Skip to contentSkip to conversation

2.6.5-rc3-mm4 x86_64 sched domains patch

PATCH Darren Hart 6 participants 10 messages Apr 8, 2004 – Apr 15, 2004

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

The current default implementations of arch_init_sched_domains constructs either a flat or two level topolology. The two level topology is built if CONFIG_NUMA is set. It seems that CONFIG_NUMA is not the appropriate flag to use for constructing a two level topology since some architectures which define CONFIG_NUMA would be better served with a flat topology. x86_64 for example will construct a two level topology with one CPU per node, causing performance problems because balancing within nodes is pointless and balancing across nodes doesn't occur as often.

This patch introduces a new CONFIG_SCHED_NUMA flag and uses it to decide between a flat or two level topology of sched_domains. The patch is minimally invasive as it primarily modifies Kconfig files and sets the appropriate default (off for x86_64, on for everything that used to export CONFIG_NUMA) and should only change the sched_domains topology constructed on x86_64 systems. I have verified this on a 4 node x86 NUMAQ, but need someone to test x86_64.

This patch is intended as a quick fix for the x86_64 problem, and doesn't solve the problem of how to build generic sched domain topologies. We can certainly conceive of various topologies for x86 systems, so even arch specific topologies may not be sufficient. Would sub-arch (ie NUMAQ) be the right way to handle different topologies, or will we be able to autodiscover the appropriate topology? I will be looking into this more, but thought some might benefit from an immediate x86_64 fix. I am very interested in hearing your ideas on this.

Regards,

Darren Hart

arch/alpha/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/alpha/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/alpha/Kconfig
@@ -519,6 +519,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server machines.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 # LARGE_VMALLOC is racy, if you *really* need it then fix it first
 config ALPHA_LARGE_VMALLOC
 	bool
arch/i386/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/i386/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/i386/Kconfig
@@ -772,6 +772,14 @@ config NUMA
 	default n if X86_PC
 	default y if (X86_NUMAQ || X86_SUMMIT)
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 # Need comments to help the hapless user trying to turn on NUMA support
 comment "NUMA (NUMA-Q) requires SMP, 64GB highmem support"
 	depends on X86_NUMAQ && (!HIGHMEM64G || !SMP)
arch/ia64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/ia64/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/ia64/Kconfig
@@ -172,6 +172,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server systems.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config VIRTUAL_MEM_MAP
 	bool "Virtual mem map"
 	default y if !IA64_HP_SIM
arch/mips/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/mips/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/mips/Kconfig
@@ -337,6 +337,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server machines.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config MAPPED_KERNEL
 	bool "Mapped kernel support"
 	depends on SGI_IP27
arch/ppc64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/ppc64/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/ppc64/Kconfig
@@ -173,6 +173,14 @@ config NUMA
 	bool "NUMA support"
 	depends on DISCONTIGMEM
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config SCHED_SMT
 	bool "SMT (Hyperthreading) scheduler support"
 	depends on SMP
arch/x86_64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/arch/x86_64/Kconfig linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/arch/x86_64/Kconfig
@@ -261,6 +261,14 @@ config NUMA
        depends on K8_NUMA
        default y
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default n
+       help
+         Enable two level sched domains hierarchy.
+         Say N if unsure.
+
 config HAVE_DEC_LOCK
 	bool
 	depends on SMP
include/linux/sched.h +2 -2
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/include/linux/sched.h linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/include/linux/sched.h
@@ -623,7 +623,7 @@ struct sched_domain {
 	.nr_balance_failed	= 0,			\
 }
 
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 /* Common values for NUMA nodes */
 #define SD_NODE_INIT (struct sched_domain) {		\
 	.span			= CPU_MASK_NONE,	\
@@ -656,7 +656,7 @@ static inline int set_cpus_allowed(task_
 
 extern unsigned long long sched_clock(void);
 
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 extern void sched_balance_exec(void);
 #else
 #define sched_balance_exec()   {}
kernel/sched.c +6 -6
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-rc3-mm4/kernel/sched.c linux-2.6.5-rc3-mm4-x86_64_arch_sched_domain/kernel/sched.c
@@ -42,7 +42,7 @@
 #include <linux/percpu.h>
 #include <linux/kthread.h>
 
