Re: [PATCH v3 3/4] mm: prevent KSM from breaking VMA merging for new VMAs

[Andrew Morton]

On Fri, 20 Jun 2025 13:48:09 +0100 Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx> wrote:

> Hi Andrew,
> 
> Sending a fix-patch for this commit due to a reported syzbot issue which
> highlighted a bug in the implementation.
> 
> I discuss the syzbot report at [0].
> 
> [0]: https://lore.kernel.org/all/a55beb72-4288-4356-9642-76ab35a2a07c@lucifer.local/
> 
> There's a very minor conflict around the map->vm_flags vs. map->flags change,
> easily resolvable, but if you need a respin let me know.

I actually saw 4 conflicts, fixed various things up and...

> @@ -2487,6 +2496,11 @@ static int __mmap_new_vma(struct mmap_state *map, struct vm_area_struct **vmap)
>  	if (error)
>  		goto free_iter_vma;
> 
> +	if (!map->check_ksm_early) {
> +		update_ksm_flags(map);
> +		vm_flags_init(vma, map->vm_flags);
> +	}
> +

Guessing map->flags was intended here, I made that change then unmade
it in the later mm-update-core-kernel-code-to-use-vm_flags_t-consistently.patch.

I'll do a full rebuild at a couple of bisection points, please check
that all landed OK.