linux-stable.git/mm, branch v5.4.259

tmpfs: verify {g,u}id mount options correctly

2023-09-23T08:59:40+00:00

[ Upstream commit 0200679fc7953177941e41c2a4241d0b6c2c5de8 ]

A while ago we received the following report:

"The other outstanding issue I noticed comes from the fact that
fsconfig syscalls may occur in a different userns than that which
called fsopen. That means that resolving the uid/gid via
current_user_ns() can save a kuid that isn't mapped in the associated
namespace when the filesystem is finally mounted. This means that it
is possible for an unprivileged user to create files owned by any
group in a tmpfs mount (since we can set the SUID bit on the tmpfs
directory), or a tmpfs that is owned by any user, including the root
group/user."

The contract for {g,u}id mount options and {g,u}id values in general set
from userspace has always been that they are translated according to the
caller's idmapping. In so far, tmpfs has been doing the correct thing.
But since tmpfs is mountable in unprivileged contexts it is also
necessary to verify that the resulting {k,g}uid is representable in the
namespace of the superblock to avoid such bugs as above.

The new mount api's cross-namespace delegation abilities are already
widely used. After having talked to a bunch of userspace this is the
most faithful solution with minimal regression risks. I know of one
users - systemd - that makes use of the new mount api in this way and
they don't set unresolable {g,u}ids. So the regression risk is minimal.

Link: https://lore.kernel.org/lkml/CALxfFW4BXhEwxR0Q5LSkg-8Vb4r2MONKCcUCVioehXQKr35eHg@mail.gmail.com
Fixes: f32356261d44 ("vfs: Convert ramfs, shmem, tmpfs, devtmpfs, rootfs to use the new mount API")
Reviewed-by: "Seth Forshee (DigitalOcean)" 
Reported-by: Seth Jenkins 
Message-Id: <20230801-vfs-fs_context-uidgid-v1-1-daf46a050bbf@kernel.org>
Signed-off-by: Christian Brauner 
Signed-off-by: Sasha Levin

mm: allow a controlled amount of unfairness in the page lock

2023-08-30T14:27:26+00:00

commit 5ef64cc8987a9211d3f3667331ba3411a94ddc79 upstream.

Commit 2a9127fcf229 ("mm: rewrite wait_on_page_bit_common() logic") made
the page locking entirely fair, in that if a waiter came in while the
lock was held, the lock would be transferred to the lockers strictly in
order.

That was intended to finally get rid of the long-reported watchdog
failures that involved the page lock under extreme load, where a process
could end up waiting essentially forever, as other page lockers stole
the lock from under it.

It also improved some benchmarks, but it ended up causing huge
performance regressions on others, simply because fair lock behavior
doesn't end up giving out the lock as aggressively, causing better
worst-case latency, but potentially much worse average latencies and
throughput.

Instead of reverting that change entirely, this introduces a controlled
amount of unfairness, with a sysctl knob to tune it if somebody needs
to.  But the default value should hopefully be good for any normal load,
allowing a few rounds of lock stealing, but enforcing the strict
ordering before the lock has been stolen too many times.

There is also a hint from Matthieu Baerts that the fair page coloring
may end up exposing an ABBA deadlock that is hidden by the usual
optimistic lock stealing, and while the unfairness doesn't fix the
fundamental issue (and I'm still looking at that), it avoids it in
practice.

The amount of unfairness can be modified by writing a new value to the
'sysctl_page_lock_unfairness' variable (default value of 5, exposed
through /proc/sys/vm/page_lock_unfairness), but that is hopefully
something we'd use mainly for debugging rather than being necessary for
any deep system tuning.

This whole issue has exposed just how critical the page lock can be, and
how contended it gets under certain locks.  And the main contention
doesn't really seem to be anything related to IO (which was the origin
of this lock), but for things like just verifying that the page file
mapping is stable while faulting in the page into a page table.

