system is otherwise idle and
waiting on page init to finish.

Helper threads operate on section-aligned ranges to both avoid false
sharing when setting the pageblock's migrate type and to avoid accessing
uninitialized buddy pages, though max order alignment is enough for the
latter.

The minimum chunk size is also a section.  There was benefit to using
multiple threads even on relatively small memory (1G) systems, and this is
the smallest size that the alignment allows.

The time (milliseconds) is the slowest node to initialize since boot
blocks until all nodes finish.  intel_pstate is loaded in active mode
without hwp and with turbo enabled, and intel_idle is active as well.

    Intel(R) Xeon(R) Platinum 8167M CPU @ 2.00GHz (Skylake, bare metal)
      2 nodes * 26 cores * 2 threads = 104 CPUs
      384G/node = 768G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --   4089.7 (  8.1)         --   1785.7 (  7.6)
       2% (  1)       1.7%   4019.3 (  1.5)       3.8%   1717.7 ( 11.8)
      12% (  6)      34.9%   2662.7 (  2.9)      79.9%    359.3 (  0.6)
      25% ( 13)      39.9%   2459.0 (  3.6)      91.2%    157.0 (  0.0)
      37% ( 19)      39.2%   2485.0 ( 29.7)      90.4%    172.0 ( 28.6)
      50% ( 26)      39.3%   2482.7 ( 25.7)      90.3%    173.7 ( 30.0)
      75% ( 39)      39.0%   2495.7 (  5.5)      89.4%    190.0 (  1.0)
     100% ( 52)      40.2%   2443.7 (  3.8)      92.3%    138.0 (  1.0)

    Intel(R) Xeon(R) CPU E5-2699C v4 @ 2.20GHz (Broadwell, kvm guest)
      1 node * 16 cores * 2 threads = 32 CPUs
      192G/node = 192G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --   1988.7 (  9.6)         --   1096.0 ( 11.5)
       3% (  1)       1.1%   1967.0 ( 17.6)       0.3%   1092.7 ( 11.0)
      12% (  4)      41.1%   1170.3 ( 14.2)      73.8%    287.0 (  3.6)
      25% (  8)      47.1%   1052.7 ( 21.9)      83.9%    177.0 ( 13.5)
      38% ( 12)      48.9%   1016.3 ( 12.1)      86.8%    144.7 (  1.5)
      50% ( 16)      48.9%   1015.7 (  8.1)      87.8%    134.0 (  4.4)
      75% ( 24)      49.1%   1012.3 (  3.1)      88.1%    130.3 (  2.3)
     100% ( 32)      49.5%   1004.0 (  5.3)      88.5%    125.7 (  2.1)

    Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz (Haswell, bare metal)
      2 nodes * 18 cores * 2 threads = 72 CPUs
      128G/node = 256G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --   1680.0 (  4.6)         --    627.0 (  4.0)
       3% (  1)       0.3%   1675.7 (  4.5)      -0.2%    628.0 (  3.6)
      11% (  4)      25.6%   1250.7 (  2.1)      67.9%    201.0 (  0.0)
      25% (  9)      30.7%   1164.0 ( 17.3)      81.8%    114.3 ( 17.7)
      36% ( 13)      31.4%   1152.7 ( 10.8)      84.0%    100.3 ( 17.9)
      50% ( 18)      31.5%   1150.7 (  9.3)      83.9%    101.0 ( 14.1)
      75% ( 27)      31.7%   1148.0 (  5.6)      84.5%     97.3 (  6.4)
     100% ( 36)      32.0%   1142.3 (  4.0)      85.6%     90.0 (  1.0)

    AMD EPYC 7551 32-Core Processor (Zen, kvm guest)
      1 node * 8 cores * 2 threads = 16 CPUs
      64G/node = 64G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --   1029.3 ( 25.1)         --    240.7 (  1.5)
       6% (  1)      -0.6%   1036.0 (  7.8)      -2.2%    246.0 (  0.0)
      12% (  2)      11.8%    907.7 (  8.6)      44.7%    133.0 (  1.0)
      25% (  4)      13.9%    886.0 ( 10.6)      62.6%     90.0 (  6.0)
      38% (  6)      17.8%    845.7 ( 14.2)      69.1%     74.3 (  3.8)
      50% (  8)      16.8%    856.0 ( 22.1)      72.9%     65.3 (  5.7)
      75% ( 12)      15.4%    871.0 ( 29.2)      79.8%     48.7 (  7.4)
     100% ( 16)      21.0%    813.7 ( 21.0)      80.5%     47.0 (  5.2)

