The following lockdep splat was observed when a certain hugetlbfs test
was run:

  ================================
  WARNING: inconsistent lock state
  4.18.0-159.el8.x86_64+debug #1 Tainted: G        W --------- -  -
  --------------------------------
  inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
  swapper/30/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
  ffffffff9acdc038 (hugetlb_lock){+.?.}, at: free_huge_page+0x36f/0xaa0
  {SOFTIRQ-ON-W} state was registered at:
    lock_acquire+0x14f/0x3b0
    _raw_spin_lock+0x30/0x70
    __nr_hugepages_store_common+0x11b/0xb30
    hugetlb_sysctl_handler_common+0x209/0x2d0
    proc_sys_call_handler+0x37f/0x450
    vfs_write+0x157/0x460
    ksys_write+0xb8/0x170
    do_syscall_64+0xa5/0x4d0
    entry_SYSCALL_64_after_hwframe+0x6a/0xdf
  irq event stamp: 691296
  hardirqs last  enabled at (691296): [<ffffffff99bb034b>] _raw_spin_unlock_irqrestore+0x4b/0x60
  hardirqs last disabled at (691295): [<ffffffff99bb0ad2>] _raw_spin_lock_irqsave+0x22/0x81
  softirqs last  enabled at (691284): [<ffffffff97ff0c63>] irq_enter+0xc3/0xe0
  softirqs last disabled at (691285): [<ffffffff97ff0ebe>] irq_exit+0x23e/0x2b0

  other info that might help us debug this:
   Possible unsafe locking scenario:

         CPU0
         ----
    lock(hugetlb_lock);
    <Interrupt>
      lock(hugetlb_lock);

   *** DEADLOCK ***
      :
  Call Trace:
   <IRQ>
   __lock_acquire+0x146b/0x48c0
   lock_acquire+0x14f/0x3b0
   _raw_spin_lock+0x30/0x70
   free_huge_page+0x36f/0xaa0
   bio_check_pages_dirty+0x2fc/0x5c0
   clone_endio+0x17f/0x670 [dm_mod]
   blk_update_request+0x276/0xe50
   scsi_end_request+0x7b/0x6a0
   scsi_io_completion+0x1c6/0x1570
   blk_done_softirq+0x22e/0x350
   __do_softirq+0x23d/0xad8
   irq_exit+0x23e/0x2b0
   do_IRQ+0x11a/0x200
   common_interrupt+0xf/0xf
   </IRQ>

Both the hugetbl_lock and the subpool lock can be acquired in
free_huge_page().  One way to solve the problem is to make both locks
irq-safe.  However, Mike Kravetz had learned that the hugetlb_lock is
held for a linear scan of ALL hugetlb pages during a cgroup reparentling
operation.  So it is just too long to have irq disabled unless we can
break hugetbl_lock down into finer-grained locks with shorter lock hold
times.

Another alternative is to defer the freeing to a workqueue job.  This
patch implements the deferred freeing by adding a free_hpage_workfn()
work function to do the actual freeing.  The free_huge_page() call in a
non-task context saves the page to be freed in the hpage_freelist linked
list in a lockless manner using the llist APIs.

The generic workqueue is used to process the work, but a dedicated
workqueue can be used instead if it is desirable to have the huge page
freed ASAP.

Thanks to Kirill Tkhai <ktkhai@virtuozzo.com> for suggesting the use of
llist APIs which simplfy the code.

Link: http://lkml.kernel.org/r/20191217170331.30893-1-longman@redhat.com
Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>