shmem: split huge pages beyond i_size under memory pressure

Even if user asked to allocate huge pages always (huge=always), we
should be able to free up some memory by splitting pages which are
partly byound i_size if memory presure comes or once we hit limit on
filesystem size (-o size=).

In order to do this we maintain per-superblock list of inodes, which
potentially have huge pages on the border of file size.

Per-fs shrinker can reclaim memory by splitting such pages.

If we hit -ENOSPC during shmem_getpage_gfp(), we try to split a page to
free up space on the filesystem and retry allocation if it succeed.

Link: http://lkml.kernel.org/r/1466021202-61880-37-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: introduce CONFIG_TRANSPARENT_HUGE_PAGECACHE

For file mappings, we don't deposit page tables on THP allocation
because it's not strictly required to implement split_huge_pmd(): we can
just clear pmd and let following page faults to reconstruct the page
table.

But Power makes use of deposited page table to address MMU quirk.

Let's hide THP page cache, including huge tmpfs, under separate config
option, so it can be forbidden on Power.

We can revert the patch later once solution for Power found.

Link: http://lkml.kernel.org/r/1466021202-61880-36-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: add support of collapse for tmpfs/shmem pages

This patch extends khugepaged to support collapse of tmpfs/shmem pages.
We share fair amount of infrastructure with anon-THP collapse.

Few design points:

  - First we are looking for VMA which can be suitable for mapping huge
    page;

  - If the VMA maps shmem file, the rest scan/collapse operations
    operates on page cache, not on page tables as in anon VMA case.

  - khugepaged_scan_shmem() finds a range which is suitable for huge
    page. The scan is lockless and shouldn't disturb system too much.

  - once the candidate for collapse is found, collapse_shmem() attempts
    to create a huge page:

      + scan over radix tree, making the range point to new huge page;

      + new huge page is not-uptodate, locked and freezed (refcount
        is 0), so nobody can touch them until we say so.

      + we swap in pages during the scan. khugepaged_scan_shmem()
        filters out ranges with more than khugepaged_max_ptes_swap
swapped out pages. It's HPAGE_PMD_NR/8 by default.

      + old pages are isolated, unmapped and put to local list in case
        to be restored back if collapse failed.

  - if collapse succeed, we retract pte page tables from VMAs where huge
    pages mapping is possible. The huge page will be mapped as PMD on
    next minor fault into the range.

Link: http://lkml.kernel.org/r/1466021202-61880-35-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: make shmem_inode_info::lock irq-safe

We are going to need to call shmem_charge() under tree_lock to get
accoutning right on collapse of small tmpfs pages into a huge one.

The problem is that tree_lock is irq-safe and lockdep is not happy, that
we take irq-unsafe lock under irq-safe[1].

Let's convert the lock to irq-safe.

[1] https://gist.github.com/kiryl/80c0149e03ed35dfaf26628b8e03cdbc

Link: http://lkml.kernel.org/r/1466021202-61880-34-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: move up_read(mmap_sem) out of khugepaged_alloc_page()

Both variants of khugepaged_alloc_page() do up_read(&mm->mmap_sem)
first: no point keep it inside the function.

Link: http://lkml.kernel.org/r/1466021202-61880-33-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: extract khugepaged from mm/huge_memory.c

khugepaged implementation grew to the point when it deserve separate
file in source.

Let's move it to mm/khugepaged.c.

Link: http://lkml.kernel.org/r/1466021202-61880-32-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem, thp: respect MADV_{NO,}HUGEPAGE for file mappings

Let's wire up existing madvise() hugepage hints for file mappings.

MADV_HUGEPAGE advise shmem to allocate huge page on page fault in the
VMA.  It only has effect if the filesystem is mounted with huge=advise
or huge=within_size.

MADV_NOHUGEPAGE prevents hugepage from being allocated on page fault in
the VMA.  It doesn't prevent a huge page from being allocated by other
means, i.e.  page fault into different mapping or write(2) into file.

Link: http://lkml.kernel.org/r/1466021202-61880-31-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: add huge pages support

Here's basic implementation of huge pages support for shmem/tmpfs.

It's all pretty streight-forward:

  - shmem_getpage() allcoates huge page if it can and try to inserd into
    radix tree with shmem_add_to_page_cache();

  - shmem_add_to_page_cache() puts the page onto radix-tree if there's
    space for it;

  - shmem_undo_range() removes huge pages, if it fully within range.
    Partial truncate of huge pages zero out this part of THP.

    This have visible effect on fallocate(FALLOC_FL_PUNCH_HOLE)
    behaviour. As we don't really create hole in this case,
    lseek(SEEK_HOLE) may have inconsistent results depending what
    pages happened to be allocated.

  - no need to change shmem_fault: core-mm will map an compound page as
    huge if VMA is suitable;

Link: http://lkml.kernel.org/r/1466021202-61880-30-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: get_unmapped_area align huge page

Provide a shmem_get_unmapped_area method in file_operations, called at
mmap time to decide the mapping address.  It could be conditional on
CONFIG_TRANSPARENT_HUGEPAGE, but save #ifdefs in other places by making
it unconditional.

shmem_get_unmapped_area() first calls the usual mm->get_unmapped_area
(which we treat as a black box, highly dependent on architecture and
config and executable layout).  Lots of conditions, and in most cases it
just goes with the address that chose; but when our huge stars are
rightly aligned, yet that did not provide a suitable address, go back to
ask for a larger arena, within which to align the mapping suitably.

There have to be some direct calls to shmem_get_unmapped_area(), not via
the file_operations: because of the way shmem_zero_setup() is called to
create a shmem object late in the mmap sequence, when MAP_SHARED is
requested with MAP_ANONYMOUS or /dev/zero.  Though this only matters
when /proc/sys/vm/shmem_huge has been set.

Link: http://lkml.kernel.org/r/1466021202-61880-29-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: prepare huge= mount option and sysfs knob

This patch adds new mount option "huge=".  It can have following values:

  - "always":
Attempt to allocate huge pages every time we need a new page;

  - "never":
Do not allocate huge pages;

  - "within_size":
Only allocate huge page if it will be fully within i_size.
Also respect fadvise()/madvise() hints;

  - "advise:
Only allocate huge pages if requested with fadvise()/madvise();

Default is "never" for now.

"mount -o remount,huge= /mountpoint" works fine after mount: remounting
huge=never will not attempt to break up huge pages at all, just stop
more from being allocated.

No new config option: put this under CONFIG_TRANSPARENT_HUGEPAGE, which
is the appropriate option to protect those who don't want the new bloat,
and with which we shall share some pmd code.

Prohibit the option when !CONFIG_TRANSPARENT_HUGEPAGE, just as mpol is
invalid without CONFIG_NUMA (was hidden in mpol_parse_str(): make it
explicit).

Allow enabling THP only if the machine has_transparent_hugepage().

But what about Shmem with no user-visible mount? SysV SHM, memfds,
shared anonymous mmaps (of /dev/zero or MAP_ANONYMOUS), GPU drivers' DRM
objects, Ashmem.  Though unlikely to suit all usages, provide sysfs knob
/sys/kernel/mm/transparent_hugepage/shmem_enabled to experiment with
huge on those.

And allow shmem_enabled two further values:

  - "deny":
For use in emergencies, to force the huge option off from
all mounts;
  - "force":
Force the huge option on for all - very useful for testing;

Based on patch by Hugh Dickins.

Link: http://lkml.kernel.org/r/1466021202-61880-28-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, rmap: account shmem thp pages

Let's add ShmemHugePages and ShmemPmdMapped fields into meminfo and
smaps.  It indicates how many times we allocate and map shmem THP.

NR_ANON_TRANSPARENT_HUGEPAGES is renamed to NR_ANON_THPS.

Link: http://lkml.kernel.org/r/1466021202-61880-27-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

truncate: handle file thp

For shmem/tmpfs we only need to tweak truncate_inode_page() and
invalidate_mapping_pages().

truncate_inode_pages_range() and invalidate_inode_pages2_range() are
adjusted to use page_to_pgoff().

Link: http://lkml.kernel.org/r/1466021202-61880-26-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

filemap: prepare find and delete operations for huge pages

For now, we would have HPAGE_PMD_NR entries in radix tree for every huge
page.  That's suboptimal and it will be changed to use Matthew's
multi-order entries later.

'add' operation is not changed, because we don't need it to implement
hugetmpfs: shmem uses its own implementation.

Link: http://lkml.kernel.org/r/1466021202-61880-25-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

radix-tree: implement radix_tree_maybe_preload_order()

The new helper is similar to radix_tree_maybe_preload(), but tries to
preload number of nodes required to insert (1 << order) continuous
naturally-aligned elements.

This is required to push huge pages into pagecache.

Link: http://lkml.kernel.org/r/1466021202-61880-24-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page-flags: relax policy for PG_mappedtodisk and PG_reclaim

These flags are in use for file THP.

Link: http://lkml.kernel.org/r/1466021202-61880-23-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: split file huge pages before paging them out

This is preparation of vmscan for file huge pages.  We cannot write out
huge pages, so we need to split them on the way out.

Link: http://lkml.kernel.org/r/1466021202-61880-22-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp, mlock: do not mlock PTE-mapped file huge pages

As with anon THP, we only mlock file huge pages if we can prove that the
page is not mapped with PTE.  This way we can avoid mlock leak into
non-mlocked vma on split.

We rely on PageDoubleMap() under lock_page() to check if the the page
may be PTE mapped.  PG_double_map is set by page_add_file_rmap() when
the page mapped with PTEs.

Link: http://lkml.kernel.org/r/1466021202-61880-21-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: file pages support for split_huge_page()

Basic scheme is the same as for anon THP.

Main differences:

  - File pages are on radix-tree, so we have head->_count offset by
    HPAGE_PMD_NR. The count got distributed to small pages during split.

  - mapping->tree_lock prevents non-lockless access to pages under split
    over radix-tree;

  - Lockless access is prevented by setting the head->_count to 0 during
    split;

  - After split, some pages can be beyond i_size. We drop them from
    radix-tree.

  - We don't setup migration entries. Just unmap pages. It helps
    handling cases when i_size is in the middle of the page: no need
    handle unmap pages beyond i_size manually.

Link: http://lkml.kernel.org/r/1466021202-61880-20-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: run vma_adjust_trans_huge() outside i_mmap_rwsem

vma_addjust_trans_huge() splits pmd if it's crossing VMA boundary.
During split we munlock the huge page which requires rmap walk.  rmap
wants to take the lock on its own.

Let's move vma_adjust_trans_huge() outside i_mmap_rwsem to fix this.

Link: http://lkml.kernel.org/r/1466021202-61880-19-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: prepare change_huge_pmd() for file thp

change_huge_pmd() has assert which is not relvant for file page.  For
shared mapping it's perfectly fine to have page table entry writable,
without explicit mkwrite.

Link: http://lkml.kernel.org/r/1466021202-61880-18-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: skip file huge pmd on copy_huge_pmd()

copy_page_range() has a check for "Don't copy ptes where a page fault
will fill them correctly." It works on VMA level.  We still copy all
page table entries from private mappings, even if they map page cache.

We can simplify copy_huge_pmd() a bit by skipping file PMDs.

We don't map file private pages with PMDs, so they only can map page
cache.  It's safe to skip them as they can be re-faulted later.

Link: http://lkml.kernel.org/r/1466021202-61880-17-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: handle file COW faults

File COW for THP is handled on pte level: just split the pmd.

It's not clear how benefitial would be allocation of huge pages on COW
faults.  And it would require some code to make them work.

I think at some point we can consider teaching khugepaged to collapse
pages in COW mappings, but allocating huge on fault is probably
overkill.

