Re: [RFC PATCH] s390: Fix runtime warning about negative pgtables_bytes

From: Martin Schwidefsky
Date: Fri Mar 02 2018 - 06:48:51 EST


On Thu, 1 Mar 2018 14:40:52 -0800
Guenter Roeck <linux@xxxxxxxxxxxx> wrote:

> When running s390 images with 'compat' processes, the following
> BUG is seen repeatedly.
>
> BUG: non-zero pgtables_bytes on freeing mm: -16384
>
> Bisect points to commit b4e98d9ac775 ("mm: account pud page tables").
> Analysis shows that init_new_context() is called with
> mm->context.asce_limit set to _REGION3_SIZE. In this situation,
> pgtables_bytes remains set to 0 and is not increased. The message is
> displayed when the affected process dies and mm_dec_nr_puds() is called.
>
> Cc: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> Cc: Heiko Carstens <heiko.carstens@xxxxxxxxxx>
> Fixes: b4e98d9ac775 ("mm: account pud page tables")
> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
> ---
> RFC since I have no idea if this is correct (or complete, for that matter).
> All I can say is that the message is no longer seen after the patch is
> applied.
>
> arch/s390/include/asm/mmu_context.h | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h
> index 65154eaa3714..6c8ce15cde7b 100644
> --- a/arch/s390/include/asm/mmu_context.h
> +++ b/arch/s390/include/asm/mmu_context.h
> @@ -63,6 +63,7 @@ static inline int init_new_context(struct task_struct *tsk,
> _ASCE_USER_BITS | _ASCE_TYPE_SEGMENT;
> /* pgd_alloc() did not account this pmd */
> mm_inc_nr_pmds(mm);
> + mm_inc_nr_puds(mm);
> }
> crst_table_init((unsigned long *) mm->pgd, pgd_entry_type(mm));
> return 0;

The page table folding always makes a knot in my brain. After staring
at the code for a while I am now convinced that your patch is correct.
On free of a 2 level page table free_pud_range() does a p4d_clear()
which is turned into a nop as the page table is folded. But the
mm_dec_nr_puds() is done unconditionally and that is where the imbalance
comes from. Adding a counter mm_inc_nr_puds to init_new_context fixes
it.

Applied to s390/linux:fixes. Thanks.

--
blue skies,
Martin.

"Reality continues to ruin my life." - Calvin.