Server-oriented distros that enable deferred page init sometimes run in
small VMs, and they still benefit even though the fraction of boot time
saved is smaller:

    AMD EPYC 7551 32-Core Processor (Zen, kvm guest)
      1 node * 2 cores * 2 threads = 4 CPUs
      16G/node = 16G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --    716.0 ( 14.0)         --     49.7 (  0.6)
      25% (  1)       1.8%    703.0 (  5.3)      -4.0%     51.7 (  0.6)
      50% (  2)       1.6%    704.7 (  1.2)      43.0%     28.3 (  0.6)
      75% (  3)       2.7%    696.7 ( 13.1)      49.7%     25.0 (  0.0)
     100% (  4)       4.1%    687.0 ( 10.4)      55.7%     22.0 (  0.0)

    Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz (Haswell, kvm guest)
      1 node * 2 cores * 2 threads = 4 CPUs
      14G/node = 14G memory

                   kernel boot                 deferred init
                   ------------------------    ------------------------
    node% (thr)    speedup  time_ms (stdev)    speedup  time_ms (stdev)
          (  0)         --    787.7 (  6.4)         --    122.3 (  0.6)
      25% (  1)       0.2%    786.3 ( 10.8)      -2.5%    125.3 (  2.1)
      50% (  2)       5.9%    741.0 ( 13.9)      37.6%     76.3 ( 19.7)
      75% (  3)       8.3%    722.0 ( 19.0)      49.9%     61.3 (  3.2)
     100% (  4)       9.3%    714.7 (  9.5)      56.4%     53.3 (  1.5)

On Josh's 96-CPU and 192G memory system:

    Without this patch series:
    [    0.487132] node 0 initialised, 23398907 pages in 292ms
    [    0.499132] node 1 initialised, 24189223 pages in 304ms
    ...
    [    0.629376] Run /sbin/init as init process

    With this patch series:
    [    0.231435] node 1 initialised, 24189223 pages in 32ms
    [    0.236718] node 0 initialised, 23398907 pages in 36ms

[1] https://static.sched.com/hosted_files/kvmforum2019/66/VMM-fast-restart_kvmforum2019.pdf

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-7-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: don't track number of pages during deferred initialization

Deferred page init used to report the number of pages initialized:

  node 0 initialised, 32439114 pages in 97ms

Tracking this makes the code more complicated when using multiple threads.
Given that the statistic probably has limited value, especially since a
zone grows on demand so that the page count can vary, just remove it.

The boot message now looks like

  node 0 deferred pages initialised in 97ms

Suggested-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-6-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

padata: add basic support for multithreaded jobs

Sometimes the kernel doesn't take full advantage of system memory
bandwidth, leading to a single CPU spending excessive time in
initialization paths where the data scales with memory size.

Multithreading naturally addresses this problem.

Extend padata, a framework that handles many parallel yet singlethreaded
jobs, to also handle multithreaded jobs by adding support for splitting up
the work evenly, specifying a minimum amount of work that's appropriate
for one helper thread to do, load balancing between helpers, and
coordinating them.

This is inspired by work from Pavel Tatashin and Steve Sistare.

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

padata: allocate work structures for parallel jobs from a pool

padata allocates per-CPU, per-instance work structs for parallel jobs.  A
do_parallel call assigns a job to a sequence number and hashes the number
to a CPU, where the job will eventually run using the corresponding work.

This approach fit with how padata used to bind a job to each CPU
round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind
parallel jobs from specific CPUs") because a work isn't bound to a
particular CPU anymore, and isn't needed at all for multithreaded jobs
because they don't have sequence numbers.

Replace the per-CPU works with a preallocated pool, which allows sharing
them between existing padata users and the upcoming multithreaded user.
The pool will also facilitate setting NUMA-aware concurrency limits with
later users.

