[Patch V2 0/3] x86, mce: Local Machine Check Exception (LMCE)

[Patch V2 0/3] x86, mce: Local Machine Check Exception (LMCE)

[Ashok Raj]

Hi Boris

Thanks for the feedback on V1. Almost all of your recommendations are
included in this update.

I haven't got a chance to test on qemu yet, but this patch fixes access to
MSR per your recommandation, so should be fine. I'm in process of making
similar changes to kvm/Qemu that i will send once I have learned to build
test it :-).

Historically machine checks on Intel X86 processors have been broadcast to all
logical processors in the system. Upcoming CPUs will support an opt-in
mechanism to request some machine checks delivered to a single logical
processor experiencing the fault.

For more details see Vol3, Chapter 15, Machine Check Architecture.

Modified to incorporate feedback from Boris on V1 patches.

Ashok Raj (3):
  x86, mce: Add LMCE definitions.
  x86, mce: Add infrastructure required to support LMCE
  x86, mce: Handling LMCE events

 Documentation/x86/x86_64/boot-options.txt |  3 ++
 arch/x86/include/asm/mce.h                | 10 ++++++
 arch/x86/include/uapi/asm/msr-index.h     |  2 ++
 arch/x86/kernel/cpu/mcheck/mce.c          | 35 ++++++++++++++----
 arch/x86/kernel/cpu/mcheck/mce_intel.c    | 60 +++++++++++++++++++++++++++++++
 5 files changed, 104 insertions(+), 6 deletions(-)

-- 
1.9.1

[Patch V2 1/3] x86, mce: Add LMCE definitions.

[Ashok Raj]

Add required definitions to support Local Machine Check Exceptions.

See http://www.intel.com/sdm Volume 3, System Programming Guide, chapter 15
for more information on MSR's and documentation on Local MCE.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
---
 arch/x86/include/asm/mce.h            | 5 +++++
 arch/x86/include/uapi/asm/msr-index.h | 2 ++
 2 files changed, 7 insertions(+)

diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index 1f5a86d..677a408 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -17,11 +17,16 @@
 #define MCG_EXT_CNT(c)		(((c) & MCG_EXT_CNT_MASK) >> MCG_EXT_CNT_SHIFT)
 #define MCG_SER_P		(1ULL<<24)   /* MCA recovery/new status bits */
 #define MCG_ELOG_P		(1ULL<<26)   /* Extended error log supported */
+#define MCG_LMCE_P		(1ULL<<27)   /* Local machine check supported */
 
 /* MCG_STATUS register defines */
 #define MCG_STATUS_RIPV  (1ULL<<0)   /* restart ip valid */
 #define MCG_STATUS_EIPV  (1ULL<<1)   /* ip points to correct instruction */
 #define MCG_STATUS_MCIP  (1ULL<<2)   /* machine check in progress */
+#define MCG_STATUS_LMCES (1ULL<<3)   /* LMCE signaled */
+
+/* MCG_EXT_CTL register defines */
+#define MCG_EXT_CTL_LMCE_EN (1ULL<<0) /* Enable LMCE */
 
 /* MCi_STATUS register defines */
 #define MCI_STATUS_VAL   (1ULL<<63)  /* valid error */
diff --git a/arch/x86/include/uapi/asm/msr-index.h b/arch/x86/include/uapi/asm/msr-index.h
index c469490..32c69d5 100644
--- a/arch/x86/include/uapi/asm/msr-index.h
+++ b/arch/x86/include/uapi/asm/msr-index.h
@@ -56,6 +56,7 @@
 #define MSR_IA32_MCG_CAP		0x00000179
 #define MSR_IA32_MCG_STATUS		0x0000017a
 #define MSR_IA32_MCG_CTL		0x0000017b
+#define MSR_IA32_MCG_EXT_CTL		0x000004d0
 
 #define MSR_OFFCORE_RSP_0		0x000001a6
 #define MSR_OFFCORE_RSP_1		0x000001a7
@@ -379,6 +380,7 @@
 #define FEATURE_CONTROL_LOCKED				(1<<0)
 #define FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX	(1<<1)
 #define FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX	(1<<2)
+#define FEATURE_CONTROL_LMCE				(1<<20)
 
 #define MSR_IA32_APICBASE		0x0000001b
 #define MSR_IA32_APICBASE_BSP		(1<<8)
-- 
1.9.1

[Patch V2 2/3] x86, mce: Add infrastructure required to support LMCE

[Ashok Raj]

Initialization and handling for LMCE
- boot time option to disable LMCE for that boot instance
- Check for capability via IA32_MCG_CAP

