Skip to contentSkip to conversation

x86: Fix sibling map with NumaChip

PATCH Superseded Daniel J Blueman 3 participants 3 messages Mar 4, 2015 x86

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Daniel J Blueman

author
#0 lore raw

PATCH v3 1/1 x86: Fix sibling map with NumaChip

On NumaChip systems, the physical processor ID assignment wasn't accounting for the number of nodes in AMD multi-module processors, giving an incorrect sibling map:

$ cd /sys/devices/system/cpu/cpu29/topology
$ grep . *
core_id:5
core_siblings:00000000,ff000000
core_siblings_list:24-31
physical_package_id:3
thread_siblings:00000000,30000000
thread_siblings_list:28-29

After fixing:

core_id:5
core_siblings:00000000,ffff0000
core_siblings_list:16-31
physical_package_id:1
thread_siblings:00000000,30000000
thread_siblings_list:28-29

v2: Fix to check for MSR availability before use, as per Boris's feedback
v3: Test against boot cpu features to correct behaviour on larger systems
with global IO

Candidate for stable.

Signed-off-by Daniel J Blueman <[email protected]>
---
 arch/x86/kernel/apic/apic_numachip.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)
arch/x86/kernel/apic/apic_numachip.c +16 -6
index c2fd21f..c1f4f36 100644
@@ -37,10 +37,12 @@ static const struct apic apic_numachip;
 static unsigned int get_apic_id(unsigned long x)
 {
 	unsigned long value;
-	unsigned int id;
+	unsigned int id = (x >> 24) & 0xff;
 
-	rdmsrl(MSR_FAM10H_NODE_ID, value);
-	id = ((x >> 24) & 0xffU) | ((value << 2) & 0xff00U);
+	if (boot_cpu_has(X86_FEATURE_NODEID_MSR)) {
+		rdmsrl(MSR_FAM10H_NODE_ID, value);
+		id |= (value << 2) & 0xff00;
+	}
 
 	return id;
 }

@@ -155,10 +157,18 @@ static int __init numachip_probe(void)
 
 static void fixup_cpu_id(struct cpuinfo_x86 *c, int node)
 {
-	if (c->phys_proc_id != node) {
-		c->phys_proc_id = node;
-		per_cpu(cpu_llc_id, smp_processor_id()) = node;
+	u64 val;
+	u32 nodes = 1;
+
+	this_cpu_write(cpu_llc_id, node);
+
+	/* Account for nodes per socket in multi-core-module processors */
+	if (cpu_has(c, X86_FEATURE_NODEID_MSR)) {
+		rdmsrl(MSR_FAM10H_NODE_ID, val);
+		nodes = ((val >> 3) & 7) + 1;
 	}
+
+	c->phys_proc_id = node / nodes;
 }
 
 static int __init numachip_system_init(void)
Signature · 1 line
1.9.1
#1 lore raw

arch/x86/kernel/apic/apic_numachip.c static const struct apic apic_numachip; · line 42

Quoted 4 lines · Daniel J Blueman

v3: Test against boot cpu features to correct behaviour on larger systems
with global IO

arch/x86/kernel/apic/apic_numachip.c +1 -0
+	if (boot_cpu_has(X86_FEATURE_NODEID_MSR)) {

Hm, so why doesn't this_cpu_has() work? Supposedly such flags ought to be symmetric on all CPUs in the system?

Thanks,

Ingo

maintainer
#2 lore raw
Quoted 2 lines · Ingo Molnar

Hm, so why doesn't this_cpu_has() work? Supposedly such flags ought to be symmetric on all CPUs in the system?

Actually, static_cpu_has_safe() should be the fastest and the safest to use in all cases.

Signature · 5 lines
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--