kfence: limit currently covered allocations when pool nearly full

One of KFENCE's main design principles is that with increasing uptime,
allocation coverage increases sufficiently to detect previously
undetected bugs.

We have observed that frequent long-lived allocations of the same source
(e.g.  pagecache) tend to permanently fill up the KFENCE pool with
increasing system uptime, thus breaking the above requirement.  The
workaround thus far had been increasing the sample interval and/or
increasing the KFENCE pool size, but is no reliable solution.

To ensure diverse coverage of allocations, limit currently covered
allocations of the same source once pool utilization reaches 75%
(configurable via `kfence.skip_covered_thresh`) or above.  The effect is
retaining reasonable allocation coverage when the pool is close to full.

A side-effect is that this also limits frequent long-lived allocations
of the same source filling up the pool permanently.

Uniqueness of an allocation for coverage purposes is based on its
(partial) allocation stack trace (the source).  A Counting Bloom filter
is used to check if an allocation is covered; if the allocation is
currently covered, the allocation is skipped by KFENCE.

Testing was done using:

(a) a synthetic workload that performs frequent long-lived
    allocations (default config values; sample_interval=1;
    num_objects=63), and

(b) normal desktop workloads on an otherwise idle machine where
    the problem was first reported after a few days of uptime
    (default config values).

In both test cases the sampled allocation rate no longer drops to zero
at any point.  In the case of (b) we observe (after 2 days uptime) 15%
unique allocations in the pool, 77% pool utilization, with 20% "skipped
allocations (covered)".

[elver@google.com: simplify and just use hash_32(), use more random stack_hash_seed]
Link: https://lkml.kernel.org/r/YU3MRGaCaJiYht5g@elver.google.com
[elver@google.com: fix 32 bit]

Link: https://lkml.kernel.org/r/20210923104803.2620285-4-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Alexander Potapenko <glider@google.com>
Cc: Aleksandr Nogikh <nogikh@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Taras Madan <tarasmadan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kfence: move saving stack trace of allocations into __kfence_alloc()

Move the saving of the stack trace of allocations into __kfence_alloc(),
so that the stack entries array can be used outside of
kfence_guarded_alloc() and we avoid potentially unwinding the stack
multiple times.

Link: https://lkml.kernel.org/r/20210923104803.2620285-3-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Alexander Potapenko <glider@google.com>
Cc: Aleksandr Nogikh <nogikh@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Taras Madan <tarasmadan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kfence: count unexpectedly skipped allocations

Maintain a counter to count allocations that are skipped due to being
incompatible (oversized, incompatible gfp flags) or no capacity.

This is to compute the fraction of allocations that could not be
serviced by KFENCE, which we expect to be rare.

Link: https://lkml.kernel.org/r/20210923104803.2620285-2-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Alexander Potapenko <glider@google.com>
Cc: Aleksandr Nogikh <nogikh@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Taras Madan <tarasmadan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

stacktrace: move filter_irq_stacks() to kernel/stacktrace.c

filter_irq_stacks() has little to do with the stackdepot implementation,
except that it is usually used by users (such as KASAN) of stackdepot to
reduce the stack trace.

However, filter_irq_stacks() itself is not useful without a stack trace
as obtained by stack_trace_save() and friends.

Therefore, move filter_irq_stacks() to kernel/stacktrace.c, so that new
users of filter_irq_stacks() do not have to start depending on
STACKDEPOT only for filter_irq_stacks().

Link: https://lkml.kernel.org/r/20210923104803.2620285-1-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Aleksandr Nogikh <nogikh@google.com>
Cc: Taras Madan <tarasmadan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/mm.h: move nr_free_buffer_pages from swap.h to mm.h

nr_free_buffer_pages could be exposed through mm.h instead of swap.h.
The advantage of this change is that it can reduce the obsolete
includes.  For example, net/ipv4/tcp.c wouldn't need swap.h any more
since it has already included mm.h.  Similarly, after checking all the
other files, it comes that tcp.c, udp.c meter.c ,...  follow the same
rule, so these files can have swap.h removed too.

Moreover, after preprocessing all the files that use
nr_free_buffer_pages, it turns out that those files have already
included mm.h.Thus, we can move nr_free_buffer_pages from swap.h to mm.h
safely.  This change will not affect the compilation of other files.

Link: https://lkml.kernel.org/r/20210912133640.1624-1-liumh1@shanghaitech.edu.cn
Signed-off-by: Mianhan Liu <liumh1@shanghaitech.edu.cn>
Cc: Jakub Kicinski <kuba@kernel.org>
CC: Ulf Hansson <ulf.hansson@linaro.org>
Cc: "David S . Miller" <davem@davemloft.net>
Cc: Simon Horman <horms@verge.net.au>
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Vlad Yasevich <vyasevich@gmail.com>
Cc: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove HARDENED_USERCOPY_FALLBACK

This has served its purpose and is no longer used.  All usercopy
violations appear to have been handled by now, any remaining instances
(or new bugs) will cause copies to be rejected.

This isn't a direct revert of commit 2d891fbc3bb6 ("usercopy: Allow
strict enforcement of whitelists"); since usercopy_fallback is
effectively 0, the fallback handling is removed too.

This also removes the usercopy_fallback module parameter on slab_common.

Link: https://github.com/KSPP/linux/issues/153
Link: https://lkml.kernel.org/r/20210921061149.1091163-1-steve@sk2.org
Signed-off-by: Stephen Kitt <steve@sk2.org>
Suggested-by: Kees Cook <keescook@chromium.org>
Acked-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Joel Stanley <joel@jms.id.au> [defconfig change]
Acked-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: James Morris <jmorris@namei.org>
Cc: "Serge E . Hallyn" <serge@hallyn.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: introduce an aged idle interface

This change introduces an aged idle interface to the existing idle sysfs
file for zram.

When CONFIG_ZRAM_MEMORY_TRACKING is enabled the idle file now also
accepts an integer argument.  This integer is the age (in seconds) of
pages to mark as idle.  The idle file still supports 'all' as it always
has.  This new approach allows for much more control over which pages
get marked as idle.

[bgeffon@google.com: use IS_ENABLED and cleanup comment]
Link: https://lkml.kernel.org/r/20210924161128.1508015-1-bgeffon@google.com
[bgeffon@google.com: Sergey's cleanup suggestions]
Link: https://lkml.kernel.org/r/20210929143056.13067-1-bgeffon@google.com
Link: https://lkml.kernel.org/r/20210923130115.1344361-1-bgeffon@google.com
Signed-off-by: Brian Geffon <bgeffon@google.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Jesse Barnes <jsbarnes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: off by one in read_block_state()

snprintf() returns the number of bytes it would have printed if there
were space.  But it does not count the NUL terminator.  So that means
that if "count == copied" then this has already overflowed by one
character.

This bug likely isn't super harmful in real life.

Link: https://lkml.kernel.org/r/20210916130404.GA25094@kili
Fixes: c0265342bff4 ("zram: introduce zram memory tracking")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram_drv: allow reclaim on bio_alloc

The read_from_bdev_async is not called on atomic context.  So GFP_NOIO
is available rather than GFP_ATOMIC.  If there were reclaimable pages
with GFP_NOIO, we can avoid allocation failure and page fault failure.

Link: https://lkml.kernel.org/r/20210908005241.28062-1-jaewon31.kim@samsung.com
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Reported-by: Yong-Taek Lee <ytk.lee@samsung.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/highmem: remove deprecated kmap_atomic

kmap_atomic() is being deprecated in favor of kmap_local_page().

Replace the uses of kmap_atomic() within the highmem code.

On profiling clear_huge_page() using ftrace an improvement of 62% was
observed on the below setup.

Setup:-
Below data has been collected on Qualcomm's SM7250 SoC THP enabled
(kernel v4.19.113) with only CPU-0(Cortex-A55) and CPU-7(Cortex-A76)
switched on and set to max frequency, also DDR set to perf governor.

FTRACE Data:-

Base data:-
Number of iterations: 48
Mean of allocation time: 349.5 us
std deviation: 74.5 us

v4 data:-
Number of iterations: 48
Mean of allocation time: 131 us
std deviation: 32.7 us

The following simple userspace experiment to allocate
100MB(BUF_SZ) of pages and writing to it gave us a good insight,
we observed an improvement of 42% in allocation and writing timings.
-------------------------------------------------------------
Test code snippet
-------------------------------------------------------------
      clock_start();
      buf = malloc(BUF_SZ); /* Allocate 100 MB of memory */

        for(i=0; i < BUF_SZ_PAGES; i++)
        {
                *((int *)(buf + (i*PAGE_SIZE))) = 1;
        }
      clock_end();
-------------------------------------------------------------

Malloc test timings for 100MB anon allocation:-

Base data:-
Number of iterations: 100
Mean of allocation time: 31831 us
std deviation: 4286 us

v4 data:-
Number of iterations: 100
Mean of allocation time: 18193 us
std deviation: 4915 us

[willy@infradead.org: fix zero_user_segments()]
Link: https://lkml.kernel.org/r/YYVhHCJcm2DM2G9u@casper.infradead.org
Link: https://lkml.kernel.org/r/20210204073255.20769-2-prathu.baronia@oneplus.com
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Prathu Baronia <prathu.baronia@oneplus.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zsmalloc.c: close race window between zs_pool_dec_isolated() and zs_unregister_migration()

There is one possible race window between zs_pool_dec_isolated() and
zs_unregister_migration() because wait_for_isolated_drain() checks the
isolated count without holding class->lock and there is no order inside
zs_pool_dec_isolated().  Thus the below race window could be possible:

  zs_pool_dec_isolated zs_unregister_migration
    check pool->destroying != 0
  pool->destroying = true;
  smp_mb();
  wait_for_isolated_drain()
    wait for pool->isolated_pages == 0
    atomic_long_dec(&pool->isolated_pages);
    atomic_long_read(&pool->isolated_pages) == 0

Since we observe the pool->destroying (false) before atomic_long_dec()
for pool->isolated_pages, waking pool->migration_wait up is missed.

Fix this by ensure checking pool->destroying happens after the
atomic_long_dec(&pool->isolated_pages).

Link: https://lkml.kernel.org/r/20210708115027.7557-1-linmiaohe@huawei.com
Fixes: 701d678599d0 ("mm/zsmalloc.c: fix race condition in zs_destroy_pool")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>
Cc: Henry Burns <henryburns@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/rmap.c: avoid double faults migrating device private pages

During migration special page table entries are installed for each page
being migrated.  These entries store the pfn and associated permissions
of ptes mapping the page being migarted.

Device-private pages use special swap pte entries to distinguish
read-only vs.  writeable pages which the migration code checks when
creating migration entries.  Normally this follows a fast path in
migrate_vma_collect_pmd() which correctly copies the permissions of
device-private pages over to migration entries when migrating pages back
to the CPU.

However the slow-path falls back to using try_to_migrate() which
unconditionally creates read-only migration entries for device-private
pages.  This leads to unnecessary double faults on the CPU as the new
pages are always mapped read-only even when they could be mapped
writeable.  Fix this by correctly copying device-private permissions in
try_to_migrate_one().

Link: https://lkml.kernel.org/r/20211018045247.3128058-1-apopple@nvidia.com
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Ralph Campbell <rcampbell@nvidia.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: indicate MEMBLOCK_DRIVER_MANAGED with IORESOURCE_SYSRAM_DRIVER_MANAGED

Let's communicate driver-managed regions to memblock, to properly teach
kexec_file with CONFIG_ARCH_KEEP_MEMBLOCK to not place images on these
memory regions.

Link: https://lkml.kernel.org/r/20211004093605.5830-6-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jianyong Wu <Jianyong.Wu@arm.com>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Shahab Vahedi <shahab@synopsys.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vineet Gupta <vgupta@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: add MEMBLOCK_DRIVER_MANAGED to mimic IORESOURCE_SYSRAM_DRIVER_MANAGED

Let's add a flag that corresponds to IORESOURCE_SYSRAM_DRIVER_MANAGED,
indicating that we're dealing with a memory region that is never
indicated in the firmware-provided memory map, but always detected and
added by a driver.

Similar to MEMBLOCK_HOTPLUG, most infrastructure has to treat such
memory regions like ordinary MEMBLOCK_NONE memory regions -- for
example, when selecting memory regions to add to the vmcore for dumping
in the crashkernel via for_each_mem_range().

However, especially kexec_file is not supposed to select such memblocks
via for_each_free_mem_range() / for_each_free_mem_range_reverse() to
place kexec images, similar to how we handle
IORESOURCE_SYSRAM_DRIVER_MANAGED without CONFIG_ARCH_KEEP_MEMBLOCK.

We'll make sure that memory hotplug code sets the flag where applicable
(IORESOURCE_SYSRAM_DRIVER_MANAGED) next.  This prepares architectures
that need CONFIG_ARCH_KEEP_MEMBLOCK, such as arm64, for virtio-mem
support.

Note that kexec *must not* indicate this memory to the second kernel and
*must not* place kexec-images on this memory.  Let's add a comment to
kexec_walk_memblock(), documenting how we handle MEMBLOCK_DRIVER_MANAGED
now just like using IORESOURCE_SYSRAM_DRIVER_MANAGED in
locate_mem_hole_callback() for kexec_walk_resources().

Also note that MEMBLOCK_HOTPLUG cannot be reused due to different
semantics:
MEMBLOCK_HOTPLUG: memory is indicated as "System RAM" in the
firmware-provided memory map and added to the system early during
boot; kexec *has to* indicate this memory to the second kernel and
can place kexec-images on this memory. After memory hotunplug,
kexec has to be re-armed. We mostly ignore this flag when
"movable_node" is not set on the kernel command line, because
then we're told to not care about hotunpluggability of such
memory regions.

MEMBLOCK_DRIVER_MANAGED: memory is not indicated as "System RAM" in
the firmware-provided memory map; this memory is always detected
and added to the system by a driver; memory might not actually be
physically hotunpluggable. kexec *must not* indicate this memory to
the second kernel and *must not* place kexec-images on this memory.

Link: https://lkml.kernel.org/r/20211004093605.5830-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jianyong Wu <Jianyong.Wu@arm.com>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Shahab Vahedi <shahab@synopsys.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vineet Gupta <vgupta@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: allow to specify flags with memblock_add_node()

We want to specify flags when hotplugging memory.  Let's prepare to pass
flags to memblock_add_node() by adjusting all existing users.

Note that when hotplugging memory the system is already up and running
and we might have concurrent memblock users: for example, while we're
hotplugging memory, kexec_file code might search for suitable memory
regions to place kexec images.  It's important to add the memory
directly to memblock via a single call with the right flags, instead of
adding the memory first and apply flags later: otherwise, concurrent
memblock users might temporarily stumble over memblocks with wrong
flags, which will be important in a follow-up patch that introduces a
new flag to properly handle add_memory_driver_managed().

Link: https://lkml.kernel.org/r/20211004093605.5830-4-david@redhat.com
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Shahab Vahedi <shahab@synopsys.com> [arch/arc]
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jianyong Wu <Jianyong.Wu@arm.com>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vineet Gupta <vgupta@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: improve MEMBLOCK_HOTPLUG documentation

The description of MEMBLOCK_HOTPLUG is currently short and consequently
misleading: we're actually dealing with a memory region that might get
hotunplugged later (i.e., the platform+firmware supports it), yet it is
indicated in the firmware-provided memory map as system ram that will
just get used by the system for any purpose when not taking special
care.  The firmware marked this memory region as a hot(un)plugged (e.g.,
hotplugged before reboot), implying that it might get hotunplugged again
later.