Incorporated feedback from Boris from V1

See http://www.intel.com/sdm Volume 3 System Programming Guide, Chapter 15
for more information on MSR's and documentation on Local MCE.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
---
 Documentation/x86/x86_64/boot-options.txt |  3 ++
 arch/x86/include/asm/mce.h                |  5 +++
 arch/x86/kernel/cpu/mcheck/mce.c          |  3 ++
 arch/x86/kernel/cpu/mcheck/mce_intel.c    | 59 +++++++++++++++++++++++++++++++
 4 files changed, 70 insertions(+)

diff --git a/Documentation/x86/x86_64/boot-options.txt b/Documentation/x86/x86_64/boot-options.txt
index 5223479..79edee0 100644
--- a/Documentation/x86/x86_64/boot-options.txt
+++ b/Documentation/x86/x86_64/boot-options.txt
@@ -31,6 +31,9 @@ Machine check
 		(e.g. BIOS or hardware monitoring applications), conflicting
 		with OS's error handling, and you cannot deactivate the agent,
 		then this option will be a help.
+   mce=no_lmce
+		Do not opt-in to Local MCE delivery. Use legacy method
+		to broadcast MCE's.
    mce=bootlog
 		Enable logging of machine checks left over from booting.
 		Disabled by default on AMD because some BIOS leave bogus ones.
diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index 677a408..8ba4d7a 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -109,6 +109,7 @@ struct mce_log {
 struct mca_config {
 	bool dont_log_ce;
 	bool cmci_disabled;
+	bool lmce_disabled;
 	bool ignore_ce;
 	bool disabled;
 	bool ser;
@@ -173,12 +174,16 @@ void cmci_clear(void);
 void cmci_reenable(void);
 void cmci_rediscover(void);
 void cmci_recheck(void);
+void lmce_clear(void);
+void lmce_enable(void);
 #else
 static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
 static inline void cmci_clear(void) {}
 static inline void cmci_reenable(void) {}
 static inline void cmci_rediscover(void) {}
 static inline void cmci_recheck(void) {}
+static inline void lmce_clear(void) {}
+static inline void lmce_enable(void) {}
 #endif
 
 #ifdef CONFIG_X86_MCE_AMD
diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index e535533..d10aada 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -1976,6 +1976,7 @@ void mce_disable_bank(int bank)
 /*
  * mce=off Disables machine check
  * mce=no_cmci Disables CMCI
+ * mce=no_lmce Disables LMCE
  * mce=dont_log_ce Clears corrected events silently, no log created for CEs.
  * mce=ignore_ce Disables polling and CMCI, corrected events are not cleared.
  * mce=TOLERANCELEVEL[,monarchtimeout] (number, see above)
@@ -1999,6 +2000,8 @@ static int __init mcheck_enable(char *str)
 		cfg->disabled = true;
 	else if (!strcmp(str, "no_cmci"))
 		cfg->cmci_disabled = true;
+	else if (!strcmp(str, "no_lmce"))
+		cfg->lmce_disabled = true;
 	else if (!strcmp(str, "dont_log_ce"))
 		cfg->dont_log_ce = true;
 	else if (!strcmp(str, "ignore_ce"))
diff --git a/arch/x86/kernel/cpu/mcheck/mce_intel.c b/arch/x86/kernel/cpu/mcheck/mce_intel.c
index b4a41cf..7d500b6 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_intel.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_intel.c
@@ -91,6 +91,37 @@ static int cmci_supported(int *banks)
 	return !!(cap & MCG_CMCI_P);
 }
 
+static bool lmce_supported(void)
+{
+	u64 cap, feature_control;
+
+	if (mca_cfg.lmce_disabled)
+		return false;
+
+	rdmsrl(MSR_IA32_MCG_CAP, cap);
+	/*
+	 * LMCE depends on recovery support in the processor.
+	 * Hence both MCG_SER_P and MCG_LMCE_P should be present in
+	 * MCG_CAP
+	 */
+	if (!((cap & (MCG_SER_P | MCG_LMCE_P)) == (MCG_SER_P | MCG_LMCE_P)))
+		return false;
+
+	/*
+	 * BIOS should indicate support for LMCE by setting
+	 * bit20 in IA32_FEATURE_CONTROL. without which touching
+	 * MCG_EXT_CTL will generate #GP fault.
+	 */
+	rdmsrl(MSR_IA32_FEATURE_CONTROL, feature_control);
+	if (((feature_control & (FEATURE_CONTROL_LOCKED |
+		FEATURE_CONTROL_LMCE)) == (FEATURE_CONTROL_LOCKED |
+				FEATURE_CONTROL_LMCE)))
+		return true;
+	else
+		return false;
+
+}
+
 bool mce_intel_cmci_poll(void)
 {
 	if (__this_cpu_read(cmci_storm_state) == CMCI_STORM_NONE)
@@ -405,6 +436,34 @@ static void intel_init_cmci(void)
 	cmci_recheck();
 }
 
