Skip to contentSkip to conversation

i383 numa: fix numaq/summit apicid conflict

PATCH In review keith mannthey 2 participants 3 messages Oct 4, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i383 numa: fix numaq/summit apicid conflict

From: Keith Mannthey <[email protected]>

This patch allows numaq to properly align cpus to their given node
during boot. Pass logical apicid to apicid_to_node and allow the summit sub-arch to use physical apicid (hard_smp_processor_id()).
Tested against numaq and summit based systems with no issues. against
2.6.18-git18.

---
 arch/i386/kernel/smpboot.c               |    2 +-
 include/asm-i386/mach-summit/mach_apic.h |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/i386/kernel/smpboot.c +1 -1
diff -urN linux-2.6.18/arch/i386/kernel/smpboot.c linux-2.6.18-git18/arch/i386/kernel/smpboot.c
@@ -648,7 +648,7 @@
 {
 	int cpu = smp_processor_id();
 	int apicid = logical_smp_processor_id();
-	int node = apicid_to_node(hard_smp_processor_id());
+	int node = apicid_to_node(apicid);
 
 	if (!node_online(node))
 		node = first_online_node;
include/asm-i386/mach-summit/mach_apic.h +1 -1
diff -urN linux-2.6.18/include/asm-i386/mach-summit/mach_apic.h linux-2.6.18-git18/include/asm-i386/mach-summit/mach_apic.h
@@ -88,7 +88,7 @@
 
 static inline int apicid_to_node(int logical_apicid)
 {
-	return apicid_2_node[logical_apicid];
+	return apicid_2_node[hard_smp_processor_id()];
 }
 
 /* Mapping from cpu number to logical apicid */
#1 lore raw
Quoted 30 lines · keith mannthey

From: Keith Mannthey <[email protected]>

This patch allows numaq to properly align cpus to their given node
during boot. Pass logical apicid to apicid_to_node and allow the summit sub-arch to use physical apicid (hard_smp_processor_id()).
Tested against numaq and summit based systems with no issues. against
2.6.18-git18.

---
 arch/i386/kernel/smpboot.c               |    2 +-
 include/asm-i386/mach-summit/mach_apic.h |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff -urN linux-2.6.18/arch/i386/kernel/smpboot.c linux-2.6.18-git18/arch/i386/kernel/smpboot.c

arch/i386/kernel/smpboot.c +1 -1
@@ -648,7 +648,7 @@
 {
 	int cpu = smp_processor_id();
 	int apicid = logical_smp_processor_id();
-	int node = apicid_to_node(hard_smp_processor_id());
+	int node = apicid_to_node(apicid);
 
 	if (!node_online(node))
 		node = first_online_node;

diff -urN linux-2.6.18/include/asm-i386/mach-summit/mach_apic.h linux-2.6.18-git18/include/asm-i386/mach-summit/mach_apic.h

include/asm-i386/mach-summit/mach_apic.h line 94

Quoted 8 lines
include/asm-i386/mach-summit/mach_apic.h +1 -1
@@ -88,7 +88,7 @@
 
 static inline int apicid_to_node(int logical_apicid)
 {
-	return apicid_2_node[logical_apicid];
+	return apicid_2_node[hard_smp_processor_id()];
 }
 
 /* Mapping from cpu number to logical apicid */

My worry here is that we might have users who are calling this about other cpus. As you have effectivly ignored the parameter on summit here.

Can we not just map the hard_smp_processor_id to its logical apic id when filling in the apicid_2_node array on summit? Such that it really does have the logical id in there?

-apw

#2 lore raw
Quoted 38 lines · Andy Whitcroft
Quoted 37 lines · keith mannthey

From: Keith Mannthey <[email protected]>

This patch allows numaq to properly align cpus to their given node
during boot. Pass logical apicid to apicid_to_node and allow the summit sub-arch to use physical apicid (hard_smp_processor_id()).
Tested against numaq and summit based systems with no issues. against
2.6.18-git18.

---
 arch/i386/kernel/smpboot.c               |    2 +-
 include/asm-i386/mach-summit/mach_apic.h |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff -urN linux-2.6.18/arch/i386/kernel/smpboot.c linux-2.6.18-git18/arch/i386/kernel/smpboot.c

arch/i386/kernel/smpboot.c +1 -1
@@ -648,7 +648,7 @@
 {
 	int cpu = smp_processor_id();
 	int apicid = logical_smp_processor_id();
-	int node = apicid_to_node(hard_smp_processor_id());
+	int node = apicid_to_node(apicid);
 
 	if (!node_online(node))
 		node = first_online_node;

diff -urN linux-2.6.18/include/asm-i386/mach-summit/mach_apic.h linux-2.6.18-git18/include/asm-i386/mach-summit/mach_apic.h

include/asm-i386/mach-summit/mach_apic.h +1 -1
@@ -88,7 +88,7 @@
 
 static inline int apicid_to_node(int logical_apicid)
 {
-	return apicid_2_node[logical_apicid];
+	return apicid_2_node[hard_smp_processor_id()];
 }
 
Quoted 4 lines

include/asm-i386/mach-summit/mach_apic.h line 94

Quoted 1 line
include/asm-i386/mach-summit/mach_apic.h +0 -0
 /* Mapping from cpu number to logical apicid */

My worry here is that we might have users who are calling this about other cpus. As you have effectivly ignored the parameter on summit here.

I sort of thought about riping out the parameter to apicid_to_node and
letting summit and numaq fill in whatever they need. apicid_to_node should only be used in map_cpu_to_logical_apicid. After this the kernel has a nice cpu to node mapping and a great interface for it.

Quoted 3 lines

Can we not just map the hard_smp_processor_id to its logical apic id when filling in the apicid_2_node array on summit? Such that it really does have the logical id in there?

The apicid_2_node array comes right from the SRAT table.... Without
constructing yet another table I don't seem to have a good way to translate logical to physical or vice versa.

Maybe making the callers of apicid_to_node pass both logical and
physical apicid will relive the worry?

Sure we can create another bootime mapping but it seems overkill.
Really only summit and numaq do anything useful with apicid_to_node and i386 numa for that matter and I don't expect any new numa sub arches....

What do you think?

Thanks,
Keith