The pool is sized according to the number of possible CPUs.  With this
limit, MAX_OBJ_NUM no longer makes sense, so remove it.

If the global pool is exhausted, a parallel job is run in the current task
instead to throttle a system trying to do too much in parallel.

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

padata: initialize earlier

padata will soon initialize the system's struct pages in parallel, so it
needs to be ready by page_alloc_init_late().

The error return from padata_driver_init() triggers an initcall warning,
so add a warning to padata_init() to avoid silent failure.

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-3-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

padata: remove exit routine

Patch series "padata: parallelize deferred page init", v3.

Deferred struct page init is a bottleneck in kernel boot--the biggest for
us and probably others.  Optimizing it maximizes availability for
large-memory systems and allows spinning up short-lived VMs as needed
without having to leave them running.  It also benefits bare metal
machines hosting VMs that are sensitive to downtime.  In projects such as
VMM Fast Restart[1], where guest state is preserved across kexec reboot,
it helps prevent application and network timeouts in the guests.

So, multithread deferred init to take full advantage of system memory
bandwidth.

Extend padata, a framework that handles many parallel singlethreaded jobs,
to handle multithreaded jobs as well by adding support for splitting up
the work evenly, specifying a minimum amount of work that's appropriate
for one helper thread to do, load balancing between helpers, and
coordinating them.  More documentation in patches 4 and 8.

This series is the first step in a project to address other memory
proportional bottlenecks in the kernel such as pmem struct page init, vfio
page pinning, hugetlb fallocate, and munmap.  Deferred page init doesn't
require concurrency limits, resource control, or priority adjustments like
these other users will because it happens during boot when the system is
otherwise idle and waiting for page init to finish.

This has been run on a variety of x86 systems and speeds up kernel boot by
4% to 49%, saving up to 1.6 out of 4 seconds.  Patch 6 has more numbers.

This patch (of 8):

padata_driver_exit() is unnecessary because padata isn't built as a module
and doesn't exit.

padata's init routine will soon allocate memory, so getting rid of the
exit function now avoids pointless code to free it.

Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Link: http://lkml.kernel.org/r/20200527173608.2885243-1-daniel.m.jordan@oracle.com
Link: http://lkml.kernel.org/r/20200527173608.2885243-2-daniel.m.jordan@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: call cond_resched() from deferred_init_memmap()

Now that deferred pages are initialized with interrupts enabled we can
replace touch_nmi_watchdog() with cond_resched(), as it was before
3a2d7fa8a3d5.

For now, we cannot do the same in deferred_grow_zone() as it is still
initializes pages with interrupts disabled.

This change fixes RCU problem described in
https://lkml.kernel.org/r/20200401104156.11564-2-david@redhat.com

[   60.474005] rcu: INFO: rcu_sched detected stalls on CPUs/tasks:
[   60.475000] rcu:  1-...0: (0 ticks this GP) idle=02a/1/0x4000000000000000 softirq=1/1 fqs=15000
[   60.475000] rcu:  (detected by 0, t=60002 jiffies, g=-1199, q=1)
[   60.475000] Sending NMI from CPU 0 to CPUs 1:
[    1.760091] NMI backtrace for cpu 1
[    1.760091] CPU: 1 PID: 20 Comm: pgdatinit0 Not tainted 4.18.0-147.9.1.el8_1.x86_64 #1
[    1.760091] Hardware name: Red Hat KVM, BIOS 1.13.0-1.module+el8.2.0+5520+4e5817f3 04/01/2014
[    1.760091] RIP: 0010:__init_single_page.isra.65+0x10/0x4f
[    1.760091] Code: 48 83 cf 63 48 89 f8 0f 1f 40 00 48 89 c6 48 89 d7 e8 6b 18 80 ff 66 90 5b c3 31 c0 b9 10 00 00 00 49 89 f8 48 c1 e6 33 f3 ab <b8> 07 00 00 00 48 c1 e2 36 41 c7 40 34 01 00 00 00 48 c1 e0 33 41
[    1.760091] RSP: 0000:ffffba783123be40 EFLAGS: 00000006
[    1.760091] RAX: 0000000000000000 RBX: fffffad34405e300 RCX: 0000000000000000
[    1.760091] RDX: 0000000000000000 RSI: 0010000000000000 RDI: fffffad34405e340
[    1.760091] RBP: 0000000033f3177e R08: fffffad34405e300 R09: 0000000000000002
[    1.760091] R10: 000000000000002b R11: ffff98afb691a500 R12: 0000000000000002
[    1.760091] R13: 0000000000000000 R14: 000000003f03ea00 R15: 000000003e10178c
[    1.760091] FS:  0000000000000000(0000) GS:ffff9c9ebeb00000(0000) knlGS:0000000000000000
[    1.760091] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    1.760091] CR2: 00000000ffffffff CR3: 000000a1cf20a001 CR4: 00000000003606e0
[    1.760091] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    1.760091] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[    1.760091] Call Trace:
[    1.760091]  deferred_init_pages+0x8f/0xbf
[    1.760091]  deferred_init_memmap+0x184/0x29d
[    1.760091]  ? deferred_free_pages.isra.97+0xba/0xba
[    1.760091]  kthread+0x112/0x130
[    1.760091]  ? kthread_flush_work_fn+0x10/0x10
[    1.760091]  ret_from_fork+0x35/0x40
[   89.123011] node 0 initialised, 1055935372 pages in 88650ms

