mm/page_alloc.c: skip setting nodemask when we are in interrupt

When we are in the interrupt context, it is irrelevant to the current task
context.  If we use current task's mems_allowed, we can be fair to alloc
pages in the fast path and fall back to slow path memory allocation when
the current node(which is the current task mems_allowed) does not have
enough memory to allocate.  In this case, it slows down the memory
allocation speed of interrupt context.  So we can skip setting the
nodemask to allow any node to allocate memory, so that fast path
allocation can success.

Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200706025921.53683-1-songmuchun@bytedance.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: fallbacks at most has 3 elements

MIGRAGE_TYPES is used to be the mark of end and there are at most 3
elements for the one dimension array.

Reduce to 3 to save little memory.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200625231022.18784-1-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: silence a KASAN false positive

kernel_init_free_pages() will use memset() on s390 to clear all pages from
kmalloc_order() which will override KASAN redzones because a redzone was
setup from the end of the allocation size to the end of the last page.
Silence it by not reporting it there.  An example of the report is,

 BUG: KASAN: slab-out-of-bounds in __free_pages_ok
 Write of size 4096 at addr 000000014beaa000
 Call Trace:
 show_stack+0x152/0x210
 dump_stack+0x1f8/0x248
 print_address_description.isra.13+0x5e/0x4d0
 kasan_report+0x130/0x178
 check_memory_region+0x190/0x218
 memset+0x34/0x60
 __free_pages_ok+0x894/0x12f0
 kfree+0x4f2/0x5e0
 unpack_to_rootfs+0x60e/0x650
 populate_rootfs+0x56/0x358
 do_one_initcall+0x1f4/0xa20
 kernel_init_freeable+0x758/0x7e8
 kernel_init+0x1c/0x170
 ret_from_fork+0x24/0x28
 Memory state around the buggy address:
 000000014bea9f00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 000000014bea9f80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>000000014beaa000: 03 fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
                    ^
 000000014beaa080: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
 000000014beaa100: fe fe fe fe fe fe fe fe fe fe fe fe fe fe

Fixes: 6471384af2a6 ("mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options")
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Vasily Gorbik <gor@linux.ibm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Link: http://lkml.kernel.org/r/20200610052154.5180-1-cai@lca.pw
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: remove unnecessary end_bitidx for [set|get]_pfnblock_flags_mask()

After previous cleanup, the end_bitidx is not necessary any more.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Link: http://lkml.kernel.org/r/20200623124201.8199-4-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: simplify pageblock bitmap access

Due to commit e58469bafd05 ("mm: page_alloc: use word-based accesses for
get/set pageblock bitmaps"), pageblock bitmap is accessed with word-based
access.  This operation could be simplified a little.

Intuitively, if we want to get a bit range [start_idx, end_idx] in a word,
we can do like this:

    mask = (1 << (end_bitidx - start_bitidx + 1)) - 1;
    ret = (word >> start_idx) & mask;

And also if we want to set a bit range [start_idx, end_idx] with flags, we
can do the same by just shift start_bitidx.

By doing so we reduce some instructions for these two helper functions:

                                Before   Patched
    set_pfnblock_flags_mask     209      198(-5%)
    get_pfnblock_flags_mask     101      87(-13%)

Since the syntax is changed a little, we need to check the whole 4-bit
migrate_type instead of part of it.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Link: http://lkml.kernel.org/r/20200623124201.8199-3-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: extract the common part in pfn_to_bitidx()

The return value calculation is the same both for SPARSEMEM or not.

Just take it out.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Link: http://lkml.kernel.org/r/20200623124201.8199-2-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: replace the definition of NR_MIGRATETYPE_BITS with PB_migratetype_bits

We already have the definition of PB_migratetype_bits and current
NR_MIGRATETYPE_BITS looks like a cyclic definition.

Just use PB_migratetype_bits is enough.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Link: http://lkml.kernel.org/r/20200623124201.8199-1-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/shuffle: remove dynamic reconfiguration

Commit e900a918b098 ("mm: shuffle initial free memory to improve
memory-side-cache utilization") promised "autodetection of a
memory-side-cache (to be added in a follow-on patch)" over a year ago.

The original series included patches [1], however, they were dropped
during review [2] to be followed-up later.

Due to lack of platforms that publish an HMAT, autodetection is currently
not implemented.  However, manual activation is actively used [3].  Let's
simplify for now and re-add when really (ever?) needed.

[1] https://lkml.kernel.org/r/154510700291.1941238.817190985966612531.stgit@dwillia2-desk3.amr.corp.intel.com
[2] https://lkml.kernel.org/r/154690326478.676627.103843791978176914.stgit@dwillia2-desk3.amr.corp.intel.com
[3] https://lkml.kernel.org/r/CAPcyv4irwGUU2x+c6b4L=KbB1dnasNKaaZd6oSpYjL9kfsnROQ@mail.gmail.com

Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Dan Williams <dan.j.williams@intel.com>
Link: http://lkml.kernel.org/r/20200624094741.9918-4-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory_hotplug: document why shuffle_zone() is relevant

It's not completely obvious why we have to shuffle the complete zone -
introduced in commit e900a918b098 ("mm: shuffle initial free memory to
improve memory-side-cache utilization") - because some sort of shuffling
is already performed when onlining pages via __free_one_page(), placing
MAX_ORDER-1 pages either to the head or the tail of the freelist.  Let's
document why we have to shuffle the complete zone when exposing larger,
contiguous physical memory areas to the buddy.

Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200624094741.9918-3-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc: remove nr_free_pagecache_pages()

nr_free_pagecache_pages() isn't used outside page_alloc.c anymore - and
the name does not really help to understand what's going on.  Let's
open-code it instead and add a comment.

Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Huang Ying <ying.huang@intel.com>
Link: http://lkml.kernel.org/r/20200619132410.23859-3-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove vm_total_pages

The global variable "vm_total_pages" is a relic from older days.  There is
only a single user that reads the variable - build_all_zonelists() - and
the first thing it does is update it.

Use a local variable in build_all_zonelists() instead and remove the
global variable.

Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Minchan Kim <minchan@kernel.org>
Link: http://lkml.kernel.org/r/20200619132410.23859-2-david@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: skip ->waternark_boost for atomic order-0 allocations

When boosting is enabled, it is observed that rate of atomic order-0
allocation failures are high due to the fact that free levels in the
system are checked with ->watermark_boost offset.  This is not a problem
for sleepable allocations but for atomic allocations which looks like
regression.

This problem is seen frequently on system setup of Android kernel running
on Snapdragon hardware with 4GB RAM size.  When no extfrag event occurred
in the system, ->watermark_boost factor is zero, thus the watermark
configurations in the system are:

   _watermark = (
          [WMARK_MIN] = 1272, --> ~5MB
          [WMARK_LOW] = 9067, --> ~36MB
          [WMARK_HIGH] = 9385), --> ~38MB
   watermark_boost = 0

After launching some memory hungry applications in Android which can cause
extfrag events in the system to an extent that ->watermark_boost can be
set to max i.e.  default boost factor makes it to 150% of high watermark.

   _watermark = (
          [WMARK_MIN] = 1272, --> ~5MB
          [WMARK_LOW] = 9067, --> ~36MB
          [WMARK_HIGH] = 9385), --> ~38MB
   watermark_boost = 14077, -->~57MB

With default system configuration, for an atomic order-0 allocation to
succeed, having free memory of ~2MB will suffice.  But boosting makes the
min_wmark to ~61MB thus for an atomic order-0 allocation to be successful
system should have minimum of ~23MB of free memory(from calculations of
zone_watermark_ok(), min = 3/4(min/2)).  But failures are observed despite
system is having ~20MB of free memory.  In the testing, this is
reproducible as early as first 300secs since boot and with furtherlowram
configurations(<2GB) it is observed as early as first 150secs since boot.

These failures can be avoided by excluding the ->watermark_boost in
watermark caluculations for atomic order-0 allocations.

[akpm@linux-foundation.org: fix comment grammar, reflow comment]
[charante@codeaurora.org: fix suggested by Mel Gorman]
Link: http://lkml.kernel.org/r/31556793-57b1-1c21-1a9d-22674d9bd938@codeaurora.org
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Link: http://lkml.kernel.org/r/1589882284-21010-1-git-send-email-charante@codeaurora.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

page_alloc: consider highatomic reserve in watermark fast

zone_watermark_fast was introduced by commit 48ee5f3696f6 ("mm,
page_alloc: shortcut watermark checks for order-0 pages").  The commit
simply checks if free pages is bigger than watermark without additional
calculation such like reducing watermark.

It considered free cma pages but it did not consider highatomic reserved.
This may incur exhaustion of free pages except high order atomic free
pages.

Assume that reserved_highatomic pageblock is bigger than watermark min,
and there are only few free pages except high order atomic free.  Because
zone_watermark_fast passes the allocation without considering high order
atomic free, normal reclaimable allocation like GFP_HIGHUSER will consume
all the free pages.  Then finally order-0 atomic allocation may fail on
allocation.

This means watermark min is not protected against non-atomic allocation.
The order-0 atomic allocation with ALLOC_HARDER unwantedly can be failed.
Additionally the __GFP_MEMALLOC allocation with ALLOC_NO_WATERMARKS also
can be failed.

To avoid the problem, zone_watermark_fast should consider highatomic
reserve.  If the actual size of high atomic free is counted accurately
like cma free, we may use it.  On this patch just use
nr_reserved_highatomic.  Additionally introduce
__zone_watermark_unusable_free to factor out common parts between
zone_watermark_fast and __zone_watermark_ok.

This is an example of ALLOC_HARDER allocation failure using v4.19 based
kernel.

 Binder:9343_3: page allocation failure: order:0, mode:0x480020(GFP_ATOMIC), nodemask=(null)
 Call trace:
 [<ffffff8008f40f8c>] dump_stack+0xb8/0xf0
 [<ffffff8008223320>] warn_alloc+0xd8/0x12c
 [<ffffff80082245e4>] __alloc_pages_nodemask+0x120c/0x1250
 [<ffffff800827f6e8>] new_slab+0x128/0x604
 [<ffffff800827b0cc>] ___slab_alloc+0x508/0x670
 [<ffffff800827ba00>] __kmalloc+0x2f8/0x310
 [<ffffff80084ac3e0>] context_struct_to_string+0x104/0x1cc
 [<ffffff80084ad8fc>] security_sid_to_context_core+0x74/0x144
 [<ffffff80084ad880>] security_sid_to_context+0x10/0x18
 [<ffffff800849bd80>] selinux_secid_to_secctx+0x20/0x28
 [<ffffff800849109c>] security_secid_to_secctx+0x3c/0x70
 [<ffffff8008bfe118>] binder_transaction+0xe68/0x454c
 Mem-Info:
 active_anon:102061 inactive_anon:81551 isolated_anon:0
  active_file:59102 inactive_file:68924 isolated_file:64
  unevictable:611 dirty:63 writeback:0 unstable:0
  slab_reclaimable:13324 slab_unreclaimable:44354
  mapped:83015 shmem:4858 pagetables:26316 bounce:0
  free:2727 free_pcp:1035 free_cma:178
 Node 0 active_anon:408244kB inactive_anon:326204kB active_file:236408kB inactive_file:275696kB unevictable:2444kB isolated(anon):0kB isolated(file):256kB mapped:332060kB dirty:252kB writeback:0kB shmem:19432kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no
 Normal free:10908kB min:6192kB low:44388kB high:47060kB active_anon:409160kB inactive_anon:325924kB active_file:235820kB inactive_file:276628kB unevictable:2444kB writepending:252kB present:3076096kB managed:2673676kB mlocked:2444kB kernel_stack:62512kB pagetables:105264kB bounce:0kB free_pcp:4140kB local_pcp:40kB free_cma:712kB
 lowmem_reserve[]: 0 0
 Normal: 505*4kB (H) 357*8kB (H) 201*16kB (H) 65*32kB (H) 1*64kB (H) 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 10236kB
 138826 total pagecache pages
 5460 pages in swap cache
 Swap cache stats: add 8273090, delete 8267506, find 1004381/4060142

This is an example of ALLOC_NO_WATERMARKS allocation failure using v4.14
based kernel.

