linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Matthew Wilcox <willy@infradead.org>
To: "Vishal Moola (Oracle)" <vishal.moola@gmail.com>
Cc: linux-mm@kvack.org, akpm@linux-foundation.org,
	sidhartha.kumar@oracle.com, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/3] mm/migrate: Convert isolate_movable_page() to use folios
Date: Mon, 23 Jan 2023 15:48:22 +0000	[thread overview]
Message-ID: <Y86sRiKAAOz2myNe@casper.infradead.org> (raw)
In-Reply-To: <Y80wKtc22eHApjEN@casper.infradead.org>

On Sun, Jan 22, 2023 at 12:46:34PM +0000, Matthew Wilcox wrote:
> On Fri, Jan 20, 2023 at 04:56:21PM -0800, Vishal Moola (Oracle) wrote:
> >  int isolate_movable_page(struct page *page, isolate_mode_t mode)
> >  {
> > +	struct folio *folio = page_folio(page);
> >  	const struct movable_operations *mops;
> >  
> >  	/*
> > @@ -71,11 +72,11 @@ int isolate_movable_page(struct page *page, isolate_mode_t mode)
> >  	 * the put_page() at the end of this block will take care of
> >  	 * release this page, thus avoiding a nasty leakage.
> >  	 */
> > -	if (unlikely(!get_page_unless_zero(page)))
> > +	if (unlikely(!folio_try_get(folio)))
> 
> This changes behaviour.  Previously when called on a tail page, the
> call failed.  Now it succeeds, getting a ref on something that at
> least was the folio head at some point.
> 
> If you're going to do this, you need to recheck that the page is still
> part of the folio after getting the ref (see gup.c for an example).
> But I think we should probably maintain the behaviour of failing on
> tail pages.
> 
> Maybe something like ...
> 
> 	if (unlikely(!get_page_unless_zero(page)))
> 		goto out;
> 	/* Refcount is zero on tail pages, so we must have a head */
> 	folio = (struct folio *)page;

I've been thinking about this some more as I don't like doing these
kinds of casts (except in the helper functions).  What do you think
to adding:

struct folio *folio_get_nontail_page(struct page *)
{
	if unlikely(!get_page_unless_zero(page))
		return NULL;
	return (struct folio *)page;
}

and then isolate_movable_page() looks like:

	struct folio *folio;
[...]

	folio = folio_get_nontail_page(page);
	if (!folio)
		goto out;

I keep thinking about how this is all going to work when we get to
one-pointer-per-page.  Telling tail pages from head pages becomes hard.
This probably becomes an out-of-line function that looks something like ..

	struct memdesc *memdesc = READ_ONCE(page->memdesc);
	struct folio *folio;

	if (!memdesc_is_folio(memdesc))
		return NULL;
	folio = memdesc_folio(memdesc);
	if (!folio_try_get(folio))
		return NULL;
	if (READ_ONCE(page->memdesc) != memdesc ||
	    folio->pfn != page_pfn(page)) {
		folio_put(folio);
		return NULL;
	}

	return folio;

(note: We need to check that page->memdesc still points to this folio after 
getting the refcount on it.  we could loop around if it fails, but
failing the entire get is OK; if memdesc changed then either before this
function was called, after this function was called or while this
function was called, the refcount on the memdesc it was pointing to
was zero)


  reply	other threads:[~2023-01-23 15:48 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-01-21  0:56 [PATCH 0/3] Convert a couple migrate functions " Vishal Moola (Oracle)
2023-01-21  0:56 ` [PATCH 1/3] mm: Add folio_movable_ops() Vishal Moola (Oracle)
2023-01-21  0:56 ` [PATCH 2/3] mm/migrate: Convert isolate_movable_page() to use folios Vishal Moola (Oracle)
2023-01-22 12:46   ` Matthew Wilcox
2023-01-23 15:48     ` Matthew Wilcox [this message]
2023-01-21  0:56 ` [PATCH 3/3] mm/migrate: Convert putback_movable_pages() " Vishal Moola (Oracle)

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Y86sRiKAAOz2myNe@casper.infradead.org \
    --to=willy@infradead.org \
    --cc=akpm@linux-foundation.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=sidhartha.kumar@oracle.com \
    --cc=vishal.moola@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox