Re: [PATCH] x86/mm/pat: Support splitting of virtual memory areas

[Nam Cao <namcao@linutronix.de>]

On Sun, Aug 25, 2024 at 05:04:44PM +0100, Lorenzo Stoakes wrote:
> On Sun, Aug 25, 2024 at 05:24:03PM GMT, Nam Cao wrote:
> 
> [snip]
> 
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index d0dfc85b209b..64067ddb8382 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -2486,6 +2486,12 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
> >  	if (err)
> >  		goto out_free_mpol;
> >
> > +	if (unlikely(vma->vm_flags & VM_PFNMAP)) {
> > +		err = track_pfn_split(vma, addr);
> > +		if (err)
> > +			goto out_vma_unlink;
> > +	}
> > +
> >  	if (new->vm_file)
> >  		get_file(new->vm_file);
> >
> > @@ -2515,6 +2521,8 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
> >  		vma_next(vmi);
> >  	return 0;
> >
> > +out_vma_unlink:
> > +	unlink_anon_vmas(vma);
> >  out_free_mpol:
> >  	mpol_put(vma_policy(new));
> >  out_free_vmi:
> > --
> > 2.39.2
> >
> 
> Right from the start the 6.11rc cycle mm-unstable and therefore -next has
> moved this function out to mm/vma.c, so you will need to make this change
> there rather than against mm/mmap.c (or whichever tree this is intended to
> come through needs to sync up, especially as there's a fairly substantial
> amount of change going on right now in VMA handling).
> 
> Sorry about that!

Ah okay, thanks for lettimg me know.

We could wait for 6.12-rc1 to be out, and then let this patch go to x86
tree. Or we could let it go to mm tree, if x86 maintainers are okay with
that?

Best regards,
Nam