Link: https://lore.kernel.org/linux-fsdevel/ed8442fd-6f54-dd84-cd4a-941e8b7ee603@MichaelLarabel.com/
Link: https://www.phoronix.com/scan.php?page=article&item=linux-50-59&num=1
Link: https://lore.kernel.org/linux-fsdevel/c560a38d-8313-51fb-b1ec-e904bd8836bc@tessares.net/
Reported-and-tested-by: Michael Larabel 
Tested-by: Matthieu Baerts 
Cc: Dave Chinner 
Cc: Matthew Wilcox 
Cc: Chris Mason 
Cc: Jan Kara 
Cc: Amir Goldstein 
Signed-off-by: Linus Torvalds 
Signed-off-by: Saeed Mirzamohammadi 
Tested-by: Maximilian Heyne 
Signed-off-by: Greg Kroah-Hartman

mm: make wait_on_page_writeback() wait for multiple pending writebacks

2023-06-28T08:18:42+00:00

commit c2407cf7d22d0c0d94cf20342b3b8f06f1d904e7 upstream.

Ever since commit 2a9127fcf229 ("mm: rewrite wait_on_page_bit_common()
logic") we've had some very occasional reports of BUG_ON(PageWriteback)
in write_cache_pages(), which we thought we already fixed in commit
073861ed77b6 ("mm: fix VM_BUG_ON(PageTail) and BUG_ON(PageWriteback)").

But syzbot just reported another one, even with that commit in place.

And it turns out that there's a simpler way to trigger the BUG_ON() than
the one Hugh found with page re-use.  It all boils down to the fact that
the page writeback is ostensibly serialized by the page lock, but that
isn't actually really true.

Yes, the people _setting_ writeback all do so under the page lock, but
the actual clearing of the bit - and waking up any waiters - happens
without any page lock.

This gives us this fairly simple race condition:

  CPU1 = end previous writeback
  CPU2 = start new writeback under page lock
  CPU3 = write_cache_pages()

  CPU1          CPU2            CPU3
  ----          ----            ----

  end_page_writeback()
    test_clear_page_writeback(page)
    ... delayed...

                lock_page();
                set_page_writeback()
                unlock_page()

                                lock_page()
                                wait_on_page_writeback();

    wake_up_page(page, PG_writeback);
    .. wakes up CPU3 ..

                                BUG_ON(PageWriteback(page));

where the BUG_ON() happens because we woke up the PG_writeback bit
becasue of the _previous_ writeback, but a new one had already been
started because the clearing of the bit wasn't actually atomic wrt the
actual wakeup or serialized by the page lock.

The reason this didn't use to happen was that the old logic in waiting
on a page bit would just loop if it ever saw the bit set again.

The nice proper fix would probably be to get rid of the whole "wait for
writeback to clear, and then set it" logic in the writeback path, and
replace it with an atomic "wait-to-set" (ie the same as we have for page
locking: we set the page lock bit with a single "lock_page()", not with
"wait for lock bit to clear and then set it").

However, out current model for writeback is that the waiting for the
writeback bit is done by the generic VFS code (ie write_cache_pages()),
but the actual setting of the writeback bit is done much later by the
filesystem ".writepages()" function.

IOW, to make the writeback bit have that same kind of "wait-to-set"
behavior as we have for page locking, we'd have to change our roughly
~50 different writeback functions.  Painful.

Instead, just make "wait_on_page_writeback()" loop on the very unlikely
situation that the PG_writeback bit is still set, basically re-instating
the old behavior.  This is very non-optimal in case of contention, but
since we only ever set the bit under the page lock, that situation is
controlled.

Reported-by: syzbot+2fc0712f8f8b8b8fa0ef@syzkaller.appspotmail.com
Fixes: 2a9127fcf229 ("mm: rewrite wait_on_page_bit_common() logic")
Acked-by: Hugh Dickins 
Cc: Andrew Morton 
Cc: Matthew Wilcox 
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm: fix VM_BUG_ON(PageTail) and BUG_ON(PageWriteback)

2023-06-28T08:18:42+00:00

commit 073861ed77b6b957c3c8d54a11dc503f7d986ceb upstream.