Link: http://lkml.kernel.org/r/1466021202-61880-16-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: handle file pages in split_huge_pmd()

Splitting THP PMD is simple: just unmap it as in DAX case.  This way we
can avoid memory overhead on page table allocation to deposit.

It's probably a good idea to try to allocation page table with
GFP_ATOMIC in __split_huge_pmd_locked() to avoid refaulting the area,
but clearing pmd should be good enough for now.

Unlike DAX, we also remove the page from rmap and drop reference.
pmd_young() is transfered to PageReferenced().

Link: http://lkml.kernel.org/r/1466021202-61880-15-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp: support file pages in zap_huge_pmd()

split_huge_pmd() for file mappings (and DAX too) is implemented by just
clearing pmd entry as we can re-fill this area from page cache on pte
level later.

This means we don't need deposit page tables when file THP is mapped.
Therefore we shouldn't try to withdraw a page table on zap_huge_pmd()
file THP PMD.

Link: http://lkml.kernel.org/r/1466021202-61880-14-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp, vmstats: add counters for huge file pages

THP_FILE_ALLOC: how many times huge page was allocated and put page
cache.

THP_FILE_MAPPED: how many times file huge page was mapped.

Link: http://lkml.kernel.org/r/1466021202-61880-13-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce do_set_pmd()

With postponed page table allocation we have chance to setup huge pages.
do_set_pte() calls do_set_pmd() if following criteria met:

 - page is compound;
 - pmd entry in pmd_none();
 - vma has suitable size and alignment;

Link: http://lkml.kernel.org/r/1466021202-61880-12-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rmap: support file thp

Naive approach: on mapping/unmapping the page as compound we update
->_mapcount on each 4k page.  That's not efficient, but it's not obvious
how we can optimize this.  We can look into optimization later.

PG_double_map optimization doesn't work for file pages since lifecycle
of file pages is different comparing to anon pages: file page can be
mapped again at any time.

Link: http://lkml.kernel.org/r/1466021202-61880-11-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: postpone page table allocation until we have page to map

The idea (and most of code) is borrowed again: from Hugh's patchset on
huge tmpfs[1].

Instead of allocation pte page table upfront, we postpone this until we
have page to map in hands.  This approach opens possibility to map the
page as huge if filesystem supports this.

Comparing to Hugh's patch I've pushed page table allocation a bit
further: into do_set_pte().  This way we can postpone allocation even in
faultaround case without moving do_fault_around() after __do_fault().

do_set_pte() got renamed to alloc_set_pte() as it can allocate page
table if required.

[1] http://lkml.kernel.org/r/alpine.LSU.2.11.1502202015090.14414@eggly.anvils

Link: http://lkml.kernel.org/r/1466021202-61880-10-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce fault_env

The idea borrowed from Peter's patch from patchset on speculative page
faults[1]:

Instead of passing around the endless list of function arguments,
replace the lot with a single structure so we can change context without
endless function signature changes.

The changes are mostly mechanical with exception of faultaround code:
filemap_map_pages() got reworked a bit.

This patch is preparation for the next one.

[1] http://lkml.kernel.org/r/20141020222841.302891540@infradead.org

Link: http://lkml.kernel.org/r/1466021202-61880-9-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: do not pass mm_struct into handle_mm_fault

We always have vma->vm_mm around.

Link: http://lkml.kernel.org/r/1466021202-61880-8-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

thp, mlock: update unevictable-lru.txt

Add description of THP handling into unevictable-lru.txt.

Link: http://lkml.kernel.org/r/1466021202-61880-7-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

khugepaged: recheck pmd after mmap_sem re-acquired

Vlastimil noted[1] that pmd can be no longer valid after we drop
mmap_sem.  We need recheck it once mmap_sem taken again.

[1] http://lkml.kernel.org/r/12918dcd-a695-c6f4-e06f-69141c5f357f@suse.cz

Link: http://lkml.kernel.org/r/1466021202-61880-6-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: fix locking inconsistency in collapse_huge_page

After creating revalidate vma function, locking inconsistency occured
due to directing the code path to wrong label.  This patch directs to
correct label and fix the inconsistency.

Related commit that caused inconsistency:
 http://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=da4360877094368f6dfe75bbe804b0f0a5d575b0

Link: http://lkml.kernel.org/r/1464956884-4644-1-git-send-email-ebru.akagunduz@gmail.com
Link: http://lkml.kernel.org/r/1466021202-61880-4-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Ebru Akagunduz <ebru.akagunduz@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: make swapin readahead under down_read of mmap_sem

Currently khugepaged makes swapin readahead under down_write.  This
patch supplies to make swapin readahead under down_read instead of
down_write.

The patch was tested with a test program that allocates 800MB of memory,
writes to it, and then sleeps.  The system was forced to swap out all.
Afterwards, the test program touches the area by writing, it skips a
page in each 20 pages of the area.

[akpm@linux-foundation.org: update comment to match new code]
[kirill.shutemov@linux.intel.com: passing 'vma' to hugepage_vma_revlidate() is useless]
Link: http://lkml.kernel.org/r/20160530095058.GA53044@black.fi.intel.com
Link: http://lkml.kernel.org/r/1466021202-61880-3-git-send-email-kirill.shutemov@linux.intel.com
Link: http://lkml.kernel.org/r/1464335964-6510-4-git-send-email-ebru.akagunduz@gmail.com
Link: http://lkml.kernel.org/r/1466021202-61880-2-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Ebru Akagunduz <ebru.akagunduz@gmail.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make swapin readahead to improve thp collapse rate

This patch makes swapin readahead to improve thp collapse rate.  When
khugepaged scanned pages, there can be a few of the pages in swap area.

With the patch THP can collapse 4kB pages into a THP when there are up
to max_ptes_swap swap ptes in a 2MB range.

The patch was tested with a test program that allocates 400B of memory,
writes to it, and then sleeps.  I force the system to swap out all.
Afterwards, the test program touches the area by writing, it skips a
page in each 20 pages of the area.

Without the patch, system did not swap in readahead.  THP rate was %65
of the program of the memory, it did not change over time.

With this patch, after 10 minutes of waiting khugepaged had collapsed
%99 of the program's memory.

[kirill.shutemov@linux.intel.com: trivial cleanup of exit path of the function]
[kirill.shutemov@linux.intel.com: __collapse_huge_page_swapin(): drop unused 'pte' parameter]
[kirill.shutemov@linux.intel.com: do not hold anon_vma lock during swap in]
Signed-off-by: Ebru Akagunduz <ebru.akagunduz@gmail.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Xie XiuQi <xiexiuqi@huawei.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make optimistic check for swapin readahead

Introduce a new sysfs integer knob
/sys/kernel/mm/transparent_hugepage/khugepaged/max_ptes_swap which makes
optimistic check for swapin readahead to increase thp collapse rate.
Before getting swapped out pages to memory, checks them and allows up to a
certain number.  It also prints out using tracepoints amount of unmapped
ptes.

[vdavydov@parallels.com: fix scan not aborted on SCAN_EXCEED_SWAP_PTE]
[sfr@canb.auug.org.au: build fix]
Link: http://lkml.kernel.org/r/20160616154503.65806e12@canb.auug.org.au
Signed-off-by: Ebru Akagunduz <ebru.akagunduz@gmail.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Xie XiuQi <xiexiuqi@huawei.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: 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/memblock.c:memblock_add_range(): if nr_new is 0 just return

If nr_new is 0 which means there's no region would be added, so just
return to the caller.

Signed-off-by: nimisolo <nimisolo@gmail.com>
Cc: Alexander Kuleshov <kuleshovmail@gmail.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Wei Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: use consistent gfp flags during readahead

Vladimir has noticed that we might declare memcg oom even during
readahead because read_pages only uses GFP_KERNEL (with mapping_gfp
restriction) while __do_page_cache_readahead uses
page_cache_alloc_readahead which adds __GFP_NORETRY to prevent from
OOMs.  This gfp mask discrepancy is really unfortunate and easily
fixable.  Drop page_cache_alloc_readahead() which only has one user and
outsource the gfp_mask logic into readahead_gfp_mask and propagate this
mask from __do_page_cache_readahead down to read_pages.

This alone would have only very limited impact as most filesystems are
implementing ->readpages and the common implementation mpage_readpages
does GFP_KERNEL (with mapping_gfp restriction) again.  We can tell it to
use readahead_gfp_mask instead as this function is called only during
readahead as well.  The same applies to read_cache_pages.

ext4 has its own ext4_mpage_readpages but the path which has pages !=
NULL can use the same gfp mask.  Btrfs, cifs, f2fs and orangefs are
doing a very similar pattern to mpage_readpages so the same can be
applied to them as well.

[akpm@linux-foundation.org: coding-style fixes]
[mhocko@suse.com: restrict gfp mask in mpage_alloc]
Link: http://lkml.kernel.org/r/20160610074223.GC32285@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/1465301556-26431-1-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Cc: Chris Mason <clm@fb.com>
Cc: Steve French <sfrench@samba.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Jan Kara <jack@suse.cz>
Cc: Mike Marshall <hubcap@omnibond.com>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Cc: Changman Lee <cm224.lee@samsung.com>
Cc: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom_reaper: make sure that mmput_async is called only when memory was reaped

Tetsuo is worried that mmput_async might still lead to a premature new
oom victim selection due to the following race:

__oom_reap_task exit_mm
  find_lock_task_mm
  atomic_inc(mm->mm_users) # = 2
  task_unlock
     task_lock
  task->mm = NULL
  up_read(&mm->mmap_sem)
< somebody write locks mmap_sem >
  task_unlock
  mmput
       atomic_dec_and_test # = 1
  exit_oom_victim
  down_read_trylock # failed - no reclaim
  mmput_async # Takes unpredictable amount of time
   < new OOM situation >

the final __mmput will be executed in the delayed context which might
happen far in the future.  Such a race is highly unlikely because the
write holder of mmap_sem would have to be an external task (all direct
holders are already killed or exiting) and it usually have to pin
mm_users in order to do anything reasonable.

We can, however, make sure that the mmput_async is only called when we
do not back off and reap some memory.  That would reduce the impact of
the delayed __mmput because the real content would be already freed.
Pin mm_count to keep it alive after we drop task_lock and before we try
to get mmap_sem.  If the mmap_sem succeeds we can try to grab mm_users
reference and then go on with unmapping the address space.

It is not clear whether this race is possible at all but it is better to
be more robust and do not pin mm_users unless we are sure we are
actually doing some real work during __oom_reap_task.

Link: http://lkml.kernel.org/r/1465306987-30297-1-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/memblock.h: Clean up code for several trivial details

Correct the function parameters alignment, since original code already
use both tabs and white spaces together for the incorrect parameters
alignment functions.

If one line can hold one statement within 80 columns, let it in one line
(original code did not consider about the tabs/spaces for 2nd line when
a statement is separated into 2 lines).

Try to let '' aligned within one macro, since all related lines are
short enough.

Remove useless statement "idx = 0;", and always assign rgn within the
'for' statement.

Link: http://lkml.kernel.org/r/1464904899-1714-1-git-send-email-chengang@emindsoft.com.cn
Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add NR_ZSMALLOC to vmstat

zram is very popular for some of the embedded world (e.g., TV, mobile
phones).  On those system, zsmalloc's consumed memory size is never
trivial (one of example from real product system, total memory: 800M,
zsmalloc consumed: 150M), so we have used this out of tree patch to
monitor system memory behavior via /proc/vmstat.

With zsmalloc in vmstat, it helps in tracking down system behavior due
to memory usage.

