hugetlb_cgroup: support noreserve mappings

Support MAP_NORESERVE accounting as part of the new counter.

For each hugepage allocation, at allocation time we check if there is a
reservation for this allocation or not.  If there is a reservation for
this allocation, then this allocation was charged at reservation time, and
we don't re-account it.  If there is no reserevation for this allocation,
we charge the appropriate hugetlb_cgroup.

The hugetlb_cgroup to uncharge for this allocation is stored in
page[3].private.  We use new APIs added in an earlier patch to set this
pointer.

Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Link: http://lkml.kernel.org/r/20200211213128.73302-6-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb_cgroup: add accounting for shared mappings

For shared mappings, the pointer to the hugetlb_cgroup to uncharge lives
in the resv_map entries, in file_region->reservation_counter.

After a call to region_chg, we charge the approprate hugetlb_cgroup, and
if successful, we pass on the hugetlb_cgroup info to a follow up
region_add call.  When a file_region entry is added to the resv_map via
region_add, we put the pointer to that cgroup in
file_region->reservation_counter.  If charging doesn't succeed, we report
the error to the caller, so that the kernel fails the reservation.

On region_del, which is when the hugetlb memory is unreserved, we also
uncharge the file_region->reservation_counter.

[akpm@linux-foundation.org: forward declare struct file_region]
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Link: http://lkml.kernel.org/r/20200211213128.73302-5-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: disable region_add file_region coalescing

A follow up patch in this series adds hugetlb cgroup uncharge info the
file_region entries in resv->regions.  The cgroup uncharge info may differ
for different regions, so they can no longer be coalesced at region_add
time.  So, disable region coalescing in region_add in this patch.

Behavior change:

Say a resv_map exists like this [0->1], [2->3], and [5->6].

Then a region_chg/add call comes in region_chg/add(f=0, t=5).

Old code would generate resv->regions: [0->5], [5->6].
New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5],
[5->6].

Special care needs to be taken to handle the resv->adds_in_progress
variable correctly.  In the past, only 1 region would be added for every
region_chg and region_add call.  But now, each call may add multiple
regions, so we can no longer increment adds_in_progress by 1 in
region_chg, or decrement adds_in_progress by 1 after region_add or
region_abort.  Instead, region_chg calls add_reservation_in_range() to
count the number of regions needed and allocates those, and that info is
passed to region_add and region_abort to decrement adds_in_progress
correctly.

We've also modified the assumption that region_add after region_chg never
fails.  region_chg now pre-allocates at least 1 region for region_add.  If
region_add needs more regions than region_chg has allocated for it, then
it may fail.

[almasrymina@google.com: fix file_region entry allocations]
Link: http://lkml.kernel.org/r/20200219012736.20363-1-almasrymina@google.com
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Greg Thelen <gthelen@google.com>
Cc: Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Link: http://lkml.kernel.org/r/20200211213128.73302-4-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb_cgroup: add reservation accounting for private mappings

Normally the pointer to the cgroup to uncharge hangs off the struct page,
and gets queried when it's time to free the page.  With hugetlb_cgroup
reservations, this is not possible.  Because it's possible for a page to
be reserved by one task and actually faulted in by another task.

The best place to put the hugetlb_cgroup pointer to uncharge for
reservations is in the resv_map.  But, because the resv_map has different
semantics for private and shared mappings, the code patch to
charge/uncharge shared and private mappings is different.  This patch
implements charging and uncharging for private mappings.

For private mappings, the counter to uncharge is in
resv_map->reservation_counter.  On initializing the resv_map this is set
to NULL.  On reservation of a region in private mapping, the tasks
hugetlb_cgroup is charged and the hugetlb_cgroup is placed is
resv_map->reservation_counter.

On hugetlb_vm_op_close, we uncharge resv_map->reservation_counter.

[akpm@linux-foundation.org: forward declare struct resv_map]
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Link: http://lkml.kernel.org/r/20200211213128.73302-3-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb_cgroup: fix hugetlb_cgroup migration

Commit c32300516047 ("hugetlb_cgroup: add interface for charge/uncharge
hugetlb reservations") mistakingly doesn't handle the migration of *both*
the reservation hugetlb_cgroup and the fault hugetlb_cgroup correctly.

What should happen is that both cgroups shuold be queried from the old
page, then both set to NULL on the old page, then both inserted into the
new page.

The mistake also creates the following warning:

mm/hugetlb_cgroup.c: In function 'hugetlb_cgroup_migrate':
mm/hugetlb_cgroup.c:777:25: warning: variable 'h_cg' set but not used
[-Wunused-but-set-variable]
  struct hugetlb_cgroup *h_cg;
                         ^~~~

Solution is to add the missing steps, namly setting the reservation
hugetlb_cgroup to NULL on the old page, and setting the fault
hugetlb_cgroup on the new page.

Fixes: c32300516047 ("hugetlb_cgroup: add interface for charge/uncharge hugetlb reservations")
Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Link: http://lkml.kernel.org/r/20200218194727.46995-1-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb_cgroup: add interface for charge/uncharge hugetlb reservations

Augments hugetlb_cgroup_charge_cgroup to be able to charge hugetlb usage
or hugetlb reservation counter.

Adds a new interface to uncharge a hugetlb_cgroup counter via
hugetlb_cgroup_uncharge_counter.

Integrates the counter with hugetlb_cgroup, via hugetlb_cgroup_init,
hugetlb_cgroup_have_usage, and hugetlb_cgroup_css_offline.

Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Link: http://lkml.kernel.org/r/20200211213128.73302-2-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb_cgroup: add hugetlb_cgroup reservation counter

These counters will track hugetlb reservations rather than hugetlb memory
faulted in.  This patch only adds the counter, following patches add the
charging and uncharging of the counter.

This is patch 1 of an 9 patch series.

Problem:

Currently tasks attempting to reserve more hugetlb memory than is
available get a failure at mmap/shmget time.  This is thanks to Hugetlbfs
Reservations [1].  However, if a task attempts to reserve more hugetlb
memory than its hugetlb_cgroup limit allows, the kernel will allow the
mmap/shmget call, but will SIGBUS the task when it attempts to fault in
the excess memory.

We have users hitting their hugetlb_cgroup limits and thus we've been
looking at this failure mode.  We'd like to improve this behavior such
that users violating the hugetlb_cgroup limits get an error on mmap/shmget
time, rather than getting SIGBUS'd when they try to fault the excess
memory in.  This gives the user an opportunity to fallback more gracefully
to non-hugetlbfs memory for example.

The underlying problem is that today's hugetlb_cgroup accounting happens
at hugetlb memory *fault* time, rather than at *reservation* time.  Thus,
enforcing the hugetlb_cgroup limit only happens at fault time, and the
offending task gets SIGBUS'd.

Proposed Solution:

A new page counter named
'hugetlb.xMB.rsvd.[limit|usage|max_usage]_in_bytes'. This counter has
slightly different semantics than
'hugetlb.xMB.[limit|usage|max_usage]_in_bytes':

- While usage_in_bytes tracks all *faulted* hugetlb memory,
  rsvd.usage_in_bytes tracks all *reserved* hugetlb memory and hugetlb
  memory faulted in without a prior reservation.

- If a task attempts to reserve more memory than limit_in_bytes allows,
  the kernel will allow it to do so.  But if a task attempts to reserve
  more memory than rsvd.limit_in_bytes, the kernel will fail this
  reservation.

This proposal is implemented in this patch series, with tests to verify
functionality and show the usage.

Alternatives considered:

1. A new cgroup, instead of only a new page_counter attached to the
   existing hugetlb_cgroup.  Adding a new cgroup seemed like a lot of code
   duplication with hugetlb_cgroup.  Keeping hugetlb related page counters
   under hugetlb_cgroup seemed cleaner as well.

2. Instead of adding a new counter, we considered adding a sysctl that
   modifies the behavior of hugetlb.xMB.[limit|usage]_in_bytes, to do
   accounting at reservation time rather than fault time.  Adding a new
   page_counter seems better as userspace could, if it wants, choose to
   enforce different cgroups differently: one via limit_in_bytes, and
   another via rsvd.limit_in_bytes.  This could be very useful if you're
   transitioning how hugetlb memory is partitioned on your system one
   cgroup at a time, for example.  Also, someone may find usage for both
   limit_in_bytes and rsvd.limit_in_bytes concurrently, and this approach
   gives them the option to do so.

Testing:
- Added tests passing.
- Used libhugetlbfs for regression testing.

[1]: https://www.kernel.org/doc/html/latest/vm/hugetlbfs_reserv.html

Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200211213128.73302-1-almasrymina@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: Use i_mmap_rwsem to address page fault/truncate race

hugetlbfs page faults can race with truncate and hole punch operations.
Current code in the page fault path attempts to handle this by 'backing
out' operations if we encounter the race.  One obvious omission in the
current code is removing a page newly added to the page cache.  This is
pretty straight forward to address, but there is a more subtle and
difficult issue of backing out hugetlb reservations.  To handle this
correctly, the 'reservation state' before page allocation needs to be
noted so that it can be properly backed out.  There are four distinct
possibilities for reservation state: shared/reserved, shared/no-resv,
private/reserved and private/no-resv.  Backing out a reservation may
require memory allocation which could fail so that needs to be taken
into account as well.

Instead of writing the required complicated code for this rare
occurrence, just eliminate the race.  i_mmap_rwsem is now held in read
mode for the duration of page fault processing.  Hold i_mmap_rwsem in
write mode when modifying i_size.  In this way, truncation can not
proceed when page faults are being processed.  In addition, i_size
will not change during fault processing so a single check can be made
to ensure faults are not beyond (proposed) end of file.  Faults can
still race with hole punch, but that race is handled by existing code
and the use of hugetlb_fault_mutex.

With this modification, checks for races with truncation in the page
fault path can be simplified and removed.  remove_inode_hugepages no
longer needs to take hugetlb_fault_mutex in the case of truncation.
Comments are expanded to explain reasoning behind locking.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Prakash Sangappa <prakash.sangappa@oracle.com>
Link: http://lkml.kernel.org/r/20200316205756.146666-3-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: use i_mmap_rwsem for more pmd sharing synchronization

Patch series "hugetlbfs: use i_mmap_rwsem for more synchronization", v2.

While discussing the issue with huge_pte_offset [1], I remembered that
there were more outstanding hugetlb races.  These issues are:

1) For shared pmds, huge PTE pointers returned by huge_pte_alloc can become
   invalid via a call to huge_pmd_unshare by another thread.
2) hugetlbfs page faults can race with truncation causing invalid global
   reserve counts and state.

A previous attempt was made to use i_mmap_rwsem in this manner as
described at [2].  However, those patches were reverted starting with [3]
due to locking issues.

To effectively use i_mmap_rwsem to address the above issues it needs to be
held (in read mode) during page fault processing.  However, during fault
processing we need to lock the page we will be adding.  Lock ordering
requires we take page lock before i_mmap_rwsem.  Waiting until after
taking the page lock is too late in the fault process for the
synchronization we want to do.

To address this lock ordering issue, the following patches change the lock
ordering for hugetlb pages.  This is not too invasive as hugetlbfs
processing is done separate from core mm in many places.  However, I don't
really like this idea.  Much ugliness is contained in the new routine
hugetlb_page_mapping_lock_write() of patch 1.