Twice now, when exercising ext4 looped on shmem huge pages, I have crashed
on the PF_ONLY_HEAD check inside PageWaiters(): ext4_finish_bio() calling
end_page_writeback() calling wake_up_page() on tail of a shmem huge page,
no longer an ext4 page at all.

The problem is that PageWriteback is not accompanied by a page reference
(as the NOTE at the end of test_clear_page_writeback() acknowledges): as
soon as TestClearPageWriteback has been done, that page could be removed
from page cache, freed, and reused for something else by the time that
wake_up_page() is reached.

https://lore.kernel.org/linux-mm/20200827122019.GC14765@casper.infradead.org/
Matthew Wilcox suggested avoiding or weakening the PageWaiters() tail
check; but I'm paranoid about even looking at an unreferenced struct page,
lest its memory might itself have already been reused or hotremoved (and
wake_up_page_bit() may modify that memory with its ClearPageWaiters()).

Then on crashing a second time, realized there's a stronger reason against
that approach.  If my testing just occasionally crashes on that check,
when the page is reused for part of a compound page, wouldn't it be much
more common for the page to get reused as an order-0 page before reaching
wake_up_page()?  And on rare occasions, might that reused page already be
marked PageWriteback by its new user, and already be waited upon?  What
would that look like?

It would look like BUG_ON(PageWriteback) after wait_on_page_writeback()
in write_cache_pages() (though I have never seen that crash myself).

Matthew Wilcox explaining this to himself:
 "page is allocated, added to page cache, dirtied, writeback starts,

  --- thread A ---
  filesystem calls end_page_writeback()
        test_clear_page_writeback()
  --- context switch to thread B ---
  truncate_inode_pages_range() finds the page, it doesn't have writeback set,
  we delete it from the page cache.  Page gets reallocated, dirtied, writeback
  starts again.  Then we call write_cache_pages(), see
  PageWriteback() set, call wait_on_page_writeback()
  --- context switch back to thread A ---
  wake_up_page(page, PG_writeback);
  ... thread B is woken, but because the wakeup was for the old use of
  the page, PageWriteback is still set.

  Devious"

And prior to 2a9127fcf229 ("mm: rewrite wait_on_page_bit_common() logic")
this would have been much less likely: before that, wake_page_function()'s
non-exclusive case would stop walking and not wake if it found Writeback
already set again; whereas now the non-exclusive case proceeds to wake.

I have not thought of a fix that does not add a little overhead: the
simplest fix is for end_page_writeback() to get_page() before calling
test_clear_page_writeback(), then put_page() after wake_up_page().

Was there a chance of missed wakeups before, since a page freed before
reaching wake_up_page() would have PageWaiters cleared?  I think not,
because each waiter does hold a reference on the page.  This bug comes
when the old use of the page, the one we do TestClearPageWriteback on,
had *no* waiters, so no additional page reference beyond the page cache
(and whoever racily freed it).  The reuse of the page has a waiter
holding a reference, and its own PageWriteback set; but the belated
wake_up_page() has woken the reuse to hit that BUG_ON(PageWriteback).

Reported-by: syzbot+3622cea378100f45d59f@syzkaller.appspotmail.com
Reported-by: Qian Cai 
Fixes: 2a9127fcf229 ("mm: rewrite wait_on_page_bit_common() logic")
Signed-off-by: Hugh Dickins 
Cc: stable@vger.kernel.org # v5.8+
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

list: add "list_del_init_careful()" to go with "list_empty_careful()"

2023-06-28T08:18:35+00:00

[ Upstream commit c6fe44d96fc1536af5b11cd859686453d1b7bfd1 ]

That gives us ordering guarantees around the pair.