 kswapd0: page allocation failure: order:0, mode:0x140000a(GFP_NOIO|__GFP_HIGHMEM|__GFP_MOVABLE), nodemask=(null)
 kswapd0 cpuset=/ mems_allowed=0
 CPU: 4 PID: 1221 Comm: kswapd0 Not tainted 4.14.113-18770262-userdebug #1
 Call trace:
 [<0000000000000000>] dump_backtrace+0x0/0x248
 [<0000000000000000>] show_stack+0x18/0x20
 [<0000000000000000>] __dump_stack+0x20/0x28
 [<0000000000000000>] dump_stack+0x68/0x90
 [<0000000000000000>] warn_alloc+0x104/0x198
 [<0000000000000000>] __alloc_pages_nodemask+0xdc0/0xdf0
 [<0000000000000000>] zs_malloc+0x148/0x3d0
 [<0000000000000000>] zram_bvec_rw+0x410/0x798
 [<0000000000000000>] zram_rw_page+0x88/0xdc
 [<0000000000000000>] bdev_write_page+0x70/0xbc
 [<0000000000000000>] __swap_writepage+0x58/0x37c
 [<0000000000000000>] swap_writepage+0x40/0x4c
 [<0000000000000000>] shrink_page_list+0xc30/0xf48
 [<0000000000000000>] shrink_inactive_list+0x2b0/0x61c
 [<0000000000000000>] shrink_node_memcg+0x23c/0x618
 [<0000000000000000>] shrink_node+0x1c8/0x304
 [<0000000000000000>] kswapd+0x680/0x7c4
 [<0000000000000000>] kthread+0x110/0x120
 [<0000000000000000>] ret_from_fork+0x10/0x18
 Mem-Info:
 active_anon:111826 inactive_anon:65557 isolated_anon:0\x0a active_file:44260 inactive_file:83422 isolated_file:0\x0a unevictable:4158 dirty:117 writeback:0 unstable:0\x0a            slab_reclaimable:13943 slab_unreclaimable:43315\x0a mapped:102511 shmem:3299 pagetables:19566 bounce:0\x0a free:3510 free_pcp:553 free_cma:0
 Node 0 active_anon:447304kB inactive_anon:262228kB active_file:177040kB inactive_file:333688kB unevictable:16632kB isolated(anon):0kB isolated(file):0kB mapped:410044kB d irty:468kB writeback:0kB shmem:13196kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no
 Normal free:14040kB min:7440kB low:94500kB high:98136kB reserved_highatomic:32768KB active_anon:447336kB inactive_anon:261668kB active_file:177572kB inactive_file:333768k           B unevictable:16632kB writepending:480kB present:4081664kB managed:3637088kB mlocked:16632kB kernel_stack:47072kB pagetables:78264kB bounce:0kB free_pcp:2280kB local_pcp:720kB free_cma:0kB        [ 4738.329607] lowmem_reserve[]: 0 0
 Normal: 860*4kB (H) 453*8kB (H) 180*16kB (H) 26*32kB (H) 34*64kB (H) 6*128kB (H) 2*256kB (H) 0*512kB 0*1024kB 0*2048kB 0*4096kB = 14232kB

This is trace log which shows GFP_HIGHUSER consumes free pages right
before ALLOC_NO_WATERMARKS.

  <...>-22275 [006] ....   889.213383: mm_page_alloc: page=00000000d2be5665 pfn=970744 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213385: mm_page_alloc: page=000000004b2335c2 pfn=970745 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213387: mm_page_alloc: page=00000000017272e1 pfn=970278 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213389: mm_page_alloc: page=00000000c4be79fb pfn=970279 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213391: mm_page_alloc: page=00000000f8a51d4f pfn=970260 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213393: mm_page_alloc: page=000000006ba8f5ac pfn=970261 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213395: mm_page_alloc: page=00000000819f1cd3 pfn=970196 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
  <...>-22275 [006] ....   889.213396: mm_page_alloc: page=00000000f6b72a64 pfn=970197 order=0 migratetype=0 nr_free=3650 gfp_flags=GFP_HIGHUSER|__GFP_ZERO
kswapd0-1207  [005] ...1   889.213398: mm_page_alloc: page= (null) pfn=0 order=0 migratetype=1 nr_free=3650 gfp_flags=GFP_NOWAIT|__GFP_HIGHMEM|__GFP_NOWARN|__GFP_MOVABLE

[jaewon31.kim@samsung.com: remove redundant code for high-order]
Link: http://lkml.kernel.org/r/20200623035242.27232-1-jaewon31.kim@samsung.com
Reported-by: Yong-Taek Lee <ytk.lee@samsung.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Yong-Taek Lee <ytk.lee@samsung.com>
Cc: Michal Hocko <mhocko@kernel.org>
Link: http://lkml.kernel.org/r/20200619235958.11283-1-jaewon31.kim@samsung.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: use unlikely() in task_capc()

Hugh noted that task_capc() could use unlikely(), as most of the time
there is no capture in progress and we are in page freeing hot path.
Indeed adding unlikely() produces assembly that better matches the
assumption and moves all the tests away from the hot path.

I have also noticed that we don't need to test for cc->direct_compaction
as the only place we set current->task_capture is compact_zone_order()
which also always sets cc->direct_compaction true.

Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Hugh Dickins <hughd@googlecom>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Li Wang <liwang@redhat.com>
Link: http://lkml.kernel.org/r/4a24f7af-3aa5-6e80-4ae6-8f253b562039@suse.cz
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: adjust kasan_stack_oob for tag-based mode

Use OOB_TAG_OFF as access offset to land the access into the next granule.

Suggested-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Elena Petrova <lenaptr@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Link: http://lkml.kernel.org/r/403b259f1de49a7a3694531c851ac28326a586a8.1596199677.git.andreyknvl@google.com
Link: http://lkml.kernel.org/r/3063ab1411e92bce36061a96e25b651212e70ba6.1596544734.git.andreyknvl@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: allow enabling stack tagging for tag-based mode

Use CONFIG_KASAN_STACK to enable stack tagging.

Note, that HWASAN short granules [1] are disabled. Supporting those will
require more kernel changes.

[1] https://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Elena Petrova <lenaptr@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Link: http://lkml.kernel.org/r/e7febb907b539c3730780df587ce0b38dc558c3d.1596199677.git.andreyknvl@google.com
Link: http://lkml.kernel.org/r/99f7d90a4237431bf5988599fb41358e92876eb0.1596544734.git.andreyknvl@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan, arm64: don't instrument functions that enable kasan

This patch prepares Software Tag-Based KASAN for stack tagging support.

With stack tagging enabled, KASAN tags stack variable in each function in
its prologue.  In start_kernel() stack variables get tagged before KASAN
is enabled via setup_arch()->kasan_init().  As the result the tags for
start_kernel()'s stack variables end up in the temporary shadow memory.
Later when KASAN gets enabled, switched to normal shadow, and starts
checking tags, this leads to false-positive reports, as proper tags are
missing in normal shadow.

Disable KASAN instrumentation for start_kernel().  Also disable it for
arm64's setup_arch() as a precaution (it doesn't have any stack variables
right now).

[andreyknvl@google.com: reorder attributes for start_kernel()]
Link: http://lkml.kernel.org/r/26fb6165a17abcf61222eda5184c030fb6b133d1.1596544734.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com> [arm64]
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Elena Petrova <lenaptr@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Link: http://lkml.kernel.org/r/55d432671a92e931ab8234b03dc36b14d4c21bfb.1596199677.git.andreyknvl@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

efi: provide empty efi_enter_virtual_mode implementation

When CONFIG_EFI is not enabled, we might get an undefined reference to
efi_enter_virtual_mode() error, if this efi_enabled() call isn't inlined
into start_kernel().  This happens in particular, if start_kernel() is
annodated with __no_sanitize_address.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Elena Petrova <lenaptr@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Link: http://lkml.kernel.org/r/6514652d3a32d3ed33d6eb5c91d0af63bf0d1a0c.1596544734.git.andreyknvl@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: don't tag stacks allocated with pagealloc

Patch series "kasan: support stack instrumentation for tag-based mode", v2.

This patch (of 5):

Prepare Software Tag-Based KASAN for stack tagging support.

With Tag-Based KASAN when kernel stacks are allocated via pagealloc (which
happens when CONFIG_VMAP_STACK is not enabled), they get tagged.  KASAN
instrumentation doesn't expect the sp register to be tagged, and this
leads to false-positive reports.

Fix by resetting the tag of kernel stack pointers after allocation.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Marco Elver <elver@google.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Cc: Elena Petrova <lenaptr@google.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Link: http://lkml.kernel.org/r/cover.1596199677.git.andreyknvl@google.com
Link: http://lkml.kernel.org/r/cover.1596544734.git.andreyknvl@google.com
Link: http://lkml.kernel.org/r/12d8c678869268dd0884b01271ab592f30792abf.1596544734.git.andreyknvl@google.com
Link: http://lkml.kernel.org/r/01c678b877755bcf29009176592402cdf6f2cb15.1596199677.git.andreyknvl@google.com
Link: https://bugzilla.kernel.org/show_bug.cgi?id=203497
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/test_kasan.c: fix KASAN unit tests for tag-based KASAN

We use tag-based KASAN, then KASAN unit tests don't detect out-of-bounds
memory access. They need to be fixed.

With tag-based KASAN, the state of each 16 aligned bytes of memory is
encoded in one shadow byte and the shadow value is tag of pointer, so
we need to read next shadow byte, the shadow value is not equal to tag
value of pointer, so that tag-based KASAN will detect out-of-bounds
memory access.

[walter-zh.wu@mediatek.com: use KASAN_SHADOW_SCALE_SIZE instead of 13]
Link: http://lkml.kernel.org/r/20200708132524.11688-1-walter-zh.wu@mediatek.com
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Link: http://lkml.kernel.org/r/20200706115039.16750-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: remove kasan_unpoison_stack_above_sp_to()

kasan_unpoison_stack_above_sp_to() is defined in kasan code but never
used.  The function was introduced as part of the commit:

   commit 9f7d416c36124667 ("kprobes: Unpoison stack in jprobe_return() for KASAN")

... where it was necessary because x86's jprobe_return() would leave
stale shadow on the stack, and was an oddity in that regard.

Since then, jprobes were removed entirely, and as of commit:

  commit 80006dbee674f9fa ("kprobes/x86: Remove jprobe implementation")

... there have been no callers of this function.

Remove the declaration and the implementation.

Signed-off-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Link: http://lkml.kernel.org/r/20200706143505.23299-1-vincenzo.frascino@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: update documentation for generic kasan

Generic KASAN will support to record the last two call_rcu() call stacks
and print them in KASAN report.  So that need to update documentation.

Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Link: http://lkml.kernel.org/r/20200601051111.1359-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: add tests for call_rcu stack recording

Test call_rcu() call stack recording and verify whether it correctly is
printed in KASAN report.

Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Link: http://lkml.kernel.org/r/20200601051045.1294-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: record and print the free track

Move free track from kasan_alloc_meta to kasan_free_meta in order to make
struct kasan_alloc_meta and kasan_free_meta size are both 16 bytes.  It is
a good size because it is the minimal redzone size and a good number of
alignment.

For free track, we make some modifications as shown below:
1) Remove the free_track from struct kasan_alloc_meta.
2) Add the free_track into struct kasan_free_meta.
3) Add a macro KASAN_KMALLOC_FREETRACK in order to check whether
   it can print free stack in KASAN report.

[1]https://bugzilla.kernel.org/show_bug.cgi?id=198437

[walter-zh.wu@mediatek.com: build fix]
Link: http://lkml.kernel.org/r/20200710162440.23887-1-walter-zh.wu@mediatek.com
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Co-developed-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Link: http://lkml.kernel.org/r/20200601051022.1230-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rcu: kasan: record and print call_rcu() call stack

Patch series "kasan: memorize and print call_rcu stack", v8.

This patchset improves KASAN reports by making them to have call_rcu()
call stack information.  It is useful for programmers to solve
use-after-free or double-free memory issue.

The KASAN report was as follows(cleaned up slightly):

BUG: KASAN: use-after-free in kasan_rcu_reclaim+0x58/0x60

Freed by task 0:
 kasan_save_stack+0x24/0x50
 kasan_set_track+0x24/0x38
 kasan_set_free_info+0x18/0x20
 __kasan_slab_free+0x10c/0x170
 kasan_slab_free+0x10/0x18
 kfree+0x98/0x270
 kasan_rcu_reclaim+0x1c/0x60

Last call_rcu():
 kasan_save_stack+0x24/0x50
 kasan_record_aux_stack+0xbc/0xd0
 call_rcu+0x8c/0x580
 kasan_rcu_uaf+0xf4/0xf8

Generic KASAN will record the last two call_rcu() call stacks and print up
to 2 call_rcu() call stacks in KASAN report.  it is only suitable for
generic KASAN.