[minchan@kernel.org: zsmalloc: follow up zsmalloc vmstat]
Link: http://lkml.kernel.org/r/20160607091737.GC23435@bbox
[akpm@linux-foundation.org: fix build with CONFIG_ZSMALLOC=m]
Link: http://lkml.kernel.org/r/1464919731-13255-1-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Sangseok Lee <sangseok.lee@lge.com>
Cc: Chanho Min <chanho.min@lge.com>
Cc: Chan Gyun Jeong <chan.jeong@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, frontswap: convert frontswap_enabled to static key

I have noticed that frontswap.h first declares "frontswap_enabled" as
extern bool variable, and then overrides it with "#define
frontswap_enabled (1)" for CONFIG_FRONTSWAP=Y or (0) when disabled.  The
bool variable isn't actually instantiated anywhere.

This all looks like an unfinished attempt to make frontswap_enabled
reflect whether a backend is instantiated.  But in the current state,
all frontswap hooks call unconditionally into frontswap.c just to check
if frontswap_ops is non-NULL.  This should at least be checked inline,
but we can further eliminate the overhead when CONFIG_FRONTSWAP is
enabled and no backend registered, using a static key that is initially
disabled, and gets enabled only upon first backend registration.

Thus, checks for "frontswap_enabled" are replaced with
"frontswap_enabled()" wrapping the static key check.  There are two
exceptions:

- xen's selfballoon_process() was testing frontswap_enabled in code guarded
  by #ifdef CONFIG_FRONTSWAP, which was effectively always true when reachable.
  The patch just removes this check. Using frontswap_enabled() does not sound
  correct here, as this can be true even without xen's own backend being
  registered.

- in SYSCALL_DEFINE2(swapon), change the check to IS_ENABLED(CONFIG_FRONTSWAP)
  as it seems the bitmap allocation cannot currently be postponed until a
  backend is registered. This means that frontswap will still have some
  memory overhead by being configured, but without a backend.

After the patch, we can expect that some functions in frontswap.c are
called only when frontswap_ops is non-NULL.  Change the checks there to
VM_BUG_ONs.  While at it, convert other BUG_ONs to VM_BUG_ONs as
frontswap has been stable for some time.

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/1463152235-9717-1-git-send-email-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,oom: remove unused argument from oom_scan_process_thread().

oom_scan_process_thread() does not use totalpages argument.
oom_badness() uses it.

Link: http://lkml.kernel.org/r/1463796041-7889-1-git-send-email-penguin-kernel@I-love.SAKURA.ne.jp
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

af_unix: charge buffers to kmemcg

Unix sockets can consume a significant amount of system memory, hence
they should be accounted to kmemcg.

Since unix socket buffers are always allocated from process context, all
we need to do to charge them to kmemcg is set __GFP_ACCOUNT in
sock->sk_allocation mask.

Eric asked:

> 1) What happens when a buffer, allocated from socket <A> lands in a
> different socket <B>, maybe owned by another user/process.
>
> Who owns it now, in term of kmemcg accounting ?

We never move memcg charges.  E.g.  if two processes from different
cgroups are sharing a memory region, each page will be charged to the
process which touched it first.  Or if two processes are working with
the same directory tree, inodes and dentries will be charged to the
first user.  The same is fair for unix socket buffers - they will be
charged to the sender.

> 2) Has performance impact been evaluated ?

I ran netperf STREAM_STREAM with default options in a kmemcg on a 4 core
x2 HT box.  The results are below:

 # clients            bandwidth (10^6bits/sec)
                    base              patched
         1      67643 +-  725      64874 +-  353    - 4.0 %
         4     193585 +- 2516     186715 +- 1460    - 3.5 %
         8     194820 +-  377     187443 +- 1229    - 3.7 %

So the accounting doesn't come for free - it takes ~4% of performance.
I believe we could optimize it by using per cpu batching not only on
charge, but also on uncharge in memcg core, but that's beyond the scope
of this patch set - I'll take a look at this later.

Anyway, if performance impact is found to be unacceptable, it is always
possible to disable kmem accounting at boot time (cgroup.memory=nokmem)
or not use memory cgroups at runtime at all (thanks to jump labels
there'll be no overhead even if they are compiled in).

Link: http://lkml.kernel.org/r/fcfe6cae27a59fbc5e40145664b3cf085a560c68.1464079538.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pipe: account to kmemcg

Pipes can consume a significant amount of system memory, hence they
should be accounted to kmemcg.

This patch marks pipe_inode_info and anonymous pipe buffer page
allocations as __GFP_ACCOUNT so that they would be charged to kmemcg.
Note, since a pipe buffer page can be "stolen" and get reused for other
purposes, including mapping to userspace, we clear PageKmemcg thus
resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which
is introduced by this patch.

A note regarding anon_pipe_buf_steal implementation.  We allow to steal
the page if its ref count equals 1.  It looks racy, but it is correct
for anonymous pipe buffer pages, because:

 - We lock out all other pipe users, because ->steal is called with
   pipe_lock held, so the page can't be spliced to another pipe from
   under us.

 - The page is not on LRU and it never was.

 - Thus a parallel thread can access it only by PFN. Although this is
   quite possible (e.g. see page_idle_get_page and balloon_page_isolate)
   this is not dangerous, because all such functions do is increase page
   ref count, check if the page is the one they are looking for, and
   decrease ref count if it isn't. Since our page is clean except for
   PageKmemcg mark, which doesn't conflict with other _mapcount users,
   the worst that can happen is we see page_count > 2 due to a transient
   ref, in which case we false-positively abort ->steal, which is still
   fine, because ->steal is not guaranteed to succeed.

Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch: x86: charge page tables to kmemcg

Page tables can bite a relatively big chunk off system memory and their
allocations are easy to trigger from userspace, so they should be
accounted to kmemcg.

This patch marks page table allocations as __GFP_ACCOUNT for x86.  Note
we must not charge allocations of kernel page tables, because they can
be shared among processes from different cgroups so accounting them to a
particular one can pin other cgroups for indefinitely long.  So we clear
__GFP_ACCOUNT flag if a page table is allocated for the kernel.

Link: http://lkml.kernel.org/r/7d5c54f6a2bcbe76f03171689440003d87e6c742.1464079538.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: teach uncharge_list to deal with kmem pages

Page table pages are batched-freed in release_pages on most
architectures.  If we want to charge them to kmemcg (this is what is
done later in this series), we need to teach mem_cgroup_uncharge_list to
handle kmem pages.

Link: http://lkml.kernel.org/r/18d5c09e97f80074ed25b97a7d0f32b95d875717.1464079538.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: charge/uncharge kmemcg from generic page allocator paths

Currently, to charge a non-slab allocation to kmemcg one has to use
alloc_kmem_pages helper with __GFP_ACCOUNT flag.  A page allocated with
this helper should finally be freed using free_kmem_pages, otherwise it
won't be uncharged.

This API suits its current users fine, but it turns out to be impossible
to use along with page reference counting, i.e.  when an allocation is
supposed to be freed with put_page, as it is the case with pipe or unix
socket buffers.

To overcome this limitation, this patch moves charging/uncharging to
generic page allocator paths, i.e.  to __alloc_pages_nodemask and
free_pages_prepare, and zaps alloc/free_kmem_pages helpers.  This way,
one can use any of the available page allocation functions to get the
allocated page charged to kmemcg - it's enough to pass __GFP_ACCOUNT,
just like in case of kmalloc and friends.  A charged page will be
automatically uncharged on free.

To make it possible, we need to mark pages charged to kmemcg somehow.
To avoid introducing a new page flag, we make use of page->_mapcount for
marking such pages.  Since pages charged to kmemcg are not supposed to
be mapped to userspace, it should work just fine.  There are other
(ab)users of page->_mapcount - buddy and balloon pages - but we don't
conflict with them.

In case kmemcg is compiled out or not used at runtime, this patch
introduces no overhead to generic page allocator paths.  If kmemcg is
used, it will be plus one gfp flags check on alloc and plus one
page->_mapcount check on free, which shouldn't hurt performance, because
the data accessed are hot.

Link: http://lkml.kernel.org/r/a9736d856f895bcb465d9f257b54efe32eda6f99.1464079538.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: cleanup kmem charge functions

 - Handle memcg_kmem_enabled check out to the caller. This reduces the
   number of function definitions making the code easier to follow. At
   the same time it doesn't result in code bloat, because all of these
   functions are used only in one or two places.

 - Move __GFP_ACCOUNT check to the caller as well so that one wouldn't
   have to dive deep into memcg implementation to see which allocations
   are charged and which are not.

 - Refresh comments.

Link: http://lkml.kernel.org/r/52882a28b542c1979fd9a033b4dc8637fc347399.1464079537.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: clean up non-standard page->_mapcount users

 - Add a proper comment to page->_mapcount.

 - Introduce a macro for generating helper functions.

 - Place all special page->_mapcount values next to each other so that
   readers can see all possible values and so we don't get duplicates.

Link: http://lkml.kernel.org/r/502f49000e0b63e6c62e338fac6b420bf34fb526.1464079537.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove pointless struct in struct page definition

This patchset implements per kmemcg accounting of page tables
(x86-only), pipe buffers, and unix socket buffers.

Patches 1-3 are just cleanups that are not supposed to introduce any
functional changes.  Patches 4 and 5 move charge/uncharge to generic
page allocator paths for the sake of accounting pipe and unix socket
buffers.  Patches 5-7 make x86 page tables, pipe buffers, and unix
socket buffers accountable.

This patch (of 8):

... to reduce indentation level thus leaving more space for comments.

Link: http://lkml.kernel.org/r/f34ffe70fce2b0b9220856437f77972d67c14275.1464079537.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmu_gather: track page size with mmu gather and force flush if page size change

This allows an arch which needs to do special handing with respect to
different page size when flushing tlb to implement the same in mmu
gather.

Link: http://lkml.kernel.org/r/1465049193-22197-3-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: change the interface for __tlb_remove_page()

This updates the generic and arch specific implementation to return true
if we need to do a tlb flush.  That means if a __tlb_remove_page
indicate a flush is needed, the page we try to remove need to be tracked
and added again after the flush.  We need to track it because we have
already update the pte to none and we can't just loop back.

This change is done to enable us to do a tlb_flush when we try to flush
a range that consists of different page sizes.  For architectures like
ppc64, we can do a range based tlb flush and we need to track page size
for that.  When we try to remove a huge page, we will force a tlb flush
and starts a new mmu gather.

[aneesh.kumar@linux.vnet.ibm.com: mm-change-the-interface-for-__tlb_remove_page-v3]
Link: http://lkml.kernel.org/r/1465049193-22197-2-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1464860389-29019-2-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: simplify hugetlb unmap

For hugetlb like THP (and unlike regular page), we do tlb flush after
dropping ptl.  Because of the above, we don't need to track force_flush
like we do now.  Instead we can simply call tlb_remove_page() which will
do the flush if needed.

No functionality change in this patch.

Link: http://lkml.kernel.org/r/1465049193-22197-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: thp: check pmd_trans_unstable() after split_huge_pmd()

split_huge_pmd() doesn't guarantee that the pmd is normal pmd pointing
to pte entries, which can be checked with pmd_trans_unstable().  Some
callers make this assertion and some do it differently and some not, so
let's do it in a unified manner.

Link: http://lkml.kernel.org/r/1464741400-12143-1-git-send-email-n-horiguchi@ah.jp.nec.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Hugh Dickins <hughd@google.com>
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/page_isolation: clean up confused code

When there is an isolated_page, post_alloc_hook() is called with page
but __free_pages() is called with isolated_page.  Since they are the
same so no problem but it's very confusing.  To reduce it, this patch
changes isolated_page to boolean type and uses page variable
consistently.