Signed-off-by: Linus Torvalds 
Stable-dep-of: 2192bba03d80 ("epoll: ep_autoremove_wake_function should use list_del_init_careful")
Signed-off-by: Sasha Levin

mm: rewrite wait_on_page_bit_common() logic

2023-06-28T08:18:35+00:00

[ Upstream commit 2a9127fcf2296674d58024f83981f40b128fffea ]

It turns out that wait_on_page_bit_common() had several problems,
ranging from just unfair behavioe due to re-queueing at the end of the
wait queue when re-trying, and an outright bug that could result in
missed wakeups (but probably never happened in practice).

This rewrites the whole logic to avoid both issues, by simply moving the
logic to check (and possibly take) the bit lock into the wakeup path
instead.

That makes everything much more straightforward, and means that we never
need to re-queue the wait entry: if we get woken up, we'll be notified
through WQ_FLAG_WOKEN, and the wait queue entry will have been removed,
and everything will have been done for us.

Link: https://lore.kernel.org/lkml/CAHk-=wjJA2Z3kUFb-5s=6+n0qbTs8ELqKFt9B3pH85a8fGD73w@mail.gmail.com/
Link: https://lore.kernel.org/lkml/alpine.LSU.2.11.2007221359450.1017@eggly.anvils/
Reported-by: Oleg Nesterov 
Reported-by: Hugh Dickins 
Cc: Michal Hocko 
Reviewed-by: Oleg Nesterov 
Signed-off-by: Linus Torvalds 
Stable-dep-of: 2192bba03d80 ("epoll: ep_autoremove_wake_function should use list_del_init_careful")
Signed-off-by: Sasha Levin

treewide: Remove uninitialized_var() usage

2023-06-09T08:29:01+00:00

commit 3f649ab728cda8038259d8f14492fe400fbab911 upstream.

Using uninitialized_var() is dangerous as it papers over real bugs[1]
(or can in the future), and suppresses unrelated compiler warnings
(e.g. "unused variable"). If the compiler thinks it is uninitialized,
either simply initialize the variable or make compiler changes.

In preparation for removing[2] the[3] macro[4], remove all remaining
needless uses with the following script:

git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \
	xargs perl -pi -e \
		's/\buninitialized_var\(([^\)]+)\)/\1/g;
		 s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;'

drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid
pathological white-space.

No outstanding warnings were found building allmodconfig with GCC 9.3.0
for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64,
alpha, and m68k.

[1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/
[2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/
[3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/
[4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/

Reviewed-by: Leon Romanovsky  # drivers/infiniband and mlx4/mlx5
Acked-by: Jason Gunthorpe  # IB
Acked-by: Kalle Valo  # wireless drivers
Reviewed-by: Chao Yu  # erofs
Signed-off-by: Kees Cook 
Signed-off-by: Greg Kroah-Hartman

mm/page_alloc: fix potential deadlock on zonelist_update_seq seqlock

2023-05-17T09:36:05+00:00

commit 1007843a91909a4995ee78a538f62d8665705b66 upstream.

syzbot is reporting circular locking dependency which involves
zonelist_update_seq seqlock [1], for this lock is checked by memory
allocation requests which do not need to be retried.

One deadlock scenario is kmalloc(GFP_ATOMIC) from an interrupt handler.

  CPU0
  ----
  __build_all_zonelists() {
    write_seqlock(&zonelist_update_seq); // makes zonelist_update_seq.seqcount odd
    // e.g. timer interrupt handler runs at this moment
      some_timer_func() {
        kmalloc(GFP_ATOMIC) {
          __alloc_pages_slowpath() {
            read_seqbegin(&zonelist_update_seq) {
              // spins forever because zonelist_update_seq.seqcount is odd
            }
          }
        }
      }
    // e.g. timer interrupt handler finishes
    write_sequnlock(&zonelist_update_seq); // makes zonelist_update_seq.seqcount even
  }

This deadlock scenario can be easily eliminated by not calling
read_seqbegin(&zonelist_update_seq) from !__GFP_DIRECT_RECLAIM allocation
requests, for retry is applicable to only __GFP_DIRECT_RECLAIM allocation
requests.  But Michal Hocko does not know whether we should go with this
approach.

