mm, proc: be more verbose about unstable VMA flags in /proc/<pid>/smaps

Patch series "THP eligibility reporting via proc".

This series of three patches aims at making THP eligibility reporting much
more robust and long term sustainable.  The trigger for the change is a
regression report [2] and the long follow up discussion.  In short the
specific application didn't have good API to query whether a particular
mapping can be backed by THP so it has used VMA flags to workaround that.
These flags represent a deep internal state of VMAs and as such they
should be used by userspace with a great deal of caution.

A similar has happened for [3] when users complained that VM_MIXEDMAP is
no longer set on DAX mappings.  Again a lack of a proper API led to an
abuse.

The first patch in the series tries to emphasise that that the semantic of
flags might change and any application consuming those should be really
careful.

The remaining two patches provide a more suitable interface to address [2]
and provide a consistent API to query the THP status both for each VMA and
process wide as well.  [1]

http://lkml.kernel.org/r/20181120103515.25280-1-mhocko@kernel.org [2]
http://lkml.kernel.org/r/http://lkml.kernel.org/r/alpine.DEB.2.21.1809241054050.224429@chino.kir.corp.google.com
[3] http://lkml.kernel.org/r/20181002100531.GC4135@quack2.suse.cz

This patch (of 3):

Even though vma flags exported via /proc/<pid>/smaps are explicitly
documented to be not guaranteed for future compatibility the warning
doesn't go far enough because it doesn't mention semantic changes to those
flags.  And they are important as well because these flags are a deep
implementation internal to the MM code and the semantic might change at
any time.

Let's consider two recent examples:
http://lkml.kernel.org/r/20181002100531.GC4135@quack2.suse.cz
: commit e1fb4a086495 "dax: remove VM_MIXEDMAP for fsdax and device dax" has
: removed VM_MIXEDMAP flag from DAX VMAs. Now our testing shows that in the
: mean time certain customer of ours started poking into /proc/<pid>/smaps
: and looks at VMA flags there and if VM_MIXEDMAP is missing among the VMA
: flags, the application just fails to start complaining that DAX support is
: missing in the kernel.

http://lkml.kernel.org/r/alpine.DEB.2.21.1809241054050.224429@chino.kir.corp.google.com
: Commit 1860033237d4 ("mm: make PR_SET_THP_DISABLE immediately active")
: introduced a regression in that userspace cannot always determine the set
: of vmas where thp is ineligible.
: Userspace relies on the "nh" flag being emitted as part of /proc/pid/smaps
: to determine if a vma is eligible to be backed by hugepages.
: Previous to this commit, prctl(PR_SET_THP_DISABLE, 1) would cause thp to
: be disabled and emit "nh" as a flag for the corresponding vmas as part of
: /proc/pid/smaps.  After the commit, thp is disabled by means of an mm
: flag and "nh" is not emitted.
: This causes smaps parsing libraries to assume a vma is eligible for thp
: and ends up puzzling the user on why its memory is not backed by thp.

In both cases userspace was relying on a semantic of a specific VMA flag.
The primary reason why that happened is a lack of a proper interface.
While this has been worked on and it will be fixed properly, it seems that
our wording could see some refinement and be more vocal about semantic
aspect of these flags as well.

Link: http://lkml.kernel.org/r/20181211143641.3503-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Jan Kara <jack@suse.cz>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Paul Oppenheimer <bepvte@gmail.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/memory_hotplug.h: remove duplicate declaration of offline_pages()

offline_pages() is already declared in this file.

Just remove the duplicated one.

Link: http://lkml.kernel.org/r/20181205031357.24769-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmu_notifier: use structure for invalidate_range_start/end calls v2

To avoid having to change many call sites everytime we want to add a
parameter use a structure to group all parameters for the mmu_notifier
invalidate_range_start/end cakks.  No functional changes with this patch.

[akpm@linux-foundation.org: coding style fixes]
Link: http://lkml.kernel.org/r/20181205053628.3210-3-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <zwisler@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Felix Kuehling <felix.kuehling@amd.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
From: Jérôme Glisse <jglisse@redhat.com>
Subject: mm/mmu_notifier: use structure for invalidate_range_start/end calls v3

fix build warning in migrate.c when CONFIG_MMU_NOTIFIER=n

Link: http://lkml.kernel.org/r/20181213171330.8489-3-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmu_notifier: use structure for invalidate_range_start/end callback

Patch series "mmu notifier contextual informations", v2.

This patchset adds contextual information, why an invalidation is
happening, to mmu notifier callback.  This is necessary for user of mmu
notifier that wish to maintains their own data structure without having to
add new fields to struct vm_area_struct (vma).

For instance device can have they own page table that mirror the process
address space.  When a vma is unmap (munmap() syscall) the device driver
can free the device page table for the range.

Today we do not have any information on why a mmu notifier call back is
happening and thus device driver have to assume that it is always an
munmap().  This is inefficient at it means that it needs to re-allocate
device page table on next page fault and rebuild the whole device driver
data structure for the range.

Other use case beside munmap() also exist, for instance it is pointless
for device driver to invalidate the device page table when the
invalidation is for the soft dirtyness tracking.  Or device driver can
optimize away mprotect() that change the page table permission access for
the range.

This patchset enables all this optimizations for device drivers.  I do not
include any of those in this series but another patchset I am posting will
leverage this.

The patchset is pretty simple from a code point of view.  The first two
patches consolidate all mmu notifier arguments into a struct so that it is
easier to add/change arguments.  The last patch adds the contextual
information (munmap, protection, soft dirty, clear, ...).

This patch (of 3):

To avoid having to change many callback definition everytime we want to
add a parameter use a structure to group all parameters for the
mmu_notifier invalidate_range_start/end callback.  No functional changes
with this patch.

[akpm@linux-foundation.org: fix drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c kerneldoc]
Link: http://lkml.kernel.org/r/20181205053628.3210-2-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Jan Kara <jack@suse.cz>
Acked-by: Jason Gunthorpe <jgg@mellanox.com> [infiniband]
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <zwisler@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Christian Koenig <christian.koenig@amd.com>
Cc: Felix Kuehling <felix.kuehling@amd.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hwpoison, memory_hotplug: allow hwpoisoned pages to be offlined

We have received a bug report that an injected MCE about faulty memory
prevents memory offline to succeed on 4.4 base kernel.  The underlying
reason was that the HWPoison page has an elevated reference count and the
migration keeps failing.  There are two problems with that.  First of all
it is dubious to migrate the poisoned page because we know that accessing
that memory is possible to fail.  Secondly it doesn't make any sense to
migrate a potentially broken content and preserve the memory corruption
over to a new location.

Oscar has found out that 4.4 and the current upstream kernels behave
slightly differently with his simply testcase

===

int main(void)
{
        int ret;
        int i;
        int fd;
        char *array = malloc(4096);
        char *array_locked = malloc(4096);

        fd = open("/tmp/data", O_RDONLY);
        read(fd, array, 4095);

        for (i = 0; i < 4096; i++)
                array_locked[i] = 'd';

        ret = mlock((void *)PAGE_ALIGN((unsigned long)array_locked), sizeof(array_locked));
        if (ret)
                perror("mlock");

        sleep (20);

        ret = madvise((void *)PAGE_ALIGN((unsigned long)array_locked), 4096, MADV_HWPOISON);
        if (ret)
                perror("madvise");

        for (i = 0; i < 4096; i++)
                array_locked[i] = 'd';

        return 0;
}
===

+ offline this memory.

In 4.4 kernels he saw the hwpoisoned page to be returned back to the LRU
list
kernel:  [<ffffffff81019ac9>] dump_trace+0x59/0x340
kernel:  [<ffffffff81019e9a>] show_stack_log_lvl+0xea/0x170
kernel:  [<ffffffff8101ac71>] show_stack+0x21/0x40
kernel:  [<ffffffff8132bb90>] dump_stack+0x5c/0x7c
kernel:  [<ffffffff810815a1>] warn_slowpath_common+0x81/0xb0
kernel:  [<ffffffff811a275c>] __pagevec_lru_add_fn+0x14c/0x160
kernel:  [<ffffffff811a2eed>] pagevec_lru_move_fn+0xad/0x100
kernel:  [<ffffffff811a334c>] __lru_cache_add+0x6c/0xb0
kernel:  [<ffffffff81195236>] add_to_page_cache_lru+0x46/0x70
kernel:  [<ffffffffa02b4373>] extent_readpages+0xc3/0x1a0 [btrfs]
kernel:  [<ffffffff811a16d7>] __do_page_cache_readahead+0x177/0x200
kernel:  [<ffffffff811a18c8>] ondemand_readahead+0x168/0x2a0
kernel:  [<ffffffff8119673f>] generic_file_read_iter+0x41f/0x660
kernel:  [<ffffffff8120e50d>] __vfs_read+0xcd/0x140
kernel:  [<ffffffff8120e9ea>] vfs_read+0x7a/0x120
kernel:  [<ffffffff8121404b>] kernel_read+0x3b/0x50
kernel:  [<ffffffff81215c80>] do_execveat_common.isra.29+0x490/0x6f0
kernel:  [<ffffffff81215f08>] do_execve+0x28/0x30
kernel:  [<ffffffff81095ddb>] call_usermodehelper_exec_async+0xfb/0x130
kernel:  [<ffffffff8161c045>] ret_from_fork+0x55/0x80

And that latter confuses the hotremove path because an LRU page is
attempted to be migrated and that fails due to an elevated reference
count.  It is quite possible that the reuse of the HWPoisoned page is some
kind of fixed race condition but I am not really sure about that.

With the upstream kernel the failure is slightly different.  The page
doesn't seem to have LRU bit set but isolate_movable_page simply fails and
do_migrate_range simply puts all the isolated pages back to LRU and
therefore no progress is made and scan_movable_pages finds same set of
pages over and over again.

Fix both cases by explicitly checking HWPoisoned pages before we even try
to get reference on the page, try to unmap it if it is still mapped.  As
explained by Naoya:

: Hwpoison code never unmapped those for no big reason because
: Ksm pages never dominate memory, so we simply didn't have strong
: motivation to save the pages.

Also put WARN_ON(PageLRU) in case there is a race and we can hit LRU
HWPoison pages which shouldn't happen but I couldn't convince myself about
that.  Naoya has noted the following:

: Theoretically no such gurantee, because try_to_unmap() doesn't have a
: guarantee of success and then memory_failure() returns immediately
: when hwpoison_user_mappings fails.
: Or the following code (comes after hwpoison_user_mappings block) also impli=
: es
: that the target page can still have PageLRU flag.
:
:         /*
:          * Torn down by someone else?
:          */
:         if (PageLRU(p) && !PageSwapCache(p) && p->mapping =3D=3D NULL) {
:                 action_result(pfn, MF_MSG_TRUNCATED_LRU, MF_IGNORED);
:                 res =3D -EBUSY;
:                 goto out;
:         }
:
: So I think it's OK to keep "if (WARN_ON(PageLRU(page)))" block in
: current version of your patch.

Link: http://lkml.kernel.org/r/20181206120135.14079-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Oscar Salvador <osalvador@suse.com>
Debugged-by: Oscar Salvador <osalvador@suse.com>
Tested-by: Oscar Salvador <osalvador@suse.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, kmemleak: little optimization while scanning

kmemleak_scan() goes through all online nodes and tries to scan all used
pages.

We can do better and use pfn_to_online_page(), so in case we have
CONFIG_MEMORY_HOTPLUG, offlined pages will be skiped automatically.  For
boxes where CONFIG_MEMORY_HOTPLUG is not present, pfn_to_online_page()
will fallback to pfn_valid().

Another little optimization is to check if the page belongs to the node we
are currently checking, so in case we have nodes interleaved we will not
check the same pfn multiple times.

I ran some tests:

Add some memory to node1 and node2 making it interleaved:

(qemu) object_add memory-backend-ram,id=ram0,size=1G
(qemu) device_add pc-dimm,id=dimm0,memdev=ram0,node=1
(qemu) object_add memory-backend-ram,id=ram1,size=1G
(qemu) device_add pc-dimm,id=dimm1,memdev=ram1,node=2
(qemu) object_add memory-backend-ram,id=ram2,size=1G
(qemu) device_add pc-dimm,id=dimm2,memdev=ram2,node=1

Then, we offline that memory:
 # for i in {32..39} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;done
 # for i in {48..55} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;don
 # for i in {40..47} ; do echo "offline" > /sys/devices/system/node/node2/memory$i/state;done

And we run kmemleak_scan:

 # echo "scan" > /sys/kernel/debug/kmemleak

before the patch:

kmemleak: time spend: 41596 us

after the patch:

kmemleak: time spend: 34899 us

[akpm@linux-foundation.org: remove stray newline, per Oscar]
Link: http://lkml.kernel.org/r/20181206131918.25099-1-osalvador@suse.de
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/ioremap: ensure break-before-make is used for huge p4d mappings

Whilst no architectures actually enable support for huge p4d mappings in
the vmap area, the code that is implemented should be using
break-before-make, as we do for pud and pmd huge entries.

Link: http://lkml.kernel.org/r/1544120495-17438-6-git-send-email-will.deacon@arm.com
Signed-off-by: Will Deacon <will.deacon@arm.com>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Chintan Pandya <cpandya@codeaurora.org>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/ioremap: ensure phys_addr actually corresponds to a physical address

The current ioremap() code uses a phys_addr variable at each level of page
table, which is confusingly offset by subtracting the base virtual address
being mapped so that adding the current virtual address back on when
iterating through the page table entries gives back the corresponding
physical address.

This is fairly confusing and results in all users of phys_addr having to
add the current virtual address back on.  Instead, this patch just updates
phys_addr when iterating over the page table entries, ensuring that it's
always up-to-date and doesn't require explicit offsetting.

Link: http://lkml.kernel.org/r/1544120495-17438-5-git-send-email-will.deacon@arm.com
Signed-off-by: Will Deacon <will.deacon@arm.com>
Tested-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Chintan Pandya <cpandya@codeaurora.org>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86/pgtable: drop pXd_none() checks from pXd_free_pYd_table()

The core code already has a check for pXd_none(), so remove it from the
architecture implementation.

Link: http://lkml.kernel.org/r/1544120495-17438-4-git-send-email-will.deacon@arm.com
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Chintan Pandya <cpandya@codeaurora.org>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arm64: mmu: drop pXd_present() checks from pXd_free_pYd_table()

The core code already has a check for pXd_none(), so remove it from the
architecture implementation.