Fixes: 3a2d7fa8a3d5 ("mm: disable interrupts while initializing deferred pages")
Reported-by: Yiqian Wei <yiwei@redhat.com>
Signed-off-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: James Morris <jmorris@namei.org>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org> [4.17+]
Link: http://lkml.kernel.org/r/20200403140952.17177-4-pasha.tatashin@soleen.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: initialize deferred pages with interrupts enabled

Initializing struct pages is a long task and keeping interrupts disabled
for the duration of this operation introduces a number of problems.

1. jiffies are not updated for long period of time, and thus incorrect time
   is reported. See proposed solution and discussion here:
   lkml/20200311123848.118638-1-shile.zhang@linux.alibaba.com
2. It prevents farther improving deferred page initialization by allowing
   intra-node multi-threading.

We are keeping interrupts disabled to solve a rather theoretical problem
that was never observed in real world (See 3a2d7fa8a3d5).

Let's keep interrupts enabled. In case we ever encounter a scenario where
an interrupt thread wants to allocate large amount of memory this early in
boot we can deal with that by growing zone (see deferred_grow_zone()) by
the needed amount before starting deferred_init_memmap() threads.

Before:
[    1.232459] node 0 initialised, 12058412 pages in 1ms

After:
[    1.632580] node 0 initialised, 12051227 pages in 436ms

Fixes: 3a2d7fa8a3d5 ("mm: disable interrupts while initializing deferred pages")
Reported-by: Shile Zhang <shile.zhang@linux.alibaba.com>
Signed-off-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: James Morris <jmorris@namei.org>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Yiqian Wei <yiwei@redhat.com>
Cc: <stable@vger.kernel.org> [4.17+]
Link: http://lkml.kernel.org/r/20200403140952.17177-3-pasha.tatashin@soleen.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/pagealloc.c: call touch_nmi_watchdog() on max order boundaries in deferred init

Patch series "initialize deferred pages with interrupts enabled", v4.

Keep interrupts enabled during deferred page initialization in order to
make code more modular and allow jiffies to update.

Original approach, and discussion can be found here:
 http://lkml.kernel.org/r/20200311123848.118638-1-shile.zhang@linux.alibaba.com

This patch (of 3):

deferred_init_memmap() disables interrupts the entire time, so it calls
touch_nmi_watchdog() periodically to avoid soft lockup splats.  Soon it
will run with interrupts enabled, at which point cond_resched() should be
used instead.

deferred_grow_zone() makes the same watchdog calls through code shared
with deferred init but will continue to run with interrupts disabled, so
it can't call cond_resched().

Pull the watchdog calls up to these two places to allow the first to be
changed later, independently of the second.  The frequency reduces from
twice per pageblock (init and free) to once per max order block.