Another deadlock scenario which syzbot is reporting is a race between
kmalloc(GFP_ATOMIC) from tty_insert_flip_string_and_push_buffer() with
port->lock held and printk() from __build_all_zonelists() with
zonelist_update_seq held.

  CPU0                                   CPU1
  ----                                   ----
  pty_write() {
    tty_insert_flip_string_and_push_buffer() {
                                         __build_all_zonelists() {
                                           write_seqlock(&zonelist_update_seq);
                                           build_zonelists() {
                                             printk() {
                                               vprintk() {
                                                 vprintk_default() {
                                                   vprintk_emit() {
                                                     console_unlock() {
                                                       console_flush_all() {
                                                         console_emit_next_record() {
                                                           con->write() = serial8250_console_write() {
      spin_lock_irqsave(&port->lock, flags);
      tty_insert_flip_string() {
        tty_insert_flip_string_fixed_flag() {
          __tty_buffer_request_room() {
            tty_buffer_alloc() {
              kmalloc(GFP_ATOMIC | __GFP_NOWARN) {
                __alloc_pages_slowpath() {
                  zonelist_iter_begin() {
                    read_seqbegin(&zonelist_update_seq); // spins forever because zonelist_update_seq.seqcount is odd
                                                             spin_lock_irqsave(&port->lock, flags); // spins forever because port->lock is held
                    }
                  }
                }
              }
            }
          }
        }
      }
      spin_unlock_irqrestore(&port->lock, flags);
                                                             // message is printed to console
                                                             spin_unlock_irqrestore(&port->lock, flags);
                                                           }
                                                         }
                                                       }
                                                     }
                                                   }
                                                 }
                                               }
                                             }
                                           }
                                           write_sequnlock(&zonelist_update_seq);
                                         }
    }
  }

This deadlock scenario can be eliminated by

  preventing interrupt context from calling kmalloc(GFP_ATOMIC)

and

  preventing printk() from calling console_flush_all()

while zonelist_update_seq.seqcount is odd.

Since Petr Mladek thinks that __build_all_zonelists() can become a
candidate for deferring printk() [2], let's address this problem by

  disabling local interrupts in order to avoid kmalloc(GFP_ATOMIC)

and

  disabling synchronous printk() in order to avoid console_flush_all()

.

As a side effect of minimizing duration of zonelist_update_seq.seqcount
being odd by disabling synchronous printk(), latency at
read_seqbegin(&zonelist_update_seq) for both !__GFP_DIRECT_RECLAIM and
__GFP_DIRECT_RECLAIM allocation requests will be reduced.  Although, from
lockdep perspective, not calling read_seqbegin(&zonelist_update_seq) (i.e.
do not record unnecessary locking dependency) from interrupt context is
still preferable, even if we don't allow calling kmalloc(GFP_ATOMIC)
inside
write_seqlock(&zonelist_update_seq)/write_sequnlock(&zonelist_update_seq)
section...

Link: https://lkml.kernel.org/r/8796b95c-3da3-5885-fddd-6ef55f30e4d3@I-love.SAKURA.ne.jp
Fixes: 3d36424b3b58 ("mm/page_alloc: fix race condition between build_all_zonelists and page allocation")
Link: https://lkml.kernel.org/r/ZCrs+1cDqPWTDFNM@alley [2]
Reported-by: syzbot 
  Link: https://syzkaller.appspot.com/bug?extid=223c7461c58c58a4cb10 [1]
Signed-off-by: Tetsuo Handa 
Acked-by: Michal Hocko 
Acked-by: Mel Gorman 
Cc: Petr Mladek 
Cc: David Hildenbrand 
Cc: Ilpo Järvinen 
Cc: John Ogness 
Cc: Patrick Daly 
Cc: Sergey Senozhatsky 
Cc: Steven Rostedt 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Greg Kroah-Hartman

mm/swap: fix swap_info_struct race between swapoff and get_swap_pages()

2023-04-20T10:07:35+00:00

commit 6fe7d6b992113719e96744d974212df3fcddc76c upstream.

