Skip to contentSkip to conversation

x86/mce: add CMCI support for Centaur CPUs

PATCH David Wang 1 participant 1 message Jun 4, 2018

lore
Share & tools
lore

David Wang

author
#0 lore raw
Quoted 61 lines

-----Original Mail-----
Sender: Borislav Petkov [mailto:[email protected]]
Time: 2018年6月1日 17:38
Receiver: David Wang <[email protected]>
CC: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Topic : Re: [PATCH] x86/mce: add CMCI support for Centaur CPUs

Quoted 1 line · David Wang

Newer Centaur support CMCI mechanism, which is compatible with INTEL

CMCI.

Quoted 14 lines
---
 arch/x86/Kconfig                 | 12 ++++++++++++
 arch/x86/kernel/cpu/mcheck/mce.c |  6 ++++++
 2 files changed, 18 insertions(+)
arch/x86/Kconfig index renamed from arch/x86/Kconfig +0 -0

dda87a3..1adff5f 100644

arch/x86/Kconfig +0 -0
@@ -1130,6 +1130,18 @@ config X86_MCE_AMD
 	   Additional support for AMD specific MCE features such as
 	   the DRAM Error Threshold.

arch/x86/Kconfig config X86_MCE_AMD · line 1141

Quoted 10 lines
arch/x86/Kconfig +9 -0
 
+config X86_MCE_CENTAUR
+	def_bool y
+	prompt "CENTAUR MCE features"
+	depends on CPU_SUP_CENTAUR && X86_MCE_INTEL
+	help
+	   Additional support for Centaur specific MCE features such as
+	   MCE broadcasting and CMCI support.
+	   New Centaur CPU support MCE broadcasting.
+	   New Centaur CPU support CMCI which is fully compliant with Intel

CMCI.

arch/x86/kernel/cpu/mcheck/mce.c static void __mcheck_cpu_init_early(struct

Quoted 12 lines
arch/x86/Kconfig +3 -0
+
+	   If unsure, say N here.
+
 config X86_ANCIENT_MCE
 	bool "Support for old Pentium 5 / WinChip machine checks"
 	depends on X86_32 && X86_MCE
arch/x86/kernel/cpu/mcheck/mce.c +0 -0
index cd76380..2ebafc7 100644
@@ -1727,6 +1727,7 @@ static void __mcheck_cpu_init_early(struct

cpuinfo_x86 *c)

arch/x86/kernel/cpu/mcheck/mce.c static void mce_centaur_feature_init(struct

Quoted 7 lines
arch/x86/kernel/cpu/mcheck/mce.c +1 -0
 	}
 }
 
+#ifdef CONFIG_X86_MCE_CENTAUR
 static void mce_centaur_feature_init(struct cpuinfo_x86 *c)  {
 	struct mca_config *cfg = &mca_cfg;
@@ -1740,7 +1741,12 @@ static void mce_centaur_feature_init(struct

cpuinfo_x86 *c)

Quoted 9 lines

arch/x86/kernel/cpu/mcheck/mce.c static void mce_centaur_feature_init(struct · line 1745

Quoted 5 lines
arch/x86/kernel/cpu/mcheck/mce.c +2 -0
 		if (cfg->monarch_timeout < 0)
 			cfg->monarch_timeout = USEC_PER_SEC;
 	}
+	mce_intel_feature_init(c);
+	mce_adjust_timer = cmci_intel_adjust_timer;

So far so good but above says "New Centaur CPU[s]" but you're calling mce_intel_feature_init() unconditionally here, for *all* centaur CPUs. Ditto for setting the CMCI handler.

There is a check for CMCI support or not In cmci_supported() function which will be called by mce_intel_feature_init.

return !!(cap & MCG_CMCI_P);
So, I didn't add Family/Model/Stepping check.

But, I am sorry to add another patch just as following:

@@ -85,7 +85,8 @@ static int cmci_supported(int *banks)
* initialization is vendor keyed and this
* makes sure none of the backdoors are entered otherwise. */
- if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+ if ((boot_cpu_data.x86_vendor != X86_VENDOR_INTEL &&
+ boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR))

                return 0;
        if (!boot_cpu_has(X86_FEATURE_APIC) || lapic_get_maxlvt() < 6)
                return 0;

	I will send patch v4 later to include this patch.
	Thank you.
Quoted 23 lines

Also mce_intel_feature_init() does more things than init CMCI so I think you wanna limit that to only intel_init_cmci(). IOW, something like the hunk below.

And frankly I'm not crazy about adding centaur code to mce_intel.c but since it is copying functionality, I think we should copy the sw support in the kernel too instead of adding a mce_centaur.c duplicate. For now at least...

Thx.

---

arch/x86/kernel/cpu/mcheck/mce_intel.c +2 -1
index d05be307d081..77d8a9b996a6 100644
@@ -506,8 +506,15 @@ static void intel_ppin_init(struct cpuinfo_x86 *c)
 
 void mce_intel_feature_init(struct cpuinfo_x86 *c)  {
-	intel_init_thermal(c);
 	intel_init_cmci();
+
+	/*

arch/x86/kernel/cpu/mcheck/mce_intel.c static void intel_ppin_init(struct cpuinfo_x86 *c) · line 520

Quoted 10 lines
arch/x86/kernel/cpu/mcheck/mce_intel.c +6 -0
+	 * Some Centaur variants support CMCI.
+	 */
+	if (c->x86_vendor == X86_VENDOR_CENTAUR)
+		return;
+
+	intel_init_thermal(c);
 	intel_init_lmce();
 	intel_ppin_init(c);
 }
 

OK. I will send patch V4 to fix this problem, just like:

void mce_intel_feature_init(struct cpuinfo_x86 *c)
{

-       intel_init_thermal(c);
-       intel_init_cmci();
-       intel_init_lmce();
-       intel_ppin_init(c);
+
+       switch (c->x86_vendor) {
+       case X86_VENDOR_INTEL:
+               intel_init_thermal(c);
+               intel_init_cmci();
+               intel_init_lmce();
+               intel_ppin_init(c);
+               break;

+ case X86_VENDOR_CENTAUR:
+ intel_init_cmci();
+ break;
+ default:
+ break;
+ }
}
Thank you.

Quoted 6 lines

--
Regards/Gruss,
Boris.

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