Fixes: 3a2d7fa8a3d5 ("mm: disable interrupts while initializing deferred pages")
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Shile Zhang <shile.zhang@linux.alibaba.com>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: James Morris <jmorris@namei.org>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Yiqian Wei <yiwei@redhat.com>
Cc: <stable@vger.kernel.org> [4.17+]
Link: http://lkml.kernel.org/r/20200403140952.17177-2-pasha.tatashin@soleen.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: restrict and formalize compound_page_dtors[]

Restrict elements in compound_page_dtors[] array per NR_COMPOUND_DTORS and
explicitly position them according to enum compound_dtor_id.  This
improves protection against possible misalignment between
compound_page_dtors[] and enum compound_dtor_id later on.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/1589795958-19317-1-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: reset the zone->watermark_boost early

Updating the zone watermarks by any means, like min_free_kbytes,
water_mark_scale_factor etc, when ->watermark_boost is set will result in
higher low and high watermarks than the user asked.

Below are the steps to reproduce the problem on system setup of Android
kernel running on Snapdragon hardware.

1) Default settings of the system are as below:

   #cat /proc/sys/vm/min_free_kbytes = 5162
   #cat /proc/zoneinfo | grep -e boost -e low -e "high " -e min -e Node
Node 0, zone   Normal
min      797
low      8340
high     8539

2) Monitor the zone->watermark_boost(by adding a debug print in the
   kernel) and whenever it is greater than zero value, write the same
   value of min_free_kbytes obtained from step 1.

   #echo 5162 > /proc/sys/vm/min_free_kbytes

3) Then read the zone watermarks in the system while the
   ->watermark_boost is zero.  This should show the same values of
   watermarks as step 1 but shown a higher values than asked.

   #cat /proc/zoneinfo | grep -e boost -e low -e "high " -e min -e Node
Node 0, zone   Normal
min      797
low      21148
high     21347

These higher values are because of updating the zone watermarks using the
macro min_wmark_pages(zone) which also adds the zone->watermark_boost.

#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] +
z->watermark_boost)

So the steps that lead to the issue are:

1) On the extfrag event, watermarks are boosted by storing the required
   value in ->watermark_boost.

2) User tries to update the zone watermarks level in the system through
   min_free_kbytes or watermark_scale_factor.

3) Later, when kswapd woke up, it resets the zone->watermark_boost to
   zero.

In step 2), we use the min_wmark_pages() macro to store the watermarks
in the zone structure thus the values are always offsetted by
->watermark_boost value. This can be avoided by resetting the
->watermark_boost to zero before it is used.

Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Link: http://lkml.kernel.org/r/1589457511-4255-1-git-send-email-charante@codeaurora.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: reset numa stats for boot pagesets

Initially, the per-cpu pagesets of each zone are set to the boot pagesets.
The real pagesets are allocated later but before that happens, page
allocations do occur and the numa stats for the boot pagesets get
incremented since they are common to all zones at that point.

The real pagesets, however, are allocated for the populated zones only.
Unpopulated zones, like those associated with memory-less nodes, continue
using the boot pageset and end up skewing the numa stats of the
corresponding node.

E.g.

  $ numactl -H
  available: 2 nodes (0-1)
  node 0 cpus: 0 1 2 3
  node 0 size: 0 MB
  node 0 free: 0 MB
  node 1 cpus: 4 5 6 7
  node 1 size: 8131 MB
  node 1 free: 6980 MB
  node distances:
  node   0   1
    0:  10  40
    1:  40  10

  $ numastat
                             node0           node1
  numa_hit                     108           56495
  numa_miss                      0               0
  numa_foreign                   0               0
  interleave_hit                 0            4537
  local_node                   108           31547
  other_node                     0           24948

Hence, the boot pageset stats need to be cleared after the real pagesets
are allocated.

After this point, the stats of the boot pagesets do not change as page
allocations requested for a memory-less node will either fail (if
__GFP_THISNODE is used) or get fulfilled by a preferred zone of a
different node based on the fallback zonelist.

[sandipan@linux.ibm.com: v3]
Link: http://lkml.kernel.org/r/20200511170356.162531-1-sandipan@linux.ibm.com
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Link: http://lkml.kernel.org/r/9c9c2d1b15e37f6e6bf32f99e3100035e90c4ac9.1588868430.git.sandipan@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: rename gfpflags_to_migratetype to gfp_migratetype for same convention

Pageblock migrate type is encoded in GFP flags, just as zone_type and
zonelist.