The si->lock must be held when deleting the si from the available list.
Otherwise, another thread can re-add the si to the available list, which
can lead to memory corruption.  The only place we have found where this
happens is in the swapoff path.  This case can be described as below:

core 0                       core 1
swapoff

del_from_avail_list(si)      waiting

try lock si->lock            acquire swap_avail_lock
                             and re-add si into
                             swap_avail_head

acquire si->lock but missing si already being added again, and continuing
to clear SWP_WRITEOK, etc.

It can be easily found that a massive warning messages can be triggered
inside get_swap_pages() by some special cases, for example, we call
madvise(MADV_PAGEOUT) on blocks of touched memory concurrently, meanwhile,
run much swapon-swapoff operations (e.g.  stress-ng-swap).

However, in the worst case, panic can be caused by the above scene.  In
swapoff(), the memory used by si could be kept in swap_info[] after
turning off a swap.  This means memory corruption will not be caused
immediately until allocated and reset for a new swap in the swapon path.
A panic message caused: (with CONFIG_PLIST_DEBUG enabled)

------------[ cut here ]------------
top: 00000000e58a3003, n: 0000000013e75cda, p: 000000008cd4451a
prev: 0000000035b1e58a, n: 000000008cd4451a, p: 000000002150ee8d
next: 000000008cd4451a, n: 000000008cd4451a, p: 000000008cd4451a
WARNING: CPU: 21 PID: 1843 at lib/plist.c:60 plist_check_prev_next_node+0x50/0x70
Modules linked in: rfkill(E) crct10dif_ce(E)...
CPU: 21 PID: 1843 Comm: stress-ng Kdump: ... 5.10.134+
Hardware name: Alibaba Cloud ECS, BIOS 0.0.0 02/06/2015
pstate: 60400005 (nZCv daif +PAN -UAO -TCO BTYPE=--)
pc : plist_check_prev_next_node+0x50/0x70
lr : plist_check_prev_next_node+0x50/0x70
sp : ffff0018009d3c30
x29: ffff0018009d3c40 x28: ffff800011b32a98
x27: 0000000000000000 x26: ffff001803908000
x25: ffff8000128ea088 x24: ffff800011b32a48
x23: 0000000000000028 x22: ffff001800875c00
x21: ffff800010f9e520 x20: ffff001800875c00
x19: ffff001800fdc6e0 x18: 0000000000000030
x17: 0000000000000000 x16: 0000000000000000
x15: 0736076307640766 x14: 0730073007380731
x13: 0736076307640766 x12: 0730073007380731
x11: 000000000004058d x10: 0000000085a85b76
x9 : ffff8000101436e4 x8 : ffff800011c8ce08
x7 : 0000000000000000 x6 : 0000000000000001
x5 : ffff0017df9ed338 x4 : 0000000000000001
x3 : ffff8017ce62a000 x2 : ffff0017df9ed340
x1 : 0000000000000000 x0 : 0000000000000000
Call trace:
 plist_check_prev_next_node+0x50/0x70
 plist_check_head+0x80/0xf0
 plist_add+0x28/0x140
 add_to_avail_list+0x9c/0xf0
 _enable_swap_info+0x78/0xb4
 __do_sys_swapon+0x918/0xa10
 __arm64_sys_swapon+0x20/0x30
 el0_svc_common+0x8c/0x220
 do_el0_svc+0x2c/0x90
 el0_svc+0x1c/0x30
 el0_sync_handler+0xa8/0xb0
 el0_sync+0x148/0x180
irq event stamp: 2082270

Now, si->lock locked before calling 'del_from_avail_list()' to make sure
other thread see the si had been deleted and SWP_WRITEOK cleared together,
will not reinsert again.