The only other way I can think of to address these issues is by catching
all the races.  After catching a race, cleanup, backout, retry ...  etc,
as needed.  This can get really ugly, especially for huge page
reservations.  At one time, I started writing some of the reservation
backout code for page faults and it got so ugly and complicated I went
down the path of adding synchronization to avoid the races.  Any other
suggestions would be welcome.

[1] https://lore.kernel.org/linux-mm/1582342427-230392-1-git-send-email-longpeng2@huawei.com/
[2] https://lore.kernel.org/linux-mm/20181222223013.22193-1-mike.kravetz@oracle.com/
[3] https://lore.kernel.org/linux-mm/20190103235452.29335-1-mike.kravetz@oracle.com
[4] https://lore.kernel.org/linux-mm/1584028670.7365.182.camel@lca.pw/
[5] https://lore.kernel.org/lkml/20200312183142.108df9ac@canb.auug.org.au/

This patch (of 2):

While looking at BUGs associated with invalid huge page map counts, it was
discovered and observed that a huge pte pointer could become 'invalid' and
point to another task's page table.  Consider the following:

A task takes a page fault on a shared hugetlbfs file and calls
huge_pte_alloc to get a ptep.  Suppose the returned ptep points to a
shared pmd.

Now, another task truncates the hugetlbfs file.  As part of truncation, it
unmaps everyone who has the file mapped.  If the range being truncated is
covered by a shared pmd, huge_pmd_unshare will be called.  For all but the
last user of the shared pmd, huge_pmd_unshare will clear the pud pointing
to the pmd.  If the task in the middle of the page fault is not the last
user, the ptep returned by huge_pte_alloc now points to another task's
page table or worse.  This leads to bad things such as incorrect page
map/reference counts or invalid memory references.

To fix, expand the use of i_mmap_rwsem as follows:
- i_mmap_rwsem is held in read mode whenever huge_pmd_share is called.
  huge_pmd_share is only called via huge_pte_alloc, so callers of
  huge_pte_alloc take i_mmap_rwsem before calling.  In addition, callers
  of huge_pte_alloc continue to hold the semaphore until finished with
  the ptep.
- i_mmap_rwsem is held in write mode whenever huge_pmd_unshare is called.

One problem with this scheme is that it requires taking i_mmap_rwsem
before taking the page lock during page faults.  This is not the order
specified in the rest of mm code.  Handling of hugetlbfs pages is mostly
isolated today.  Therefore, we use this alternative locking order for
PageHuge() pages.

         mapping->i_mmap_rwsem
           hugetlb_fault_mutex (hugetlbfs specific page fault mutex)
             page->flags PG_locked (lock_page)

To help with lock ordering issues, hugetlb_page_mapping_lock_write() is
introduced to write lock the i_mmap_rwsem associated with a page.

In most cases it is easy to get address_space via vma->vm_file->f_mapping.
However, in the case of migration or memory errors for anon pages we do
not have an associated vma.  A new routine _get_hugetlb_page_mapping()
will use anon_vma to get address_space in these cases.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Prakash Sangappa <prakash.sangappa@oracle.com>
Link: http://lkml.kernel.org/r/20200316205756.146666-2-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memblock.c: remove redundant assignment to variable max_addr

The variable max_addr is being initialized with a value that is never read
and it is being updated later with a new value.  The initialization is
redundant and can be removed.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200228235003.112718-1-colin.king@canonical.com
Addresses-Coverity: ("Unused value")
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mempolicy: require at least one nodeid for MPOL_PREFERRED

Using an empty (malformed) nodelist that is not caught during mount option
parsing leads to a stack-out-of-bounds access.

The option string that was used was: "mpol=prefer:,".  However,
MPOL_PREFERRED requires a single node number, which is not being provided
here.

Add a check that 'nodes' is not empty after parsing for MPOL_PREFERRED's
nodeid.

Fixes: 095f1fc4ebf3 ("mempolicy: rework shmem mpol parsing and display")
Reported-by: Entropy Moe <3ntr0py1337@gmail.com>
Reported-by: syzbot+b055b1a6b2b958707a21@syzkaller.appspotmail.com
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: syzbot+b055b1a6b2b958707a21@syzkaller.appspotmail.com
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Link: http://lkml.kernel.org/r/89526377-7eb6-b662-e1d8-4430928abde9@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mempolicy: use VM_BUG_ON_VMA in queue_pages_test_walk()

The VM_BUG_ON() is already used by queue_pages_test_walk(), it sounds
better to dump more debug information by using VM_BUG_ON_VMA() to help
debugging.

Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: "Li Xinhai" <lixinhai.lxh@gmail.com>
Cc: Qian Cai <cai@lca.pw>
Link: http://lkml.kernel.org/r/1579068565-110432-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mempolicy: check hugepage migration is supported by arch in vma_migratable()

vma_migratable() is called to check if pages in vma can be migrated before
go ahead to further actions.  Currently it is used in below code path:

- task_numa_work
- mbind
- move_pages

For hugetlb mapping, whether vma is migratable or not is determined by:
- CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
- arch_hugetlb_migration_supported

Issue: current code only checks for CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
alone, and no code should use it directly.  (note that current code in
vma_migratable don't cause failure or bug because
unmap_and_move_huge_page() will catch unsupported hugepage and handle it
properly)

This patch checks the two factors by hugepage_migration_supported for
impoving code logic and robustness.  It will enable early bail out of
hugepage migration procedure, but because currently all architecture
supporting hugepage migration is able to support all page size, we would
not see performance gain with this patch applied.

vma_migratable() is moved to mm/mempolicy.c, because of the circular
reference of mempolicy.h and hugetlb.h cause defining it as inline not
feasible.

Signed-off-by: Li Xinhai <lixinhai.lxh@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Link: http://lkml.kernel.org/r/1579786179-30633-1-git-send-email-lixinhai.lxh@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mempolicy: support MPOL_MF_STRICT for huge page mapping

MPOL_MF_STRICT is used in mbind() for purposes:

(1) MPOL_MF_STRICT is set alone without MPOL_MF_MOVE or
    MPOL_MF_MOVE_ALL, to check if there is misplaced page and return -EIO;

(2) MPOL_MF_STRICT is set with MPOL_MF_MOVE or MPOL_MF_MOVE_ALL, to
    check if there is misplaced page which is failed to isolate, or page
    is success on isolate but failed to move, and return -EIO.

For non hugepage mapping, (1) and (2) are implemented as expectation.  For
hugepage mapping, (1) is not implemented.  And in (2), the part about
failed to isolate and report -EIO is not implemented.

This patch implements the missed parts for hugepage mapping.  Benefits
with it applied:

- User space can apply same code logic to handle mbind() on hugepage and
  non hugepage mapping;

- Reliably using MPOL_MF_STRICT alone to check whether there is
  misplaced page or not when bind policy on address range, especially for
  address range which contains both hugepage and non hugepage mapping.

Analysis of potential impact to existing users:

- If MPOL_MF_STRICT alone was previously used, hugetlb pages not
  following the memory policy would not cause an EIO error.  After this
  change, hugetlb pages are treated like all other pages.  If
  MPOL_MF_STRICT alone is used and hugetlb pages do not follow memory
  policy an EIO error will be returned.

- For users who using MPOL_MF_STRICT with MPOL_MF_MOVE or
  MPOL_MF_MOVE_ALL, the semantic about some pages could not be moved will
  not be changed by this patch, because failed to isolate and failed to
  move have same effects to users, so their existing code will not be
  impacted.

In mbind man page, the note about 'MPOL_MF_STRICT is ignored on huge page
mappings' can be removed after this patch is applied.

Mike:

: The current behavior with MPOL_MF_STRICT and hugetlb pages is inconsistent
: and does not match documentation (as described above).  The special
: behavior for hugetlb pages ideally should have been removed when hugetlb
: page migration was introduced.  It is unlikely that anyone relies on
: today's inconsistent behavior, and removing one more case of special
: handling for hugetlb pages is a good thing.

Signed-off-by: Li Xinhai <lixinhai.lxh@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: linux-man <linux-man@vger.kernel.org>
Link: http://lkml.kernel.org/r/1581559627-6206-1-git-send-email-lixinhai.lxh@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction.c: clean code by removing unnecessary assignment

Previously 0 was assigned to variable 'last_migrated_pfn'.  But the
variable is not read after that, so the assignment can be removed.

Signed-off-by: Mateusz Nosek <mateusznosek0@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/20200318174509.15021-1-mateusznosek0@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction: Disable compact_unevictable_allowed on RT

Since commit 5bbe3547aa3ba ("mm: allow compaction of unevictable pages")
it is allowed to examine mlocked pages and compact them by default.  On
-RT even minor pagefaults are problematic because it may take a few 100us
to resolve them and until then the task is blocked.

Make compact_unevictable_allowed = 0 default and issue a warning on RT if
it is changed.

[bigeasy@linutronix.de: v5]
Link: https://lore.kernel.org/linux-mm/20190710144138.qyn4tuttdq6h7kqx@linutronix.de/
Link: http://lkml.kernel.org/r/20200319165536.ovi75tsr2seared4@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Iurii Zaikin <yzaikin@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: https://lore.kernel.org/linux-mm/20190710144138.qyn4tuttdq6h7kqx@linutronix.de/
Link: http://lkml.kernel.org/r/20200303202225.nhqc3v5gwlb7x6et@linutronix.de
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction: really limit compact_unevictable_allowed to 0 and 1

The proc file `compact_unevictable_allowed' should allow 0 and 1 only, the
`extra*' attribues have been set properly but without
proc_dointvec_minmax() as the `proc_handler' the limit will not be
enforced.

Use proc_dointvec_minmax() as the `proc_handler' to enfoce the valid
specified range.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Iurii Zaikin <yzaikin@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/20200303202054.gsosv7fsx2ma3cic@linutronix.de
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, compaction: fully assume capture is not NULL in compact_zone_order()

Dan reports:

The patch 5e1f0f098b46: "mm, compaction: capture a page under direct
compaction" from Mar 5, 2019, leads to the following Smatch complaint:

    mm/compaction.c:2321 compact_zone_order()
     error: we previously assumed 'capture' could be null (see line 2313)

mm/compaction.c
  2288  static enum compact_result compact_zone_order(struct zone *zone, int order,
  2289                  gfp_t gfp_mask, enum compact_priority prio,
  2290                  unsigned int alloc_flags, int classzone_idx,
  2291                  struct page **capture)
                                      ^^^^^^^

  2313 if (capture)
                    ^^^^^^^
Check for NULL

  2314 current->capture_control = &capc;
  2315
  2316 ret = compact_zone(&cc, &capc);
  2317
  2318 VM_BUG_ON(!list_empty(&cc.freepages));
  2319 VM_BUG_ON(!list_empty(&cc.migratepages));
  2320
  2321 *capture = capc.page;
                ^^^^^^^^
Unchecked dereference.

