mm: track mapcount of large folios in single value

Let's track the mapcount of large folios in a single value.  The mapcount
of a large folio currently corresponds to the sum of the entire mapcount
and all page mapcounts.

This sum is what we actually want to know in folio_mapcount() and it is
also sufficient for implementing folio_mapped().

With PTE-mapped THP becoming more important and more widely used, we want
to avoid looping over all pages of a folio just to obtain the mapcount of
large folios.  The comment "In the common case, avoid the loop when no
pages mapped by PTE" in folio_total_mapcount() does no longer hold for
mTHP that are always mapped by PTE.

Further, we are planning on using folio_mapcount() more frequently, and
might even want to remove page mapcounts for large folios in some kernel
configs.  Therefore, allow for reading the mapcount of large folios
efficiently and atomically without looping over any pages.

Maintain the mapcount also for hugetlb pages for simplicity.  Use the new
mapcount to implement folio_mapcount() and folio_mapped().  Make
page_mapped() simply call folio_mapped().  We can now get rid of
folio_large_is_mapped().

_nr_pages_mapped is now only used in rmap code and for debugging purposes.
Keep folio_nr_pages_mapped() around, but document that its use should be
limited to rmap internals and debugging purposes.

This change implies one additional atomic add/sub whenever
mapping/unmapping (parts of) a large folio.

As we now batch RMAP operations for PTE-mapped THP during fork(), during
unmap/zap, and when PTE-remapping a PMD-mapped THP, and we adjust the
large mapcount for a PTE batch only once, the added overhead in the common
case is small.  Only when unmapping individual pages of a large folio
(e.g., during COW), the overhead might be bigger in comparison, but it's
essentially one additional atomic operation.

Note that before the new mapcount would overflow, already our refcount
would overflow: each mapping requires a folio reference.  Extend the
focumentation of folio_mapcount().

Link: https://lkml.kernel.org/r/20240409192301.907377-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/rmap: add fast-path for small folios when adding/removing/duplicating

Let's add a fast-path for small folios to all relevant rmap functions.
Note that only RMAP_LEVEL_PTE applies.

This is a preparation for tracking the mapcount of large folios in a
single value.

Link: https://lkml.kernel.org/r/20240409192301.907377-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/rmap: always inline anon/file rmap duplication of a single PTE

As we grow the code, the compiler might make stupid decisions and
unnecessarily degrade fork() performance.  Let's make sure to always
inline functions that operate on a single PTE so the compiler will always
optimize out the loop and avoid a function call.

This is a preparation for maintining a total mapcount for large folios.

Link: https://lkml.kernel.org/r/20240409192301.907377-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: allow for detecting underflows with page_mapcount() again

Patch series "mm: mapcount for large folios + page_mapcount() cleanups".

This series tracks the mapcount of large folios in a single value, so it
can be read efficiently and atomically, just like the mapcount of small
folios.

folio_mapcount() is then used in a couple more places, most notably to
reduce false negatives in folio_likely_mapped_shared(), and many users of
page_mapcount() are cleaned up (that's maybe why you got CCed on the full
series, sorry sh+xtensa folks!  :) ).

The remaining s390x user and one KSM user of page_mapcount() are getting
removed separately on the list right now.  I have patches to handle the
other KSM one, the khugepaged one and the kpagecount one; as they are not
as "obvious", I will send them out separately in the future.  Once that is
all in place, I'm planning on moving page_mapcount() into
fs/proc/task_mmu.c, the remaining user for the time being (and we can
discuss at LSF/MM details on that :) ).

I proposed the mapcount for large folios (previously called total
mapcount) originally in part of [1] and I later included it in [2] where
it is a requirement.  In the meantime, I changed the patch a bit so I
dropped all RB's.  During the discussion of [1], Peter Xu correctly raised
that this additional tracking might affect the performance when PMD->PTE
remapping THPs.  In the meantime.  I addressed that by batching RMAP
operations during fork(), unmap/zap and when PMD->PTE remapping THPs.

Running some of my micro-benchmarks [3] (fork,munmap,cow-byte,remap) on 1
GiB of memory backed by folios with the same order, I observe the
following on an Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz tuned for
reproducible results as much as possible:

Standard deviation is mostly < 1%, except for order-9, where it's < 2% for
fork() and munmap().

(1) Small folios are not affected (< 1%) in all 4 microbenchmarks.
(2) Order-4 folios are not affected (< 1%) in all 4 microbenchmarks. A bit
    weird comapred to the other orders ...
(3) PMD->PTE remapping of order-9 THPs is not affected (< 1%)
(4) COW-byte (COWing a single page by writing a single byte) is not
    affected for any order (< 1 %). The page copy_fault overhead dominates
    everything.
(5) fork() is mostly not affected (< 1%), except order-2, where we have
    a slowdown of ~4%. Already for order-3 folios, we're down to a slowdown
    of < 1%.
(6) munmap() sees a slowdown by < 3% for some orders (order-5,
    order-6, order-9), but less for others (< 1% for order-4 and order-8,
    < 2% for order-2, order-3, order-7).

Especially the fork() and munmap() benchmark are sensitive to each added
instruction and other system noise, so I suspect some of the change and
observed weirdness (order-4) is due to code layout changes and other
factors, but not really due to the added atomics.

So in the common case where we can batch, the added atomics don't really
make a big difference, especially in light of the recent improvements for
large folios that we recently gained due to batching.  Surprisingly, for
some cases where we cannot batch (e.g., COW), the added atomics don't seem
to matter, because other overhead dominates.

My fork and munmap micro-benchmarks don't cover cases where we cannot
batch-process bigger parts of large folios.  As this is not the common
case, I'm not worrying about that right now.

Future work is batching RMAP operations during swapout and folio
migration.

[1] https://lore.kernel.org/all/20230809083256.699513-1-david@redhat.com/
[2] https://lore.kernel.org/all/20231124132626.235350-1-david@redhat.com/
[3] https://gitlab.com/davidhildenbrand/scratchspace/-/raw/main/pte-mapped-folio-benchmarks.c?ref_type=heads

This patch (of 18):

Commit 53277bcf126d ("mm: support page_mapcount() on page_has_type()
pages") made it impossible to detect mapcount underflows by treating any
negative raw mapcount value as a mapcount of 0.

We perform such underflow checks in zap_present_folio_ptes() and
zap_huge_pmd(), which would currently no longer trigger.

Let's check against PAGE_MAPCOUNT_RESERVE instead by using
page_type_has_type(), like page_has_type() would, so we can still catch
some underflows.

[david@redhat.com: make page_mapcount() slighly more efficient]
Link: https://lkml.kernel.org/r/1af4fd61-7926-47c8-be45-833c0dbec08b@redhat.com
Link: https://lkml.kernel.org/r/20240409192301.907377-1-david@redhat.com
Link: https://lkml.kernel.org/r/20240409192301.907377-2-david@redhat.com
Fixes: 53277bcf126d ("mm: support page_mapcount() on page_has_type() pages")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yin Fengwei <fengwei.yin@intel.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: follow_pte() improvements

follow_pte() is now our main function to lookup PTEs in VM_PFNMAP/VM_IO
VMAs.  Let's perform some more sanity checks to make this exported
function harder to abuse.

Further, extend the doc a bit, it still focuses on the KVM use case with
MMU notifiers.  Drop the KVM+follow_pfn() comment, follow_pfn() is no
more, and we have other users nowadays.

Also extend the doc regarding refcounted pages and the interaction with
MMU notifiers.

KVM is one example that uses MMU notifiers and can deal with refcounted
pages properly.  VFIO is one example that doesn't use MMU notifiers, and
to prevent use-after-free, rejects refcounted pages: pfn_valid(pfn) &&
!PageReserved(pfn_to_page(pfn)).  Protection changes are less of a concern
for users like VFIO: the behavior is similar to longterm-pinning a page,
and getting the PTE protection changed afterwards.

The primary concern with refcounted pages is use-after-free, which callers
should be aware of.

Link: https://lkml.kernel.org/r/20240410155527.474777-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Fei Li <fei1.li@intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <seanjc@google.com>
Cc: Yonghua Huang <yonghua.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: pass VMA instead of MM to follow_pte()

... and centralize the VM_IO/VM_PFNMAP sanity check in there. We'll
now also perform these sanity checks for direct follow_pte()
invocations.

For generic_access_phys(), we might now check multiple times: nothing to
worry about, really.

Link: https://lkml.kernel.org/r/20240410155527.474777-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Sean Christopherson <seanjc@google.com> [KVM]
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Fei Li <fei1.li@intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Yonghua Huang <yonghua.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

drivers/virt/acrn: fix PFNMAP PTE checks in acrn_vm_ram_map()

Patch series "mm: follow_pte() improvements and acrn follow_pte() fixes".

Patch #1 fixes a bunch of issues I spotted in the acrn driver.  It
compiles, that's all I know.  I'll appreciate some review and testing from
acrn folks.

Patch #2+#3 improve follow_pte(), passing a VMA instead of the MM, adding
more sanity checks, and improving the documentation.  Gave it a quick test
on x86-64 using VM_PAT that ends up using follow_pte().

This patch (of 3):

We currently miss handling various cases, resulting in a dangerous
follow_pte() (previously follow_pfn()) usage.

(1) We're not checking PTE write permissions.

Maybe we should simply always require pte_write() like we do for
pin_user_pages_fast(FOLL_WRITE)? Hard to tell, so let's check for
ACRN_MEM_ACCESS_WRITE for now.

(2) We're not rejecting refcounted pages.

As we are not using MMU notifiers, messing with refcounted pages is
dangerous and can result in use-after-free. Let's make sure to reject them.

(3) We are only looking at the first PTE of a bigger range.

We only lookup a single PTE, but memmap->len may span a larger area.
Let's loop over all involved PTEs and make sure the PFN range is
actually contiguous. Reject everything else: it couldn't have worked
either way, and rather made use access PFNs we shouldn't be accessing.

Link: https://lkml.kernel.org/r/20240410155527.474777-1-david@redhat.com
Link: https://lkml.kernel.org/r/20240410155527.474777-2-david@redhat.com
Fixes: 8a6e85f75a83 ("virt: acrn: obtain pa from VMA with PFNMAP flag")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Fei Li <fei1.li@intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Yonghua Huang <yonghua.huang@intel.com>
Cc: Sean Christopherson <seanjc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm,swap: add document about RCU read lock and swapoff interaction

During reviewing a patch to fix the race condition between
free_swap_and_cache() and swapoff() [1], it was found that the document
about how to prevent racing with swapoff isn't clear enough.  Especially
RCU read lock can prevent swapoff from freeing data structures.  So, the
document is added as comments.