Whether we consider this information depends on the "movable_node"
kernel commandline parameter: only with "movable_node" set, we'll try
keeping this memory hotunpluggable, for example, by not serving early
allocations from this memory region and by letting the buddy manage it
using the ZONE_MOVABLE.

Let's make this clearer by extending the documentation.

Note: kexec *has to* indicate this memory to the second kernel.  With
"movable_node" set, we don't want to place kexec-images on this memory.
Without "movable_node" set, we don't care and can place kexec-images on
this memory.  In both cases, after successful memory hotunplug, kexec
has to be re-armed to update the memory map for the second kernel and to
place the kexec-images somewhere else.

Link: https://lkml.kernel.org/r/20211004093605.5830-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jianyong Wu <Jianyong.Wu@arm.com>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Shahab Vahedi <shahab@synopsys.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vineet Gupta <vgupta@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: handle memblock_add_node() failures in add_memory_resource()

Patch series "mm/memory_hotplug: full support for add_memory_driver_managed() with CONFIG_ARCH_KEEP_MEMBLOCK", v2.

Architectures that require CONFIG_ARCH_KEEP_MEMBLOCK=y, such as arm64,
don't cleanly support add_memory_driver_managed() yet.  Most
prominently, kexec_file can still end up placing kexec images on such
driver-managed memory, resulting in undesired behavior, for example,
having kexec images located on memory not part of the firmware-provided
memory map.

Teaching kexec to not place images on driver-managed memory is
especially relevant for virtio-mem.  Details can be found in commit
7b7b27214bba ("mm/memory_hotplug: introduce
add_memory_driver_managed()").

Extend memblock with a new flag and set it from memory hotplug code when
applicable.  This is required to fully support virtio-mem on arm64,
making also kexec_file behave like on x86-64.

This patch (of 2):

If memblock_add_node() fails, we're most probably running out of memory.
While this is unlikely to happen, it can happen and having memory added
without a memblock can be problematic for architectures that use
memblock to detect valid memory.  Let's fail in a nice way instead of
silently ignoring the error.

Link: https://lkml.kernel.org/r/20211004093605.5830-1-david@redhat.com
Link: https://lkml.kernel.org/r/20211004093605.5830-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Jianyong Wu <Jianyong.Wu@arm.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Vineet Gupta <vgupta@kernel.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Shahab Vahedi <shahab@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86: remove memory hotplug support on X86_32

CONFIG_MEMORY_HOTPLUG was marked BROKEN over one year and we just
restricted it to 64 bit.  Let's remove the unused x86 32bit
implementation and simplify the Kconfig.

Link: https://lkml.kernel.org/r/20210929143600.49379-7-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Alex Shi <alexs@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: remove stale function declarations

These functions no longer exist.

Link: https://lkml.kernel.org/r/20210929143600.49379-6-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Alex Shi <alexs@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: remove HIGHMEM leftovers

We don't support CONFIG_MEMORY_HOTPLUG on 32 bit and consequently not
HIGHMEM.  Let's remove any leftover code -- including the unused
"status_change_nid_high" field part of the memory notifier.

Link: https://lkml.kernel.org/r/20210929143600.49379-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Alex Shi <alexs@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: restrict CONFIG_MEMORY_HOTPLUG to 64 bit

32 bit support is broken in various ways: for example, we can online
memory that should actually go to ZONE_HIGHMEM to ZONE_MOVABLE or in
some cases even to one of the other kernel zones.

We marked it BROKEN in commit b59d02ed0869 ("mm/memory_hotplug: disable
the functionality for 32b") almost one year ago.  According to that
commit it might be broken at least since 2017.  Further, there is hardly
a sane use case nowadays.

Let's just depend completely on 64bit, dropping the "BROKEN" dependency
to make clear that we are not going to support it again.  Next, we'll
remove some HIGHMEM leftovers from memory hotplug code to clean up.

Link: https://lkml.kernel.org/r/20210929143600.49379-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Alex Shi <alexs@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: remove CONFIG_MEMORY_HOTPLUG_SPARSE

CONFIG_MEMORY_HOTPLUG depends on CONFIG_SPARSEMEM, so there is no need for
CONFIG_MEMORY_HOTPLUG_SPARSE anymore; adjust all instances to use
CONFIG_MEMORY_HOTPLUG and remove CONFIG_MEMORY_HOTPLUG_SPARSE.

Link: https://lkml.kernel.org/r/20210929143600.49379-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Shuah Khan <skhan@linuxfoundation.org> [kselftest]
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Oscar Salvador <osalvador@suse.de>
Cc: Alex Shi <alexs@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: remove CONFIG_X86_64_ACPI_NUMA dependency from CONFIG_MEMORY_HOTPLUG

Patch series "mm/memory_hotplug: Kconfig and 32 bit cleanups".

Some cleanups around CONFIG_MEMORY_HOTPLUG, including removing 32 bit
leftovers of memory hotplug support.

This patch (of 6):

SPARSEMEM is the only possible memory model for x86-64, FLATMEM is not
possible:

config ARCH_FLATMEM_ENABLE
def_bool y
depends on X86_32 && !NUMA

And X86_64_ACPI_NUMA (obviously) only supports x86-64:

config X86_64_ACPI_NUMA
def_bool y
depends on X86_64 && NUMA && ACPI && PCI

Let's just remove the CONFIG_X86_64_ACPI_NUMA dependency, as it does no
longer make sense.

Link: https://lkml.kernel.org/r/20210929143600.49379-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Alex Shi <alexs@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug.rst: document the "auto-movable" online policy

Commit e83a437faa62 ("mm/memory_hotplug: introduce "auto-movable" online
policy") introduced a new memory online policy to automatically select a
zone for memory blocks to be onlined.  It added a way to set the active
online policy and tunables for the auto-movable online policy.

Follow-up commits tweaked the "auto-movable" policy to also consider
memory device details when selecting zones for memory blocks to be
onlined.

Let's document the new toggles and how the two online policies we have
work.

[david@redhat.com: updates]
Link: https://lkml.kernel.org/r/20211011082058.6076-4-david@redhat.com
Link: https://lkml.kernel.org/r/20210930144117.23641-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug.rst: fix wrong /sys/module/memory_hotplug/parameters/ path

We accidentially added a superfluous "s".

Link: https://lkml.kernel.org/r/20210930144117.23641-3-david@redhat.com
Fixes: ac3332c44767 ("memory-hotplug.rst: complete admin-guide overhaul")
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug.rst: fix two instances of "movablecore" that should be "movable_node"

Patch series "memory-hotplug.rst: document the "auto-movable" online
policy".

Now that the memory-hotplug.rst overhaul is upstream, proper
documentation for the "auto-movable" online policy, documenting all new
toggles and options.  Along, two fixes for the original overhaul.

This patch (of 3):

We really want to refer to the "movable_node" kernel command line
parameter here.

Link: https://lkml.kernel.org/r/20210930144117.23641-2-david@redhat.com
Fixes: ac3332c44767 ("memory-hotplug.rst: complete admin-guide overhaul")
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: add static qualifier for online_policy_to_str()

online_policy_to_str is only used in memory_hotplug.c and should be
defined as static.

Link: https://lkml.kernel.org/r/20210913024534.26161-1-tangyizhou@huawei.com
Signed-off-by: Tang Yizhou <tangyizhou@huawei.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests/vm: make MADV_POPULATE_(READ|WRITE) use in-tree headers

The madv_populate selftest currently builds with a warning when the
local installed headers (via the distribution) don't include
MADV_POPULATE_READ and MADV_POPULATE_WRITE.  The warning is correct,
because the test cannot locate the necessary header.

The reason is that the in-tree installed headers (usr/include) have a
"linux" instead of a "sys" subdirectory.

Including "linux/mman.h" instead of "sys/mman.h" doesn't work (e.g.,
mmap() and madvise() are not defined that way).  The only thing that
seems to work is including "linux/mman.h" in addition to "sys/mman.h".

We can get rid of our availability check and simplify.

Link: https://lkml.kernel.org/r/20211015165758.41374-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reported-by: Shuah Khan <skhan@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmstat.c: make extfrag_index show more pretty

fragmentation_index may return -1000 and the corresponding formated
value showed by seq_printf will take a negative signatrue, but other
positive formated values don't take a positive signatrue, so the output
becomes unaligned.

before:
  Node 0, zone      DMA -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000
  Node 0, zone    DMA32 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000
  Node 0, zone   Normal -1.000 -1.000 -1.000 -1.000 0.931 0.966 0.983 0.992 0.996 0.998 0.999

after this patch:
  Node 0, zone      DMA -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000
  Node 0, zone    DMA32 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000
  Node 0, zone   Normal -1.000 -1.000 -1.000 -1.000  0.931  0.966  0.983  0.992  0.996  0.998  0.999

Link: https://lkml.kernel.org/r/20211019103241.134797-1-linf@wangsu.com
Signed-off-by: Lin Feng <linf@wangsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmstat: annotate data race for zone->free_area[order].nr_free

KCSAN reports a data-race on v5.10 which also exists on mainline:

  BUG: KCSAN: data-race in extfrag_for_order+0x33/0x2d0

  race at unknown origin, with read to 0xffff9ee9bfffab48 of 8 bytes by task 34 on cpu 1:
   extfrag_for_order+0x33/0x2d0
   kcompactd+0x5f0/0xce0
   kthread+0x1f9/0x220
   ret_from_fork+0x22/0x30

  Reported by Kernel Concurrency Sanitizer on:
  CPU: 1 PID: 34 Comm: kcompactd0 Not tainted 5.10.0+ #2
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014

Access to zone->free_area[order].nr_free in extfrag_for_order() and
frag_show_print() is lockless.  That's intentional and the stats are a
rough estimate anyway.  Annotate them with data_race().

[liushixin2@huawei.com: add comments]
Link: https://lkml.kernel.org/r/20210918084655.2696522-1-liushixin2@huawei.com
Link: https://lkml.kernel.org/r/20210908015606.3999871-1-liushixin2@huawei.com
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests: vm: add KSM huge pages merging time test

Add test case of KSM merging time using mostly huge pages

Link: https://lkml.kernel.org/r/20211013044045.360251-1-pedrodemargomes@gmail.com
Signed-off-by: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Zhansaya Bagdauletkyzy <zhansayabagdaulet@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftest/vm: fix ksm selftest to run with different NUMA topologies

Platforms can have non-contiguous NUMA nodes like below

   #numactl  -H
  available: 2 nodes (0,8)
  .....
  node distances:
  node   0   8
    0:  10  40
    8:  40  10

   #numactl  -H
  available: 1 nodes (1)
  ....
  node distances:
  node   1
    1:  10

Hence update the test to not assume the presence of Node 0 and 1 and
also use numa_num_configured_nodes() instead of numa_max_node for
finding whether to skip the test.

Link: https://lkml.kernel.org/r/20210914141414.350759-1-aneesh.kumar@linux.ibm.com
Fixes: 82e717ad3501 ("selftests: vm: add KSM merging across nodes test")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Zhansaya Bagdauletkyzy <zhansayabagdaulet@gmail.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Tyler Hicks <tyhicks@linux.microsoft.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: nommu: kill arch_get_unmapped_area()

When nommu, the arch_get_unmapped_area() will not be called, just kill
it.

Link: https://lkml.kernel.org/r/20210910061906.36299-1-wangkefeng.wang@huawei.com
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/readahead.c: fix incorrect comments for get_init_ra_size

In fact, formated values returned by get_init_ra_size are not that
intuitive.  This patch make the comments reflect its truth.

Link: https://lkml.kernel.org/r/20211019104812.135602-1-linf@wangsu.com
Signed-off-by: Lin Feng <linf@wangsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: fix incorrect unmap behavior for private pages

When truncating pagecache on file THP, the private pages of a process
should not be unmapped mapping.  This incorrect behavior on a dynamic
shared libraries which will cause related processes to happen core dump.

A simple test for a DSO (Prerequisite is the DSO mapped in file THP):

    int main(int argc, char *argv[])
    {
int fd;

fd = open(argv[1], O_WRONLY);
if (fd < 0) {
perror("open");
}

close(fd);
return 0;
    }

The test only to open a target DSO, and do nothing.  But this operation
will lead one or more process to happen core dump.  This patch mainly to
fix this bug.

Link: https://lkml.kernel.org/r/20211025092134.18562-3-rongwei.wang@linux.alibaba.com
Fixes: eb6ecbed0aa2 ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs")
Signed-off-by: Rongwei Wang <rongwei.wang@linux.alibaba.com>
Tested-by: Xu Yu <xuyu@linux.alibaba.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Song Liu <song@kernel.org>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Collin Fijalkovich <cfijalkovich@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: lock filemap when truncating page cache

Patch series "fix two bugs for file THP".

This patch (of 2):

Transparent huge page has supported read-only non-shmem files.  The
file- backed THP is collapsed by khugepaged and truncated when written
(for shared libraries).

However, there is a race when multiple writers truncate the same page
cache concurrently.

In that case, subpage(s) of file THP can be revealed by find_get_entry
in truncate_inode_pages_range, which will trigger PageTail BUG_ON in
truncate_inode_page, as follows:

    page:000000009e420ff2 refcount:1 mapcount:0 mapping:0000000000000000 index:0x7ff pfn:0x50c3ff
    head:0000000075ff816d order:9 compound_mapcount:0 compound_pincount:0
    flags: 0x37fffe0000010815(locked|uptodate|lru|arch_1|head)
    raw: 37fffe0000000000 fffffe0013108001 dead000000000122 dead000000000400
    raw: 0000000000000001 0000000000000000 00000000ffffffff 0000000000000000
    head: 37fffe0000010815 fffffe001066bd48 ffff000404183c20 0000000000000000
    head: 0000000000000600 0000000000000000 00000001ffffffff ffff000c0345a000
    page dumped because: VM_BUG_ON_PAGE(PageTail(page))
    ------------[ cut here ]------------
    kernel BUG at mm/truncate.c:213!
    Internal error: Oops - BUG: 0 [#1] SMP
    Modules linked in: xfs(E) libcrc32c(E) rfkill(E) ...
    CPU: 14 PID: 11394 Comm: check_madvise_d Kdump: ...
    Hardware name: ECS, BIOS 0.0.0 02/06/2015
    pstate: 60400005 (nZCv daif +PAN -UAO -TCO BTYPE=--)
    Call trace:
     truncate_inode_page+0x64/0x70
     truncate_inode_pages_range+0x550/0x7e4
     truncate_pagecache+0x58/0x80
     do_dentry_open+0x1e4/0x3c0
     vfs_open+0x38/0x44
     do_open+0x1f0/0x310
     path_openat+0x114/0x1dc
     do_filp_open+0x84/0x134
     do_sys_openat2+0xbc/0x164
     __arm64_sys_openat+0x74/0xc0
     el0_svc_common.constprop.0+0x88/0x220
     do_el0_svc+0x30/0xa0
     el0_svc+0x20/0x30
     el0_sync_handler+0x1a4/0x1b0
     el0_sync+0x180/0x1c0
    Code: aa0103e0 900061e1 910ec021 9400d300 (d4210000)

This patch mainly to lock filemap when one enter truncate_pagecache(),
avoiding truncating the same page cache concurrently.