  2322 current->capture_control = NULL;
  2323

In practice this is not an issue, as the only caller path passes non-NULL
capture:

__alloc_pages_direct_compact()
  struct page *page = NULL;
  try_to_compact_pages(capture = &page);
    compact_zone_order(capture = capture);

So let's remove the unnecessary check, which should also make Smatch happy.

Fixes: 5e1f0f098b46 ("mm, compaction: capture a page under direct compaction")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/18b0df3c-0589-d96c-23fa-040798fee187@suse.cz
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,thp,compaction,cma: allow THP migration for CMA allocations

The code to implement THP migrations already exists, and the code for CMA
to clear out a region of memory already exists.

Only a few small tweaks are needed to allow CMA to move THP memory when
attempting an allocation from alloc_contig_range.

With these changes, migrating THPs from a CMA area works when allocating a
1GB hugepage from CMA memory.

[riel@surriel.com: fix hugetlbfs pages per Mike, cleanup per Vlastimil]
Link: http://lkml.kernel.org/r/20200228104700.0af2f18d@imladris.surriel.com
Signed-off-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Link: http://lkml.kernel.org/r/20200227213238.1298752-2-riel@surriel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,compaction,cma: add alloc_contig flag to compact_control

Patch series "fix THP migration for CMA allocations", v2.

Transparent huge pages are allocated with __GFP_MOVABLE, and can end up in
CMA memory blocks.  Transparent huge pages also have most of the
infrastructure in place to allow migration.

However, a few pieces were missing, causing THP migration to fail when
attempting to use CMA to allocate 1GB hugepages.

With these patches in place, THP migration from CMA blocks seems to work,
both for anonymous THPs and for tmpfs/shmem THPs.

This patch (of 2):

Add information to struct compact_control to indicate that the allocator
would really like to clear out this specific part of memory, used by for
example CMA.

Signed-off-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Link: http://lkml.kernel.org/r/20200227213238.1298752-1-riel@surriel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests: vm: drop dependencies on page flags from mlock2 tests

It was noticed that mlock2 tests are failing after 9c4e6b1a7027f ("mm,
mlock, vmscan: no more skipping pagevecs") because the patch has changed
the timing on when the page is added to the unevictable LRU list and thus
gains the unevictable page flag.

The test was just too dependent on the implementation details which were
true at the time when it was introduced.  Page flags and the timing when
they are set is something no userspace should ever depend on.  The test
should be testing only for the user observable contract of the tested
syscalls.  Those are defined pretty well for the mlock and there are other
means for testing them.  In fact this is already done and testing for page
flags can be safely dropped to achieve the aimed purpose.  Present bits
can be checked by /proc/<pid>/smaps RSS field and the locking state by
VmFlags although I would argue that Locked: field would be more
appropriate.

Drop all the page flag machinery and considerably simplify the test.  This
should be more robust for future kernel changes while checking the
promised contract is still valid.

Fixes: 9c4e6b1a7027f ("mm, mlock, vmscan: no more skipping pagevecs")
Reported-by: Rafael Aquini <aquini@redhat.com>
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Rafael Aquini <aquini@redhat.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Eric B Munson <emunson@akamai.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200324154218.GS19542@dhcp22.suse.cz
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: do_try_to_free_pages(): clean code by removing unnecessary assignment

sc->memcg_low_skipped resets skipped_deactivate to 0 but this is not
needed as this code path is never reachable with skipped_deactivate != 0
due to previous sc->skipped_deactivate branch.

[mhocko@kernel.org: rewrite changelog]
Signed-off-by: Mateusz Nosek <mateusznosek0@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Link: http://lkml.kernel.org/r/20200319165938.23354-1-mateusznosek0@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: make may_enter_fs bool in shrink_page_list()

This gives some size improvement:

$size mm/vmscan.o (before)
   text    data     bss     dec     hex filename
  53670   24123      12   77805   12fed mm/vmscan.o

$size mm/vmscan.o (after)
   text    data     bss     dec     hex filename
  53648   24123      12   77783   12fd7 mm/vmscan.o

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/Message-ID:
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: clean code by removing unnecessary assignment

Previously 0 was assigned to variable 'lruvec_size', but the variable was
never read later.  So the assignment can be removed.

Fixes: f87bccde6a7d ("mm/vmscan: remove unused lru_pages argument")
Signed-off-by: Mateusz Nosek <mateusznosek0@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200229214022.11853-1-mateusznosek0@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: fix data races using kswapd_classzone_idx

pgdat->kswapd_classzone_idx could be accessed concurrently in
wakeup_kswapd().  Plain writes and reads without any lock protection
result in data races.  Fix them by adding a pair of READ|WRITE_ONCE() as
well as saving a branch (compilers might well optimize the original code
in an unintentional way anyway).  While at it, also take care of
pgdat->kswapd_order and non-kswapd threads in allow_direct_reclaim().  The
data races were reported by KCSAN,

 BUG: KCSAN: data-race in wakeup_kswapd / wakeup_kswapd

 write to 0xffff9f427ffff2dc of 4 bytes by task 7454 on cpu 13:
  wakeup_kswapd+0xf1/0x400
  wakeup_kswapd at mm/vmscan.c:3967
  wake_all_kswapds+0x59/0xc0
  wake_all_kswapds at mm/page_alloc.c:4241
  __alloc_pages_slowpath+0xdcc/0x1290
  __alloc_pages_slowpath at mm/page_alloc.c:4512
  __alloc_pages_nodemask+0x3bb/0x450
  alloc_pages_vma+0x8a/0x2c0
  do_anonymous_page+0x16e/0x6f0
  __handle_mm_fault+0xcd5/0xd40
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40

 1 lock held by mtest01/7454:
  #0: ffff9f425afe8808 (&mm->mmap_sem#2){++++}, at:
 do_page_fault+0x143/0x6f9
 do_user_addr_fault at arch/x86/mm/fault.c:1405
 (inlined by) do_page_fault at arch/x86/mm/fault.c:1539
 irq event stamp: 6944085
 count_memcg_event_mm+0x1a6/0x270
 count_memcg_event_mm+0x119/0x270
 __do_softirq+0x34c/0x57c
 irq_exit+0xa2/0xc0

 read to 0xffff9f427ffff2dc of 4 bytes by task 7472 on cpu 38:
  wakeup_kswapd+0xc8/0x400
  wake_all_kswapds+0x59/0xc0
  __alloc_pages_slowpath+0xdcc/0x1290
  __alloc_pages_nodemask+0x3bb/0x450
  alloc_pages_vma+0x8a/0x2c0
  do_anonymous_page+0x16e/0x6f0
  __handle_mm_fault+0xcd5/0xd40
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40

 1 lock held by mtest01/7472:
  #0: ffff9f425a9ac148 (&mm->mmap_sem#2){++++}, at:
 do_page_fault+0x143/0x6f9
 irq event stamp: 6793561
 count_memcg_event_mm+0x1a6/0x270
 count_memcg_event_mm+0x119/0x270
 __do_softirq+0x34c/0x57c
 irq_exit+0xa2/0xc0

 BUG: KCSAN: data-race in kswapd / wakeup_kswapd

 write to 0xffff90973ffff2dc of 4 bytes by task 820 on cpu 6:
  kswapd+0x27c/0x8d0
  kthread+0x1e0/0x200
  ret_from_fork+0x27/0x50

 read to 0xffff90973ffff2dc of 4 bytes by task 6299 on cpu 0:
  wakeup_kswapd+0xf3/0x450
  wake_all_kswapds+0x59/0xc0
  __alloc_pages_slowpath+0xdcc/0x1290
  __alloc_pages_nodemask+0x3bb/0x450
  alloc_pages_vma+0x8a/0x2c0
  do_anonymous_page+0x170/0x700
  __handle_mm_fault+0xc9f/0xd00
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40

Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Marco Elver <elver@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/1582749472-5171-1-git-send-email-cai@lca.pw
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: remove cpu online notification for now

kswapd kernel thread starts either with a CPU affinity set to the full cpu
mask of its target node or without any affinity at all if the node is
CPUless.  There is a cpu hotplug callback (kswapd_cpu_online) that
implements an elaborate way to update this mask when a cpu is onlined.

It is not really clear whether there is any actual benefit from this
scheme. Completely CPU-less NUMA nodes rarely gain a new CPU during
runtime. Drop the code for that reason. If there is a real usecase then
we can resurrect and simplify the code.

[mhocko@suse.com rewrite changelog]

Suggested-by: Michal Hocko <mhocko@suse.org>
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/20200218224422.3407-1-richardw.yang@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: replace open codings to NUMA_NO_NODE

The commit 98fa15f34cb3 ("mm: replace all open encodings for
NUMA_NO_NODE") did the replacement across the kernel tree, but we got
some more in vmscan.c since then.

Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/1581568298-45317-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmpressure: use mem_cgroup_is_root API

Use mem_cgroup_is_root() API to check if memcg is root memcg instead of
open coding.

Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/1581398649-125989-2-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmpressure: don't need call kfree if kstrndup fails

When kstrndup fails, no memory was allocated and we can exit directly.

[david@redhat.com: reword changelog]
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/1581398649-125989-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: simplify page_is_buddy() for better code readability

Simplify page_is_buddy() to reduce the redundant code for better code
readability.

Signed-off-by: chenqiwu <chenqiwu@xiaomi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Link: http://lkml.kernel.org/r/1583853751-5525-1-git-send-email-qiwuchen55@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: micro-optimisation Remove unnecessary branch

Previously if branch condition was false, the assignment was not executed.
The assignment can be safely executed even when the condition is false
and it is not incorrect as it assigns the value of 'nodemask' to
'ac.nodemask' which already has the same value.

So as the assignment can be executed unconditionally, the branch can be
removed.

Signed-off-by: Mateusz Nosek <mateusznosek0@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200307225335.31300-1-mateusznosek0@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: use free_area_empty() instead of open-coding

Use free_area_empty() API to replace list_empty() for better code
readability.

Signed-off-by: chenqiwu <chenqiwu@xiaomi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: http://lkml.kernel.org/r/1583674354-7713-1-git-send-email-qiwuchen55@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, pagealloc: micro-optimisation: save two branches on hot page allocation path

This patch makes ALLOC_KSWAPD equal to __GFP_KSWAPD_RECLAIM (cast to int).

Thanks to that code like:

    if (gfp_mask & __GFP_KSWAPD_RECLAIM)
    alloc_flags |= ALLOC_KSWAPD;

can be changed to:

    alloc_flags |= (__force int) (gfp_mask &__GFP_KSWAPD_RECLAIM);

Thanks to this one branch less is generated in the assembly.

In case of ALLOC_KSWAPD flag two branches are saved, first one in code
that always executes in the beginning of page allocation and the second
one in loop in page allocator slowpath.

Signed-off-by: Mateusz Nosek <mateusznosek0@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/20200304162118.14784-1-mateusznosek0@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: increase default min_free_kbytes bound

Currently, the vm.min_free_kbytes sysctl value is capped at a hardcoded
64M in init_per_zone_wmark_min (unless it is overridden by khugepaged
initialization).

This value has not been modified since 2005, and enterprise-grade systems
now frequently have hundreds of GB of RAM and multiple 10, 40, or even 100
GB NICs.  We have seen page allocation failures on heavily loaded systems
related to NIC drivers.  These issues were resolved by an increase to
vm.min_free_kbytes.