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 #define cpu_to_node_mask(cpu) node_to_cpumask(cpu_to_node(cpu))
 #else
 #define cpu_to_node_mask(cpu) (cpu_online_map)
@@ -1142,7 +1142,7 @@ enum idle_type
 };
 
 #ifdef CONFIG_SMP
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 /*
  * If dest_cpu is allowed for this process, migrate the task to it.
  * This is accomplished by forcing the cpu_allowed mask to only
@@ -1241,7 +1241,7 @@ void sched_balance_exec(void)
 out:
 	put_cpu();
 }
-#endif /* CONFIG_NUMA */
+#endif /* CONFIG_SCHED_NUMA */
 
 /*
  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
@@ -3461,7 +3461,7 @@ extern void __init arch_init_sched_domai
 #else
 static struct sched_group sched_group_cpus[NR_CPUS];
 static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 static struct sched_group sched_group_nodes[MAX_NUMNODES];
 static DEFINE_PER_CPU(struct sched_domain, node_domains);
 static void __init arch_init_sched_domains(void)
@@ -3532,7 +3532,7 @@ static void __init arch_init_sched_domai
 	}
 }
 
-#else /* !CONFIG_NUMA */
+#else /* !CONFIG_SCHED_NUMA */
 static void __init arch_init_sched_domains(void)
 {
 	int i;
@@ -3570,7 +3570,7 @@ static void __init arch_init_sched_domai
 	}
 }
 
-#endif /* CONFIG_NUMA */
+#endif /* CONFIG_SCHED_NUMA */
 #endif /* ARCH_HAS_SCHED_DOMAIN */
 
 #define SCHED_DOMAIN_DEBUG
#1 lore raw
Quoted 11 lines · Darren Hart

The current default implementations of arch_init_sched_domains constructs either a flat or two level topolology. The two level topology is built if CONFIG_NUMA is set. It seems that CONFIG_NUMA is not the appropriate flag to use for constructing a two level topology since some architectures which define CONFIG_NUMA would be better served with a flat topology. x86_64 for example will construct a two level topology with one CPU per node, causing performance problems because balancing within nodes is pointless and balancing across nodes doesn't occur as often.

This is correct, although I don't know why there would be performance problems. The rebalance in the degenerate node-local domain should be basically unmeasurable. It would be nice to get rid of it at some time. I have code to prune off degenerate domains, which I will submit soonish.

The NUMA rebalance should occur more often than the old numasched did, but perhaps with some recent Altix-centric changes to the generic setup, this is no longer the case.

The STREAM performance problem is due mainly to the more conservative nature of balancing, which is otherwise a good thing. I think we can fix this in the short term by having x86_64 balance between nodes more often. In the long term, we can merge Ingo's balance on clone stuff, and the interested people can play with that.

Quoted 9 lines

This patch introduces a new CONFIG_SCHED_NUMA flag and uses it to decide between a flat or two level topology of sched_domains. The patch is minimally invasive as it primarily modifies Kconfig files and sets the appropriate default (off for x86_64, on for everything that used to export CONFIG_NUMA) and should only change the sched_domains topology constructed on x86_64 systems. I have verified this on a 4 node x86 NUMAQ, but need someone to test x86_64.

I guess I can't see a big problem with this, other than more complexity. In the long run, we should obviously have the arch code set up optimal domains depending on the machine and config.

Quoted 10 lines

This patch is intended as a quick fix for the x86_64 problem, and doesn't solve the problem of how to build generic sched domain topologies. We can certainly conceive of various topologies for x86 systems, so even arch specific topologies may not be sufficient. Would sub-arch (ie NUMAQ) be the right way to handle different topologies, or will we be able to autodiscover the appropriate topology? I will be looking into this more, but thought some might benefit from an immediate x86_64 fix. I am very interested in hearing your ideas on this.