[1] https://lore.kernel.org/linux-mm/c8fe62d0-78b8-527a-5bef-ee663ccdc37a@huawei.com/

Link: https://lkml.kernel.org/r/20240407065450.498821-1-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Miaohe Lin <linmiaohe@huawei.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/mmap: make accountable_mapping return bool

accountable_mapping() can return bool, so change it.

Link: https://lkml.kernel.org/r/20240407063843.804274-1-gehao@kylinos.cn
Signed-off-by: Hao Ge <gehao@kylinos.cn>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/mmap: make vma_wants_writenotify return bool

vma_wants_writenotify() should return bool, so change it.

Link: https://lkml.kernel.org/r/20240407062653.803142-1-gehao@kylinos.cn
Signed-off-by: Hao Ge <gehao@kylinos.cn>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

memory tier: create CPUless memory tiers after obtaining HMAT info

The current implementation treats emulated memory devices, such as CXL1.1
type3 memory, as normal DRAM when they are emulated as normal memory
(E820_TYPE_RAM).  However, these emulated devices have different
characteristics than traditional DRAM, making it important to distinguish
them.  Thus, we modify the tiered memory initialization process to
introduce a delay specifically for CPUless NUMA nodes.  This delay ensures
that the memory tier initialization for these nodes is deferred until HMAT
information is obtained during the boot process.  Finally, demotion tables
are recalculated at the end.

* late_initcall(memory_tier_late_init);
  Some device drivers may have initialized memory tiers between
  `memory_tier_init()` and `memory_tier_late_init()`, potentially bringing
  online memory nodes and configuring memory tiers.  They should be
  excluded in the late init.

* Handle cases where there is no HMAT when creating memory tiers
  There is a scenario where a CPUless node does not provide HMAT
  information.  If no HMAT is specified, it falls back to using the
  default DRAM tier.

* Introduce another new lock `default_dram_perf_lock` for adist
  calculation In the current implementation, iterating through CPUlist
  nodes requires holding the `memory_tier_lock`.  However,
  `mt_calc_adistance()` will end up trying to acquire the same lock,
  leading to a potential deadlock.  Therefore, we propose introducing a
  standalone `default_dram_perf_lock` to protect `default_dram_perf_*`.
  This approach not only avoids deadlock but also prevents holding a large
  lock simultaneously.

* Upgrade `set_node_memory_tier` to support additional cases, including
  default DRAM, late CPUless, and hot-plugged initializations.  To cover
  hot-plugged memory nodes, `mt_calc_adistance()` and
  `mt_find_alloc_memory_type()` are moved into `set_node_memory_tier()` to
  handle cases where memtype is not initialized and where HMAT information
  is available.

* Introduce `default_memory_types` for those memory types that are not
  initialized by device drivers.  Because late initialized memory and
  default DRAM memory need to be managed, a default memory type is created
  for storing all memory types that are not initialized by device drivers
  and as a fallback.

Link: https://lkml.kernel.org/r/20240405000707.2670063-3-horenchuang@bytedance.com
Signed-off-by: Ho-Ren (Jack) Chuang <horenchuang@bytedance.com>
Signed-off-by: Hao Xiang <hao.xiang@bytedance.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Gregory Price <gourry.memverge@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ravi Jonnalagadda <ravis.opensrc@micron.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawie.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

memory tier: dax/kmem: introduce an abstract layer for finding, allocating, and putting memory types

Patch series "Improved Memory Tier Creation for CPUless NUMA Nodes", v11.

When a memory device, such as CXL1.1 type3 memory, is emulated as normal
memory (E820_TYPE_RAM), the memory device is indistinguishable from normal
DRAM in terms of memory tiering with the current implementation.  The
current memory tiering assigns all detected normal memory nodes to the
same DRAM tier.  This results in normal memory devices with different
attributions being unable to be assigned to the correct memory tier,
leading to the inability to migrate pages between different types of
memory.
https://lore.kernel.org/linux-mm/PH0PR08MB7955E9F08CCB64F23963B5C3A860A@PH0PR08MB7955.namprd08.prod.outlook.com/T/

This patchset automatically resolves the issues.  It delays the
initialization of memory tiers for CPUless NUMA nodes until they obtain
HMAT information and after all devices are initialized at boot time,
eliminating the need for user intervention.  If no HMAT is specified, it
falls back to using `default_dram_type`.

Example usecase:
We have CXL memory on the host, and we create VMs with a new system memory
device backed by host CXL memory.  We inject CXL memory performance
attributes through QEMU, and the guest now sees memory nodes with
performance attributes in HMAT.  With this change, we enable the guest
kernel to construct the correct memory tiering for the memory nodes.

This patch (of 2):

Since different memory devices require finding, allocating, and putting
memory types, these common steps are abstracted in this patch, enhancing
the scalability and conciseness of the code.

Link: https://lkml.kernel.org/r/20240405000707.2670063-1-horenchuang@bytedance.com
Link: https://lkml.kernel.org/r/20240405000707.2670063-2-horenchuang@bytedance.com
Signed-off-by: Ho-Ren (Jack) Chuang <horenchuang@bytedance.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawie.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Gregory Price <gourry.memverge@gmail.com>
Cc: Hao Xiang <hao.xiang@bytedance.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ravi Jonnalagadda <ravis.opensrc@micron.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: set pageblock_order to HPAGE_PMD_ORDER in case with !CONFIG_HUGETLB_PAGE but THP enabled

As Vlastimil suggested in previous discussion[1], it doesn't make sense to
set pageblock_order as MAX_PAGE_ORDER when hugetlbfs is not enabled and
THP is enabled.  Instead, it should be set to HPAGE_PMD_ORDER.

[1] https://lore.kernel.org/all/76457ec5-d789-449b-b8ca-dcb6ceb12445@suse.cz/
Link: https://lkml.kernel.org/r/3d57d253070035bdc0f6d6e5681ce1ed0e1934f7.1712286863.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: convert free_zone_device_page to free_zone_device_folio

Both callers already have a folio; pass it in and save a few calls to
compound_head().

Link: https://lkml.kernel.org/r/20240405153228.2563754-6-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: combine __folio_put_small, __folio_put_large and __folio_put

It's now obvious that __folio_put_small() and __folio_put_large() do
almost exactly the same thing.  Inline them both into __folio_put().

Link: https://lkml.kernel.org/r/20240405153228.2563754-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: inline destroy_large_folio() into __folio_put_large()

destroy_large_folio() has only one caller, move its contents there.

Link: https://lkml.kernel.org/r/20240405153228.2563754-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: combine free_the_page() and free_unref_page()

The pcp_allowed_order() check in free_the_page() was only being skipped by
__folio_put_small() which is about to be rearranged.

Link: https://lkml.kernel.org/r/20240405153228.2563754-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: free non-hugetlb large folios in a batch

Patch series "Clean up __folio_put()".

With all the changes over the last few years, __folio_put_small and
__folio_put_large have become almost identical to each other ...  except
you can't tell because they're spread over two files.  Rearrange it all so
that you can tell, and then inline them both into __folio_put().

This patch (of 5):

free_unref_folios() can now handle non-hugetlb large folios, so keep
normal large folios in the batch.  hugetlb folios still need to be handled
specially.

[peterx@redhat.com: fix panic]
Link: https://lkml.kernel.org/r/ZikjPB0Dt5HA8-uL@x1n
Link: https://lkml.kernel.org/r/20240405153228.2563754-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20240405153228.2563754-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: convert pagecache_isize_extended to use a folio

Remove four hidden calls to compound_head().  Also exit early if the
filesystem block size is >= PAGE_SIZE instead of just equal to PAGE_SIZE.

Link: https://lkml.kernel.org/r/20240405180038.2618624-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Pankaj Raghav <p.raghav@samsung.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/hugetlb: pass correct order_per_bit to cma_declare_contiguous_nid

The hugetlb_cma code passes 0 in the order_per_bit argument to
cma_declare_contiguous_nid (the alignment, computed using the page order,
is correctly passed in).

This causes a bit in the cma allocation bitmap to always represent a 4k
page, making the bitmaps potentially very large, and slower.

It would create bitmaps that would be pretty big.  E.g.  for a 4k page
size on x86, hugetlb_cma=64G would mean a bitmap size of (64G / 4k) / 8
== 2M.  With HUGETLB_PAGE_ORDER as order_per_bit, as intended, this
would be (64G / 2M) / 8 == 4k.  So, that's quite a difference.

Also, this restricted the hugetlb_cma area to ((PAGE_SIZE <<
MAX_PAGE_ORDER) * 8) * PAGE_SIZE (e.g.  128G on x86) , since
bitmap_alloc uses normal page allocation, and is thus restricted by
MAX_PAGE_ORDER.  Specifying anything about that would fail the CMA
initialization.

So, correctly pass in the order instead.

Link: https://lkml.kernel.org/r/20240404162515.527802-2-fvdl@google.com
Fixes: cf11e85fc08c ("mm: hugetlb: optionally allocate gigantic hugepages using cma")
Signed-off-by: Frank van der Linden <fvdl@google.com>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/cma: drop incorrect alignment check in cma_init_reserved_mem

cma_init_reserved_mem uses IS_ALIGNED to check if the size represented by
one bit in the cma allocation bitmask is aligned with
CMA_MIN_ALIGNMENT_BYTES (pageblock size).

However, this is too strict, as this will fail if order_per_bit >
pageblock_order, which is a valid configuration.

We could check IS_ALIGNED both ways, but since both numbers are powers of
two, no check is needed at all.

Link: https://lkml.kernel.org/r/20240404162515.527802-1-fvdl@google.com
Fixes: de9e14eebf33 ("drivers: dma-contiguous: add initialization from device tree")
Signed-off-by: Frank van der Linden <fvdl@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests/mm: fix additional build errors for selftests

These build errors only occur if one fails to first run "make headers".
However, that is a non-obvious and instrusive requirement, and so there
was a discussion on how to get rid of it [1].  This uses that solution.

These two files were created by taking a snapshot of the generated header
files that are created via "make headers".  These two files were copied
from ./usr/include/linux/ to ./tools/include/uapi/linux/ .

That fixes the selftests/mm build on today's Arch Linux (which required
the userfaultfd.h) and Ubuntu 23.04 (which additionally required memfd.h).

[1] https://lore.kernel.org/all/783a4178-1dec-4e30-989a-5174b8176b09@redhat.com/

Link: https://lkml.kernel.org/r/20240328033418.203790-3-jhubbard@nvidia.com
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Muhammad Usama Anjum <usama.anjum@collabora.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests: break the dependency upon local header files

Patch series "Fix selftests/mm build without requiring "make headers"".

As mentioned in each patch, this implements the solution that we discussed
in December 2023, in [1].  This turned out to be very clean and easy.  It
should also be quite easy to maintain.