This patch increases the hardcoded value by a factor of 4 as a temporary
solution.

Further work to make the calculation of vm.min_free_kbytes more consistent
throughout the kernel would be desirable.

As an example of the inconsistency of the current method, this value is
recalculated by init_per_zone_wmark_min() in the case of memory hotplug
which will override the value set by set_recommended_min_free_kbytes()
called during khugepaged initialization even if khugepaged remains
enabled, however an on/off toggle of khugepaged will then recalculate and
set the value via set_recommended_min_free_kbytes().

Signed-off-by: Joel Savitz <jsavitz@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Rafael Aquini <aquini@redhat.com>
Link: http://lkml.kernel.org/r/20200220150103.5183-1-jsavitz@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: add test for invalid size in memmove

Test negative size in memmove in order to verify whether it correctly get
KASAN report.

Casting negative numbers to size_t would indeed turn up as a large size_t,
so it will have out-of-bounds bug and be detected by KASAN.

[walter-zh.wu@mediatek.com: fix -Wstringop-overflow warning]
Link: http://lkml.kernel.org/r/20200311134244.13016-1-walter-zh.wu@mediatek.com
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: kernel test robot <lkp@intel.com>
Link: http://lkml.kernel.org/r/20191112065313.7060-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: detect negative size in memory operation function

Patch series "fix the missing underflow in memory operation function", v4.

The patchset helps to produce a KASAN report when size is negative in
memory operation functions.  It is helpful for programmer to solve an
undefined behavior issue.  Patch 1 based on Dmitry's review and
suggestion, patch 2 is a test in order to verify the patch 1.

[1]https://bugzilla.kernel.org/show_bug.cgi?id=199341
[2]https://lore.kernel.org/linux-arm-kernel/20190927034338.15813-1-walter-zh.wu@mediatek.com/

This patch (of 2):

KASAN missed detecting size is a negative number in memset(), memcpy(),
and memmove(), it will cause out-of-bounds bug.  So needs to be detected
by KASAN.

If size is a negative number, then it has a reason to be defined as
out-of-bounds bug type.  Casting negative numbers to size_t would indeed
turn up as a large size_t and its value will be larger than ULONG_MAX/2,
so that this can qualify as out-of-bounds.

KASAN report is shown below:

 BUG: KASAN: out-of-bounds in kmalloc_memmove_invalid_size+0x70/0xa0
 Read of size 18446744073709551608 at addr ffffff8069660904 by task cat/72

 CPU: 2 PID: 72 Comm: cat Not tainted 5.4.0-rc1-next-20191004ajb-00001-gdb8af2f372b2-dirty #1
 Hardware name: linux,dummy-virt (DT)
 Call trace:
  dump_backtrace+0x0/0x288
  show_stack+0x14/0x20
  dump_stack+0x10c/0x164
  print_address_description.isra.9+0x68/0x378
  __kasan_report+0x164/0x1a0
  kasan_report+0xc/0x18
  check_memory_region+0x174/0x1d0
  memmove+0x34/0x88
  kmalloc_memmove_invalid_size+0x70/0xa0

[1] https://bugzilla.kernel.org/show_bug.cgi?id=199341

[cai@lca.pw: fix -Wdeclaration-after-statement warn]
Link: http://lkml.kernel.org/r/1583509030-27939-1-git-send-email-cai@lca.pw
[peterz@infradead.org: fix objtool warning]
Link: http://lkml.kernel.org/r/20200305095436.GV2596@hirez.programming.kicks-ass.net
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Link: http://lkml.kernel.org/r/20191112065302.7015-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse.c: allocate memmap preferring the given node

When allocating memmap for hot added memory with the classic sparse, the
specified 'nid' is ignored in populate_section_memmap().

While in allocating memmap for the classic sparse during boot, the node
given by 'nid' is preferred.  And VMEMMAP prefers the node of 'nid' in
both boot stage and memory hot adding.  So seems no reason to not respect
the node of 'nid' for the classic sparse when hot adding memory.

Use kvmalloc_node instead to use the passed in 'nid'.

Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Link: http://lkml.kernel.org/r/20200316125625.GH3486@MiWiFi-R3L-srv
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse.c: use kvmalloc/kvfree to alloc/free memmap for the classic sparse

This change makes populate_section_memmap()/depopulate_section_memmap
much simpler.

Suggested-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200316125450.GG3486@MiWiFi-R3L-srv
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: rename pfn_present() to pfn_in_present_section()

After introducing mem sub section concept, pfn_present() loses its literal
meaning, and will not be necessary a truth on partial populated mem
section.

Since all of the callers use it to judge an absent section, it is better
to rename pfn_present() as pfn_in_present_section().

Signed-off-by: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Leonardo Bras <leonardo@linux.ibm.com>
Cc: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Cc: Nathan Lynch <nathanl@linux.ibm.com>
Link: http://lkml.kernel.org/r/1581919110-29575-1-git-send-email-kernelfans@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparsemem: get address to page struct instead of address to pfn

memmap should be the address to page struct instead of address to pfn.

As mentioned by David, if system memory and devmem sit within a section,
the mismatch address would lead kdump to dump unexpected memory.

Since sub-section only works for SPARSEMEM_VMEMMAP, pfn_to_page() is valid
to get the page struct address at this point.

Fixes: ba72b4c8cf60 ("mm/sparsemem: support sub-section hotplug")
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Baoquan He <bhe@redhat.com>
Link: http://lkml.kernel.org/r/20200210005048.10437-1-richardw.yang@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests: add MREMAP_DONTUNMAP selftest

Add a few simple self tests for the new flag MREMAP_DONTUNMAP, they are
simple smoke tests which also demonstrate the behavior.

[akpm@linux-foundation.org: convert eight-spaces to hard tabs]
[bgeffon@google.com: v7]
Link: http://lkml.kernel.org/r/20200221174248.244748-2-bgeffon@google.com
[akpm@linux-foundation.org: coding style fixes]
Signed-off-by: Brian Geffon <bgeffon@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: "Michael S . Tsirkin" <mst@redhat.com>
Cc: Brian Geffon <bgeffon@google.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Will Deacon <will@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Sonny Rao <sonnyrao@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Jesse Barnes <jsbarnes@google.com>
Cc: Nathan Chancellor <natechancellor@gmail.com>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Lokesh Gidra <lokeshgidra@google.com>
Link: http://lkml.kernel.org/r/20200218173221.237674-2-bgeffon@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mremap: add MREMAP_DONTUNMAP to mremap()

When remapping an anonymous, private mapping, if MREMAP_DONTUNMAP is set,
the source mapping will not be removed.  The remap operation will be
performed as it would have been normally by moving over the page tables to
the new mapping.  The old vma will have any locked flags cleared, have no
pagetables, and any userfaultfds that were watching that range will
continue watching it.

For a mapping that is shared or not anonymous, MREMAP_DONTUNMAP will cause
the mremap() call to fail.  Because MREMAP_DONTUNMAP always results in
moving a VMA you MUST use the MREMAP_MAYMOVE flag, it's not possible to
resize a VMA while also moving with MREMAP_DONTUNMAP so old_len must
always be equal to the new_len otherwise it will return -EINVAL.

We hope to use this in Chrome OS where with userfaultfd we could write an
anonymous mapping to disk without having to STOP the process or worry
about VMA permission changes.

This feature also has a use case in Android, Lokesh Gidra has said that
"As part of using userfaultfd for GC, We'll have to move the physical
pages of the java heap to a separate location.  For this purpose mremap
will be used.  Without the MREMAP_DONTUNMAP flag, when I mremap the java
heap, its virtual mapping will be removed as well.  Therefore, we'll
require performing mmap immediately after.  This is not only time
consuming but also opens a time window where a native thread may call mmap
and reserve the java heap's address range for its own usage.  This flag
solves the problem."

[bgeffon@google.com: v6]
Link: http://lkml.kernel.org/r/20200218173221.237674-1-bgeffon@google.com
[bgeffon@google.com: v7]
Link: http://lkml.kernel.org/r/20200221174248.244748-1-bgeffon@google.com
Signed-off-by: Brian Geffon <bgeffon@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Lokesh Gidra <lokeshgidra@google.com>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: "Michael S . Tsirkin" <mst@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Will Deacon <will@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Sonny Rao <sonnyrao@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Jesse Barnes <jsbarnes@google.com>
Cc: Nathan Chancellor <natechancellor@gmail.com>
Cc: Florian Weimer <fweimer@redhat.com>
Link: http://lkml.kernel.org/r/20200207201856.46070-1-bgeffon@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mmap: add trace point of vm_unmapped_area

Even on 64 bit kernel, the mmap failure can happen for a 32 bit task.
Virtual memory space shortage of a task on mmap is reported to userspace
as -ENOMEM.  It can be confused as physical memory shortage of overall
system.

The vm_unmapped_area can be called to by some drivers or other kernel core
system like filesystem.  In my platform, GPU driver calls to
vm_unmapped_area and the driver returns -ENOMEM even in GPU side shortage.
It can be hard to distinguish which code layer returns the -ENOMEM.

Create mmap trace file and add trace point of vm_unmapped_area.

i.e.)
277.156599: vm_unmapped_area: addr=77e0d03000 err=0 total_vm=0x17014b flags=0x1 len=0x400000 lo=0x8000 hi=0x7878c27000 mask=0x0 ofs=0x1
342.838740: vm_unmapped_area: addr=0 err=-12 total_vm=0xffb08 flags=0x0 len=0x100000 lo=0x40000000 hi=0xfffff000 mask=0x0 ofs=0x22

[akpm@linux-foundation.org: prefix address printk with 0x, per Matthew]
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@suse.de>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200320055823.27089-3-jaewon31.kim@samsung.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmap: remove inline of vm_unmapped_area

Patch series "mm: mmap: add mmap trace point", v3.

Create mmap trace file and add trace point of vm_unmapped_area().

This patch (of 2):

In preparation for next patch remove inline of vm_unmapped_area and move
code to mmap.c.  There is no logical change.

Also remove unmapped_area[_topdown] out of mm.h, there is no code
calling to them.

Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/20200320055823.27089-2-jaewon31.kim@samsung.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory.c: clarify a confusing comment for vm_iomap_memory

The param "start" actually referes to the physical memory start, which is
to be mapped into virtual area vma.  And it is the field vma->vm_start
which stands for the start of the area.

Most of the time, we do not read through whole implementation of a
function but only the definition and essential comments.  Accurate
comments are definitely the base stone.

Signed-off-by: Wang Wenhu <wenhu.wang@vivo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200318052206.105104-1-wenhu.wang@vivo.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: clarify a confusing comment for remap_pfn_range()

It really made me scratch my head.  Replace the comment with an accurate
and consistent description.

The parameter pfn actually refers to the page frame number which is
right-shifted by PAGE_SHIFT from the physical address.