SGI want to do sub arch domains so they can do specific things with their systems. I don't really care what the arch code does with them, but it would be wise to only specialise it when there is a genuine need. I'm glad you'll be looking into it, thanks.

Nick

#2 lore raw

Can SLIT/SRAT be used here to define topology for the generic case?

SRAT is being used by i386 to build zonelists, but not for the scheduler - any good reason why?

--Shai

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Nick Piggin
Sent: Thursday, April 08, 2004 16:42
To: Darren Hart
Cc: lkml; [email protected]; Martin J Bligh; Rick Lindsley; [email protected]; Ingo Molnar
Subject: Re: 2.6.5-rc3-mm4 x86_64 sched domains patch

Quoted 11 lines · Darren Hart

The current default implementations of arch_init_sched_domains constructs either a flat or two level topolology. The two level topology is built if CONFIG_NUMA is set. It seems that CONFIG_NUMA is not the appropriate flag to use for constructing a two level topology since some architectures which define CONFIG_NUMA would be better served with a flat topology. x86_64 for example will construct a two level topology with one CPU per node, causing performance problems because balancing within nodes is pointless and balancing across nodes doesn't occur as often.

This is correct, although I don't know why there would be performance problems. The rebalance in the degenerate node-local domain should be basically unmeasurable. It would be nice to get rid of it at some time. I have code to prune off degenerate domains, which I will submit soonish.

The NUMA rebalance should occur more often than the old numasched did, but perhaps with some recent Altix-centric changes to the generic setup, this is no longer the case.

The STREAM performance problem is due mainly to the more conservative nature of balancing, which is otherwise a good thing. I think we can fix this in the short term by having x86_64 balance between nodes more often. In the long term, we can merge Ingo's balance on clone stuff, and the interested people can play with that.

Quoted 9 lines

This patch introduces a new CONFIG_SCHED_NUMA flag and uses it to decide between a flat or two level topology of sched_domains. The patch is minimally invasive as it primarily modifies Kconfig files and sets the appropriate default (off for x86_64, on for everything that used to export CONFIG_NUMA) and should only change the sched_domains topology constructed on x86_64 systems. I have verified this on a 4 node x86 NUMAQ, but need someone to test x86_64.

I guess I can't see a big problem with this, other than more complexity. In the long run, we should obviously have the arch code set up optimal domains depending on the machine and config.

Quoted 10 lines

This patch is intended as a quick fix for the x86_64 problem, and doesn't solve the problem of how to build generic sched domain topologies. We can certainly conceive of various topologies for x86 systems, so even arch specific topologies may not be sufficient. Would sub-arch (ie NUMAQ) be the right way to handle different topologies, or will we be able to autodiscover the appropriate topology? I will be looking into this more, but thought some might benefit from an immediate x86_64 fix. I am very interested in hearing your ideas on this.

SGI want to do sub arch domains so they can do specific things with their systems. I don't really care what the arch code does with them, but it would be wise to only specialise it when there is a genuine need. I'm glad you'll be looking into it, thanks.

Nick

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#3 lore raw

Can SLIT/SRAT be used here to define topology for the generic case?

SRAT is being used by i386 to build zonelists, but not for the scheduler - any good reason why?

I can think of some possible reasons, but I'm not familiar with SLIT/SRAT ... can you describe it for me?

Rick

#4 lore raw
In reply to [email protected]
Quoted 4 lines

Can SLIT/SRAT be used here to define topology for the generic case?

SRAT is being used by i386 to build zonelists, but not for the scheduler - any good reason why?

Because it's not generic to all machines.

M.

#5 lore raw
In reply to Darren Hart
Quoted 2 lines · On Thu, 08 Apr 2004 16:22:09 -0700 Darren Hart <[email protected]> wrote:

This patch is intended as a quick fix for the x86_64 problem, and

Ingo's latest tweaks seemed to already cure STREAM, but some more tuning is probably a good idea agreed.

Quoted 7 lines

doesn't solve the problem of how to build generic sched domain topologies. We can certainly conceive of various topologies for x86 systems, so even arch specific topologies may not be sufficient. Would sub-arch (ie NUMAQ) be the right way to handle different topologies, or will we be able to autodiscover the appropriate topology? I will be looking into this more, but thought some might benefit from an immediate x86_64 fix. I am very interested in hearing your ideas on this.