Link: https://lkml.kernel.org/r/20211025092134.18562-1-rongwei.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/20211025092134.18562-2-rongwei.wang@linux.alibaba.com
Fixes: eb6ecbed0aa2 ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs")
Signed-off-by: Xu Yu <xuyu@linux.alibaba.com>
Signed-off-by: Rongwei Wang <rongwei.wang@linux.alibaba.com>
Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Tested-by: Song Liu <song@kernel.org>
Cc: Collin Fijalkovich <cfijalkovich@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests/vm/transhuge-stress: fix ram size thinko

When executing transhuge-stress with an argument to specify the virtual
memory size for testing, the ram size is reported as 0, e.g.

  transhuge-stress 384
  thp-mmap: allocate 192 transhuge pages, using 384 MiB virtual memory and 0 MiB of ram
  thp-mmap: 0.184 s/loop, 0.957 ms/page,   2090.265 MiB/s  192 succeed,    0 failed

This appears to be due to a thinko in commit 0085d61fe05e
("selftests/vm/transhuge-stress: stress test for memory compaction"),
where, at a guess, the intent was to base "xyz MiB of ram" on `ram`
size.

Here are results after using `ram` size:

  thp-mmap: allocate 192 transhuge pages, using 384 MiB virtual memory and 14 MiB of ram

Link: https://lkml.kernel.org/r/20210825135843.29052-1-george_davis@mentor.com
Fixes: 0085d61fe05e ("selftests/vm/transhuge-stress: stress test for memory compaction")
Signed-off-by: George G. Davis <davis.george@siemens.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Eugeniu Rosca <erosca@de.adit-jv.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: migrate: make demotion knob depend on migration

The memory demotion needs to call migrate_pages() to do the jobs.  And
it is controlled by a knob, however, the knob doesn't depend on
CONFIG_MIGRATION.  The knob could be truned on even though MIGRATION is
disabled, this will not cause any crash since migrate_pages() would just
return -ENOSYS.  But it is definitely not optimal to go through demotion
path then retry regular swap every time.

And it doesn't make too much sense to have the knob visible to the users
when !MIGRATION.  Move the related code from mempolicy.[h|c] to
migrate.[h|c].

Link: https://lkml.kernel.org/r/20211015005559.246709-1-shy828301@gmail.com
Signed-off-by: Yang Shi <shy828301@gmail.com>
Acked-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/migrate: de-duplicate migrate_reason strings

In order to remove the need to manually keep three different files in
synch, provide a common definition of the mapping between enum
migrate_reason, and the associated strings for each enum item.

1. Use the tracing system's mapping of enums to strings, by redefining
   and reusing the MIGRATE_REASON and supporting macros, and using that
   to populate the string array in mm/debug.c.

2. Move enum migrate_reason to migrate_mode.h. This is not strictly
   necessary for this patch, but migrate mode and migrate reason go
   together, so this will slightly clarify things.

Link: https://lkml.kernel.org/r/20210922041755.141817-2-jhubbard@nvidia.com
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Reviewed-by: Weizhao Ouyang <o451686892@gmail.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs: extend the definition of hugepages parameter to support node allocation

We can specify the number of hugepages to allocate at boot.  But the
hugepages is balanced in all nodes at present.  In some scenarios, we
only need hugepages in one node.  For example: DPDK needs hugepages
which are in the same node as NIC.

If DPDK needs four hugepages of 1G size in node1 and system has 16 numa
nodes we must reserve 64 hugepages on the kernel cmdline.  But only four
hugepages are used.  The others should be free after boot.  If the
system memory is low(for example: 64G), it will be an impossible task.

So extend the hugepages parameter to support specifying hugepages on a
specific node.  For example add following parameter:

  hugepagesz=1G hugepages=0:1,1:3

It will allocate 1 hugepage in node0 and 3 hugepages in node1.

Link: https://lkml.kernel.org/r/20211005054729.86457-1-yaozhenguo1@gmail.com
Signed-off-by: Zhenguo Yao <yaozhenguo1@gmail.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Zhenguo Yao <yaozhenguo1@gmail.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mark the OOM reaper thread as freezable

The OOM reaper alters user address space which might theoretically alter
the snapshot if reaping is allowed to happen after the freezer quiescent
state.  To this end, the reaper kthread uses wait_event_freezable()
while waiting for any work so that it cannot run while the system
freezes.

However, the current implementation doesn't respect the freezer because
all kernel threads are created with the PF_NOFREEZE flag, so they are
automatically excluded from freezing operations.  This means that the
OOM reaper can race with system snapshotting if it has work to do while
the system is being frozen.

Fix this by adding a set_freezable() call which will clear the
PF_NOFREEZE flag and thus make the OOM reaper visible to the freezer.

Please note that the OOM reaper altering the snapshot this way is mostly
a theoretical concern and has not been observed in practice.

Link: https://lkml.kernel.org/r/20210921165758.6154-1-sultan@kerneltoast.com
Link: https://lkml.kernel.org/r/20210918233920.9174-1-sultan@kerneltoast.com
Fixes: aac453635549 ("mm, oom: introduce oom reaper")
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: use memblock_free for freeing virtual pointers

Rename memblock_free_ptr() to memblock_free() and use memblock_free()
when freeing a virtual pointer so that memblock_free() will be a
counterpart of memblock_alloc()

The callers are updated with the below semantic patch and manual
addition of (void *) casting to pointers that are represented by
unsigned long variables.

    @@
    identifier vaddr;
    expression size;
    @@
    (
    - memblock_phys_free(__pa(vaddr), size);
    + memblock_free(vaddr, size);
    |
    - memblock_free_ptr(vaddr, size);
    + memblock_free(vaddr, size);
    )

[sfr@canb.auug.org.au: fixup]
Link: https://lkml.kernel.org/r/20211018192940.3d1d532f@canb.auug.org.au
Link: https://lkml.kernel.org/r/20210930185031.18648-7-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Juergen Gross <jgross@suse.com>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: rename memblock_free to memblock_phys_free

Since memblock_free() operates on a physical range, make its name
reflect it and rename it to memblock_phys_free(), so it will be a
logical counterpart to memblock_phys_alloc().

The callers are updated with the below semantic patch:

    @@
    expression addr;
    expression size;
    @@
    - memblock_free(addr, size);
    + memblock_phys_free(addr, size);

Link: https://lkml.kernel.org/r/20210930185031.18648-6-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Juergen Gross <jgross@suse.com>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: stop aliasing __memblock_free_late with memblock_free_late

memblock_free_late() is a NOP wrapper for __memblock_free_late(), there
is no point to keep this indirection.

Drop the wrapper and rename __memblock_free_late() to
memblock_free_late().

Link: https://lkml.kernel.org/r/20210930185031.18648-5-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Juergen Gross <jgross@suse.com>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: drop memblock_free_early_nid() and memblock_free_early()

memblock_free_early_nid() is unused and memblock_free_early() is an
alias for memblock_free().

Replace calls to memblock_free_early() with calls to memblock_free() and
remove memblock_free_early() and memblock_free_early_nid().

Link: https://lkml.kernel.org/r/20210930185031.18648-4-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Juergen Gross <jgross@suse.com>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

xen/x86: free_p2m_page: use memblock_free_ptr() to free a virtual pointer

free_p2m_page() wrongly passes a virtual pointer to memblock_free() that
treats it as a physical address.

Call memblock_free_ptr() instead that gets a virtual address to free the
memory.

Link: https://lkml.kernel.org/r/20210930185031.18648-3-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch_numa: simplify numa_distance allocation

Patch series "memblock: cleanup memblock_free interface", v2.

This is the fix for memblock freeing APIs mismatch [1].

The first patch is a cleanup of numa_distance allocation in arch_numa
I've spotted during the conversion.  The second patch is a fix for Xen
memory freeing on some of the error paths.

[1] https://lore.kernel.org/all/CAHk-=wj9k4LZTz+svCxLYs5Y1=+yKrbAUArH1+ghyG3OLd8VVg@mail.gmail.com

This patch (of 6):

Memory allocation of numa_distance uses memblock_phys_alloc_range()
without actual range limits, converts the returned physical address to
virtual and then only uses the virtual address for further
initialization.

Simplify this by replacing memblock_phys_alloc_range() with
memblock_alloc().

Link: https://lkml.kernel.org/r/20210930185031.18648-1-rppt@kernel.org
Link: https://lkml.kernel.org/r/20210930185031.18648-2-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Juergen Gross <jgross@suse.com>
Cc: Shahab Vahedi <Shahab.Vahedi@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page-types.c: print file offset in hexadecimal

In page list mode (with -l and -L option), virtual address and physical
address are printed in hexadecimal, but file offset is not, which is
confusing, so let's align it.

Link: https://lkml.kernel.org/r/20211004061325.1525902-4-naoya.horiguchi@linux.dev
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Bin Wang <wangbin224@huawei.com>
Cc: Changbin Du <changbin.du@intel.com>
Cc: Christian Hansen <chansen3@cisco.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page-types.c: move show_file() to summary output

Currently file info from show_file() is printed out within page list
like below, but this is inconvenient a little to utilize the page list
from other scripts (maybe needs additional filtering).

    $ ./page-types -f page-types.c -l
    foffset offset  len     flags
    page-types.c Inode: 15108680 Size: 30953 (8 pages)
    Modify: Sat Oct  2 23:11:20 2021 (2399 seconds ago)
    Access: Sat Oct  2 23:11:28 2021 (2391 seconds ago)
    0       d9f59e  1       ___U_lA____________________________________
    1       1031eb5 1       __RU_l_____________________________________
    2       13bf717 1       __RU_l_____________________________________
    3       13ac333 1       ___U_lA____________________________________
    4       d9f59f  1       __RU_l_____________________________________
    5       183fd49 1       ___U_lA____________________________________
    6       13cbf69 1       ___U_lA____________________________________
    7       d9ef05  1       ___U_lA____________________________________

                 flags      page-count       MB  symbolic-flags                     long-symbolic-flags
    0x000000000000002c               3        0  __RU_l_____________________________________        referenced,uptodate,lru
    0x0000000000000068               5        0  ___U_lA____________________________________        uptodate,lru,active
                 total               8        0

With this patch file info is printed out in summary part like below:

    $ ./page-types -f page-types.c -l
    foffset offset  len     flags
    0       d9f59e  1       ___U_lA_____________________________________
    1       1031eb5 1       __RU_l______________________________________
    2       13bf717 1       __RU_l______________________________________
    3       13ac333 1       ___U_lA_____________________________________
    4       d9f59f  1       __RU_l______________________________________
    5       183fd49 1       ___U_lA_____________________________________
    6       13cbf69 1       ___U_lA_____________________________________

    page-types.c Inode: 15108680 Size: 30953 (8 pages)
    Modify: Sat Oct  2 23:11:20 2021 (2435 seconds ago)
    Access: Sat Oct  2 23:11:28 2021 (2427 seconds ago)

                 flags      page-count       MB  symbolic-flags                     long-symbolic-flags
    0x000000000000002c               3        0  __RU_l______________________________________       referenced,uptodate,lru
    0x0000000000000068               4        0  ___U_lA_____________________________________       uptodate,lru,active
                 total               7        0

Link: https://lkml.kernel.org/r/20211004061325.1525902-3-naoya.horiguchi@linux.dev
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Bin Wang <wangbin224@huawei.com>
Cc: Changbin Du <changbin.du@intel.com>
Cc: Christian Hansen <chansen3@cisco.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page-types.c: make walk_file() aware of address range option

Patch series "tools/vm/page-types.c: a few improvements".

This patchset adds some improvements on tools/vm/page-types.c.  Patch
1/3 makes -a option (specify address range) work with -f (file cache
mode).  Patch 2/3 and 3/3 are to fix minor formatting issues of this
tool.  These would make life a little easier for the users of this tool.

Please see individual patches for more details about specific issues.

This patch (of 3):

-a|--addr option is used to limit the range of address to be scanned for
page status.  It works now for physical address space (dafult mode) or for
virtual address space (with -p option), but not for file address space
(with -f option).  So make walk_file() aware of -a option.

Link: https://lkml.kernel.org/r/20211004061325.1525902-1-naoya.horiguchi@linux.dev
Link: https://lkml.kernel.org/r/20211004061325.1525902-2-naoya.horiguchi@linux.dev
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Christian Hansen <chansen3@cisco.com>
Cc: Changbin Du <changbin.du@intel.com>
Cc: Bin Wang <wangbin224@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/vm/page_owner_sort.c: count and sort by mem

When viewing page owner information, we may be more concerned about the
total memory rather than the times of stack appears.  Therefore, the
following adjustments are made:

1. Added the statistics on the total number of pages.

2. Added the optional parameter "-m" to configure the program to sort by
   memory (total pages).

The general output of page_owner is as follows:

Page allocated via order XXX, ...
PFN XXX ...
 // Detailed stack

Page allocated via order XXX, ...
PFN XXX ...
 // Detailed stack

The original page_owner_sort ignores PFN rows, puts the remaining rows
in buf, counts the times of buf, and finally sorts them according to the
times.  General output:

XXX times:
Page allocated via order XXX, ...
 // Detailed stack

Now, we use regexp to extract the page order value from the buf, and
count the total pages for the buf.  General output:

XXX times, XXX pages:
Page allocated via order XXX, ...
 // Detailed stack

By default, it is still sorted by the times of buf; If you want to sort
by the pages nums of buf, use the new -m parameter.

Link: https://lkml.kernel.org/r/1631678242-41033-1-git-send-email-weizhenliang@huawei.com
Signed-off-by: Zhenliang Wei <weizhenliang@huawei.com>
Cc: Tang Bin <tangbin@cmss.chinamobile.com>
Cc: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Cc: Zhenliang Wei <weizhenliang@huawei.com>
Cc: Xiaoming Ni <nixiaoming@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmpressure: fix data-race with memcg->socket_pressure

When reading memcg->socket_pressure in mem_cgroup_under_socket_pressure()
and writing memcg->socket_pressure in vmpressure() at the same time, the
following data-race occurs:

  BUG: KCSAN: data-race in __sk_mem_reduce_allocated / vmpressure

  write to 0xffff8881286f4938 of 8 bytes by task 24550 on cpu 3:
   vmpressure+0x218/0x230 mm/vmpressure.c:307
   shrink_node_memcgs+0x2b9/0x410 mm/vmscan.c:2658
   shrink_node+0x9d2/0x11d0 mm/vmscan.c:2769
   shrink_zones+0x29f/0x470 mm/vmscan.c:2972
   do_try_to_free_pages+0x193/0x6e0 mm/vmscan.c:3027
   try_to_free_mem_cgroup_pages+0x1c0/0x3f0 mm/vmscan.c:3345
   reclaim_high mm/memcontrol.c:2440 [inline]
   mem_cgroup_handle_over_high+0x18b/0x4d0 mm/memcontrol.c:2624
   tracehook_notify_resume include/linux/tracehook.h:197 [inline]
   exit_to_user_mode_loop kernel/entry/common.c:164 [inline]
   exit_to_user_mode_prepare+0x110/0x170 kernel/entry/common.c:191
   syscall_exit_to_user_mode+0x16/0x30 kernel/entry/common.c:266
   ret_from_fork+0x15/0x30 arch/x86/entry/entry_64.S:289

  read to 0xffff8881286f4938 of 8 bytes by interrupt on cpu 1:
   mem_cgroup_under_socket_pressure include/linux/memcontrol.h:1483 [inline]
   sk_under_memory_pressure include/net/sock.h:1314 [inline]
   __sk_mem_reduce_allocated+0x1d2/0x270 net/core/sock.c:2696
   __sk_mem_reclaim+0x44/0x50 net/core/sock.c:2711
   sk_mem_reclaim include/net/sock.h:1490 [inline]
   ......
   net_rx_action+0x17a/0x480 net/core/dev.c:6864
   __do_softirq+0x12c/0x2af kernel/softirq.c:298
   run_ksoftirqd+0x13/0x20 kernel/softirq.c:653
   smpboot_thread_fn+0x33f/0x510 kernel/smpboot.c:165
   kthread+0x1fc/0x220 kernel/kthread.c:292
   ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:296

Fix it by using READ_ONCE() and WRITE_ONCE() to read and write
memcg->socket_pressure.

