2003-11-05 21:16:16

by Jesse Barnes

Subject: [PATCH] use NODES_SHIFT to calculate ZONE_SHIFT

Now that we have a proper NODES_SHIFT value, we need to use it to define
ZONE_SHIFT otherwise we'll spill over 8 bits if we have more than 85
nodes. How does this look? The '+2' should really be
log2(MAX_NR_NODES), but I think this is an improvement over what was
there.

Thanks,
Jesse

===== include/linux/mm.h 1.133 vs edited =====
--- 1.133/include/linux/mm.h Sun Oct 5 01:07:49 2003
+++ edited/include/linux/mm.h Tue Nov 4 16:45:33 2003
@@ -322,8 +322,10 @@
/*
* The zone field is never updated after free_area_init_core()
* sets it, so none of the operations on it need to be atomic.
+ * We'll have up to log2(MAX_NUMNODES * MAX_NR_ZONES) zones
+ * total, so we use NODES_SHIFT here to get enough bits.
*/
-#define ZONE_SHIFT (BITS_PER_LONG - 8)
+#define ZONE_SHIFT (BITS_PER_LONG - (NODES_SHIFT + 2))

struct zone;
extern struct zone *zone_table[];

2003-11-07 22:28:26

by Jesse Barnes

Subject: Re: [PATCH] use NODES_SHIFT to calculate ZONE_SHIFT

On Fri, Nov 07, 2003 at 01:58:06PM -0800, Andrew Morton wrote:
> [email protected] (Jesse Barnes) wrote:
> >
> > Now that we have a proper NODES_SHIFT value, we need to use it to define
> > ZONE_SHIFT otherwise we'll spill over 8 bits if we have more than 85
> > nodes. How does this look? The '+2' should really be
> > log2(MAX_NR_NODES), but I think this is an improvement over what was
> > there.
>
> You mean log2(MAX_NR_ZONES).

Yep, sorry.

> How about we do it this way, so at least the duplicated information is on
> adjacent lines, and they are unlikely to get out of sync?

Yeah, this looks great.

Thanks,
Jesse

2003-11-07 22:33:29

by Andrew Morton

Subject: Re: [PATCH] use NODES_SHIFT to calculate ZONE_SHIFT

[email protected] (Jesse Barnes) wrote:
>
> Now that we have a proper NODES_SHIFT value, we need to use it to define
> ZONE_SHIFT otherwise we'll spill over 8 bits if we have more than 85
> nodes. How does this look? The '+2' should really be
> log2(MAX_NR_NODES), but I think this is an improvement over what was
> there.

You mean log2(MAX_NR_ZONES).

How about we do it this way, so at least the duplicated information is on
adjacent lines, and they are unlikely to get out of sync?



diff -puN include/linux/mm.h~ZONE_SHIFT-from-NODES_SHIFT include/linux/mm.h
--- 25/include/linux/mm.h~ZONE_SHIFT-from-NODES_SHIFT Fri Nov 7 13:51:22 2003
+++ 25-akpm/include/linux/mm.h Fri Nov 7 13:55:11 2003
@@ -322,8 +322,10 @@ static inline void put_page(struct page
/*
* The zone field is never updated after free_area_init_core()
* sets it, so none of the operations on it need to be atomic.
+ * We'll have up to log2(MAX_NUMNODES * MAX_NR_ZONES) zones
+ * total, so we use NODES_SHIFT here to get enough bits.
*/
-#define ZONE_SHIFT (BITS_PER_LONG - 8)
+#define ZONE_SHIFT (BITS_PER_LONG - NODES_SHIFT - MAX_NR_ZONES_SHIFT)

struct zone;
extern struct zone *zone_table[];
diff -puN include/linux/mmzone.h~ZONE_SHIFT-from-NODES_SHIFT include/linux/mmzone.h
--- 25/include/linux/mmzone.h~ZONE_SHIFT-from-NODES_SHIFT Fri Nov 7 13:51:49 2003
+++ 25-akpm/include/linux/mmzone.h Fri Nov 7 13:57:19 2003
@@ -159,7 +159,10 @@ struct zone {
#define ZONE_DMA 0
#define ZONE_NORMAL 1
#define ZONE_HIGHMEM 2
-#define MAX_NR_ZONES 3
+
+#define MAX_NR_ZONES 3 /* Sync this with MAX_NR_ZONES_SHIFT */
+#define MAX_NR_ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
+
#define GFP_ZONEMASK 0x03

/*

_