This feature considers the size of struct kasan_alloc_meta and
kasan_free_meta, we try to optimize the structure layout and size, lets it
get better memory consumption.

[1]https://bugzilla.kernel.org/show_bug.cgi?id=198437
[2]https://groups.google.com/forum/#!searchin/kasan-dev/better$20stack$20traces$20for$20rcu%7Csort:date/kasan-dev/KQsjT_88hDE/7rNUZprRBgAJ

This patch (of 4):

This feature will record the last two call_rcu() call stacks and prints up
to 2 call_rcu() call stacks in KASAN report.

When call_rcu() is called, we store the call_rcu() call stack into slub
alloc meta-data, so that the KASAN report can print rcu stack.

[1]https://bugzilla.kernel.org/show_bug.cgi?id=198437
[2]https://groups.google.com/forum/#!searchin/kasan-dev/better$20stack$20traces$20for$20rcu%7Csort:date/kasan-dev/KQsjT_88hDE/7rNUZprRBgAJ

[walter-zh.wu@mediatek.com: build fix]
Link: http://lkml.kernel.org/r/20200710162401.23816-1-walter-zh.wu@mediatek.com
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Link: http://lkml.kernel.org/r/20200710162123.23713-1-walter-zh.wu@mediatek.com
Link: http://lkml.kernel.org/r/20200601050847.1096-1-walter-zh.wu@mediatek.com
Link: http://lkml.kernel.org/r/20200601050927.1153-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: update required compiler versions in documentation

Updates the recently changed compiler requirements for KASAN.  In
particular, we require GCC >= 8.3.0, and add a note that Clang 11 supports
OOB detection of globals.

Fixes: 7b861a53e46b ("kasan: Bump required compiler version")
Fixes: acf7b0bf7dcf ("kasan: Fix required compiler version")
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Daniel Axtens <dja@axtens.net>
Link: http://lkml.kernel.org/r/20200629104157.3242503-2-elver@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kasan: improve and simplify Kconfig.kasan

Turn 'KASAN' into a menuconfig, to avoid cluttering its parent menu with
the suboptions if enabled.  Use 'if KASAN ...  endif' instead of having to
'depend on KASAN' for each entry.

Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Daniel Axtens <dja@axtens.net>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Link: http://lkml.kernel.org/r/20200629104157.3242503-1-elver@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c: remove BUG() from the find_va_links()

Get rid of BUG() macro, that should be used only when a critical situation
happens and a system is not able to function anymore.

Replace it with WARN() macro instead, dump some extra information about
start/end addresses of both VAs which overlap.  Such overlap data can help
to figure out what happened making further analysis easier.  For example
if both areas are identical it could mean a double free.

A recovery process consists of declining all further steps regarding
inserting of conflicting overlap range.  In that sense find_va_links() now
can return NULL, so its return value has to be checked by callers.

Side effect of such process is it can leak memory, but it is better than
just killing a machine for no good reason.  Apart of that a debugging
process can be done on alive system.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20200711104531.12242-1-urezki@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmalloc: remove redundant assignment in unmap_kernel_range_noflush()

'addr' is set to 'start' and then a few lines afterwards 'start' is set to
'addr'.  Remove the second asignment.

Fixes: 2ba3e6947aed ("mm/vmalloc: track which page-table levels were modified")
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Joerg Roedel <jroedel@suse.de>
Link: http://lkml.kernel.org/r/20200707163226.374685-1-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc: update the header about KVA rework

Reflect information about the author, date and year when the KVA rework
was done.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200622195821.4796-1-urezki@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc: switch to "propagate()" callback

An augment_tree_propagate_from() function uses its own implementation that
populates a tree from the specified node toward a root node.

On the other hand the RB_DECLARE_CALLBACKS_MAX macro provides the
"propagate()" callback that does exactly the same.  Having two similar
functions does not make sense and is redundant.

Reuse "built in" functionality to the macros.  So the code size gets
reduced.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200527205054.1696-3-urezki@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc: simplify augment_tree_propagate_check()

This function is for debug purpose only.  Currently it uses recursion for
tree traversal, checking an augmented value of each node to find out if it
is valid or not.

The recursion can corrupt the stack because the tree can be huge if
synthetic tests are applied.  To prevent it, navigate the tree from bottom
to upper levels using a regular list instead, because nodes are linked
among each other also.  It is faster and without recursion.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200527205054.1696-2-urezki@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc: simplify merge_or_add_vmap_area()

Currently when a VA is deallocated and is about to be placed back to the
tree, it can be either: merged with next/prev neighbors or inserted if not
coalesced.

On those steps the tree can be populated several times.  For example when
both neighbors are merged.  It can be avoided and simplified in fact.

Therefore do it only once when VA points to final merged area, after all
manipulations: merging/removing/inserting.

Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200527205054.1696-1-urezki@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc: convert to XArray

The radix tree of vmap blocks is simpler to express as an XArray.  Reduces
both the text and data sizes of the object file and eliminates a user of
the radix tree preload API.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Link: http://lkml.kernel.org/r/20200603171448.5894-1-willy@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: cleanup the code surrounding memory_present()

After removal of CONFIG_HAVE_MEMBLOCK_NODE_MAP we have two equivalent
functions that call memory_present() for each region in memblock.memory:
sparse_memory_present_with_active_regions() and membocks_present().

Moreover, all architectures have a call to either of these functions
preceding the call to sparse_init() and in the most cases they are called
one after the other.

Mark the regions from memblock.memory as present during sparce_init() by
making sparse_init() call memblocks_present(), make memblocks_present()
and memory_present() functions static and remove redundant
sparse_memory_present_with_active_regions() function.

Also remove no longer required HAVE_MEMORY_PRESENT configuration option.

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200712083130.22919-1-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: only sub-section aligned range would be populated

There are two code path which invoke __populate_section_memmap()

  * sparse_init_nid()
  * sparse_add_section()

For both case, we are sure the memory range is sub-section aligned.

  * we pass PAGES_PER_SECTION to sparse_init_nid()
  * we check range by check_pfn_span() before calling
    sparse_add_section()

Also, the counterpart of __populate_section_memmap(), we don't do such
calculation and check since the range is checked by check_pfn_span() in
__remove_pages().

Clear the calculation and check to keep it simple and comply with its
counterpart.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200703031828.14645-1-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: never partially remove memmap for early section

For early sections, its memmap is handled specially even sub-section is
enabled.  The memmap could only be populated as a whole.

Quoted from the comment of section_activate():

    * The early init code does not consider partially populated
    * initial sections, it simply assumes that memory will never be
    * referenced.  If we hot-add memory into such a section then we
    * do not need to populate the memmap and can simply reuse what
    * is already there.

While current section_deactivate() breaks this rule.  When hot-remove a
sub-section, section_deactivate() would depopulate its memmap.  The
consequence is if we hot-add this subsection again, its memmap never get
proper populated.

We can reproduce the case by following steps:

1. Hacking qemu to allow sub-section early section

:   diff --git a/hw/i386/pc.c b/hw/i386/pc.c
:   index 51b3050d01..c6a78d83c0 100644
:   --- a/hw/i386/pc.c
:   +++ b/hw/i386/pc.c
:   @@ -1010,7 +1010,7 @@ void pc_memory_init(PCMachineState *pcms,
:            }
:
:            machine->device_memory->base =
:   -            ROUND_UP(0x100000000ULL + x86ms->above_4g_mem_size, 1 * GiB);
:   +            0x100000000ULL + x86ms->above_4g_mem_size;
:
:            if (pcmc->enforce_aligned_dimm) {
:                /* size device region assuming 1G page max alignment per slot */

2. Bootup qemu with PSE disabled and a sub-section aligned memory size

   Part of the qemu command would look like this:

   sudo x86_64-softmmu/qemu-system-x86_64 \
       --enable-kvm -cpu host,pse=off \
       -m 4160M,maxmem=20G,slots=1 \
       -smp sockets=2,cores=16 \
       -numa node,nodeid=0,cpus=0-1 -numa node,nodeid=1,cpus=2-3 \
       -machine pc,nvdimm \
       -nographic \
       -object memory-backend-ram,id=mem0,size=8G \
       -device nvdimm,id=vm0,memdev=mem0,node=0,addr=0x144000000,label-size=128k

3. Re-config a pmem device with sub-section size in guest

   ndctl create-namespace --force --reconfig=namespace0.0 --mode=devdax --size=16M

Then you would see the following call trace:

   pmem0: detected capacity change from 0 to 16777216
   BUG: unable to handle page fault for address: ffffec73c51000b4
   #PF: supervisor write access in kernel mode
   #PF: error_code(0x0002) - not-present page
   PGD 81ff8067 P4D 81ff8067 PUD 81ff7067 PMD 1437cb067 PTE 0
   Oops: 0002 [#1] SMP NOPTI
   CPU: 16 PID: 1348 Comm: ndctl Kdump: loaded Tainted: G        W         5.8.0-rc2+ #24
   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.4
   RIP: 0010:memmap_init_zone+0x154/0x1c2
   Code: 77 16 f6 40 10 02 74 10 48 03 48 08 48 89 cb 48 c1 eb 0c e9 3a ff ff ff 48 89 df 48 c1 e7 06 48f
   RSP: 0018:ffffbdc7011a39b0 EFLAGS: 00010282
   RAX: ffffec73c5100088 RBX: 0000000000144002 RCX: 0000000000144000
   RDX: 0000000000000004 RSI: 007ffe0000000000 RDI: ffffec73c5100080
   RBP: 027ffe0000000000 R08: 0000000000000001 R09: ffff9f8d38f6d708
   R10: ffffec73c0000000 R11: 0000000000000000 R12: 0000000000000004
   R13: 0000000000000001 R14: 0000000000144200 R15: 0000000000000000
   FS:  00007efe6b65d780(0000) GS:ffff9f8d3f780000(0000) knlGS:0000000000000000
   CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
   CR2: ffffec73c51000b4 CR3: 000000007d718000 CR4: 0000000000340ee0
   Call Trace:
    move_pfn_range_to_zone+0x128/0x150
    memremap_pages+0x4e4/0x5a0
    devm_memremap_pages+0x1e/0x60
    dev_dax_probe+0x69/0x160 [device_dax]
    really_probe+0x298/0x3c0
    driver_probe_device+0xe1/0x150
    ? driver_allows_async_probing+0x50/0x50
    bus_for_each_drv+0x7e/0xc0
    __device_attach+0xdf/0x160
    bus_probe_device+0x8e/0xa0
    device_add+0x3b9/0x740
    __devm_create_dev_dax+0x127/0x1c0
    __dax_pmem_probe+0x1f2/0x219 [dax_pmem_core]
    dax_pmem_probe+0xc/0x1b [dax_pmem]
    nvdimm_bus_probe+0x69/0x1c0 [libnvdimm]
    really_probe+0x147/0x3c0
    driver_probe_device+0xe1/0x150
    device_driver_attach+0x53/0x60
    bind_store+0xd1/0x110
    kernfs_fop_write+0xce/0x1b0
    vfs_write+0xb6/0x1a0
    ksys_write+0x5f/0xe0
    do_syscall_64+0x4d/0x90
    entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: ba72b4c8cf60 ("mm/sparsemem: support sub-section hotplug")
Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Link: http://lkml.kernel.org/r/20200625223534.18024-1-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

selftests: add mincore() tests

Add a test suite for the mincore() syscall.  It tests most of its use
cases as well as its interface.

Tests implemented:

  - basic interface test
  - behavior on anonymous mappings
  - behavior on anonymous mappings with huge tlb pages
  - file-backed mapping with a regular file
  - file-backed mapping with a tmpfs file

Signed-off-by: Ricardo Cañuelo <ricardo.canuelo@collabora.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200728100450.4065-1-ricardo.canuelo@collabora.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mremap: start addresses are properly aligned

After previous cleanup, extent is the minimal step for both source and
destination.  This means when extent is HPAGE_PMD_SIZE or PMD_SIZE,
old_addr and new_addr are properly aligned too.

Since these two functions are only invoked in move_page_tables, it is safe
to remove the check now.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Thomas Hellstrom (VMware) <thomas_os@shipmail.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Link: http://lkml.kernel.org/r/20200708095028.41706-4-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mremap: calculate extent in one place

Page tables is moved on the base of PMD.  This requires both source and
destination range should meet the requirement.

Current code works well since move_huge_pmd() and move_normal_pmd() would
check old_addr and new_addr again.  And then return to move_ptes() if the
either of them is not aligned.

Instead of calculating the extent separately, it is better to calculate in
one place, so we know it is not necessary to try move pmd.  By doing so,
the logic seems a little clear.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Thomas Hellstrom (VMware) <thomas_os@shipmail.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Link: http://lkml.kernel.org/r/20200708095028.41706-3-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mremap: it is sure to have enough space when extent meets requirement

Patch series "mm/mremap: cleanup move_page_tables() a little", v5.

move_page_tables() tries to move page table by PMD or PTE.

The root reason is if it tries to move PMD, both old and new range should
be PMD aligned.  But current code calculate old range and new range
separately.  This leads to some redundant check and calculation.

This cleanup tries to consolidate the range check in one place to reduce
some extra range handling.

This patch (of 3):

old_end is passed to these two functions to check whether there is enough
space to do the move, while this check is done before invoking these
functions.

These two functions only would be invoked when extent meets the
requirement and there is one check before invoking these functions:

    if (extent > old_end - old_addr)
        extent = old_end - old_addr;

This implies (old_end - old_addr) won't fail the check in these two
functions.

Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Thomas Hellstrom (VMware) <thomas_os@shipmail.org>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Wei Yang <richard.weiyang@linux.alibaba.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Link: http://lkml.kernel.org/r/20200710092835.56368-1-richard.weiyang@linux.alibaba.com
Link: http://lkml.kernel.org/r/20200710092835.56368-2-richard.weiyang@linux.alibaba.com
Link: http://lkml.kernel.org/r/20200708095028.41706-1-richard.weiyang@linux.alibaba.com
Link: http://lkml.kernel.org/r/20200708095028.41706-2-richard.weiyang@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unnecessary wrapper function do_mmap_pgoff()

The current split between do_mmap() and do_mmap_pgoff() was introduced in
commit 1fcfd8db7f82 ("mm, mpx: add "vm_flags_t vm_flags" arg to
do_mmap_pgoff()") to support MPX.

The wrapper function do_mmap_pgoff() always passed 0 as the value of the
vm_flags argument to do_mmap().  However, MPX support has subsequently
been removed from the kernel and there were no more direct callers of
do_mmap(); all calls were going via do_mmap_pgoff().

Simplify the code by removing do_mmap_pgoff() and changing all callers to
directly call do_mmap(), which now no longer takes a vm_flags argument.

Signed-off-by: Peter Collingbourne <pcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200727194109.1371462-1-pcc@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mmap: merge vma after call_mmap() if possible

The vm_flags may be changed after call_mmap() because drivers may set some
flags for their own purpose.  As a result, we failed to merge the adjacent
vma due to the different vm_flags as userspace can't pass in the same one.
Try to merge vma after call_mmap() to fix this issue.

Signed-off-by: Hongxiang Lou <louhongxiang@huawei.com>
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1594954065-23733-1-git-send-email-linmiaohe@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arm64/mm: enable vmem_altmap support for vmemmap mappings

Device memory ranges when getting hot added into ZONE_DEVICE, might
require their vmemmap mapping's backing memory to be allocated from their
own range instead of consuming system memory.  This prevents large system
memory usage for potentially large device memory ranges.  Device driver
communicates this request via vmem_altmap structure.  Architecture needs
to take this request into account while creating and tearing down vemmmap
mappings.

This enables vmem_altmap support in vmemmap_populate() and vmemmap_free()
which includes vmemmap_populate_basepages() used for ARM64_16K_PAGES and
ARM64_64K_PAGES configs.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Jia He <justin.he@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Hsin-Yi Wang <hsinyi@chromium.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Tony Luck <tony.luck@intel.com>
Link: http://lkml.kernel.org/r/1594004178-8861-4-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparsemem: enable vmem_altmap support in vmemmap_alloc_block_buf()

There are many instances where vmemap allocation is often switched between
regular memory and device memory just based on whether altmap is available
or not.  vmemmap_alloc_block_buf() is used in various platforms to
allocate vmemmap mappings.  Lets also enable it to handle altmap based
device memory allocation along with existing regular memory allocations.
This will help in avoiding the altmap based allocation switch in many
places.  To summarize there are two different methods to call
vmemmap_alloc_block_buf().

vmemmap_alloc_block_buf(size, node, NULL)   /* Allocate from system RAM */
vmemmap_alloc_block_buf(size, node, altmap) /* Allocate from altmap */

This converts altmap_alloc_block_buf() into a static function, drops it's
entry from the header and updates Documentation/vm/memory-model.rst.

Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Jia He <justin.he@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Will Deacon <will@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.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: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Hsin-Yi Wang <hsinyi@chromium.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yu Zhao <yuzhao@google.com>
Link: http://lkml.kernel.org/r/1594004178-8861-3-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparsemem: enable vmem_altmap support in vmemmap_populate_basepages()

Patch series "arm64: Enable vmemmap mapping from device memory", v4.

This series enables vmemmap backing memory allocation from device memory
ranges on arm64.  But before that, it enables vmemmap_populate_basepages()
and vmemmap_alloc_block_buf() to accommodate struct vmem_altmap based
alocation requests.

This patch (of 3):

vmemmap_populate_basepages() is used across platforms to allocate backing
memory for vmemmap mapping.  This is used as a standard default choice or
as a fallback when intended huge pages allocation fails.  This just
creates entire vmemmap mapping with base pages (PAGE_SIZE).

On arm64 platforms, vmemmap_populate_basepages() is called instead of the
platform specific vmemmap_populate() when ARM64_SWAPPER_USES_SECTION_MAPS
is not enabled as in case for ARM64_16K_PAGES and ARM64_64K_PAGES configs.

At present vmemmap_populate_basepages() does not support allocating from
driver defined struct vmem_altmap while trying to create vmemmap mapping
for a device memory range.  It prevents ARM64_16K_PAGES and
ARM64_64K_PAGES configs on arm64 from supporting device memory with
vmemap_altmap request.

This enables vmem_altmap support in vmemmap_populate_basepages() unlocking
device memory allocation for vmemap mapping on arm64 platforms with 16K or
64K base page configs.

Each architecture should evaluate and decide on subscribing device memory
based base page allocation through vmemmap_populate_basepages().  Hence
lets keep it disabled on all archs in order to preserve the existing
semantics.  A subsequent patch enables it on arm64.

Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Jia He <justin.he@arm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Will Deacon <will@kernel.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Hsin-Yi Wang <hsinyi@chromium.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Link: http://lkml.kernel.org/r/1594004178-8861-1-git-send-email-anshuman.khandual@arm.com
Link: http://lkml.kernel.org/r/1594004178-8861-2-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: adjust vm_committed_as_batch according to vm overcommit policy

When checking a performance change for will-it-scale scalability mmap test
[1], we found very high lock contention for spinlock of percpu counter
'vm_committed_as':

    94.14%     0.35%  [kernel.kallsyms]         [k] _raw_spin_lock_irqsave
    48.21% _raw_spin_lock_irqsave;percpu_counter_add_batch;__vm_enough_memory;mmap_region;do_mmap;
    45.91% _raw_spin_lock_irqsave;percpu_counter_add_batch;__do_munmap;

Actually this heavy lock contention is not always necessary.  The
'vm_committed_as' needs to be very precise when the strict
OVERCOMMIT_NEVER policy is set, which requires a rather small batch number
for the percpu counter.

So keep 'batch' number unchanged for strict OVERCOMMIT_NEVER policy, and
lift it to 64X for OVERCOMMIT_ALWAYS and OVERCOMMIT_GUESS policies.  Also
add a sysctl handler to adjust it when the policy is reconfigured.

Benchmark with the same testcase in [1] shows 53% improvement on a 8C/16T
desktop, and 2097%(20X) on a 4S/72C/144T server.  We tested with test
platforms in 0day (server, desktop and laptop), and 80%+ platforms shows
improvements with that test.  And whether it shows improvements depends on
if the test mmap size is bigger than the batch number computed.

And if the lift is 16X, 1/3 of the platforms will show improvements,
though it should help the mmap/unmap usage generally, as Michal Hocko
mentioned:

: I believe that there are non-synthetic worklaods which would benefit from
: a larger batch.  E.g.  large in memory databases which do large mmaps
: during startups from multiple threads.

[1] https://lore.kernel.org/lkml/20200305062138.GI5972@shao2-debian/

Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Qian Cai <cai@lca.pw>
Cc: Kees Cook <keescook@chromium.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/1589611660-89854-4-git-send-email-feng.tang@intel.com
Link: http://lkml.kernel.org/r/1592725000-73486-4-git-send-email-feng.tang@intel.com
Link: http://lkml.kernel.org/r/1594389708-60781-5-git-send-email-feng.tang@intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

percpu_counter: add percpu_counter_sync()

percpu_counter's accuracy is related to its batch size.  For a
percpu_counter with a big batch, its deviation could be big, so when the
counter's batch is runtime changed to a smaller value for better accuracy,
there could also be requirment to reduce the big deviation.

So add a percpu-counter sync function to be run on each CPU.

Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Tim Chen <tim.c.chen@intel.com>
Link: http://lkml.kernel.org/r/1594389708-60781-4-git-send-email-feng.tang@intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/util.c: make vm_memory_committed() more accurate

percpu_counter_sum_positive() will provide more accurate info.

As with percpu_counter_read_positive(), in worst case the deviation could
be 'batch * nr_cpus', which is totalram_pages/256 for now, and will be
more when the batch gets enlarged.

Its time cost is about 800 nanoseconds on a 2C/4T platform and 2~3
microseconds on a 2S/36C/72T Skylake server in normal case, and in worst
case where vm_committed_as's spinlock is under severe contention, it costs
30~40 microseconds for the 2S/36C/72T Skylake sever, which should be fine
for its only two users: /proc/meminfo and HyperV balloon driver's status
trace per second.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com> # for /proc/meminfo
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Qian Cai <cai@lca.pw>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/1592725000-73486-3-git-send-email-feng.tang@intel.com
Link: http://lkml.kernel.org/r/1594389708-60781-3-git-send-email-feng.tang@intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc/meminfo: avoid open coded reading of vm_committed_as

Patch series "make vm_committed_as_batch aware of vm overcommit policy", v6.

When checking a performance change for will-it-scale scalability mmap test
[1], we found very high lock contention for spinlock of percpu counter
'vm_committed_as':

    94.14%     0.35%  [kernel.kallsyms]         [k] _raw_spin_lock_irqsave
    48.21% _raw_spin_lock_irqsave;percpu_counter_add_batch;__vm_enough_memory;mmap_region;do_mmap;
    45.91% _raw_spin_lock_irqsave;percpu_counter_add_batch;__do_munmap;

Actually this heavy lock contention is not always necessary.  The
'vm_committed_as' needs to be very precise when the strict
OVERCOMMIT_NEVER policy is set, which requires a rather small batch number
for the percpu counter.

So keep 'batch' number unchanged for strict OVERCOMMIT_NEVER policy, and
enlarge it for not-so-strict OVERCOMMIT_ALWAYS and OVERCOMMIT_GUESS
policies.

Benchmark with the same testcase in [1] shows 53% improvement on a 8C/16T
desktop, and 2097%(20X) on a 4S/72C/144T server.  And for that case,
whether it shows improvements depends on if the test mmap size is bigger
than the batch number computed.

We tested 10+ platforms in 0day (server, desktop and laptop).  If we lift
it to 64X, 80%+ platforms show improvements, and for 16X lift, 1/3 of the
platforms will show improvements.

And generally it should help the mmap/unmap usage,as Michal Hocko
mentioned:

: I believe that there are non-synthetic worklaods which would benefit
: from a larger batch. E.g. large in memory databases which do large
: mmaps during startups from multiple threads.

Note: There are some style complain from checkpatch for patch 4, as sysctl
handler declaration follows the similar format of sibling functions

[1] https://lore.kernel.org/lkml/20200305062138.GI5972@shao2-debian/

This patch (of 4):

Use the existing vm_memory_committed() instead, which is also convenient
for future change.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Qian Cai <cai@lca.pw>
Cc: Kees Cook <keescook@chromium.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/1594389708-60781-1-git-send-email-feng.tang@intel.com
Link: http://lkml.kernel.org/r/1594389708-60781-2-git-send-email-feng.tang@intel.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmap: optimize a branch judgment in ksys_mmap_pgoff()

Look at the pseudo code below.  It's very clear that, the judgement
"!is_file_hugepages(file)" at 3) is duplicated to the one at 1), we can
use "else if" to avoid it.  And the assignment "retval = -EINVAL" at 2) is
only needed by the branch 3), because "retval" will be overwritten at 4).

