Re: [PATCH 2/3] userfaultfd: convert __mcopy_atomic_hugetlb() to use a folio

[zhangpeng (AS)]

On 2023/3/14 16:31, Matthew Wilcox wrote:

> On Tue, Mar 14, 2023 at 03:37:33AM +0000, Peng Zhang wrote:
> > +++ b/include/linux/mm.h
> > @@ -3546,9 +3546,8 @@ extern void copy_user_huge_page(struct page *dst, struct page *src,
> >   				unsigned long addr_hint,
> >   				struct vm_area_struct *vma,
> >   				unsigned int pages_per_huge_page);
> > -extern long copy_huge_page_from_user(struct page *dst_page,
> > +extern long copy_large_folio_from_user(struct folio *dst_folio,
> You can drop the 'extern'.

Got it.

> > +++ b/mm/memory.c
> > @@ -5769,26 +5769,28 @@ void copy_user_huge_page(struct page *dst, struct page *src,
> >   	process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg);
> >   }
> >   
> > -long copy_huge_page_from_user(struct page *dst_page,
> > +long copy_large_folio_from_user(struct folio *dst_folio,
> >   				const void __user *usr_src,
> > -				unsigned int pages_per_huge_page,
> >   				bool allow_pagefault)
> >   {
> >   	void *page_kaddr;
> >   	unsigned long i, rc = 0;
> > -	unsigned long ret_val = pages_per_huge_page * PAGE_SIZE;
> > +	unsigned int nr_pages = folio_nr_pages(dst_folio);
> > +	unsigned long ret_val = nr_pages * PAGE_SIZE;
> >   	struct page *subpage;
> > +	struct folio *inner_folio;
> What is an 'inner folio'?
>
> > -	for (i = 0; i < pages_per_huge_page; i++) {
> > -		subpage = nth_page(dst_page, i);
> > +	for (i = 0; i < nr_pages; i++) {
> > +		subpage = folio_page(dst_folio, i);
> > +		inner_folio = page_folio(subpage);
> >   		if (allow_pagefault)
> > -			page_kaddr = kmap(subpage);
> > +			page_kaddr = kmap_local_folio(inner_folio, 0);
> This doesn't do what you think it does.  Did you test this?
>
> >   		else
> >   			page_kaddr = kmap_atomic(subpage);
> Pretty sure all this should be converted to kmap_local and the atomic
> bits should go away.
>
> >   		rc = copy_from_user(page_kaddr,
> >   				usr_src + i * PAGE_SIZE, PAGE_SIZE);
> >   		if (allow_pagefault)
> > -			kunmap(subpage);
> > +			kunmap_local(page_kaddr);
> >   		else
> >   			kunmap_atomic(page_kaddr);
> >   
> > @@ -5796,7 +5798,7 @@ long copy_huge_page_from_user(struct page *dst_page,
> >   		if (rc)
> >   			break;
> >   
> > -		flush_dcache_page(subpage);
> > +		flush_dcache_folio(inner_folio);
> The flush should probably be pulled outside the loop.
>
> > +			err = copy_large_folio_from_user(folio,
> > +						(const void __user *) src_addr, true);
> I wonder if this shouldn't be 'copy_folio_from_user()'.  after all,
> it'll work for any size folio, right?

Thanks for your review.

I'll rename copy_large_folio_from_user() to copy_folio_from_user().
I'll delete the inner_folio. kmap() and kmap_atomic() will be
converted to the following code.

	page_kaddr = kmap_local_page(subpage);
	if (!allow_pagefault)
		pagefault_disable();
	rc = copy_from_user(page_kaddr,
			usr_src + i * PAGE_SIZE, PAGE_SIZE);
	if (!allow_pagefault)
		pagefault_enable();
	kunmap_local(page_kaddr);

flush_dcache_folio() will be placed outside the loop.

I'll fix all this in a v2 of this patch series.


Thanks,
Peng.