Re: [PATCH v4 1/2] mm,hwpoison: fix race with compound page allocation

[Naoya Horiguchi <nao.horiguchi@gmail.com>]

On Mon, May 17, 2021 at 12:12:50PM +0200, Oscar Salvador wrote:
> On Mon, May 17, 2021 at 01:54:00PM +0900, Naoya Horiguchi wrote:
> > From: Naoya Horiguchi <naoya.horiguchi@nec.com>
> > 
> > When hugetlb page fault (under overcommitting situation) and
> > memory_failure() race, VM_BUG_ON_PAGE() is triggered by the following race:
> > 
> >     CPU0:                           CPU1:
> > 
> >                                     gather_surplus_pages()
> >                                       page = alloc_surplus_huge_page()
> >     memory_failure_hugetlb()
> >       get_hwpoison_page(page)
> >         __get_hwpoison_page(page)
> >           get_page_unless_zero(page)
> >                                       zero = put_page_testzero(page)
> >                                       VM_BUG_ON_PAGE(!zero, page)
> >                                       enqueue_huge_page(h, page)
> >       put_page(page)
> > 
> > __get_hwpoison_page() only checks page refcount before taking additional
>                                   ^^ the?                      ^^ an
> > one for memory error handling, which is wrong because there's a time
> > window where compound pages have non-zero refcount during initialization.
> > 
> > So makes __get_hwpoison_page() check page status a bit more for a few
>      ^^ make
> > types of compound pages. PageSlab() check is added because otherwise
> > "non anonymous thp" path is wrongly chosen.
> 
> This is no longer true with this patch, is it? What happened here?

Sorry, we need remove this. I dropped the changes for PageSlab
because I'm still not sure about what to do (I'd like to do it
in a separate work).

> 
> >  static int __get_hwpoison_page(struct page *page)
> >  {
> >  	struct page *head = compound_head(page);
> > +	int ret = 0;
> > +
> > +#ifdef CONFIG_HUGETLB_PAGE
> > +	spin_lock(&hugetlb_lock);
> > +	if (PageHuge(head) && (HPageFreed(head) || HPageMigratable(head)))
> > +		ret = get_page_unless_zero(head);
> > +	spin_unlock(&hugetlb_lock);
> > +	if (ret > 0)
> > +		return ret;
> > +#endif
> 
> I am kind of fine with this, but I wonder whether it makes sense to hide this
> details into helper (with an empty stub for non-hugetlb pages)?

OK, I will do this.

> 
> >  	if (!PageHuge(head) && PageTransHuge(head)) {
> This !PageHuge could go?

I'll update this, too.

Thanks,
Naoya Horiguchi