No functional change, but it can reduce the code size. Maybe more clearer?
Before:
text    data     bss     dec     hex filename
28733    1590       1   30324    7674 mm/mmap.o

After:
text    data     bss     dec     hex filename
28701    1590       1   30292    7654 mm/mmap.o

====pseudo code====:
if (!(flags & MAP_ANONYMOUS)) {
...
1) if (is_file_hugepages(file))
len = ALIGN(len, huge_page_size(hstate_file(file)));
2) retval = -EINVAL;
3) if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
goto out_fput;
} else if (flags & MAP_HUGETLB) {
...
}
...

4) retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
out_fput:
...
return retval;

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200705080112.1405-1-thunder.leizhen@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move p?d_alloc_track to separate header file

The functions are only used in two source files, so there is no need for
them to be in the global <linux/mm.h> header.  Move them to the new
<linux/pgalloc-track.h> header and include it only where needed.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200609120533.25867-1-joro@8bytes.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move lib/ioremap.c to mm/

The functionality in lib/ioremap.c deals with pagetables, vmalloc and
caches, so it naturally belongs to mm/ Moving it there will also allow
declaring p?d_alloc_track functions in an header file inside mm/ rather
than having those declarations in include/linux/mm.h

Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-8-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

asm-generic: pgalloc: provide generic pgd_free()

Most architectures define pgd_free() as a wrapper for free_page().

Provide a generic version in asm-generic/pgalloc.h and enable its use for
most architectures.

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k]
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-7-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

asm-generic: pgalloc: provide generic pud_alloc_one() and pud_free_one()

Several architectures define pud_alloc_one() as a wrapper for
__get_free_page() and pud_free() as a wrapper for free_page().

Provide a generic implementation in asm-generic/pgalloc.h and use it where
appropriate.

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-6-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

asm-generic: pgalloc: provide generic pmd_alloc_one() and pmd_free_one()

For most architectures that support >2 levels of page tables,
pmd_alloc_one() is a wrapper for __get_free_pages(), sometimes with
__GFP_ZERO and sometimes followed by memset(0) instead.

More elaborate versions on arm64 and x86 account memory for the user page
tables and call to pgtable_pmd_page_ctor() as the part of PMD page
initialization.

Move the arm64 version to include/asm-generic/pgalloc.h and use the
generic version on several architectures.

The pgtable_pmd_page_ctor() is a NOP when ARCH_ENABLE_SPLIT_PMD_PTLOCK is
not enabled, so there is no functional change for most architectures
except of the addition of __GFP_ACCOUNT for allocation of user page
tables.

The pmd_free() is a wrapper for free_page() in all the cases, so no
functional change here.

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-5-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

xtensa: switch to generic version of pte allocation

xtensa clears PTEs during allocation of the page tables and pte_clear()
sets the PTE to a non-zero value.  Splitting ptes_clear() helper out of
pte_alloc_one() and pte_alloc_one_kernel() allows reuse of base generic
allocation methods (__pte_alloc_one() and __pte_alloc_one_kernel()) and
the common GFP mask for page table allocations.

The pte_free() and pte_free_kernel() implementations on xtensa are
identical to the generic ones and can be dropped.

[jcmvbkbc@gmail.com: xtensa: fix closing endif comment]
Link: http://lkml.kernel.org/r/20200721024751.1257-1-jcmvbkbc@gmail.com
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-4-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

opeinrisc: switch to generic version of pte allocation

Replace pte_alloc_one(), pte_free() and pte_free_kernel() with the generic
implementation.  The only actual functional change is the addition of
__GFP_ACCOUT for the allocation of the user page tables.

The pte_alloc_one_kernel() is kept back because its implementation on
openrisc is different than the generic one.

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Acked-by: Stafford Horne <shorne@gmail.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-3-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unneeded includes of <asm/pgalloc.h>

Patch series "mm: cleanup usage of <asm/pgalloc.h>"

Most architectures have very similar versions of pXd_alloc_one() and
pXd_free_one() for intermediate levels of page table.  These patches add
generic versions of these functions in <asm-generic/pgalloc.h> and enable
use of the generic functions where appropriate.

In addition, functions declared and defined in <asm/pgalloc.h> headers are
used mostly by core mm and early mm initialization in arch and there is no
actual reason to have the <asm/pgalloc.h> included all over the place.
The first patch in this series removes unneeded includes of
<asm/pgalloc.h>

In the end it didn't work out as neatly as I hoped and moving
pXd_alloc_track() definitions to <asm-generic/pgalloc.h> would require
unnecessary changes to arches that have custom page table allocations, so
I've decided to move lib/ioremap.c to mm/ and make pgalloc-track.h local
to mm/.

This patch (of 8):

In most cases <asm/pgalloc.h> header is required only for allocations of
page table memory.  Most of the .c files that include that header do not
use symbols declared in <asm/pgalloc.h> and do not require that header.

As for the other header files that used to include <asm/pgalloc.h>, it is
possible to move that include into the .c file that actually uses symbols
from <asm/pgalloc.h> and drop the include from the header file.

The process was somewhat automated using

sed -i -E '/[<"]asm\/pgalloc\.h/d' \
                $(grep -L -w -f /tmp/xx \
                        $(git grep -E -l '[<"]asm/pgalloc\.h'))

where /tmp/xx contains all the symbols defined in
arch/*/include/asm/pgalloc.h.

[rppt@linux.ibm.com: fix powerpc warning]

Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k]
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Matthew Wilcox <willy@infradead.org>
Link: http://lkml.kernel.org/r/20200627143453.31835-1-rppt@kernel.org
Link: http://lkml.kernel.org/r/20200627143453.31835-2-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory.c: make remap_pfn_range() reject unaligned addr

This function implicitly assumes that the addr passed in is page aligned.
A non page aligned addr could ultimately cause a kernel bug in
remap_pte_range as the exit condition in the logic loop may never be
satisfied.  This patch documents the need for the requirement, as well as
explicitly adds a check for it.

Signed-off-by: Alex Zhang <zhangalex@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200617233512.177519-1-zhangalex@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove redundant check non_swap_entry()

In zap_pte_range(), the check for non_swap_entry() and
is_device_private_entry() is unnecessary since the latter is sufficient to
determine if the page is a device private page.  Remove the test for
non_swap_entry() to simplify the code and for clarity.

Signed-off-by: Ralph Campbell <rcampbell@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Acked-by: David Hildenbrand <david@redhat.com>
Link: http://lkml.kernel.org/r/20200615175405.4613-1-rcampbell@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_counter.c: fix protection usage propagation

When workload runs in cgroups that aren't directly below root cgroup and
their parent specifies reclaim protection, it may end up ineffective.

The reason is that propagate_protected_usage() is not called in all
hierarchy up.  All the protected usage is incorrectly accumulated in the
workload's parent.  This means that siblings_low_usage is overestimated
and effective protection underestimated.  Even though it is transitional
phenomenon (uncharge path does correct propagation and fixes the wrong
children_low_usage), it can undermine the intended protection
unexpectedly.

We have noticed this problem while seeing a swap out in a descendant of a
protected memcg (intermediate node) while the parent was conveniently
under its protection limit and the memory pressure was external to that
hierarchy.  Michal has pinpointed this down to the wrong
siblings_low_usage which led to the unwanted reclaim.

The fix is simply updating children_low_usage in respective ancestors also
in the charging path.

Fixes: 230671533d64 ("mm: memory.low hierarchical behavior")
Signed-off-by: Michal Koutný <mkoutny@suse.com>
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: <stable@vger.kernel.org> [4.18+]
Link: http://lkml.kernel.org/r/20200803153231.15477-1-mhocko@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: don't count limit-setting reclaim as memory pressure

When an outside process lowers one of the memory limits of a cgroup (or
uses the force_empty knob in cgroup1), direct reclaim is performed in the
context of the write(), in order to directly enforce the new limit and
have it being met by the time the write() returns.

Currently, this reclaim activity is accounted as memory pressure in the
cgroup that the writer(!) belongs to.  This is unexpected.  It
specifically causes problems for senpai
(https://github.com/facebookincubator/senpai), which is an agent that
routinely adjusts the memory limits and performs associated reclaim work
in tens or even hundreds of cgroups running on the host.  The cgroup that
senpai is running in itself will report elevated levels of memory
pressure, even though it itself is under no memory shortage or any sort of
distress.

Move the psi annotation from the central cgroup reclaim function to
callsites in the allocation context, and thereby no longer count any
limit-setting reclaim as memory pressure.  If the newly set limit causes
the workload inside the cgroup into direct reclaim, that of course will
continue to count as memory pressure.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Roman Gushchin <guro@fb.com>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200728135210.379885-2-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: restore proper dirty throttling when memory.high changes

Commit 8c8c383c04f6 ("mm: memcontrol: try harder to set a new
memory.high") inadvertently removed a callback to recalculate the
writeback cache size in light of a newly configured memory.high limit.

Without letting the writeback cache know about a potentially heavily
reduced limit, it may permit too many dirty pages, which can cause
unnecessary reclaim latencies or even avoidable OOM situations.

This was spotted while reading the code, it hasn't knowingly caused any
problems in practice so far.

Fixes: 8c8c383c04f6 ("mm: memcontrol: try harder to set a new memory.high")
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/20200728135210.379885-1-hannes@cmpxchg.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg, oom: check memcg margin for parallel oom

Memcg oom killer invocation is synchronized by the global oom_lock and
tasks are sleeping on the lock while somebody is selecting the victim or
potentially race with the oom_reaper is releasing the victim's memory.
This can result in a pointless oom killer invocation because a waiter
might be racing with the oom_reaper

        P1              oom_reaper              P2
                        oom_reap_task           mutex_lock(oom_lock)
                                                out_of_memory # no victim because we have one already
                        __oom_reap_task_mm      mute_unlock(oom_lock)
 mutex_lock(oom_lock)
                        set MMF_OOM_SKIP
 select_bad_process
 # finds a new victim

The page allocator prevents from this race by trying to allocate after the
lock can be acquired (in __alloc_pages_may_oom) which acts as a last
minute check.  Moreover page allocator simply doesn't block on the
oom_lock and simply retries the whole reclaim process.

Memcg oom killer should do the last minute check as well.  Call
mem_cgroup_margin to do that.  Trylock on the oom_lock could be done as
well but this doesn't seem to be necessary at this stage.

[mhocko@kernel.org: commit log]

Suggested-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Chris Down <chris@chrisdown.name>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Link: http://lkml.kernel.org/r/1594735034-19190-1-git-send-email-laoar.shao@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: decouple e{low,min} state mutations from protection checks

mem_cgroup_protected currently is both used to set effective low and min
and return a mem_cgroup_protection based on the result.  As a user, this
can be a little unexpected: it appears to be a simple predicate function,
if not for the big warning in the comment above about the order in which
it must be executed.

This change makes it so that we separate the state mutations from the
actual protection checks, which makes it more obvious where we need to be
careful mutating internal state, and where we are simply checking and
don't need to worry about that.

[mhocko@suse.com - don't check protection on root memcgs]

Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Yafang Shao <laoar.shao@gmail.com>
Link: http://lkml.kernel.org/r/ff3f915097fcee9f6d7041c084ef92d16aaeb56a.1594638158.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: avoid stale protection values when cgroup is above protection

Patch series "mm, memcg: memory.{low,min} reclaim fix & cleanup", v4.

This series contains a fix for a edge case in my earlier protection
calculation patches, and a patch to make the area overall a little more
robust to hopefully help avoid this in future.

This patch (of 2):

A cgroup can have both memory protection and a memory limit to isolate it
from its siblings in both directions - for example, to prevent it from
being shrunk below 2G under high pressure from outside, but also from
growing beyond 4G under low pressure.

Commit 9783aa9917f8 ("mm, memcg: proportional memory.{low,min} reclaim")
implemented proportional scan pressure so that multiple siblings in excess
of their protection settings don't get reclaimed equally but instead in
accordance to their unprotected portion.

During limit reclaim, this proportionality shouldn't apply of course:
there is no competition, all pressure is from within the cgroup and should
be applied as such.  Reclaim should operate at full efficiency.

However, mem_cgroup_protected() never expected anybody to look at the
effective protection values when it indicated that the cgroup is above its
protection.  As a result, a query during limit reclaim may return stale
protection values that were calculated by a previous reclaim cycle in
which the cgroup did have siblings.

When this happens, reclaim is unnecessarily hesitant and potentially slow
to meet the desired limit.  In theory this could lead to premature OOM
kills, although it's not obvious this has occurred in practice.

Workaround the problem by special casing reclaim roots in
mem_cgroup_protection.  These memcgs are never participating in the
reclaim protection because the reclaim is internal.

We have to ignore effective protection values for reclaim roots because
mem_cgroup_protected might be called from racing reclaim contexts with
different roots.  Calculation is relying on root -> leaf tree traversal
therefore top-down reclaim protection invariants should hold.  The only
exception is the reclaim root which should have effective protection set
to 0 but that would be problematic for the following setup:

 Let's have global and A's reclaim in parallel:
  |
  A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
  |\
  | C (low = 1G, usage = 2.5G)
  B (low = 1G, usage = 0.5G)

 for A reclaim we have
 B.elow = B.low
 C.elow = C.low

 For the global reclaim
 A.elow = A.low
 B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
 C.elow = min(C.usage, C.low)

 With the effective values resetting we have A reclaim
 A.elow = 0
 B.elow = B.low
 C.elow = C.low

 and global reclaim could see the above and then
 B.elow = C.elow = 0 because children_low_usage > A.elow

Which means that protected memcgs would get reclaimed.

In future we would like to make mem_cgroup_protected more robust against
racing reclaim contexts but that is likely more complex solution than this
simple workaround.

[hannes@cmpxchg.org - large part of the changelog]
[mhocko@suse.com - workaround explanation]
[chris@chrisdown.name - retitle]

Fixes: 9783aa9917f8 ("mm, memcg: proportional memory.{low,min} reclaim")
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/cover.1594638158.git.chris@chrisdown.name
Link: http://lkml.kernel.org/r/044fb8ecffd001c7905d27c0c2ad998069fdc396.1594638158.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: unify reclaim retry limits with page allocator

Reclaim retries have been set to 5 since the beginning of time in
commit 66e1707bc346 ("Memory controller: add per cgroup LRU and
reclaim").  However, we now have a generally agreed-upon standard for
page reclaim: MAX_RECLAIM_RETRIES (currently 16), added many years later
in commit 0a0337e0d1d1 ("mm, oom: rework oom detection").

In the absence of a compelling reason to declare an OOM earlier in memcg
context than page allocator context, it seems reasonable to supplant
MEM_CGROUP_RECLAIM_RETRIES with MAX_RECLAIM_RETRIES, making the page
allocator and memcg internals more similar in semantics when reclaim
fails to produce results, avoiding premature OOMs or throttling.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/da557856c9c7654308eaff4eedc1952a95e8df5f.1594640214.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: reclaim more aggressively before high allocator throttling

Patch series "mm, memcg: reclaim harder before high throttling", v2.

This patch (of 2):

In Facebook production, we've seen cases where cgroups have been put into
allocator throttling even when they appear to have a lot of slack file
caches which should be trivially reclaimable.

Looking more closely, the problem is that we only try a single cgroup
reclaim walk for each return to usermode before calculating whether or not
we should throttle.  This single attempt doesn't produce enough pressure
to shrink for cgroups with a rapidly growing amount of file caches prior
to entering allocator throttling.

As an example, we see that threads in an affected cgroup are stuck in
allocator throttling:

    # for i in $(cat cgroup.threads); do
    >     grep over_high "/proc/$i/stack"
    > done
    [<0>] mem_cgroup_handle_over_high+0x10b/0x150
    [<0>] mem_cgroup_handle_over_high+0x10b/0x150
    [<0>] mem_cgroup_handle_over_high+0x10b/0x150

...however, there is no I/O pressure reported by PSI, despite a lot of
slack file pages:

    # cat memory.pressure
    some avg10=78.50 avg60=84.99 avg300=84.53 total=5702440903
    full avg10=78.50 avg60=84.99 avg300=84.53 total=5702116959
    # cat io.pressure
    some avg10=0.00 avg60=0.00 avg300=0.00 total=78051391
    full avg10=0.00 avg60=0.00 avg300=0.00 total=78049640
    # grep _file memory.stat
    inactive_file 1370939392
    active_file 661635072

This patch changes the behaviour to retry reclaim either until the current
task goes below the 10ms grace period, or we are making no reclaim
progress at all.  In the latter case, we enter reclaim throttling as
before.

To a user, there's no intuitive reason for the reclaim behaviour to differ
from hitting memory.high as part of a new allocation, as opposed to
hitting memory.high because someone lowered its value.  As such this also
brings an added benefit: it unifies the reclaim behaviour between the two.

There's precedent for this behaviour: we already do reclaim retries when
writing to memory.{high,max}, in max reclaim, and in the page allocator
itself.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Roman Gushchin <guro@fb.com>
Link: http://lkml.kernel.org/r/cover.1594640214.git.chris@chrisdown.name
Link: http://lkml.kernel.org/r/a4e23b59e9ef499b575ae73a8120ee089b7d3373.1594640214.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: avoid workload stalls when lowering memory.high

Memory.high limit is implemented in a way such that the kernel penalizes
all threads which are allocating a memory over the limit.  Forcing all
threads into the synchronous reclaim and adding some artificial delays
allows to slow down the memory consumption and potentially give some time
for userspace oom handlers/resource control agents to react.

It works nicely if the memory usage is hitting the limit from below,
however it works sub-optimal if a user adjusts memory.high to a value way
below the current memory usage.  It basically forces all workload threads
(doing any memory allocations) into the synchronous reclaim and sleep.
This makes the workload completely unresponsive for a long period of time
and can also lead to a system-wide contention on lru locks.  It can happen
even if the workload is not actually tight on memory and has, for example,
a ton of cold pagecache.

In the current implementation writing to memory.high causes an atomic
update of page counter's high value followed by an attempt to reclaim
enough memory to fit into the new limit.  To fix the problem described
above, all we need is to change the order of execution: try to push the
memory usage under the limit first, and only then set the new high limit.

Reported-by: Domas Mituzas <domas@fb.com>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Chris Down <chris@chrisdown.name>
Link: http://lkml.kernel.org/r/20200709194718.189231-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: switch to static_branch_likely() in memcg_kmem_enabled()

Currently memcg_kmem_enabled() is optimized for the kernel memory
accounting being off.  It was so for a long time, and arguably the reason
behind was that the kernel memory accounting was initially an opt-in
feature.  However, now it's on by default on both cgroup v1 and cgroup v2,
and it's on for all cgroups.  So let's switch over to
static_branch_likely() to reflect this fact.

Unlikely there is a significant performance difference, as the cost of a
memory allocation and its accounting significantly exceeds the cost of a
jump.  However, the conversion makes the code look more logically.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pekka Enberg <penberg@kernel.org>
Link: http://lkml.kernel.org/r/20200707173612.124425-3-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: slab: rename (un)charge_slab_page() to (un)account_slab_page()

charge_slab_page() and uncharge_slab_page() are not related anymore to
memcg charging and uncharging.  In order to make their names less
confusing, let's rename them to account_slab_page() and
unaccount_slab_page() respectively.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pekka Enberg <penberg@kernel.org>
Link: http://lkml.kernel.org/r/20200707173612.124425-2-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: remove unused argument by charge_slab_page()

charge_slab_page() is not using the gfp argument anymore,
remove it.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Link: http://lkml.kernel.org/r/20200707173612.124425-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: account kernel stack per node

Currently the kernel stack is being accounted per-zone.  There is no need
to do that.  In addition due to being per-zone, memcg has to keep a
separate MEMCG_KERNEL_STACK_KB.  Make the stat per-node and deprecate
MEMCG_KERNEL_STACK_KB as memcg_stat_item is an extension of
node_stat_item.  In addition localize the kernel stack stats updates to
account_kernel_stack().

Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Link: http://lkml.kernel.org/r/20200630161539.1759185-1-shakeelb@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/cgroup: add memcg_slabinfo.py tool

Add a drgn-based tool to display slab information for a given memcg.  Can
replace cgroup v1 memory.kmem.slabinfo interface on cgroup v2, but in a
more flexiable way.

Currently supports only SLUB configuration, but SLAB can be trivially
added later.

Output example:
$ sudo ./tools/cgroup/memcg_slabinfo.py /sys/fs/cgroup/user.slice/user-111017.slice/user\@111017.service
shmem_inode_cache     92     92    704   46    8 : tunables    0    0    0 : slabdata      2      2      0
eventpoll_pwq         56     56     72   56    1 : tunables    0    0    0 : slabdata      1      1      0
eventpoll_epi         32     32    128   32    1 : tunables    0    0    0 : slabdata      1      1      0
kmalloc-8              0      0      8  512    1 : tunables    0    0    0 : slabdata      0      0      0
kmalloc-96             0      0     96   42    1 : tunables    0    0    0 : slabdata      0      0      0
kmalloc-2048           0      0   2048   16    8 : tunables    0    0    0 : slabdata      0      0      0
kmalloc-64           128    128     64   64    1 : tunables    0    0    0 : slabdata      2      2      0
mm_struct            160    160   1024   32    8 : tunables    0    0    0 : slabdata      5      5      0
signal_cache          96     96   1024   32    8 : tunables    0    0    0 : slabdata      3      3      0
sighand_cache         45     45   2112   15    8 : tunables    0    0    0 : slabdata      3      3      0
files_cache          138    138    704   46    8 : tunables    0    0    0 : slabdata      3      3      0
task_delay_info      153    153     80   51    1 : tunables    0    0    0 : slabdata      3      3      0
task_struct           27     27   3520    9    8 : tunables    0    0    0 : slabdata      3      3      0
radix_tree_node       56     56    584   28    4 : tunables    0    0    0 : slabdata      2      2      0
btrfs_inode          140    140   1136   28    8 : tunables    0    0    0 : slabdata      5      5      0
kmalloc-1024          64     64   1024   32    8 : tunables    0    0    0 : slabdata      2      2      0
kmalloc-192           84     84    192   42    2 : tunables    0    0    0 : slabdata      2      2      0
inode_cache           54     54    600   27    4 : tunables    0    0    0 : slabdata      2      2      0
kmalloc-128            0      0    128   32    1 : tunables    0    0    0 : slabdata      0      0      0
kmalloc-512           32     32    512   32    4 : tunables    0    0    0 : slabdata      1      1      0
skbuff_head_cache     32     32    256   32    2 : tunables    0    0    0 : slabdata      1      1      0
sock_inode_cache      46     46    704   46    8 : tunables    0    0    0 : slabdata      1      1      0
cred_jar             378    378    192   42    2 : tunables    0    0    0 : slabdata      9      9      0
proc_inode_cache      96     96    672   24    4 : tunables    0    0    0 : slabdata      4      4      0
dentry               336    336    192   42    2 : tunables    0    0    0 : slabdata      8      8      0
filp                 697    864    256   32    2 : tunables    0    0    0 : slabdata     27     27      0
anon_vma             644    644     88   46    1 : tunables    0    0    0 : slabdata     14     14      0
pid                 1408   1408     64   64    1 : tunables    0    0    0 : slabdata     22     22      0
vm_area_struct      1200   1200    200   40    2 : tunables    0    0    0 : slabdata     30     30      0

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200623174037.3951353-20-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kselftests: cgroup: add kernel memory accounting tests

Add some tests to cover the kernel memory accounting functionality.  These
are covering some issues (and changes) we had recently.

1) A test which allocates a lot of negative dentries, checks memcg slab
   statistics, creates memory pressure by setting memory.max to some low
   value and checks that some number of slabs was reclaimed.

2) A test which covers side effects of memcg destruction: it creates
   and destroys a large number of sub-cgroups, each containing a
   multi-threaded workload which allocates and releases some kernel
   memory.  Then it checks that the charge ans memory.stats do add up on
   the parent level.

3) A test which reads /proc/kpagecgroup and implicitly checks that it
   doesn't crash the system.

4) A test which spawns a large number of threads and checks that the
   kernel stacks accounting works as expected.

