Re: [PATCH] sparc: Use vmemmap_populate_hugepages for vmemmap_populate

[Tao pilgrim]

On Thu, Dec 18, 2025 at 9:17 PM Mike Rapoport <rppt@xxxxxxxxxx> wrote:
>
> On Thu, Dec 18, 2025 at 08:29:27PM +0800, Tao pilgrim wrote:
> > On Thu, Dec 18, 2025 at 8:11 PM Mike Rapoport <rppt@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Dec 18, 2025 at 07:45:52PM +0800, Tao pilgrim wrote:
> > > > On Thu, Dec 18, 2025 at 7:14 PM Mike Rapoport <rppt@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Wed, Dec 17, 2025 at 08:08:58PM +0800, chengkaitao wrote:
> > > > > > From: Chengkaitao <chengkaitao@xxxxxxxxxx>
> > > > > >
> > > > > > 1. Added the vmemmap_false_pmd function to accommodate architectures
> > > > > > that do not support basepages.
> > > > > > 2. In the SPARC architecture, reimplemented vmemmap_populate using
> > > > > > vmemmap_populate_hugepages.
> > > > >
> > > > > What problem this patch is trying to solve?
> > > >
> > > > Following the pattern established by other architectures, vmemmap_populate
> > > > is reimplemented using vmemmap_populate_hugepages in the SPARC architecture,
> > > > which fulfills the intended purpose of vmemmap_populate_hugepages.
> > >
> > > But what's wrong with the current sparc::vmemmap_populate()?
> >
> > The current implementation of sparc::vmemmap_populate() is correct
> > and functional.
> >
> > However, I believe that generalizing vmemmap_populate_hugepages()
> > still results in certain architectures requiring special-case handling
> > branches, which is contrary to our goal of a unified implementation.
> >
> > Another benefit is that it allows us to eliminate the external
> > declarations of vmemmap_p?d_populate functions and convert
> > them to static functions.
>
> If the goals are to provide a unified implementation and eliminate the
> external declarations of vmemmap_p?d_populate() functions, I'd suggest to
> send all this a single series rather than a single patch that updates sparc
> implementation.
>
> Regarding the ability of sparc to fallback to vmemmap_populate_basepages(),
> it does not seem to me that this is an architectural limitation. sparc uses
> base pages in the kernel page tables, so it should be able to use them in
> vmemmap as well.

If SPARC can use base pages in vmemmap, I might remove the
vmemmap_false_pmd implementation. Thanks for your review.

> > > > The functional logic remains unchanged.
> > > >
> > > > > > Signed-off-by: Chengkaitao <chengkaitao@xxxxxxxxxx>
> > > > > > ---
> >

-- 
Yours,
Kaitao Cheng