Link: http://lkml.kernel.org/r/1544120495-17438-3-git-send-email-will.deacon@arm.com
Signed-off-by: Will Deacon <will.deacon@arm.com>
Cc: Chintan Pandya <cpandya@codeaurora.org>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ioremap: rework pXd_free_pYd_page() API

The recently merged API for ensuring break-before-make on page-table
entries when installing huge mappings in the vmalloc/ioremap region is
fairly counter-intuitive, resulting in the arch freeing functions (e.g.
pmd_free_pte_page()) being called even on entries that aren't present.
This resulted in a minor bug in the arm64 implementation, giving rise to
spurious VM_WARN messages.

This patch moves the pXd_present() checks out into the core code,
refactoring the callsites at the same time so that we avoid the complex
conjunctions when determining whether or not we can put down a huge
mapping.

Link: http://lkml.kernel.org/r/1544120495-17438-2-git-send-email-will.deacon@arm.com
Signed-off-by: Will Deacon <will.deacon@arm.com>
Reviewed-by: Toshi Kani <toshi.kani@hpe.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Chintan Pandya <cpandya@codeaurora.org>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/filemap.c: remove useless check in pagecache_get_page()

page always is not NULL, so we may remove this useless check.

Link: http://lkml.kernel.org/r/154419752044.18559.2452963074922917720.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Cyrill Gorcunov <gorcunov@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

/proc/kpagecount: return 0 for special pages that are never mapped

Certain pages that are never mapped to userspace have a type indicated in
the page_type field of their struct pages (e.g.  PG_buddy).  page_type
overlaps with _mapcount so set the count to 0 and avoid calling
page_mapcount() for these pages.

[anthony.yznaga@oracle.com: incorporate feedback from Matthew Wilcox]
Link: http://lkml.kernel.org/r/1544481313-27318-1-git-send-email-anthony.yznaga@oracle.com
Link: http://lkml.kernel.org/r/1543963526-27917-1-git-send-email-anthony.yznaga@oracle.com
Signed-off-by: Anthony Yznaga <anthony.yznaga@oracle.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Miles Chen <miles.chen@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page-types.c: fix "kpagecount returned fewer pages than expected" failures

Because kpagecount_read() fakes success if map counts are not being
collected, clamp the page count passed to it by walk_pfn() to the pages
value returned by the preceding call to kpageflags_read().

Link: http://lkml.kernel.org/r/1543962269-26116-1-git-send-email-anthony.yznaga@oracle.com
Fixes: 7f1d23e60718 ("tools/vm/page-types.c: include shared map counts")
Signed-off-by: Anthony Yznaga <anthony.yznaga@oracle.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: drop uneeded __meminit and __meminitdata

Since commit 03e85f9d5f1 ("mm/page_alloc: Introduce
free_area_init_core_hotplug"), some functions changed to only be called
during system initialization.  Concretly, free_area_init_node() and the
functions that hang from it.

Also, some variables are no longer used after the system has gone
through initialization.  So this could be considered as a late clean-up
for that patch.

This patch changes the functions from __meminit to __init, and the
variables from __meminitdata to __initdata.

In return, we get some KBs back:

Before:
  Freeing unused kernel image memory: 2472K

After:
  Freeing unused kernel image memory: 2480K

Link: http://lkml.kernel.org/r/20181204111507.4808-1-osalvador@suse.de
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page-writeback.c: don't break integrity writeback on ->writepage() error

write_cache_pages() is used in both background and integrity writeback
scenarios by various filesystems.  Background writeback is mostly
concerned with cleaning a certain number of dirty pages based on various
mm heuristics.  It may not write the full set of dirty pages or wait for
I/O to complete.  Integrity writeback is responsible for persisting a set
of dirty pages before the writeback job completes.  For example, an
fsync() call must perform integrity writeback to ensure data is on disk
before the call returns.

write_cache_pages() unconditionally breaks out of its processing loop in
the event of a ->writepage() error.  This is fine for background
writeback, which had no strict requirements and will eventually come
around again.  This can cause problems for integrity writeback on
filesystems that might need to clean up state associated with failed page
writeouts.  For example, XFS performs internal delayed allocation
accounting before returning a ->writepage() error, where applicable.  If
the current writeback happens to be associated with an unmount and
write_cache_pages() completes the writeback prematurely due to error, the
filesystem is unmounted in an inconsistent state if dirty+delalloc pages
still exist.

To handle this problem, update write_cache_pages() to always process the
full set of pages for integrity writeback regardless of ->writepage()
errors.  Save the first encountered error and return it to the caller once
complete.  This facilitates XFS (or any other fs that expects integrity
writeback to process the entire set of dirty pages) to clean up its
internal state completely in the event of persistent mapping errors.
Background writeback continues to exit on the first error encountered.

[akpm@linux-foundation.org: fix typo in comment]
Link: http://lkml.kernel.org/r/20181116134304.32440-1-bfoster@redhat.com
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/show_mem.c: drop pgdat_resize_lock in show_mem()

Function show_mem() is used to print system memory status when user
requires or fail to allocate memory.  Generally, this is a best effort
information so any races with memory hotplug (or very theoretically an
early initialization) should be tolerable and the worst that could happen
is to print an imprecise node state.

Drop the resize lock because this is the only place which might hold the
lock from the interrupt context and so all other callers might use a
simple spinlock.  Even though this doesn't solve any real issue it makes
the code easier to follow and tiny more effective.

Link: http://lkml.kernel.org/r/20181129235532.9328-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hmm.c: remove set but not used variable 'devmem'

Fixes gcc '-Wunused-but-set-variable' warning:

mm/hmm.c: In function 'hmm_devmem_ref_kill':
mm/hmm.c:995:21: warning:
 variable 'devmem' set but not used [-Wunused-but-set-variable]

It not used any more since 35d39f953d4e ("mm, hmm: replace
hmm_devmem_pages_create() with devm_memremap_pages()")

Link: http://lkml.kernel.org/r/1543629971-128377-1-git-send-email-yuehaibing@huawei.com
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hotplug: move init_currently_empty_zone() under zone_span_lock protection

During online_pages phase, pgdat->nr_zones will be updated in case this
zone is empty.

Currently the online_pages phase is protected by the global locks
(device_device_hotplug_lock and mem_hotplug_lock), which ensures there is
no contention during the update of nr_zones.

These global locks introduces scalability issues (especially the second
one), which slow down code relying on get_online_mems().  This is also a
preparation for not having to rely on get_online_mems() but instead some
more fine grained locks.

The patch moves init_currently_empty_zone under both zone_span_writelock
and pgdat_resize_lock because both the pgdat state is changed (nr_zones)
and the zone's start_pfn.  Also this patch changes the documentation of
node_size_lock to include the protection of nr_zones.

Link: http://lkml.kernel.org/r/20181203205016.14123-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, sparse: pass nid instead of pgdat to sparse_add_one_section()

Since the information needed in sparse_add_one_section() is node id to
allocate proper memory, it is not necessary to pass its pgdat.

This patch changes the prototype of sparse_add_one_section() to pass node
id directly.  This is intended to reduce misleading that
sparse_add_one_section() would touch pgdat.

Link: http://lkml.kernel.org/r/20181204085657.20472-2-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, sparse: drop pgdat_resize_lock in sparse_add/remove_one_section()

pgdat_resize_lock is used to protect pgdat's memory region information
like: node_start_pfn, node_present_pages, etc.  While in function
sparse_add/remove_one_section(), pgdat_resize_lock is used to protect
initialization/release of one mem_section.  This looks not proper.

These code paths are currently protected by mem_hotplug_lock currently but
should there ever be any reason for locking at the sparse layer a
dedicated lock should be introduced.

Following is the current call trace of sparse_add/remove_one_section()

    mem_hotplug_begin()
    arch_add_memory()
       add_pages()
           __add_pages()
               __add_section()
                   sparse_add_one_section()
    mem_hotplug_done()

    mem_hotplug_begin()
    arch_remove_memory()
        __remove_pages()
            __remove_section()
                sparse_remove_one_section()
    mem_hotplug_done()

The comment above the pgdat_resize_lock also mentions "Holding this will
also guarantee that any pfn_valid() stays that way.", which is true with
the current implementation and false after this patch.  But current
implementation doesn't meet this comment.  There isn't any pfn walkers to
take the lock so this looks like a relict from the past.  This patch also
removes this comment.

[richard.weiyang@gmail.com: v4]
Link: http://lkml.kernel.org/r/20181204085657.20472-1-richard.weiyang@gmail.com
[mhocko@suse.com: changelog suggestion]
Link: http://lkml.kernel.org/r/20181128091243.19249-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove pte_lock_deinit()

Pagetable page doesn't touch page->mapping or have any used field that
overlaps with it.  No need to clear mapping in dtor.  In fact, doing so
might mask problems that otherwise would be detected by bad_page().

Link: http://lkml.kernel.org/r/20181128235525.58780-1-yuzhao@google.com
Signed-off-by: Yu Zhao <yuzhao@google.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Keith Busch <keith.busch@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: writeback throttle

If there are lots of write IO with flash device, it could have a
wearout problem of storage. To overcome the problem, admin needs
to design write limitation to guarantee flash health
for entire product life.

This patch creates a new knob "writeback_limit" for zram.

writeback_limit's default value is 0 so that it doesn't limit
any writeback. If admin want to measure writeback count in a
certain period, he could know it via /sys/block/zram0/bd_stat's
3rd column.

If admin want to limit writeback as per-day 400M, he could do it
like below.

MB_SHIFT=20
4K_SHIFT=12
echo $((400<<MB_SHIFT>>4K_SHIFT)) > \
/sys/block/zram0/writeback_limit.

If admin want to allow further write again, he could do it like below

echo 0 > /sys/block/zram0/writeback_limit

If admin want to see remaining writeback budget,

cat /sys/block/zram0/writeback_limit

The writeback_limit count will reset whenever you reset zram (e.g., system
reboot, echo 1 > /sys/block/zramX/reset) so keeping how many of writeback
happened until you reset the zram to allocate extra writeback budget in
next setting is user's job.

[minchan@kernel.org: v4]
Link: http://lkml.kernel.org/r/20181203024045.153534-8-minchan@kernel.org
Link: http://lkml.kernel.org/r/20181127055429.251614-8-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: add bd_stat statistics

bd_stat represents things that happened in the backing device.  Currently
it supports bd_counts, bd_reads and bd_writes which are helpful to
understand wearout of flash and memory saving.

[minchan@kernel.org: v4]
Link: http://lkml.kernel.org/r/20181203024045.153534-7-minchan@kernel.org
Link: http://lkml.kernel.org/r/20181127055429.251614-7-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: support idle/huge page writeback

Add a new feature "zram idle/huge page writeback".  In the zram-swap use
case, zram usually has many idle/huge swap pages.  It's pointless to keep
them in memory (ie, zram).

To solve this problem, this feature introduces idle/huge page writeback to
the backing device so the goal is to save more memory space on embedded
systems.

Normal sequence to use idle/huge page writeback feature is as follows,

while (1) {
        # mark allocated zram slot to idle
        echo all > /sys/block/zram0/idle
        # leave system working for several hours
        # Unless there is no access for some blocks on zram,
# they are still IDLE marked pages.

        echo "idle" > /sys/block/zram0/writeback
or/and
echo "huge" > /sys/block/zram0/writeback
        # write the IDLE or/and huge marked slot into backing device
# and free the memory.
}

Per the discussion at
https://lore.kernel.org/lkml/20181122065926.GG3441@jagdpanzerIV/T/#u,

This patch removes direct incommpressibe page writeback feature
(d2afd25114f4 ("zram: write incompressible pages to backing device")).

Below concerns from Sergey:
== &< ==

"IDLE writeback" is superior to "incompressible writeback".

"incompressible writeback" is completely unpredictable and uncontrollable;
it depens on data patterns and compression algorithms.  While "IDLE
writeback" is predictable.

I even suspect, that, *ideally*, we can remove "incompressible writeback".
"IDLE pages" is a super set which also includes "incompressible" pages.
So, technically, we still can do "incompressible writeback" from "IDLE
writeback" path; but a much more reasonable one, based on a page idling
period.

I understand that you want to keep "direct incompressible writeback"
around.  ZRAM is especially popular on devices which do suffer from flash
wearout, so I can see "incompressible writeback" path becoming a dead
code, long term.

== &< ==

Below concerns from Minchan:
== &< ==

My concern is if we enable CONFIG_ZRAM_WRITEBACK in this implementation,
both hugepage/idlepage writeck will turn on.  However someuser want to
enable only idlepage writeback so we need to introduce turn on/off knob
for hugepage or new CONFIG_ZRAM_IDLEPAGE_WRITEBACK for those usecase.  I
don't want to make it complicated *if possible*.

Long term, I imagine we need to make VM aware of new swap hierarchy a
little bit different with as-is.  For example, first high priority swap
can return -EIO or -ENOCOMP, swap try to fallback to next lower priority
swap device.  With that, hugepage writeback will work tranparently.

So we could regard it as regression because incompressible pages doesn't
go to backing storage automatically.  Instead, user should do it via "echo
huge" > /sys/block/zram/writeback" manually.

== &< ==

Link: http://lkml.kernel.org/r/20181127055429.251614-6-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Joey Pabalinas <joeypabalinas@gmail.com>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: introduce ZRAM_IDLE flag

To support idle page writeback with upcoming patches, this patch
introduces a new ZRAM_IDLE flag.

Userspace can mark zram slots as "idle" via
"echo all > /sys/block/zramX/idle"
which marks every allocated zram slot as ZRAM_IDLE.
User could see it by /sys/kernel/debug/zram/zram0/block_state.

          300    75.033841 ...i
          301    63.806904 s..i
          302    63.806919 ..hi

Once there is IO for the slot, the mark will be disappeared.

  300    75.033841 ...
          301    63.806904 s..i
          302    63.806919 ..hi

Therefore, 300th block is idle zpage. With this feature,
user can how many zram has idle pages which are waste of memory.

Link: http://lkml.kernel.org/r/20181127055429.251614-5-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: refactor flags and writeback stuff

Rename some variables and restructure some code for better readability in
writeback and zs_free_page.

Link: http://lkml.kernel.org/r/20181127055429.251614-4-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: fix double free backing device

If blkdev_get fails, we shouldn't do blkdev_put.  Otherwise, kernel emits
below log.  This patch fixes it.

