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

[Yu Zhao]

On Mon, Sep 12, 2022 at 12:55:08AM -0600, Yu Zhao wrote:
> 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.
  ^^^^^^ continue (without releasing mmap_lock)

> 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!
> 
> ---
>  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