linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Matthew Wilcox <willy@infradead.org>
To: Christopher Lameter <cl@linux.com>
Cc: linux-mm@kvack.org
Subject: Re: [PATCH] slub: Remove use of page->counter
Date: Wed, 11 Apr 2018 11:26:06 -0700	[thread overview]
Message-ID: <20180411182606.GA22494@bombadil.infradead.org> (raw)
In-Reply-To: <alpine.DEB.2.20.1804101702240.30842@nuc-kabylake>

On Tue, Apr 10, 2018 at 05:03:17PM -0500, Christopher Lameter wrote:
> On Tue, 10 Apr 2018, Matthew Wilcox wrote:
> 
> > > Is this aligned on a doubleword boundary? Maybe move the refcount below
> > > the flags field?
> >
> > You need freelist and _mapcount to be in the same dword.  There's no
> > space to put them both in dword 0, so that's used for flags and mapping
> > / s_mem.  Then freelist, mapcount and refcount are in dword 1 (on 64-bit),
> > or freelist & mapcount are in dword 1 on 32-bit.  After that, 32 and 64-bit
> > no longer line up on the same dword boundaries.
> 
> Well its no longer clear from the definitions that this must be the case.
> Clarify that in the next version?

I had a Thought.  And it seems to work:

struct page {
	unsigned long flags;
	union {		/* Five words */
		struct {	/* Page cache & anonymous pages */
			struct list_head lru;
			unsigned long private;
			struct address_space *mapping;
			pgoff_t index;
		};
		struct {	/* Slab / Slob / Slub */
			struct page *next;
			void *freelist;
			union {
				unsigned int active;		/* slab */
				struct {			/* slub */
					unsigned inuse:16;
					unsigned objects:15;
					unsigned frozen:1;
				};
				int units;			/* slob */
			};
#ifdef CONFIG_64BIT
			int pages;
#endif
			void *s_mem;
			struct kmem_cache *slab_cache;
		};
		struct rcu_head rcu_head;
		... tail pages, page tables, etc, etc ...
	};
	union {
		atomic_t _mapcount;
		unsigned int page_type;
#ifdef CONFIG_64BIT
		unsigned int pobjects;			/* slab */
#else
		struct {
			short int pages;
			short int pobjects;
		};
#endif
	};
	atomic_t _refcount;
	struct mem_cgroup *mem_cgroup;
};

Now everybody gets 5 contiguous words to use as they want with the only
caveat that they can't use bit 0 of the first word (PageTail).  It looks
a little messy to split up pages & pobjects like that -- as far as I
can see there's no reason we couldn't make them unsigned short on 64BIT?
pages is always <= pobjects, and pobjects is limited to 2^15.

  reply	other threads:[~2018-04-11 18:26 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-04-10 19:54 Matthew Wilcox
2018-04-10 20:47 ` Christopher Lameter
2018-04-10 20:57   ` Matthew Wilcox
2018-04-10 22:03     ` Christopher Lameter
2018-04-11 18:26       ` Matthew Wilcox [this message]
2018-04-16 13:53         ` Matthew Wilcox
2018-04-16 15:08           ` Christopher Lameter

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=20180411182606.GA22494@bombadil.infradead.org \
    --to=willy@infradead.org \
    --cc=cl@linux.com \
    --cc=linux-mm@kvack.org \
    /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