Link: http://lkml.kernel.org/r/1466150259-27727-10-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.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/page_alloc: introduce post allocation processing on page allocator

This patch is motivated from Hugh and Vlastimil's concern [1].

There are two ways to get freepage from the allocator.  One is using
normal memory allocation API and the other is __isolate_free_page()
which is internally used for compaction and pageblock isolation.  Later
usage is rather tricky since it doesn't do whole post allocation
processing done by normal API.

One problematic thing I already know is that poisoned page would not be
checked if it is allocated by __isolate_free_page().  Perhaps, there
would be more.

We could add more debug logic for allocated page in the future and this
separation would cause more problem.  I'd like to fix this situation at
this time.  Solution is simple.  This patch commonize some logic for
newly allocated page and uses it on all sites.  This will solve the
problem.

[1] http://marc.info/?i=alpine.LSU.2.11.1604270029350.7066%40eggly.anvils%3E

[iamjoonsoo.kim@lge.com: mm-page_alloc-introduce-post-allocation-processing-on-page-allocator-v3]
Link: http://lkml.kernel.org/r/1464230275-25791-7-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1466150259-27727-9-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1464230275-25791-7-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_owner: use stackdepot to store stacktrace

Currently, we store each page's allocation stacktrace on corresponding
page_ext structure and it requires a lot of memory.  This causes the
problem that memory tight system doesn't work well if page_owner is
enabled.  Moreover, even with this large memory consumption, we cannot
get full stacktrace because we allocate memory at boot time and just
maintain 8 stacktrace slots to balance memory consumption.  We could
increase it to more but it would make system unusable or change system
behaviour.

To solve the problem, this patch uses stackdepot to store stacktrace.
It obviously provides memory saving but there is a drawback that
stackdepot could fail.

stackdepot allocates memory at runtime so it could fail if system has
not enough memory.  But, most of allocation stack are generated at very
early time and there are much memory at this time.  So, failure would
not happen easily.  And, one failure means that we miss just one page's
allocation stacktrace so it would not be a big problem.  In this patch,
when memory allocation failure happens, we store special stracktrace
handle to the page that is failed to save stacktrace.  With it, user can
guess memory usage properly even if failure happens.

Memory saving looks as following.  (4GB memory system with page_owner)
(before the patch -> after the patch)

static allocation:
92274688 bytes -> 25165824 bytes

dynamic allocation after boot + kernel build:
0 bytes -> 327680 bytes

total:
92274688 bytes -> 25493504 bytes

72% reduction in total.

Note that implementation looks complex than someone would imagine
because there is recursion issue.  stackdepot uses page allocator and
page_owner is called at page allocation.  Using stackdepot in page_owner
could re-call page allcator and then page_owner.  That is a recursion.
To detect and avoid it, whenever we obtain stacktrace, recursion is
checked and page_owner is set to dummy information if found.  Dummy
information means that this page is allocated for page_owner feature
itself (such as stackdepot) and it's understandable behavior for user.

[iamjoonsoo.kim@lge.com: mm-page_owner-use-stackdepot-to-store-stacktrace-v3]
Link: http://lkml.kernel.org/r/1464230275-25791-6-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1466150259-27727-7-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1464230275-25791-6-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page_owner: increase temporary buffer size

Page owner will be changed to store more deep stacktrace so current
temporary buffer size isn't enough.  Increase it.

Link: http://lkml.kernel.org/r/1464230275-25791-5-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_owner: introduce split_page_owner and replace manual handling

split_page() calls set_page_owner() to set up page_owner to each pages.
But, it has a drawback that head page and the others have different
stacktrace because callsite of set_page_owner() is slightly differnt.
To avoid this problem, this patch copies head page's page_owner to the
others.  It needs to introduce new function, split_page_owner() but it
also remove the other function, get_page_owner_gfp() so looks good to
do.

Link: http://lkml.kernel.org/r/1464230275-25791-4-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_owner: copy last_migrate_reason in copy_page_owner()

Currently, copy_page_owner() doesn't copy all the owner information.  It
skips last_migrate_reason because copy_page_owner() is used for
migration and it will be properly set soon.  But, following patch will
use copy_page_owner() and this skip will cause the problem that
allocated page has uninitialied last_migrate_reason.  To prevent it,
this patch also copy last_migrate_reason in copy_page_owner().

Link: http://lkml.kernel.org/r/1464230275-25791-3-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_owner: initialize page owner without holding the zone lock

It's not necessary to initialized page_owner with holding the zone lock.
It would cause more contention on the zone lock although it's not a big
problem since it is just debug feature.  But, it is better than before
so do it.  This is also preparation step to use stackdepot in page owner
feature.  Stackdepot allocates new pages when there is no reserved space
and holding the zone lock in this case will cause deadlock.

Link: http://lkml.kernel.org/r/1464230275-25791-2-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction: split freepages without holding the zone lock

We don't need to split freepages with holding the zone lock.  It will
cause more contention on zone lock so not desirable.

[rientjes@google.com: if __isolate_free_page() fails, avoid adding to freelist so we don't call map_pages() with it]
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1606211447001.43430@chino.kir.corp.google.com
Link: http://lkml.kernel.org/r/1464230275-25791-1-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: use OBJ_TAG_BIT for bit shifter

Static check warns using tag as bit shifter.  It doesn't break current
working but not good for redability.  Let's use OBJ_TAG_BIT as bit
shifter instead of OBJ_ALLOCATED_TAG.

Link: http://lkml.kernel.org/r/20160607045146.GF26230@bbox
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: use __GFP_MOVABLE for memory allocation

Zsmalloc is ready for page migration so zram can use __GFP_MOVABLE from
now on.

I did test to see how it helps to make higher order pages.  Test
scenario is as follows.

KVM guest, 1G memory, ext4 formated zram block device,

  for i in `seq 1 8`;
  do
          dd if=/dev/vda1 of=mnt/test$i.txt bs=128M count=1 &
  done

  wait `pidof dd`

  for i in `seq 1 2 8`;
  do
          rm -rf mnt/test$i.txt
  done
  fstrim -v mnt

  echo "init"
  cat /proc/buddyinfo

  echo "compaction"
  echo 1 > /proc/sys/vm/compact_memory
  cat /proc/buddyinfo

old:

  init
  Node 0, zone      DMA    208    120     51     41     11      0      0      0      0      0      0
  Node 0, zone    DMA32  16380  13777   9184   3805    789     54      3      0      0      0      0
  compaction
  Node 0, zone      DMA    132     82     40     39     16      2      1      0      0      0      0
  Node 0, zone    DMA32   5219   5526   4969   3455   1831    677    139     15      0      0      0

new:

  init
  Node 0, zone      DMA    379    115     97     19      2      0      0      0      0      0      0
  Node 0, zone    DMA32  18891  16774  10862   3947    637     21      0      0      0      0      0
  compaction
  Node 0, zone      DMA    214     66     87     29     10      3      0      0      0      0      0
  Node 0, zone    DMA32   1612   3139   3154   2469   1745    990    384     94      7      0      0

As you can see, compaction made so many high-order pages. Yay!

Link: http://lkml.kernel.org/r/1464736881-24886-13-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: page migration support

This patch introduces run-time migration feature for zspage.

For migration, VM uses page.lru field so it would be better to not use
page.next field which is unified with page.lru for own purpose.  For
that, firstly, we can get first object offset of the page via runtime
calculation instead of using page.index so we can use page.index as link
for page chaining instead of page.next.

In case of huge object, it stores handle to page.index instead of next
link of page chaining because huge object doesn't need to next link for
page chaining.  So get_next_page need to identify huge object to return
NULL.  For it, this patch uses PG_owner_priv_1 flag of the page flag.

For migration, it supports three functions

* zs_page_isolate

It isolates a zspage which includes a subpage VM want to migrate from
class so anyone cannot allocate new object from the zspage.

We could try to isolate a zspage by the number of subpage so subsequent
isolation trial of other subpage of the zpsage shouldn't fail.  For
that, we introduce zspage.isolated count.  With that, zs_page_isolate
can know whether zspage is already isolated or not for migration so if
it is isolated for migration, subsequent isolation trial can be
successful without trying further isolation.

* zs_page_migrate

First of all, it holds write-side zspage->lock to prevent migrate other
subpage in zspage.  Then, lock all objects in the page VM want to
migrate.  The reason we should lock all objects in the page is due to
race between zs_map_object and zs_page_migrate.

  zs_map_object zs_page_migrate

  pin_tag(handle)
  obj = handle_to_obj(handle)
  obj_to_location(obj, &page, &obj_idx);

write_lock(&zspage->lock)
if (!trypin_tag(handle))
goto unpin_object

  zspage = get_zspage(page);
  read_lock(&zspage->lock);

If zs_page_migrate doesn't do trypin_tag, zs_map_object's page can be
stale by migration so it goes crash.

If it locks all of objects successfully, it copies content from old page
to new one, finally, create new zspage chain with new page.  And if it's
last isolated subpage in the zspage, put the zspage back to class.

* zs_page_putback

It returns isolated zspage to right fullness_group list if it fails to
migrate a page.  If it find a zspage is ZS_EMPTY, it queues zspage
freeing to workqueue.  See below about async zspage freeing.

This patch introduces asynchronous zspage free.  The reason to need it
is we need page_lock to clear PG_movable but unfortunately, zs_free path
should be atomic so the apporach is try to grab page_lock.  If it got
page_lock of all of pages successfully, it can free zspage immediately.
Otherwise, it queues free request and free zspage via workqueue in
process context.

If zs_free finds the zspage is isolated when it try to free zspage, it
delays the freeing until zs_page_putback finds it so it will free free
the zspage finally.

In this patch, we expand fullness_list from ZS_EMPTY to ZS_FULL.  First
of all, it will use ZS_EMPTY list for delay freeing.  And with adding
ZS_FULL list, it makes to identify whether zspage is isolated or not via
list_empty(&zspage->list) test.

[minchan@kernel.org: zsmalloc: keep first object offset in struct page]
Link: http://lkml.kernel.org/r/1465788015-23195-1-git-send-email-minchan@kernel.org
[minchan@kernel.org: zsmalloc: zspage sanity check]
Link: http://lkml.kernel.org/r/20160603010129.GC3304@bbox
Link: http://lkml.kernel.org/r/1464736881-24886-12-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: use freeobj for index

Zsmalloc stores first free object's <PFN, obj_idx> position into freeobj
in each zspage.  If we change it with index from first_page instead of
position, it makes page migration simple because we don't need to
correct other entries for linked list if a page is migrated out.

Link: http://lkml.kernel.org/r/1464736881-24886-11-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: separate free_zspage from putback_zspage

Currently, putback_zspage does free zspage under class->lock if fullness
become ZS_EMPTY but it makes trouble to implement locking scheme for new
zspage migration.  So, this patch is to separate free_zspage from
putback_zspage and free zspage out of class->lock which is preparation
for zspage migration.

Link: http://lkml.kernel.org/r/1464736881-24886-10-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: introduce zspage structure

We have squeezed meta data of zspage into first page's descriptor.  So,
to get meta data from subpage, we should get first page first of all.
But it makes trouble to implment page migration feature of zsmalloc
because any place where to get first page from subpage can be raced with
first page migration.  IOW, first page it got could be stale.  For
preventing it, I have tried several approahces but it made code
complicated so finally, I concluded to separate metadata from first
page.  Of course, it consumes more memory.  IOW, 16bytes per zspage on
32bit at the moment.  It means we lost 1% at *worst case*(40B/4096B)
which is not bad I think at the cost of maintenance.

Link: http://lkml.kernel.org/r/1464736881-24886-9-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: factor page chain functionality out

For page migration, we need to create page chain of zspage dynamically
so this patch factors it out from alloc_zspage.