Link: https://lkml.kernel.org/r/20211025082843.671690-1-songyuanzheng@huawei.com
Signed-off-by: Yuanzheng Song <songyuanzheng@huawei.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Alex Shi <alexs@kernel.org>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: delay waking of tasks throttled on NOPROGRESS

Tracing indicates that tasks throttled on NOPROGRESS are woken
prematurely resulting in occasional massive spikes in direct reclaim
activity.  This patch wakes tasks throttled on NOPROGRESS if reclaim
efficiency is at least 12%.

Link: https://lkml.kernel.org/r/20211022144651.19914-9-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: increase the timeout if page reclaim is not making progress

Tracing of the stutterp workload showed the following delays

      1 usect_delayed=124000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=128000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=176000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=536000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=544000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=556000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=624000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=716000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usect_delayed=772000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usect_delayed=512000 reason=VMSCAN_THROTTLE_NOPROGRESS
     16 usect_delayed=120000 reason=VMSCAN_THROTTLE_NOPROGRESS
     53 usect_delayed=116000 reason=VMSCAN_THROTTLE_NOPROGRESS
    116 usect_delayed=112000 reason=VMSCAN_THROTTLE_NOPROGRESS
   5907 usect_delayed=108000 reason=VMSCAN_THROTTLE_NOPROGRESS
  71741 usect_delayed=104000 reason=VMSCAN_THROTTLE_NOPROGRESS

All the throttling hit the full timeout and then there was wakeup delays
meaning that the wakeups are premature as no other reclaimer such as
kswapd has made progress.  This patch increases the maximum timeout.

Link: https://lkml.kernel.org/r/20211022144651.19914-8-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: centralise timeout values for reclaim_throttle

Neil Brown raised concerns about callers of reclaim_throttle specifying
a timeout value.  The original timeout values to congestion_wait() were
probably pulled out of thin air or copy&pasted from somewhere else.
This patch centralises the timeout values and selects a timeout based on
the reason for reclaim throttling.  These figures are also pulled out of
the same thin air but better values may be derived

Running a workload that is throttling for inappropriate periods and
tracing mm_vmscan_throttled can be used to pick a more appropriate
value.  Excessive throttling would pick a lower timeout where as
excessive CPU usage in reclaim context would select a larger timeout.
Ideally a large value would always be used and the wakeups would occur
before a timeout but that requires careful testing.

Link: https://lkml.kernel.org/r/20211022144651.19914-7-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: remove the throttling logic from the page allocator

The page allocator stalls based on the number of pages that are waiting
for writeback to start but this should now be redundant.
shrink_inactive_list() will wake flusher threads if the LRU tail are
unqueued dirty pages so the flusher should be active.  If it fails to
make progress due to pages under writeback not being completed quickly
then it should stall on VMSCAN_THROTTLE_WRITEBACK.

Link: https://lkml.kernel.org/r/20211022144651.19914-6-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/writeback: throttle based on page writeback instead of congestion

do_writepages throttles on congestion if the writepages() fails due to a
lack of memory but congestion_wait() is partially broken as the
congestion state is not updated for all BDIs.

This patch stalls waiting for a number of pages to complete writeback
that located on the local node.  The main weakness is that there is no
correlation between the location of the inode's pages and locality but
that is still better than congestion_wait.

Link: https://lkml.kernel.org/r/20211022144651.19914-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: throttle reclaim when no progress is being made

Memcg reclaim throttles on congestion if no reclaim progress is made.
This makes little sense, it might be due to writeback or a host of other
factors.

For !memcg reclaim, it's messy.  Direct reclaim primarily is throttled
in the page allocator if it is failing to make progress.  Kswapd
throttles if too many pages are under writeback and marked for immediate
reclaim.

This patch explicitly throttles if reclaim is failing to make progress.

[vbabka@suse.cz: Remove redundant code]

Link: https://lkml.kernel.org/r/20211022144651.19914-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: throttle reclaim and compaction when too may pages are isolated

Page reclaim throttles on congestion if too many parallel reclaim
instances have isolated too many pages.  This makes no sense, excessive
parallelisation has nothing to do with writeback or congestion.

This patch creates an additional workqueue to sleep on when too many
pages are isolated.  The throttled tasks are woken when the number of
isolated pages is reduced or a timeout occurs.  There may be some false
positive wakeups for GFP_NOIO/GFP_NOFS callers but the tasks will
throttle again if necessary.

[shy828301@gmail.com: Wake up from compaction context]
[vbabka@suse.cz: Account number of throttled tasks only for writeback]

Link: https://lkml.kernel.org/r/20211022144651.19914-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan: throttle reclaim until some writeback completes if congested

Patch series "Remove dependency on congestion_wait in mm/", v5.

This series that removes all calls to congestion_wait in mm/ and deletes
wait_iff_congested.  It's not a clever implementation but
congestion_wait has been broken for a long time [1].

Even if congestion throttling worked, it was never a great idea.  While
excessive dirty/writeback pages at the tail of the LRU is one
possibility that reclaim may be slow, there is also the problem of too
many pages being isolated and reclaim failing for other reasons
(elevated references, too many pages isolated, excessive LRU contention
etc).

This series replaces the "congestion" throttling with 3 different types.

 - If there are too many dirty/writeback pages, sleep until a timeout or
   enough pages get cleaned

 - If too many pages are isolated, sleep until enough isolated pages are
   either reclaimed or put back on the LRU

 - If no progress is being made, direct reclaim tasks sleep until
   another task makes progress with acceptable efficiency.

This was initially tested with a mix of workloads that used to trigger
corner cases that no longer work.  A new test case was created called
"stutterp" (pagereclaim-stutterp-noreaders in mmtests) using a freshly
created XFS filesystem.  Note that it may be necessary to increase the
timeout of ssh if executing remotely as ssh itself can get throttled and
the connection may timeout.

stutterp varies the number of "worker" processes from 4 up to NR_CPUS*4
to check the impact as the number of direct reclaimers increase.  It has
four types of worker.

 - One "anon latency" worker creates small mappings with mmap() and
   times how long it takes to fault the mapping reading it 4K at a time

 - X file writers which is fio randomly writing X files where the total
   size of the files add up to the allowed dirty_ratio. fio is allowed
   to run for a warmup period to allow some file-backed pages to
   accumulate. The duration of the warmup is based on the best-case
   linear write speed of the storage.

 - Y file readers which is fio randomly reading small files

 - Z anon memory hogs which continually map (100-dirty_ratio)% of memory

 - Total estimated WSS = (100+dirty_ration) percentage of memory

X+Y+Z+1 == NR_WORKERS varying from 4 up to NR_CPUS*4

The intent is to maximise the total WSS with a mix of file and anon
memory where some anonymous memory must be swapped and there is a high
likelihood of dirty/writeback pages reaching the end of the LRU.

The test can be configured to have no background readers to stress
dirty/writeback pages.  The results below are based on having zero
readers.

The short summary of the results is that the series works and stalls
until some event occurs but the timeouts may need adjustment.

The test results are not broken down by patch as the series should be
treated as one block that replaces a broken throttling mechanism with a
working one.

Finally, three machines were tested but I'm reporting the worst set of
results.  The other two machines had much better latencies for example.

First the results of the "anon latency" latency

  stutterp
                                5.15.0-rc1             5.15.0-rc1
                                   vanilla mm-reclaimcongest-v5r4
  Amean     mmap-4      31.4003 (   0.00%)   2661.0198 (-8374.52%)
  Amean     mmap-7      38.1641 (   0.00%)    149.2891 (-291.18%)
  Amean     mmap-12     60.0981 (   0.00%)    187.8105 (-212.51%)
  Amean     mmap-21    161.2699 (   0.00%)    213.9107 ( -32.64%)
  Amean     mmap-30    174.5589 (   0.00%)    377.7548 (-116.41%)
  Amean     mmap-48   8106.8160 (   0.00%)   1070.5616 (  86.79%)
  Stddev    mmap-4      41.3455 (   0.00%)  27573.9676 (-66591.66%)
  Stddev    mmap-7      53.5556 (   0.00%)   4608.5860 (-8505.23%)
  Stddev    mmap-12    171.3897 (   0.00%)   5559.4542 (-3143.75%)
  Stddev    mmap-21   1506.6752 (   0.00%)   5746.2507 (-281.39%)
  Stddev    mmap-30    557.5806 (   0.00%)   7678.1624 (-1277.05%)
  Stddev    mmap-48  61681.5718 (   0.00%)  14507.2830 (  76.48%)
  Max-90    mmap-4      31.4243 (   0.00%)     83.1457 (-164.59%)
  Max-90    mmap-7      41.0410 (   0.00%)     41.0720 (  -0.08%)
  Max-90    mmap-12     66.5255 (   0.00%)     53.9073 (  18.97%)
  Max-90    mmap-21    146.7479 (   0.00%)    105.9540 (  27.80%)
  Max-90    mmap-30    193.9513 (   0.00%)     64.3067 (  66.84%)
  Max-90    mmap-48    277.9137 (   0.00%)    591.0594 (-112.68%)
  Max       mmap-4    1913.8009 (   0.00%) 299623.9695 (-15555.96%)
  Max       mmap-7    2423.9665 (   0.00%) 204453.1708 (-8334.65%)
  Max       mmap-12   6845.6573 (   0.00%) 221090.3366 (-3129.64%)
  Max       mmap-21  56278.6508 (   0.00%) 213877.3496 (-280.03%)
  Max       mmap-30  19716.2990 (   0.00%) 216287.6229 (-997.00%)
  Max       mmap-48 477923.9400 (   0.00%) 245414.8238 (  48.65%)

For most thread counts, the time to mmap() is unfortunately increased.
In earlier versions of the series, this was lower but a large number of
throttling events were reaching their timeout increasing the amount of
inefficient scanning of the LRU.  There is no prioritisation of reclaim
tasks making progress based on each tasks rate of page allocation versus
progress of reclaim.  The variance is also impacted for high worker
counts but in all cases, the differences in latency are not
statistically significant due to very large maximum outliers.  Max-90
shows that 90% of the stalls are comparable but the Max results show the
massive outliers which are increased to to stalling.

It is expected that this will be very machine dependant.  Due to the
test design, reclaim is difficult so allocations stall and there are
variances depending on whether THPs can be allocated or not.  The amount
of memory will affect exactly how bad the corner cases are and how often
they trigger.  The warmup period calculation is not ideal as it's based
on linear writes where as fio is randomly writing multiple files from
multiple tasks so the start state of the test is variable.  For example,
these are the latencies on a single-socket machine that had more memory

  Amean     mmap-4      42.2287 (   0.00%)     49.6838 * -17.65%*
  Amean     mmap-7     216.4326 (   0.00%)     47.4451 *  78.08%*
  Amean     mmap-12   2412.0588 (   0.00%)     51.7497 (  97.85%)
  Amean     mmap-21   5546.2548 (   0.00%)     51.8862 (  99.06%)
  Amean     mmap-30   1085.3121 (   0.00%)     72.1004 (  93.36%)

The overall system CPU usage and elapsed time is as follows

                    5.15.0-rc3  5.15.0-rc3
                       vanilla mm-reclaimcongest-v5r4
  Duration User        6989.03      983.42
  Duration System      7308.12      799.68
  Duration Elapsed     2277.67     2092.98

The patches reduce system CPU usage by 89% as the vanilla kernel is rarely
stalling.

The high-level /proc/vmstats show

                                       5.15.0-rc1     5.15.0-rc1
                                          vanilla mm-reclaimcongest-v5r2
  Ops Direct pages scanned          1056608451.00   503594991.00
  Ops Kswapd pages scanned           109795048.00   147289810.00
  Ops Kswapd pages reclaimed          63269243.00    31036005.00
  Ops Direct pages reclaimed          10803973.00     6328887.00
  Ops Kswapd efficiency %                   57.62          21.07
  Ops Kswapd velocity                    48204.98       57572.86
  Ops Direct efficiency %                    1.02           1.26
  Ops Direct velocity                   463898.83      196845.97

Kswapd scanned less pages but the detailed pattern is different.  The
vanilla kernel scans slowly over time where as the patches exhibits
burst patterns of scan activity.  Direct reclaim scanning is reduced by
52% due to stalling.

The pattern for stealing pages is also slightly different.  Both kernels
exhibit spikes but the vanilla kernel when reclaiming shows pages being
reclaimed over a period of time where as the patches tend to reclaim in
spikes.  The difference is that vanilla is not throttling and instead
scanning constantly finding some pages over time where as the patched
kernel throttles and reclaims in spikes.

  Ops Percentage direct scans               90.59          77.37

For direct reclaim, vanilla scanned 90.59% of pages where as with the
patches, 77.37% were direct reclaim due to throttling

  Ops Page writes by reclaim           2613590.00     1687131.00

Page writes from reclaim context are reduced.

  Ops Page writes anon                 2932752.00     1917048.00

And there is less swapping.

  Ops Page reclaim immediate         996248528.00   107664764.00

The number of pages encountered at the tail of the LRU tagged for
immediate reclaim but still dirty/writeback is reduced by 89%.

  Ops Slabs scanned                     164284.00      153608.00

Slab scan activity is similar.

ftrace was used to gather stall activity

  Vanilla
  -------
      1 writeback_wait_iff_congested: usec_timeout=100000 usec_delayed=16000
      2 writeback_wait_iff_congested: usec_timeout=100000 usec_delayed=12000
      8 writeback_wait_iff_congested: usec_timeout=100000 usec_delayed=8000
     29 writeback_wait_iff_congested: usec_timeout=100000 usec_delayed=4000
  82394 writeback_wait_iff_congested: usec_timeout=100000 usec_delayed=0

The fast majority of wait_iff_congested calls do not stall at all.  What
is likely happening is that cond_resched() reschedules the task for a
short period when the BDI is not registering congestion (which it never
will in this test setup).

      1 writeback_congestion_wait: usec_timeout=100000 usec_delayed=120000
      2 writeback_congestion_wait: usec_timeout=100000 usec_delayed=132000
      4 writeback_congestion_wait: usec_timeout=100000 usec_delayed=112000
    380 writeback_congestion_wait: usec_timeout=100000 usec_delayed=108000
    778 writeback_congestion_wait: usec_timeout=100000 usec_delayed=104000

congestion_wait if called always exceeds the timeout as there is no
trigger to wake it up.

Bottom line: Vanilla will throttle but it's not effective.

