Possible partial miss in pages needed for zone's memory map?

Possible partial miss in pages needed for zone's memory map?

[Johannes Weiner]

Hi,

I stumbled over the following in the zone initialization code.  Please
let me know if I overlooked something here.

---

From: Johannes Weiner <hannes@saeurebad.de>
Subject: [PATCH] Don't drop a partial page in a zone's memory map size

In a zone's present pages number, account for all pages occupied by the
memory map, including a partial.

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
---

--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3378,7 +3378,8 @@ static void __paginginit free_area_init_
 		 * is used by this zone for memmap. This affects the watermark
 		 * and per-cpu initialisations
 		 */
-		memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
+		memmap_pages =
+			PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
 		if (realsize >= memmap_pages) {
 			realsize -= memmap_pages;
 			printk(KERN_DEBUG

Re: Possible partial miss in pages needed for zone's memory map?

[Andrew Morton]

On Sat, 17 May 2008 14:19:15 +0200 Johannes Weiner <hannes@saeurebad.de> wrote:

> Hi,
> 
> I stumbled over the following in the zone initialization code.  Please
> let me know if I overlooked something here.
> 

hm, no takers.  Let's add linux-mm.

> 
> From: Johannes Weiner <hannes@saeurebad.de>
> Subject: [PATCH] Don't drop a partial page in a zone's memory map size
> 
> In a zone's present pages number, account for all pages occupied by the
> memory map, including a partial.
> 
> Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
> ---
> 
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3378,7 +3378,8 @@ static void __paginginit free_area_init_
>  		 * is used by this zone for memmap. This affects the watermark
>  		 * and per-cpu initialisations
>  		 */
> -		memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
> +		memmap_pages =
> +			PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
>  		if (realsize >= memmap_pages) {
>  			realsize -= memmap_pages;
>  			printk(KERN_DEBUG

I looked in there for 30 seconds and collapsed in confusion over which
variables are in which units.

Hint: never ever name a variable or a /proc file or your cat 
or anything else anything dimensionless like "size".  It can always be
replaced with something which communicates its units.  zones_nrbytes,
zholes_nrpages, etc.

Re: Possible partial miss in pages needed for zone's memory map?

[KAMEZAWA Hiroyuki]

On Mon, 19 May 2008 23:19:37 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Sat, 17 May 2008 14:19:15 +0200 Johannes Weiner <hannes@saeurebad.de> wrote:
> 
> > Hi,
> > 
> > I stumbled over the following in the zone initialization code.  Please
> > let me know if I overlooked something here.
> > 
> 
> hm, no takers.  Let's add linux-mm.
> 
> > 
> > From: Johannes Weiner <hannes@saeurebad.de>
> > Subject: [PATCH] Don't drop a partial page in a zone's memory map size
> > 
> > In a zone's present pages number, account for all pages occupied by the
> > memory map, including a partial.
> > 
> > Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
> > ---
> > 
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -3378,7 +3378,8 @@ static void __paginginit free_area_init_
> >  		 * is used by this zone for memmap. This affects the watermark
> >  		 * and per-cpu initialisations
> >  		 */
> > -		memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
> > +		memmap_pages =
> > +			PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
> >  		if (realsize >= memmap_pages) {
> >  			realsize -= memmap_pages;
> >  			printk(KERN_DEBUG
> 
> I looked in there for 30 seconds and collapsed in confusion over which
> variables are in which units.
> 
Hmm, size * sizeof(struct page) is  multiple of PAGE_SIZE in many case.
Becasue "size" is always alinged to (1 << MAX_ORDER -1) (I believe...).

ex.) In x86 case,
   (1 << (MAX_ORDER(11) - 1)) * 4 (sizeof(long)) = (1 << 12) = PAGE_SIZE.

But not sure on other archs with various params.
I think above fix is correct.

Thanks,
-Kame