This should also make Peter Zijlstra happy, because it directly addresses
the root cause of his "NAK NAK NAK" reply [2].  :)

[1] https://lore.kernel.org/all/783a4178-1dec-4e30-989a-5174b8176b09@redhat.com/
[2] https://lore.kernel.org/lkml/20231103121652.GA6217@noisy.programming.kicks-ass.net/

This patch (of 2):

Use tools/include/uapi/ files instead.  These are obtained by taking a
snapshot: run "make headers" at the top level, then copy the desired
header file into the appropriate subdir in tools/uapi/.

This was discussed and solved in [1].

However, even before copying any additional files there, there are already
quite a few in tools/include/uapi already.  And these will immediately fix
a number of selftests/mm build failures.

So this patch:

a) Adds TOOLS_INCLUDES to selftests/lib.mk, so that all selftests can
   immediately and easily include the snapshotted header files.

b) Uses $(TOOLS_INCLUDES) in the selftests/mm build.  On today's Arch
   Linux, this already fixes all build errors except for a few
   userfaultfd.h (those will be addressed in a subsequent patch).

[1] https://lore.kernel.org/all/783a4178-1dec-4e30-989a-5174b8176b09@redhat.com/

Link: https://lkml.kernel.org/r/20240328033418.203790-1-jhubbard@nvidia.com
Link: https://lkml.kernel.org/r/20240328033418.203790-2-jhubbard@nvidia.com
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Muhammad Usama Anjum <usama.anjum@collabora.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

hugetlb: convert hugetlb_wp() to use struct vm_fault

hugetlb_wp() can use the struct vm_fault passed in from hugetlb_fault().
This alleviates the stack by consolidating 5 variables into a single
struct.

[vishal.moola@gmail.com: simplify hugetlb_wp() arguments]
Link: https://lkml.kernel.org/r/ZhQtoFNZBNwBCeXn@fedora
Link: https://lkml.kernel.org/r/20240401202651.31440-4-vishal.moola@gmail.com
Signed-off-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

hugetlb: convert hugetlb_no_page() to use struct vm_fault

hugetlb_no_page() can use the struct vm_fault passed in from
hugetlb_fault(). This alleviates the stack by consolidating 7
variables into a single struct.

[vishal.moola@gmail.com:  simplify hugetlb_no_page() arguments]
Link: https://lkml.kernel.org/r/ZhQtN8y5zud8iI1u@fedora
Link: https://lkml.kernel.org/r/20240401202651.31440-3-vishal.moola@gmail.com
Signed-off-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

hugetlb: convert hugetlb_fault() to use struct vm_fault

Patch series "Hugetlb fault path to use struct vm_fault", v2.

This patchset converts the hugetlb fault path to use struct vm_fault.
This helps make the code more readable, and alleviates the stack by
allowing us to consolidate many fault-related variables into an individual
pointer.

This patch (of 3):

Now that hugetlb_fault() has a vm_fault available for fault tracking, use
it throughout.  This cleans up the code by removing 2 variables, and
prepares hugetlb_fault() to take in a struct vm_fault argument.

Link: https://lkml.kernel.org/r/20240401202651.31440-1-vishal.moola@gmail.com
Link: https://lkml.kernel.org/r/20240401202651.31440-2-vishal.moola@gmail.com
Signed-off-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Muchun Song <muchun.song@linux.dev>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/ksm: remove redundant code in ksm_fork

Since commit 3c6f33b7273a ("mm/ksm: support fork/exec for prctl"), when a
child process is forked, the MMF_VM_MERGE_ANY flag will be inherited in
mm_init(). So, it's unnecessary to set the flag in ksm_fork().

Link: https://lkml.kernel.org/r/20240402024934.1093361-1-tujinjiang@huawei.com
Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Stefan Roesch <shr@devkernel.io>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: use "GUP-fast" instead "fast GUP" in remaining comments

Let's fixup the remaining comments to consistently call that thing
"GUP-fast".  With this change, we consistently call it "GUP-fast".

Link: https://lkml.kernel.org/r/20240402125516.223131-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Mike Rapoport (IBM) <rppt@kernel.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/treewide: rename CONFIG_HAVE_FAST_GUP to CONFIG_HAVE_GUP_FAST

Nowadays, we call it "GUP-fast", the external interface includes functions
like "get_user_pages_fast()", and we renamed all internal functions to
reflect that as well.

Let's make the config option reflect that.

Link: https://lkml.kernel.org/r/20240402125516.223131-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Mike Rapoport (IBM) <rppt@kernel.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/gup: consistently name GUP-fast functions

Patch series "mm/gup: consistently call it GUP-fast".

Some cleanups around function names, comments and the config option of
"GUP-fast" -- GUP without "lock" safety belts on.

With this cleanup it's easy to judge which functions are GUP-fast
specific.  We now consistently call it "GUP-fast", avoiding mixing it with
"fast GUP", "lockless", or simply "gup" (which I always considered
confusing in the ode).

So the magic now happens in functions that contain "gup_fast", whereby
gup_fast() is the entry point into that magic.  Comments consistently
reference either "GUP-fast" or "gup_fast()".

This patch (of 3):

Let's consistently call the "fast-only" part of GUP "GUP-fast" and rename
all relevant internal functions to start with "gup_fast", to make it
clearer that this is not ordinary GUP.  The current mixture of "lockless",
"gup" and "gup_fast" is confusing.

Further, avoid the term "huge" when talking about a "leaf" -- for example,
we nowadays check pmd_leaf() because pmd_huge() is gone.  For the
"hugepd"/"hugepte" stuff, it's part of the name ("is_hugepd"), so that
stays.

What remains is the "external" interface:
* get_user_pages_fast_only()
* get_user_pages_fast()
* pin_user_pages_fast()

The high-level internal functions for GUP-fast (+slow fallback) are now:
* internal_get_user_pages_fast() -> gup_fast_fallback()
* lockless_pages_from_mm() -> gup_fast()

The basic GUP-fast walker functions:
* gup_pgd_range() -> gup_fast_pgd_range()
* gup_p4d_range() -> gup_fast_p4d_range()
* gup_pud_range() -> gup_fast_pud_range()
* gup_pmd_range() -> gup_fast_pmd_range()
* gup_pte_range() -> gup_fast_pte_range()
* gup_huge_pgd()  -> gup_fast_pgd_leaf()
* gup_huge_pud()  -> gup_fast_pud_leaf()
* gup_huge_pmd()  -> gup_fast_pmd_leaf()

The weird hugepd stuff:
* gup_huge_pd() -> gup_fast_hugepd()
* gup_hugepte() -> gup_fast_hugepte()

The weird devmap stuff:
* __gup_device_huge_pud() -> gup_fast_devmap_pud_leaf()
* __gup_device_huge_pmd   -> gup_fast_devmap_pmd_leaf()
* __gup_device_huge()     -> gup_fast_devmap_leaf()
* undo_dev_pagemap()      -> gup_fast_undo_dev_pagemap()

Helper functions:
* unpin_user_pages_lockless() -> gup_fast_unpin_user_pages()
* gup_fast_folio_allowed() is already properly named
* gup_fast_permitted() is already properly named

With "gup_fast()", we now even have a function that is referred to in
comment in mm/mmu_gather.c.

Link: https://lkml.kernel.org/r/20240402125516.223131-1-david@redhat.com
Link: https://lkml.kernel.org/r/20240402125516.223131-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Mike Rapoport (IBM) <rppt@kernel.org>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

hugetlb: convert alloc_buddy_hugetlb_folio to use a folio

While this function returned a folio, it was still using __alloc_pages()
and __free_pages().  Use __folio_alloc() and put_folio() instead.  This
actually removes a call to compound_head(), but more importantly, it
prepares us for the move to memdescs.

Link: https://lkml.kernel.org/r/20240402200656.913841-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove struct page from get_shadow_from_swap_cache

We don't actually use any parts of struct page; all we do is check the
value of the pointer.  So give the pointer the appropriate name & type.

Link: https://lkml.kernel.org/r/20240402201659.918308-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

x86: mm: accelerate pagefault when badaccess

The access_error() of vma is already checked under per-VMA lock, if it is
a bad access, directly handle error, no need to retry with mmap_lock
again.  In order to release the correct lock, pass the mm_struct into
bad_area_access_error().  If mm is NULL, release vma lock, or release
mmap_lock.  Since the page faut is handled under per-VMA lock, count it as
a vma lock event with VMA_LOCK_SUCCESS.

Link: https://lkml.kernel.org/r/20240403083805.1818160-8-wangkefeng.wang@huawei.com
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

s390: mm: accelerate pagefault when badaccess

The vm_flags of vma already checked under per-VMA lock, if it is a bad
access, directly handle error, no need to retry with mmap_lock again.
Since the page faut is handled under per-VMA lock, count it as a vma lock
event with VMA_LOCK_SUCCESS.

Link: https://lkml.kernel.org/r/20240403083805.1818160-7-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

riscv: mm: accelerate pagefault when badaccess

The access_error() of vma already checked under per-VMA lock, if it is a
bad access, directly handle error, no need to retry with mmap_lock again.
Since the page faut is handled under per-VMA lock, count it as a vma lock
event with VMA_LOCK_SUCCESS.

[wangkefeng.wang@huawei.com: use `cause' rather than SIGSEGV, per Alexandre]
Link: https://lkml.kernel.org/r/ac978061-ce1a-40a4-8b0a-61883b42bea7@huawei.com
Link: https://lkml.kernel.org/r/20240403083805.1818160-6-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Tested-by: Alexandre Ghiti <alexghiti@rivosinc.com
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

powerpc: mm: accelerate pagefault when badaccess

The access_[pkey]_error() of vma already checked under per-VMA lock, if it
is a bad access, directly handle error, no need to retry with mmap_lock
again.  In order to release the correct lock, pass the mm_struct into
bad_access_pkey()/bad_access(), if mm is NULL, release vma lock, or
release mmap_lock.  Since the page faut is handled under per-VMA lock,
count it as a vma lock event with VMA_LOCK_SUCCESS.

Link: https://lkml.kernel.org/r/20240403083805.1818160-5-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

arm: mm: accelerate pagefault when VM_FAULT_BADACCESS

The vm_flags of vma already checked under per-VMA lock, if it is a bad
access, directly set fault to VM_FAULT_BADACCESS and handle error, no need
to retry with mmap_lock again.  Since the page faut is handled under
per-VMA lock, count it as a vma lock event with VMA_LOCK_SUCCESS.

Link: https://lkml.kernel.org/r/20240403083805.1818160-4-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

arm64: mm: accelerate pagefault when VM_FAULT_BADACCESS

The vm_flags of vma already checked under per-VMA lock, if it is a bad
access, directly set fault to VM_FAULT_BADACCESS and handle error, no need
to retry with mmap_lock again, the latency time reduces 34% in 'lat_sig -P
1 prot lat_sig' from lmbench testcase.