  WARNING: CPU: 0 PID: 1893 at fs/block_dev.c:1828 blkdev_put+0x105/0x120
  Modules linked in:
  CPU: 0 PID: 1893 Comm: swapoff Not tainted 4.19.0+ #453
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
  RIP: 0010:blkdev_put+0x105/0x120
  Call Trace:
    __x64_sys_swapoff+0x46d/0x490
    do_syscall_64+0x5a/0x190
    entry_SYSCALL_64_after_hwframe+0x49/0xbe
  irq event stamp: 4466
  hardirqs last  enabled at (4465):  __free_pages_ok+0x1e3/0x490
  hardirqs last disabled at (4466):  trace_hardirqs_off_thunk+0x1a/0x1c
  softirqs last  enabled at (3420):  __do_softirq+0x333/0x446
  softirqs last disabled at (3407):  irq_exit+0xd1/0xe0

Link: http://lkml.kernel.org/r/20181127055429.251614-3-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Joey Pabalinas <joeypabalinas@gmail.com>
Cc: <stable@vger.kernel.org> [4.14+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: fix lockdep warning of free block handling

Patch series "zram idle page writeback", v3.

Inherently, swap device has many idle pages which are rare touched since
it was allocated.  It is never problem if we use storage device as swap.
However, it's just waste for zram-swap.

This patchset supports zram idle page writeback feature.

* Admin can define what is idle page "no access since X time ago"
* Admin can define when zram should writeback them
* Admin can define when zram should stop writeback to prevent wearout

Details are in each patch's description.

This patch (of 7):

  ================================
  WARNING: inconsistent lock state
  4.19.0+ #390 Not tainted
  --------------------------------
  inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
  zram_verify/2095 [HC0[0]:SC1[1]:HE1:SE0] takes:
  00000000b1828693 (&(&zram->bitmap_lock)->rlock){+.?.}, at: put_entry_bdev+0x1e/0x50
  {SOFTIRQ-ON-W} state was registered at:
    _raw_spin_lock+0x2c/0x40
    zram_make_request+0x755/0xdc9
    generic_make_request+0x373/0x6a0
    submit_bio+0x6c/0x140
    __swap_writepage+0x3a8/0x480
    shrink_page_list+0x1102/0x1a60
    shrink_inactive_list+0x21b/0x3f0
    shrink_node_memcg.constprop.99+0x4f8/0x7e0
    shrink_node+0x7d/0x2f0
    do_try_to_free_pages+0xe0/0x300
    try_to_free_pages+0x116/0x2b0
    __alloc_pages_slowpath+0x3f4/0xf80
    __alloc_pages_nodemask+0x2a2/0x2f0
    __handle_mm_fault+0x42e/0xb50
    handle_mm_fault+0x55/0xb0
    __do_page_fault+0x235/0x4b0
    page_fault+0x1e/0x30
  irq event stamp: 228412
  hardirqs last  enabled at (228412): [<ffffffff98245846>] __slab_free+0x3e6/0x600
  hardirqs last disabled at (228411): [<ffffffff98245625>] __slab_free+0x1c5/0x600
  softirqs last  enabled at (228396): [<ffffffff98e0031e>] __do_softirq+0x31e/0x427
  softirqs last disabled at (228403): [<ffffffff98072051>] irq_exit+0xd1/0xe0

  other info that might help us debug this:
   Possible unsafe locking scenario:

         CPU0
         ----
    lock(&(&zram->bitmap_lock)->rlock);
    <Interrupt>
      lock(&(&zram->bitmap_lock)->rlock);

   *** DEADLOCK ***

  no locks held by zram_verify/2095.

  stack backtrace:
  CPU: 5 PID: 2095 Comm: zram_verify Not tainted 4.19.0+ #390
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
  Call Trace:
   <IRQ>
   dump_stack+0x67/0x9b
   print_usage_bug+0x1bd/0x1d3
   mark_lock+0x4aa/0x540
   __lock_acquire+0x51d/0x1300
   lock_acquire+0x90/0x180
   _raw_spin_lock+0x2c/0x40
   put_entry_bdev+0x1e/0x50
   zram_free_page+0xf6/0x110
   zram_slot_free_notify+0x42/0xa0
   end_swap_bio_read+0x5b/0x170
   blk_update_request+0x8f/0x340
   scsi_end_request+0x2c/0x1e0
   scsi_io_completion+0x98/0x650
   blk_done_softirq+0x9e/0xd0
   __do_softirq+0xcc/0x427
   irq_exit+0xd1/0xe0
   do_IRQ+0x93/0x120
   common_interrupt+0xf/0xf
   </IRQ>

With writeback feature, zram_slot_free_notify could be called in softirq
context by end_swap_bio_read.  However, bitmap_lock is not aware of that
so lockdep yell out:

  get_entry_bdev
  spin_lock(bitmap->lock);
  irq
  softirq
  end_swap_bio_read
  zram_slot_free_notify
  zram_slot_lock <-- deadlock prone
  zram_free_page
  put_entry_bdev
  spin_lock(bitmap->lock); <-- deadlock prone

With akpm's suggestion (i.e.  bitmap operation is already atomic), we
could remove bitmap lock.  It might fail to find a empty slot if serious
contention happens.  However, it's not severe problem because huge page
writeback has already possiblity to fail if there is severe memory
pressure.  Worst case is just keeping the incompressible in memory, not
storage.

The other problem is zram_slot_lock in zram_slot_slot_free_notify.  To
make it safe is this patch introduces zram_slot_trylock where
zram_slot_free_notify uses it.  Although it's rare to be contented, this
patch adds new debug stat "miss_free" to keep monitoring how often it
happens.

Link: http://lkml.kernel.org/r/20181127055429.251614-2-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memblock.c: skip kmemleak for kasan_init()

Kmemleak does not play well with KASAN (tested on both HPE Apollo 70 and
Huawei TaiShan 2280 aarch64 servers).

After calling start_kernel()->setup_arch()->kasan_init(), kmemleak early
log buffer went from something like 280 to 260000 which caused kmemleak
disabled and crash dump memory reservation failed.  The multitude of
kmemleak_alloc() calls is from nested loops while KASAN is setting up full
memory mappings, so let early kmemleak allocations skip those
memblock_alloc_internal() calls came from kasan_init() given that those
early KASAN memory mappings should not reference to other memory.  Hence,
no kmemleak false positives.

kasan_init
  kasan_map_populate [1]
    kasan_pgd_populate [2]
      kasan_pud_populate [3]
        kasan_pmd_populate [4]
          kasan_pte_populate [5]
            kasan_alloc_zeroed_page
              memblock_alloc_try_nid
                memblock_alloc_internal
                  kmemleak_alloc

[1] for_each_memblock(memory, reg)
[2] while (pgdp++, addr = next, addr != end)
[3] while (pudp++, addr = next, addr != end && pud_none(READ_ONCE(*pudp)))
[4] while (pmdp++, addr = next, addr != end && pmd_none(READ_ONCE(*pmdp)))
[5] while (ptep++, addr = next, addr != end && pte_none(READ_ONCE(*ptep)))

Link: http://lkml.kernel.org/r/1543442925-17794-1-git-send-email-cai@gmx.us
Signed-off-by: Qian Cai <cai@gmx.us>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel, resource: check for IORESOURCE_SYSRAM in release_mem_region_adjustable

This is a preparation for the next patch.

Currently, we only call release_mem_region_adjustable() in __remove_pages
if the zone is not ZONE_DEVICE, because resources that belong to HMM/devm
are being released by themselves with devm_release_mem_region.

Since we do not want to touch any zone/page stuff during the removing of
the memory (but during the offlining), we do not want to check for the
zone here.  So we need another way to tell release_mem_region_adjustable()
to not realease the resource in case it belongs to HMM/devm.

HMM/devm acquires/releases a resource through
devm_request_mem_region/devm_release_mem_region.

These resources have the flag IORESOURCE_MEM, while resources acquired by
hot-add memory path (register_memory_resource()) contain
IORESOURCE_SYSTEM_RAM.

So, we can check for this flag in release_mem_region_adjustable, and if
the resource does not contain such flag, we know that we are dealing with
a HMM/devm resource, so we can back off.

Link: http://lkml.kernel.org/r/20181127162005.15833-3-osalvador@suse.de
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.com>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: add nid parameter to arch_remove_memory

Patch series "Do not touch pages in hot-remove path", v2.

This patchset aims for two things:

 1) A better definition about offline and hot-remove stage
 2) Solving bugs where we can access non-initialized pages
    during hot-remove operations [2] [3].

This is achieved by moving all page/zone handling to the offline
stage, so we do not need to access pages when hot-removing memory.

[1] https://patchwork.kernel.org/cover/10691415/
[2] https://patchwork.kernel.org/patch/10547445/
[3] https://www.spinics.net/lists/linux-mm/msg161316.html

This patch (of 5):

This is a preparation for the following-up patches.  The idea of passing
the nid is that it will allow us to get rid of the zone parameter
afterwards.

Link: http://lkml.kernel.org/r/20181127162005.15833-2-osalvador@suse.de
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: check nr_initialised with PAGES_PER_SECTION directly in defer_init()

When DEFERRED_STRUCT_PAGE_INIT is configured, only the first section of
each node's highest zone is initialized before defer stage.

static_init_pgcnt is used to store the number of pages like this:

    pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION,
                                              pgdat->node_spanned_pages);

because we don't want to overflow zone's range.

But this is not necessary, since defer_init() is called like this:

  memmap_init_zone()
    for pfn in [start_pfn, end_pfn)
      defer_init(pfn, end_pfn)

In case (pgdat->node_spanned_pages < PAGES_PER_SECTION), the loop would
stop before calling defer_init().

BTW, comparing PAGES_PER_SECTION with node_spanned_pages is not correct,
since nr_initialised is zone based instead of node based.  Even
node_spanned_pages is bigger than PAGES_PER_SECTION, its highest zone
would have pages less than PAGES_PER_SECTION.

Link: http://lkml.kernel.org/r/20181122094807.6985-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: put_and_wait_on_page_locked() while page is migrated

Waiting on a page migration entry has used wait_on_page_locked() all along
since 2006: but you cannot safely wait_on_page_locked() without holding a
reference to the page, and that extra reference is enough to make
migrate_page_move_mapping() fail with -EAGAIN, when a racing task faults
on the entry before migrate_page_move_mapping() gets there.

And that failure is retried nine times, amplifying the pain when trying to
migrate a popular page.  With a single persistent faulter, migration
sometimes succeeds; with two or three concurrent faulters, success becomes
much less likely (and the more the page was mapped, the worse the overhead
of unmapping and remapping it on each try).

This is especially a problem for memory offlining, where the outer level
retries forever (or until terminated from userspace), because a heavy
refault workload can trigger an endless loop of migration failures.
wait_on_page_locked() is the wrong tool for the job.

David Herrmann (but was he the first?) noticed this issue in 2014:
https://marc.info/?l=linux-mm&m=140110465608116&w=2

Tim Chen started a thread in August 2017 which appears relevant:
https://marc.info/?l=linux-mm&m=150275941014915&w=2 where Kan Liang went
on to implicate __migration_entry_wait():
https://marc.info/?l=linux-mm&m=150300268411980&w=2 and the thread ended
up with the v4.14 commits: 2554db916586 ("sched/wait: Break up long wake
list walk") 11a19c7b099f ("sched/wait: Introduce wakeup boomark in
wake_up_page_bit")

Baoquan He reported "Memory hotplug softlock issue" 14 November 2018:
https://marc.info/?l=linux-mm&m=154217936431300&w=2

We have all assumed that it is essential to hold a page reference while
waiting on a page lock: partly to guarantee that there is still a struct
page when MEMORY_HOTREMOVE is configured, but also to protect against
reuse of the struct page going to someone who then holds the page locked
indefinitely, when the waiter can reasonably expect timely unlocking.

But in fact, so long as wait_on_page_bit_common() does the put_page(), and
is careful not to rely on struct page contents thereafter, there is no
need to hold a reference to the page while waiting on it.  That does mean
that this case cannot go back through the loop: but that's fine for the
page migration case, and even if used more widely, is limited by the "Stop
walking if it's locked" optimization in wake_page_function().

Add interface put_and_wait_on_page_locked() to do this, using "behavior"
enum in place of "lock" arg to wait_on_page_bit_common() to implement it.
No interruptible or killable variant needed yet, but they might follow: I
have a vague notion that reporting -EINTR should take precedence over
return from wait_on_page_bit_common() without knowing the page state, so
arrange it accordingly - but that may be nothing but pedantic.

__migration_entry_wait() still has to take a brief reference to the page,
prior to calling put_and_wait_on_page_locked(): but now that it is dropped
before waiting, the chance of impeding page migration is very much
reduced.  Should we perhaps disable preemption across this?

shrink_page_list()'s __ClearPageLocked(): that was a surprise!  This
survived a lot of testing before that showed up.  PageWaiters may have
been set by wait_on_page_bit_common(), and the reference dropped, just
before shrink_page_list() succeeds in freezing its last page reference: in
such a case, unlock_page() must be used.  Follow the suggestion from
Michal Hocko, just revert a978d6f52106 ("mm: unlockless reclaim") now:
that optimization predates PageWaiters, and won't buy much these days; but
we can reinstate it for the !PageWaiters case if anyone notices.

It does raise the question: should vmscan.c's is_page_cache_freeable() and
__remove_mapping() now treat a PageWaiters page as if an extra reference
were held?  Perhaps, but I don't think it matters much, since
shrink_page_list() already had to win its trylock_page(), so waiters are
not very common there: I noticed no difference when trying the bigger
change, and it's surely not needed while put_and_wait_on_page_locked() is
only used for page migration.

[willy@infradead.org: add put_and_wait_on_page_locked() kerneldoc]
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1811261121330.1116@eggly.anvils
Signed-off-by: Hugh Dickins <hughd@google.com>
Reported-by: Baoquan He <bhe@redhat.com>
Tested-by: Baoquan He <bhe@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: David Herrmann <dh.herrmann@gmail.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Nick Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: add oom victim's memcg to the oom context information

The current oom report doesn't display victim's memcg context during the
global OOM situation.  While this information is not strictly needed, it
can be really helpful for containerized environments to locate which
container has lost a process.  Now that we have a single line for the oom
context, we can trivially add both the oom memcg (this can be either
global_oom or a specific memcg which hits its hard limits) and task_memcg
which is the victim's memcg.

Below is the single line output in the oom report after this patch.