Link: http://lkml.kernel.org/r/1464736881-24886-8-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: use accessor

Upcoming patch will change how to encode zspage meta so for easy review,
this patch wraps code to access metadata as accessor.

Link: http://lkml.kernel.org/r/1464736881-24886-7-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: use bit_spin_lock

Use kernel standard bit spin-lock instead of custom mess.  Even, it has
a bug which doesn't disable preemption.  The reason we don't have any
problem is that we have used it during preemption disable section by
class->lock spinlock.  So no need to go to stable.

Link: http://lkml.kernel.org/r/1464736881-24886-6-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zsmalloc: keep max_object in size_class

Every zspage in a size_class has same number of max objects so we could
move it to a size_class.

Link: http://lkml.kernel.org/r/1464736881-24886-5-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: balloon: use general non-lru movable page feature

Now, VM has a feature to migrate non-lru movable pages so balloon
doesn't need custom migration hooks in migrate.c and compaction.c.

Instead, this patch implements the page->mapping->a_ops->
{isolate|migrate|putback} functions.

With that, we could remove hooks for ballooning in general migration
functions and make balloon compaction simple.

[akpm@linux-foundation.org: compaction.h requires that the includer first include node.h]
Link: http://lkml.kernel.org/r/1464736881-24886-4-git-send-email-minchan@kernel.org
Signed-off-by: Gioh Kim <gi-oh.kim@profitbricks.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Rafael Aquini <aquini@redhat.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: migrate: support non-lru movable page migration

We have allowed migration for only LRU pages until now and it was enough
to make high-order pages.  But recently, embedded system(e.g., webOS,
android) uses lots of non-movable pages(e.g., zram, GPU memory) so we
have seen several reports about troubles of small high-order allocation.
For fixing the problem, there were several efforts (e,g,.  enhance
compaction algorithm, SLUB fallback to 0-order page, reserved memory,
vmalloc and so on) but if there are lots of non-movable pages in system,
their solutions are void in the long run.

So, this patch is to support facility to change non-movable pages with
movable.  For the feature, this patch introduces functions related to
migration to address_space_operations as well as some page flags.

If a driver want to make own pages movable, it should define three
functions which are function pointers of struct
address_space_operations.

1. bool (*isolate_page) (struct page *page, isolate_mode_t mode);

What VM expects on isolate_page function of driver is to return *true*
if driver isolates page successfully.  On returing true, VM marks the
page as PG_isolated so concurrent isolation in several CPUs skip the
page for isolation.  If a driver cannot isolate the page, it should
return *false*.

Once page is successfully isolated, VM uses page.lru fields so driver
shouldn't expect to preserve values in that fields.

2. int (*migratepage) (struct address_space *mapping,
struct page *newpage, struct page *oldpage, enum migrate_mode);

After isolation, VM calls migratepage of driver with isolated page.  The
function of migratepage is to move content of the old page to new page
and set up fields of struct page newpage.  Keep in mind that you should
indicate to the VM the oldpage is no longer movable via
__ClearPageMovable() under page_lock if you migrated the oldpage
successfully and returns 0.  If driver cannot migrate the page at the
moment, driver can return -EAGAIN.  On -EAGAIN, VM will retry page
migration in a short time because VM interprets -EAGAIN as "temporal
migration failure".  On returning any error except -EAGAIN, VM will give
up the page migration without retrying in this time.

Driver shouldn't touch page.lru field VM using in the functions.

3. void (*putback_page)(struct page *);

If migration fails on isolated page, VM should return the isolated page
to the driver so VM calls driver's putback_page with migration failed
page.  In this function, driver should put the isolated page back to the
own data structure.

4. non-lru movable page flags

There are two page flags for supporting non-lru movable page.

* PG_movable

Driver should use the below function to make page movable under
page_lock.

void __SetPageMovable(struct page *page, struct address_space *mapping)

It needs argument of address_space for registering migration family
functions which will be called by VM.  Exactly speaking, PG_movable is
not a real flag of struct page.  Rather than, VM reuses page->mapping's
lower bits to represent it.

#define PAGE_MAPPING_MOVABLE 0x2
page->mapping = page->mapping | PAGE_MAPPING_MOVABLE;

so driver shouldn't access page->mapping directly.  Instead, driver
should use page_mapping which mask off the low two bits of page->mapping
so it can get right struct address_space.

For testing of non-lru movable page, VM supports __PageMovable function.
However, it doesn't guarantee to identify non-lru movable page because
page->mapping field is unified with other variables in struct page.  As
well, if driver releases the page after isolation by VM, page->mapping
doesn't have stable value although it has PAGE_MAPPING_MOVABLE (Look at
__ClearPageMovable).  But __PageMovable is cheap to catch whether page
is LRU or non-lru movable once the page has been isolated.  Because LRU
pages never can have PAGE_MAPPING_MOVABLE in page->mapping.  It is also
good for just peeking to test non-lru movable pages before more
expensive checking with lock_page in pfn scanning to select victim.

For guaranteeing non-lru movable page, VM provides PageMovable function.
Unlike __PageMovable, PageMovable functions validates page->mapping and
mapping->a_ops->isolate_page under lock_page.  The lock_page prevents
sudden destroying of page->mapping.

Driver using __SetPageMovable should clear the flag via
__ClearMovablePage under page_lock before the releasing the page.

* PG_isolated

To prevent concurrent isolation among several CPUs, VM marks isolated
page as PG_isolated under lock_page.  So if a CPU encounters PG_isolated
non-lru movable page, it can skip it.  Driver doesn't need to manipulate
the flag because VM will set/clear it automatically.  Keep in mind that
if driver sees PG_isolated page, it means the page have been isolated by
VM so it shouldn't touch page.lru field.  PG_isolated is alias with
PG_reclaim flag so driver shouldn't use the flag for own purpose.

[opensource.ganesh@gmail.com: mm/compaction: remove local variable is_lru]
Link: http://lkml.kernel.org/r/20160618014841.GA7422@leo-test
Link: http://lkml.kernel.org/r/1464736881-24886-3-git-send-email-minchan@kernel.org
Signed-off-by: Gioh Kim <gi-oh.kim@profitbricks.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Ganesh Mahendran <opensource.ganesh@gmail.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rafael Aquini <aquini@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: John Einar Reitan <john.reitan@foss.arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use put_page() to free page instead of putback_lru_page()

Recently, I got many reports about perfermance degradation in embedded
system(Android mobile phone, webOS TV and so on) and easy fork fail.

The problem was fragmentation caused by zram and GPU driver mainly.
With memory pressure, their pages were spread out all of pageblock and
it cannot be migrated with current compaction algorithm which supports
only LRU pages.  In the end, compaction cannot work well so reclaimer
shrinks all of working set pages.  It made system very slow and even to
fail to fork easily which requires order-[2 or 3] allocations.

Other pain point is that they cannot use CMA memory space so when OOM
kill happens, I can see many free pages in CMA area, which is not memory
efficient.  In our product which has big CMA memory, it reclaims zones
too exccessively to allocate GPU and zram page although there are lots
of free space in CMA so system becomes very slow easily.

To solve these problem, this patch tries to add facility to migrate
non-lru pages via introducing new functions and page flags to help
migration.

struct address_space_operations {
..
..
bool (*isolate_page)(struct page *, isolate_mode_t);
void (*putback_page)(struct page *);
..
}

new page flags

PG_movable
PG_isolated

For details, please read description in "mm: migrate: support non-lru
movable page migration".

Originally, Gioh Kim had tried to support this feature but he moved so I
took over the work.  I took many code from his work and changed a little
bit and Konstantin Khlebnikov helped Gioh a lot so he should deserve to
have many credit, too.

And I should mention Chulmin who have tested this patchset heavily so I
can find many bugs from him.  :)

Thanks, Gioh, Konstantin and Chulmin!

This patchset consists of five parts.

1. clean up migration
  mm: use put_page to free page instead of putback_lru_page

2. add non-lru page migration feature
  mm: migrate: support non-lru movable page migration

3. rework KVM memory-ballooning
  mm: balloon: use general non-lru movable page feature

4. zsmalloc refactoring for preparing page migration
  zsmalloc: keep max_object in size_class
  zsmalloc: use bit_spin_lock
  zsmalloc: use accessor
  zsmalloc: factor page chain functionality out
  zsmalloc: introduce zspage structure
  zsmalloc: separate free_zspage from putback_zspage
  zsmalloc: use freeobj for index

5. zsmalloc page migration
  zsmalloc: page migration support
  zram: use __GFP_MOVABLE for memory allocation

This patch (of 12):

Procedure of page migration is as follows:

First of all, it should isolate a page from LRU and try to migrate the
page.  If it is successful, it releases the page for freeing.
Otherwise, it should put the page back to LRU list.

For LRU pages, we have used putback_lru_page for both freeing and
putback to LRU list.  It's okay because put_page is aware of LRU list so
if it releases last refcount of the page, it removes the page from LRU
list.  However, It makes unnecessary operations (e.g., lru_cache_add,
pagevec and flags operations.  It would be not significant but no worth
to do) and harder to support new non-lru page migration because put_page
isn't aware of non-lru page's data structure.

To solve the problem, we can add new hook in put_page with PageMovable
flags check but it can increase overhead in hot path and needs new
locking scheme to stabilize the flag check with put_page.

So, this patch cleans it up to divide two semantic(ie, put and putback).
If migration is successful, use put_page instead of putback_lru_page and
use putback_lru_page only on failure.  That makes code more readable and
doesn't add overhead in put_page.

Comment from Vlastimil
 "Yeah, and compaction (perhaps also other migration users) has to drain
  the lru pvec...  Getting rid of this stuff is worth even by itself."

Link: http://lkml.kernel.org/r/1464736881-24886-2-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: drop gfp_t from zcomp_strm_alloc()

We now allocate streams from CPU_UP hot-plug path, there are no
context-dependent stream allocations anymore and we can schedule from
zcomp_strm_alloc().  Use GFP_KERNEL directly and drop a gfp_t parameter.

Link: http://lkml.kernel.org/r/20160531122017.2878-9-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
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>

zram: add more compression algorithms

Add "deflate", "lz4hc", "842" algorithms to the list of known
compression backends.  The real availability of those algorithms,
however, depends on the corresponding CONFIG_CRYPTO_FOO config options.

[sergey.senozhatsky@gmail.com: zram-add-more-compression-algorithms-v3]
Link: http://lkml.kernel.org/r/20160604024902.11778-7-sergey.senozhatsky@gmail.com
Link: http://lkml.kernel.org/r/20160531122017.2878-8-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
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>

zram: delete custom lzo/lz4

Remove lzo/lz4 backends, we use crypto API now.

[sergey.senozhatsky@gmail.com: zram-delete-custom-lzo-lz4-v3]
Link: http://lkml.kernel.org/r/20160604024902.11778-6-sergey.senozhatsky@gmail.com
Link: http://lkml.kernel.org/r/20160531122017.2878-7-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
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>

zram: cosmetic: cleanup documentation

zram documentation is a mix of different styles: spaces, tabs, tabs +
spaces, etc.  Clean it up.

Link: http://lkml.kernel.org/r/20160531122017.2878-6-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: use crypto api to check alg availability

There is no way to get a string with all the crypto comp algorithms
supported by the crypto comp engine, so we need to maintain our own
backends list.  At the same time we additionally need to use
crypto_has_comp() to make sure that the user has requested a compression
algorithm that is recognized by the crypto comp engine.  Relying on
/proc/crypto is not an options here, because it does not show
not-yet-inserted compression modules.