Since the page fault is handled under per-VMA lock, count it as a vma lock
event with VMA_LOCK_SUCCESS.

Link: https://lkml.kernel.org/r/20240403083805.1818160-3-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

arm64: mm: cleanup __do_page_fault()

Patch series "arch/mm/fault: accelerate pagefault when badaccess", v2.

After VMA lock-based page fault handling enabled, if bad access met
under per-vma lock, it will fallback to mmap_lock-based handling,
so it leads to unnessary mmap lock and vma find again. A test from
lmbench shows 34% improve after this changes on arm64,

  lat_sig -P 1 prot lat_sig 0.29194 -> 0.19198

This patch (of 7):

The __do_page_fault() only calls handle_mm_fault() after vm_flags checked,
and it is only called by do_page_fault(), let's squash it into
do_page_fault() to cleanup code.

Link: https://lkml.kernel.org/r/20240403083805.1818160-1-wangkefeng.wang@huawei.com
Link: https://lkml.kernel.org/r/20240403083805.1818160-2-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: madvise: avoid split during MADV_PAGEOUT and MADV_COLD

Rework madvise_cold_or_pageout_pte_range() to avoid splitting any large
folio that is fully and contiguously mapped in the pageout/cold vm range.
This change means that large folios will be maintained all the way to swap
storage.  This both improves performance during swap-out, by eliding the
cost of splitting the folio, and sets us up nicely for maintaining the
large folio when it is swapped back in (to be covered in a separate
series).

Folios that are not fully mapped in the target range are still split, but
note that behavior is changed so that if the split fails for any reason
(folio locked, shared, etc) we now leave it as is and move to the next pte
in the range and continue work on the proceeding folios.  Previously any
failure of this sort would cause the entire operation to give up and no
folios mapped at higher addresses were paged out or made cold.  Given
large folios are becoming more common, this old behavior would have likely
lead to wasted opportunities.

While we are at it, change the code that clears young from the ptes to use
ptep_test_and_clear_young(), via the new mkold_ptes() batch helper
function.  This is more efficent than get_and_clear/modify/set, especially
for contpte mappings on arm64, where the old approach would require
unfolding/refolding and the new approach can be done in place.

Link: https://lkml.kernel.org/r/20240408183946.2991168-8-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Barry Song <v-songbaohua@oppo.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Gao Xiang <xiang@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: vmscan: avoid split during shrink_folio_list()

Now that swap supports storing all mTHP sizes, avoid splitting large
folios before swap-out.  This benefits performance of the swap-out path by
eliding split_folio_to_list(), which is expensive, and also sets us up for
swapping in large folios in a future series.

If the folio is partially mapped, we continue to split it since we want to
avoid the extra IO overhead and storage of writing out pages
uneccessarily.

THP_SWPOUT and THP_SWPOUT_FALLBACK counters should continue to count
events only for PMD-mappable folios to avoid user confusion.  THP_SWPOUT
already has the appropriate guard.  Add a guard for THP_SWPOUT_FALLBACK.
It may be appropriate to add per-size counters in future.

Link: https://lkml.kernel.org/r/20240408183946.2991168-7-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Barry Song <v-songbaohua@oppo.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Gao Xiang <xiang@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: swap: allow storage of all mTHP orders

Multi-size THP enables performance improvements by allocating large,
pte-mapped folios for anonymous memory.  However I've observed that on an
arm64 system running a parallel workload (e.g.  kernel compilation) across
many cores, under high memory pressure, the speed regresses.  This is due
to bottlenecking on the increased number of TLBIs added due to all the
extra folio splitting when the large folios are swapped out.

Therefore, solve this regression by adding support for swapping out mTHP
without needing to split the folio, just like is already done for
PMD-sized THP.  This change only applies when CONFIG_THP_SWAP is enabled,
and when the swap backing store is a non-rotating block device.  These are
the same constraints as for the existing PMD-sized THP swap-out support.

Note that no attempt is made to swap-in (m)THP here - this is still done
page-by-page, like for PMD-sized THP.  But swapping-out mTHP is a
prerequisite for swapping-in mTHP.

The main change here is to improve the swap entry allocator so that it can
allocate any power-of-2 number of contiguous entries between [1, (1 <<
PMD_ORDER)].  This is done by allocating a cluster for each distinct order
and allocating sequentially from it until the cluster is full.  This
ensures that we don't need to search the map and we get no fragmentation
due to alignment padding for different orders in the cluster.  If there is
no current cluster for a given order, we attempt to allocate a free
cluster from the list.  If there are no free clusters, we fail the
allocation and the caller can fall back to splitting the folio and
allocates individual entries (as per existing PMD-sized THP fallback).

The per-order current clusters are maintained per-cpu using the existing
infrastructure.  This is done to avoid interleving pages from different
tasks, which would prevent IO being batched.  This is already done for the
order-0 allocations so we follow the same pattern.

As is done for order-0 per-cpu clusters, the scanner now can steal order-0
entries from any per-cpu-per-order reserved cluster.  This ensures that
when the swap file is getting full, space doesn't get tied up in the
per-cpu reserves.

This change only modifies swap to be able to accept any order mTHP.  It
doesn't change the callers to elide doing the actual split.  That will be
done in separate changes.

Link: https://lkml.kernel.org/r/20240408183946.2991168-6-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gao Xiang <xiang@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: swap: update get_swap_pages() to take folio order

We are about to allow swap storage of any mTHP size.  To prepare for that,
let's change get_swap_pages() to take a folio order parameter instead of
nr_pages.  This makes the interface self-documenting; a power-of-2 number
of pages must be provided.  We will also need the order internally so this
simplifies accessing it.

Link: https://lkml.kernel.org/r/20240408183946.2991168-5-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Gao Xiang <xiang@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: swap: simplify struct percpu_cluster

struct percpu_cluster stores the index of cpu's current cluster and the
offset of the next entry that will be allocated for the cpu.  These two
pieces of information are redundant because the cluster index is just
(offset / SWAPFILE_CLUSTER).  The only reason for explicitly keeping the
cluster index is because the structure used for it also has a flag to
indicate "no cluster".  However this data structure also contains a spin
lock, which is never used in this context, as a side effect the code
copies the spinlock_t structure, which is questionable coding practice in
my view.

So let's clean this up and store only the next offset, and use a sentinal
value (SWAP_NEXT_INVALID) to indicate "no cluster".  SWAP_NEXT_INVALID is
chosen to be 0, because 0 will never be seen legitimately; The first page
in the swap file is the swap header, which is always marked bad to prevent
it from being allocated as an entry.  This also prevents the cluster to
which it belongs being marked free, so it will never appear on the free
list.

This change saves 16 bytes per cpu.  And given we are shortly going to
extend this mechanism to be per-cpu-AND-per-order, we will end up saving
16 * 9 = 144 bytes per cpu, which adds up if you have 256 cpus in the
system.

Link: https://lkml.kernel.org/r/20240408183946.2991168-4-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gao Xiang <xiang@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: swap: free_swap_and_cache_nr() as batched free_swap_and_cache()

Now that we no longer have a convenient flag in the cluster to determine
if a folio is large, free_swap_and_cache() will take a reference and lock
a large folio much more often, which could lead to contention and (e.g.)
failure to split large folios, etc.

Let's solve that problem by batch freeing swap and cache with a new
function, free_swap_and_cache_nr(), to free a contiguous range of swap
entries together.  This allows us to first drop a reference to each swap
slot before we try to release the cache folio.  This means we only try to
release the folio once, only taking the reference and lock once - much
better than the previous 512 times for the 2M THP case.

Contiguous swap entries are gathered in zap_pte_range() and
madvise_free_pte_range() in a similar way to how present ptes are already
gathered in zap_pte_range().

While we are at it, let's simplify by converting the return type of both
functions to void.  The return value was used only by zap_pte_range() to
print a bad pte, and was ignored by everyone else, so the extra reporting
wasn't exactly guaranteed.  We will still get the warning with most of the
information from get_swap_device().  With the batch version, we wouldn't
know which pte was bad anyway so could print the wrong one.

[ryan.roberts@arm.com: fix a build warning on parisc]
Link: https://lkml.kernel.org/r/20240409111840.3173122-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20240408183946.2991168-3-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Gao Xiang <xiang@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: swap: remove CLUSTER_FLAG_HUGE from swap_cluster_info:flags

Patch series "Swap-out mTHP without splitting", v7.

This series adds support for swapping out multi-size THP (mTHP) without
needing to first split the large folio via
split_huge_page_to_list_to_order().  It closely follows the approach
already used to swap-out PMD-sized THP.

There are a couple of reasons for swapping out mTHP without splitting:

  - Performance: It is expensive to split a large folio and under
    extreme memory pressure some workloads regressed performance when
    using 64K mTHP vs 4K small folios because of this extra cost in the
    swap-out path.  This series not only eliminates the regression but
    makes it faster to swap out 64K mTHP vs 4K small folios.

  - Memory fragmentation avoidance: If we can avoid splitting a large
    folio memory is less likely to become fragmented, making it easier to
    re-allocate a large folio in future.

  - Performance: Enables a separate series [7] to swap-in whole mTHPs,
    which means we won't lose the TLB-efficiency benefits of mTHP once the
    memory has been through a swap cycle.

I've done what I thought was the smallest change possible, and as a
result, this approach is only employed when the swap is backed by a
non-rotating block device (just as PMD-sized THP is supported today).
Discussion against the RFC concluded that this is sufficient.

Performance Testing
===================

I've run some swap performance tests on Ampere Altra VM (arm64) with 8
CPUs.  The VM is set up with a 35G block ram device as the swap device and
the test is run from inside a memcg limited to 40G memory.  I've then run
`usemem` from vm-scalability with 70 processes, each allocating and
writing 1G of memory.  I've repeated everything 6 times and taken the mean
performance improvement relative to 4K page baseline:

| alloc size |                baseline |           + this series |
|            | mm-unstable (~v6.9-rc1) |                         |
|:-----------|------------------------:|------------------------:|
| 4K Page    |                    0.0% |                    1.3% |
| 64K THP    |                  -13.6% |                   46.3% |
| 2M THP     |                   91.4% |                   89.6% |

So with this change, the 64K swap performance goes from a 14% regression to a
46% improvement. While 2M shows a small regression I'm confident that this is
just noise.