Patch series
------------

Kswapd throttle activity was always due to scanning pages tagged for
immediate reclaim at the tail of the LRU

      1 usec_timeout=100000 usect_delayed=72000 reason=VMSCAN_THROTTLE_WRITEBACK
      4 usec_timeout=100000 usect_delayed=20000 reason=VMSCAN_THROTTLE_WRITEBACK
      5 usec_timeout=100000 usect_delayed=12000 reason=VMSCAN_THROTTLE_WRITEBACK
      6 usec_timeout=100000 usect_delayed=16000 reason=VMSCAN_THROTTLE_WRITEBACK
     11 usec_timeout=100000 usect_delayed=100000 reason=VMSCAN_THROTTLE_WRITEBACK
     11 usec_timeout=100000 usect_delayed=8000 reason=VMSCAN_THROTTLE_WRITEBACK
     94 usec_timeout=100000 usect_delayed=0 reason=VMSCAN_THROTTLE_WRITEBACK
    112 usec_timeout=100000 usect_delayed=4000 reason=VMSCAN_THROTTLE_WRITEBACK

The majority of events did not stall or stalled for a short period.
Roughly 16% of stalls reached the timeout before expiry.  For direct
reclaim, the number of times stalled for each reason were

   6624 reason=VMSCAN_THROTTLE_ISOLATED
  93246 reason=VMSCAN_THROTTLE_NOPROGRESS
  96934 reason=VMSCAN_THROTTLE_WRITEBACK

The most common reason to stall was due to excessive pages tagged for
immediate reclaim at the tail of the LRU followed by a failure to make
forward.  A relatively small number were due to too many pages isolated
from the LRU by parallel threads

For VMSCAN_THROTTLE_ISOLATED, the breakdown of delays was

      9 usec_timeout=20000 usect_delayed=4000 reason=VMSCAN_THROTTLE_ISOLATED
     12 usec_timeout=20000 usect_delayed=16000 reason=VMSCAN_THROTTLE_ISOLATED
     83 usec_timeout=20000 usect_delayed=20000 reason=VMSCAN_THROTTLE_ISOLATED
   6520 usec_timeout=20000 usect_delayed=0 reason=VMSCAN_THROTTLE_ISOLATED

Most did not stall at all.  A small number reached the timeout.

For VMSCAN_THROTTLE_NOPROGRESS, the breakdown of stalls were all over
the map

      1 usec_timeout=500000 usect_delayed=324000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usec_timeout=500000 usect_delayed=332000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usec_timeout=500000 usect_delayed=348000 reason=VMSCAN_THROTTLE_NOPROGRESS
      1 usec_timeout=500000 usect_delayed=360000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=228000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=260000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=340000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=364000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=372000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=428000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=460000 reason=VMSCAN_THROTTLE_NOPROGRESS
      2 usec_timeout=500000 usect_delayed=464000 reason=VMSCAN_THROTTLE_NOPROGRESS
      3 usec_timeout=500000 usect_delayed=244000 reason=VMSCAN_THROTTLE_NOPROGRESS
      3 usec_timeout=500000 usect_delayed=252000 reason=VMSCAN_THROTTLE_NOPROGRESS
      3 usec_timeout=500000 usect_delayed=272000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=188000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=268000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=328000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=380000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=392000 reason=VMSCAN_THROTTLE_NOPROGRESS
      4 usec_timeout=500000 usect_delayed=432000 reason=VMSCAN_THROTTLE_NOPROGRESS
      5 usec_timeout=500000 usect_delayed=204000 reason=VMSCAN_THROTTLE_NOPROGRESS
      5 usec_timeout=500000 usect_delayed=220000 reason=VMSCAN_THROTTLE_NOPROGRESS
      5 usec_timeout=500000 usect_delayed=412000 reason=VMSCAN_THROTTLE_NOPROGRESS
      5 usec_timeout=500000 usect_delayed=436000 reason=VMSCAN_THROTTLE_NOPROGRESS
      6 usec_timeout=500000 usect_delayed=488000 reason=VMSCAN_THROTTLE_NOPROGRESS
      7 usec_timeout=500000 usect_delayed=212000 reason=VMSCAN_THROTTLE_NOPROGRESS
      7 usec_timeout=500000 usect_delayed=300000 reason=VMSCAN_THROTTLE_NOPROGRESS
      7 usec_timeout=500000 usect_delayed=316000 reason=VMSCAN_THROTTLE_NOPROGRESS
      7 usec_timeout=500000 usect_delayed=472000 reason=VMSCAN_THROTTLE_NOPROGRESS
      8 usec_timeout=500000 usect_delayed=248000 reason=VMSCAN_THROTTLE_NOPROGRESS
      8 usec_timeout=500000 usect_delayed=356000 reason=VMSCAN_THROTTLE_NOPROGRESS
      8 usec_timeout=500000 usect_delayed=456000 reason=VMSCAN_THROTTLE_NOPROGRESS
      9 usec_timeout=500000 usect_delayed=124000 reason=VMSCAN_THROTTLE_NOPROGRESS
      9 usec_timeout=500000 usect_delayed=376000 reason=VMSCAN_THROTTLE_NOPROGRESS
      9 usec_timeout=500000 usect_delayed=484000 reason=VMSCAN_THROTTLE_NOPROGRESS
     10 usec_timeout=500000 usect_delayed=172000 reason=VMSCAN_THROTTLE_NOPROGRESS
     10 usec_timeout=500000 usect_delayed=420000 reason=VMSCAN_THROTTLE_NOPROGRESS
     10 usec_timeout=500000 usect_delayed=452000 reason=VMSCAN_THROTTLE_NOPROGRESS
     11 usec_timeout=500000 usect_delayed=256000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=112000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=116000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=144000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=152000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=264000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=384000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=424000 reason=VMSCAN_THROTTLE_NOPROGRESS
     12 usec_timeout=500000 usect_delayed=492000 reason=VMSCAN_THROTTLE_NOPROGRESS
     13 usec_timeout=500000 usect_delayed=184000 reason=VMSCAN_THROTTLE_NOPROGRESS
     13 usec_timeout=500000 usect_delayed=444000 reason=VMSCAN_THROTTLE_NOPROGRESS
     14 usec_timeout=500000 usect_delayed=308000 reason=VMSCAN_THROTTLE_NOPROGRESS
     14 usec_timeout=500000 usect_delayed=440000 reason=VMSCAN_THROTTLE_NOPROGRESS
     14 usec_timeout=500000 usect_delayed=476000 reason=VMSCAN_THROTTLE_NOPROGRESS
     16 usec_timeout=500000 usect_delayed=140000 reason=VMSCAN_THROTTLE_NOPROGRESS
     17 usec_timeout=500000 usect_delayed=232000 reason=VMSCAN_THROTTLE_NOPROGRESS
     17 usec_timeout=500000 usect_delayed=240000 reason=VMSCAN_THROTTLE_NOPROGRESS
     17 usec_timeout=500000 usect_delayed=280000 reason=VMSCAN_THROTTLE_NOPROGRESS
     18 usec_timeout=500000 usect_delayed=404000 reason=VMSCAN_THROTTLE_NOPROGRESS
     20 usec_timeout=500000 usect_delayed=148000 reason=VMSCAN_THROTTLE_NOPROGRESS
     20 usec_timeout=500000 usect_delayed=216000 reason=VMSCAN_THROTTLE_NOPROGRESS
     20 usec_timeout=500000 usect_delayed=468000 reason=VMSCAN_THROTTLE_NOPROGRESS
     21 usec_timeout=500000 usect_delayed=448000 reason=VMSCAN_THROTTLE_NOPROGRESS
     23 usec_timeout=500000 usect_delayed=168000 reason=VMSCAN_THROTTLE_NOPROGRESS
     23 usec_timeout=500000 usect_delayed=296000 reason=VMSCAN_THROTTLE_NOPROGRESS
     25 usec_timeout=500000 usect_delayed=132000 reason=VMSCAN_THROTTLE_NOPROGRESS
     25 usec_timeout=500000 usect_delayed=352000 reason=VMSCAN_THROTTLE_NOPROGRESS
     26 usec_timeout=500000 usect_delayed=180000 reason=VMSCAN_THROTTLE_NOPROGRESS
     27 usec_timeout=500000 usect_delayed=284000 reason=VMSCAN_THROTTLE_NOPROGRESS
     28 usec_timeout=500000 usect_delayed=164000 reason=VMSCAN_THROTTLE_NOPROGRESS
     29 usec_timeout=500000 usect_delayed=136000 reason=VMSCAN_THROTTLE_NOPROGRESS
     30 usec_timeout=500000 usect_delayed=200000 reason=VMSCAN_THROTTLE_NOPROGRESS
     30 usec_timeout=500000 usect_delayed=400000 reason=VMSCAN_THROTTLE_NOPROGRESS
     31 usec_timeout=500000 usect_delayed=196000 reason=VMSCAN_THROTTLE_NOPROGRESS
     32 usec_timeout=500000 usect_delayed=156000 reason=VMSCAN_THROTTLE_NOPROGRESS
     33 usec_timeout=500000 usect_delayed=224000 reason=VMSCAN_THROTTLE_NOPROGRESS
     35 usec_timeout=500000 usect_delayed=128000 reason=VMSCAN_THROTTLE_NOPROGRESS
     35 usec_timeout=500000 usect_delayed=176000 reason=VMSCAN_THROTTLE_NOPROGRESS
     36 usec_timeout=500000 usect_delayed=368000 reason=VMSCAN_THROTTLE_NOPROGRESS
     36 usec_timeout=500000 usect_delayed=496000 reason=VMSCAN_THROTTLE_NOPROGRESS
     37 usec_timeout=500000 usect_delayed=312000 reason=VMSCAN_THROTTLE_NOPROGRESS
     38 usec_timeout=500000 usect_delayed=304000 reason=VMSCAN_THROTTLE_NOPROGRESS
     40 usec_timeout=500000 usect_delayed=288000 reason=VMSCAN_THROTTLE_NOPROGRESS
     43 usec_timeout=500000 usect_delayed=408000 reason=VMSCAN_THROTTLE_NOPROGRESS
     55 usec_timeout=500000 usect_delayed=416000 reason=VMSCAN_THROTTLE_NOPROGRESS
     56 usec_timeout=500000 usect_delayed=76000 reason=VMSCAN_THROTTLE_NOPROGRESS
     58 usec_timeout=500000 usect_delayed=120000 reason=VMSCAN_THROTTLE_NOPROGRESS
     59 usec_timeout=500000 usect_delayed=208000 reason=VMSCAN_THROTTLE_NOPROGRESS
     61 usec_timeout=500000 usect_delayed=68000 reason=VMSCAN_THROTTLE_NOPROGRESS
     71 usec_timeout=500000 usect_delayed=192000 reason=VMSCAN_THROTTLE_NOPROGRESS
     71 usec_timeout=500000 usect_delayed=480000 reason=VMSCAN_THROTTLE_NOPROGRESS
     79 usec_timeout=500000 usect_delayed=60000 reason=VMSCAN_THROTTLE_NOPROGRESS
     82 usec_timeout=500000 usect_delayed=320000 reason=VMSCAN_THROTTLE_NOPROGRESS
     82 usec_timeout=500000 usect_delayed=92000 reason=VMSCAN_THROTTLE_NOPROGRESS
     85 usec_timeout=500000 usect_delayed=64000 reason=VMSCAN_THROTTLE_NOPROGRESS
     85 usec_timeout=500000 usect_delayed=80000 reason=VMSCAN_THROTTLE_NOPROGRESS
     88 usec_timeout=500000 usect_delayed=84000 reason=VMSCAN_THROTTLE_NOPROGRESS
     90 usec_timeout=500000 usect_delayed=160000 reason=VMSCAN_THROTTLE_NOPROGRESS
     90 usec_timeout=500000 usect_delayed=292000 reason=VMSCAN_THROTTLE_NOPROGRESS
     94 usec_timeout=500000 usect_delayed=56000 reason=VMSCAN_THROTTLE_NOPROGRESS
    118 usec_timeout=500000 usect_delayed=88000 reason=VMSCAN_THROTTLE_NOPROGRESS
    119 usec_timeout=500000 usect_delayed=72000 reason=VMSCAN_THROTTLE_NOPROGRESS
    126 usec_timeout=500000 usect_delayed=108000 reason=VMSCAN_THROTTLE_NOPROGRESS
    146 usec_timeout=500000 usect_delayed=52000 reason=VMSCAN_THROTTLE_NOPROGRESS
    148 usec_timeout=500000 usect_delayed=36000 reason=VMSCAN_THROTTLE_NOPROGRESS
    148 usec_timeout=500000 usect_delayed=48000 reason=VMSCAN_THROTTLE_NOPROGRESS
    159 usec_timeout=500000 usect_delayed=28000 reason=VMSCAN_THROTTLE_NOPROGRESS
    178 usec_timeout=500000 usect_delayed=44000 reason=VMSCAN_THROTTLE_NOPROGRESS
    183 usec_timeout=500000 usect_delayed=40000 reason=VMSCAN_THROTTLE_NOPROGRESS
    237 usec_timeout=500000 usect_delayed=100000 reason=VMSCAN_THROTTLE_NOPROGRESS
    266 usec_timeout=500000 usect_delayed=32000 reason=VMSCAN_THROTTLE_NOPROGRESS
    313 usec_timeout=500000 usect_delayed=24000 reason=VMSCAN_THROTTLE_NOPROGRESS
    347 usec_timeout=500000 usect_delayed=96000 reason=VMSCAN_THROTTLE_NOPROGRESS
    470 usec_timeout=500000 usect_delayed=20000 reason=VMSCAN_THROTTLE_NOPROGRESS
    559 usec_timeout=500000 usect_delayed=16000 reason=VMSCAN_THROTTLE_NOPROGRESS
    964 usec_timeout=500000 usect_delayed=12000 reason=VMSCAN_THROTTLE_NOPROGRESS
   2001 usec_timeout=500000 usect_delayed=104000 reason=VMSCAN_THROTTLE_NOPROGRESS
   2447 usec_timeout=500000 usect_delayed=8000 reason=VMSCAN_THROTTLE_NOPROGRESS
   7888 usec_timeout=500000 usect_delayed=4000 reason=VMSCAN_THROTTLE_NOPROGRESS
  22727 usec_timeout=500000 usect_delayed=0 reason=VMSCAN_THROTTLE_NOPROGRESS
  51305 usec_timeout=500000 usect_delayed=500000 reason=VMSCAN_THROTTLE_NOPROGRESS

The full timeout is often hit but a large number also do not stall at
all.  The remainder slept a little allowing other reclaim tasks to make
progress.

While this timeout could be further increased, it could also negatively
impact worst-case behaviour when there is no prioritisation of what task
should make progress.