Signed-off-by: WANG Wenhu <wenhu.wang@vivo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200310073955.43415-1-wenhu.wang@vivo.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/userfaultfd: honor FAULT_FLAG_KILLABLE in fault path

Userfaultfd fault path was by default killable even if the caller does not
have FAULT_FLAG_KILLABLE.  That makes sense before in that when with gup
we don't have FAULT_FLAG_KILLABLE properly set before.  Now after previous
patch we've got FAULT_FLAG_KILLABLE applied even for gup code so it should
also make sense to let userfaultfd to honor the FAULT_FLAG_KILLABLE.

Because we're unconditionally setting FAULT_FLAG_KILLABLE in gup code
right now, this patch should have no functional change.  It also cleaned
the code a little bit by introducing some helpers.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160300.9941-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: allow to react to fatal signals

The existing gup code does not react to the fatal signals in many code
paths.  For example, in one retry path of gup we're still using
down_read() rather than down_read_killable().  Also, when doing page
faults we don't pass in FAULT_FLAG_KILLABLE as well, which means that
within the faulting process we'll wait in non-killable way as well.  These
were spotted by Linus during the code review of some other patches.

Let's allow the gup code to react to fatal signals to improve the
responsiveness of threads when during gup and being killed.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160256.9887-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: allow VM_FAULT_RETRY for multiple times

This is the gup counterpart of the change that allows the VM_FAULT_RETRY
to happen for more than once.  One thing to mention is that we must check
the fatal signal here before retry because the GUP can be interrupted by
that, otherwise we can loop forever.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220195357.16371-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: allow VM_FAULT_RETRY for multiple times

The idea comes from a discussion between Linus and Andrea [1].

Before this patch we only allow a page fault to retry once.  We achieved
this by clearing the FAULT_FLAG_ALLOW_RETRY flag when doing
handle_mm_fault() the second time.  This was majorly used to avoid
unexpected starvation of the system by looping over forever to handle the
page fault on a single page.  However that should hardly happen, and after
all for each code path to return a VM_FAULT_RETRY we'll first wait for a
condition (during which time we should possibly yield the cpu) to happen
before VM_FAULT_RETRY is really returned.

This patch removes the restriction by keeping the FAULT_FLAG_ALLOW_RETRY
flag when we receive VM_FAULT_RETRY.  It means that the page fault handler
now can retry the page fault for multiple times if necessary without the
need to generate another page fault event.  Meanwhile we still keep the
FAULT_FLAG_TRIED flag so page fault handler can still identify whether a
page fault is the first attempt or not.

Then we'll have these combinations of fault flags (only considering
ALLOW_RETRY flag and TRIED flag):

  - ALLOW_RETRY and !TRIED:  this means the page fault allows to
                             retry, and this is the first try

  - ALLOW_RETRY and TRIED:   this means the page fault allows to
                             retry, and this is not the first try

  - !ALLOW_RETRY and !TRIED: this means the page fault does not allow
                             to retry at all

  - !ALLOW_RETRY and TRIED:  this is forbidden and should never be used

In existing code we have multiple places that has taken special care of
the first condition above by checking against (fault_flags &
FAULT_FLAG_ALLOW_RETRY).  This patch introduces a simple helper to detect
the first retry of a page fault by checking against both (fault_flags &
FAULT_FLAG_ALLOW_RETRY) and !(fault_flag & FAULT_FLAG_TRIED) because now
even the 2nd try will have the ALLOW_RETRY set, then use that helper in
all existing special paths.  One example is in __lock_page_or_retry(), now
we'll drop the mmap_sem only in the first attempt of page fault and we'll
keep it in follow up retries, so old locking behavior will be retained.

This will be a nice enhancement for current code [2] at the same time a
supporting material for the future userfaultfd-writeprotect work, since in
that work there will always be an explicit userfault writeprotect retry
for protected pages, and if that cannot resolve the page fault (e.g., when
userfaultfd-writeprotect is used in conjunction with swapped pages) then
we'll possibly need a 3rd retry of the page fault.  It might also benefit
other potential users who will have similar requirement like userfault
write-protection.

GUP code is not touched yet and will be covered in follow up patch.

Please read the thread below for more information.

[1] https://lore.kernel.org/lkml/20171102193644.GB22686@redhat.com/
[2] https://lore.kernel.org/lkml/20181230154648.GB9832@redhat.com/

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160246.9790-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce FAULT_FLAG_INTERRUPTIBLE

handle_userfaultfd() is currently the only one place in the kernel page
fault procedures that can respond to non-fatal userspace signals.  It was
trying to detect such an allowance by checking against USER & KILLABLE
flags, which was "un-official".

In this patch, we introduced a new flag (FAULT_FLAG_INTERRUPTIBLE) to show
that the fault handler allows the fault procedure to respond even to
non-fatal signals.  Meanwhile, add this new flag to the default fault
flags so that all the page fault handlers can benefit from the new flag.
With that, replacing the userfault check to this one.

Since the line is getting even longer, clean up the fault flags a bit too
to ease TTY users.

Although we've got a new flag and applied it, we shouldn't have any
functional change with this patch so far.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220195348.16302-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce FAULT_FLAG_DEFAULT

Although there're tons of arch-specific page fault handlers, most of them
are still sharing the same initial value of the page fault flags.  Say,
merely all of the page fault handlers would allow the fault to be retried,
and they also allow the fault to respond to SIGKILL.

Let's define a default value for the fault flags to replace those initial
page fault flags that were copied over.  With this, it'll be far easier to
introduce new fault flag that can be used by all the architectures instead
of touching all the archs.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160238.9694-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd: don't retake mmap_sem to emulate NOPAGE

This patch removes the risk path in handle_userfault() then we will be
sure that the callers of handle_mm_fault() will know that the VMAs might
have changed.  Meanwhile with previous patch we don't lose responsiveness
as well since the core mm code now can handle the nonfatal userspace
signals even if we return VM_FAULT_RETRY.

Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160234.9646-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: return faster for non-fatal signals in user mode faults

The idea comes from the upstream discussion between Linus and Andrea:

  https://lore.kernel.org/lkml/20171102193644.GB22686@redhat.com/

A summary to the issue: there was a special path in handle_userfault() in
the past that we'll return a VM_FAULT_NOPAGE when we detected non-fatal
signals when waiting for userfault handling.  We did that by reacquiring
the mmap_sem before returning.  However that brings a risk in that the
vmas might have changed when we retake the mmap_sem and even we could be
holding an invalid vma structure.

This patch is a preparation of removing that special path by allowing the
page fault to return even faster if we were interrupted by a non-fatal
signal during a user-mode page fault handling routine.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160230.9598-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh/mm: use helper fault_signal_pending()

Let SH to use the new fault_signal_pending() helper.  Here we'll need to
move the up_read() out because that's actually needed as long as !RETRY
cases.  At the meantime we can drop all the rest of up_read()s now (which
seems to be cleaner).

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160226.9550-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc/mm: use helper fault_signal_pending()

Let powerpc code to use the new helper, by moving the signal handling
earlier before the retry logic.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220160222.9422-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arm64/mm: use helper fault_signal_pending()

Let the arm64 fault handling to use the new fault_signal_pending() helper,
by moving the signal handling out of the retry logic.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155927.9264-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arc/mm: use helper fault_signal_pending()

Let ARC to use the new helper fault_signal_pending() by moving the signal
check out of the retry logic as standalone.  This should also helps to
simplify the code a bit.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155843.9172-1-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86/mm: use helper fault_signal_pending()

Let's move the fatal signal check even earlier so that we can directly use
the new fault_signal_pending() in x86 mm code.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155353.8676-5-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce fault_signal_pending()

For most architectures, we've got a quick path to detect fatal signal
after a handle_mm_fault().  Introduce a helper for that quick path.

It cleans the current codes a bit so we don't need to duplicate the same
check across archs.  More importantly, this will be an unified place that
we handle the signal immediately right after an interrupted page fault, so
it'll be much easier for us if we want to change the behavior of handling
signals later on for all the archs.

Note that currently only part of the archs are using this new helper,
because some archs have their own way to handle signals.  In the follow up
patches, we'll try to apply this helper to all the rest of archs.

Another note is that the "regs" parameter in the new helper is not used
yet.  It'll be used very soon.  Now we kept it in this patch only to avoid
touching all the archs again in the follow up patches.

[peterx@redhat.com: fix sparse warnings]
Link: http://lkml.kernel.org/r/20200311145921.GD479302@xz-x1
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155353.8676-4-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: fix __get_user_pages() on fault retry of hugetlb

When follow_hugetlb_page() returns with *locked==0, it means we've got a
VM_FAULT_RETRY within the fauling process and we've released the mmap_sem.
When that happens, we should stop and bail out.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155353.8676-3-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: rename "nonblocking" to "locked" where proper

Patch series "mm: Page fault enhancements", v6.

This series contains cleanups and enhancements to current page fault
logic.  The whole idea comes from the discussion between Andrea and Linus
on the bug reported by syzbot here:

  https://lkml.org/lkml/2017/11/2/833

Basically it does two things:

  (a) Allows the page fault logic to be more interactive on not only
      SIGKILL, but also the rest of userspace signals, and,

  (b) Allows the page fault retry (VM_FAULT_RETRY) to happen for more
      than once.

For (a): with the changes we should be able to react faster when page
faults are working in parallel with userspace signals like SIGSTOP and
SIGCONT (and more), and with that we can remove the buggy part in
userfaultfd and benefit the whole page fault mechanism on faster signal
processing to reach the userspace.

For (b), we should be able to allow the page fault handler to loop for
even more than twice.  Some context: for now since we have
FAULT_FLAG_ALLOW_RETRY we can allow to retry the page fault once with the
same interrupt context, however never more than twice.  This can be not
only a potential cleanup to remove this assumption since AFAIU the code
itself doesn't really have this twice-only limitation (though that should
be a protective approach in the past), at the same time it'll greatly
simplify future works like userfaultfd write-protect where it's possible
to retry for more than twice (please have a look at [1] below for a
possible user that might require the page fault to be handled for a third
time; if we can remove the retry limitation we can simply drop that patch
and those complexity).

This patch (of 16):

There's plenty of places around __get_user_pages() that has a parameter
"nonblocking" which does not really mean that "it won't block" (because it
can really block) but instead it shows whether the mmap_sem is released by
up_read() during the page fault handling mostly when VM_FAULT_RETRY is
returned.

We have the correct naming in e.g.  get_user_pages_locked() or
get_user_pages_remote() as "locked", however there're still many places
that are using the "nonblocking" as name.

Renaming the places to "locked" where proper to better suite the
functionality of the variable.  While at it, fixing up some of the
comments accordingly.

Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Brian Geffon <bgeffon@google.com>
Reviewed-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Martin Cracauer <cracauer@cons.org>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com>
Cc: Bobby Powers <bobbypowers@gmail.com>
Cc: Maya Gokhale <gokhale2@llnl.gov>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Marty McFadden <mcfadden8@llnl.gov>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Denis Plotnikov <dplotnikov@virtuozzo.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Link: http://lkml.kernel.org/r/20200220155353.8676-2-peterx@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add pagemap.h to the fine documentation

The documentation currently does not include the deathless prose written
to describe functions in pagemap.h because it's not included in any rst
file.  Fix up the mismatches between parameter names and the documentation
and add the file to mm-api.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Link: http://lkml.kernel.org/r/20200221220045.24989-1-willy@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vma: make is_vma_temporary_stack() available for general use

Currently the declaration and definition for is_vma_temporary_stack() are
scattered.  Lets make is_vma_temporary_stack() helper available for
general use and also drop the declaration from (include/linux/huge_mm.h)
which is no longer required.  While at this, rename this as
vma_is_temporary_stack() in line with existing helpers.  This should not
cause any functional change.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1582782965-3274-4-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vma: make vma_is_foreign() available for general use

Idea of a foreign VMA with respect to the present context is very generic.
But currently there are two identical definitions for this in powerpc and
x86 platforms.  Lets consolidate those redundant definitions while making
vma_is_foreign() available for general use later.  This should not cause
any functional change.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Link: http://lkml.kernel.org/r/1582782965-3274-3-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vma: move VM_NO_KHUGEPAGED into generic header

Patch series "mm/vma: some more minor changes", v2.

The motivation here is to consolidate VMA flags and helpers in generic
memory header and reduce code duplication when ever applicable.  If there
are other possible similar instances which might be missing here, please
do let me me know.  I will be happy to incorporate them.

This patch (of 3):

Move VM_NO_KHUGEPAGED into generic header (include/linux/mm.h).  This just
makes sure that no VMA flag is scattered in individual function files any
longer.  While at this, fix an old comment which is no longer valid.  This
should not cause any functional change.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1582782965-3274-2-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mapping_dirty_helpers: update huge page-table entry callbacks

Following the update of pagewalk code commit a07984d48146 ("mm: pagewalk:
add p4d_entry() and pgd_entry()") we can modify the mapping_dirty_helpers'
huge page-table entry callbacks to avoid splitting when a huge pud or -pmd
is encountered.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Steven Price <steven.price@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200203154305.15045-1-thomas_os@shipmail.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg: make memory.oom.group tolerable to task migration

If a task is getting moved out of the OOMing cgroup, it might result in
unexpected OOM killings if memory.oom.group is used anywhere in the cgroup
tree.

Imagine the following example:

          A (oom.group = 1)
         / \
  (OOM) B   C

Let's say B's memory.max is exceeded and it's OOMing.  The OOM killer
selects a task in B as a victim, but someone asynchronously moves the task
into C.  mem_cgroup_get_oom_group() will iterate over all ancestors of C
up to the root cgroup.  In theory it had to stop at the oom_domain level -
the memory cgroup which is OOMing.  But because B is not an ancestor of C,
it's not happening.  Instead it chooses A (because it's oom.group is set),
and kills all tasks in A.  This behavior is wrong because the OOM happened
in B, so there is no reason to kill anything outside.

Fix this by checking it the memory cgroup to which the task belongs is a
descendant of the oom_domain.  If not, memory.oom.group should be ignored,
and the OOM killer should kill only the victim task.

Reported-by: Dan Schatzberg <dschatzberg@fb.com>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: http://lkml.kernel.org/r/20200316223510.3176148-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent mem_cgroup_protected store tearing

The read side of this is all protected, but we can still tear if multiple
iterations of mem_cgroup_protected are going at the same time.

There's some intentional racing in mem_cgroup_protected which is ok, but
load/store tearing should be avoided.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/d1e9fbc0379fe8db475d82c8b6fbe048876e12ae.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent memory.swap.max load tearing

The write side of this is xchg()/smp_mb(), so that's all good.  Just a few
sites missing a READ_ONCE.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/bbec2c3d822217334855c8877a9d28b2a6d395fb.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent memory.min load/store tearing

This can be set concurrently with reads, which may cause the wrong value
to be propagated.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/e809b4e6b0c1626dac6945970de06409a180ee65.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent memory.low load/store tearing

This can be set concurrently with reads, which may cause the wrong value
to be propagated.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/448206f44b0fa7be9dad2ca2601d2bcb2c0b7844.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent memory.max load tearing

This one is a bit more nuanced because we have memcg_max_mutex, which is
mostly just used for enforcing invariants, but we still need to READ_ONCE
since (despite its name) it doesn't really protect memory.max access.

On write (page_counter_set_max() and memory_max_write()) we use xchg(),
which uses smp_mb(), so that's already fine.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/50a31e5f39f8ae6c8fb73966ba1455f0924e8f44.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: prevent memory.high load/store tearing

A mem_cgroup's high attribute can be concurrently set at the same time as
we are trying to read it -- for example, if we are in memory_high_write at
the same time as we are trying to do high reclaim.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/2f66f7038ed1d4688e59de72b627ae0ea52efa83.1584034301.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c: make mem_cgroup_id_get_many() __maybe_unused

mem_cgroup_id_get_many() is currently used only when MMU or MEMCG_SWAP
configuration options are enabled.  Having them disabled triggers the
following warning at compile time:

  linux/mm/memcontrol.c:4797:13: warning: `mem_cgroup_id_get_many' defined but not used [-Wunused-function]
   static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)

Make mem_cgroup_id_get_many() __maybe_unused to address the issue.

Signed-off-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Chris Down <chris@chrisdown.name>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200305164354.48147-1-vincenzo.frascino@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: css_tryget_online cleanups

Currently multiple locations in memcg code, css_tryget_online() is being
used. However it doesn't matter whether the cgroup is online for the
callers. Online used to matter when we had reparenting on offlining and
we needed a way to prevent new ones from showing up.

The failure case for couple of these css_tryget_online usage is to
fallback to root_mem_cgroup which kind of make bypassing the memcg
limits possible for some workloads. For example creating an inotify
group in a subcontainer and then deleting that container after moving the
process to a different container will make all the event objects
allocated for that group to the root_mem_cgroup. So, using
css_tryget_online() is dangerous for such cases.

Two locations still use the online version. The swapin of offlined
memcg's pages and the memcg kmem cache creation. The kmem cache indeed
needs the online version as the kernel does the reparenting of memcg
kmem caches. For the swapin case, it has been left for later as the
fallback is not really that concerning.

With swap accounting enabled, if the memcg of the swapped out page is
not online then the memcg extracted from the given 'mm' will be charged
and if 'mm' is NULL then root memcg will be charged.  However I could
not find a code path where the given 'mm' will be NULL for swap-in
case.

Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/20200302203109.179417-1-shakeelb@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: recursive memory.low protection

Right now, the effective protection of any given cgroup is capped by its
own explicit memory.low setting, regardless of what the parent says.  The
reasons for this are mostly historical and ease of implementation: to make
delegation of memory.low safe, effective protection is the min() of all
memory.low up the tree.

Unfortunately, this limitation makes it impossible to protect an entire
subtree from another without forcing the user to make explicit protection
allocations all the way to the leaf cgroups - something that is highly
undesirable in real life scenarios.

Consider memory in a data center host.  At the cgroup top level, we have a
distinction between system management software and the actual workload the
system is executing.  Both branches are further subdivided into individual
services, job components etc.

We want to protect the workload as a whole from the system management
software, but that doesn't mean we want to protect and prioritize
individual workload wrt each other.  Their memory demand can vary over
time, and we'd want the VM to simply cache the hottest data within the
workload subtree.  Yet, the current memory.low limitations force us to
allocate a fixed amount of protection to each workload component in order
to get protection from system management software in general.  This
results in very inefficient resource distribution.

Another concern with mandating downward allocation is that, as the
complexity of the cgroup tree grows, it gets harder for the lower levels
to be informed about decisions made at the host-level.  Consider a
container inside a namespace that in turn creates its own nested tree of
cgroups to run multiple workloads.  It'd be extremely difficult to
configure memory.low parameters in those leaf cgroups that on one hand
balance pressure among siblings as the container desires, while also
reflecting the host-level protection from e.g.  rpm upgrades, that lie
beyond one or more delegation and namespacing points in the tree.

It's highly unusual from a cgroup interface POV that nested levels have to
be aware of and reflect decisions made at higher levels for them to be
effective.

To enable such use cases and scale configurability for complex trees, this
patch implements a resource inheritance model for memory that is similar
to how the CPU and the IO controller implement work-conserving resource
allocations: a share of a resource allocated to a subree always applies to
the entire subtree recursively, while allowing, but not mandating,
children to further specify distribution rules.

That means that if protection is explicitly allocated among siblings,
those configured shares are being followed during page reclaim just like
they are now.  However, if the memory.low set at a higher level is not
fully claimed by the children in that subtree, the "floating" remainder is
applied to each cgroup in the tree in proportion to its size.  Since
reclaim pressure is applied in proportion to size as well, each child in
that tree gets the same boost, and the effect is neutral among siblings -
with respect to each other, they behave as if no memory control was
enabled at all, and the VM simply balances the memory demands optimally
within the subtree.  But collectively those cgroups enjoy a boost over the
cgroups in neighboring trees.

E.g.  a leaf cgroup with a memory.low setting of 0 no longer means that
it's not getting a share of the hierarchically assigned resource, just
that it doesn't claim a fixed amount of it to protect from its siblings.

This allows us to recursively protect one subtree (workload) from another
(system management), while letting subgroups compete freely among each
other - without having to assign fixed shares to each leaf, and without
nested groups having to echo higher-level settings.

The floating protection composes naturally with fixed protection.
Consider the following example tree:

A            A: low = 2G
               / \          A1: low = 1G
              A1 A2         A2: low = 0G

As outside pressure is applied to this tree, A1 will enjoy a fixed
protection from A2 of 1G, but the remaining, unclaimed 1G from A is split
evenly among A1 and A2, coming out to 1.5G and 0.5G.

There is a slight risk of regressing theoretical setups where the
top-level cgroups don't know about the true budgeting and set bogusly high
"bypass" values that are meaningfully allocated down the tree.  Such
setups would rely on unclaimed protection to be discarded, and
distributing it would change the intended behavior.  Be safe and hide the
new behavior behind a mount option, 'memory_recursiveprot'.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Chris Down <chris@chrisdown.name>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Michal Koutný <mkoutny@suse.com>
Link: http://lkml.kernel.org/r/20200227195606.46212-4-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: clean up and document effective low/min calculations

The effective protection of any given cgroup is a somewhat complicated
construct that depends on the ancestor's configuration, siblings'
configurations, as well as current memory utilization in all these groups.
It's done this way to satisfy hierarchical delegation requirements while
also making the configuration semantics flexible and expressive in complex
real life scenarios.

Unfortunately, all the rules and requirements are sparsely documented, and
the code is a little too clever in merging different scenarios into a
single min() expression.  This makes it hard to reason about the
implementation and avoid breaking semantics when making changes to it.

This patch documents each semantic rule individually and splits out the
handling of the overcommit case from the regular case.

Michal Koutný also points out that the points of equilibrium as described
in the existing example scenarios aren't actually accurate.  Delete these
examples for now to avoid confusion.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Michal Koutný <mkoutny@suse.com>
Link: http://lkml.kernel.org/r/20200227195606.46212-3-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: fix memory.low proportional distribution

Patch series "mm: memcontrol: recursive memory.low protection", v3.

The current memory.low (and memory.min) semantics require protection to be
assigned to a cgroup in an untinterrupted chain from the top-level cgroup
all the way to the leaf.

