Re: [PATCH] mm/page_alloc: fix alignment for alloc_contig_pages_noprof()

[Michal Hocko <mhocko@suse.com>]

On Thu 25-09-25 11:22:26, David Hildenbrand wrote:
> On 25.09.25 10:14, Michal Hocko wrote:
> > On Wed 24-09-25 14:19:37, David Hildenbrand wrote:
> > > On 24.09.25 14:01, Michal Hocko wrote:
> > [...]
> > > > I thought we have concluded that there is no user expecting a specific
> > > > alignment. Or have I misunderstood that?
> > > 
> > > We concluded that for users that don't pass in non-power-2 requests.
> > 
> > Ok, my misreading of the discussion then.
> > 
> > > hugetlb->folio_alloc_gigantic()->alloc_contig_pages_noprof() certainly
> > > expects a suitable alignment :)
> > 
> > True that.
> > 
> > So effectivelly there is nothing to be done here right. Those that
> > provide power-two requests will get what they have been getting and
> > those non-existent others
> 
> I think kfence might be doing a non-power-of-two allocation. TDX maybe as
> well.
> 
> But neither seems to depend on some alignment, so all good.
> 
> > will get whatever as we do not provide any
> > guarantee in that case. Maybe something we could add as a comment.
> 
> We currently have
> 
> "The allocated memory is always aligned to a page boundary. If nr_pages is a
> power of two, then allocated range is also guaranteed to be aligned to same
> nr_pages (e.g. 1GB request would be aligned to 1GB)."
> 
> What would be your suggestion?

Suggestion to self to read through the doc before suggesting editings
and suggestion to the patch to not do anything.

Thanks for the patience with me.
-- 
Michal Hocko
SUSE Labs