Skip to contentSkip to conversation

x86: devicetree: enable multiprocessing in DT

PATCH Superseded Ivan Gorinov 2 participants 2 messages Mar 1, 2018 – Mar 2, 2018 x86

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 2/3 x86: devicetree: enable multiprocessing in DT

Adding code to register the processors described in Device Tree. APIC ID is specified in 'intel-apic_id' propery as used in U-Boot. First address specified in 'reg' is used as default APIC ID.

---
 arch/x86/kernel/devicetree.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)
arch/x86/kernel/devicetree.c +24 -2
index 44189ee..ef1cd85 100644
@@ -130,6 +130,29 @@ static void __init dtb_setup_hpet(void)
 #endif
 }
 
+static void __init dtb_cpu_setup(void)
+{
+	struct device_node *dn;
+	struct resource r;
+	const void *prop;
+	int apic_id, version;
+	int ret;
+
+	version = GET_APIC_VERSION(apic_read(APIC_LVR));
+	for_each_node_by_type(dn, "cpu") {
+		prop = of_get_property(dn, "intel,apic-id", NULL);
+		if (prop) {
+			apic_id = be32_to_cpup(prop);
+		} else {
+			ret = of_address_to_resource(dn, 0, &r);
+			if (WARN_ON_ONCE(ret))
+				continue;
+			apic_id = r.start;
+		}
+		generic_processor_info(apic_id, version);
+	}
+}
+
 static void __init dtb_lapic_setup(void)
 {
 #ifdef CONFIG_X86_LOCAL_APIC
@@ -153,8 +176,6 @@ static void __init dtb_lapic_setup(void)
 	smp_found_config = 1;
 	pic_mode = 1;
 	register_lapic_address(r.start);
-	generic_processor_info(boot_cpu_physical_apicid,
-			       GET_APIC_VERSION(apic_read(APIC_LVR)));
 #endif
 }
 
@@ -256,6 +277,7 @@ static void __init dtb_ioapic_setup(void) {}
 static void __init dtb_apic_setup(void)
 {
 	dtb_lapic_setup();
+	dtb_cpu_setup();
 	dtb_ioapic_setup();
 }
 
Signature · 1 line
2.7.4
#1 lore raw

On Thu, 1 Mar 2018, Ivan Gorinov wrote:

The patch order is wrong. Fixes first and then features simply because you cannot test the feature patch standalone as the fix (3/3) is missing ....

Quoted 1 line

Adding code to register the processors described in Device Tree.

'Adding code...' is a pointless filler phrase.

Quoted 1 line

APIC ID is specified in 'intel-apic_id' propery as used in U-Boot.

s/propery/property/

Also this has nothing to do with U-Boot. The DT properties are described in the DT bindings and not specified in U-Boot.

Where is the matching devicetree binding documented? This should be a separate patch and needs to go into Documentation/devicetree/x86/

Please Cc the DT maintainers as we need their ack for that.

Quoted 1 line

First address specified in 'reg' is used as default APIC ID.

Changelogs should describe the context/problem and the approach how this is fixed or made working. Something like this:

The current x86 device tree implementation does not support SMP.

Use the new DT bindings for describing CPUs and their APIC resources.

Hmm?

Quoted 22 lines
---
 arch/x86/kernel/devicetree.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)
arch/x86/kernel/devicetree.c +9 -0
index 44189ee..ef1cd85 100644
@@ -130,6 +130,29 @@ static void __init dtb_setup_hpet(void)
 #endif
 }
 
+static void __init dtb_cpu_setup(void)
+{
+	struct device_node *dn;
+	struct resource r;
+	const void *prop;
+	int apic_id, version;
+	int ret;
+
+	version = GET_APIC_VERSION(apic_read(APIC_LVR));

arch/x86/kernel/devicetree.c static void __init dtb_setup_hpet(void) · line 151

Quoted 10 lines
arch/x86/kernel/devicetree.c +10 -0
+	for_each_node_by_type(dn, "cpu") {
+		prop = of_get_property(dn, "intel,apic-id", NULL);
+		if (prop) {
+			apic_id = be32_to_cpup(prop);
+		} else {
+			ret = of_address_to_resource(dn, 0, &r);
+			if (WARN_ON_ONCE(ret))
+				continue;
+			apic_id = r.start;
+		}

What kind of logic is this? If a CPU node does not have apic id property then it's invalid. That else clause is just voodoo programming at least without a proper comment explaining it.

Thanks,

tglx