Re: [PATCH 1/2] arm64/mm: fix incorrect file_map_count for non-leaf pmd/pud

[Liu Shixin]

On 2022/11/17 12:09, Anshuman Khandual wrote:
>
> On 11/16/22 14:08, Liu Shixin wrote:
>> The page table check trigger BUG_ON() unexpectedly when collapse hugepage:
>>
>>  ------------[ cut here ]------------
>>  kernel BUG at mm/page_table_check.c:82!
>>  Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
>>  Dumping ftrace buffer:
>>     (ftrace buffer empty)
>>  Modules linked in:
>>  CPU: 6 PID: 68 Comm: khugepaged Not tainted 6.1.0-rc3+ #750
>>  Hardware name: linux,dummy-virt (DT)
>>  pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>>  pc : page_table_check_clear.isra.0+0x258/0x3f0
>>  lr : page_table_check_clear.isra.0+0x240/0x3f0
>> [...]
>>  Call trace:
>>   page_table_check_clear.isra.0+0x258/0x3f0
>>   __page_table_check_pmd_clear+0xbc/0x108
>>   pmdp_collapse_flush+0xb0/0x160
>>   collapse_huge_page+0xa08/0x1080
>>   hpage_collapse_scan_pmd+0xf30/0x1590
>>   khugepaged_scan_mm_slot.constprop.0+0x52c/0xac8
>>   khugepaged+0x338/0x518
>>   kthread+0x278/0x2f8
>>   ret_from_fork+0x10/0x20
>> [...]
>>
>> Since pmd_user_accessible_page() doesn't check if a pmd is leaf, it
>> decrease file_map_count for a non-leaf pmd comes from collapse_huge_page().
>> and so trigger BUG_ON() unexpectedly.
> Could you please provide the pmd_val() on the pmd entry, that triggers this
> BUG_ON() here ? Only additional thing pmd_leaf() ensures, is that the entry
> is not a table one.
>
> #define pmd_leaf(pmd)           (pmd_present(pmd) && !pmd_table(pmd))
>
> collapse_huge_page() pmd is non-leaf because it has table bit on ?
The pmd_val is 0x80000004c367003. It is indeed a table entry. collapse_huge_page()
will replace page table of page granularity with block granularity. Before this replace,
it will call pmdp_collapse_flush() to clear the table pmd. In this function, the table pmd
do the check unexpectdly and trigger the BUG_ON().

diff --git a/mm/pgtable-generic.c b/mm/pgtable-generic.c
index 90ab721a12a8..a5c2380bac4d 100644
--- a/mm/pgtable-generic.c
+++ b/mm/pgtable-generic.c
@@ -221,6 +221,7 @@ pmd_t pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
 
        VM_BUG_ON(address & ~HPAGE_PMD_MASK);
        VM_BUG_ON(pmd_trans_huge(*pmdp));
+       pr_err("pmd_val is 0x%lx\n", pmd_val(*pmdp));
        pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
 
        /* collapse entails shooting down ptes not pmd */

Thanks,
Liu Shixin
.

>
>> Fix this problem by using pmd_leaf() insteal of pmd_present() in
>> pmd_user_accessible_page(). Moreover, use pud_leaf() for
>> pud_user_accessible_page() too.
>>
>> Fixes: 42b2547137f5 ("arm64/mm: enable ARCH_SUPPORTS_PAGE_TABLE_CHECK")
>> Reported-by: Denys Vlasenko <dvlasenk@xxxxxxxxxx>
>> Signed-off-by: Liu Shixin <liushixin2@xxxxxxxxxx>
>> ---
>>  arch/arm64/include/asm/pgtable.h | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
>> index 71a1af42f0e8..edf6625ce965 100644
>> --- a/arch/arm64/include/asm/pgtable.h
>> +++ b/arch/arm64/include/asm/pgtable.h
>> @@ -863,12 +863,12 @@ static inline bool pte_user_accessible_page(pte_t pte)
>>  
>>  static inline bool pmd_user_accessible_page(pmd_t pmd)
>>  {
>> -	return pmd_present(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
>> +	return pmd_leaf(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
>>  }
>>  
>>  static inline bool pud_user_accessible_page(pud_t pud)
>>  {
>> -	return pud_present(pud) && pud_user(pud);
>> +	return pud_leaf(pud) && pud_user(pud);
>>  }
>>  #endif
>>  
> .
>