Received-SPF: pass (google.com: domain of linux-kernel+bounces-191094-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Mon, 27 May 2024 16:26:47 +0100
From: Matthew Wilcox <willy@infradead.org>
To: David Hildenbrand <david@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	Andrew Morton <akpm@linux-foundation.org>,
	Mike Rapoport <rppt@kernel.org>, Minchan Kim <minchan@kernel.org>,
	Sergey Senozhatsky <senozhatsky@chromium.org>,
	Hyeonggon Yoo <42.hyeyoo@gmail.com>
Subject: Re: [PATCH v1 2/6] mm: allow reuse of the lower 16 bit of the page
 type with an actual type
Message-ID: <ZlSmNzgkeYaCrPWc@casper.infradead.org>
References: <20240527141454.113132-1-david@redhat.com>
 <20240527141454.113132-3-david@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240527141454.113132-3-david@redhat.com>

On Mon, May 27, 2024 at 04:14:50PM +0200, David Hildenbrand wrote:
> As long as the owner sets a page type first, we can allow reuse of the
> lower 18 bit: sufficient to store an offset into a 64 KiB page, which

You say 18 here and 16 below.

> is the maximum base page size in *common* configurations (ignoring the
> 256 KiB variant). Restrict it to the head page.
> 
> We'll use that for zsmalloc next, to set a proper type while still
> reusing that field to store information (offset into a base page) that
> cannot go elsewhere for now.
> 
> Fear of running out of bits for storing the actual type? Actually, we
> don't need one bit per type, we could store a single value instead.
> Further, we could likely limit PAGE_TYPE_BASE to a single (highest) bit.

We could, but it's more instructions to check.

> +++ b/include/linux/page-flags.h
> @@ -945,14 +945,18 @@ PAGEFLAG_FALSE(HasHWPoisoned, has_hwpoisoned)
>   */
>  
>  #define PAGE_TYPE_BASE	0xf0000000
> -/* Reserve		0x0000007f to catch underflows of _mapcount */
> -#define PAGE_MAPCOUNT_RESERVE	-128
> -#define PG_buddy	0x00000080
> -#define PG_offline	0x00000100
> -#define PG_table	0x00000200
> -#define PG_guard	0x00000400
> -#define PG_hugetlb	0x00000800
> -#define PG_slab		0x00001000
> +/*
> + * Reserve		0x0000ffff to catch underflows of _mapcount and
> + * allow owners that set a type to reuse the lower 16 bit for their own
> + * purposes.
> + */
> +#define PAGE_MAPCOUNT_RESERVE	-65536

I think my original comment was misleading.  This should be:

 * Reserve 0xffff0000 - 0xfffffffe to catch _mapcount underflow.

How about we start at the top end and let people extend down?  ie:

#define PAGE_TYPE_BASE	0x80000000
#define PG_buddy	0x40000000
#define PG_offline	0x20000000
#define PG_table	0x10000000
#define PG_guard	0x08000000
#define PG_hugetlb	0x04000000
#define PG_slab		0x02000000
#define PAGE_MAPCOUNT_RESERVE	(~0x0000ffff)

Now we can see that we have 9 flags remaining, which should last until
we can have proper memdesc typing.