5) A test which checks that living charged slab objects are not
   preventing the memory cgroup from being released after being deleted by
   a user.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200623174037.3951353-19-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: use a single set of kmem_caches for all allocations

Instead of having two sets of kmem_caches: one for system-wide and
non-accounted allocations and the second one shared by all accounted
allocations, we can use just one.

The idea is simple: space for obj_cgroup metadata can be allocated on
demand and filled only for accounted allocations.

It allows to remove a bunch of code which is required to handle kmem_cache
clones for accounted allocations.  There is no more need to create them,
accumulate statistics, propagate attributes, etc.  It's a quite
significant simplification.

Also, because the total number of slab_caches is reduced almost twice (not
all kmem_caches have a memcg clone), some additional memory savings are
expected.  On my devvm it additionally saves about 3.5% of slab memory.

[guro@fb.com: fix build on MIPS]
Link: http://lkml.kernel.org/r/20200717214810.3733082-1-guro@fb.com
Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Naresh Kamboju <naresh.kamboju@linaro.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-18-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: remove redundant check in memcg_accumulate_slabinfo()

memcg_accumulate_slabinfo() is never called with a non-root kmem_cache as
a first argument, so the is_root_cache(s) check is redundant and can be
removed without any functional change.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-17-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: deprecate slab_root_caches

Currently there are two lists of kmem_caches:
1) slab_caches, which contains all kmem_caches,
2) slab_root_caches, which contains only root kmem_caches.

And there is some preprocessor magic to have a single list if
CONFIG_MEMCG_KMEM isn't enabled.

It was required earlier because the number of non-root kmem_caches was
proportional to the number of memory cgroups and could reach really big
values.  Now, when it cannot exceed the number of root kmem_caches, there
is really no reason to maintain two lists.

We never iterate over the slab_root_caches list on any hot paths, so it's
perfectly fine to iterate over slab_caches and filter out non-root
kmem_caches.

It allows to remove a lot of config-dependent code and two pointers from
the kmem_cache structure.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-16-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: remove memcg_kmem_get_cache()

The memcg_kmem_get_cache() function became really trivial, so let's just
inline it into the single call point: memcg_slab_pre_alloc_hook().

It will make the code less bulky and can also help the compiler to
generate a better code.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-15-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: simplify memcg cache creation

Because the number of non-root kmem_caches doesn't depend on the number of
memory cgroups anymore and is generally not very big, there is no more
need for a dedicated workqueue.

Also, as there is no more need to pass any arguments to the
memcg_create_kmem_cache() except the root kmem_cache, it's possible to
just embed the work structure into the kmem_cache and avoid the dynamic
allocation of the work structure.

This will also simplify the synchronization: for each root kmem_cache
there is only one work.  So there will be no more concurrent attempts to
create a non-root kmem_cache for a root kmem_cache: the second and all
following attempts to queue the work will fail.

On the kmem_cache destruction path there is no more need to call the
expensive flush_workqueue() and wait for all pending works to be finished.
Instead, cancel_work_sync() can be used to cancel/wait for only one work.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-14-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: use a single set of kmem_caches for all accounted allocations

This is fairly big but mostly red patch, which makes all accounted slab
allocations use a single set of kmem_caches instead of creating a separate
set for each memory cgroup.

Because the number of non-root kmem_caches is now capped by the number of
root kmem_caches, there is no need to shrink or destroy them prematurely.
They can be perfectly destroyed together with their root counterparts.
This allows to dramatically simplify the management of non-root
kmem_caches and delete a ton of code.

This patch performs the following changes:
1) introduces memcg_params.memcg_cache pointer to represent the
   kmem_cache which will be used for all non-root allocations
2) reuses the existing memcg kmem_cache creation mechanism
   to create memcg kmem_cache on the first allocation attempt
3) memcg kmem_caches are named <kmemcache_name>-memcg,
   e.g. dentry-memcg
4) simplifies memcg_kmem_get_cache() to just return memcg kmem_cache
   or schedule it's creation and return the root cache
5) removes almost all non-root kmem_cache management code
   (separate refcounter, reparenting, shrinking, etc)
6) makes slab debugfs to display root_mem_cgroup css id and never
   show :dead and :deact flags in the memcg_slabinfo attribute.

Following patches in the series will simplify the kmem_cache creation.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-13-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: move memcg_kmem_bypass() to memcontrol.h

To make the memcg_kmem_bypass() function available outside of the
memcontrol.c, let's move it to memcontrol.h.  The function is small and
nicely fits into static inline sort of functions.

It will be used from the slab code.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-12-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: deprecate memory.kmem.slabinfo

Deprecate memory.kmem.slabinfo.

An empty file will be presented if corresponding config options are
enabled.

The interface is implementation dependent, isn't present in cgroup v2, and
is generally useful only for core mm debugging purposes.  In other words,
it doesn't provide any value for the absolute majority of users.

A drgn-based replacement can be found in
tools/cgroup/memcg_slabinfo.py.  It does support cgroup v1 and v2,
mimics memory.kmem.slabinfo output and also allows to get any
additional information without a need to recompile the kernel.

If a drgn-based solution is too slow for a task, a bpf-based tracing tool
can be used, which can easily keep track of all slab allocations belonging
to a memory cgroup.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-11-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: charge individual slab objects instead of pages

Switch to per-object accounting of non-root slab objects.

Charging is performed using obj_cgroup API in the pre_alloc hook.
Obj_cgroup is charged with the size of the object and the size of
metadata: as now it's the size of an obj_cgroup pointer.  If the amount of
memory has been charged successfully, the actual allocation code is
executed.  Otherwise, -ENOMEM is returned.

In the post_alloc hook if the actual allocation succeeded, corresponding
vmstats are bumped and the obj_cgroup pointer is saved.  Otherwise, the
charge is canceled.

On the free path obj_cgroup pointer is obtained and used to uncharge the
size of the releasing object.

Memcg and lruvec counters are now representing only memory used by active
slab objects and do not include the free space.  The free space is shared
and doesn't belong to any specific cgroup.

Global per-node slab vmstats are still modified from
(un)charge_slab_page() functions.  The idea is to keep all slab pages
accounted as slab pages on system level.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-10-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: save obj_cgroup for non-root slab objects

Store the obj_cgroup pointer in the corresponding place of
page->obj_cgroups for each allocated non-root slab object.  Make sure that
each allocated object holds a reference to obj_cgroup.

Objcg pointer is obtained from the memcg->objcg dereferencing in
memcg_kmem_get_cache() and passed from pre_alloc_hook to post_alloc_hook.
Then in case of successful allocation(s) it's getting stored in the
page->obj_cgroups vector.

The objcg obtaining part look a bit bulky now, but it will be simplified
by next commits in the series.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-9-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: allocate obj_cgroups for non-root slab pages

Allocate and release memory to store obj_cgroup pointers for each non-root
slab page. Reuse page->mem_cgroup pointer to store a pointer to the
allocated space.

This commit temporarily increases the memory footprint of the kernel memory
accounting. To store obj_cgroup pointers we'll need a place for an
objcg_pointer for each allocated object. However, the following patches
in the series will enable sharing of slab pages between memory cgroups,
which will dramatically increase the total slab utilization. And the final
memory footprint will be significantly smaller than before.

To distinguish between obj_cgroups and memcg pointers in case when it's
not obvious which one is used (as in page_cgroup_ino()), let's always set
the lowest bit in the obj_cgroup case. The original obj_cgroups
pointer is marked to be ignored by kmemleak, which otherwise would
report a memory leak for each allocated vector.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-8-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg/slab: obj_cgroup API

Obj_cgroup API provides an ability to account sub-page sized kernel
objects, which potentially outlive the original memory cgroup.

The top-level API consists of the following functions:
  bool obj_cgroup_tryget(struct obj_cgroup *objcg);
  void obj_cgroup_get(struct obj_cgroup *objcg);
  void obj_cgroup_put(struct obj_cgroup *objcg);

  int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
  void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);

  struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg);
  struct obj_cgroup *get_obj_cgroup_from_current(void);

Object cgroup is basically a pointer to a memory cgroup with a per-cpu
reference counter.  It substitutes a memory cgroup in places where it's
necessary to charge a custom amount of bytes instead of pages.

All charged memory rounded down to pages is charged to the corresponding
memory cgroup using __memcg_kmem_charge().

It implements reparenting: on memcg offlining it's getting reattached to
the parent memory cgroup.  Each online memory cgroup has an associated
active object cgroup to handle new allocations and the list of all
attached object cgroups.  On offlining of a cgroup this list is reparented
and for each object cgroup in the list the memcg pointer is swapped to the
parent memory cgroup.  It prevents long-living objects from pinning the
original memory cgroup in the memory.

The implementation is based on byte-sized per-cpu stocks.  A sub-page
sized leftover is stored in an atomic field, which is a part of obj_cgroup
object.  So on cgroup offlining the leftover is automatically reparented.

memcg->objcg is rcu protected.  objcg->memcg is a raw pointer, which is
always pointing at a memory cgroup, but can be atomically swapped to the
parent memory cgroup.  So a user must ensure the lifetime of the
cgroup, e.g.  grab rcu_read_lock or css_set_lock.

Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200623174037.3951353-7-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: decouple reference counting from page accounting

The reference counting of a memcg is currently coupled directly to how
many 4k pages are charged to it.  This doesn't work well with Roman's new
slab controller, which maintains pools of objects and doesn't want to keep
an extra balance sheet for the pages backing those objects.

This unusual refcounting design (reference counts usually track pointers
to an object) is only for historical reasons: memcg used to not take any
css references and simply stalled offlining until all charges had been
reparented and the page counters had dropped to zero.  When we got rid of
the reparenting requirement, the simple mechanical translation was to take
a reference for every charge.

More historical context can be found in commit e8ea14cc6ead ("mm:
memcontrol: take a css reference for each charged page"), commit
64f219938941 ("mm: memcontrol: remove obsolete kmemcg pinning tricks") and
commit b2052564e66d ("mm: memcontrol: continue cache reclaim from offlined
groups").

The new slab controller exposes the limitations in this scheme, so let's
switch it to a more idiomatic reference counting model based on actual
kernel pointers to the memcg:

- The per-cpu stock holds a reference to the memcg its caching

- User pages hold a reference for their page->mem_cgroup. Transparent
  huge pages will no longer acquire tail references in advance, we'll
  get them if needed during the split.

- Kernel pages hold a reference for their page->mem_cgroup

- Pages allocated in the root cgroup will acquire and release css
  references for simplicity. css_get() and css_put() optimize that.

- The current memcg_charge_slab() already hacked around the per-charge
  references; this change gets rid of that as well.

- tcp accounting will handle reference in mem_cgroup_sk_{alloc,free}

Roman:
1) Rebased on top of the current mm tree: added css_get() in
   mem_cgroup_charge(), dropped mem_cgroup_try_charge() part
2) I've reformatted commit references in the commit log to make
   checkpatch.pl happy.

[hughd@google.com: remove css_put_many() from __mem_cgroup_clear_mc()]
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2007302011450.2347@eggly.anvils
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200623174037.3951353-6-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: slub: implement SLUB version of obj_to_index()

This commit implements SLUB version of the obj_to_index() function, which
will be required to calculate the offset of obj_cgroup in the obj_cgroups
vector to store/obtain the objcg ownership data.

To make it faster, let's repeat the SLAB's trick introduced by commit
6a2d7a955d8d ("SLAB: use a multiply instead of a divide in
obj_to_index()") and avoid an expensive division.

Vlastimil Babka noticed, that SLUB does have already a similar function
called slab_index(), which is defined only if SLUB_DEBUG is enabled.  The
function does a similar math, but with a division, and it also takes a
page address instead of a page pointer.

Let's remove slab_index() and replace it with the new helper
__obj_to_index(), which takes a page address.  obj_to_index() will be a
simple wrapper taking a page pointer and passing page_address(page) into
__obj_to_index().

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-5-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg: convert vmstat slab counters to bytes

In order to prepare for per-object slab memory accounting, convert
NR_SLAB_RECLAIMABLE and NR_SLAB_UNRECLAIMABLE vmstat items to bytes.

To make it obvious, rename them to NR_SLAB_RECLAIMABLE_B and
NR_SLAB_UNRECLAIMABLE_B (similar to NR_KERNEL_STACK_KB).

Internally global and per-node counters are stored in pages, however memcg
and lruvec counters are stored in bytes.  This scheme may look weird, but
only for now.  As soon as slab pages will be shared between multiple
cgroups, global and node counters will reflect the total number of slab
pages.  However memcg and lruvec counters will be used for per-memcg slab
memory tracking, which will take separate kernel objects in the account.
Keeping global and node counters in pages helps to avoid additional
overhead.