This problem exists in versions after stable 5.10.y.

Link: https://lkml.kernel.org/r/20230404154716.23058-1-rongwei.wang@linux.alibaba.com
Fixes: a2468cc9bfdff ("swap: choose swap device according to numa node")
Tested-by: Yongchen Yin 
Signed-off-by: Rongwei Wang 
Cc: Bagas Sanjaya 
Cc: Matthew Wilcox (Oracle) 
Cc: Aaron Lu 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Greg Kroah-Hartman

mm/thp: check and bail out if page in deferred queue already

2023-03-11T15:44:05+00:00

commit 81e506bec9be1eceaf5a2c654e28ba5176ef48d8 upstream.

Kernel build regression with LLVM was reported here:
https://lore.kernel.org/all/Y1GCYXGtEVZbcv%2F5@dev-arch.thelio-3990X/ with
commit f35b5d7d676e ("mm: align larger anonymous mappings on THP
boundaries").  And the commit f35b5d7d676e was reverted.

It turned out the regression is related with madvise(MADV_DONTNEED)
was used by ld.lld. But with none PMD_SIZE aligned parameter len.
trace-bpfcc captured:
531607  531732  ld.lld          do_madvise.part.0 start: 0x7feca9000000, len: 0x7fb000, behavior: 0x4
531607  531793  ld.lld          do_madvise.part.0 start: 0x7fec86a00000, len: 0x7fb000, behavior: 0x4

If the underneath physical page is THP, the madvise(MADV_DONTNEED) can
trigger split_queue_lock contention raised significantly. perf showed
following data:
    14.85%     0.00%  ld.lld           [kernel.kallsyms]           [k]
       entry_SYSCALL_64_after_hwframe
           11.52%
                entry_SYSCALL_64_after_hwframe
                do_syscall_64
                __x64_sys_madvise
                do_madvise.part.0
                zap_page_range
                unmap_single_vma
                unmap_page_range
                page_remove_rmap
                deferred_split_huge_page
                __lock_text_start
                native_queued_spin_lock_slowpath

If THP can't be removed from rmap as whole THP, partial THP will be
removed from rmap by removing sub-pages from rmap.  Even the THP head page
is added to deferred queue already, the split_queue_lock will be acquired
and check whether the THP head page is in the queue already.  Thus, the
contention of split_queue_lock is raised.

Before acquire split_queue_lock, check and bail out early if the THP
head page is in the queue already. The checking without holding
split_queue_lock could race with deferred_split_scan, but it doesn't
impact the correctness here.

Test result of building kernel with ld.lld:
commit 7b5a0b664ebe (parent commit of f35b5d7d676e):
time -f "\t%E real,\t%U user,\t%S sys" make LD=ld.lld -skj96 allmodconfig all
        6:07.99 real,   26367.77 user,  5063.35 sys

commit f35b5d7d676e:
time -f "\t%E real,\t%U user,\t%S sys" make LD=ld.lld -skj96 allmodconfig all
        7:22.15 real,   26235.03 user,  12504.55 sys

commit f35b5d7d676e with the fixing patch:
time -f "\t%E real,\t%U user,\t%S sys" make LD=ld.lld -skj96 allmodconfig all
        6:08.49 real,   26520.15 user,  5047.91 sys

Link: https://lkml.kernel.org/r/20221223135207.2275317-1-fengwei.yin@intel.com
Signed-off-by: Yin Fengwei 
Tested-by: Nathan Chancellor 
Acked-by: David Rientjes 
Reviewed-by: "Huang, Ying" 
Cc: Feng Tang 
Cc: Matthew Wilcox 
Cc: Rik van Riel 
Cc: Xing Zhengjun 
Cc: Yang Shi 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Greg Kroah-Hartman