[1] https://lore.kernel.org/linux-mm/20231010142111.3997780-1-ryan.roberts@arm.com/
[2] https://lore.kernel.org/linux-mm/20231017161302.2518826-1-ryan.roberts@arm.com/
[3] https://lore.kernel.org/linux-mm/20231025144546.577640-1-ryan.roberts@arm.com/
[4] https://lore.kernel.org/linux-mm/20240311150058.1122862-1-ryan.roberts@arm.com/
[5] https://lore.kernel.org/linux-mm/20240327144537.4165578-1-ryan.roberts@arm.com/
[6] https://lore.kernel.org/linux-mm/20240403114032.1162100-1-ryan.roberts@arm.com/
[7] https://lore.kernel.org/linux-mm/20240304081348.197341-1-21cnbao@gmail.com/
[8] https://lore.kernel.org/linux-mm/CAGsJ_4yMOow27WDvN2q=E4HAtDd2PJ=OQ5Pj9DG+6FLWwNuXUw@mail.gmail.com/
[9] https://lore.kernel.org/linux-mm/579d5127-c763-4001-9625-4563a9316ac3@redhat.com/

This patch (of 7):

As preparation for supporting small-sized THP in the swap-out path,
without first needing to split to order-0, Remove the CLUSTER_FLAG_HUGE,
which, when present, always implies PMD-sized THP, which is the same as
the cluster size.

The only use of the flag was to determine whether a swap entry refers to a
single page or a PMD-sized THP in swap_page_trans_huge_swapped().  Instead
of relying on the flag, we now pass in order, which originates from the
folio's order.  This allows the logic to work for folios of any order.

The one snag is that one of the swap_page_trans_huge_swapped() call sites
does not have the folio.  But it was only being called there to shortcut a
call __try_to_reclaim_swap() in some cases.  __try_to_reclaim_swap() gets
the folio and (via some other functions) calls
swap_page_trans_huge_swapped().  So I've removed the problematic call site
and believe the new logic should be functionally equivalent.

That said, removing the fast path means that we will take a reference and
trylock a large folio much more often, which we would like to avoid.  The
next patch will solve this.

Removing CLUSTER_FLAG_HUGE also means we can remove split_swap_cluster()
which used to be called during folio splitting, since
split_swap_cluster()'s only job was to remove the flag.

Link: https://lkml.kernel.org/r/20240408183946.2991168-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20240408183946.2991168-2-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Chris Li <chrisl@kernel.org>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Gao Xiang <xiang@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: page_alloc: use the correct THP order for THP PCP

Commit 44042b449872 ("mm/page_alloc: allow high-order pages to be stored
on the per-cpu lists") extends the PCP allocator to store THP pages, and
it determines whether to cache THP pages in PCP by comparing with
pageblock_order.  But the pageblock_order is not always equal to THP
order.  It might also be MAX_PAGE_ORDER, which could prevent PCP from
caching THP pages.

Therefore, using HPAGE_PMD_ORDER instead to determine the need for caching
THP for PCP will fix this issue

Link: https://lkml.kernel.org/r/a25c9e14cd03907d5978b60546a69e6aa3fc2a7d.1712151833.git.baolin.wang@linux.alibaba.com
Fixes: 44042b449872 ("mm/page_alloc: allow high-order pages to be stored on the per-cpu lists")
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Reviewed-by: Barry Song <baohua@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: convert smaps_pmd_entry to use a folio

Replace two calls to compound_head() with one.

Link: https://lkml.kernel.org/r/20240403171456.1445117-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: pass a folio to smaps_page_accumulate()

Both callers already have a folio; pass it in instead of doing the
conversion each time.

Link: https://lkml.kernel.org/r/20240403171456.1445117-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: convert smaps_page_accumulate to use a folio

Replaces three calls to compound_head() with one.  Shrinks the function
from 2614 bytes to 1112 bytes in an allmodconfig build.

Link: https://lkml.kernel.org/r/20240403171456.1445117-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: convert gather_stats to use a folio

Patch series "Use folio APIs in procfs".

We're down to very few users of the PageFoo macros, with proc being a
major user.

After this patchset and another patchset I have for khugepaged, we can get
rid of PageActive, PageReadahead and PageSwapBacked.  This patchset has
the usual advantages in its own right of removing hidden calls to
compound_head().  We have the page table lock, so the mapcount & refcount
are stable and there can't be any races with folios suddenly becoming tail
pages.

This patch (of 4):

Replaces six calls to compound_head() with one.  Shrinks the function from
5054 bytes to 1756 bytes in an allmodconfig build.

Link: https://lkml.kernel.org/r/20240403171456.1445117-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20240403171456.1445117-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: generate PAGE_IDLE_FLAG definitions

If CONFIG_PAGE_IDLE_FLAG is not set, we can use FOLIO_FLAG_FALSE() to
generate these definitions.

Link: https://lkml.kernel.org/r/20240402201252.917342-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove page_idle and page_young wrappers

All users have now been converted to the folio equivalents, so remove the
page wrappers.

Link: https://lkml.kernel.org/r/20240402201252.917342-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: convert smaps_account() to use a folio

Replace seven calls to compound_head() with one.

Link: https://lkml.kernel.org/r/20240402201252.917342-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: convert clear_refs_pte_range to use a folio

Patch series "Remove page_idle and page_young wrappers".

There are only a couple of places left using the page wrappers for idle &
young tracking.  Convert the two users in proc and then we can remove the
wrappers.  That enables the further simplification of autogenerating the
definitions when CONFIG_PAGE_IDLE_FLAG is disabled.

This patch (of 4):

Replaces four calls to compound_head() with two.

Link: https://lkml.kernel.org/r/20240402201252.917342-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20240402201252.917342-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: use a folio throughout hpage_collapse_scan_file()

Replace the use of pages with folios.  Saves a few calls to
compound_head() and removes some uses of obsolete functions.

Link: https://lkml.kernel.org/r/20240403171838.1445826-8-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: use a folio throughout collapse_file()

Pull folios from the page cache instead of pages.  Half of this work had
been done already, but we were still operating on pages for a large chunk
of this function.  There is no attempt in this patch to handle large
folios that are smaller than a THP; that will have to wait for a future
patch.

[willy@infradead.org: the unlikely() is embedded in IS_ERR()]
Link: https://lkml.kernel.org/r/ZhIWX8K0E2tSyMSr@casper.infradead.org
Link: https://lkml.kernel.org/r/20240403171838.1445826-7-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: remove hpage from collapse_file()

Use new_folio throughout where we had been using hpage.

Link: https://lkml.kernel.org/r/20240403171838.1445826-6-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: pass a folio to __collapse_huge_page_copy()

Simplify the body of __collapse_huge_page_copy() while I'm looking at
it.

Link: https://lkml.kernel.org/r/20240403171838.1445826-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: remove hpage from collapse_huge_page()

Work purely in terms of the folio.  Removes a call to compound_head()
in put_page().

Link: https://lkml.kernel.org/r/20240403171838.1445826-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: convert alloc_charge_hpage to alloc_charge_folio

Both callers want to deal with a folio, so return a folio from this
function.

Link: https://lkml.kernel.org/r/20240403171838.1445826-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

khugepaged: inline hpage_collapse_alloc_folio()

Patch series "khugepaged folio conversions".

We've been kind of hacking piecemeal at converting khugepaged to use
folios instead of compound pages, and so this patchset is a little larger
than it should be as I undo some of our wrong moves in the past.  In
particular, collapse_file() now consistently uses 'new_folio' for the
freshly allocated folio and 'folio' for the one that's currently in use.

This patch (of 7):

This function has one caller, and the combined function is simpler to
read, reason about and modify.

Link: https://lkml.kernel.org/r/20240403171838.1445826-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20240403171838.1445826-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests/mm: mremap_test: use sscanf to parse /proc/self/maps

Enforce consistency across files by avoiding two separate functions to
parse /proc/self/maps, replacing them with a simple sscanf().

Link: https://lkml.kernel.org/r/20240330173557.2697684-4-dev.jain@arm.com
Signed-off-by: Dev Jain <dev.jain@arm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests/mm: mremap_test: optimize execution time from minutes to seconds using chunkwise memcmp

Mismatch index is currently being checked by a brute force iteration over
the buffer.  Instead, break the comparison into O(sqrt(n)) number of
chunks, with the chunk size of this order only, where n is the size of the
buffer.  Do a brute-force iteration to print to stdout only when the
highly optimized memcmp() library function returns a mismatch in the
chunk.  The time complexity of this algorithm is O(sqrt(n)) * t, where t
is the time taken by memcmp(); for our test conditions, it is safe to
assume t to be small.

Link: https://lkml.kernel.org/r/20240330173557.2697684-3-dev.jain@arm.com
Signed-off-by: Dev Jain <dev.jain@arm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests/mm: mremap_test: optimize using pre-filled random array and memcpy

Patch series "selftests/mm: mremap_test: Optimizations and style fixes".

The mremap_test, in a worst case controlled by the -t flag, does a for
loop iteration in orders of GB.  Without compromising on the stdout
report, the aim is to reduce this time.

A pre-filled random buffer is allocated based on the seed, replacing
repetitive rand() calls.  The byte pattern in the memory locations is set
through memcpy() from the random buffer.

Replacing the loop for printing the mismatch index to stdout, employ an
efficient algorithm by breaking the comparison into chunks, use the highly
optimized memcmp() library function, and when a mismatch does occur, only
then do a brute force iteration.

Also, use sscanf() to parse /proc/self/maps for consistency across files.

Execution time results (x86 system):
./mremap_test
Original: 3 seconds
After change: 0.8 seconds

./mremap_test -t100
Original: 17 seconds
After change: 2 seconds

./mremap_test -t0 (worst case):
Original: 9:40 minutes
After change: 45 seconds

This patch (of 3):

Allocate a pre-filled random buffer using the seed.  Replace iterative
copying of the random sequence to buffers using the highly optimized
library function memcpy().

Link: https://lkml.kernel.org/r/20240330173557.2697684-1-dev.jain@arm.com
Link: https://lkml.kernel.org/r/20240330173557.2697684-2-dev.jain@arm.com
Signed-off-by: Dev Jain <dev.jain@arm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

memory: remove the now superfluous sentinel element from ctl_table array

This commit comes at the tail end of a greater effort to remove the empty
elements at the end of the ctl_table arrays (sentinels) which will reduce
the overall build time size of the kernel and run time memory bloat by ~64
bytes per sentinel (further information Link :
https://lore.kernel.org/all/ZO5Yx5JFogGi%2FcBo@bombadil.infradead.org/)

Remove sentinel from all files under mm/ that register a sysctl table.

Link: https://lkml.kernel.org/r/20240328-jag-sysctl_remset_misc-v1-1-47c1463b3af2@samsung.com
Signed-off-by: Joel Granados <j.granados@samsung.com>
Reviewed-by: Muchun Song <muchun.song@linux.dev>
Reviewed-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: rename vma_pgoff_address back to vma_address

With all callers converted, we can use the nice shorter name.  Take this
opportunity to reorder the arguments to the logical order (larger object
first).

Link: https://lkml.kernel.org/r/20240328225831.1765286-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove vma_address()

Convert the three remaining callers to call vma_pgoff_address() directly.
This removes an ambiguity where we'd check just one page if passed a tail
page and all N pages if passed a head page.