The size of slab memory shouldn't exceed 4Gb on 32-bit machines, so it
will fit into atomic_long_t we use for vmstats.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-4-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg: prepare for byte-sized vmstat items

To implement per-object slab memory accounting, we need to convert slab
vmstat counters to bytes.  Actually, out of 4 levels of counters: global,
per-node, per-memcg and per-lruvec only two last levels will require
byte-sized counters.  It's because global and per-node counters will be
counting the number of slab pages, and per-memcg and per-lruvec will be
counting the amount of memory taken by charged slab objects.

Converting all vmstat counters to bytes or even all slab counters to bytes
would introduce an additional overhead.  So instead let's store global and
per-node counters in pages, and memcg and lruvec counters in bytes.

To make the API clean all access helpers (both on the read and write
sides) are dealing with bytes.

To avoid back-and-forth conversions a new flavor of read-side helpers is
introduced, which always returns values in pages: node_page_state_pages()
and global_node_page_state_pages().

Actually new helpers are just reading raw values.  Old helpers are simple
wrappers, which will complain on an attempt to read byte value, because at
the moment no one actually needs bytes.

Thanks to Johannes Weiner for the idea of having the byte-sized API on top
of the page-sized internal storage.

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-3-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcg: factor out memcg- and lruvec-level changes out of __mod_lruvec_state()

Patch series "The new cgroup slab memory controller", v7.

The patchset moves the accounting from the page level to the object level.
It allows to share slab pages between memory cgroups.  This leads to a
significant win in the slab utilization (up to 45%) and the corresponding
drop in the total kernel memory footprint.  The reduced number of
unmovable slab pages should also have a positive effect on the memory
fragmentation.

The patchset makes the slab accounting code simpler: there is no more need
in the complicated dynamic creation and destruction of per-cgroup slab
caches, all memory cgroups use a global set of shared slab caches.  The
lifetime of slab caches is not more connected to the lifetime of memory
cgroups.

The more precise accounting does require more CPU, however in practice the
difference seems to be negligible.  We've been using the new slab
controller in Facebook production for several months with different
workloads and haven't seen any noticeable regressions.  What we've seen
were memory savings in order of 1 GB per host (it varied heavily depending
on the actual workload, size of RAM, number of CPUs, memory pressure,
etc).

The third version of the patchset added yet another step towards the
simplification of the code: sharing of slab caches between accounted and
non-accounted allocations.  It comes with significant upsides (most
noticeable, a complete elimination of dynamic slab caches creation) but
not without some regression risks, so this change sits on top of the
patchset and is not completely merged in.  So in the unlikely event of a
noticeable performance regression it can be reverted separately.

The slab memory accounting works in exactly the same way for SLAB and
SLUB.  With both allocators the new controller shows significant memory
savings, with SLUB the difference is bigger.  On my 16-core desktop
machine running Fedora 32 the size of the slab memory measured after the
start of the system was lower by 58% and 38% with SLUB and SLAB
correspondingly.

As an estimation of a potential CPU overhead, below are results of
slab_bulk_test01 test, kindly provided by Jesper D.  Brouer.  He also
helped with the evaluation of results.

The test can be found here: https://github.com/netoptimizer/prototype-kernel/
The smallest number in each row should be picked for a comparison.

SLUB-patched - bulk-API
 - SLUB-patched : bulk_quick_reuse objects=1 : 187 -  90 - 224  cycles(tsc)
 - SLUB-patched : bulk_quick_reuse objects=2 : 110 -  53 - 133  cycles(tsc)
 - SLUB-patched : bulk_quick_reuse objects=3 :  88 -  95 -  42  cycles(tsc)
 - SLUB-patched : bulk_quick_reuse objects=4 :  91 -  85 -  36  cycles(tsc)
 - SLUB-patched : bulk_quick_reuse objects=8 :  32 -  66 -  32  cycles(tsc)

SLUB-original -  bulk-API
 - SLUB-original: bulk_quick_reuse objects=1 :  87 -  87 - 142  cycles(tsc)
 - SLUB-original: bulk_quick_reuse objects=2 :  52 -  53 -  53  cycles(tsc)
 - SLUB-original: bulk_quick_reuse objects=3 :  42 -  42 -  91  cycles(tsc)
 - SLUB-original: bulk_quick_reuse objects=4 :  91 -  37 -  37  cycles(tsc)
 - SLUB-original: bulk_quick_reuse objects=8 :  31 -  79 -  76  cycles(tsc)

SLAB-patched -  bulk-API
 - SLAB-patched : bulk_quick_reuse objects=1 :  67 -  67 - 140  cycles(tsc)
 - SLAB-patched : bulk_quick_reuse objects=2 :  55 -  46 -  46  cycles(tsc)
 - SLAB-patched : bulk_quick_reuse objects=3 :  93 -  94 -  39  cycles(tsc)
 - SLAB-patched : bulk_quick_reuse objects=4 :  35 -  88 -  85  cycles(tsc)
 - SLAB-patched : bulk_quick_reuse objects=8 :  30 -  30 -  30  cycles(tsc)

SLAB-original-  bulk-API
 - SLAB-original: bulk_quick_reuse objects=1 : 143 - 136 -  67  cycles(tsc)
 - SLAB-original: bulk_quick_reuse objects=2 :  45 -  46 -  46  cycles(tsc)
 - SLAB-original: bulk_quick_reuse objects=3 :  38 -  39 -  39  cycles(tsc)
 - SLAB-original: bulk_quick_reuse objects=4 :  35 -  87 -  87  cycles(tsc)
 - SLAB-original: bulk_quick_reuse objects=8 :  29 -  66 -  30  cycles(tsc)

This patch (of 19):

To convert memcg and lruvec slab counters to bytes there must be a way to
change these counters without touching node counters.  Factor out
__mod_memcg_lruvec_state() out of __mod_lruvec_state().

Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20200623174037.3951353-1-guro@fb.com
Link: http://lkml.kernel.org/r/20200623174037.3951353-2-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kmem: make memcg_kmem_enabled() irreversible

Historically the kernel memory accounting was an opt-in feature, which
could be enabled for individual cgroups.  But now it's not true, and it's
on by default both on cgroup v1 and cgroup v2.  And as long as a user has
at least one non-root memory cgroup, the kernel memory accounting is on.
So in most setups it's either always on (if memory cgroups are in use and
kmem accounting is not disabled), either always off (otherwise).

memcg_kmem_enabled() is used in many places to guard the kernel memory
accounting code.  If memcg_kmem_enabled() can reverse from returning true
to returning false (as now), we can't rely on it on release paths and have
to check if it was on before.

If we'll make memcg_kmem_enabled() irreversible (always returning true
after returning it for the first time), it'll make the general logic more
simple and robust.  It also will allow to guard some checks which
otherwise would stay unguarded.

Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Link: http://lkml.kernel.org/r/20200702180926.1330769-1-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: support 64-bit inums per-sb

The default is still set to inode32 for backwards compatibility, but
system administrators can opt in to the new 64-bit inode numbers by
either:

1. Passing inode64 on the command line when mounting, or
2. Configuring the kernel with CONFIG_TMPFS_INODE64=y

The inode64 and inode32 names are used based on existing precedent from
XFS.

[hughd@google.com: Kconfig fixes]
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2008011928010.13320@eggly.anvils
Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Jeff Layton <jlayton@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/8b23758d0c66b5e2263e08baf9c4b6a7565cbd8f.1594661218.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: per-superblock i_ino support

Patch series "tmpfs: inode: Reduce risk of inum overflow", v7.

In Facebook production we are seeing heavy i_ino wraparounds on tmpfs.  On
affected tiers, in excess of 10% of hosts show multiple files with
different content and the same inode number, with some servers even having
as many as 150 duplicated inode numbers with differing file content.

This causes actual, tangible problems in production.  For example, we have
complaints from those working on remote caches that their application is
reporting cache corruptions because it uses (device, inodenum) to
establish the identity of a particular cache object, but because it's not
unique any more, the application refuses to continue and reports cache
corruption.  Even worse, sometimes applications may not even detect the
corruption but may continue anyway, causing phantom and hard to debug
behaviour.

In general, userspace applications expect that (device, inodenum) should
be enough to be uniquely point to one inode, which seems fair enough.  One
might also need to check the generation, but in this case:

1. That's not currently exposed to userspace
   (ioctl(...FS_IOC_GETVERSION...) returns ENOTTY on tmpfs);
2. Even with generation, there shouldn't be two live inodes with the
   same inode number on one device.

In order to mitigate this, we take a two-pronged approach:

1. Moving inum generation from being global to per-sb for tmpfs. This
   itself allows some reduction in i_ino churn. This works on both 64-
   and 32- bit machines.
2. Adding inode{64,32} for tmpfs. This fix is supported on machines with
   64-bit ino_t only: we allow users to mount tmpfs with a new inode64
   option that uses the full width of ino_t, or CONFIG_TMPFS_INODE64.

You can see how this compares to previous related patches which didn't
implement this per-superblock:

- https://patchwork.kernel.org/patch/11254001/
- https://patchwork.kernel.org/patch/11023915/

This patch (of 2):

get_next_ino has a number of problems:

- It uses and returns a uint, which is susceptible to become overflowed
  if a lot of volatile inodes that use get_next_ino are created.
- It's global, with no specificity per-sb or even per-filesystem. This
  means it's not that difficult to cause inode number wraparounds on a
  single device, which can result in having multiple distinct inodes
  with the same inode number.

This patch adds a per-superblock counter that mitigates the second case.
This design also allows us to later have a specific i_ino size per-device,
for example, allowing users to choose whether to use 32- or 64-bit inodes
for each tmpfs mount.  This is implemented in the next commit.

For internal shmem mounts which may be less tolerant to spinlock delays,
we implement a percpu batching scheme which only takes the stat_lock at
each batch boundary.

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Amir Goldstein <amir73il@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Jeff Layton <jlayton@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/cover.1594661218.git.chris@chrisdown.name
Link: http://lkml.kernel.org/r/1986b9d63b986f08ec07a4aa4b2275e718e47d8a.1594661218.git.chris@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_io.c: use blk_io_schedule() for avoiding task hung in sync io

swap_readpage() does the sync io for one page, the io is not big,
normally, the io can be finished quickly, but it may take long time or
wait forever in case of io failure or discard.

This patch uses blk_io_schedule() instead of io_schedule() to avoid task
hung and crash (when set /proc/sys/kernel/hung_task_panic) when the above
exception occurs.

This is similar to the hung task avoidance in submit_bio_wait(),
blk_execute_rq() and __blkdev_direct_IO().

Signed-off-by: Xianting Tian <xianting_tian@126.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Hugh Dickins <hughd@google.com>
Link: http://lkml.kernel.org/r/1596461807-21087-1-git-send-email-xianting_tian@126.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: swap: fix kerneldoc of swap_vma_readahead()

Fix W=1 compile warnings (invalid kerneldoc):

    mm/swap_state.c:742: warning: Function parameter or member 'fentry' not described in 'swap_vma_readahead'
    mm/swap_state.c:742: warning: Excess function parameter 'entry' description in 'swap_vma_readahead'

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20200728171109.28687-2-krzk@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap_slots.c: remove redundant check for swap_slot_cache_initialized

Because enable_swap_slots_cache can only become true in
enable_swap_slots_cache(), and depends on swap_slot_cache_initialized is
true before.  That means, when enable_swap_slots_cache is true,
swap_slot_cache_initialized is true also.

So the condition:
"swap_slot_cache_enabled && swap_slot_cache_initialized"
can be reduced to "swap_slot_cache_enabled"

And in mathematics:
"!swap_slot_cache_enabled || !swap_slot_cache_initialized"
is equal to "!(swap_slot_cache_enabled && swap_slot_cache_initialized)"

So no functional change.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Link: http://lkml.kernel.org/r/20200430061143.450-4-thunder.leizhen@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap_slots.c: simplify enable_swap_slots_cache()

Whether swap_slot_cache_initialized is true or false,
__reenable_swap_slots_cache() is always called.  To make this meaning
clear, leave only one call to __reenable_swap_slots_cache().  This also
make it clearer what extra needs be done when swap_slot_cache_initialized
is false.

No functional change.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Link: http://lkml.kernel.org/r/20200430061143.450-3-thunder.leizhen@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>