[PATCH v2 07/19] mm/fork: Accept huge pfnmap entries

[Peter Xu]

On Thu, Aug 29, 2024 at 09:44:01PM +0200, David Hildenbrand wrote:
> On 29.08.24 20:26, Peter Xu wrote:
> > On Thu, Aug 29, 2024 at 05:10:42PM +0200, David Hildenbrand wrote:
> > > On 26.08.24 22:43, Peter Xu wrote:
> > > > Teach the fork code to properly copy pfnmaps for pmd/pud levels.  Pud is
> > > > much easier, the write bit needs to be persisted though for writable and
> > > > shared pud mappings like PFNMAP ones, otherwise a follow up write in either
> > > > parent or child process will trigger a write fault.
> > > > 
> > > > Do the same for pmd level.
> > > > 
> > > > Signed-off-by: Peter Xu <peterx at redhat.com>
> > > > ---
> > > >    mm/huge_memory.c | 29 ++++++++++++++++++++++++++---
> > > >    1 file changed, 26 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > > > index e2c314f631f3..15418ffdd377 100644
> > > > --- a/mm/huge_memory.c
> > > > +++ b/mm/huge_memory.c
> > > > @@ -1559,6 +1559,24 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
> > > >    	pgtable_t pgtable = NULL;
> > > >    	int ret = -ENOMEM;
> > > > +	pmd = pmdp_get_lockless(src_pmd);
> > > > +	if (unlikely(pmd_special(pmd))) {
> > > 
> > > I assume I have to clean up your mess here as well?
> > 
> > Can you leave meaningful and explicit comment?  I'll try to address.
> 
> Sorry Peter, but I raised all that as reply to v1. For example, I stated
> that vm_normal_page_pmd() already *exist* and why these pmd_special() checks
> should be kept there.

We discussed the usage of pmd_page() but I don't think this is clear you
suggest it to be used there.  IOW, copy_huge_pmd() doesn't use
vm_normal_page_pmd() yet so far and I'm not sure whether it's always safe.

E.g. at least one thing I spot is vm_normal_page_pmd() returns NULL for
huge zeropage pmd but here in fork() we need to take a ref with
mm_get_huge_zero_folio().

> 
> I hear you, you're not interested in cleaning that up. So at this point it's
> easier for me to clean it up myself.

It might be easier indeed you provide a patch that you think the best.

Then I'll leave that to you, and I'll send the solo fixup patch to be
squashed soon to the list.

-- 
Peter Xu