Also add better kernel-doc for vma_pgoff_address().

Link: https://lkml.kernel.org/r/20240328225831.1765286-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: correct page_mapped_in_vma() for large folios

Patch series "Unify vma_address and vma_pgoff_address".

The current vma_address() pretends that the ambiguity between head & tail
page is an advantage.  If you pass a head page to vma_address(), it will
operate on all pages in the folio, while if you pass a tail page, it will
operate on a single page.  That's not what any of the callers actually
want, so first convert all callers to use vma_pgoff_address() and then
rename vma_pgoff_address() to vma_address().

This patch (of 3):

If 'page' is the first page of a large folio then vma_address() will scan
for any page in the entire folio.  This can lead to page_mapped_in_vma()
returning true if some of the tail pages are mapped and the head page is
not.  This could lead to memory failure choosing to kill a task
unnecessarily.

Link: https://lkml.kernel.org/r/20240328225831.1765286-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20240328225831.1765286-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: huge_memory: add the missing folio_test_pmd_mappable() for THP split statistics

Now the mTHP can also be split or added into the deferred list, so add
folio_test_pmd_mappable() validation for PMD mapped THP, to avoid
confusion with PMD mapped THP related statistics.

[baolin.wang@linux.alibaba.com: check THP earlier in case folio is split, per Lance]
Link: https://lkml.kernel.org/r/b99f8cb14bc85fdb6ab43721d1331cb5ebed2581.1713771041.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/a5341defeef27c9ac7b85c97f030f93e4368bbc1.1711694852.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: support multi-size THP numa balancing

Now the anonymous page allocation already supports multi-size THP (mTHP),
but the numa balancing still prohibits mTHP migration even though it is an
exclusive mapping, which is unreasonable.

Allow scanning mTHP:
Commit 859d4adc3415 ("mm: numa: do not trap faults on shared data section
pages") skips shared CoW pages' NUMA page migration to avoid shared data
segment migration. In addition, commit 80d47f5de5e3 ("mm: don't try to
NUMA-migrate COW pages that have other uses") change to use page_count()
to avoid GUP pages migration, that will also skip the mTHP numa scanning.
Theoretically, we can use folio_maybe_dma_pinned() to detect the GUP
issue, although there is still a GUP race, the issue seems to have been
resolved by commit 80d47f5de5e3. Meanwhile, use the folio_likely_mapped_shared()
to skip shared CoW pages though this is not a precise sharers count. To
check if the folio is shared, ideally we want to make sure every page is
mapped to the same process, but doing that seems expensive and using
the estimated mapcount seems can work when running autonuma benchmark.

Allow migrating mTHP:
As mentioned in the previous thread[1], large folios (including THP) are
more susceptible to false sharing issues among threads than 4K base page,
leading to pages ping-pong back and forth during numa balancing, which is
currently not easy to resolve. Therefore, as a start to support mTHP numa
balancing, we can follow the PMD mapped THP's strategy, that means we can
reuse the 2-stage filter in should_numa_migrate_memory() to check if the
mTHP is being heavily contended among threads (through checking the CPU id
and pid of the last access) to avoid false sharing at some degree. Thus,
we can restore all PTE maps upon the first hint page fault of a large folio
to follow the PMD mapped THP's strategy. In the future, we can continue to
optimize the NUMA balancing algorithm to avoid the false sharing issue with
large folios as much as possible.

Performance data:
Machine environment: 2 nodes, 128 cores Intel(R) Xeon(R) Platinum
Base: 2024-03-25 mm-unstable branch
Enable mTHP to run autonuma-benchmark

mTHP:16K
Base Patched
numa01 numa01
224.70 143.48
numa01_THREAD_ALLOC numa01_THREAD_ALLOC
118.05 47.43
numa02 numa02
13.45 9.29
numa02_SMT numa02_SMT
14.80 7.50

mTHP:64K
Base Patched
numa01 numa01
216.15 114.40
numa01_THREAD_ALLOC numa01_THREAD_ALLOC
115.35 47.41
numa02 numa02
13.24 9.25
numa02_SMT numa02_SMT
14.67 7.34

mTHP:128K
Base Patched
numa01 numa01
205.13 144.45
numa01_THREAD_ALLOC numa01_THREAD_ALLOC
112.93 41.88
numa02 numa02
13.16 9.18
numa02_SMT numa02_SMT
14.81 7.49

[1] https://lore.kernel.org/all/20231117100745.fnpijbk4xgmals3k@techsingularity.net/

[baolin.wang@linux.alibaba.com: v3]
Link: https://lkml.kernel.org/r/c33a5c0b0a0323b1f8ed53772f50501f4b196e25.1712132950.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/d28d276d599c26df7f38c9de8446f60e22dd1950.1711683069.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: factor out the numa mapping rebuilding into a new helper

Patch series "support multi-size THP numa balancing", v2.

This patchset tries to support mTHP numa balancing, as a simple solution
to start, the NUMA balancing algorithm for mTHP will follow the THP
strategy as the basic support.  Please find details in each patch.

This patch (of 2):

To support large folio's numa balancing, factor out the numa mapping
rebuilding into a new helper as a preparation.

Link: https://lkml.kernel.org/r/cover.1712132950.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/cover.1711683069.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/8bc2586bdd8dbbe6d83c09b77b360ec8fcac3736.1711683069.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: alloc_anon_folio: avoid doing vma_thp_gfp_mask in fallback cases

Fallback rates surpassing 90% have been observed on phones utilizing 64KiB
CONT-PTE mTHP.  In these scenarios, when one out of every 16 PTEs fails to
allocate large folios, the remaining 15 PTEs fallback.  Consequently,
invoking vma_thp_gfp_mask seems redundant in such cases.  Furthermore,
abstaining from its use can also contribute to improved code readability.

Link: https://lkml.kernel.org/r/20240329073750.20012-1-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Acked-by: Yu Zhao <yuzhao@google.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: David Rientjes <rientjes@google.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Itaru Kitayama <itaru.kitayama@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yin Fengwei <fengwei.yin@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

zram: add max_pages param to recompression

Introduce "max_pages" param to recompress device attribute which sets an
upper limit on the number of entries (pages) zram attempts to recompress
(in this particular recompression call).  S/W recompression can be quite
expensive so limiting the number of pages recompress touches can be quite
helpful.

Link: https://lkml.kernel.org/r/20240329094050.2815699-1-senozhatsky@chromium.org
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Acked-by: Brian Geffon <bgeffon@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: init_mlocked_on_free_v3

Implements the "init_mlocked_on_free" boot option. When this boot option
is enabled, any mlock'ed pages are zeroed on free. If
the pages are munlock'ed beforehand, no initialization takes place.
This boot option is meant to combat the performance hit of
"init_on_free" as reported in commit 6471384af2a6 ("mm: security:
introduce init_on_alloc=1 and init_on_free=1 boot options"). With
"init_mlocked_on_free=1" only relevant data is freed while everything
else is left untouched by the kernel. Correspondingly, this patch
introduces no performance hit for unmapping non-mlock'ed memory. The
unmapping overhead for purely mlocked memory was measured to be
approximately 13%. Realistically, most systems mlock only a fraction of
the total memory so the real-world system overhead should be close to
zero.

Optimally, userspace programs clear any key material or other
confidential memory before exit and munlock the according memory
regions. If a program crashes, userspace key managers fail to do this
job. Accordingly, no munlock operations are performed so the data is
caught and zeroed by the kernel. Should the program not crash, all
memory will ideally be munlocked so no overhead is caused.

CONFIG_INIT_MLOCKED_ON_FREE_DEFAULT_ON can be set to enable
"init_mlocked_on_free" by default.

Link: https://lkml.kernel.org/r/20240329145605.149917-1-yjnworkstation@gmail.com
Signed-off-by: York Jasper Niebuhr <yjnworkstation@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: York Jasper Niebuhr <yjnworkstation@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftest/mm: ksm_functional_tests: extend test case for ksm fork/exec

This extends test_prctl_fork() and test_prctl_fork_exec() to make sure
that deduplication really happens, instead of only testing the
MMF_VM_MERGE_ANY flag is set.

[colin.i.king@gmail.com: fix spelling mistake in ksft_test_result_skip message]
Link: https://lkml.kernel.org/r/20240402081537.1365939-1-colin.i.king@gmail.com
Link: https://lkml.kernel.org/r/20240328111010.1502191-4-tujinjiang@huawei.com
Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Stefan Roesch <shr@devkernel.io>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftest/mm: ksm_functional_tests: refactor mmap_and_merge_range()

In order to extend test_prctl_fork() and test_prctl_fork_exec() to make
sure that deduplication really happens, mmap_and_merge_range() needs to be
refactored.

Firstly, mmap_and_merge_range() will be called with no need to call enable
KSM by madvise or prctl.  So, switch the 'bool use_prctl' parameter to
enum ksm_merge_mode.

Secondly, mmap_and_merge_range() will be called in child process in the
two testcases, it isn't appropriate to call ksft_test_result_{fail, skip},
because the global variables ksft_{fail, skip} aren't consistent with the
parent process.  Thus, convert calls of ksft_test_result_{fail, skip} to
ksft_print_msg(), return differrent error according to the two cases, and
rename mmap_and_merge_range() to __mmap_and_merge_range().  For existing
callers, introduce new mmap_and_merge_range() to handle different return
values of __mmap_and_merge_range().

Link: https://lkml.kernel.org/r/20240328111010.1502191-3-tujinjiang@huawei.com
Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Stefan Roesch <shr@devkernel.io>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/ksm: fix ksm exec support for prctl

Patch series "mm/ksm: fix ksm exec support for prctl", v4.

commit 3c6f33b7273a ("mm/ksm: support fork/exec for prctl") inherits
MMF_VM_MERGE_ANY flag when a task calls execve().  However, it doesn't
create the mm_slot, so ksmd will not try to scan this task.  The first
patch fixes the issue.

The second patch refactors to prepare for the third patch.  The third
patch extends the selftests of ksm to verfity the deduplication really
happens after fork/exec inherits ths KSM setting.

This patch (of 3):

commit 3c6f33b7273a ("mm/ksm: support fork/exec for prctl") inherits
MMF_VM_MERGE_ANY flag when a task calls execve().  Howerver, it doesn't
create the mm_slot, so ksmd will not try to scan this task.

To fix it, allocate and add the mm_slot to ksm_mm_head in __bprm_mm_init()
when the mm has MMF_VM_MERGE_ANY flag.