- global oom context information:

oom-kill:constraint=<constraint>,nodemask=<nodemask>,cpuset=<cpuset>,mems_allowed=<mems_allowed>,global_oom,task_memcg=<memcg>,task=<comm>,pid=<pid>,uid=<uid>

- memcg oom context information:

oom-kill:constraint=<constraint>,nodemask=<nodemask>,cpuset=<cpuset>,mems_allowed=<mems_allowed>,oom_memcg=<memcg>,task_memcg=<memcg>,task=<comm>,pid=<pid>,uid=<uid>

[penguin-kernel@I-love.SAKURA.ne.jp: use pr_cont() in mem_cgroup_print_oom_context()]
Link: http://lkml.kernel.org/r/201812190723.wBJ7NdkN032628@www262.sakura.ne.jp
Link: http://lkml.kernel.org/r/1542799799-36184-2-git-send-email-ufo19890607@gmail.com
Signed-off-by: yuzhoujian <yuzhoujian@didichuxing.com>
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: Roman Gushchin <guro@fb.com>
Cc: Yang Shi <yang.s@alibaba-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: reorganize the oom report in dump_header

OOM report contains several sections.  The first one is the allocation
context that has triggered the OOM.  Then we have cpuset context followed
by the stack trace of the OOM path.  The tird one is the OOM memory
information.  Followed by the current memory state of all system tasks.
At last, we will show oom eligible tasks and the information about the
chosen oom victim.

One thing that makes parsing more awkward than necessary is that we do not
have a single and easily parsable line about the oom context.  This patch
is reorganizing the oom report to

1) who invoked oom and what was the allocation request

[  515.902945] tuned invoked oom-killer: gfp_mask=0x6200ca(GFP_HIGHUSER_MOVABLE), order=0, oom_score_adj=0

2) OOM stack trace

[  515.904273] CPU: 24 PID: 1809 Comm: tuned Not tainted 4.20.0-rc3+ #3
[  515.905518] Hardware name: Inspur SA5212M4/YZMB-00370-107, BIOS 4.1.10 11/14/2016
[  515.906821] Call Trace:
[  515.908062]  dump_stack+0x5a/0x73
[  515.909311]  dump_header+0x55/0x28c
[  515.914260]  oom_kill_process+0x2d8/0x300
[  515.916708]  out_of_memory+0x145/0x4a0
[  515.917932]  __alloc_pages_slowpath+0x7d2/0xa16
[  515.919157]  __alloc_pages_nodemask+0x277/0x290
[  515.920367]  filemap_fault+0x3d0/0x6c0
[  515.921529]  ? filemap_map_pages+0x2b8/0x420
[  515.922709]  ext4_filemap_fault+0x2c/0x40 [ext4]
[  515.923884]  __do_fault+0x20/0x80
[  515.925032]  __handle_mm_fault+0xbc0/0xe80
[  515.926195]  handle_mm_fault+0xfa/0x210
[  515.927357]  __do_page_fault+0x233/0x4c0
[  515.928506]  do_page_fault+0x32/0x140
[  515.929646]  ? page_fault+0x8/0x30
[  515.930770]  page_fault+0x1e/0x30

3) OOM memory information

[  515.958093] Mem-Info:
[  515.959647] active_anon:26501758 inactive_anon:1179809 isolated_anon:0
 active_file:4402672 inactive_file:483963 isolated_file:1344
 unevictable:0 dirty:4886753 writeback:0 unstable:0
 slab_reclaimable:148442 slab_unreclaimable:18741
 mapped:1347 shmem:1347 pagetables:58669 bounce:0
 free:88663 free_pcp:0 free_cma:0
...

4) current memory state of all system tasks

[  516.079544] [    744]     0   744     9211     1345   114688       82             0 systemd-journal
[  516.082034] [    787]     0   787    31764        0   143360       92             0 lvmetad
[  516.084465] [    792]     0   792    10930        1   110592      208         -1000 systemd-udevd
[  516.086865] [   1199]     0  1199    13866        0   131072      112         -1000 auditd
[  516.089190] [   1222]     0  1222    31990        1   110592      157             0 smartd
[  516.091477] [   1225]     0  1225     4864       85    81920       43             0 irqbalance
[  516.093712] [   1226]     0  1226    52612        0   258048      426             0 abrtd
[  516.112128] [   1280]     0  1280   109774       55   299008      400             0 NetworkManager
[  516.113998] [   1295]     0  1295    28817       37    69632       24             0 ksmtuned
[  516.144596] [  10718]     0 10718  2622484  1721372 15998976   267219             0 panic
[  516.145792] [  10719]     0 10719  2622484  1164767  9818112    53576             0 panic
[  516.146977] [  10720]     0 10720  2622484  1174361  9904128    53709             0 panic
[  516.148163] [  10721]     0 10721  2622484  1209070 10194944    54824             0 panic
[  516.149329] [  10722]     0 10722  2622484  1745799 14774272    91138             0 panic

5) oom context (contrains and the chosen victim).

oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0-1,task=panic,pid=10737,uid=0

An admin can easily get the full oom context at a single line which
makes parsing much easier.

Link: http://lkml.kernel.org/r/1542799799-36184-1-git-send-email-ufo19890607@gmail.com
Signed-off-by: yuzhoujian <yuzhoujian@didichuxing.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: Yang Shi <yang.s@alibaba-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make free_reserved_area() return "const char *"

and propagate through down the call stack.

Link: http://lkml.kernel.org/r/20181124091411.GC10969@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/debug.c: make "migrate_reason_names[]" const char *

Those strings are immutable as well.

Link: http://lkml.kernel.org/r/20181124090508.GB10877@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmzone.c: make "migratetype_names" const char *

Those strings are immutable in fact.

Link: http://lkml.kernel.org/r/20181124090327.GA10877@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: reclaim small amounts of memory when an external fragmentation event occurs

An external fragmentation event was previously described as

    When the page allocator fragments memory, it records the event using
    the mm_page_alloc_extfrag event. If the fallback_order is smaller
    than a pageblock order (order-9 on 64-bit x86) then it's considered
    an event that will cause external fragmentation issues in the future.

The kernel reduces the probability of such events by increasing the
watermark sizes by calling set_recommended_min_free_kbytes early in the
lifetime of the system.  This works reasonably well in general but if
there are enough sparsely populated pageblocks then the problem can still
occur as enough memory is free overall and kswapd stays asleep.

This patch introduces a watermark_boost_factor sysctl that allows a zone
watermark to be temporarily boosted when an external fragmentation causing
events occurs.  The boosting will stall allocations that would decrease
free memory below the boosted low watermark and kswapd is woken if the
calling context allows to reclaim an amount of memory relative to the size
of the high watermark and the watermark_boost_factor until the boost is
cleared.  When kswapd finishes, it wakes kcompactd at the pageblock order
to clean some of the pageblocks that may have been affected by the
fragmentation event.  kswapd avoids any writeback, slab shrinkage and swap
from reclaim context during this operation to avoid excessive system
disruption in the name of fragmentation avoidance.  Care is taken so that
kswapd will do normal reclaim work if the system is really low on memory.

This was evaluated using the same workloads as "mm, page_alloc: Spread
allocations across zones before introducing fragmentation".

1-socket Skylake machine
config-global-dhp__workload_thpfioscale XFS (no special madvise)
4 fio threads, 1 THP allocating thread
--------------------------------------

4.20-rc3 extfrag events < order 9:   804694
4.20-rc3+patch:                      408912 (49% reduction)
4.20-rc3+patch1-4:                    18421 (98% reduction)

                                   4.20.0-rc3             4.20.0-rc3
                                 lowzone-v5r8             boost-v5r8
Amean     fault-base-1      653.58 (   0.00%)      652.71 (   0.13%)
Amean     fault-huge-1        0.00 (   0.00%)      178.93 * -99.00%*

                              4.20.0-rc3             4.20.0-rc3
                            lowzone-v5r8             boost-v5r8
Percentage huge-1        0.00 (   0.00%)        5.12 ( 100.00%)

Note that external fragmentation causing events are massively reduced by
this path whether in comparison to the previous kernel or the vanilla
kernel.  The fault latency for huge pages appears to be increased but that
is only because THP allocations were successful with the patch applied.

1-socket Skylake machine
global-dhp__workload_thpfioscale-madvhugepage-xfs (MADV_HUGEPAGE)
-----------------------------------------------------------------

4.20-rc3 extfrag events < order 9:  291392
4.20-rc3+patch:                     191187 (34% reduction)
4.20-rc3+patch1-4:                   13464 (95% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                 lowzone-v5r8             boost-v5r8
Min       fault-base-1      912.00 (   0.00%)      905.00 (   0.77%)
Min       fault-huge-1      127.00 (   0.00%)      135.00 (  -6.30%)
Amean     fault-base-1     1467.55 (   0.00%)     1481.67 (  -0.96%)
Amean     fault-huge-1     1127.11 (   0.00%)     1063.88 *   5.61%*

                              4.20.0-rc3             4.20.0-rc3
                            lowzone-v5r8             boost-v5r8
Percentage huge-1       77.64 (   0.00%)       83.46 (   7.49%)

As before, massive reduction in external fragmentation events, some jitter
on latencies and an increase in THP allocation success rates.

2-socket Haswell machine
config-global-dhp__workload_thpfioscale XFS (no special madvise)
4 fio threads, 5 THP allocating threads
----------------------------------------------------------------

4.20-rc3 extfrag events < order 9:  215698
4.20-rc3+patch:                     200210 (7% reduction)
4.20-rc3+patch1-4:                   14263 (93% reduction)

                                   4.20.0-rc3             4.20.0-rc3
                                 lowzone-v5r8             boost-v5r8
Amean     fault-base-5     1346.45 (   0.00%)     1306.87 (   2.94%)
Amean     fault-huge-5     3418.60 (   0.00%)     1348.94 (  60.54%)

                              4.20.0-rc3             4.20.0-rc3
                            lowzone-v5r8             boost-v5r8
Percentage huge-5        0.78 (   0.00%)        7.91 ( 910.64%)

There is a 93% reduction in fragmentation causing events, there is a big
reduction in the huge page fault latency and allocation success rate is
higher.

2-socket Haswell machine
global-dhp__workload_thpfioscale-madvhugepage-xfs (MADV_HUGEPAGE)
-----------------------------------------------------------------

4.20-rc3 extfrag events < order 9: 166352
4.20-rc3+patch:                    147463 (11% reduction)
4.20-rc3+patch1-4:                  11095 (93% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                 lowzone-v5r8             boost-v5r8
Amean     fault-base-5     6217.43 (   0.00%)     7419.67 * -19.34%*
Amean     fault-huge-5     3163.33 (   0.00%)     3263.80 (  -3.18%)

                              4.20.0-rc3             4.20.0-rc3
                            lowzone-v5r8             boost-v5r8
Percentage huge-5       95.14 (   0.00%)       87.98 (  -7.53%)

There is a large reduction in fragmentation events with some jitter around
the latencies and success rates.  As before, the high THP allocation
success rate does mean the system is under a lot of pressure.  However, as
the fragmentation events are reduced, it would be expected that the
long-term allocation success rate would be higher.

Link: http://lkml.kernel.org/r/20181123114528.28802-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Zi Yan <zi.yan@cs.rutgers.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use alloc_flags to record if kswapd can wake

This is a preparation patch that copies the GFP flag __GFP_KSWAPD_RECLAIM
into alloc_flags.  This is a preparation patch only that avoids having to
pass gfp_mask through a long callchain in a future patch.

Note that the setting in the fast path happens in alloc_flags_nofragment()
and it may be claimed that this has nothing to do with ALLOC_NO_FRAGMENT.
That's true in this patch but is not true later so it's done now for
easier review to show where the flag needs to be recorded.

No functional change.

[mgorman@techsingularity.net: ALLOC_KSWAPD flag needs to be applied in the !CONFIG_ZONE_DMA32 case]
Link: http://lkml.kernel.org/r/20181126143503.GO23260@techsingularity.net
Link: http://lkml.kernel.org/r/20181123114528.28802-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zi Yan <zi.yan@cs.rutgers.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move zone watermark accesses behind an accessor

This is a preparation patch only, no functional change.

Link: http://lkml.kernel.org/r/20181123114528.28802-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Zi Yan <zi.yan@cs.rutgers.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: spread allocations across zones before introducing fragmentation

Patch series "Fragmentation avoidance improvements", v5.

It has been noted before that fragmentation avoidance (aka
anti-fragmentation) is not perfect. Given sufficient time or an adverse
workload, memory gets fragmented and the long-term success of high-order
allocations degrades. This series defines an adverse workload, a definition
of external fragmentation events (including serious) ones and a series
that reduces the level of those fragmentation events.

The details of the workload and the consequences are described in more
detail in the changelogs. However, from patch 1, this is a high-level
summary of the adverse workload. The exact details are found in the
mmtests implementation.

The broad details of the workload are as follows;

1. Create an XFS filesystem (not specified in the configuration but done
   as part of the testing for this patch)
2. Start 4 fio threads that write a number of 64K files inefficiently.
   Inefficiently means that files are created on first access and not
   created in advance (fio parameterr create_on_open=1) and fallocate
   is not used (fallocate=none). With multiple IO issuers this creates
   a mix of slab and page cache allocations over time. The total size
   of the files is 150% physical memory so that the slabs and page cache
   pages get mixed
3. Warm up a number of fio read-only threads accessing the same files
   created in step 2. This part runs for the same length of time it
   took to create the files. It'll fault back in old data and further
   interleave slab and page cache allocations. As it's now low on
   memory due to step 2, fragmentation occurs as pageblocks get
   stolen.
4. While step 3 is still running, start a process that tries to allocate
   75% of memory as huge pages with a number of threads. The number of
   threads is based on a (NR_CPUS_SOCKET - NR_FIO_THREADS)/4 to avoid THP
   threads contending with fio, any other threads or forcing cross-NUMA
   scheduling. Note that the test has not been used on a machine with less
   than 8 cores. The benchmark records whether huge pages were allocated
   and what the fault latency was in microseconds
5. Measure the number of events potentially causing external fragmentation,
   the fault latency and the huge page allocation success rate.
6. Cleanup

Overall the series reduces external fragmentation causing events by over 94%
on 1 and 2 socket machines, which in turn impacts high-order allocation
success rates over the long term. There are differences in latencies and
high-order allocation success rates. Latencies are a mixed bag as they
are vulnerable to exact system state and whether allocations succeeded
so they are treated as a secondary metric.

Patch 1 uses lower zones if they are populated and have free memory
instead of fragmenting a higher zone. It's special cased to
handle a Normal->DMA32 fallback with the reasons explained
in the changelog.

Patch 2-4 boosts watermarks temporarily when an external fragmentation
event occurs. kswapd wakes to reclaim a small amount of old memory
and then wakes kcompactd on completion to recover the system
slightly. This introduces some overhead in the slowpath. The level
of boosting can be tuned or disabled depending on the tolerance
for fragmentation vs allocation latency.