Example:

 modprobe zram
 cat /proc/crypto | grep -i lz4
 modprobe lz4
 cat /proc/crypto | grep -i lz4
name         : lz4
driver       : lz4-generic
module       : lz4

So the user can't tell exactly if the lz4 is really supported from
/proc/crypto output, unless someone or something has loaded it.

This patch also adds crypto_has_comp() to zcomp_available_show().  We
store all the compression algorithms names in zcomp's `backends' array,
regardless the CONFIG_CRYPTO_FOO configuration, but show only those that
are also supported by crypto engine.  This helps user to know the exact
list of compression algorithms that can be used.

Example:
  module lz4 is not loaded yet, but is supported by the crypto
  engine. /proc/crypto has no information on this module, while
  zram's `comp_algorithm' lists it:

 cat /proc/crypto | grep -i lz4

 cat /sys/block/zram0/comp_algorithm
[lzo] lz4 deflate lz4hc 842

We still use the `backends' array to determine if the requested
compression backend is known to crypto api.  This array, however, may not
contain some entries, therefore as the last step we call crypto_has_comp()
function which attempts to insmod the requested compression algorithm to
determine if crypto api supports it.  The advantage of this method is that
now we permit the usage of out-of-tree crypto compression modules
(implementing S/W or H/W compression).

[sergey.senozhatsky@gmail.com: zram-use-crypto-api-to-check-alg-availability-v3]
Link: http://lkml.kernel.org/r/20160604024902.11778-4-sergey.senozhatsky@gmail.com
Link: http://lkml.kernel.org/r/20160531122017.2878-5-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-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: switch to crypto compress API

We don't have an idle zstreams list anymore and our write path now works
absolutely differently, preventing preemption during compression.  This
removes possibilities of read paths preempting writes at wrong places
(which could badly affect the performance of both paths) and at the same
time opens the door for a move from custom LZO/LZ4 compression backends
implementation to a more generic one, using crypto compress API.

Joonsoo Kim [1] attempted to do this a while ago, but faced with the
need of introducing a new crypto API interface.  The root cause was the
fact that crypto API compression algorithms require a compression stream
structure (in zram terminology) for both compression and decompression
ops, while in reality only several of compression algorithms really need
it.  This resulted in a concept of context-less crypto API compression
backends [2].  Both write and read paths, though, would have been
executed with the preemption enabled, which in the worst case could have
resulted in a decreased worst-case performance, e.g.  consider the
following case:

CPU0

zram_write()
  spin_lock()
    take the last idle stream
  spin_unlock()

<< preempted >>

zram_read()
  spin_lock()
   no idle streams
  spin_unlock()
  schedule()

resuming zram_write compression()

but it took me some time to realize that, and it took even longer to
evolve zram and to make it ready for crypto API.  The key turned out to be
-- drop the idle streams list entirely.  Without the idle streams list we
are free to use compression algorithms that require compression stream for
decompression (read), because streams are now placed in per-cpu data and
each write path has to disable preemption for compression op, almost
completely eliminating the aforementioned case (technically, we still have
a small chance, because write path has a fast and a slow paths and the
slow path is executed with the preemption enabled; but the frequency of
failed fast path is too low).

TEST
====

- 4 CPUs, x86_64 system
- 3G zram, lzo
- fio tests: read, randread, write, randwrite, rw, randrw

test script [3] command:
 ZRAM_SIZE=3G LOG_SUFFIX=XXXX FIO_LOOPS=5 ./zram-fio-test.sh

                   BASE           PATCHED
jobs1
READ:           2527.2MB/s  2482.7MB/s
READ:           2102.7MB/s  2045.0MB/s
WRITE:          1284.3MB/s  1324.3MB/s
WRITE:          1080.7MB/s  1101.9MB/s
READ:           430125KB/s  437498KB/s
WRITE:          430538KB/s  437919KB/s
READ:           399593KB/s  403987KB/s
WRITE:          399910KB/s  404308KB/s
jobs2
READ:           8133.5MB/s  7854.8MB/s
READ:           7086.6MB/s  6912.8MB/s
WRITE:          3177.2MB/s  3298.3MB/s
WRITE:          2810.2MB/s  2871.4MB/s
READ:           1017.6MB/s  1023.4MB/s
WRITE:          1018.2MB/s  1023.1MB/s
READ:           977836KB/s  984205KB/s
WRITE:          979435KB/s  985814KB/s
jobs3
READ:           13557MB/s  13391MB/s
READ:           11876MB/s  11752MB/s
WRITE:          4641.5MB/s  4682.1MB/s
WRITE:          4164.9MB/s  4179.3MB/s
READ:           1453.8MB/s  1455.1MB/s
WRITE:          1455.1MB/s  1458.2MB/s
READ:           1387.7MB/s  1395.7MB/s
WRITE:          1386.1MB/s  1394.9MB/s
jobs4
READ:           20271MB/s  20078MB/s
READ:           18033MB/s  17928MB/s
WRITE:          6176.8MB/s  6180.5MB/s
WRITE:          5686.3MB/s  5705.3MB/s
READ:           2009.4MB/s  2006.7MB/s
WRITE:          2007.5MB/s  2004.9MB/s
READ:           1929.7MB/s  1935.6MB/s
WRITE:          1926.8MB/s  1932.6MB/s
jobs5
READ:           18823MB/s  19024MB/s
READ:           18968MB/s  19071MB/s
WRITE:          6191.6MB/s  6372.1MB/s
WRITE:          5818.7MB/s  5787.1MB/s
READ:           2011.7MB/s  1981.3MB/s
WRITE:          2011.4MB/s  1980.1MB/s
READ:           1949.3MB/s  1935.7MB/s
WRITE:          1940.4MB/s  1926.1MB/s
jobs6
READ:           21870MB/s  21715MB/s
READ:           19957MB/s  19879MB/s
WRITE:          6528.4MB/s  6537.6MB/s
WRITE:          6098.9MB/s  6073.6MB/s
READ:           2048.6MB/s  2049.9MB/s
WRITE:          2041.7MB/s  2042.9MB/s
READ:           2013.4MB/s  1990.4MB/s
WRITE:          2009.4MB/s  1986.5MB/s
jobs7
READ:           21359MB/s  21124MB/s
READ:           19746MB/s  19293MB/s
WRITE:          6660.4MB/s  6518.8MB/s
WRITE:          6211.6MB/s  6193.1MB/s
READ:           2089.7MB/s  2080.6MB/s
WRITE:          2085.8MB/s  2076.5MB/s
READ:           2041.2MB/s  2052.5MB/s
WRITE:          2037.5MB/s  2048.8MB/s
jobs8
READ:           20477MB/s  19974MB/s
READ:           18922MB/s  18576MB/s
WRITE:          6851.9MB/s  6788.3MB/s
WRITE:          6407.7MB/s  6347.5MB/s
READ:           2134.8MB/s  2136.1MB/s
WRITE:          2132.8MB/s  2134.4MB/s
READ:           2074.2MB/s  2069.6MB/s
WRITE:          2087.3MB/s  2082.4MB/s
jobs9
READ:           19797MB/s  19994MB/s
READ:           18806MB/s  18581MB/s
WRITE:          6878.7MB/s  6822.7MB/s
WRITE:          6456.8MB/s  6447.2MB/s
READ:           2141.1MB/s  2154.7MB/s
WRITE:          2144.4MB/s  2157.3MB/s
READ:           2084.1MB/s  2085.1MB/s
WRITE:          2091.5MB/s  2092.5MB/s
jobs10
READ:           19794MB/s  19784MB/s
READ:           18794MB/s  18745MB/s
WRITE:          6984.4MB/s  6676.3MB/s
WRITE:          6532.3MB/s  6342.7MB/s
READ:           2150.6MB/s  2155.4MB/s
WRITE:          2156.8MB/s  2161.5MB/s
READ:           2106.4MB/s  2095.6MB/s
WRITE:          2109.7MB/s  2098.4MB/s

                                    BASE                       PATCHED
jobs1                              perfstat
stalled-cycles-frontend     102,480,595,419 (  41.53%)   114,508,864,804 (  46.92%)
stalled-cycles-backend       51,941,417,832 (  21.05%)    46,836,112,388 (  19.19%)
instructions                283,612,054,215 (    1.15)   283,918,134,959 (    1.16)
branches                     56,372,560,385 ( 724.923)    56,449,814,753 ( 733.766)
branch-misses                   374,826,000 (   0.66%)       326,935,859 (   0.58%)
jobs2                              perfstat
stalled-cycles-frontend     155,142,745,777 (  40.99%)   164,170,979,198 (  43.82%)
stalled-cycles-backend       70,813,866,387 (  18.71%)    66,456,858,165 (  17.74%)
instructions                463,436,648,173 (    1.22)   464,221,890,191 (    1.24)
branches                     91,088,733,902 ( 760.088)    91,278,144,546 ( 769.133)
branch-misses                   504,460,363 (   0.55%)       394,033,842 (   0.43%)
jobs3                              perfstat
stalled-cycles-frontend     201,300,397,212 (  39.84%)   223,969,902,257 (  44.44%)
stalled-cycles-backend       87,712,593,974 (  17.36%)    81,618,888,712 (  16.19%)
instructions                642,869,545,023 (    1.27)   644,677,354,132 (    1.28)
branches                    125,724,560,594 ( 690.682)   126,133,159,521 ( 694.542)
branch-misses                   527,941,798 (   0.42%)       444,782,220 (   0.35%)
jobs4                              perfstat
stalled-cycles-frontend     246,701,197,429 (  38.12%)   280,076,030,886 (  43.29%)
stalled-cycles-backend      119,050,341,112 (  18.40%)   110,955,641,671 (  17.15%)
instructions                822,716,962,127 (    1.27)   825,536,969,320 (    1.28)
branches                    160,590,028,545 ( 688.614)   161,152,996,915 ( 691.068)
branch-misses                   650,295,287 (   0.40%)       550,229,113 (   0.34%)
jobs5                              perfstat
stalled-cycles-frontend     298,958,462,516 (  38.30%)   344,852,200,358 (  44.16%)
stalled-cycles-backend      137,558,742,122 (  17.62%)   129,465,067,102 (  16.58%)
instructions              1,005,714,688,752 (    1.29) 1,007,657,999,432 (    1.29)
branches                    195,988,773,962 ( 697.730)   196,446,873,984 ( 700.319)
branch-misses                   695,818,940 (   0.36%)       624,823,263 (   0.32%)
jobs6                              perfstat
stalled-cycles-frontend     334,497,602,856 (  36.71%)   387,590,419,779 (  42.38%)
stalled-cycles-backend      163,539,365,335 (  17.95%)   152,640,193,639 (  16.69%)
instructions              1,184,738,177,851 (    1.30) 1,187,396,281,677 (    1.30)
branches                    230,592,915,640 ( 702.902)   231,253,802,882 ( 702.356)
branch-misses                   747,934,786 (   0.32%)       643,902,424 (   0.28%)
jobs7                              perfstat
stalled-cycles-frontend     396,724,684,187 (  37.71%)   460,705,858,952 (  43.84%)
stalled-cycles-backend      188,096,616,496 (  17.88%)   175,785,787,036 (  16.73%)
instructions              1,364,041,136,608 (    1.30) 1,366,689,075,112 (    1.30)
branches                    265,253,096,936 ( 700.078)   265,890,524,883 ( 702.839)
branch-misses                   784,991,589 (   0.30%)       729,196,689 (   0.27%)
jobs8                              perfstat
stalled-cycles-frontend     440,248,299,870 (  36.92%)   509,554,793,816 (  42.46%)
stalled-cycles-backend      222,575,930,616 (  18.67%)   213,401,248,432 (  17.78%)
instructions              1,542,262,045,114 (    1.29) 1,545,233,932,257 (    1.29)
branches                    299,775,178,439 ( 697.666)   300,528,458,505 ( 694.769)
branch-misses                   847,496,084 (   0.28%)       748,794,308 (   0.25%)
jobs9                              perfstat
stalled-cycles-frontend     506,269,882,480 (  37.86%)   592,798,032,820 (  44.43%)
stalled-cycles-backend      253,192,498,861 (  18.93%)   233,727,666,185 (  17.52%)
instructions              1,721,985,080,913 (    1.29) 1,724,666,236,005 (    1.29)
branches                    334,517,360,255 ( 694.134)   335,199,758,164 ( 697.131)
branch-misses                   873,496,730 (   0.26%)       815,379,236 (   0.24%)
jobs10                             perfstat
stalled-cycles-frontend     549,063,363,749 (  37.18%)   651,302,376,662 (  43.61%)
stalled-cycles-backend      281,680,986,810 (  19.07%)   277,005,235,582 (  18.55%)
instructions              1,901,859,271,180 (    1.29) 1,906,311,064,230 (    1.28)
branches                    369,398,536,153 ( 694.004)   370,527,696,358 ( 688.409)
branch-misses                   967,929,335 (   0.26%)       890,125,056 (   0.24%)

