Re: [PATCH v2 5/5] mm/madvise: eliminate very confusing manipulation of prev VMA

[Vlastimil Babka]

On 6/24/25 15:16, Lorenzo Stoakes wrote:
> Hi Andrew,
> 
> stress-ng picked up a very subtle bug here. Please apply this fix-patch to
> correct it!
> 
> Cheers, Lorenzo
> 
> ----8<----
> From a04de12e1deb50e708392b0e716612f26c6b386a Mon Sep 17 00:00:00 2001
> From: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
> Date: Tue, 24 Jun 2025 14:14:19 +0100
> Subject: [PATCH] mm/madvise: fix very subtle bug
> 
> With thanks to stress-ng --madvise :)
> 
> vma may have been updated (in the modify call in madvise_update_vma()), so
> we can't assign prev = vma, we must first reassign vma to
> madv_behavior->vma to account for this, before setting prev = vma.

So glad we eliminated very confusing manipulation of prev VMA :) sigh...

LGTM (not worth much it seems...)

> ---
>  mm/madvise.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 4491bf080f55..c467ee42596f 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -1665,8 +1665,8 @@ int madvise_walk_vmas(struct madvise_behavior *madv_behavior)
>  			vma = NULL;
>  			madv_behavior->lock_dropped = false;
>  		} else {
> -			prev = vma;
>  			vma = madv_behavior->vma;
> +			prev = vma;
>  		}
> 
>  		if (vma && range->end < vma->vm_end)
> --
> 2.50.0