In practice, we want to protect entire cgroup subtrees from each other
(system management software vs.  workload), but we would like the VM to
balance memory optimally *within* each subtree, without having to make
explicit weight allocations among individual components.  The current
semantics make that impossible.

They also introduce unmanageable complexity into more advanced resource
trees.  For example:

          host root
          `- system.slice
             `- rpm upgrades
             `- logging
          `- workload.slice
             `- a container
                `- system.slice
                `- workload.slice
                   `- job A
                      `- component 1
                      `- component 2
                   `- job B

At a host-level perspective, we would like to protect the outer
workload.slice subtree as a whole from rpm upgrades, logging etc.  But for
that to be effective, right now we'd have to propagate it down through the
container, the inner workload.slice, into the job cgroup and ultimately
the component cgroups where memory is actually, physically allocated.
This may cross several tree delegation points and namespace boundaries,
which make such a setup near impossible.

CPU and IO on the other hand are already distributed recursively.  The
user would simply configure allowances at the host level, and they would
apply to the entire subtree without any downward propagation.

To enable the above-mentioned usecases and bring memory in line with other
resource controllers, this patch series extends memory.low/min such that
settings apply recursively to the entire subtree.  Users can still assign
explicit shares in subgroups, but if they don't, any ancestral protection
will be distributed such that children compete freely amongst each other -
as if no memory control were enabled inside the subtree - but enjoy
protection from neighboring trees.

In the above example, the user would then be able to configure shares of
CPU, IO and memory at the host level to comprehensively protect and
isolate the workload.slice as a whole from system.slice activity.

Patch #1 fixes an existing bug that can give a cgroup tree more protection
than it should receive as per ancestor configuration.

Patch #2 simplifies and documents the existing code to make it easier to
reason about the changes in the next patch.

Patch #3 finally implements recursive memory protection semantics.

Because of a risk of regressing legacy setups, the new semantics are
hidden behind a cgroup2 mount option, 'memory_recursiveprot'.

More details in patch #3.

This patch (of 3):

When memory.low is overcommitted - i.e.  the children claim more
protection than their shared ancestor grants them - the allowance is
distributed in proportion to how much each sibling uses their own declared
protection:

low_usage = min(memory.low, memory.current)
elow = parent_elow * (low_usage / siblings_low_usage)

However, siblings_low_usage is not the sum of all low_usages. It sums
up the usages of *only those cgroups that are within their memory.low*
That means that low_usage can be *bigger* than siblings_low_usage, and
consequently the total protection afforded to the children can be
bigger than what the ancestor grants the subtree.

Consider three groups where two are in excess of their protection:

  A/memory.low = 10G
  A/A1/memory.low = 10G, memory.current = 20G
  A/A2/memory.low = 10G, memory.current = 20G
  A/A3/memory.low = 10G, memory.current =  8G
  siblings_low_usage = 8G (only A3 contributes)

  A1/elow = parent_elow(10G) * low_usage(10G) / siblings_low_usage(8G) = 12.5G -> 10G
  A2/elow = parent_elow(10G) * low_usage(10G) / siblings_low_usage(8G) = 12.5G -> 10G
  A3/elow = parent_elow(10G) * low_usage(8G) / siblings_low_usage(8G) = 10.0G

  (the 12.5G are capped to the explicit memory.low setting of 10G)

With that, the sum of all awarded protection below A is 30G, when A
only grants 10G for the entire subtree.

What does this mean in practice? A1 and A2 would still be in excess of
their 10G allowance and would be reclaimed, whereas A3 would not. As
they eventually drop below their protection setting, they would be
counted in siblings_low_usage again and the error would right itself.

When reclaim was applied in a binary fashion (cgroup is reclaimed when
it's above its protection, otherwise it's skipped) this would actually
work out just fine. However, since 1bc63fb1272b ("mm, memcg: make scan
aggression always exclude protection"), reclaim pressure is scaled to
how much a cgroup is above its protection. As a result this
calculation error unduly skews pressure away from A1 and A2 toward the
rest of the system.

But why did we do it like this in the first place?

The reasoning behind exempting groups in excess from
siblings_low_usage was to go after them first during reclaim in an
overcommitted subtree:

  A/memory.low = 2G, memory.current = 4G
  A/A1/memory.low = 3G, memory.current = 2G
  A/A2/memory.low = 1G, memory.current = 2G

  siblings_low_usage = 2G (only A1 contributes)
  A1/elow = parent_elow(2G) * low_usage(2G) / siblings_low_usage(2G) = 2G
  A2/elow = parent_elow(2G) * low_usage(1G) / siblings_low_usage(2G) = 1G

While the children combined are overcomitting A and are technically
both at fault, A2 is actively declaring unprotected memory and we
would like to reclaim that first.

However, while this sounds like a noble goal on the face of it, it
doesn't make much difference in actual memory distribution: Because A
is overcommitted, reclaim will not stop once A2 gets pushed back to
within its allowance; we'll have to reclaim A1 either way. The end
result is still that protection is distributed proportionally, with A1
getting 3/4 (1.5G) and A2 getting 1/4 (0.5G) of A's allowance.

[ If A weren't overcommitted, it wouldn't make a difference since each
  cgroup would just get the protection it declares:

  A/memory.low = 2G, memory.current = 3G
  A/A1/memory.low = 1G, memory.current = 1G
  A/A2/memory.low = 1G, memory.current = 2G

  With the current calculation:

  siblings_low_usage = 1G (only A1 contributes)
  A1/elow = parent_elow(2G) * low_usage(1G) / siblings_low_usage(1G) = 2G -> 1G
  A2/elow = parent_elow(2G) * low_usage(1G) / siblings_low_usage(1G) = 2G -> 1G

  Including excess groups in siblings_low_usage:

  siblings_low_usage = 2G
  A1/elow = parent_elow(2G) * low_usage(1G) / siblings_low_usage(2G) = 1G -> 1G
  A2/elow = parent_elow(2G) * low_usage(1G) / siblings_low_usage(2G) = 1G -> 1G ]

Simplify the calculation and fix the proportional reclaim bug by
including excess cgroups in siblings_low_usage.

After this patch, the effective memory.low distribution from the
example above would be as follows:

  A/memory.low = 10G
  A/A1/memory.low = 10G, memory.current = 20G
  A/A2/memory.low = 10G, memory.current = 20G
  A/A3/memory.low = 10G, memory.current =  8G
  siblings_low_usage = 28G

  A1/elow = parent_elow(10G) * low_usage(10G) / siblings_low_usage(28G) = 3.5G
  A2/elow = parent_elow(10G) * low_usage(10G) / siblings_low_usage(28G) = 3.5G
  A3/elow = parent_elow(10G) * low_usage(8G) / siblings_low_usage(28G) = 2.8G

Fixes: 1bc63fb1272b ("mm, memcg: make scan aggression always exclude protection")
Fixes: 230671533d64 ("mm: memory.low hierarchical behavior")
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Michal Koutný <mkoutny@suse.com>
Link: http://lkml.kernel.org/r/20200227195606.46212-2-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: rename (__)memcg_kmem_(un)charge_memcg() to __memcg_kmem_(un)charge()

Drop the _memcg suffix from (__)memcg_kmem_(un)charge functions.  It's
shorter and more obvious.

These are the most basic functions which are just (un)charging the given
cgroup with the given amount of pages.

Also fix up the corresponding comments.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-7-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: cache page number in memcg_(un)charge_slab()

There are many places in memcg_charge_slab() and memcg_uncharge_slab()
which are calculating the number of pages to charge, css references to
grab etc depending on the order of the slab page.

Let's simplify the code by calculating it once and caching in the local
variable.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-6-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: switch to nr_pages in (__)memcg_kmem_charge_memcg()

These functions are charging the given number of kernel pages to the given
memory cgroup.  The number doesn't have to be a power of two.  Let's make
them to take the unsigned int nr_pages as an argument instead of the page
order.

It makes them look consistent with the corresponding uncharge functions
and functions like: mem_cgroup_charge_skmem(memcg, nr_pages).

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-5-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: rename memcg_kmem_(un)charge() into memcg_kmem_(un)charge_page()

Rename (__)memcg_kmem_(un)charge() into (__)memcg_kmem_(un)charge_page()
to better reflect what they are actually doing:

1) call __memcg_kmem_(un)charge_memcg() to actually charge or uncharge
   the current memcg

2) set or clear the PageKmemcg flag

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-4-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: cleanup memcg_kmem_uncharge_memcg() arguments

Drop the unused page argument and put the memcg pointer at the first
place.  This make the function consistent with its peers:
__memcg_kmem_uncharge_memcg(), memcg_kmem_charge_memcg(), etc.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-3-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: cleanup (__)memcg_kmem_charge_memcg() arguments

Patch series "mm: memcg: kmem API cleanup", v2.

This patchset aims to clean up the kernel memory charging API.  It doesn't
bring any functional changes, just removes unused arguments, renames some
functions and fixes some comments.

Currently it's not obvious which functions are most basic
(memcg_kmem_(un)charge_memcg()) and which are based on them
(memcg_kmem_(un)charge()).  The patchset renames these functions and
removes unused arguments:

TL;DR:
was:
  memcg_kmem_charge_memcg(page, gfp, order, memcg)
  memcg_kmem_uncharge_memcg(memcg, nr_pages)
  memcg_kmem_charge(page, gfp, order)
  memcg_kmem_uncharge(page, order)

now:
  memcg_kmem_charge(memcg, gfp, nr_pages)
  memcg_kmem_uncharge(memcg, nr_pages)
  memcg_kmem_charge_page(page, gfp, order)
  memcg_kmem_uncharge_page(page, order)

This patch (of 6):

The first argument of memcg_kmem_charge_memcg() and
__memcg_kmem_charge_memcg() is the page pointer and it's not used.  Let's
drop it.

Memcg pointer is passed as the last argument.  Move it to the first place
for consistency with other memcg functions, e.g.
__memcg_kmem_uncharge_memcg() or try_charge().

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200109202659.752357-2-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: use mem_cgroup_from_obj()

Sometimes we need to get a memcg pointer from a charged kernel object.
The right way to get it depends on whether it's a proper slab object or
it's backed by raw pages (e.g.  it's a vmalloc alloction).  In the first
case the kmem_cache->memcg_params.memcg indirection should be used; in
other cases it's just page->mem_cgroup.

To simplify this task and hide the implementation details let's use the
mem_cgroup_from_obj() helper, which takes a pointer to any kernel object
and returns a valid memcg pointer or NULL.

Passing a kernel address rather than a pointer to a page will allow to use
this helper for per-object (rather than per-page) tracked objects in the
future.