The patch doesn't apply against 2.6.5-mm5 anymore. Can you generate a new patch? I will test it then.

Also it will need merging with the patch that adds SMT support for IA32e machines on x86-64.

-Andi

#6 lore raw
Quoted 11 lines · Andi Kleen
Quoted 1 line · On Thu, 08 Apr 2004 16:22:09 -0700 Darren Hart <[email protected]> wrote:

This patch is intended as a quick fix for the x86_64 problem, and

Ingo's latest tweaks seemed to already cure STREAM, but some more tuning is probably a good idea agreed.

Where is STREAM versus other kernels? You said you got best performance on a custom 2.4 kernel. Do we match that?

How is your performance for other things? I recall you may have told me about some other (smaller) issues you were seeing?

#7 lore raw
Quoted 9 lines · On Thu, 15 Apr 2004 00:14:19 +1000 Nick Piggin <[email protected]> wrote:
Quoted 8 lines · Andi Kleen
Quoted 1 line · On Thu, 08 Apr 2004 16:22:09 -0700 Darren Hart <[email protected]> wrote:

This patch is intended as a quick fix for the x86_64 problem, and

Quoted 7 lines
Quoted 3 lines

Ingo's latest tweaks seemed to already cure STREAM, but some more tuning is probably a good idea agreed.

Where is STREAM versus other kernels? You said you got best performance on a custom 2.4 kernel. Do we match that?

Differences were below the measurement error, so I consider it fixed.

Quoted 4 lines

How is your performance for other things? I recall you may have told me about some other (smaller) issues you were seeing?

I haven't tested much yet. I can compare kernel compilations later.

Also I'm still somewhat hoping that the IBM benchmark team will take a stab at it - they are much better than me at running many tests.

-Andi

#8 lore raw
Quoted 10 lines · Andi Kleen
Quoted 3 lines · On Thu, 15 Apr 2004 00:14:19 +1000 Nick Piggin <[email protected]> wrote:

Where is STREAM versus other kernels? You said you got best performance on a custom 2.4 kernel. Do we match that?

Differences were below the measurement error, so I consider it fixed.

great.

Quoted 8 lines
Quoted 4 lines

How is your performance for other things? I recall you may have told me about some other (smaller) issues you were seeing?

I haven't tested much yet. I can compare kernel compilations later.

That would be good. I don't expect you to do all the work, but Opteron being a non traditional NUMA, and me doing most of my testing on an old NUMAQ makes them quite important.

Even if you just got some results for a couple of random benchmarks would be great.

Quoted 2 lines

Also I'm still somewhat hoping that the IBM benchmark team will take a stab at it - they are much better than me at running many tests.

Well we've survived OSDL's STP tests as far as I know. A couple of regressions were found and fixed there, so that was good.

#9 lore raw
In reply to Andi Kleen
Quoted 9 lines · Andi Kleen
Quoted 1 line · On Thu, 08 Apr 2004 16:22:09 -0700 Darren Hart <[email protected]> wrote:

This patch is intended as a quick fix for the x86_64 problem, and

Ingo's latest tweaks seemed to already cure STREAM, but some more tuning is probably a good idea agreed.
...
The patch doesn't apply against 2.6.5-mm5 anymore. Can you generate a new patch? I will test it then.

Find below the patch updated for akpm's 2.6.5-mm5-1.bz2 patch. As with the previous patch I verified it works properly on a 4 node, 16 CPU NUMA-Q. Please test both CONFIG_SCHED_NUMA=n (the improved case, default) and CONFIG_SCHED_NUMA=y (pre-patch equivalent) on x86_64, and thanks!

Quoted 3 lines

Also it will need merging with the patch that adds SMT support for IA32e machines on x86-64.

Where is this patch?