For VMSCAN_THROTTLE_WRITEBACK, the breakdown was

      1 usec_timeout=100000 usect_delayed=44000 reason=VMSCAN_THROTTLE_WRITEBACK
      2 usec_timeout=100000 usect_delayed=76000 reason=VMSCAN_THROTTLE_WRITEBACK
      3 usec_timeout=100000 usect_delayed=80000 reason=VMSCAN_THROTTLE_WRITEBACK
      5 usec_timeout=100000 usect_delayed=48000 reason=VMSCAN_THROTTLE_WRITEBACK
      5 usec_timeout=100000 usect_delayed=84000 reason=VMSCAN_THROTTLE_WRITEBACK
      6 usec_timeout=100000 usect_delayed=72000 reason=VMSCAN_THROTTLE_WRITEBACK
      7 usec_timeout=100000 usect_delayed=88000 reason=VMSCAN_THROTTLE_WRITEBACK
     11 usec_timeout=100000 usect_delayed=56000 reason=VMSCAN_THROTTLE_WRITEBACK
     12 usec_timeout=100000 usect_delayed=64000 reason=VMSCAN_THROTTLE_WRITEBACK
     16 usec_timeout=100000 usect_delayed=92000 reason=VMSCAN_THROTTLE_WRITEBACK
     24 usec_timeout=100000 usect_delayed=68000 reason=VMSCAN_THROTTLE_WRITEBACK
     28 usec_timeout=100000 usect_delayed=32000 reason=VMSCAN_THROTTLE_WRITEBACK
     30 usec_timeout=100000 usect_delayed=60000 reason=VMSCAN_THROTTLE_WRITEBACK
     30 usec_timeout=100000 usect_delayed=96000 reason=VMSCAN_THROTTLE_WRITEBACK
     32 usec_timeout=100000 usect_delayed=52000 reason=VMSCAN_THROTTLE_WRITEBACK
     42 usec_timeout=100000 usect_delayed=40000 reason=VMSCAN_THROTTLE_WRITEBACK
     77 usec_timeout=100000 usect_delayed=28000 reason=VMSCAN_THROTTLE_WRITEBACK
     99 usec_timeout=100000 usect_delayed=36000 reason=VMSCAN_THROTTLE_WRITEBACK
    137 usec_timeout=100000 usect_delayed=24000 reason=VMSCAN_THROTTLE_WRITEBACK
    190 usec_timeout=100000 usect_delayed=20000 reason=VMSCAN_THROTTLE_WRITEBACK
    339 usec_timeout=100000 usect_delayed=16000 reason=VMSCAN_THROTTLE_WRITEBACK
    518 usec_timeout=100000 usect_delayed=12000 reason=VMSCAN_THROTTLE_WRITEBACK
    852 usec_timeout=100000 usect_delayed=8000 reason=VMSCAN_THROTTLE_WRITEBACK
   3359 usec_timeout=100000 usect_delayed=4000 reason=VMSCAN_THROTTLE_WRITEBACK
   7147 usec_timeout=100000 usect_delayed=0 reason=VMSCAN_THROTTLE_WRITEBACK
  83962 usec_timeout=100000 usect_delayed=100000 reason=VMSCAN_THROTTLE_WRITEBACK

The majority hit the timeout in direct reclaim context although a
sizable number did not stall at all.  This is very different to kswapd
where only a tiny percentage of stalls due to writeback reached the
timeout.

Bottom line, the throttling appears to work and the wakeup events may
limit worst case stalls.  There might be some grounds for adjusting
timeouts but it's likely futile as the worst-case scenarios depend on
the workload, memory size and the speed of the storage.  A better
approach to improve the series further would be to prioritise tasks
based on their rate of allocation with the caveat that it may be very
expensive to track.

This patch (of 5):

Page reclaim throttles on wait_iff_congested under the following
conditions:

 - kswapd is encountering pages under writeback and marked for immediate
   reclaim implying that pages are cycling through the LRU faster than
   pages can be cleaned.

 - Direct reclaim will stall if all dirty pages are backed by congested
   inodes.

wait_iff_congested is almost completely broken with few exceptions.
This patch adds a new node-based workqueue and tracks the number of
throttled tasks and pages written back since throttling started.  If
enough pages belonging to the node are written back then the throttled
tasks will wake early.  If not, the throttled tasks sleeps until the
timeout expires.

[neilb@suse.de: Uninterruptible sleep and simpler wakeups]
[hdanton@sina.com: Avoid race when reclaim starts]
[vbabka@suse.cz: vmstat irq-safe api, clarifications]

Link: https://lore.kernel.org/linux-mm/45d8b7a6-8548-65f5-cccf-9f451d4ae3d4@kernel.dk/
Link: https://lkml.kernel.org/r/20211022144651.19914-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20211022144651.19914-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: NeilBrown <neilb@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: "Darrick J . Wong" <djwong@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: fix -Wunused-but-set-variable warning

We fix the following warning when building kernel with W=1:

  mm/vmscan.c:1362:6: warning: variable 'err' set but not used [-Wunused-but-set-variable]

Link: https://lkml.kernel.org/r/20210924181218.21165-1-songkai01@inspur.com
Signed-off-by: Kai Song <songkai01@inspur.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_isolation: guard against possible putback unisolated page

Isolating a free page in an isolated pageblock is expected to always
work as watermarks don't apply here.

But if __isolate_free_page() failed, due to condition changes, the page
will be left on the free list.  And the page will be put back to free
list again via __putback_isolated_page().  This may trigger
VM_BUG_ON_PAGE() on page->flags checking in __free_one_page() if
PageReported is set.  Or we will corrupt the free list because
list_add() will be called for pages already on another list.

Add a VM_WARN_ON() to complain about this change.

Link: https://lkml.kernel.org/r/20210914114508.23725-1-linmiaohe@huawei.com
Fixes: 3c605096d315 ("mm/page_alloc: restrict max order of merging on isolated pageblock")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_isolation: fix potential missing call to unset_migratetype_isolate()

In start_isolate_page_range() undo path, pfn_to_online_page() just
checks the first pfn in a pageblock while __first_valid_page() will
traverse the pageblock until the first online pfn is found.  So we may
miss the call to unset_migratetype_isolate() in undo path and pages will
remain isolated unexpectedly.

Fix this by calling undo_isolate_page_range() and this will also help to
simplify the code further.  Note we shouldn't ever trigger it because
MAX_ORDER-1 aligned pfn ranges shouldn't contain memory holes now.

Link: https://lkml.kernel.org/r/20210914114348.15569-1-linmiaohe@huawei.com
Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd/selftests: fix calculation of expected ioctls

Today, we assert that the ioctls the kernel reports as supported for a
registration match a precomputed list.  We decide which ioctls are
supported by examining the memory type.  Then, in several locations we
"fix up" this list by adding or removing things this initial decision
got wrong.

What ioctls the kernel reports is actually a function of several things:
- The memory type
- Kernel feature support (e.g., no writeprotect on aarch64)
- The registration type (e.g., CONTINUE only supported for MINOR mode)

So, we can't fully compute this at the start, in set_test_type.  It
varies per test, depending on what registration mode(s) those tests use.

Instead, introduce a new function which computes the correct list.  This
centralizes the add/remove of ioctls depending on these function inputs
in one place, so we don't have to repeat ourselves in various tests.

Not only is the resulting code a bit shorter, but it fixes a real bug in
the existing code: previously, we would incorrectly require the
writeprotect ioctl to be present on aarch64, where it isn't actually
supported.

Link: https://lkml.kernel.org/r/20210930212309.4001967-4-axelrasmussen@google.com
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd/selftests: fix feature support detection

Before any tests are run, in set_test_type, we decide what feature(s) we
are going to be testing, based upon our command line arguments.
However, the supported features are not just a function of the memory
type being used, so this is broken.

For instance, consider writeprotect support.  It is "normally" supported
for anonymous memory, but furthermore it requires that the kernel has
CONFIG_HAVE_ARCH_USERFAULTFD_WP.  So, it is *not* supported at all on
aarch64, for example.

So, this fixes this by querying the kernel for the set of features it
supports in set_test_type, by opening a userfaultfd and issuing a
UFFDIO_API ioctl.  Based upon the reported features, we toggle what
tests are enabled.

Link: https://lkml.kernel.org/r/20210930212309.4001967-3-axelrasmussen@google.com
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

userfaultfd/selftests: don't rely on GNU extensions for random numbers

Patch series "Small userfaultfd selftest fixups", v2.

This patch (of 3):

Two arguments for doing this:

First, and maybe most importantly, the resulting code is significantly
shorter / simpler.

Then, we avoid using GNU libc extensions.  Why does this matter? It
makes testing userfaultfd with the selftest easier e.g.  on distros
which use something other than glibc (e.g., Alpine, which uses musl);
basically, it makes the test more portable.

Link: https://lkml.kernel.org/r/20210930212309.4001967-2-axelrasmussen@google.com
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: remove unnecessary set_page_count in prep_compound_gigantic_page

In commit 7118fc2906e2 ("hugetlb: address ref count racing in
prep_compound_gigantic_page"), page_ref_freeze is used to atomically
zero the ref count of tail pages iff they are 1.  The unconditional call
to set_page_count(0) was left in the code.  This call is after
page_ref_freeze so it is really a noop.

Remove redundant and unnecessary set_page_count call.

Link: https://lkml.kernel.org/r/20211026220635.35187-1-mike.kravetz@oracle.com
Fixes: 7118fc2906e29 ("hugetlb: address ref count racing in prep_compound_gigantic_page")
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Suggested-by: Pasha Tatashin <pasha.tatashin@soleen.com>
Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: remove redundant VM_BUG_ON() in add_reservation_in_range()

When calling hugetlb_resv_map_add(), we've guaranteed that the parameter
'to' is always larger than 'from', so it never returns a negative value
from hugetlb_resv_map_add().  Thus remove the redundant VM_BUG_ON().

Link: https://lkml.kernel.org/r/2b565552f3d06753da1e8dda439c0d96d6d9a5a3.1634797639.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: remove redundant validation in has_same_uncharge_info()

The callers of has_same_uncharge_info() has accessed the original
file_region and new file_region, and they are impossible to be NULL now.

So we can remove the file_region validation in has_same_uncharge_info()
to simplify the code.

Link: https://lkml.kernel.org/r/97fc68d3f8d34f63c204645e10d7a718997e50b7.1634797639.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: replace the obsolete hugetlb_instantiation_mutex in the comments

After commit 8382d914ebf7 ("mm, hugetlb: improve page-fault
scalability"), the hugetlb_instantiation_mutex lock had been replaced by
hugetlb_fault_mutex_table to serializes faults on the same logical page.

Thus update the obsolete hugetlb_instantiation_mutex related comments.

Link: https://lkml.kernel.org/r/4b3febeae37455ff7b74aa0aad16cc6909cf0926.1634797639.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb_cgroup: remove unused hugetlb_cgroup_from_counter macro

Patch series "Some cleanups and improvements for hugetlb".

This patchset does some cleanups and improvements for hugetlb and
hugetlb_cgroup.

This patch (of 4):

Since commit 726b7bbeafd4 ("hugetlb_cgroup: fix illegal access to
memory"), the hugetlb_cgroup_from_counter() macro is not used any more,
remove it.

Link: https://lkml.kernel.org/r/cover.1634797639.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/f03b29b801fa9942466ab15334ec09988e124ae6.1634797639.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove duplicate include in hugepage-mremap.c

Remove duplicate includes 'unistd.h' included in
 '/tools/testing/selftests/vm/hugepage-mremap.c'  is duplicated.It is also
 included on 23 line.

Link: https://lkml.kernel.org/r/20211018102336.869726-1-ran.jianping@zte.com.cn
Signed-off-by: Ran Jianping <ran.jianping@zte.com.cn>
Reported-by: Zeal Robot <zealci@zte.com.cn>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: support node specified when using cma for gigantic hugepages

Now the size of CMA area for gigantic hugepages runtime allocation is
balanced for all online nodes, but we also want to specify the size of
CMA per-node, or only one node in some cases, which are similar with
patch [1].

For example, on some multi-nodes systems, each node's memory can be
different, allocating the same size of CMA for each node is not suitable
for the low-memory nodes.  Meanwhile some workloads like DPDK mentioned
by Zhenguo in patch [1] only need hugepages in one node.

On the other hand, we have some machines with multiple types of memory,
like DRAM and PMEM (persistent memory).  On this system, we may want to
specify all the hugepages only on DRAM node, or specify the proportion
of DRAM node and PMEM node, to tuning the performance of the workloads.

Thus this patch adds node format for 'hugetlb_cma' parameter to support
specifying the size of CMA per-node.  An example is as follows:

  hugetlb_cma=0:5G,2:5G

which means allocating 5G size of CMA area on node 0 and node 2
respectively.  And the users should use the node specific sysfs file to
allocate the gigantic hugepages if specified the CMA size on that node.

Link: https://lkml.kernel.org/r/20211005054729.86457-1-yaozhenguo1@gmail.com
Link: https://lkml.kernel.org/r/bb790775ca60bb8f4b26956bb3f6988f74e075c7.1634261144.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hugepages: add hugetlb vma mremap() test

[almasrymina@google.com: v8]
Link: https://lkml.kernel.org/r/20211014200542.4126947-2-almasrymina@google.com
[wanjiabing@vivo.com: remove duplicated include in hugepage-mremap]
Link: https://lkml.kernel.org/r/20211021122944.8857-1-wanjiabing@vivo.com
Link: https://lkml.kernel.org/r/20211013195825.3058275-2-almasrymina@google.com
Signed-off-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Wan Jiabing <wanjiabing@vivo.com>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ken Chen <kenchen@google.com>
Cc: Chris Kennelly <ckennelly@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Kirill Shutemov <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, hugepages: add mremap() support for hugepage backed vma

Support mremap() for hugepage backed vma segment by simply repositioning
page table entries.  The page table entries are repositioned to the new
virtual address on mremap().

Hugetlb mremap() support is of course generic; my motivating use case is
a library (hugepage_text), which reloads the ELF text of executables in
hugepages.  This significantly increases the execution performance of
said executables.

Restrict the mremap operation on hugepages to up to the size of the
original mapping as the underlying hugetlb reservation is not yet
capable of handling remapping to a larger size.

During the mremap() operation we detect pmd_share'd mappings and we
unshare those during the mremap().  On access and fault the sharing is
established again.

Link: https://lkml.kernel.org/r/20211013195825.3058275-1-almasrymina@google.com
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ken Chen <kenchen@google.com>
Cc: Chris Kennelly <ckennelly@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Kirill Shutemov <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: khugepaged: recalculate min_free_kbytes after stopping khugepaged

When initializing transparent huge pages, min_free_kbytes would be
calculated according to what khugepaged expected.

So when transparent huge pages get disabled, min_free_kbytes should be
recalculated instead of the higher value set by khugepaged.

Link: https://lkml.kernel.org/r/1633937809-16558-1-git-send-email-liangcaifan19@gmail.com
Signed-off-by: Liangcai Fan <liangcaifan19@gmail.com>
Signed-off-by: Chunyan Zhang <zhang.lyra@gmail.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add hugetlb demote page support

Demote page functionality will split a huge page into a number of huge
pages of a smaller size.  For example, on x86 a 1GB huge page can be
demoted into 512 2M huge pages.  Demotion is done 'in place' by simply
splitting the huge page.

Added '*_for_demote' wrappers for remove_hugetlb_page,
destroy_compound_hugetlb_page and prep_compound_gigantic_page for use by
demote code.

[mike.kravetz@oracle.com: v4]
Link: https://lkml.kernel.org/r/6ca29b8e-527c-d6ec-900e-e6a43e4f8b73@oracle.com
Link: https://lkml.kernel.org/r/20211007181918.136982-6-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Nghia Le <nghialm78@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add demote bool to gigantic page routines

The routines remove_hugetlb_page and destroy_compound_gigantic_page will
remove a gigantic page and make the set of base pages ready to be
returned to a lower level allocator.  In the process of doing this, they
make all base pages reference counted.

The routine prep_compound_gigantic_page creates a gigantic page from a
set of base pages.  It assumes that all these base pages are reference
counted.

During demotion, a gigantic page will be split into huge pages of a
smaller size.  This logically involves use of the routines,
remove_hugetlb_page, and destroy_compound_gigantic_page followed by
prep_compound*_page for each smaller huge page.

When pages are reference counted (ref count >= 0), additional
speculative ref counts could be taken as described in previous commits
[1] and [2].  This could result in errors while demoting a huge page.
Quite a bit of code would need to be created to handle all possible
issues.

