Re: [PATCH] mm/huge_memory.c: introduce split_unmapped_folio_to_order

[David Hildenbrand (Red Hat)]

On 13.11.25 17:24, Zi Yan wrote:
> On 13 Nov 2025, at 11:02, Lorenzo Stoakes wrote:
>
> > On Thu, Nov 13, 2025 at 04:36:01PM +0100, Francois Dugast wrote:
> > > Hi Balbir,
> > >
> > > On Wed, Nov 12, 2025 at 03:46:33PM +1100, Balbir Singh wrote:
> > > > Unmapped was added as a parameter to __folio_split() and related
> > > > call sites to support splitting of folios already in the midst
> > > > of a migration. This special case arose for device private folio
> > > > migration since during migration there could be a disconnect between
> > > > source and destination on the folio size.
> > > >
> > > > Introduce split_unmapped_folio_to_order() to handle this special case.
> > > > This in turn removes the special casing introduced by the unmapped
> > > > parameter in __folio_split().
> > >
> > > Such a helper would be needed in drm_pagemap_migrate_to_devmem when
> > > reallocating a device folio to smaller pages.
> > >
> > > Could we export it (EXPORT_SYMBOL)?
>
> drm_pagemap_migrate_to_devmem() is a function defined in tree, you
> just need to include huge_mm.h to use split_unmapped_folio_to_order().
> Why do you need to export split_unmapped_folio_to_order()?

I guess because DRM_GPUSVM is tristate, so can be built as a module. 
IIUC, that's where drm_pagemap_migrate_to_devmem() ends up.

> > As a rule we don't export things from core mm. And certainly not to non-GPL
> > modules.
> >
> > Unless David feels very differently or there's some enormously compelling
> > reason for it I'd really rather we didn't.

We'd need a pretty good reason to go down that path indeed :)

--
Cheers

David