Re: [PATCH v2] mm/vma: don't remove VMA from rmap if pgoff unchanged

[Lance Yang <lance.yang@linux.dev>]

On 2026/10/2 22:05, Lorenzo Stoakes (ARM) wrote:
> On Thu, Oct 01, 2026 at 11:45:11PM +0800, Lance Yang wrote:
[...]
>>> +static void anon_rmap_tree_update_inplace(struct anon_vma_chain *avc)
>>> +{
>>> +#ifdef CONFIG_DEBUG_VM_RB
>>> +	avc->cached_vma_last = avc_last_pgoff(avc);
>>> +#endif
>>> +	/* Propagate all the way up the tree. */
>>
>> Nit: propagate() can stop early when rb_subtree_last is unchanged ...
>>
>> Maybe:
>>
>> /* Update the subtree maximum and propagate any changes up the tree. */
> 
> I think in this case it's ok to be a bit blurry about it :P it deciding not to
> unnecessary work is fine but I don't want to put too much in there.
> 
> The point is as a simple sign or pointer to help somebody wondering wtf that's
> for even if it's not quite the full story!
> 
> Hopefully that's ok? :)

Fair enough. Let’s leave it as-is ;)

> 
>>
>>> +	__anon_rmap_tree_augment.propagate(&avc->rb, NULL);
>>> +}
>>> +
>> [...]
>>
>> Acked-by: Lance Yang <lance.yang@linux.dev>
> 
> Thanks :)
> 
>>
>> Hammered it with VMA churn (split/merge/mremap/madvise/fork) + concurrent
>> rmap walks + hwpoison injection. Nothing complained :D
>>
>> Tested-by: Lance Yang <lance.yang@linux.dev>
> 
> Thanks, very much appreciated! :)

Cheers!