Currently we use gfp_zone() and gfp_zonelist() to extract related
information, it would be proper to use the same naming convention for
migrate type.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Link: http://lkml.kernel.org/r/20200329080823.7735-1-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: use NODE_MASK_NONE in build_zonelists()

Slightly simplify the code by initializing user_mask with NODE_MASK_NONE,
instead of later calling nodes_clear().  This saves a line of code.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Link: http://lkml.kernel.org/r/20200330220840.21228-1-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: integrate classzone_idx and high_zoneidx

classzone_idx is just different name for high_zoneidx now.  So, integrate
them and add some comment to struct alloc_context in order to reduce
future confusion about the meaning of this variable.

The accessor, ac_classzone_idx() is also removed since it isn't needed
after integration.

In addition to integration, this patch also renames high_zoneidx to
highest_zoneidx since it represents more precise meaning.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ye Xiaolong <xiaolong.ye@intel.com>
Link: http://lkml.kernel.org/r/1587095923-7515-3-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: use ac->high_zoneidx for classzone_idx

Patch series "integrate classzone_idx and high_zoneidx", v5.

This patchset is followup of the problem reported and discussed two years
ago [1, 2].  The problem this patchset solves is related to the
classzone_idx on the NUMA system.  It causes a problem when the lowmem
reserve protection exists for some zones on a node that do not exist on
other nodes.

This problem was reported two years ago, and, at that time, the solution
got general agreements [2].  But it was not upstreamed.

[1]: http://lkml.kernel.org/r/20180102063528.GG30397@yexl-desktop
[2]: http://lkml.kernel.org/r/1525408246-14768-1-git-send-email-iamjoonsoo.kim@lge.com

This patch (of 2):

Currently, we use classzone_idx to calculate lowmem reserve proetection
for an allocation request.  This classzone_idx causes a problem on NUMA
systems when the lowmem reserve protection exists for some zones on a node
that do not exist on other nodes.

Before further explanation, I should first clarify how to compute the
classzone_idx and the high_zoneidx.

- ac->high_zoneidx is computed via the arcane gfp_zone(gfp_mask) and
  represents the index of the highest zone the allocation can use

- classzone_idx was supposed to be the index of the highest zone on the
  local node that the allocation can use, that is actually available in
  the system

Think about following example.  Node 0 has 4 populated zone,
DMA/DMA32/NORMAL/MOVABLE.  Node 1 has 1 populated zone, NORMAL.  Some
zones, such as MOVABLE, doesn't exist on node 1 and this makes following
difference.

Assume that there is an allocation request whose gfp_zone(gfp_mask) is the
zone, MOVABLE.  Then, it's high_zoneidx is 3.  If this allocation is
initiated on node 0, it's classzone_idx is 3 since actually
available/usable zone on local (node 0) is MOVABLE.  If this allocation is
initiated on node 1, it's classzone_idx is 2 since actually
available/usable zone on local (node 1) is NORMAL.

You can see that classzone_idx of the allocation request are different
according to their starting node, even if their high_zoneidx is the same.

Think more about these two allocation requests.  If they are processed on
local, there is no problem.  However, if allocation is initiated on node 1
are processed on remote, in this example, at the NORMAL zone on node 0,
due to memory shortage, problem occurs.  Their different classzone_idx
leads to different lowmem reserve and then different min watermark.  See
the following example.

root@ubuntu:/sys/devices/system/memory# cat /proc/zoneinfo
Node 0, zone      DMA
  per-node stats
...
  pages free     3965
        min      5
        low      8
        high     11
        spanned  4095
        present  3998
        managed  3977
        protection: (0, 2961, 4928, 5440)
...
Node 0, zone    DMA32
  pages free     757955
        min      1129
        low      1887
        high     2645
        spanned  1044480
        present  782303
        managed  758116
        protection: (0, 0, 1967, 2479)
...
Node 0, zone   Normal
  pages free     459806
        min      750
        low      1253
        high     1756
        spanned  524288
        present  524288
        managed  503620
        protection: (0, 0, 0, 4096)
...
Node 0, zone  Movable
  pages free     130759
        min      195
        low      326
        high     457
        spanned  1966079
        present  131072
        managed  131072
        protection: (0, 0, 0, 0)
...
Node 1, zone      DMA
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 0, 1006, 1006)
Node 1, zone    DMA32
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 0, 1006, 1006)
Node 1, zone   Normal
  per-node stats
...
  pages free     233277
        min      383
        low      640
        high     897
        spanned  262144
        present  262144
        managed  257744
        protection: (0, 0, 0, 0)
...
Node 1, zone  Movable
  pages free     0
        min      0
        low      0
        high     0
        spanned  262144
        present  0
        managed  0
        protection: (0, 0, 0, 0)

