[PATCH] - Allow architectures to increase size of MAX_NR_MEMBLKS

[PATCH] - Allow architectures to increase size of MAX_NR_MEMBLKS

[Jack Steiner]

This fixes a problem that occurs on system with >64 nodes. 

Previously, MAX_NR_MEMBLKS was defined as BITS_PER_LONG. This
patch allows an architecture to override this definition by
defining a value in the arch-specific asm-xxx/mmzone.h file.





# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
#	           ChangeSet	1.1401  -> 1.1402 
#	include/linux/mmzone.h	1.45    -> 1.46   
#	include/asm-ia64/mmzone.h	1.6     -> 1.7    
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/11/05	steiner@attica.americas.sgi.com	1.1402
# Allow architectures to supply their own value for MAX_NR_MEMBLKS.
# The previous value (64) is not large enough for some systems.
# --------------------------------------------
#



diff -Nru a/include/asm-ia64/mmzone.h b/include/asm-ia64/mmzone.h
--- a/include/asm-ia64/mmzone.h	Wed Nov  5 12:07:34 2003
+++ b/include/asm-ia64/mmzone.h	Wed Nov  5 12:07:34 2003
@@ -27,6 +27,8 @@
 # define NR_MEMBLKS		(NR_NODES)
 #endif
 
+#define MAX_NR_MEMBLKS		NR_MEMBLKS
+
 extern unsigned long max_low_pfn;
 
 #define pfn_valid(pfn)		(((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
diff -Nru a/include/linux/mmzone.h b/include/linux/mmzone.h
--- a/include/linux/mmzone.h	Wed Nov  5 12:07:34 2003
+++ b/include/linux/mmzone.h	Wed Nov  5 12:07:34 2003
@@ -287,12 +287,6 @@
 extern void setup_per_zone_pages_min(void);
 
 
-#ifdef CONFIG_NUMA
-#define MAX_NR_MEMBLKS	BITS_PER_LONG /* Max number of Memory Blocks */
-#else /* !CONFIG_NUMA */
-#define MAX_NR_MEMBLKS	1
-#endif /* CONFIG_NUMA */
-
 #include <linux/topology.h>
 /* Returns the number of the current Node. */
 #define numa_node_id()		(cpu_to_node(smp_processor_id()))
@@ -322,6 +316,14 @@
 #endif
 
 #endif /* !CONFIG_DISCONTIGMEM */
+
+#ifdef CONFIG_NUMA
+#ifndef MAX_NR_MEMBLKS
+#define MAX_NR_MEMBLKS	BITS_PER_LONG /* Max number of Memory Blocks */
+#endif
+#else /* !CONFIG_NUMA */
+#define MAX_NR_MEMBLKS	1
+#endif /* CONFIG_NUMA */
 
 #if NODES_SHIFT > MAX_NODES_SHIFT
 #error NODES_SHIFT > MAX_NODES_SHIFT




-- 
Thanks

Jack Steiner (steiner@sgi.com)          651-683-5302
Principal Engineer                      SGI - Silicon Graphics, Inc.

Re: [PATCH] - Allow architectures to increase size of MAX_NR_MEMBLKS

[Christoph Hellwig]

On Wed, Nov 05, 2003 at 12:19:11PM -0600, Jack Steiner wrote:
> This fixes a problem that occurs on system with >64 nodes. 
> 
> Previously, MAX_NR_MEMBLKS was defined as BITS_PER_LONG. This
> patch allows an architecture to override this definition by
> defining a value in the arch-specific asm-xxx/mmzone.h file.

IMHO this is too much clutter.  Just make it mandatory for the
architectures to define their own MAX_NR_MEMBLKS in the numa case.

Or actually even better just have a 

#ifndef MAX_NR_MEMBLKS
#define MAX_NR_MEMBLKS	1
#endif

in the generic code and let every architecture that wants to override
it do so.

Re: [PATCH] - Allow architectures to increase size of MAX_NR_MEMBLKS

[Jack Steiner]

On Thu, Nov 06, 2003 at 08:57:35AM +0000, Christoph Hellwig wrote:
> On Wed, Nov 05, 2003 at 12:19:11PM -0600, Jack Steiner wrote:
> > This fixes a problem that occurs on system with >64 nodes. 
> > 
> > Previously, MAX_NR_MEMBLKS was defined as BITS_PER_LONG. This
> > patch allows an architecture to override this definition by
> > defining a value in the arch-specific asm-xxx/mmzone.h file.
> 
> IMHO this is too much clutter.  Just make it mandatory for the
> architectures to define their own MAX_NR_MEMBLKS in the numa case.
> 
> Or actually even better just have a 
> 
> #ifndef MAX_NR_MEMBLKS
> #define MAX_NR_MEMBLKS	1
> #endif
> 
> in the generic code and let every architecture that wants to override
> it do so.


I like this idea but wanted to minimize risk to other architectures 
at this point in 2.6.

AFAICT, the only other achitecture that would require change would 
be i386. I'm not familar with the i386 numa architecture. It
looks like the change would be to add the definition of
MAX_NR_MEMBLKS to include/asm-i386/numaq.h.

What is the value of MAX_NR_MEMBLKS for NUMAQ? 

Are any architectures affected that I am overlooking???

I'm ok with either approach. What do other think??


-- 
Thanks

Jack Steiner (steiner@sgi.com)          651-683-5302
Principal Engineer                      SGI - Silicon Graphics, Inc.