Re: [PATCH] mm: memory: fix /proc/meminfo reporting for MLOCK_ONFAULT

[Lucian Grijincu <lucian@fb.com>]

On 9/13/19, 04:18, "Souptick Joarder" <jrdr.linux@gmail.com> wrote:

    On Fri, Sep 13, 2019 at 4:49 AM Lucian Adrian Grijincu <lucian@fb.com> wrote:
    >
    > As pages are faulted in MLOCK_ONFAULT correctly updates
    > /proc/self/smaps, but doesn't update /proc/meminfo's Mlocked field.
    >
    > - Before this /proc/meminfo fields didn't change as pages were faulted in:
    > diff --git a/mm/memory.c b/mm/memory.c
    > index e0c232fe81d9..7e8dc3ed4e89 100644
    > --- a/mm/memory.c
    > +++ b/mm/memory.c
    > @@ -3311,6 +3311,9 @@ vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
    >         } else {
    >                 inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
    >                 page_add_file_rmap(page, false);
    > +               if ((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED &&
    > +                               !PageTransCompound(page))
    
    Do we need to check against VM_SPECIAL ?
    
I think you're right. mlock/mlock2 already checks and doesn't set VM_LOCKED if VM_SPECIAL is set: https://github.com/torvalds/linux/blob/v5.2/mm/mlock.c#L519-L533

/*
 * mlock_fixup  - handle mlock[all]/munlock[all] requests.
 *
 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
 * munlock is a no-op.  However, for some special vmas, we go ahead and
 * populate the ptes.
 *
 * For vmas that pass the filters, merge/split as appropriate.
 */
static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
	unsigned long start, unsigned long end, vm_flags_t newflags)
{
	struct mm_struct *mm = vma->vm_mm;
	pgoff_t pgoff;
	int nr_pages;
	int ret = 0;
	int lock = !!(newflags & VM_LOCKED);
	vm_flags_t old_flags = vma->vm_flags;

	if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
	    vma_is_dax(vma))
		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
		goto out;


I got thrown off by this check https://github.com/torvalds/linux/blob/v5.2/mm/swap.c#L454-L469


void lru_cache_add_active_or_unevictable(struct page *page,
					 struct vm_area_struct *vma)
{
	VM_BUG_ON_PAGE(PageLRU(page), page);

	if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
		SetPageActive(page);
	else if (!TestSetPageMlocked(page)) {
		/*
		 * We use the irq-unsafe __mod_zone_page_stat because this
		 * counter is not modified from interrupt context, and the pte
		 * lock is held(spinlock), which implies preemption disabled.
		 */
		__mod_zone_page_state(page_zone(page), NR_MLOCK,
				    hpage_nr_pages(page));
		count_vm_event(UNEVICTABLE_PGMLOCKED);

I'll remove VM_SPECIAL and re-submit.

--
Lucian