                            BASE           PATCHED
seconds elapsed        79.421641008 78.735285546
seconds elapsed        61.471246133 60.869085949
seconds elapsed        62.317058173 62.224188495
seconds elapsed        60.030739363 60.081102518
seconds elapsed        74.070398362 74.317582865
seconds elapsed        84.985953007 85.414364176
seconds elapsed        97.724553255 98.173311344
seconds elapsed        109.488066758 110.268399318
seconds elapsed        122.768189405 122.967164498
seconds elapsed        135.130035105 136.934770801

On my other system (8 x86_64 CPUs, short version of test results):

                            BASE           PATCHED
seconds elapsed        19.518065994 19.806320662
seconds elapsed        15.172772749 15.594718291
seconds elapsed        13.820925970 13.821708564
seconds elapsed        13.293097816 14.585206405
seconds elapsed        16.207284118 16.064431606
seconds elapsed        17.958376158 17.771825767
seconds elapsed        19.478009164 19.602961508
seconds elapsed        21.347152811 21.352318709
seconds elapsed        24.478121126 24.171088735
seconds elapsed        26.865057442 26.767327618

So performance-wise the numbers are quite similar.

Also update zcomp interface to be more aligned with the crypto API.

[1] http://marc.info/?l=linux-kernel&m=144480832108927&w=2
[2] http://marc.info/?l=linux-kernel&m=145379613507518&w=2
[3] https://github.com/sergey-senozhatsky/zram-perf-test

Link: http://lkml.kernel.org/r/20160531122017.2878-3-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Suggested-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: rename zstrm find-release functions

This has started as a 'add zlib support' work, but after some thinking I
saw no blockers for a bigger change -- a switch to crypto API.

We don't have an idle zstreams list anymore and our write path now works
absolutely differently, preventing preemption during compression.  This
removes possibilities of read paths preempting writes at wrong places
and opens the door for a move from custom LZO/LZ4 compression backends
implementation to a more generic one, using crypto compress API.

This patch set also eliminates the need of a new context-less crypto API
interface, which was quite hard to sell, so we can move along faster.

benchmarks:

(x86_64, 4GB, zram-perf script)

perf reported run-time fio (max jobs=3).  I performed fio test with the
increasing number of parallel jobs (max to 3) on a 3G zram device, using
`static' data and the following crypto comp algorithms:

842, deflate, lz4, lz4hc, lzo

the output was:

 - test running time (which can tell us what algorithms performs faster)

and

 - zram mm_stat (which tells the compressed memory size, max used memory, etc).

It's just for information.  for example, LZ4HC has twice the running
time of LZO, but the compressed memory size is: 23592960 vs 34603008
bytes.

  test-fio-zram-842
     197.907655282 seconds time elapsed
     201.623142884 seconds time elapsed
     226.854291345 seconds time elapsed
  test-fio-zram-DEFLATE
     253.259516155 seconds time elapsed
     258.148563401 seconds time elapsed
     290.251909365 seconds time elapsed
  test-fio-zram-LZ4
      27.022598717 seconds time elapsed
      29.580522717 seconds time elapsed
      33.293463430 seconds time elapsed
  test-fio-zram-LZ4HC
      56.393954615 seconds time elapsed
      74.904659747 seconds time elapsed
     101.940998564 seconds time elapsed
  test-fio-zram-LZO
      28.155948075 seconds time elapsed
      30.390036330 seconds time elapsed
      34.455773159 seconds time elapsed

zram mm_stat-s (max fio jobs=3)

  test-fio-zram-842
  mm_stat (jobs1): 3221225472 673185792 690266112        0 690266112        0        0
  mm_stat (jobs2): 3221225472 673185792 690266112        0 690266112        0        0
  mm_stat (jobs3): 3221225472 673185792 690266112        0 690266112        0        0
  test-fio-zram-DEFLATE
  mm_stat (jobs1): 3221225472  24379392  37761024        0  37761024        0        0
  mm_stat (jobs2): 3221225472  24379392  37761024        0  37761024        0        0
  mm_stat (jobs3): 3221225472  24379392  37761024        0  37761024        0        0
  test-fio-zram-LZ4
  mm_stat (jobs1): 3221225472  23592960  37761024        0  37761024        0        0
  mm_stat (jobs2): 3221225472  23592960  37761024        0  37761024        0        0
  mm_stat (jobs3): 3221225472  23592960  37761024        0  37761024        0        0
  test-fio-zram-LZ4HC
  mm_stat (jobs1): 3221225472  23592960  37761024        0  37761024        0        0
  mm_stat (jobs2): 3221225472  23592960  37761024        0  37761024        0        0
  mm_stat (jobs3): 3221225472  23592960  37761024        0  37761024        0        0
  test-fio-zram-LZO
  mm_stat (jobs1): 3221225472  34603008  50335744        0  50335744        0        0
  mm_stat (jobs2): 3221225472  34603008  50335744        0  50335744        0        0
  mm_stat (jobs3): 3221225472  34603008  50335744        0  50339840        0        0

This patch (of 8):

We don't perform any zstream idle list lookup anymore, so
zcomp_strm_find()/zcomp_strm_release() names are not representative.

Rename to zcomp_stream_get()/zcomp_stream_put().

Link: http://lkml.kernel.org/r/20160531122017.2878-2-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
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>

powerpc/mm: check for irq disabled() only if DEBUG_VM is enabled

We don't need to check this always.  The idea here is to capture the
wrong usage of find_linux_pte_or_hugepte and we can do that by
occasionally running with DEBUG_VM enabled.

Link: http://lkml.kernel.org/r/1464692688-6612-2-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/mmdebug.h: add VM_WARN which maps to WARN()

This enables us to do VM_WARN(condition, "warn message");

Link: http://lkml.kernel.org/r/1464692688-6612-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: oom: add memcg to oom_control

It's a part of oom context just like allocation order and nodemask, so
let's move it to oom_control instead of passing it in the argument list.

Link: http://lkml.kernel.org/r/40e03fd7aaf1f55c75d787128d6d17c5a71226c2.1464358556.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
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: zap ZONE_OOM_LOCKED

Not used since oom_lock was instroduced.

Link: http://lkml.kernel.org/r/1464358093-22663-1-git-send-email-vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: use zone_can_shift() for sysfs valid_zones attribute

Since zone_can_shift() is being used to validate the target zone during
onlining, it should also be used to determine the content of
valid_zones.

Link: http://lkml.kernel.org/r/1462816419-4479-4-git-send-email-arbab@linux.vnet.ibm.com
Signed-off-by: Reza Arbab <arbab@linux.vnet.ibm.com>
Reviewd-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrew Banman <abanman@sgi.com>
Cc: Chen Yucong <slaoub@gmail.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Zhang Zhen <zhenzhang.zhang@huawei.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: more general validation of zone during online

When memory is onlined, we are only able to rezone from ZONE_MOVABLE to
ZONE_KERNEL, or from (ZONE_MOVABLE - 1) to ZONE_MOVABLE.

To be more flexible, use the following criteria instead; to online
memory from zone X into zone Y,

* Any zones between X and Y must be unused.
* If X is lower than Y, the onlined memory must lie at the end of X.
* If X is higher than Y, the onlined memory must lie at the start of X.

Add zone_can_shift() to make this determination.

Link: http://lkml.kernel.org/r/1462816419-4479-3-git-send-email-arbab@linux.vnet.ibm.com
Signed-off-by: Reza Arbab <arbab@linux.vnet.ibm.com>
Reviewd-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrew Banman <abanman@sgi.com>
Cc: Chen Yucong <slaoub@gmail.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Zhang Zhen <zhenzhang.zhang@huawei.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: add move_pfn_range()

Add move_pfn_range(), a wrapper to call move_pfn_range_left() or
move_pfn_range_right().

No functional change. This will be utilized by a later patch.

Link: http://lkml.kernel.org/r/1462816419-4479-2-git-send-email-arbab@linux.vnet.ibm.com
Signed-off-by: Reza Arbab <arbab@linux.vnet.ibm.com>
Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrew Banman <abanman@sgi.com>
Cc: Chen Yucong <slaoub@gmail.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Zhang Zhen <zhenzhang.zhang@huawei.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/init: fix zone boundary creation

As a part of memory initialisation the architecture passes an array to
free_area_init_nodes() which specifies the max PFN of each memory zone.
This array is not necessarily monotonic (due to unused zones) so this
array is parsed to build monotonic lists of the min and max PFN for each
zone.  ZONE_MOVABLE is special cased here as its limits are managed by
the mm subsystem rather than the architecture.  Unfortunately, this
special casing is broken when ZONE_MOVABLE is the not the last zone in
the zone list.  The core of the issue is:

if (i == ZONE_MOVABLE)
continue;
arch_zone_lowest_possible_pfn[i] =
arch_zone_highest_possible_pfn[i-1];

As ZONE_MOVABLE is skipped the lowest_possible_pfn of the next zone will
be set to zero.  This patch fixes this bug by adding explicitly tracking
where the next zone should start rather than relying on the contents
arch_zone_highest_possible_pfn[].

Thie is low priority.  To get bitten by this you need to enable a zone
that appears after ZONE_MOVABLE in the zone_type enum.  As far as I can
tell this means running a kernel with ZONE_DEVICE or ZONE_CMA enabled,
so I can't see this affecting too many people.

I only noticed this because I've been fiddling with ZONE_DEVICE on
powerpc and 4.6 broke my test kernel.  This bug, in conjunction with the
changes in Taku Izumi's kernelcore=mirror patch (d91749c1dda71) and
powerpc being the odd architecture which initialises max_zone_pfn[] to
~0ul instead of 0 caused all of system memory to be placed into
ZONE_DEVICE at boot, followed a panic since device memory cannot be used
for kernel allocations.  I've already submitted a patch to fix the
powerpc specific bits, but I figured this should be fixed too.

Link: http://lkml.kernel.org/r/1462435033-15601-1-git-send-email-oohall@gmail.com
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.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/memcontrol.c: remove the useless parameter for mc_handle_swap_pte

It seems like this parameter has never been used since being introduced
by 90254a65833b ("memcg: clean up move charge").  Not a big deal because
I assume the function would get inlined into the caller anyway but why
not get rid of it.