Patch 5 stalls some movable allocation requests to let kswapd from patch 4
make some progress. The duration of the stalls is very low but it
is possible to tune the system to avoid fragmentation events if
larger stalls can be tolerated.

The bulk of the improvement in fragmentation avoidance is from patches
1-4 but patch 5 can deal with a rare corner case and provides the option
of tuning a system for THP allocation success rates in exchange for
some stalls to control fragmentation.

This patch (of 5):

The page allocator zone lists are iterated based on the watermarks of each
zone which does not take anti-fragmentation into account.  On x86, node 0
may have multiple zones while other nodes have one zone.  A consequence is
that tasks running on node 0 may fragment ZONE_NORMAL even though
ZONE_DMA32 has plenty of free memory.  This patch special cases the
allocator fast path such that it'll try an allocation from a lower local
zone before fragmenting a higher zone.  In this case, stealing of
pageblocks or orders larger than a pageblock are still allowed in the fast
path as they are uninteresting from a fragmentation point of view.

This was evaluated using a benchmark designed to fragment memory before
attempting THP allocations.  It's implemented in mmtests as the following
configurations

configs/config-global-dhp__workload_thpfioscale
configs/config-global-dhp__workload_thpfioscale-defrag
configs/config-global-dhp__workload_thpfioscale-madvhugepage

e.g. from mmtests
./run-mmtests.sh --run-monitor --config configs/config-global-dhp__workload_thpfioscale test-run-1

The broad details of the workload are as follows;

1. Create an XFS filesystem (not specified in the configuration but done
   as part of the testing for this patch).
2. Start 4 fio threads that write a number of 64K files inefficiently.
   Inefficiently means that files are created on first access and not
   created in advance (fio parameter create_on_open=1) and fallocate
   is not used (fallocate=none). With multiple IO issuers this creates
   a mix of slab and page cache allocations over time. The total size
   of the files is 150% physical memory so that the slabs and page cache
   pages get mixed.
3. Warm up a number of fio read-only processes accessing the same files
   created in step 2. This part runs for the same length of time it
   took to create the files. It'll refault old data and further
   interleave slab and page cache allocations. As it's now low on
   memory due to step 2, fragmentation occurs as pageblocks get
   stolen.
4. While step 3 is still running, start a process that tries to allocate
   75% of memory as huge pages with a number of threads. The number of
   threads is based on a (NR_CPUS_SOCKET - NR_FIO_THREADS)/4 to avoid THP
   threads contending with fio, any other threads or forcing cross-NUMA
   scheduling. Note that the test has not been used on a machine with less
   than 8 cores. The benchmark records whether huge pages were allocated
   and what the fault latency was in microseconds.
5. Measure the number of events potentially causing external fragmentation,
   the fault latency and the huge page allocation success rate.
6. Cleanup the test files.

Note that due to the use of IO and page cache that this benchmark is not
suitable for running on large machines where the time to fragment memory
may be excessive.  Also note that while this is one mix that generates
fragmentation that it's not the only mix that generates fragmentation.
Differences in workload that are more slab-intensive or whether SLUB is
used with high-order pages may yield different results.

When the page allocator fragments memory, it records the event using the
mm_page_alloc_extfrag ftrace event.  If the fallback_order is smaller than
a pageblock order (order-9 on 64-bit x86) then it's considered to be an
"external fragmentation event" that may cause issues in the future.
Hence, the primary metric here is the number of external fragmentation
events that occur with order < 9.  The secondary metric is allocation
latency and huge page allocation success rates but note that differences
in latencies and what the success rate also can affect the number of
external fragmentation event which is why it's a secondary metric.

1-socket Skylake machine
config-global-dhp__workload_thpfioscale XFS (no special madvise)
4 fio threads, 1 THP allocating thread
--------------------------------------

4.20-rc3 extfrag events < order 9:   804694
4.20-rc3+patch:                      408912 (49% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                      vanilla           lowzone-v5r8
Amean     fault-base-1      662.92 (   0.00%)      653.58 *   1.41%*
Amean     fault-huge-1        0.00 (   0.00%)        0.00 (   0.00%)

                              4.20.0-rc3             4.20.0-rc3
                                 vanilla           lowzone-v5r8
Percentage huge-1        0.00 (   0.00%)        0.00 (   0.00%)

Fault latencies are slightly reduced while allocation success rates remain
at zero as this configuration does not make any special effort to allocate
THP and fio is heavily active at the time and either filling memory or
keeping pages resident.  However, a 49% reduction of serious fragmentation
events reduces the changes of external fragmentation being a problem in
the future.

Vlastimil asked during review for a breakdown of the allocation types
that are falling back.

vanilla
   3816 MIGRATE_UNMOVABLE
 800845 MIGRATE_MOVABLE
     33 MIGRATE_UNRECLAIMABLE

patch
    735 MIGRATE_UNMOVABLE
 408135 MIGRATE_MOVABLE
     42 MIGRATE_UNRECLAIMABLE

The majority of the fallbacks are due to movable allocations and this is
consistent for the workload throughout the series so will not be presented
again as the primary source of fallbacks are movable allocations.

Movable fallbacks are sometimes considered "ok" to fallback because they
can be migrated.  The problem is that they can fill an
unmovable/reclaimable pageblock causing those allocations to fallback
later and polluting pageblocks with pages that cannot move.  If there is a
movable fallback, it is pretty much guaranteed to affect an
unmovable/reclaimable pageblock and while it might not be enough to
actually cause a unmovable/reclaimable fallback in the future, we cannot
know that in advance so the patch takes the only option available to it.
Hence, it's important to control them.  This point is also consistent
throughout the series and will not be repeated.

1-socket Skylake machine
global-dhp__workload_thpfioscale-madvhugepage-xfs (MADV_HUGEPAGE)
-----------------------------------------------------------------

4.20-rc3 extfrag events < order 9:  291392
4.20-rc3+patch:                     191187 (34% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                      vanilla           lowzone-v5r8
Amean     fault-base-1     1495.14 (   0.00%)     1467.55 (   1.85%)
Amean     fault-huge-1     1098.48 (   0.00%)     1127.11 (  -2.61%)

thpfioscale Percentage Faults Huge
                              4.20.0-rc3             4.20.0-rc3
                                 vanilla           lowzone-v5r8
Percentage huge-1       78.57 (   0.00%)       77.64 (  -1.18%)

Fragmentation events were reduced quite a bit although this is known
to be a little variable. The latencies and allocation success rates
are similar but they were already quite high.

2-socket Haswell machine
config-global-dhp__workload_thpfioscale XFS (no special madvise)
4 fio threads, 5 THP allocating threads
----------------------------------------------------------------

4.20-rc3 extfrag events < order 9:  215698
4.20-rc3+patch:                     200210 (7% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                      vanilla           lowzone-v5r8
Amean     fault-base-5     1350.05 (   0.00%)     1346.45 (   0.27%)
Amean     fault-huge-5     4181.01 (   0.00%)     3418.60 (  18.24%)

                              4.20.0-rc3             4.20.0-rc3
                                 vanilla           lowzone-v5r8
Percentage huge-5        1.15 (   0.00%)        0.78 ( -31.88%)

The reduction of external fragmentation events is slight and this is
partially due to the removal of __GFP_THISNODE in commit ac5b2c18911f
("mm: thp: relax __GFP_THISNODE for MADV_HUGEPAGE mappings") as THP
allocations can now spill over to remote nodes instead of fragmenting
local memory.

2-socket Haswell machine
global-dhp__workload_thpfioscale-madvhugepage-xfs (MADV_HUGEPAGE)
-----------------------------------------------------------------

4.20-rc3 extfrag events < order 9: 166352
4.20-rc3+patch:                    147463 (11% reduction)

thpfioscale Fault Latencies
                                   4.20.0-rc3             4.20.0-rc3
                                      vanilla           lowzone-v5r8
Amean     fault-base-5     6138.97 (   0.00%)     6217.43 (  -1.28%)
Amean     fault-huge-5     2294.28 (   0.00%)     3163.33 * -37.88%*

thpfioscale Percentage Faults Huge
                              4.20.0-rc3             4.20.0-rc3
                                 vanilla           lowzone-v5r8
Percentage huge-5       96.82 (   0.00%)       95.14 (  -1.74%)

There was a slight reduction in external fragmentation events although the
latencies were higher.  The allocation success rate is high enough that
the system is struggling and there is quite a lot of parallel reclaim and
compaction activity.  There is also a certain degree of luck on whether
processes start on node 0 or not for this patch but the relevance is
reduced later in the series.

Overall, the patch reduces the number of external fragmentation causing
events so the success of THP over long periods of time would be improved
for this adverse workload.

Link: http://lkml.kernel.org/r/20181123114528.28802-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: drop "online" parameter from add_memory_resource()

Userspace should always be in charge of how to online memory and if memory
should be onlined automatically in the kernel.  Let's drop the parameter
to overwrite this - XEN passes memhp_auto_online, just like add_memory(),
so we can directly use that instead internally.

Link: http://lkml.kernel.org/r/20181123123740.27652-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Acked-by: Juergen Gross <jgross@suse.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Arun KS <arunks@codeaurora.org>
Cc: Mathieu Malaterre <malat@debian.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/base/memory.c: remove an unnecessary check on NR_MEM_SECTIONS

In cb5e39b8038b ("drivers: base: refactor add_memory_section() to
add_memory_block()"), add_memory_block() is introduced, which is only
invoked in memory_dev_init().

When combining these two loops in memory_dev_init() and
add_memory_block(), they looks like this:

    for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block)
        for (j = i;
    (j < i + sections_per_block) && j < NR_MEM_SECTIONS;
    j++)

Since it is sure the (i < NR_MEM_SECTIONS) and j sits in its own memory
block, the check of (j < NR_MEM_SECTIONS) is not necessary.

This patch just removes this check.

Link: http://lkml.kernel.org/r/20181123222811.18216-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Seth Jennings <sjenning@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: replace usage of __memblock_free_early() with memblock_free()

__memblock_free_early() is only used by the convenience wrappers, so
essentially we wrap a call to memblock_free() twice.  Replace calls of
__memblock_free_early() with calls to memblock_free() and drop the former.

Link: http://lkml.kernel.org/r/20181125102940.GE28634@rapoport-lnx
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Wentao Wang <witallwang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: deduplicate __memblock_free_early() and memblock_free()

Link: http://lkml.kernel.org/r/C8ECE1B7A767434691FEEFA3A01765D72AFB8E78@MX203CL03.corp.emc.com
Signed-off-by: Wentao Wang <witallwang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: use a single function to free page

There are multiple places of freeing a page, they all do the same things
so a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Link: http://lkml.kernel.org/r/20181119134834.17765-3-aaron.lu@intel.com
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Pankaj gupta <pagupta@redhat.com>
Cc: Pawel Staszewski <pstaszewski@itcare.pl>
Cc: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: free order-0 pages through PCP in page_frag_free()

page_frag_free() calls __free_pages_ok() to free the page back to Buddy.
This is OK for high order page, but for order-0 pages, it misses the
optimization opportunity of using Per-Cpu-Pages and can cause zone lock
contention when called frequently.

Pawel Staszewski recently shared his result of 'how Linux kernel handles
normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the
lock contention comes from page allocator:

  mlx5e_poll_tx_cq
  |
   --16.34%--napi_consume_skb
             |
             |--12.65%--__free_pages_ok
             |          |
             |           --11.86%--free_one_page
             |                     |
             |                     |--10.10%--queued_spin_lock_slowpath
             |                     |
             |                      --0.65%--_raw_spin_lock
             |
             |--1.55%--page_frag_free
             |
              --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is
not effective in this workload and pages have to go through the page
allocator.  The lock contention happens during mlx5 DMA TX completion
cycle.  And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and
thought this is an lock contention for high order pages but Jesper
explained in detail that __free_pages_ok() here are actually freeing
order-0 pages because mlx5 is using order-0 pages to satisfy its page pool
allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
  -> skb_free_frag()
    -> page_frag_free()
And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send
the being freed page to PCP if it's an order-0 page, or directly to Buddy
if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his
workload and Jesper has noticed a 7% performance improvement using a micro
benchmark and lock contention is gone.  Ilias' test on a 'low' speed 1Gbit
interface on an cortex-a53 shows ~11% performance boost testing with
64byte packets and __free_pages_ok() disappeared from perf top.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

[akpm@linux-foundation.org: add comment]
Link: http://lkml.kernel.org/r/20181120014544.GB10657@intel.com
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
Reported-by: Pawel Staszewski <pstaszewski@itcare.pl>
Analysed-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tested-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Acked-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Pankaj gupta <pagupta@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hmm: mark hmm_devmem_{add, add_resource} EXPORT_SYMBOL_GPL

At Maintainer Summit, Greg brought up a topic I proposed around
EXPORT_SYMBOL_GPL usage.  The motivation was considerations for when
EXPORT_SYMBOL_GPL is warranted and the criteria for taking the exceptional
step of reclassifying an existing export.  Specifically, I wanted to make
the case that although the line is fuzzy and hard to specify in abstract
terms, it is nonetheless clear that devm_memremap_pages() and HMM
(Heterogeneous Memory Management) have crossed it.  The
devm_memremap_pages() facility should have been EXPORT_SYMBOL_GPL from the
beginning, and HMM as a derivative of that functionality should have
naturally picked up that designation as well.

Contrary to typical rules, the HMM infrastructure was merged upstream with
zero in-tree consumers.  There was a promise at the time that those users
would be merged "soon", but it has been over a year with no drivers
arriving.  While the Nouveau driver is about to belatedly make good on
that promise it is clear that HMM was targeted first and foremost at an
out-of-tree consumer.

HMM is derived from devm_memremap_pages(), a facility Christoph and I
spearheaded to support persistent memory.  It combines a device lifetime
model with a dynamically created 'struct page' / memmap array for any
physical address range.  It enables coordination and control of the many
code paths in the kernel built to interact with memory via 'struct page'
objects.  With HMM the integration goes even deeper by allowing device
drivers to hook and manipulate page fault and page free events.

One interpretation of when EXPORT_SYMBOL is suitable is when it is
exporting stable and generic leaf functionality.  The
devm_memremap_pages() facility continues to see expanding use cases,
peer-to-peer DMA being the most recent, with no clear end date when it
will stop attracting reworks and semantic changes.  It is not suitable to
export devm_memremap_pages() as a stable 3rd party driver API due to the
fact that it is still changing and manipulates core behavior.  Moreover,
it is not in the best interest of the long term development of the core
memory management subsystem to permit any external driver to effectively
define its own system-wide memory management policies with no
encouragement to engage with upstream.