Instead of dealing with the possibility of speculative ref counts, avoid
the possibility by keeping ref counts at zero during the demote process.
Add a boolean 'demote' to the routines remove_hugetlb_page,
destroy_compound_gigantic_page and prep_compound_gigantic_page.  If the
boolean is set, the remove and destroy routines will not reference count
pages and the prep routine will not expect reference counted pages.

'*_for_demote' wrappers of the routines will be added in a subsequent
patch where this functionality is used.

[1] https://lore.kernel.org/linux-mm/20210622021423.154662-3-mike.kravetz@oracle.com/
[2] https://lore.kernel.org/linux-mm/20210809184832.18342-3-mike.kravetz@oracle.com/

Link: https://lkml.kernel.org/r/20211007181918.136982-5-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Nghia Le <nghialm78@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: be sure to free demoted CMA pages to CMA

When huge page demotion is fully implemented, gigantic pages can be
demoted to a smaller huge page size.  For example, on x86 a 1G page can
be demoted to 512 2M pages.  However, gigantic pages can potentially be
allocated from CMA.  If a gigantic page which was allocated from CMA is
demoted, the corresponding demoted pages needs to be returned to CMA.

Use the new interface cma_pages_valid() to determine if a non-gigantic
hugetlb page should be freed to CMA.  Also, clear mapping field of these
pages as expected by cma_release.

This also requires a change to CMA region creation for gigantic pages.
CMA uses a per-region bit map to track allocations.  When setting up the
region, you specify how many pages each bit represents.  Currently, only
gigantic pages are allocated/freed from CMA so the region is set up such
that one bit represents a gigantic page size allocation.

With demote, a gigantic page (allocation) could be split into smaller
size pages.  And, these smaller size pages will be freed to CMA.  So,
since the per-region bit map needs to be set up to represent the
smallest allocation/free size, it now needs to be set to the smallest
huge page size which can be freed to CMA.

Unfortunately, we set up the CMA region for huge pages before we set up
huge pages sizes (hstates).  So, technically we do not know the smallest
huge page size as this can change via command line options and
architecture specific code.  Therefore, at region setup time we use
HUGETLB_PAGE_ORDER as the smallest possible huge page size that can be
given back to CMA.  It is possible that this value is sub-optimal for
some architectures/config options.  If needed, this can be addressed in
follow on work.

Link: https://lkml.kernel.org/r/20211007181918.136982-4-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Nghia Le <nghialm78@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/cma: add cma_pages_valid to determine if pages are in CMA

Add new interface cma_pages_valid() which indicates if the specified
pages are part of a CMA region.  This interface will be used in a
subsequent patch by hugetlb code.

In order to keep the same amount of DEBUG information, a pr_debug() call
was added to cma_pages_valid().  In the case where the page passed to
cma_release is not in cma region, the debug message will be printed from
cma_pages_valid as opposed to cma_release.

Link: https://lkml.kernel.org/r/20211007181918.136982-3-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Nghia Le <nghialm78@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: add demote hugetlb page sysfs interfaces

Patch series "hugetlb: add demote/split page functionality", v4.

The concurrent use of multiple hugetlb page sizes on a single system is
becoming more common.  One of the reasons is better TLB support for
gigantic page sizes on x86 hardware.  In addition, hugetlb pages are
being used to back VMs in hosting environments.

When using hugetlb pages to back VMs, it is often desirable to
preallocate hugetlb pools.  This avoids the delay and uncertainty of
allocating hugetlb pages at VM startup.  In addition, preallocating huge
pages minimizes the issue of memory fragmentation that increases the
longer the system is up and running.

In such environments, a combination of larger and smaller hugetlb pages
are preallocated in anticipation of backing VMs of various sizes.  Over
time, the preallocated pool of smaller hugetlb pages may become depleted
while larger hugetlb pages still remain.  In such situations, it is
desirable to convert larger hugetlb pages to smaller hugetlb pages.

Converting larger to smaller hugetlb pages can be accomplished today by
first freeing the larger page to the buddy allocator and then allocating
the smaller pages.  For example, to convert 50 GB pages on x86:

  gb_pages=`cat .../hugepages-1048576kB/nr_hugepages`
  m2_pages=`cat .../hugepages-2048kB/nr_hugepages`
  echo $(($gb_pages - 50)) > .../hugepages-1048576kB/nr_hugepages
  echo $(($m2_pages + 25600)) > .../hugepages-2048kB/nr_hugepages

On an idle system this operation is fairly reliable and results are as
expected.  The number of 2MB pages is increased as expected and the time
of the operation is a second or two.

However, when there is activity on the system the following issues
arise:

1) This process can take quite some time, especially if allocation of
   the smaller pages is not immediate and requires migration/compaction.

2) There is no guarantee that the total size of smaller pages allocated
   will match the size of the larger page which was freed. This is
   because the area freed by the larger page could quickly be
   fragmented.

In a test environment with a load that continually fills the page cache
with clean pages, results such as the following can be observed:

  Unexpected number of 2MB pages allocated: Expected 25600, have 19944
  real    0m42.092s
  user    0m0.008s
  sys     0m41.467s

To address these issues, introduce the concept of hugetlb page demotion.
Demotion provides a means of 'in place' splitting of a hugetlb page to
pages of a smaller size.  This avoids freeing pages to buddy and then
trying to allocate from buddy.

Page demotion is controlled via sysfs files that reside in the per-hugetlb
page size and per node directories.

 - demote_size
        Target page size for demotion, a smaller huge page size. File
        can be written to chose a smaller huge page size if multiple are
        available.

 - demote
        Writable number of hugetlb pages to be demoted

To demote 50 GB huge pages, one would:

  cat .../hugepages-1048576kB/free_hugepages   /* optional, verify free pages */
  cat .../hugepages-1048576kB/demote_size      /* optional, verify target size */
  echo 50 > .../hugepages-1048576kB/demote

Only hugetlb pages which are free at the time of the request can be
demoted.  Demotion does not add to the complexity of surplus pages and
honors reserved huge pages.  Therefore, when a value is written to the
sysfs demote file, that value is only the maximum number of pages which
will be demoted.  It is possible fewer will actually be demoted.  The
recently introduced per-hstate mutex is used to synchronize demote
operations with other operations that modify hugetlb pools.

Real world use cases
--------------------
The above scenario describes a real world use case where hugetlb pages
are used to back VMs on x86.  Both issues of long allocation times and
not necessarily getting the expected number of smaller huge pages after
a free and allocate cycle have been experienced.  The occurrence of
these issues is dependent on other activity within the host and can not
be predicted.

This patch (of 5):

Two new sysfs files are added to demote hugtlb pages.  These files are
both per-hugetlb page size and per node.  Files are:

  demote_size - The size in Kb that pages are demoted to. (read-write)
  demote - The number of huge pages to demote. (write-only)

By default, demote_size is the next smallest huge page size.  Valid huge
page sizes less than huge page size may be written to this file.  When
huge pages are demoted, they are demoted to this size.

Writing a value to demote will result in an attempt to demote that
number of hugetlb pages to an appropriate number of demote_size pages.

NOTE: Demote interfaces are only provided for huge page sizes if there
is a smaller target demote huge page size.  For example, on x86 1GB huge
pages will have demote interfaces.  2MB huge pages will not have demote
interfaces.

This patch does not provide full demote functionality.  It only provides
the sysfs interfaces.

It also provides documentation for the new interfaces.

[mike.kravetz@oracle.com: n_mask initialization does not need to be protected by the mutex]
Link: https://lkml.kernel.org/r/0530e4ef-2492-5186-f919-5db68edea654@oracle.com
Link: https://lkml.kernel.org/r/20211007181918.136982-2-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: David Rientjes <rientjes@google.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>
Cc: Nghia Le <nghialm78@gmail.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: drop __unmap_hugepage_range definition from hugetlb.h

Remove __unmap_hugepage_range() from the header file, because it is only
used in hugetlb.c.

Link: https://lkml.kernel.org/r/20210917165108.9341-1-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Suggested-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: hwpoison: handle non-anonymous THP correctly

Currently hwpoison doesn't handle non-anonymous THP, but since v4.8 THP
support for tmpfs and read-only file cache has been added.  They could
be offlined by split THP, just like anonymous THP.

Link: https://lkml.kernel.org/r/20211020210755.23964-7-shy828301@gmail.com
Signed-off-by: Yang Shi <shy828301@gmail.com>
Acked-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: shmem: don't truncate page if memory failure happens

The current behavior of memory failure is to truncate the page cache
regardless of dirty or clean.  If the page is dirty the later access
will get the obsolete data from disk without any notification to the
users.  This may cause silent data loss.  It is even worse for shmem
since shmem is in-memory filesystem, truncating page cache means
discarding data blocks.  The later read would return all zero.

The right approach is to keep the corrupted page in page cache, any
later access would return error for syscalls or SIGBUS for page fault,
until the file is truncated, hole punched or removed.  The regular
storage backed filesystems would be more complicated so this patch is
focused on shmem.  This also unblock the support for soft offlining
shmem THP.

[arnd@arndb.de: fix uninitialized variable use in me_pagecache_clean()]
Link: https://lkml.kernel.org/r/20211022064748.4173718-1-arnd@kernel.org
Link: https://lkml.kernel.org/r/20211020210755.23964-6-shy828301@gmail.com
Signed-off-by: Yang Shi <shy828301@gmail.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: hwpoison: refactor refcount check handling

Memory failure will report failure if the page still has extra pinned
refcount other than from hwpoison after the handler is done.  Actually
the check is not necessary for all handlers, so move the check into
specific handlers.  This would make the following keeping shmem page in
page cache patch easier.

There may be expected extra pin for some cases, for example, when the
page is dirty and in swapcache.

Link: https://lkml.kernel.org/r/20211020210755.23964-5-shy828301@gmail.com
Signed-off-by: Yang Shi <shy828301@gmail.com>
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Suggested-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: filemap: coding style cleanup for filemap_map_pmd()

Patch series "Solve silent data loss caused by poisoned page cache (shmem/tmpfs)", v5.

When discussing the patch that splits page cache THP in order to offline
the poisoned page, Noaya mentioned there is a bigger problem [1] that
prevents this from working since the page cache page will be truncated
if uncorrectable errors happen.  By looking this deeper it turns out
this approach (truncating poisoned page) may incur silent data loss for
all non-readonly filesystems if the page is dirty.  It may be worse for
in-memory filesystem, e.g.  shmem/tmpfs since the data blocks are
actually gone.

To solve this problem we could keep the poisoned dirty page in page
cache then notify the users on any later access, e.g.  page fault,
read/write, etc.  The clean page could be truncated as is since they can
be reread from disk later on.

The consequence is the filesystems may find poisoned page and manipulate
it as healthy page since all the filesystems actually don't check if the
page is poisoned or not in all the relevant paths except page fault.  In
general, we need make the filesystems be aware of poisoned page before
we could keep the poisoned page in page cache in order to solve the data
loss problem.

To make filesystems be aware of poisoned page we should consider:

 - The page should be not written back: clearing dirty flag could
   prevent from writeback.

 - The page should not be dropped (it shows as a clean page) by drop
   caches or other callers: the refcount pin from hwpoison could prevent
   from invalidating (called by cache drop, inode cache shrinking, etc),
   but it doesn't avoid invalidation in DIO path.

 - The page should be able to get truncated/hole punched/unlinked: it
   works as it is.

 - Notify users when the page is accessed, e.g. read/write, page fault
   and other paths (compression, encryption, etc).

The scope of the last one is huge since almost all filesystems need do
it once a page is returned from page cache lookup.  There are a couple
of options to do it:

 1. Check hwpoison flag for every path, the most straightforward way.

 2. Return NULL for poisoned page from page cache lookup, the most
    callsites check if NULL is returned, this should have least work I
    think. But the error handling in filesystems just return -ENOMEM,
    the error code will incur confusion to the users obviously.

 3. To improve #2, we could return error pointer, e.g. ERR_PTR(-EIO),
    but this will involve significant amount of code change as well
    since all the paths need check if the pointer is ERR or not just
    like option #1.

I did prototypes for both #1 and #3, but it seems #3 may require more
changes than #1.  For #3 ERR_PTR will be returned so all the callers
need to check the return value otherwise invalid pointer may be
dereferenced, but not all callers really care about the content of the
page, for example, partial truncate which just sets the truncated range
in one page to 0.  So for such paths it needs additional modification if
ERR_PTR is returned.  And if the callers have their own way to handle
the problematic pages we need to add a new FGP flag to tell FGP
functions to return the pointer to the page.

It may happen very rarely, but once it happens the consequence (data
corruption) could be very bad and it is very hard to debug.  It seems
this problem had been slightly discussed before, but seems no action was
taken at that time.  [2]

As the aforementioned investigation, it needs huge amount of work to
solve the potential data loss for all filesystems.  But it is much
easier for in-memory filesystems and such filesystems actually suffer
more than others since even the data blocks are gone due to truncating.
So this patchset starts from shmem/tmpfs by taking option #1.

TODO:
* The unpoison has been broken since commit 0ed950d1f281 ("mm,hwpoison: make
  get_hwpoison_page() call get_any_page()"), and this patch series make
  refcount check for unpoisoning shmem page fail.
* Expand to other filesystems.  But I haven't heard feedback from filesystem
  developers yet.

Patch breakdown:
Patch #1: cleanup, depended by patch #2
Patch #2: fix THP with hwpoisoned subpage(s) PMD map bug
Patch #3: coding style cleanup
Patch #4: refactor and preparation.
Patch #5: keep the poisoned page in page cache and handle such case for all
          the paths.
Patch #6: the previous patches unblock page cache THP split, so this patch
          add page cache THP split support.

This patch (of 4):

A minor cleanup to the indent.

Link: https://lkml.kernel.org/r/20211020210755.23964-1-shy828301@gmail.com
Link: https://lkml.kernel.org/r/20211020210755.23964-4-shy828301@gmail.com
Signed-off-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_failure: constify static mm_walk_ops

The only usage of hwp_walk_ops is to pass its address to
walk_page_range() which takes a pointer to const mm_walk_ops as
argument.

Make it const to allow the compiler to put it in read-only memory.

Link: https://lkml.kernel.org/r/20211014075042.17174-3-rikard.falkeborn@gmail.com
Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Acked-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix data race in PagePoisoned()

PagePoisoned() accesses page->flags which can be updated concurrently:

  | BUG: KCSAN: data-race in next_uptodate_page / unlock_page
  |
  | write (marked) to 0xffffea00050f37c0 of 8 bytes by task 1872 on cpu 1:
  |  instrument_atomic_write           include/linux/instrumented.h:87 [inline]
  |  clear_bit_unlock_is_negative_byte include/asm-generic/bitops/instrumented-lock.h:74 [inline]
  |  unlock_page+0x102/0x1b0           mm/filemap.c:1465
  |  filemap_map_pages+0x6c6/0x890     mm/filemap.c:3057
  |  ...
  | read to 0xffffea00050f37c0 of 8 bytes by task 1873 on cpu 0:
  |  PagePoisoned                   include/linux/page-flags.h:204 [inline]
  |  PageReadahead                  include/linux/page-flags.h:382 [inline]
  |  next_uptodate_page+0x456/0x830 mm/filemap.c:2975
  |  ...
  | CPU: 0 PID: 1873 Comm: systemd-udevd Not tainted 5.11.0-rc4-00001-gf9ce0be71d1f #1

To avoid the compiler tearing or otherwise optimizing the access, use
READ_ONCE() to access flags.