[mhocko@suse.com: wrote changelog]
Link: http://lkml.kernel.org/r/20160525151831.GJ20132@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/1464145026-26693-1-git-send-email-roy.qing.li@gmail.com
Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/slab: use list_move instead of list_del/list_add

Using list_move() instead of list_del() + list_add() to avoid needlessly
poisoning the next and prev values.

Link: http://lkml.kernel.org/r/1468929772-9174-1-git-send-email-weiyj_lk@163.com
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
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: faster kmalloc_array(), kcalloc()

When both arguments to kmalloc_array() or kcalloc() are known at compile
time then their product is known at compile time but search for kmalloc
cache happens at runtime not at compile time.

Link: http://lkml.kernel.org/r/20160627213454.GA2440@p183.telecom.by
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slab: do not panic on invalid gfp_mask

Both SLAB and SLUB BUG() when a caller provides an invalid gfp_mask.
This is a rather harsh way to announce a non-critical issue.  Allocator
is free to ignore invalid flags.  Let's simply replace BUG() by
dump_stack to tell the offender and fixup the mask to move on with the
allocation request.

This is an example for kmalloc(GFP_KERNEL|__GFP_HIGHMEM) from a test
module:

  Unexpected gfp: 0x2 (__GFP_HIGHMEM). Fixing up to gfp: 0x24000c0 (GFP_KERNEL). Fix your code!
  CPU: 0 PID: 2916 Comm: insmod Tainted: G           O    4.6.0-slabgfp2-00002-g4cdfc2ef4892-dirty #936
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
  Call Trace:
    dump_stack+0x67/0x90
    cache_alloc_refill+0x201/0x617
    kmem_cache_alloc_trace+0xa7/0x24a
    ? 0xffffffffa0005000
    mymodule_init+0x20/0x1000 [test_slab]
    do_one_initcall+0xe7/0x16c
    ? rcu_read_lock_sched_held+0x61/0x69
    ? kmem_cache_alloc_trace+0x197/0x24a
    do_init_module+0x5f/0x1d9
    load_module+0x1a3d/0x1f21
    ? retint_kernel+0x2d/0x2d
    SyS_init_module+0xe8/0x10e
    ? SyS_init_module+0xe8/0x10e
    do_syscall_64+0x68/0x13f
    entry_SYSCALL64_slow_path+0x25/0x25

Link: http://lkml.kernel.org/r/1465548200-11384-2-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slab: make GFP_SLAB_BUG_MASK information more human readable

printk offers %pGg for quite some time so let's use it to get a human
readable list of invalid flags.

The original output would be
  [  429.191962] gfp: 2

after the change
  [  429.191962] Unexpected gfp: 0x2 (__GFP_HIGHMEM)

Link: http://lkml.kernel.org/r/1465548200-11384-1-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: SLUB freelist randomization

Implements freelist randomization for the SLUB allocator.  It was
previous implemented for the SLAB allocator.  Both use the same
configuration option (CONFIG_SLAB_FREELIST_RANDOM).

The list is randomized during initialization of a new set of pages.  The
order on different freelist sizes is pre-computed at boot for
performance.  Each kmem_cache has its own randomized freelist.

This security feature reduces the predictability of the kernel SLUB
allocator against heap overflows rendering attacks much less stable.

For example these attacks exploit the predictability of the heap:
 - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU)
 - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95)

Performance results:

slab_test impact is between 3% to 4% on average for 100000 attempts
without smp.  It is a very focused testing, kernbench show the overall
impact on the system is way lower.

Before:

  Single thread testing
  =====================
  1. Kmalloc: Repeatedly allocate then free test
  100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles
  100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles
  100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles
  100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles
  100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles
  100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles
  100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles
  100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles
  100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles
  100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles
  2. Kmalloc: alloc/free test
  100000 times kmalloc(8)/kfree -> 70 cycles
  100000 times kmalloc(16)/kfree -> 70 cycles
  100000 times kmalloc(32)/kfree -> 70 cycles
  100000 times kmalloc(64)/kfree -> 70 cycles
  100000 times kmalloc(128)/kfree -> 70 cycles
  100000 times kmalloc(256)/kfree -> 69 cycles
  100000 times kmalloc(512)/kfree -> 70 cycles
  100000 times kmalloc(1024)/kfree -> 73 cycles
  100000 times kmalloc(2048)/kfree -> 72 cycles
  100000 times kmalloc(4096)/kfree -> 71 cycles

After:

  Single thread testing
  =====================
  1. Kmalloc: Repeatedly allocate then free test
  100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles
  100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles
  100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles
  100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles
  100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles
  100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles
  100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles
  100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles
  100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles
  100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles
  2. Kmalloc: alloc/free test
  100000 times kmalloc(8)/kfree -> 66 cycles
  100000 times kmalloc(16)/kfree -> 66 cycles
  100000 times kmalloc(32)/kfree -> 66 cycles
  100000 times kmalloc(64)/kfree -> 66 cycles
  100000 times kmalloc(128)/kfree -> 65 cycles
  100000 times kmalloc(256)/kfree -> 67 cycles
  100000 times kmalloc(512)/kfree -> 67 cycles
  100000 times kmalloc(1024)/kfree -> 64 cycles
  100000 times kmalloc(2048)/kfree -> 67 cycles
  100000 times kmalloc(4096)/kfree -> 67 cycles

Kernbench, before:

  Average Optimal load -j 12 Run (std deviation):
  Elapsed Time 101.873 (1.16069)
  User Time 1045.22 (1.60447)
  System Time 88.969 (0.559195)
  Percent CPU 1112.9 (13.8279)
  Context Switches 189140 (2282.15)
  Sleeps 99008.6 (768.091)

After:

  Average Optimal load -j 12 Run (std deviation):
  Elapsed Time 102.47 (0.562732)
  User Time 1045.3 (1.34263)
  System Time 88.311 (0.342554)
  Percent CPU 1105.8 (6.49444)
  Context Switches 189081 (2355.78)
  Sleeps 99231.5 (800.358)

Link: http://lkml.kernel.org/r/1464295031-26375-3-git-send-email-thgarnie@google.com
Signed-off-by: Thomas Garnier <thgarnie@google.com>
Reviewed-by: Kees Cook <keescook@chromium.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: reorganize SLAB freelist randomization

The kernel heap allocators are using a sequential freelist making their
allocation predictable.  This predictability makes kernel heap overflow
easier to exploit.  An attacker can careful prepare the kernel heap to
control the following chunk overflowed.

For example these attacks exploit the predictability of the heap:
 - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU)
 - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95)

***Problems that needed solving:
 - Randomize the Freelist (singled linked) used in the SLUB allocator.
 - Ensure good performance to encourage usage.
 - Get best entropy in early boot stage.

***Parts:
 - 01/02 Reorganize the SLAB Freelist randomization to share elements
   with the SLUB implementation.
 - 02/02 The SLUB Freelist randomization implementation. Similar approach
   than the SLAB but tailored to the singled freelist used in SLUB.

***Performance data:

slab_test impact is between 3% to 4% on average for 100000 attempts
without smp.  It is a very focused testing, kernbench show the overall
impact on the system is way lower.

Before:

  Single thread testing
  =====================
  1. Kmalloc: Repeatedly allocate then free test
  100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles
  100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles
  100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles
  100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles
  100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles
  100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles
  100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles
  100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles
  100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles
  100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles
  2. Kmalloc: alloc/free test
  100000 times kmalloc(8)/kfree -> 70 cycles
  100000 times kmalloc(16)/kfree -> 70 cycles
  100000 times kmalloc(32)/kfree -> 70 cycles
  100000 times kmalloc(64)/kfree -> 70 cycles
  100000 times kmalloc(128)/kfree -> 70 cycles
  100000 times kmalloc(256)/kfree -> 69 cycles
  100000 times kmalloc(512)/kfree -> 70 cycles
  100000 times kmalloc(1024)/kfree -> 73 cycles
  100000 times kmalloc(2048)/kfree -> 72 cycles
  100000 times kmalloc(4096)/kfree -> 71 cycles

After:

  Single thread testing
  =====================
  1. Kmalloc: Repeatedly allocate then free test
  100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles
  100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles
  100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles
  100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles
  100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles
  100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles
  100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles
  100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles
  100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles
  100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles
  2. Kmalloc: alloc/free test
  100000 times kmalloc(8)/kfree -> 66 cycles
  100000 times kmalloc(16)/kfree -> 66 cycles
  100000 times kmalloc(32)/kfree -> 66 cycles
  100000 times kmalloc(64)/kfree -> 66 cycles
  100000 times kmalloc(128)/kfree -> 65 cycles
  100000 times kmalloc(256)/kfree -> 67 cycles
  100000 times kmalloc(512)/kfree -> 67 cycles
  100000 times kmalloc(1024)/kfree -> 64 cycles
  100000 times kmalloc(2048)/kfree -> 67 cycles
  100000 times kmalloc(4096)/kfree -> 67 cycles

Kernbench, before:

  Average Optimal load -j 12 Run (std deviation):
  Elapsed Time 101.873 (1.16069)
  User Time 1045.22 (1.60447)
  System Time 88.969 (0.559195)
  Percent CPU 1112.9 (13.8279)
  Context Switches 189140 (2282.15)
  Sleeps 99008.6 (768.091)

After:

  Average Optimal load -j 12 Run (std deviation):
  Elapsed Time 102.47 (0.562732)
  User Time 1045.3 (1.34263)
  System Time 88.311 (0.342554)
  Percent CPU 1105.8 (6.49444)
  Context Switches 189081 (2355.78)
  Sleeps 99231.5 (800.358)

This patch (of 2):

This commit reorganizes the previous SLAB freelist randomization to
prepare for the SLUB implementation.  It moves functions that will be
shared to slab_common.

The entropy functions are changed to align with the SLUB implementation,
now using get_random_(int|long) functions.  These functions were chosen
because they provide a bit more entropy early on boot and better
performance when specific arch instructions are not available.

[akpm@linux-foundation.org: fix build]
Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com
Signed-off-by: Thomas Garnier <thgarnie@google.com>
Reviewed-by: Kees Cook <keescook@chromium.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>

fs/fs-writeback.c: inode writeback list tracking tracepoints

The per-sb inode writeback list tracks inodes currently under writeback
to facilitate efficient sync processing.  In particular, it ensures that
sync only needs to walk through a list of inodes that were cleaned by
the sync.

Add a couple tracepoints to help identify when inodes are added/removed
to and from the writeback lists.  Piggyback off of the writeback
lazytime tracepoint template as it already tracks the relevant inode
information.

Link: http://lkml.kernel.org/r/1466594593-6757-3-git-send-email-bfoster@redhat.com
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <dchinner@redhat.com>
cc: Josef Bacik <jbacik@fb.com>
Cc: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/fs-writeback.c: add a new writeback list for sync

wait_sb_inodes() currently does a walk of all inodes in the filesystem
to find dirty one to wait on during sync.  This is highly inefficient
and wastes a lot of CPU when there are lots of clean cached inodes that
we don't need to wait on.

To avoid this "all inode" walk, we need to track inodes that are
currently under writeback that we need to wait for.  We do this by
adding inodes to a writeback list on the sb when the mapping is first
tagged as having pages under writeback.  wait_sb_inodes() can then walk
this list of "inodes under IO" and wait specifically just for the inodes
that the current sync(2) needs to wait for.

Define a couple helpers to add/remove an inode from the writeback list
and call them when the overall mapping is tagged for or cleared from
writeback.  Update wait_sb_inodes() to walk only the inodes under
writeback due to the sync.

With this change, filesystem sync times are significantly reduced for
fs' with largely populated inode caches and otherwise no other work to
do.  For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem
with a ~10m entry inode cache, sync times are reduced from ~7.3s to less
than 0.1s when the filesystem is fully clean.

Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>