-- Darren

arch/alpha/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/alpha/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/alpha/Kconfig
@@ -519,6 +519,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server machines.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 # LARGE_VMALLOC is racy, if you *really* need it then fix it first
 config ALPHA_LARGE_VMALLOC
 	bool
arch/i386/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/i386/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/i386/Kconfig
@@ -724,6 +724,14 @@ config NUMA
 	default n if X86_PC
 	default y if (X86_NUMAQ || X86_SUMMIT)
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 # Need comments to help the hapless user trying to turn on NUMA support
 comment "NUMA (NUMA-Q) requires SMP, 64GB highmem support"
 	depends on X86_NUMAQ && (!HIGHMEM64G || !SMP)
arch/ia64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/ia64/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/ia64/Kconfig
@@ -172,6 +172,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server systems.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config VIRTUAL_MEM_MAP
 	bool "Virtual mem map"
 	default y if !IA64_HP_SIM
arch/mips/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/mips/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/mips/Kconfig
@@ -337,6 +337,14 @@ config NUMA
 	  Access).  This option is for configuring high-end multiprocessor
 	  server machines.  If in doubt, say N.
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config MAPPED_KERNEL
 	bool "Mapped kernel support"
 	depends on SGI_IP27
arch/ppc64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/ppc64/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/ppc64/Kconfig
@@ -173,6 +173,14 @@ config NUMA
 	bool "NUMA support"
 	depends on DISCONTIGMEM
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default y
+       help
+         Enable two level sched domains hierarchy.
+         Say Y if unsure.
+
 config SCHED_SMT
 	bool "SMT (Hyperthreading) scheduler support"
 	depends on SMP
arch/x86_64/Kconfig +8 -0
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/arch/x86_64/Kconfig linux-2.6.5-mm5-x86_64_arch_sched_domain/arch/x86_64/Kconfig
@@ -261,6 +261,14 @@ config NUMA
        depends on K8_NUMA
        default y
 
+config SCHED_NUMA
+       bool "Two level sched domains"
+       depends on NUMA
+       default n
+       help
+         Enable two level sched domains hierarchy.
+         Say N if unsure.
+
 config HAVE_DEC_LOCK
 	bool
 	depends on SMP
include/linux/sched.h +1 -1
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/include/linux/sched.h linux-2.6.5-mm5-x86_64_arch_sched_domain/include/linux/sched.h
@@ -670,7 +670,7 @@ struct sched_domain {
 	.nr_balance_failed	= 0,			\
 }
 
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 /* Common values for NUMA nodes */
 #define SD_NODE_INIT (struct sched_domain) {		\
 	.span			= CPU_MASK_NONE,	\
kernel/sched.c +4 -4
diff -aurpN -X /home/dvhart/.diff.exclude linux-2.6.5-mm5/kernel/sched.c linux-2.6.5-mm5-x86_64_arch_sched_domain/kernel/sched.c
@@ -45,7 +45,7 @@
 #include <linux/seq_file.h>
 #include <linux/times.h>
 
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 #define cpu_to_node_mask(cpu) node_to_cpumask(cpu_to_node(cpu))
 #else
 #define cpu_to_node_mask(cpu) (cpu_online_map)
@@ -3735,7 +3735,7 @@ extern void __init arch_init_sched_domai
 #else
 static struct sched_group sched_group_cpus[NR_CPUS];
 static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
-#ifdef CONFIG_NUMA
+#ifdef CONFIG_SCHED_NUMA
 static struct sched_group sched_group_nodes[MAX_NUMNODES];
 static DEFINE_PER_CPU(struct sched_domain, node_domains);
 static void __init arch_init_sched_domains(void)
@@ -3806,7 +3806,7 @@ static void __init arch_init_sched_domai
 	}
 }
 
-#else /* !CONFIG_NUMA */
+#else /* !CONFIG_SCHED_NUMA */
 static void __init arch_init_sched_domains(void)
 {
 	int i;
@@ -3845,7 +3845,7 @@ static void __init arch_init_sched_domai
 	}
 }
 
-#endif /* CONFIG_NUMA */
+#endif /* CONFIG_SCHED_NUMA */
 #endif /* ARCH_HAS_SCHED_DOMAIN */
 
 #define SCHED_DOMAIN_DEBUG