I am also concerned that HMM was designed in a way to minimize further
engagement with the core-MM.  That, with these hooks in place,
device-drivers are free to implement their own policies without much
consideration for whether and how the core-MM could grow to meet that
need.  Going forward not only should HMM be EXPORT_SYMBOL_GPL, but the
core-MM should be allowed the opportunity and stimulus to change and
address these new use cases as first class functionality.

Original changelog:

hmm_devmem_add(), and hmm_devmem_add_resource() duplicated
devm_memremap_pages() and are now simple now wrappers around the core
facility to inject a dev_pagemap instance into the global pgmap_radix and
hook page-idle events.  The devm_memremap_pages() interface is base
infrastructure for HMM.  HMM has more and deeper ties into the kernel
memory management implementation than base ZONE_DEVICE which is itself a
EXPORT_SYMBOL_GPL facility.

Originally, the HMM page structure creation routines copied the
devm_memremap_pages() code and reused ZONE_DEVICE.  A cleanup to unify the
implementations was discussed during the initial review:
http://lkml.iu.edu/hypermail/linux/kernel/1701.2/00812.html Recent work to
extend devm_memremap_pages() for the peer-to-peer-DMA facility enabled
this cleanup to move forward.

In addition to the integration with devm_memremap_pages() HMM depends on
other GPL-only symbols:

    mmu_notifier_unregister_no_release
    percpu_ref
    region_intersects
    __class_create

It goes further to consume / indirectly expose functionality that is not
exported to any other driver:

    alloc_pages_vma
    walk_page_range

HMM is derived from devm_memremap_pages(), and extends deep core-kernel
fundamentals. Similar to devm_memremap_pages(), mark its entry points
EXPORT_SYMBOL_GPL().

[logang@deltatee.com: PCI/P2PDMA: match interface changes to devm_memremap_pages()]
Link: http://lkml.kernel.org/r/20181130225911.2900-1-logang@deltatee.com
Link: http://lkml.kernel.org/r/154275560565.76910.15919297436557795278.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: "Jérôme Glisse" <jglisse@redhat.com>
Cc: Balbir Singh <bsingharora@gmail.com>,
Cc: Michal Hocko <mhocko@suse.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hmm: replace hmm_devmem_pages_create() with devm_memremap_pages()

Commit e8d513483300 ("memremap: change devm_memremap_pages interface to
use struct dev_pagemap") refactored devm_memremap_pages() to allow a
dev_pagemap instance to be supplied.  Passing in a dev_pagemap interface
simplifies the design of pgmap type drivers in that they can rely on
container_of() to lookup any private data associated with the given
dev_pagemap instance.

In addition to the cleanups this also gives hmm users multi-order-radix
improvements that arrived with commit ab1b597ee0e4 "mm,
devm_memremap_pages: use multi-order radix for ZONE_DEVICE lookups"

As part of the conversion to the devm_memremap_pages() method of
handling the percpu_ref relative to when pages are put, the percpu_ref
completion needs to move to hmm_devmem_ref_exit().  See 71389703839e
("mm, zone_device: Replace {get, put}_zone_device_page...") for details.

Link: http://lkml.kernel.org/r/154275560053.76910.10870962637383152392.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hmm: use devm semantics for hmm_devmem_{add, remove}

devm semantics arrange for resources to be torn down when
device-driver-probe fails or when device-driver-release completes.
Similar to devm_memremap_pages() there is no need to support an explicit
remove operation when the users properly adhere to devm semantics.

Note that devm_kzalloc() automatically handles allocating node-local
memory.

Link: http://lkml.kernel.org/r/154275559545.76910.9186690723515469051.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Cc: "Jérôme Glisse" <jglisse@redhat.com>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, devm_memremap_pages: add MEMORY_DEVICE_PRIVATE support

In preparation for consolidating all ZONE_DEVICE enabling via
devm_memremap_pages(), teach it how to handle the constraints of
MEMORY_DEVICE_PRIVATE ranges.

[jglisse@redhat.com: call move_pfn_range_to_zone for MEMORY_DEVICE_PRIVATE]
Link: http://lkml.kernel.org/r/154275559036.76910.12434636179931292607.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Reported-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, devm_memremap_pages: fix shutdown handling

The last step before devm_memremap_pages() returns success is to allocate
a release action, devm_memremap_pages_release(), to tear the entire setup
down.  However, the result from devm_add_action() is not checked.

Checking the error from devm_add_action() is not enough.  The api
currently relies on the fact that the percpu_ref it is using is killed by
the time the devm_memremap_pages_release() is run.  Rather than continue
this awkward situation, offload the responsibility of killing the
percpu_ref to devm_memremap_pages_release() directly.  This allows
devm_memremap_pages() to do the right thing relative to init failures and
shutdown.

Without this change we could fail to register the teardown of
devm_memremap_pages().  The likelihood of hitting this failure is tiny as
small memory allocations almost always succeed.  However, the impact of
the failure is large given any future reconfiguration, or disable/enable,
of an nvdimm namespace will fail forever as subsequent calls to
devm_memremap_pages() will fail to setup the pgmap_radix since there will
be stale entries for the physical address range.

An argument could be made to require that the ->kill() operation be set in
the @pgmap arg rather than passed in separately.  However, it helps code
readability, tracking the lifetime of a given instance, to be able to grep
the kill routine directly at the devm_memremap_pages() call site.

Link: http://lkml.kernel.org/r/154275558526.76910.7535251937849268605.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Fixes: e8d513483300 ("memremap: change devm_memremap_pages interface...")
Reviewed-by: "Jérôme Glisse" <jglisse@redhat.com>
Reported-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, devm_memremap_pages: kill mapping "System RAM" support

Given the fact that devm_memremap_pages() requires a percpu_ref that is
torn down by devm_memremap_pages_release() the current support for mapping
RAM is broken.

Support for remapping "System RAM" has been broken since the beginning and
there is no existing user of this this code path, so just kill the support
and make it an explicit error.

This cleanup also simplifies a follow-on patch to fix the error path when
setting a devm release action for devm_memremap_pages_release() fails.

Link: http://lkml.kernel.org/r/154275557997.76910.14689813630968180480.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: "Jérôme Glisse" <jglisse@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, devm_memremap_pages: mark devm_memremap_pages() EXPORT_SYMBOL_GPL

devm_memremap_pages() is a facility that can create struct page entries
for any arbitrary range and give drivers the ability to subvert core
aspects of page management.

Specifically the facility is tightly integrated with the kernel's memory
hotplug functionality.  It injects an altmap argument deep into the
architecture specific vmemmap implementation to allow allocating from
specific reserved pages, and it has Linux specific assumptions about page
structure reference counting relative to get_user_pages() and
get_user_pages_fast().  It was an oversight and a mistake that this was
not marked EXPORT_SYMBOL_GPL from the outset.

Again, devm_memremap_pagex() exposes and relies upon core kernel internal
assumptions and will continue to evolve along with 'struct page', memory
hotplug, and support for new memory types / topologies.  Only an in-kernel
GPL-only driver is expected to keep up with this ongoing evolution.  This
interface, and functionality derived from this interface, is not suitable
for kernel-external drivers.

Link: http://lkml.kernel.org/r/154275557457.76910.16923571232582744134.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: "Jérôme Glisse" <jglisse@redhat.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: change the order of MIGRATE_RECLAIMABLE/MIGRATE_MOVABLE in fallbacks

In the enum migratetype definition, MIGRATE_MOVABLE is before
MIGRATE_RECLAIMABLE.  Change the order of them to match the enumeration's
order.

Link: http://lkml.kernel.org/r/20181121085821.3442-1-sjhuang@iluvatar.ai
Signed-off-by: Huang Shijie <sjhuang@iluvatar.ai>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: convert userfaultfd_ctx::refcount to refcount_t

Reference counters should use refcount_t rather than atomic_t, since the
refcount_t implementation can prevent overflows, reducing the
exploitability of reference leak bugs.  userfaultfd_ctx::refcount is a
reference counter with the usual semantics, so convert it to refcount_t.

Note: I replaced the BUG() on incrementing a 0 refcount with just
refcount_inc(), since part of the semantics of refcount_t is that that
incrementing a 0 refcount is not allowed; with CONFIG_REFCOUNT_FULL,
refcount_inc() already checks for it and warns.

Link: http://lkml.kernel.org/r/20181115003916.63381-1-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap: use nr_node_ids for avail_lists in swap_info_struct

Since a2468cc9bfdf ("swap: choose swap device according to numa node"),
avail_lists field of swap_info_struct is changed to an array with
MAX_NUMNODES elements.  This made swap_info_struct size increased to 40KiB
and needs an order-4 page to hold it.

This is not optimal in that:
1 Most systems have way less than MAX_NUMNODES(1024) nodes so it
  is a waste of memory;
2 It could cause swapon failure if the swap device is swapped on
  after system has been running for a while, due to no order-4
  page is available as pointed out by Vasily Averin.

Solve the above two issues by using nr_node_ids(which is the actual
possible node number the running system has) for avail_lists instead of
MAX_NUMNODES.

nr_node_ids is unknown at compile time so can't be directly used when
declaring this array.  What I did here is to declare avail_lists as zero
element array and allocate space for it when allocating space for
swap_info_struct.  The reason why keep using array but not pointer is
plist_for_each_entry needs the field to be part of the struct, so pointer
will not work.

This patch is on top of Vasily Averin's fix commit.  I think the use of
kvzalloc for swap_info_struct is still needed in case nr_node_ids is
really big on some systems.

Link: http://lkml.kernel.org/r/20181115083847.GA11129@intel.com
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vasily Averin <vvs@virtuozzo.com>
Cc: Huang Ying <ying.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: return NODE_RECLAIM_NOSCAN in node_reclaim() when CONFIG_NUMA is n

Commit fa5e084e43eb ("vmscan: do not unconditionally treat zones that
fail zone_reclaim() as full") changed the return value of
node_reclaim().  The original return value 0 means NODE_RECLAIM_SOME
after this commit.

While the return value of node_reclaim() when CONFIG_NUMA is n is not
changed.  This will leads to call zone_watermark_ok() again.

This patch fixes the return value by adjusting to NODE_RECLAIM_NOSCAN.
Since node_reclaim() is only called in page_alloc.c, move it to
mm/internal.h.

Link: http://lkml.kernel.org/r/20181113080436.22078-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove managed_page_count_lock spinlock

Now that totalram_pages and managed_pages are atomic varibles, no need of
managed_page_count spinlock.  The lock had really a weak consistency
guarantee.  It hasn't been used for anything but the update but no reader
actually cares about all the values being updated to be in sync.

Link: http://lkml.kernel.org/r/1542090790-21750-5-git-send-email-arunks@codeaurora.org
Signed-off-by: Arun KS <arunks@codeaurora.org>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: David Hildenbrand <david@redhat.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: convert totalram_pages and totalhigh_pages variables to atomic

totalram_pages and totalhigh_pages are made static inline function.

Main motivation was that managed_page_count_lock handling was complicating
things.  It was discussed in length here,
https://lore.kernel.org/patchwork/patch/995739/#1181785 So it seemes
better to remove the lock and convert variables to atomic, with preventing
poteintial store-to-read tearing as a bonus.

[akpm@linux-foundation.org: coding style fixes]
Link: http://lkml.kernel.org/r/1542090790-21750-4-git-send-email-arunks@codeaurora.org
Signed-off-by: Arun KS <arunks@codeaurora.org>
Suggested-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: convert zone->managed_pages to atomic variable

totalram_pages, zone->managed_pages and totalhigh_pages updates are
protected by managed_page_count_lock, but readers never care about it.
Convert these variables to atomic to avoid readers potentially seeing a
store tear.

This patch converts zone->managed_pages.  Subsequent patches will convert
totalram_panges, totalhigh_pages and eventually managed_page_count_lock
will be removed.

Main motivation was that managed_page_count_lock handling was complicating
things.  It was discussed in length here,
https://lore.kernel.org/patchwork/patch/995739/#1181785 So it seemes
better to remove the lock and convert variables to atomic, with preventing
poteintial store-to-read tearing as a bonus.

Link: http://lkml.kernel.org/r/1542090790-21750-3-git-send-email-arunks@codeaurora.org
Signed-off-by: Arun KS <arunks@codeaurora.org>
Suggested-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: reference totalram_pages and managed_pages once per function

Patch series "mm: convert totalram_pages, totalhigh_pages and managed
pages to atomic", v5.

This series converts totalram_pages, totalhigh_pages and
zone->managed_pages to atomic variables.

totalram_pages, zone->managed_pages and totalhigh_pages updates are
protected by managed_page_count_lock, but readers never care about it.
Convert these variables to atomic to avoid readers potentially seeing a
store tear.

Main motivation was that managed_page_count_lock handling was complicating
things.  It was discussed in length here,
https://lore.kernel.org/patchwork/patch/995739/#1181785 It seemes better
to remove the lock and convert variables to atomic.  With the change,
preventing poteintial store-to-read tearing comes as a bonus.

This patch (of 4):

This is in preparation to a later patch which converts totalram_pages and
zone->managed_pages to atomic variables.  Please note that re-reading the
value might lead to a different value and as such it could lead to
unexpected behavior.  There are no known bugs as a result of the current
code but it is better to prevent from them in principle.

Link: http://lkml.kernel.org/r/1542090790-21750-2-git-send-email-arunks@codeaurora.org
Signed-off-by: Arun KS <arunks@codeaurora.org>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove reset of pcp->counter in pageset_init()

per_cpu_pageset is cleared by memset, it is not necessary to reset it
again.