Link: https://lkml.kernel.org/r/20240328111010.1502191-1-tujinjiang@huawei.com
Link: https://lkml.kernel.org/r/20240328111010.1502191-2-tujinjiang@huawei.com
Fixes: 3c6f33b7273a ("mm/ksm: support fork/exec for prctl")
Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Stefan Roesch <shr@devkernel.io>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

selftests/x86: add placement guard gap test for shstk

The existing shadow stack test for guard gaps just checks that new
mappings are not placed in an existing mapping's guard gap.  Add one that
checks that new mappings are not placed such that preexisting mappings are
in the new mappings guard gap.

Link: https://lkml.kernel.org/r/20240326021656.202649-15-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

x86/mm: care about shadow stack guard gap during placement

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The longstanding behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

Now that the vm_flags is passed into the arch get_unmapped_area()'s, and
vm_unmapped_area() is ready to consider it, have VM_SHADOW_STACK's get
guard gap consideration for scenario 2.

Link: https://lkml.kernel.org/r/20240326021656.202649-14-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

x86/mm: implement HAVE_ARCH_UNMAPPED_AREA_VMFLAGS

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The longstanding behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

Add x86 arch implementations of arch_get_unmapped_area_vmflags/_topdown()
so future changes can allow the guard gap of type of vma being placed to
be taken into account.  This will be used for shadow stack memory.

Link: https://lkml.kernel.org/r/20240326021656.202649-13-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: take placement mappings gap into account

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The longstanding behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

For MAP_GROWSDOWN/VM_GROWSDOWN and MAP_GROWSUP/VM_GROWSUP this has not
been a problem in practice because applications place these kinds of
mappings very early, when there is not many mappings to find a space
between.  But for shadow stacks, they may be placed throughout the
lifetime of the application.

Use the start_gap field to find a space that includes the guard gap for
the new mapping.  Take care to not interfere with the alignment.

Link: https://lkml.kernel.org/r/20240326021656.202649-12-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

treewide: use initializer for struct vm_unmapped_area_info

Future changes will need to add a new member to struct
vm_unmapped_area_info.  This would cause trouble for any call site that
doesn't initialize the struct.  Currently every caller sets each member
manually, so if new ones are added they will be uninitialized and the core
code parsing the struct will see garbage in the new member.

It could be possible to initialize the new member manually to 0 at each
call site.  This and a couple other options were discussed.  Having some
struct vm_unmapped_area_info instances not zero initialized will put those
sites at risk of feeding garbage into vm_unmapped_area(), if the
convention is to zero initialize the struct and any new field addition
missed a call site that initializes each field manually.  So it is useful
to do things similar across the kernel.

The consensus (see links) was that in general the best way to accomplish
taking into account both code cleanliness and minimizing the chance of
introducing bugs, was to do C99 static initialization.  As in: struct
vm_unmapped_area_info info = {};

With this method of initialization, the whole struct will be zero
initialized, and any statements setting fields to zero will be unneeded.
The change should not leave cleanup at the call sides.

While iterating though the possible solutions a few archs kindly acked
other variations that still zero initialized the struct.  These sites have
been modified in previous changes using the pattern acked by the
respective arch.

So to be reduce the chance of bugs via uninitialized fields, perform a
tree wide change using the consensus for the best general way to do this
change.  Use C99 static initializing to zero the struct and remove and
statements that simply set members to zero.

Link: https://lkml.kernel.org/r/20240326021656.202649-11-rick.p.edgecombe@intel.com
Link: https://lore.kernel.org/lkml/202402280912.33AEE7A9CF@keescook/#t
Link: https://lore.kernel.org/lkml/j7bfvig3gew3qruouxrh7z7ehjjafrgkbcmg6tcghhfh3rhmzi@wzlcoecgy5rs/
Link: https://lore.kernel.org/lkml/ec3e377a-c0a0-4dd3-9cb9-96517e54d17e@csgroup.eu/
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

powerpc: use initializer for struct vm_unmapped_area_info

Future changes will need to add a new member to struct
vm_unmapped_area_info.  This would cause trouble for any call site that
doesn't initialize the struct.  Currently every caller sets each member
manually, so if new members are added they will be uninitialized and the
core code parsing the struct will see garbage in the new member.

It could be possible to initialize the new member manually to 0 at each
call site.  This and a couple other options were discussed, and a working
consensus (see links) was that in general the best way to accomplish this
would be via static initialization with designated member initiators.
Having some struct vm_unmapped_area_info instances not zero initialized
will put those sites at risk of feeding garbage into vm_unmapped_area() if
the convention is to zero initialize the struct and any new member
addition misses a call site that initializes each member manually.

It could be possible to leave the code mostly untouched, and just change
the line:
struct vm_unmapped_area_info info
to:
struct vm_unmapped_area_info info = {};

However, that would leave cleanup for the members that are manually set to
zero, as it would no longer be required.

So to be reduce the chance of bugs via uninitialized members, instead
simply continue the process to initialize the struct this way tree wide.
This will zero any unspecified members.  Move the member initializers to
the struct declaration when they are known at that time.  Leave the
members out that were manually initialized to zero, as this would be
redundant for designated initializers.

Link: https://lkml.kernel.org/r/20240326021656.202649-10-rick.p.edgecombe@intel.com
Link: https://lore.kernel.org/lkml/202402280912.33AEE7A9CF@keescook/#t
Link: https://lore.kernel.org/lkml/j7bfvig3gew3qruouxrh7z7ehjjafrgkbcmg6tcghhfh3rhmzi@wzlcoecgy5rs/
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

parisc: use initializer for struct vm_unmapped_area_info

Future changes will need to add a new member to struct
vm_unmapped_area_info.  This would cause trouble for any call site that
doesn't initialize the struct.  Currently every caller sets each member
manually, so if new members are added they will be uninitialized and the
core code parsing the struct will see garbage in the new member.

It could be possible to initialize the new member manually to 0 at each
call site.  This and a couple other options were discussed, and a working
consensus (see links) was that in general the best way to accomplish this
would be via static initialization with designated member initiators.
Having some struct vm_unmapped_area_info instances not zero initialized
will put those sites at risk of feeding garbage into vm_unmapped_area() if
the convention is to zero initialize the struct and any new member
addition misses a call site that initializes each member manually.

It could be possible to leave the code mostly untouched, and just change
the line:
struct vm_unmapped_area_info info
to:
struct vm_unmapped_area_info info = {};

However, that would leave cleanup for the members that are manually set
to zero, as it would no longer be required.

So to be reduce the chance of bugs via uninitialized members, instead
simply continue the process to initialize the struct this way tree wide.
This will zero any unspecified members.  Move the member initializers to
the struct declaration when they are known at that time.  Leave the
members out that were manually initialized to zero, as this would be
redundant for designated initializers.

Link: https://lkml.kernel.org/r/20240326021656.202649-9-rick.p.edgecombe@intel.com
Link: https://lore.kernel.org/lkml/202402280912.33AEE7A9CF@keescook/#t
Link: https://lore.kernel.org/lkml/j7bfvig3gew3qruouxrh7z7ehjjafrgkbcmg6tcghhfh3rhmzi@wzlcoecgy5rs/
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

csky: use initializer for struct vm_unmapped_area_info

Future changes will need to add a new member to struct
vm_unmapped_area_info.  This would cause trouble for any call site that
doesn't initialize the struct.  Currently every caller sets each member
manually, so if new members are added they will be uninitialized and the
core code parsing the struct will see garbage in the new member.

It could be possible to initialize the new member manually to 0 at each
call site.  This and a couple other options were discussed, and a working
consensus (see links) was that in general the best way to accomplish this
would be via static initialization with designated member initiators.
Having some struct vm_unmapped_area_info instances not zero initialized
will put those sites at risk of feeding garbage into vm_unmapped_area() if
the convention is to zero initialize the struct and any new member
addition misses a call site that initializes each member manually.

It could be possible to leave the code mostly untouched, and just change
the line:
struct vm_unmapped_area_info info
to:
struct vm_unmapped_area_info info = {};

However, that would leave cleanup for the members that are manually set to
zero, as it would no longer be required.

So to be reduce the chance of bugs via uninitialized members, instead
simply continue the process to initialize the struct this way tree wide.
This will zero any unspecified members.  Move the member initializers to
the struct declaration when they are known at that time.  Leave the
members out that were manually initialized to zero, as this would be
redundant for designated initializers.

Link: https://lkml.kernel.org/r/20240326021656.202649-8-rick.p.edgecombe@intel.com
Link: https://lore.kernel.org/lkml/202402280912.33AEE7A9CF@keescook/#t
Link: https://lore.kernel.org/lkml/j7bfvig3gew3qruouxrh7z7ehjjafrgkbcmg6tcghhfh3rhmzi@wzlcoecgy5rs/
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Guo Ren <guoren@kernel.org>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Guo Ren <guoren@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

thp: add thp_get_unmapped_area_vmflags()

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The longstanding behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

Add a THP implementations of the vm_flags variant of get_unmapped_area().
Future changes will call this from mmap.c in the do_mmap() path to allow
shadow stacks to be placed with consideration taken for the start guard
gap.  Shadow stack memory is always private and anonymous and so special
guard gap logic is not needed in a lot of caseis, but it can be mapped by
THP, so needs to be handled.

Link: https://lkml.kernel.org/r/20240326021656.202649-7-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: use get_unmapped_area_vmflags()

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The long standing behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

Use mm_get_unmapped_area_vmflags() in the do_mmap() so future changes can
cause shadow stack mappings to be placed with a guard gap.  Also use the
THP variant that takes vm_flags, such that THP shadow stack can get the
same treatment.  Adjust the vm_flags calculation to happen earlier so that
the vm_flags can be passed into __get_unmapped_area().

Link: https://lkml.kernel.org/r/20240326021656.202649-6-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove export for get_unmapped_area()

The mm/mmap.c function get_unmapped_area() is not used from any modules,
so it doesn't need to be exported.  Remove the export.

Link: https://lkml.kernel.org/r/20240326021656.202649-5-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: introduce arch_get_unmapped_area_vmflags()

When memory is being placed, mmap() will take care to respect the guard
gaps of certain types of memory (VM_SHADOWSTACK, VM_GROWSUP and
VM_GROWSDOWN).  In order to ensure guard gaps between mappings, mmap()
needs to consider two things:

 1. That the new mapping isn't placed in an any existing mappings guard
    gaps.
 2. That the new mapping isn't placed such that any existing mappings
    are not in *its* guard gaps.

The longstanding behavior of mmap() is to ensure 1, but not take any care
around 2.  So for example, if there is a PAGE_SIZE free area, and a mmap()
with a PAGE_SIZE size, and a type that has a guard gap is being placed,
mmap() may place the shadow stack in the PAGE_SIZE free area.  Then the
mapping that is supposed to have a guard gap will not have a gap to the
adjacent VMA.

