khugepaged: adjust VM_BUG_ON_MM() in __khugepaged_enter()

syzbot crashes on the VM_BUG_ON_MM(khugepaged_test_exit(mm), mm) in
__khugepaged_enter(): yes, when one thread is about to dump core, has set
core_state, and is waiting for others, another might do something calling
__khugepaged_enter(), which now crashes because I lumped the core_state
test (known as "mmget_still_valid") into khugepaged_test_exit().  I still
think it's best to lump them together, so just in this exceptional case,
check mm->mm_users directly instead of khugepaged_test_exit().

Fixes: bbe98f9 ("khugepaged: khugepaged_test_exit() check mmget_still_valid()")
Reported-by: syzbot <[email protected]>
Signed-off-by: Hugh Dickins <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Acked-by: Yang Shi <[email protected]>
Cc: "Kirill A. Shutemov" <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Song Liu <[email protected]>
Cc: Mike Kravetz <[email protected]>
Cc: Eric Dumazet <[email protected]>
Cc: <[email protected]>	[4.8+]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Linus Torvalds <[email protected]>

mm/khugepaged.c

@@ -466,7 +466,7 @@ int __khugepaged_enter(struct mm_struct *mm)
 		return -ENOMEM;
 
 	/* __khugepaged_exit() must not run from under us */
-	VM_BUG_ON_MM(khugepaged_test_exit(mm), mm);
+	VM_BUG_ON_MM(atomic_read(&mm->mm_users) == 0, mm);
 	if (unlikely(test_and_set_bit(MMF_VM_HUGEPAGE, &mm->flags))) {
 		free_mm_slot(mm_slot);
 		return 0;