Link: http://lkml.kernel.org/r/20181021023920.5501-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: do not clear numa_node association after hot_remove

Per-cpu numa_node provides a default node for each possible cpu.  The
association gets initialized during the boot when the architecture
specific code explores cpu->NUMA affinity.  When the whole NUMA node is
removed though we are clearing this association

try_offline_node
  check_and_unmap_cpu_on_node
    unmap_cpu_on_node
      numa_clear_node
        numa_set_node(cpu, NUMA_NO_NODE)

This means that whoever calls cpu_to_node for a cpu associated with such a
node will get NUMA_NO_NODE.  This is problematic for two reasons.  First
it is fragile because __alloc_pages_node would simply blow up on an
out-of-bound access.  We have encountered this when loading kvm module

  BUG: unable to handle kernel paging request at 00000000000021c0
  IP: __alloc_pages_nodemask+0x93/0xb70
  PGD 800000ffe853e067 PUD 7336bbc067 PMD 0
  Oops: 0000 [#1] SMP
  [...]
  CPU: 88 PID: 1223749 Comm: modprobe Tainted: G        W          4.4.156-94.64-default #1
  RIP: __alloc_pages_nodemask+0x93/0xb70
  RSP: 0018:ffff887354493b40  EFLAGS: 00010202
  RAX: 00000000000021c0 RBX: 0000000000000000 RCX: 0000000000000000
  RDX: 0000000000000000 RSI: 0000000000000002 RDI: 00000000014000c0
  RBP: 00000000014000c0 R08: ffffffffffffffff R09: 0000000000000000
  R10: ffff88fffc89e790 R11: 0000000000014000 R12: 0000000000000101
  R13: ffffffffa0772cd4 R14: ffffffffa0769ac0 R15: 0000000000000000
  FS:  00007fdf2f2f1700(0000) GS:ffff88fffc880000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000000021c0 CR3: 00000077205ee000 CR4: 0000000000360670
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
    alloc_vmcs_cpu+0x3d/0x90 [kvm_intel]
    hardware_setup+0x781/0x849 [kvm_intel]
    kvm_arch_hardware_setup+0x28/0x190 [kvm]
    kvm_init+0x7c/0x2d0 [kvm]
    vmx_init+0x1e/0x32c [kvm_intel]
    do_one_initcall+0xca/0x1f0
    do_init_module+0x5a/0x1d7
    load_module+0x1393/0x1c90
    SYSC_finit_module+0x70/0xa0
    entry_SYSCALL_64_fastpath+0x1e/0xb7
  DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x1e/0xb7

on an older kernel but the code is basically the same in the current Linus
tree as well.  alloc_vmcs_cpu could use alloc_pages_nodemask which would
recognize NUMA_NO_NODE and use alloc_pages_node which would translate it
to numa_mem_id but that is wrong as well because it would use a cpu
affinity of the local CPU which might be quite far from the original node.
It is also reasonable to expect that cpu_to_node will provide a sane
value and there might be many more callers like that.

The second problem is that __register_one_node relies on cpu_to_node to
properly associate cpus back to the node when it is onlined.  We do not
want to lose that link as there is no arch independent way to get it from
the early boot time AFAICS.

Drop the whole check_and_unmap_cpu_on_node machinery and keep the
association to fix both issues.  The NODE_DATA(nid) is not deallocated so
it will stay in place and if anybody wants to allocate from that node then
a fallback node will be used.

Thanks to Vlastimil Babka for his live system debugging skills that helped
debugging the issue.

Link: http://lkml.kernel.org/r/20181108100413.966-1-mhocko@kernel.org
Fixes: e13fe8695c57 ("cpu-hotplug,memory-hotplug: clear cpu_to_node() when offlining the node")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Debugged-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmap.c: remove verify_mm_writelocked()

We should get rid of this function.  It no longer serves its purpose.
This is a historical artifact from 2005 where do_brk was called outside of
the core mm.  We do have a proper abstraction in vm_brk_flags and that one
does the locking properly so there is no need to use this function.

Link: http://lkml.kernel.org/r/20181108174856.10811-1-tiny.windzz@gmail.com
Signed-off-by: Yangtao Li <tiny.windzz@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ksm: replace jhash2 with xxhash

Replace jhash2 with xxhash.

Perf numbers:
Intel(R) Xeon(R) CPU E5-2420 v2 @ 2.20GHz
ksm: crc32c   hash() 12081 MB/s
ksm: xxh64    hash()  8770 MB/s
ksm: xxh32    hash()  4529 MB/s
ksm: jhash2   hash()  1569 MB/s

Sioh Lee did some testing:

crc32c_intel: 1084.10ns
crc32c (no hardware acceleration): 7012.51ns
xxhash32: 2227.75ns
xxhash64: 1413.16ns
jhash2: 5128.30ns

As jhash2 always will be slower (for data size like PAGE_SIZE).  Don't use
it in ksm at all.

Use only xxhash for now, because for using crc32c, cryptoapi must be
initialized first - that requires some tricky solution to work well in all
situations.

Link: http://lkml.kernel.org/r/20181023182554.23464-3-nefelim4ag@gmail.com
Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Signed-off-by: leesioh <solee@os.korea.ac.kr>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Reviewed-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

xxHash: create arch dependent 32/64-bit xxhash()

Patch series "Currently used jhash are slow enough and replace it allow as
to make KSM", v8.

Apeed (in kernel):
        ksm: crc32c   hash() 12081 MB/s
        ksm: xxh64    hash()  8770 MB/s
        ksm: xxh32    hash()  4529 MB/s
        ksm: jhash2   hash()  1569 MB/s

Sioh Lee's testing (copy from other mail):

Test platform: openstack cloud platform (NEWTON version)
Experiment node: openstack based cloud compute node (CPU: xeon E5-2620 v3, memory 64gb)
VM: (2 VCPU, RAM 4GB, DISK 20GB) * 4
Linux kernel: 4.14 (latest version)
KSM setup - sleep_millisecs: 200ms, pages_to_scan: 200

Experiment process:
Firstly, we turn off KSM and launch 4 VMs.  Then we turn on the KSM and
measure the checksum computation time until full_scans become two.

The experimental results (the experimental value is the average of the measured values)
crc32c_intel: 1084.10ns
crc32c (no hardware acceleration): 7012.51ns
xxhash32: 2227.75ns
xxhash64: 1413.16ns
jhash2: 5128.30ns

In summary, the result shows that crc32c_intel has advantages over all of
the hash function used in the experiment.  (decreased by 84.54% compared
to crc32c, 78.86% compared to jhash2, 51.33% xxhash32, 23.28% compared to
xxhash64) the results are similar to those of Timofey.

But, use only xxhash for now, because for using crc32c, cryptoapi must be
initialized first - that require some tricky solution to work good in all
situations.

So:

- First patch implement compile time pickup of fastest implementation of
  xxhash for target platform.

- The second patch replaces jhash2 with xxhash

This patch (of 2):

xxh32() - fast on both 32/64-bit platforms
xxh64() - fast only on 64-bit platform

Create xxhash() which will pick up the fastest version at compile time.

Link: http://lkml.kernel.org/r/20181023182554.23464-2-nefelim4ag@gmail.com
Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Reviewed-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: leesioh <solee@os.korea.ac.kr>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: only report isolation failures when offlining memory

Heiko has complained that his log is swamped by warnings from
has_unmovable_pages

[   20.536664] page dumped because: has_unmovable_pages
[   20.536792] page:000003d081ff4080 count:1 mapcount:0 mapping:000000008ff88600 index:0x0 compound_mapcount: 0
[   20.536794] flags: 0x3fffe0000010200(slab|head)
[   20.536795] raw: 03fffe0000010200 0000000000000100 0000000000000200 000000008ff88600
[   20.536796] raw: 0000000000000000 0020004100000000 ffffffff00000001 0000000000000000
[   20.536797] page dumped because: has_unmovable_pages
[   20.536814] page:000003d0823b0000 count:1 mapcount:0 mapping:0000000000000000 index:0x0
[   20.536815] flags: 0x7fffe0000000000()
[   20.536817] raw: 07fffe0000000000 0000000000000100 0000000000000200 0000000000000000
[   20.536818] raw: 0000000000000000 0000000000000000 ffffffff00000001 0000000000000000

which are not triggered by the memory hotplug but rather CMA allocator.
The original idea behind dumping the page state for all call paths was
that these messages will be helpful debugging failures.  From the above it
seems that this is not the case for the CMA path because we are lacking
much more context.  E.g the second reported page might be a CMA allocated
page.  It is still interesting to see a slab page in the CMA area but it
is hard to tell whether this is bug from the above output alone.

Address this issue by dumping the page state only on request.  Both
start_isolate_page_range and has_unmovable_pages already have an argument
to ignore hwpoison pages so make this argument more generic and turn it
into flags and allow callers to combine non-default modes into a mask.
While we are at it, has_unmovable_pages call from
is_pageblock_removable_nolock (sysfs removable file) is questionable to
report the failure so drop it from there as well.

Link: http://lkml.kernel.org/r/20181218092802.31429-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: be more verbose for memory offline failures

There is only very limited information printed when the memory offlining
fails:

[ 1984.506184] rac1 kernel: memory offlining [mem 0x82600000000-0x8267fffffff] failed due to signal backoff

This tells us that the failure is triggered by the userspace intervention
but it doesn't tell us much more about the underlying reason.  It might be
that the page migration failes repeatedly and the userspace timeout
expires and send a signal or it might be some of the earlier steps
(isolation, memory notifier) takes too long.

If the migration failes then it would be really helpful to see which page
that and its state.  The same applies to the isolation phase.  If we fail
to isolate a page from the allocator then knowing the state of the page
would be helpful as well.

Dump the page state that fails to get isolated or migrated.  This will
tell us more about the failure and what to focus on during debugging.

[akpm@linux-foundation.org: add missing printk arg]
[mhocko@suse.com: tweak dump_page() `reason' text]
Link: http://lkml.kernel.org/r/20181116083020.20260-6-mhocko@kernel.org
Link: http://lkml.kernel.org/r/20181107101830.17405-6-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: print reason for the offlining failure

The memory offlining failure reporting is inconsistent and insufficient.
Some error paths simply do not report the failure to the log at all.  When
we do report there are no details about the reason of the failure and
there are several of them which makes memory offlining failures hard to
debug.

Make sure that the
memory offlining [mem %#010llx-%#010llx] failed
message is printed for all failures and also provide a short textual
reason for the failure e.g.

[ 1984.506184] rac1 kernel: memory offlining [mem 0x82600000000-0x8267fffffff] failed due to signal backoff

this tells us that the offlining has failed because of a signal pending
aka user intervention.

[akpm@linux-foundation.org: tweak messages a bit]
Link: http://lkml.kernel.org/r/20181107101830.17405-5-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memory_hotplug: drop pointless block alignment checks from __offline_pages

This function is never called from a context which would provide
misaligned pfn range so drop the pointless check.

Link: http://lkml.kernel.org/r/20181107101830.17405-4-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: lower the printk loglevel for __dump_page messages

__dump_page messages use KERN_EMERG resp.  KERN_ALERT loglevel (this is
the case since 2004).  Most callers of this function are really detecting
a critical page state and BUG right after.  On the other hand the function
is called also from contexts which just want to inform about the page
state and those would rather not disrupt logs that much (e.g.  some
systems route these messages to the normal console).

Reduce the loglevel to KERN_WARNING to make dump_page easier to reuse for
other contexts while those messages will still make it to the kernel log
in most setups.  Even if the loglevel setup filters warnings away those
paths that are really critical already print the more targeted error or
panic and that should make it to the kernel log.

[mhocko@kernel.org: fix __dump_page()]
Link: http://lkml.kernel.org/r/20181212142540.GA7378@dhcp22.suse.cz
[akpm@linux-foundation.org: s/KERN_WARN/KERN_WARNING/, per Michal]
Link: http://lkml.kernel.org/r/20181107101830.17405-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: print more information about mapping in __dump_page

I have been promissing to improve memory offlining failures debugging for
quite some time.  As things stand now we get only very limited information
in the kernel log when the offlining fails.  It is usually only

[ 1984.506184] rac1 kernel: memory offlining [mem 0x82600000000-0x8267fffffff] failed

with no further details.  We do not know what exactly fails and for what
reason.  Whenever I was forced to debug such a failure I've always had to
do a debugging patch to tell me more.  We can enable some tracepoints but
it would be much better to get a better picture without using them.

This patch series does 2 things.  The first one is to make dump_page more
usable by printing more information about the mapping patch 1.  Then it
reduces the log level from emerg to warning so that this function is
usable from less critical context patch 2.  Then I have added more
detailed information about the offlining failure patch 4 and finally add
dump_page to isolation and offlining migration paths.  Patch 3 is a
trivial cleanup.

This patch (of 6):

__dump_page prints the mapping pointer but that is quite unhelpful for
many reports because the pointer itself only helps to distinguish anon/ksm
mappings from other ones (because of lowest bits set).  Sometimes it would
be much more helpful to know what kind of mapping that is actually and if
we know this is a file mapping then also try to resolve the dentry name.

[dan.carpenter@oracle.com: fix a width vs precision bug in printk]
Link: http://lkml.kernel.org/r/20181123072135.gqvblm2vdujbvfjs@kili.mountain
[mhocko@kernel.org: use %dp to print dentry]
Link: http://lkml.kernel.org/r/20181125080834.GB12455@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/20181107101830.17405-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/readahead.c: simplify get_next_ra_size()

It's a trivial simplification for get_next_ra_size() and clear enough for
humans to understand.

It also fixes potential overflow if ra->size(< ra_pages) is too large.

Link: http://lkml.kernel.org/r/1540707206-19649-1-git-send-email-hsiangkao@aol.com
Signed-off-by: Gao Xiang <hsiangkao@aol.com>
Reviewed-by: Fengguang Wu <fengguang.wu@intel.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

writeback: don't decrement wb->refcnt if !wb->bdi

This happened while running in qemu-system-aarch64, the AMBA PL011 UART
driver when enabling CONFIG_DEBUG_TEST_DRIVER_REMOVE.
arch_initcall(pl011_init) came before subsys_initcall(default_bdi_init),
devtmpfs' handle_remove() crashes because the reference count is a NULL
pointer only because wb->bdi hasn't been initialized yet.

Rework so that wb_put have an extra check if wb->bdi before decrement
wb->refcnt and also add a WARN_ON_ONCE to get a warning if it happens
again in other drivers.

Link: http://lkml.kernel.org/r/20181030113545.30999-2-anders.roxell@linaro.org
Fixes: 52ebea749aae ("writeback: make backing_dev_info host cgroup-specific bdi_writebacks")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Co-developed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmu_notifier.c: remove mmu_notifier_synchronize()

Contrary to its name, mmu_notifier_synchronize() does not synchronize the
notifier's SRCU instance, but rather waits for RCU callbacks to finish.
i.e.  it invokes rcu_barrier().  The RCU documentation is quite clear on
this matter, explicitly calling out that rcu_barrier() does not imply
synchronize_rcu().

As there are no callers of mmu_notifier_synchronize() and it's unclear
whether any user of mmu_notifier_call_srcu() will ever want to barrier on
their callbacks, simply remove the function.

Link: http://lkml.kernel.org/r/20181106134705.14197-1-sean.j.christopherson@intel.com
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hotplug: optimize clear_hwpoisoned_pages()

In hot remove, we try to clear poisoned pages, but a small optimization to
check if num_poisoned_pages is 0 helps remove the iteration through
nr_pages.