In order to take the start gap into account, the maple tree search needs
to know the size of start gap the new mapping will need.  The call chain
from do_mmap() to the actual maple tree search looks like this:

do_mmap(size, vm_flags, map_flags, ..)
mm/mmap.c:get_unmapped_area(size, map_flags, ...)
arch_get_unmapped_area(size, map_flags, ...)
vm_unmapped_area(struct vm_unmapped_area_info)

One option would be to add another MAP_ flag to mean a one page start gap
(as is for shadow stack), but this consumes a flag unnecessarily.  Another
option could be to simply increase the size passed in do_mmap() by the
start gap size, and adjust after the fact, but this will interfere with
the alignment requirements passed in struct vm_unmapped_area_info, and
unknown to mmap.c.  Instead, introduce variants of
arch_get_unmapped_area/_topdown() that take vm_flags.  In future changes,
these variants can be used in mmap.c:get_unmapped_area() to allow the
vm_flags to be passed through to vm_unmapped_area(), while preserving the
normal arch_get_unmapped_area/_topdown() for the existing callers.

Link: https://lkml.kernel.org/r/20240326021656.202649-4-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: switch mm->get_unmapped_area() to a flag

The mm_struct contains a function pointer *get_unmapped_area(), which is
set to either arch_get_unmapped_area() or arch_get_unmapped_area_topdown()
during the initialization of the mm.

Since the function pointer only ever points to two functions that are
named the same across all arch's, a function pointer is not really
required.  In addition future changes will want to add versions of the
functions that take additional arguments.  So to save a pointers worth of
bytes in mm_struct, and prevent adding additional function pointers to
mm_struct in future changes, remove it and keep the information about
which get_unmapped_area() to use in a flag.

Add the new flag to MMF_INIT_MASK so it doesn't get clobbered on fork by
mmf_init_flags().  Most MM flags get clobbered on fork.  In the
pre-existing behavior mm->get_unmapped_area() would get copied to the new
mm in dup_mm(), so not clobbering the flag preserves the existing behavior
around inheriting the topdown-ness.

Introduce a helper, mm_get_unmapped_area(), to easily convert code that
refers to the old function pointer to instead select and call either
arch_get_unmapped_area() or arch_get_unmapped_area_topdown() based on the
flag.  Then drop the mm->get_unmapped_area() function pointer.  Leave the
get_unmapped_area() pointer in struct file_operations alone.  The main
purpose of this change is to reorganize in preparation for future changes,
but it also converts the calls of mm->get_unmapped_area() from indirect
branches into a direct ones.

The stress-ng bigheap benchmark calls realloc a lot, which calls through
get_unmapped_area() in the kernel.  On x86, the change yielded a ~1%
improvement there on a retpoline config.

In testing a few x86 configs, removing the pointer unfortunately didn't
result in any actual size reductions in the compiled layout of mm_struct.
But depending on compiler or arch alignment requirements, the change could
shrink the size of mm_struct.

Link: https://lkml.kernel.org/r/20240326021656.202649-3-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Mark Brown <broonie@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

proc: refactor pde_get_unmapped_area as prep

Patch series "Cover a guard gap corner case", v4.

In working on x86’s shadow stack feature, I came across some limitations
around the kernel’s handling of guard gaps.  AFAICT these limitations
are not too important for the traditional stack usage of guard gaps, but
have bigger impact on shadow stack’s usage.  And now in addition to x86,
we have two other architectures implementing shadow stack like features
that plan to use guard gaps.  I wanted to see about addressing them, but I
have not worked on mmap() placement related code before, so would greatly
appreciate if people could take a look and point me in the right
direction.

The nature of the limitations of concern is as follows. In order to ensure
guard gaps between mappings, mmap() would need to consider two things:
 1. That the new mapping isn’t placed in an any existing mapping’s guard
    gap.
 2. That the new mapping isn’t placed such that any existing mappings are
    not in *its* guard gaps
Currently mmap never considers (2), and (1) is not considered in some
situations.

When not passing an address hint, or passing one without
MAP_FIXED_NOREPLACE, (1) is enforced.  With MAP_FIXED_NOREPLACE, (1) is
not enforced.  With MAP_FIXED, (1) is not considered, but this seems to be
expected since MAP_FIXED can already clobber existing mappings.  For
MAP_FIXED_NOREPLACE I would have guessed it should respect the guard gaps
of existing mappings, but it is probably a little ambiguous.

In this series I just tried to add enforcement of (2) for the normal (no
address hint) case and only for the newer shadow stack memory (not
stacks).  The reason is that with the no-address-hint situation, landing
next to a guard gap could come up naturally and so be more influencable by
attackers such that two shadow stacks could be adjacent without a guard
gap.  Where as the address-hint scenarios would require more control -
being able to call mmap() with specific arguments.  As for why not just
fix the other corner cases anyway, I thought it might have some greater
possibility of affecting existing apps.

This patch (of 14):

Future changes will perform a treewide change to remove the indirect
branch that is involved in calling mm->get_unmapped_area().  After doing
this, the function will no longer be able to be handled as a function
pointer.  To make the treewide change diff cleaner and easier to review,
refactor pde_get_unmapped_area() such that mm->get_unmapped_area() is
called without being stored in a local function pointer.  With this in
refactoring, follow on changes will be able to simply replace the call
site with a future function that calls it directly.

Link: https://lkml.kernel.org/r/20240326021656.202649-1-rick.p.edgecombe@intel.com
Link: https://lkml.kernel.org/r/20240326021656.202649-2-rick.p.edgecombe@intel.com
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Deepak Gupta <debug@rivosinc.com>
Cc: H. Peter Anvin (Intel) <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

userfaultfd: early return in dup_userfaultfd()

When vma->vm_userfaultfd_ctx.ctx is NULL, vma->vm_flags should have
cleared __VM_UFFD_FLAGS. Therefore, there is no need to down_write or
clear the flag, which will affect fork performance. Fix this by
returning early if octx is NULL in dup_userfaultfd().

By applying this patch we can get a 1.3% performance improvement for
lmbench fork_prot. Results are as follows:
                   base      early return
Process fork+exit: 419.1106  413.4804

Link: https://lkml.kernel.org/r/20240327090835.3232629-1-zhangpeng362@huawei.com
Signed-off-by: ZhangPeng <zhangpeng362@huawei.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Lokesh Gidra <lokeshgidra@google.com>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove __set_page_dirty_nobuffers()

There are no more callers of __set_page_dirty_nobuffers(), remove it.

Link: https://lkml.kernel.org/r/20240327143008.3739435-1-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: remove "prot" parameter from move_pte()

The "prot" parameter is unused, and using it instead of what's stored in
that particular PTE would very likely be wrong.  Let's simply remove it.

Link: https://lkml.kernel.org/r/20240327143301.741807-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andreas Larsson <andreas@gaisler.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: optimize CONFIG_PER_VMA_LOCK member placement in vm_area_struct

Currently, we end up wasting some memory in each vm_area_struct. Pahole
states that:
[...]
int                        vm_lock_seq;          /*    40     4 */

/* XXX 4 bytes hole, try to pack */

struct vma_lock *          vm_lock;              /*    48     8 */
bool                       detached;             /*    56     1 */

/* XXX 7 bytes hole, try to pack */
[...]

Let's reduce the holes and memory wastage by moving the bool:
[...]
bool                       detached;             /*    40     1 */

/* XXX 3 bytes hole, try to pack */

int                        vm_lock_seq;          /*    44     4 */
struct vma_lock *          vm_lock;              /*    48     8 */
[...]

Effectively shrinking the vm_area_struct with CONFIG_PER_VMA_LOCK by
8 byte.

Likely, we could place "detached" in the lowest bit of vm_lock, but at
least on 64bit that won't really make a difference, so keep it simple.

Link: https://lkml.kernel.org/r/20240327143548.744070-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

filemap: remove __set_page_dirty()

All callers have been converted to use folios; remove this wrapper.

Link: https://lkml.kernel.org/r/20240327185447.1076689-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: use rwsem assertion macros for mmap_lock

This slightly strengthens our write assertion when lockdep is disabled.
It also downgrades us from BUG_ON to WARN_ON, but I think that's an
improvement.  I don't think dumping the mm_struct was all that valuable;
the call chain is what's important.

Link: https://lkml.kernel.org/r/20240327190701.1082560-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: allow anon exclusive check over hugetlb tail pages

PageAnonExclusive() used to forbid tail pages for hugetlbfs, as that used
to be called mostly in hugetlb specific paths and the head page was
guaranteed.

As we move forward towards merging hugetlb paths into generic mm, we may
start to pass in tail hugetlb pages (when with cont-pte/cont-pmd huge
pages) for such check.  Allow it to properly fetch the head, in which case
the anon-exclusiveness of the head will always represents the tail page.

There's already a sign of it when we look at the GUP-fast which already
contain the hugetlb processing altogether: we used to have a specific
commit 5805192c7b72 ("mm/gup: handle cont-PTE hugetlb pages correctly in
gup_must_unshare() via GUP-fast") covering that area.  Now with this more
generic change, that can also go away.

[akpm@linux-foundation.org: simplify PageAnonExclusive(), per Matthew]
Link: https://lkml.kernel.org/r/Zg3u5Sh9EbbYPhaI@casper.infradead.org
Link: https://lkml.kernel.org/r/20240403013249.1418299-2-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: WANG Xuerui <kernel@xen0n.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/gup: handle hugetlb in the generic follow_page_mask code

Now follow_page() is ready to handle hugetlb pages in whatever form, and
over all architectures.  Switch to the generic code path.

Time to retire hugetlb_follow_page_mask(), following the previous
retirement of follow_hugetlb_page() in 4849807114b8.

There may be a slight difference of how the loops run when processing slow
GUP over a large hugetlb range on cont_pte/cont_pmd supported archs: each
loop of __get_user_pages() will resolve one pgtable entry with the patch
applied, rather than relying on the size of hugetlb hstate, the latter may
cover multiple entries in one loop.

A quick performance test on an aarch64 VM on M1 chip shows 15% degrade
over a tight loop of slow gup after the path switched.  That shouldn't be
a problem because slow-gup should not be a hot path for GUP in general:
when page is commonly present, fast-gup will already succeed, while when
the page is indeed missing and require a follow up page fault, the slow
gup degrade will probably buried in the fault paths anyway.  It also
explains why slow gup for THP used to be very slow before 57edfcfd3419
("mm/gup: accelerate thp gup even for "pages != NULL"") lands, the latter
not part of a performance analysis but a side benefit.  If the performance
will be a concern, we can consider handle CONT_PTE in follow_page().

Before that is justified to be necessary, keep everything clean and simple.

Link: https://lkml.kernel.org/r/20240327152332.950956-14-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Tested-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrew Jones <andrew.jones@linux.dev>
Cc: Aneesh Kumar K.V (IBM) <aneesh.kumar@kernel.org>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Mike Rapoport (IBM)" <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Rik van Riel <riel@surriel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>