Re: [PATCH v14 67/70] mm/vmscan: Use vma iterator instead of vm_next

From: Liam Howlett
Date: Mon Sep 12 2022 - 03:15:13 EST


* Yu Zhao <yuzhao@xxxxxxxxxx> [220912 02:55]:
> On Tue, Sep 06, 2022 at 07:49:05PM +0000, Liam Howlett wrote:
> > Use the vma iterator in in get_next_vma() instead of the linked list.
> >
> > Suggested-by: Yu Zhao <yuzhao@xxxxxxxxxx>
>
> Apologies for the bad suggestion.
>
> > --- a/mm/vmscan.c
> > +++ b/mm/vmscan.c
> > @@ -3776,23 +3776,14 @@ static bool get_next_vma(unsigned long mask, unsigned long size, struct mm_walk
> > {
> > unsigned long start = round_up(*vm_end, size);
> > unsigned long end = (start | ~mask) + 1;
> > + VMA_ITERATOR(vmi, args->mm, start);
> >
> > VM_WARN_ON_ONCE(mask & size);
> > VM_WARN_ON_ONCE((start & mask) != (*vm_start & mask));
> >
> > - while (args->vma) {
> > - if (start >= args->vma->vm_end) {
> > - args->vma = args->vma->vm_next;
> > + for_each_vma_range(vmi, args->vma, end) {
> > + if (should_skip_vma(args->vma->vm_start, args->vma->vm_end, args))
> > continue;
> > - }
> > -
> > - if (end && end <= args->vma->vm_start)
> > - return false;
>
> Here the original code leaves args->vma pointing the first vma out of
> the range [start, end). This allows the caller (page table walker) to
> resume at that vma, if it chooses to.
>
> With for_each_vma_range(), under the same condition, args->vma is set to
> NULL. And the page table walker may terminate prematurely. Apparently I
> overlooked until I was told MGLRU in mm-unstable is slower than itself
> on 6.0-rc4 yesterday.
>
> > -
> > - if (should_skip_vma(args->vma->vm_start, args->vma->vm_end, args)) {
> > - args->vma = args->vma->vm_next;
> > - continue;
> > - }
> >
> > *vm_start = max(start, args->vma->vm_start);
> > *vm_end = min(end - 1, args->vma->vm_end - 1) + 1;
>
> The following should work properly. Please take a look. Thanks!
>

Thanks Yu. This looks good to me and the explanation makes sense.

> ---
> mm/vmscan.c | 12 +++---------
> 1 file changed, 3 insertions(+), 9 deletions(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 11a86d47e85e..b22d3efe3031 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -3776,23 +3776,17 @@ static bool get_next_vma(unsigned long mask, unsigned long size, struct mm_walk
> {
> unsigned long start = round_up(*vm_end, size);
> unsigned long end = (start | ~mask) + 1;
> + VMA_ITERATOR(vmi, args->mm, start);
>
> VM_WARN_ON_ONCE(mask & size);
> VM_WARN_ON_ONCE((start & mask) != (*vm_start & mask));
>
> - while (args->vma) {
> - if (start >= args->vma->vm_end) {
> - args->vma = args->vma->vm_next;
> - continue;
> - }
> -
> + for_each_vma(vmi, args->vma) {
> if (end && end <= args->vma->vm_start)
> return false;
>
> - if (should_skip_vma(args->vma->vm_start, args->vma->vm_end, args)) {
> - args->vma = args->vma->vm_next;
> + if (should_skip_vma(args->vma->vm_start, args->vma->vm_end, args))
> continue;
> - }
>
> *vm_start = max(start, args->vma->vm_start);
> *vm_end = min(end - 1, args->vma->vm_end - 1) + 1;
> --
> 2.37.2.789.g6183377224-goog