- static min watermark for the NORMAL zone on node 0 is 750.

- lowmem reserve for the request with classzone idx 3 at the NORMAL on
  node 0 is 4096.

- lowmem reserve for the request with classzone idx 2 at the NORMAL on
  node 0 is 0.

So, overall min watermark is:
allocation initiated on node 0 (classzone_idx 3): 750 + 4096 = 4846
allocation initiated on node 1 (classzone_idx 2): 750 + 0 = 750

Allocation initiated on node 1 will have some precedence than allocation
initiated on node 0 because min watermark of the former allocation is
lower than the other.  So, allocation initiated on node 1 could succeed on
node 0 when allocation initiated on node 0 could not, and, this could
cause too many numa_miss allocation.  Then, performance could be
downgraded.

Recently, there was a regression report about this problem on CMA patches
since CMA memory are placed in ZONE_MOVABLE by those patches.  I checked
that problem is disappeared with this fix that uses high_zoneidx for
classzone_idx.

http://lkml.kernel.org/r/20180102063528.GG30397@yexl-desktop

Using high_zoneidx for classzone_idx is more consistent way than previous
approach because system's memory layout doesn't affect anything to it.
With this patch, both classzone_idx on above example will be 3 so will
have the same min watermark.

allocation initiated on node 0: 750 + 4096 = 4846
allocation initiated on node 1: 750 + 4096 = 4846

One could wonder if there is a side effect that allocation initiated on
node 1 will use higher bar when allocation is handled on local since
classzone_idx could be higher than before.  It will not happen because the
zone without managed page doesn't contributes lowmem_reserve at all.

Reported-by: Ye Xiaolong <xiaolong.ye@intel.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Ye Xiaolong <xiaolong.ye@intel.com>
Reviewed-by: Baoquan He <bhe@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/1587095923-7515-1-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1587095923-7515-2-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmstat.c: do not show lowmem reserve protection information of empty zone

Because the lowmem reserve protection of a zone can't tell anything if the
zone is empty, except of adding one more line in /proc/zoneinfo.

Let's remove it from that zone's showing.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200402140113.3696-4-bhe@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: clear out zone->lowmem_reserve[] if the zone is empty

When requesting memory allocation from a specific zone is not satisfied,
it will fall to lower zone to try allocating memory.  In this case, lower
zone's ->lowmem_reserve[] will help protect its own memory resource.  The
higher the relevant ->lowmem_reserve[] is, the harder the upper zone can
get memory from this lower zone.

However, this protection mechanism should be applied to populated zone,
but not an empty zone. So filling ->lowmem_reserve[] for empty zone is
not necessary, and may mislead people that it's valid data in that zone.

Node 2, zone      DMA
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 0, 1024, 1024)
Node 2, zone    DMA32
  pages free     0
        min      0
        low      0
        high     0
        spanned  0
        present  0
        managed  0
        protection: (0, 0, 1024, 1024)
Node 2, zone   Normal
  per-node stats
      nr_inactive_anon 0
      nr_active_anon 143
      nr_inactive_file 0
      nr_active_file 0
      nr_unevictable 0
      nr_slab_reclaimable 45
      nr_slab_unreclaimable 254

Here clear out zone->lowmem_reserve[] if zone is empty.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200402140113.3696-3-bhe@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: only tune sysctl_lowmem_reserve_ratio value once when changing it

Patch series "improvements about lowmem_reserve and /proc/zoneinfo", v2.

