Re: [RFC PATCH v2] mm/rmap: make num_children and num_active_vmas update in internally

[Lorenzo Stoakes]

On Mon, Sep 08, 2025 at 02:05:04PM +0000, Yajun Deng wrote:
> If the anon_vma_alloc() is called, the num_children of the parent of
> the anon_vma will be updated. But this operation occurs outside of
> anon_vma_alloc(). There are two callers, one has itself as its parent,
> while another has a real parent. That means they have the same logic.

No they do not. As I explained to you at length.

>
> The update of num_active_vmas and vma->anon_vma are not performed
> together. These operations should be performed under a function.
>
> Add an __anon_vma_alloc() function that implements anon_vma_alloc().
> If the caller has a real parent, called __anon_vma_alloc() and pass
> the parent to it. If it not, called anon_vma_alloc() directly. It will
> set the parent and root of the anon_vma and also updates the num_children
> of its parent anon_vma.

Doing exactly what I told you not to do...

>
> Introduce vma_attach_anon() and vma_detach_anon() to update
> num_active_vmas with vma->anon_vma together.
>
> Signed-off-by: Yajun Deng <yajun.deng@xxxxxxxxx>

NAK.

There's so much wrong with this and you've ignored review Liam and I have
spent time giving you (a resource which I have very little of), it's simply
not a good use of my time to look at this further.

Please abandon this idea, it's not good, and you're not implementing it
well.

Thanks, Lorenzo