The caller is still responsible to ensure that the returned memcg isn't
going away underneath: take the rcu read lock, cgroup mutex etc; depending
on the context.

mem_cgroup_from_kmem() defined in mm/list_lru.c is now obsolete and can be
removed.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200117203609.3146239-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c: allocate shrinker_map on appropriate NUMA node

The shrinker_map may be touched from any cpu (e.g., a bit there may be set
by a task running everywhere) but kswapd is always bound to specific node.
So allocate shrinker_map from the related NUMA node to respect its NUMA
locality.  Also, this follows generic way we use for allocation of memcg's
per-node data.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Roman Gushchin <guro@fb.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/fff0e636-4c36-ed10-281c-8cdb0687c839@virtuozzo.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: fix build error around the usage of kmem_caches

When I manually set default n to MEMCG_KMEM in init/Kconfig, bellow error
occurs,

  mm/slab_common.c: In function 'memcg_slab_start':
  mm/slab_common.c:1530:30: error: 'struct mem_cgroup' has no member named
  'kmem_caches'
    return seq_list_start(&memcg->kmem_caches, *pos);
                                ^
  mm/slab_common.c: In function 'memcg_slab_next':
  mm/slab_common.c:1537:32: error: 'struct mem_cgroup' has no member named
  'kmem_caches'
    return seq_list_next(p, &memcg->kmem_caches, pos);
                                  ^
  mm/slab_common.c: In function 'memcg_slab_show':
  mm/slab_common.c:1551:16: error: 'struct mem_cgroup' has no member named
  'kmem_caches'
    if (p == memcg->kmem_caches.next)
                  ^
    CC      arch/x86/xen/smp.o
  mm/slab_common.c: In function 'memcg_slab_start':
  mm/slab_common.c:1531:1: warning: control reaches end of non-void function
  [-Wreturn-type]
   }
   ^
  mm/slab_common.c: In function 'memcg_slab_next':
  mm/slab_common.c:1538:1: warning: control reaches end of non-void function
  [-Wreturn-type]
   }
   ^

That's because kmem_caches is defined only when CONFIG_MEMCG_KMEM is set,
while memcg_slab_start() will use it no matter CONFIG_MEMCG_KMEM is defined
or not.

By the way, the reason I mannuly undefined CONFIG_MEMCG_KMEM is to verify
whether my some other code change is still stable when CONFIG_MEMCG_KMEM is
not set. Unfortunately, the existing code has been already unstable since
v4.11.

Fixes: bc2791f857e1 ("slab: link memcg kmem_caches on their associated memory cgroup")
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Link: http://lkml.kernel.org/r/1580970260-2045-1-git-send-email-laoar.shao@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap_state.c: use the same way to count page in [add_to|delete_from]_swap_cache

add_to_swap_cache() and delete_from_swap_cache() are counterparts, while
currently they use different ways to count pages.

It doesn't break anything because we only have two sizes for PageAnon, but
this is confusing and not good practice.

This patch corrects it by making both functions use hpage_nr_pages().

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200315012920.2687-1-richard.weiyang@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: swap: use smp_mb__after_atomic() to order LRU bit set

Memory barrier is needed after setting LRU bit, but smp_mb() is too
strong.  Some architectures, i.e.  x86, imply memory barrier with atomic
operations, so replacing it with smp_mb__after_atomic() sounds better,
which is nop on strong ordered machines, and full memory barriers on
others.  With this change the vm-scalability cases would perform better on
x86, I saw total 6% improvement with this patch and previous inline fix.

The test data (lru-file-readtwice throughput) against v5.6-rc4:
mainline w/ inline fix w/ both (adding this)
150MB 154MB 159MB

Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs")
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: http://lkml.kernel.org/r/1584500541-46817-2-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: swap: make page_evictable() inline

When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping
pagevecs") to our 4.9 kernel, our test bench noticed around 10% down with
a couple of vm-scalability's test cases (lru-file-readonce,
lru-file-readtwice and lru-file-mmap-read).  I didn't see that much down
on my VM (32c-64g-2nodes).  It might be caused by the test configuration,
which is 32c-256g with NUMA disabled and the tests were run in root memcg,
so the tests actually stress only one inactive and active lru.  It sounds
not very usual in mordern production environment.

That commit did two major changes:
1. Call page_evictable()
2. Use smp_mb to force the PG_lru set visible

It looks they contribute the most overhead.  The page_evictable() is a
function which does function prologue and epilogue, and that was used by
page reclaim path only.  However, lru add is a very hot path, so it sounds
better to make it inline.  However, it calls page_mapping() which is not
inlined either, but the disassemble shows it doesn't do push and pop
operations and it sounds not very straightforward to inline it.

Other than this, it sounds smp_mb() is not necessary for x86 since
SetPageLRU is atomic which enforces memory barrier already, replace it
with smp_mb__after_atomic() in the following patch.

With the two fixes applied, the tests can get back around 5% on that test
bench and get back normal on my VM.  Since the test bench configuration is
not that usual and I also saw around 6% up on the latest upstream, so it
sounds good enough IMHO.

The below is test data (lru-file-readtwice throughput) against the v5.6-rc4:
mainline w/ inline fix
          150MB            154MB

With this patch the throughput gets 2.67% up.  The data with using
smp_mb__after_atomic() is showed in the following patch.

Shakeel Butt did the below test:

On a real machine with limiting the 'dd' on a single node and reading 100
GiB sparse file (less than a single node).  Just ran a single instance to
not cause the lru lock contention.  The cmdline used is "dd if=file-100GiB
of=/dev/null bs=4k".  Ran the cmd 10 times with drop_caches in between and
measured the time it took.

Without patch: 56.64143 +- 0.672 sec

With patches: 56.10 +- 0.21 sec

[akpm@linux-foundation.org: move page_evictable() to internal.h]
Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs")
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: http://lkml.kernel.org/r/1584500541-46817-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap_slots.c: assign|reset cache slot by value directly

Currently we use a tmp pointer, pentry, to transfer and reset swap cache
slot, which is a little redundant.  Swap cache slot stores the entry value
directly, assign and reset it by value would be straight forward.

Also this patch merges the else and if, since this is the only case we
refill and repeat swap cache.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Link: http://lkml.kernel.org/r/20200311055352.50574-1-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swapfile: fix data races in try_to_unuse()

si->inuse_pages could be accessed concurrently as noticed by KCSAN,

 write to 0xffff98b00ebd04dc of 4 bytes by task 82262 on cpu 92:
  swap_range_free+0xbe/0x230
  swap_range_free at mm/swapfile.c:719
  swapcache_free_entries+0x1be/0x250
  free_swap_slot+0x1c8/0x220
  __swap_entry_free.constprop.19+0xa3/0xb0
  free_swap_and_cache+0x53/0xa0
  unmap_page_range+0x7e0/0x1ce0
  unmap_single_vma+0xcd/0x170
  unmap_vmas+0x18b/0x220
  exit_mmap+0xee/0x220
  mmput+0xe7/0x240
  do_exit+0x598/0xfd0
  do_group_exit+0x8b/0x180
  get_signal+0x293/0x13d0
  do_signal+0x37/0x5d0
  prepare_exit_to_usermode+0x1b7/0x2c0
  ret_from_intr+0x32/0x42

 read to 0xffff98b00ebd04dc of 4 bytes by task 82499 on cpu 46:
  try_to_unuse+0x86b/0xc80
  try_to_unuse at mm/swapfile.c:2185
  __x64_sys_swapoff+0x372/0xd40
  do_syscall_64+0x91/0xb05
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

The plain reads in try_to_unuse() are outside si->lock critical section
which result in data races that could be dangerous to be used in a loop.
Fix them by adding READ_ONCE().

Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Marco Elver <elver@google.com>
Cc: Hugh Dickins <hughd@google.com>
Link: http://lkml.kernel.org/r/1582578903-29294-1-git-send-email-cai@lca.pw
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap.c: not necessary to export __pagevec_lru_add()

__pagevec_lru_add() is only used in mm directory now.

Remove the export symbol.

Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200126011436.22979-1-richardw.yang@linux.intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swapfile.c: fix comments for swapcache_prepare

The -EEXIST returned by __swap_duplicate means there is a swap cache
instead -EBUSY

Signed-off-by: Chen Wandun <chenwandun@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200212145754.27123-1-chenwandun@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: fix omission of check on FOLL_LONGTERM in gup fast path

FOLL_LONGTERM is a special case of FOLL_PIN.  It suggests a pin which is
going to be given to hardware and can't move.  It would truncate CMA
permanently and should be excluded.

In gup slow path, where
__gup_longterm_locked->check_and_migrate_cma_pages() handles
FOLL_LONGTERM, but in fast path, there lacks such a check, which means a
possible leak of CMA page to longterm pinned.

Place a check in try_grab_compound_head() in the fast path to fix the
leak, and if FOLL_LONGTERM happens on CMA, it will fall back to slow path
to migrate the page.

Some note about the check: Huge page's subpages have the same migrate type
due to either allocation from a free_list[] or alloc_contig_range() with
param MIGRATE_MOVABLE.  So it is enough to check on a single subpage by
is_migrate_cma_page(subpage)

Signed-off-by: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Link: http://lkml.kernel.org/r/1584876733-17405-3-git-send-email-kernelfans@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup: rename nr as nr_pinned in get_user_pages_fast()

To better reflect the held state of pages and make code self-explaining,
rename nr as nr_pinned.

Signed-off-by: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Link: http://lkml.kernel.org/r/1584876733-17405-2-git-send-email-kernelfans@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/gup/writeback: add callbacks for inaccessible pages

With the introduction of protected KVM guests on s390 there is now a
concept of inaccessible pages.  These pages need to be made accessible
before the host can access them.

While cpu accesses will trigger a fault that can be resolved, I/O accesses
will just fail.  We need to add a callback into architecture code for
places that will do I/O, namely when writeback is started or when a page
reference is taken.

This is not only to enable paging, file backing etc, it is also necessary
to protect the host against a malicious user space.  For example a bad
QEMU could simply start direct I/O on such protected memory.  We do not
want userspace to be able to trigger I/O errors and thus the logic is
"whenever somebody accesses that page (gup) or does I/O, make sure that
this page can be accessed".  When the guest tries to access that page we
will wait in the page fault handler for writeback to have finished and for
the page_ref to be the expected value.

On s390x the function is not supposed to fail, so it is ok to use a
WARN_ON on failure.  If we ever need some more finegrained handling we can
tackle this when we know the details.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Acked-by: Will Deacon <will@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200306132537.783769-3-imbrenda@linux.ibm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: dump_page(): additional diagnostics for huge pinned pages

As part of pin_user_pages() and related API calls, pages are "dma-pinned".
For the case of compound pages of order > 1, the per-page accounting of
dma pins is accomplished via the 3rd struct page in the compound page.  In
order to support debugging of any pin_user_pages()- related problems,
enhance dump_page() so as to report the pin count in that case.

Documentation/core-api/pin_user_pages.rst is also updated accordingly.

Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200211001536.1027652-13-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: improve dump_page() for compound pages

There was no protection against a corrupted struct page having an
implausible compound_head().  Sanity check that a compound page has a head
within reach of the maximum allocatable page (this will need to be
adjusted if one of the plans to allocate 1GB pages comes to fruition).  In
addition,

 - Print the mapping pointer using %p insted of %px.  The actual value of
   the pointer can be read out of the raw page dump and using %p gives a
   chance to correlate it with an earlier printk of the mapping pointer
 - Print the mapping pointer from the head page, not the tail page
   (the tail ->mapping pointer may be in use for other purposes, eg part
   of a list_head)
 - Print the order of the page for compound pages
 - Dump the raw head page as well as the raw page
 - Print the refcount from the head page, not the tail page

Suggested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Co-developed-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200211001536.1027652-12-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>