This patch (of 3):

When people write to /proc/sys/vm/lowmem_reserve_ratio to change
sysctl_lowmem_reserve_ratio[], setup_per_zone_lowmem_reserve() is called
to recalculate all ->lowmem_reserve[] for each zone of all nodes as below:

static void setup_per_zone_lowmem_reserve(void)
{
...
for_each_online_pgdat(pgdat) {
for (j = 0; j < MAX_NR_ZONES; j++) {
...
while (idx) {
...
if (sysctl_lowmem_reserve_ratio[idx] < 1) {
sysctl_lowmem_reserve_ratio[idx] = 0;
lower_zone->lowmem_reserve[j] = 0;
                                } else {
...
}
}
}
}

Meanwhile, here, sysctl_lowmem_reserve_ratio[idx] will be tuned if its
value is smaller than '1'.  As we know, sysctl_lowmem_reserve_ratio[] is
set for zone without regarding to which node it belongs to.  That means
the tuning will be done on all nodes, even though it has been done in the
first node.

And the tuning will be done too even when init_per_zone_wmark_min() calls
setup_per_zone_lowmem_reserve(), where actually nobody tries to change
sysctl_lowmem_reserve_ratio[].

So now move the tuning into lowmem_reserve_ratio_sysctl_handler(), to make
code logic more reasonable.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/20200402140113.3696-1-bhe@redhat.com
Link: http://lkml.kernel.org/r/20200402140113.3696-2-bhe@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: remove unused free_bootmem_with_active_regions

Since commit 397dc00e249ec64e10 ("mips: sgi-ip27: switch from DISCONTIGMEM
to SPARSEMEM"), the last caller of free_bootmem_with_active_regions() was
gone.  Now no user calls it any more.

Let's remove it.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200402143455.5145-1-bhe@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,page_alloc,cma: conditionally prefer cma pageblocks for movable allocations

Currently a cma area is barely used by the page allocator because it's
used only as a fallback from movable, however kswapd tries hard to make
sure that the fallback path isn't used.

This results in a system evicting memory and pushing data into swap, while
lots of CMA memory is still available.  This happens despite the fact that
alloc_contig_range is perfectly capable of moving any movable allocations
out of the way of an allocation.

To effectively use the cma area let's alter the rules: if the zone has
more free cma pages than the half of total free pages in the zone, use cma
pageblocks first and fallback to movable blocks in the case of failure.

[guro@fb.com: ifdef the cma-specific code]
Link: http://lkml.kernel.org/r/20200311225832.GA178154@carbon.DHCP.thefacebook.com
Co-developed-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Qian Cai <cai@lca.pw>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Link: http://lkml.kernel.org/r/20200306150102.3e77354b@imladris.surriel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: extract check_[new|free]_page_bad() common part to page_bad_reason()

We share similar code in check_[new|free]_page_bad() to get the page's bad
reason.

Let's extract it and reduce code duplication.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200411220357.9636-6-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: rename free_pages_check() to check_free_page()

free_pages_check() is the counterpart of check_new_page().  Rename it to
use the same naming convention.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200411220357.9636-5-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: rename free_pages_check_bad() to check_free_page_bad()

free_pages_check_bad() is the counterpart of check_new_page_bad().  Rename
it to use the same naming convention.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200411220357.9636-4-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: bad_flags is not necessary for bad_page()

After commit 5b57b8f22709 ("mm/debug.c: always print flags in
dump_page()"), page->flags is always printed for a bad page.  It is not
necessary to have bad_flags any more.

Suggested-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200411220357.9636-3-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: bad_[reason|flags] is not necessary when PageHWPoison

Patch series "mm/page_alloc.c: cleanup on check page", v3.

This patchset does some cleanup related to check page.

1. Remove unnecessary bad_reason assignment
2. Remove bad_flags to bad_page()
3. Rename function for naming convention
4. Extract common part to check page

Thanks for suggestions from David Rientjes and Anshuman Khandual.

This patch (of 5):

Since function returns directly, bad_[reason|flags] is not used any where.
And move this to the first.

This is a following cleanup for commit e570f56cccd21 ("mm:
check_new_page_bad() directly returns in __PG_HWPOISON case")

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/20200411220357.9636-2-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>