[akpm@linux-foundation.org: tweak comment text]
Link: http://lkml.kernel.org/r/20181102120001.4526-1-bsingharora@gmail.com
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_owner: clamp read count to PAGE_SIZE

The (root-only) page owner read might allocate a large size of memory with
a large read count.  Allocation fails can easily occur when doing high
order allocations.

Clamp buffer size to PAGE_SIZE to avoid arbitrary size allocation
and avoid allocation fails due to high order allocation.

[akpm@linux-foundation.org: use min_t()]
Link: http://lkml.kernel.org/r/1541091607-27402-1-git-send-email-miles.chen@mediatek.com
Signed-off-by: Miles Chen <miles.chen@mediatek.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Joe Perches <joe@perches.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/slab.h: fix sparse warning in kmalloc_type()

Multiple people have reported the following sparse warning:

./include/linux/slab.h:332:43: warning: dubious: x & !y

The minimal fix would be to change the logical & to boolean &&, which
emits the same code, but Andrew has suggested that the branch-avoiding
tricks are maybe not worthwile.  David Laight provided a nice comparison
of disassembly of multiple variants, which shows that the current version
produces a 4 deep dependency chain, and fixing the sparse warning by
changing logical and to multiplication emits an IMUL, making it even more
expensive.

The code as rewritten by this patch yielded the best disassembly, with a
single predictable branch for the most common case, and a ternary operator
for the rest, which gcc seems to compile without a branch or cmov by
itself.

The result should be more readable, without a sparse warning and probably
also faster for the common case.

Link: http://lkml.kernel.org/r/80340595-d7c5-97b9-4f6c-23fa893a91e9@suse.cz
Fixes: 1291523f2c1d ("mm, slab/slub: introduce kmalloc-reclaimable caches")
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Bart Van Assche <bvanassche@acm.org>
Reported-by: Darryl T. Agostinelli <dagostinelli@gmail.com>
Reported-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Suggested-by: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/slub.c: record final state of slub action in deactivate_slab()

If __cmpxchg_double_slab() fails and (l != m), current code records
transition states of slub action.

Update the action after __cmpxchg_double_slab() success to record the
final state.

[akpm@linux-foundation.org: more whitespace cleanup]
Link: http://lkml.kernel.org/r/20181107013119.3816-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/slub.c: page is always non-NULL in node_match()

node_match() is a static function and is only invoked in slub.c.

In all three places, `page' is ensured to be valid.

Link: http://lkml.kernel.org/r/20181106150245.1668-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/slub.c: remove validation on cpu_slab in __flush_cpu_slab()

cpu_slab is a per cpu variable which is allocated in all or none.  If a
cpu_slab failed to be allocated, the slub is not usable.

We could use cpu_slab without validation in __flush_cpu_slab().

Link: http://lkml.kernel.org/r/20181103141218.22844-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, slab: remove unnecessary unlikely()

WARN_ON() already contains an unlikely(), so it's not necessary to use
unlikely.

Also change WARN_ON() back to WARN_ON_ONCE() to avoid potentially
spamming dmesg with user-triggerable large allocations.

[akpm@linux-foundation.org: s/WARN_ON/WARN_ON_ONCE/, per Vlastimil]
Link: http://lkml.kernel.org/r/20181104125028.3572-1-tiny.windzz@gmail.com
Signed-off-by: Yangtao Li <tiny.windzz@gmail.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: don't clear bh uptodate for block read

For sync io read in ocfs2_read_blocks_sync(), first clear bh uptodate flag
and submit the io, second wait io done, last check whether bh uptodate, if
not return io error.

If two sync io for the same bh were issued, it could be the first io done
and set uptodate flag, but just before check that flag, the second io came
in and cleared uptodate, then ocfs2_read_blocks_sync() for the first io
will return IO error.

Indeed it's not necessary to clear uptodate flag, as the io end handler
end_buffer_read_sync() will set or clear it based on io succeed or failed.

The following message was found from a nfs server but the underlying
storage returned no error.

[4106438.567376] (nfsd,7146,3):ocfs2_get_suballoc_slot_bit:2780 ERROR: read block 1238823695 failed -5
[4106438.567569] (nfsd,7146,3):ocfs2_get_suballoc_slot_bit:2812 ERROR: status = -5
[4106438.567611] (nfsd,7146,3):ocfs2_test_inode_bit:2894 ERROR: get alloc slot and bit failed -5
[4106438.567643] (nfsd,7146,3):ocfs2_test_inode_bit:2932 ERROR: status = -5
[4106438.567675] (nfsd,7146,3):ocfs2_get_dentry:94 ERROR: test inode bit failed -5

Same issue in non sync read ocfs2_read_blocks(), fixed it as well.

Link: http://lkml.kernel.org/r/20181121020023.3034-4-junxiao.bi@oracle.com
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Jun Piao <piaojun@huawei.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: clear journal dirty flag after shutdown journal

Dirty flag of the journal should be cleared at the last stage of umount,
if do it before jbd2_journal_destroy(), then some metadata in uncommitted
transaction could be lost due to io error, but as dirty flag of journal
was already cleared, we can't find that until run a full fsck.  This may
cause system panic or other corruption.

Link: http://lkml.kernel.org/r/20181121020023.3034-3-junxiao.bi@oracle.com
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Jun Piao <piaojun@huawei.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@versity.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: fix panic due to unrecovered local alloc

mount.ocfs2 ignore the inconsistent error that journal is clean but
local alloc is unrecovered.  After mount, local alloc not empty, then
reserver cluster didn't alloc a new local alloc window, reserveration
map is empty(ocfs2_reservation_map.m_bitmap_len = 0), that triggered the
following panic.

This issue was reported at

  https://oss.oracle.com/pipermail/ocfs2-devel/2015-May/010854.html

and was advised to fixed during mount.  But this is a very unusual
inconsistent state, usually journal dirty flag should be cleared at the
last stage of umount until every other things go right.  We may need do
further debug to check that.  Any way to avoid possible futher
corruption, mount should be abort and fsck should be run.

  (mount.ocfs2,1765,1):ocfs2_load_local_alloc:353 ERROR: Local alloc hasn't been recovered!
  found = 6518, set = 6518, taken = 8192, off = 15912372
  ocfs2: Mounting device (202,64) on (node 0, slot 3) with ordered data mode.
  o2dlm: Joining domain 89CEAC63CC4F4D03AC185B44E0EE0F3F ( 0 1 2 3 4 5 6 8 ) 8 nodes
  ocfs2: Mounting device (202,80) on (node 0, slot 3) with ordered data mode.
  o2hb: Region 89CEAC63CC4F4D03AC185B44E0EE0F3F (xvdf) is now a quorum device
  o2net: Accepted connection from node yvwsoa17p (num 7) at 172.22.77.88:7777
  o2dlm: Node 7 joins domain 64FE421C8C984E6D96ED12C55FEE2435 ( 0 1 2 3 4 5 6 7 8 ) 9 nodes
  o2dlm: Node 7 joins domain 89CEAC63CC4F4D03AC185B44E0EE0F3F ( 0 1 2 3 4 5 6 7 8 ) 9 nodes
  ------------[ cut here ]------------
  kernel BUG at fs/ocfs2/reservations.c:507!
  invalid opcode: 0000 [#1] SMP
  Modules linked in: ocfs2 rpcsec_gss_krb5 auth_rpcgss nfsv4 nfs fscache lockd grace ocfs2_dlmfs ocfs2_stack_o2cb ocfs2_dlm ocfs2_nodemanager ocfs2_stackglue configfs sunrpc ipt_REJECT nf_reject_ipv4 nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables ib_ipoib rdma_ucm ib_ucm ib_uverbs ib_umad rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 ovmapi ppdev parport_pc parport xen_netfront fb_sys_fops sysimgblt sysfillrect syscopyarea acpi_cpufreq pcspkr i2c_piix4 i2c_core sg ext4 jbd2 mbcache2 sr_mod cdrom xen_blkfront pata_acpi ata_generic ata_piix floppy dm_mirror dm_region_hash dm_log dm_mod
  CPU: 0 PID: 4349 Comm: startWebLogic.s Not tainted 4.1.12-124.19.2.el6uek.x86_64 #2
  Hardware name: Xen HVM domU, BIOS 4.4.4OVM 09/06/2018
  task: ffff8803fb04e200 ti: ffff8800ea4d8000 task.ti: ffff8800ea4d8000
  RIP: 0010:[<ffffffffa05e96a8>]  [<ffffffffa05e96a8>] __ocfs2_resv_find_window+0x498/0x760 [ocfs2]
  Call Trace:
    ocfs2_resmap_resv_bits+0x10d/0x400 [ocfs2]
    ocfs2_claim_local_alloc_bits+0xd0/0x640 [ocfs2]
    __ocfs2_claim_clusters+0x178/0x360 [ocfs2]
    ocfs2_claim_clusters+0x1f/0x30 [ocfs2]
    ocfs2_convert_inline_data_to_extents+0x634/0xa60 [ocfs2]
    ocfs2_write_begin_nolock+0x1c6/0x1da0 [ocfs2]
    ocfs2_write_begin+0x13e/0x230 [ocfs2]
    generic_perform_write+0xbf/0x1c0
    __generic_file_write_iter+0x19c/0x1d0
    ocfs2_file_write_iter+0x589/0x1360 [ocfs2]
    __vfs_write+0xb8/0x110
    vfs_write+0xa9/0x1b0
    SyS_write+0x46/0xb0
    system_call_fastpath+0x18/0xd7
  Code: ff ff 8b 75 b8 39 75 b0 8b 45 c8 89 45 98 0f 84 e5 fe ff ff 45 8b 74 24 18 41 8b 54 24 1c e9 56 fc ff ff 85 c0 0f 85 48 ff ff ff <0f> 0b 48 8b 05 cf c3 de ff 48 ba 00 00 00 00 00 00 00 10 48 85
  RIP   __ocfs2_resv_find_window+0x498/0x760 [ocfs2]
   RSP <ffff8800ea4db668>
  ---[ end trace 566f07529f2edf3c ]---
  Kernel panic - not syncing: Fatal exception
  Kernel Offset: disabled

Link: http://lkml.kernel.org/r/20181121020023.3034-2-junxiao.bi@oracle.com
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Acked-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Jun Piao <piaojun@huawei.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: improve ocfs2 Makefile

Included file path was hard-wired in the ocfs2 makefile, which might
causes some confusion when compiling ocfs2 as an external module.

Say if we compile ocfs2 module as following.
cp -r /kernel/tree/fs/ocfs2 /other/dir/ocfs2
cd /other/dir/ocfs2
make -C /path/to/kernel_source M=`pwd` modules

Acutally, the compiler wil try to find included file in
/kernel/tree/fs/ocfs2, rather than the directory /other/dir/ocfs2.

To fix this little bug, we introduce the var $(src) provided by kbuild.
$(src) means the absolute path of the running kbuild file.

Link: http://lkml.kernel.org/r/20181108085546.15149-1-lchen@suse.com
Signed-off-by: Larry Chen <lchen@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: remove set but not used variable 'lastzero'

lastzero is not used after setting its value.  It is safe to remove the
unused variable.

Link: http://lkml.kernel.org/r/1540296942-24533-1-git-send-email-zhongjiang@huawei.com
Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: dlmfs: remove set but not used variable 'status'

status is not used after setting its value.  It is safe to remove the
unused variable.

Link: http://lkml.kernel.org/r/1540300179-26697-1-git-send-email-zhongjiang@huawei.com
Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Reviewed-by: Jun Piao <piaojun@huawei.com>
Acked-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ocfs2: optimize the reading of heartbeat data

Reading heartbeat data from lowest node rather than from zero, in cases
where the node is not defined from zero, can reduce the number of sectors
read.

Here is a simple test data obtained with 'iostat -dmx dm-5 2', with
two nodes in the cluster, node number 10, 20, respectively.

Before optimization:
Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
dm-5              0.00     0.00    0.50    0.50     0.01     0.00    11.00     0.00    1.00    1.00    1.00   1.50   0.15

After the optimization:
Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
dm-5              0.00     0.00    0.50    0.50     0.00     0.00     6.00     0.00    0.50    1.00    0.00   0.50   0.05

Link: http://lkml.kernel.org/r/99fe4988-69ac-3615-a218-3042fe6fbe72@huawei.com
Signed-off-by: Jia Guo <guojia12@huawei.com>
Reviewed-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Acked-by: Joseph Qi <jiangqi903@gmail.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

debugobjects: call debug_objects_mem_init eariler

The current value of the early boot static pool size, 1024 is not big
enough for systems with large number of CPUs with timer or/and workqueue
objects selected.  As the results, systems have 60+ CPUs with both timer
and workqueue objects enabled could trigger "ODEBUG: Out of memory.
ODEBUG disabled".

Some debug objects are allocated during the early boot.  Enabling some
options like timers or workqueue objects may increase the size required
significantly with large number of CPUs.  For example,

CONFIG_DEBUG_OBJECTS_TIMERS:
No. CPUs x 2 (worker pool) objects:
start_kernel
  workqueue_init_early
    init_worker_pool
      init_timer_key
        debug_object_init

plus No. CPUs objects (CONFIG_HIGH_RES_TIMERS):
sched_init
  hrtick_rq_init
    hrtimer_init

CONFIG_DEBUG_OBJECTS_WORK:
No. CPUs objects:
vmalloc_init
  __init_work

plus No. CPUs x 6 (workqueue) objects:
workqueue_init_early
  alloc_workqueue
    __alloc_workqueue_key
      alloc_and_link_pwqs
        init_pwq

Also, plus No. CPUs objects:
perf_event_init
  __init_srcu_struct
    init_srcu_struct_fields
      init_srcu_struct_nodes
        __init_work

However, none of the things are actually used or required before
debug_objects_mem_init() is invoked, so just move the call right before
vmalloc_init().

According to tglx, "the reason why the call is at this place in
start_kernel() is historical.  It's because back in the days when
debugobjects were added the memory allocator was enabled way later than
today."

Link: http://lkml.kernel.org/r/20181126102407.1836-1-cai@gmx.us
Signed-off-by: Qian Cai <cai@gmx.us>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: lib: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/871s6wcswb.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Rich Felker <dalias@libc.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: kernel: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/8736rccswn.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: cpu: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/874lbscswy.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: shmobile: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/875zw8csxa.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: sh5: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/877egocsxl.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: sh4a: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

Link: http://lkml.kernel.org/r/878t14csxy.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: sh4: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text, excepting ${LINUX}/arch/sh/kernel/cpu/sh4/softfloat.c which is not
GPL license

Link: http://lkml.kernel.org/r/87a7lkcsya.wl-kuninori.morimoto.gx@renesas.com
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>