Link: https://lore.kernel.org/all/20210826144157.GA26950@xsang-OptiPlex-9020/
Link: https://lkml.kernel.org/r/20210913113542.2658064-1-elver@google.com
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Will Deacon <will@kernel.org>
Cc: Marco Elver <elver@google.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: use clamp() to simplify code

This patch uses clamp() to simplify code in init_per_zone_wmark_min().

Link: https://lkml.kernel.org/r/20211021034830.1049150-1-bobo.shaobowang@huawei.com
Signed-off-by: Wang ShaoBo <bobo.shaobowang@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Wei Yongjun <weiyongjun1@huawei.com>
Cc: Li Bin <huawei.libin@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: page_alloc: use migrate_disable() in drain_local_pages_wq()

drain_local_pages_wq() disables preemption to avoid CPU migration during
CPU hotplug and can't use cpus_read_lock().

Using migrate_disable() works here, too.  The scheduler won't take the
CPU offline until the task left the migrate-disable section.  The
problem with disabled preemption here is that drain_local_pages()
acquires locks which are turned into sleeping locks on PREEMPT_RT and
can't be acquired with disabled preemption.

Use migrate_disable() in drain_local_pages_wq().

Link: https://lkml.kernel.org/r/20211015210933.viw6rjvo64qtqxn4@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

s390: use generic version of arch_is_kernel_initmem_freed()

The generic version of arch_is_kernel_initmem_freed() now does the same
as s390 version.

Remove the s390 version.

Link: https://lkml.kernel.org/r/b6feb5dfe611a322de482762fc2df3a9eece70c7.1633001016.git.christophe.leroy@csgroup.eu
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc: use generic version of arch_is_kernel_initmem_freed()

The generic version of arch_is_kernel_initmem_freed() now does the same
as powerpc version.

Remove the powerpc version.

Link: https://lkml.kernel.org/r/c53764eb45d41491e2b21da2e7812239897dbebb.1633001016.git.christophe.leroy@csgroup.eu
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make generic arch_is_kernel_initmem_freed() do what it says

Commit 7a5da02de8d6 ("locking/lockdep: check for freed initmem in
static_obj()") added arch_is_kernel_initmem_freed() which is supposed to
report whether an object is part of already freed init memory.

For the time being, the generic version of
arch_is_kernel_initmem_freed() always reports 'false', allthough
free_initmem() is generically called on all architectures.

Therefore, change the generic version of arch_is_kernel_initmem_freed()
to check whether free_initmem() has been called.  If so, then check if a
given address falls into init memory.

To ease the use of system_state, move it out of line into its only
caller which is lockdep.c

Link: https://lkml.kernel.org/r/1d40783e676e07858be97d881f449ee7ea8adfb1.1633001016.git.christophe.leroy@csgroup.eu
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: create a new system state and fix core_kernel_text()

core_kernel_text() considers that until system_state in at least
SYSTEM_RUNNING, init memory is valid.

But init memory is freed a few lines before setting SYSTEM_RUNNING, so
we have a small period of time when core_kernel_text() is wrong.

Create an intermediate system state called SYSTEM_FREEING_INIT that is
set before starting freeing init memory, and use it in
core_kernel_text() to report init memory invalid earlier.

Link: https://lkml.kernel.org/r/9ecfdee7dd4d741d172cb93ff1d87f1c58127c9a.1633001016.git.christophe.leroy@csgroup.eu
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@ozlabs.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: show watermark_boost of zone in zoneinfo

min/low/high_wmark_pages(z) is defined as

  (z->_watermark[WMARK_MIN/LOW/HIGH] + z->watermark_boost)

If kswapd is frequently woken up due to the increase of
min/low/high_wmark_pages, printing watermark_boost can quickly locate
whether watermark_boost or _watermark[WMARK_MIN/LOW/HIGH] caused
min/low/high_wmark_pages to increase.

Link: https://lkml.kernel.org/r/1632472566-12246-1-git-send-email-liangcaifan19@gmail.com
Signed-off-by: Liangcai Fan <liangcaifan19@gmail.com>
Cc: Chunyan Zhang <zhang.lyra@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: detect allocation forbidden by cpuset and bail out early

There was a report that starting an Ubuntu in docker while using cpuset
to bind it to movable nodes (a node only has movable zone, like a node
for hotplug or a Persistent Memory node in normal usage) will fail due
to memory allocation failure, and then OOM is involved and many other
innocent processes got killed.

It can be reproduced with command:

    $ docker run -it --rm --cpuset-mems 4 ubuntu:latest bash -c "grep Mems_allowed /proc/self/status"

(where node 4 is a movable node)

  runc:[2:INIT] invoked oom-killer: gfp_mask=0x500cc2(GFP_HIGHUSER|__GFP_ACCOUNT), order=0, oom_score_adj=0
  CPU: 8 PID: 8291 Comm: runc:[2:INIT] Tainted: G        W I E     5.8.2-0.g71b519a-default #1 openSUSE Tumbleweed (unreleased)
  Hardware name: Dell Inc. PowerEdge R640/0PHYDR, BIOS 2.6.4 04/09/2020
  Call Trace:
   dump_stack+0x6b/0x88
   dump_header+0x4a/0x1e2
   oom_kill_process.cold+0xb/0x10
   out_of_memory.part.0+0xaf/0x230
   out_of_memory+0x3d/0x80
   __alloc_pages_slowpath.constprop.0+0x954/0xa20
   __alloc_pages_nodemask+0x2d3/0x300
   pipe_write+0x322/0x590
   new_sync_write+0x196/0x1b0
   vfs_write+0x1c3/0x1f0
   ksys_write+0xa7/0xe0
   do_syscall_64+0x52/0xd0
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

  Mem-Info:
  active_anon:392832 inactive_anon:182 isolated_anon:0
   active_file:68130 inactive_file:151527 isolated_file:0
   unevictable:2701 dirty:0 writeback:7
   slab_reclaimable:51418 slab_unreclaimable:116300
   mapped:45825 shmem:735 pagetables:2540 bounce:0
   free:159849484 free_pcp:73 free_cma:0
  Node 4 active_anon:1448kB inactive_anon:0kB active_file:0kB inactive_file:0kB unevictable:0kB isolated(anon):0kB isolated(file):0kB mapped:0kB dirty:0kB writeback:0kB shmem:0kB shmem_thp: 0kB shmem_pmdmapped: 0kB anon_thp: 0kB writeback_tmp:0kB all_unreclaimable? no
  Node 4 Movable free:130021408kB min:9140kB low:139160kB high:269180kB reserved_highatomic:0KB active_anon:1448kB inactive_anon:0kB active_file:0kB inactive_file:0kB unevictable:0kB writepending:0kB present:130023424kB managed:130023424kB mlocked:0kB kernel_stack:0kB pagetables:0kB bounce:0kB free_pcp:292kB local_pcp:84kB free_cma:0kB
  lowmem_reserve[]: 0 0 0 0 0
  Node 4 Movable: 1*4kB (M) 0*8kB 0*16kB 1*32kB (M) 0*64kB 0*128kB 1*256kB (M) 1*512kB (M) 1*1024kB (M) 0*2048kB 31743*4096kB (M) = 130021156kB

  oom-kill:constraint=CONSTRAINT_CPUSET,nodemask=(null),cpuset=docker-9976a269caec812c134fa317f27487ee36e1129beba7278a463dd53e5fb9997b.scope,mems_allowed=4,global_oom,task_memcg=/system.slice/containerd.service,task=containerd,pid=4100,uid=0
  Out of memory: Killed process 4100 (containerd) total-vm:4077036kB, anon-rss:51184kB, file-rss:26016kB, shmem-rss:0kB, UID:0 pgtables:676kB oom_score_adj:0
  oom_reaper: reaped process 8248 (docker), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
  oom_reaper: reaped process 2054 (node_exporter), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
  oom_reaper: reaped process 1452 (systemd-journal), now anon-rss:0kB, file-rss:8564kB, shmem-rss:4kB
  oom_reaper: reaped process 2146 (munin-node), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
  oom_reaper: reaped process 8291 (runc:[2:INIT]), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB

The reason is that in this case, the target cpuset nodes only have
movable zone, while the creation of an OS in docker sometimes needs to
allocate memory in non-movable zones (dma/dma32/normal) like
GFP_HIGHUSER, and the cpuset limit forbids the allocation, then
out-of-memory killing is involved even when normal nodes and movable
nodes both have many free memory.

The OOM killer cannot help to resolve the situation as there is no
usable memory for the request in the cpuset scope.  The only reasonable
measure to take is to fail the allocation right away and have the caller
to deal with it.

So add a check for cases like this in the slowpath of allocation, and
bail out early returning NULL for the allocation.

As page allocation is one of the hottest path in kernel, this check will
hurt all users with sane cpuset configuration, add a static branch check
and detect the abnormal config in cpuset memory binding setup so that
the extra check cost in page allocation is not paid by everyone.

[thanks to Micho Hocko and David Rientjes for suggesting not handling
 it inside OOM code, adding cpuset check, refining comments]

Link: https://lkml.kernel.org/r/1632481657-68112-1-git-send-email-feng.tang@intel.com
Signed-off-by: Feng Tang <feng.tang@intel.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Zefan Li <lizefan.x@bytedance.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: do not acquire zone lock in is_free_buddy_page()

Grabbing zone lock in is_free_buddy_page() gives a wrong sense of
safety, and has potential performance implications when zone is
experiencing lock contention.

In any case, if a caller needs a stable result, it should grab zone lock
before calling this function.

Link: https://lkml.kernel.org/r/20210922152833.4023972-1-eric.dumazet@gmail.com
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move fold_vm_numa_events() to fix NUMA without SMP

If CONFIG_NUMA=y, but CONFIG_SMP=n (e.g. sh/migor_defconfig):

    sh4-linux-gnu-ld: mm/vmstat.o: in function `vmstat_start': vmstat.c:(.text+0x97c): undefined reference to `fold_vm_numa_events'
    sh4-linux-gnu-ld: drivers/base/node.o: in function `node_read_vmstat': node.c:(.text+0x140): undefined reference to `fold_vm_numa_events'
    sh4-linux-gnu-ld: drivers/base/node.o: in function `node_read_numastat': node.c:(.text+0x1d0): undefined reference to `fold_vm_numa_events'

Fix this by moving fold_vm_numa_events() outside the SMP-only section.

Link: https://lkml.kernel.org/r/9d16ccdd9ef32803d7100c84f737de6a749314fb.1631781495.git.geert+renesas@glider.be
Fixes: f19298b9516c1a03 ("mm/vmstat: convert NUMA statistics to basic NUMA counters")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Gon Solo <gonsolo@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move node_reclaim_distance to fix NUMA without SMP

Patch series "Fix NUMA without SMP".

SuperH is the only architecture which still supports NUMA without SMP,
for good reasons (various memories scattered around the address space,
each with varying latencies).

This series fixes two build errors due to variables and functions used
by the NUMA code being provided by SMP-only source files or sections.

This patch (of 2):

If CONFIG_NUMA=y, but CONFIG_SMP=n (e.g. sh/migor_defconfig):

    sh4-linux-gnu-ld: mm/page_alloc.o: in function `get_page_from_freelist':
    page_alloc.c:(.text+0x2c24): undefined reference to `node_reclaim_distance'

Fix this by moving the declaration of node_reclaim_distance from an
SMP-only to a generic file.

Link: https://lkml.kernel.org/r/cover.1631781495.git.geert+renesas@glider.be
Link: https://lkml.kernel.org/r/6432666a648dde85635341e6c918cee97c97d264.1631781495.git.geert+renesas@glider.be
Fixes: a55c7454a8c887b2 ("sched/topology: Improve load balancing on AMD EPYC systems")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Suggested-by: Matt Fleming <matt@codeblueprint.co.uk>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Cc: Rich Felker <dalias@libc.org>
Cc: Gon Solo <gonsolo@gmail.com>
Cc: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: use accumulated load when building node fallback list

In build_zonelists(), when the fallback list is built for the nodes, the
node load gets reinitialized during each iteration.  This results in
nodes with same distances occupying the same slot in different node
fallback lists rather than appearing in the intended round- robin
manner.  This results in one node getting picked for allocation more
compared to other nodes with the same distance.

As an example, consider a 4 node system with the following distance
matrix.

  Node 0  1  2  3
  ----------------
  0    10 12 32 32
  1    12 10 32 32
  2    32 32 10 12
  3    32 32 12 10

For this case, the node fallback list gets built like this:

  Node  Fallback list
  ---------------------
  0     0 1 2 3
  1     1 0 3 2
  2     2 3 0 1
  3     3 2 0 1 <-- Unexpected fallback order

In the fallback list for nodes 2 and 3, the nodes 0 and 1 appear in the
same order which results in more allocations getting satisfied from node
0 compared to node 1.

The effect of this on remote memory bandwidth as seen by stream
benchmark is shown below:

  Case 1: Bandwidth from cores on nodes 2 & 3 to memory on nodes 0 & 1
(numactl -m 0,1 ./stream_lowOverhead ... --cores <from 2, 3>)
  Case 2: Bandwidth from cores on nodes 0 & 1 to memory on nodes 2 & 3
(numactl -m 2,3 ./stream_lowOverhead ... --cores <from 0, 1>)

  ----------------------------------------
BANDWIDTH (MB/s)
      TEST Case 1 Case 2
  ----------------------------------------
      COPY 57479.6 110791.8
     SCALE 55372.9 105685.9
       ADD 50460.6 96734.2
    TRIADD 50397.6 97119.1
  ----------------------------------------

The bandwidth drop in Case 1 occurs because most of the allocations get
satisfied by node 0 as it appears first in the fallback order for both
nodes 2 and 3.

This can be fixed by accumulating the node load in build_zonelists()
rather than reinitializing it during each iteration.  With this the
nodes with the same distance rightly get assigned in the round robin
manner.

In fact this was how it was originally until commit f0c0b2b808f2
("change zonelist order: zonelist order selection logic") dropped the
load accumulation and resorted to initializing the load during each
iteration.

While zonelist ordering was removed by commit c9bff3eebc09 ("mm,
page_alloc: rip out ZONELIST_ORDER_ZONE"), the change to the node load
accumulation in build_zonelists() remained.  So essentially this patch
reverts back to the accumulated node load logic.

After this fix, the fallback order gets built like this:

  Node Fallback list
  ------------------
  0    0 1 2 3
  1    1 0 3 2
  2    2 3 0 1
  3    3 2 1 0 <-- Note the change here

The bandwidth in Case 1 improves and matches Case 2 as shown below.

  ----------------------------------------
BANDWIDTH (MB/s)
      TEST Case 1 Case 2
  ----------------------------------------
      COPY 110438.9 110107.2
     SCALE 105930.5 105817.5
       ADD 97005.1 96159.8
    TRIADD 97441.5 96757.1
  ----------------------------------------

The correctness of the fallback list generation has been verified for
the above node configuration where the node 3 starts as memory-less node
and comes up online only during memory hotplug.

[bharata@amd.com: Added changelog, review, test validation]

Link: https://lkml.kernel.org/r/20210830121603.1081-3-bharata@amd.com
Fixes: f0c0b2b808f2 ("change zonelist order: zonelist order selection logic")
Signed-off-by: Krupa Ramakrishnan <krupa.ramakrishnan@amd.com>
Co-developed-by: Sadagopan Srinivasan <Sadagopan.Srinivasan@amd.com>
Signed-off-by: Sadagopan Srinivasan <Sadagopan.Srinivasan@amd.com>
Signed-off-by: Bharata B Rao <bharata@amd.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>