+void intel_init_lmce(void)
+{
+	u64 val;
+
+	if (!lmce_supported())
+		return;
+
+	rdmsrl(MSR_IA32_MCG_EXT_CTL, val);
+	val |= MCG_EXT_CTL_LMCE_EN;
+	wrmsrl(MSR_IA32_MCG_EXT_CTL, val);
+}
+
+/*
+ * Disable LMCE on this CPU for all banks it owns when it goes down.
+ * This allows other CPUs to claim the banks on rediscovery.
+ */
+void lmce_clear(void)
+{
+	u64 val;
+
+	if (!lmce_supported())
+		return;
+
+	rdmsrl(MSR_IA32_MCG_EXT_CTL, val);
+	val &= ~MCG_EXT_CTL_LMCE_EN;
+	wrmsrl(MSR_IA32_MCG_EXT_CTL, val);
+}
+
 void mce_intel_feature_init(struct cpuinfo_x86 *c)
 {
 	intel_init_thermal(c);
-- 
1.9.1

[Patch V2 3/3] x86, mce: Handling LMCE events

[Ashok Raj]

This patch has handling changes to do_machine_check() to process MCE
signaled as local MCE. Typically only recoverable errors (SRAR) type
error will be Signaled as LMCE. But architecture does not restrict to
only those errors.

When errors are signaled as LMCE, there is no need for the MCE handler to
perform rendezvous with other logical processors unlike earlier processors
that would broadcast machine check errors.

See http://www.intel.com/sdm Volume 3, Chapter 15 for more information
on MSR's and documentation on Local MCE.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
---
 arch/x86/kernel/cpu/mcheck/mce.c       | 32 ++++++++++++++++++++++++++------
 arch/x86/kernel/cpu/mcheck/mce_intel.c |  1 +
 2 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index d10aada..3d71daf 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -1047,6 +1047,7 @@ void do_machine_check(struct pt_regs *regs, long error_code)
 	char *msg = "Unknown";
 	u64 recover_paddr = ~0ull;
 	int flags = MF_ACTION_REQUIRED;
+	int lmce = 0;
 
 	prev_state = ist_enter(regs);
 
@@ -1074,11 +1075,20 @@ void do_machine_check(struct pt_regs *regs, long error_code)
 		kill_it = 1;
 
 	/*
-	 * Go through all the banks in exclusion of the other CPUs.
-	 * This way we don't report duplicated events on shared banks
-	 * because the first one to see it will clear it.
+	 * Check if this MCE is signaled to only this logical processor
 	 */
-	order = mce_start(&no_way_out);
+	if (m.mcgstatus & MCG_STATUS_LMCES)
+		lmce = 1;
+	else {
+		/*
+		 * Go through all the banks in exclusion of the other CPUs.
+		 * This way we don't report duplicated events on shared banks
+		 * because the first one to see it will clear it.
+		 * If this is a Local MCE, then no need to perform rendezvous.
+		 */
+		order = mce_start(&no_way_out);
+	}
+
 	for (i = 0; i < cfg->banks; i++) {
 		__clear_bit(i, toclear);
 		if (!test_bit(i, valid_banks))
@@ -1155,8 +1165,18 @@ void do_machine_check(struct pt_regs *regs, long error_code)
 	 * Do most of the synchronization with other CPUs.
 	 * When there's any problem use only local no_way_out state.
 	 */
-	if (mce_end(order) < 0)
-		no_way_out = worst >= MCE_PANIC_SEVERITY;
+	if (!lmce) {
+		if (mce_end(order) < 0)
+			no_way_out = worst >= MCE_PANIC_SEVERITY;
+	} else {
+		/*
+		 * Local MCE skipped calling mce_reign()
+		 * If we found a fatal error, we need to panic here.
+		 */
+		 if (worst >= MCE_PANIC_SEVERITY && mca_cfg.tolerant < 3)
+			mce_panic("Machine check from unknown source",
+				NULL, NULL);
+	}
 
 	/*
 	 * At insane "tolerant" levels we take no action. Otherwise
diff --git a/arch/x86/kernel/cpu/mcheck/mce_intel.c b/arch/x86/kernel/cpu/mcheck/mce_intel.c
index 7d500b6..47b2a2b 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_intel.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_intel.c
@@ -468,4 +468,5 @@ void mce_intel_feature_init(struct cpuinfo_x86 *c)
 {
 	intel_init_thermal(c);
 	intel_init_cmci();
+	intel_init_lmce();
 }
-- 
1.9.1

Re: [Patch V2 2/3] x86, mce: Add infrastructure required to support LMCE

[Borislav Petkov]

On Tue, Jun 02, 2015 at 10:49:53AM -0700, Ashok Raj wrote:
> Initialization and handling for LMCE
> - boot time option to disable LMCE for that boot instance
> - Check for capability via IA32_MCG_CAP
> 
> Incorporated feedback from Boris from V1
> 
> See http://www.intel.com/sdm Volume 3 System Programming Guide, Chapter 15
> for more information on MSR's and documentation on Local MCE.
> 
> Signed-off-by: Ashok Raj <ashok.raj@intel.com>
> ---
>  Documentation/x86/x86_64/boot-options.txt |  3 ++
>  arch/x86/include/asm/mce.h                |  5 +++
>  arch/x86/kernel/cpu/mcheck/mce.c          |  3 ++
>  arch/x86/kernel/cpu/mcheck/mce_intel.c    | 59 +++++++++++++++++++++++++++++++

I did a bunch of improvements while applying:

---
From: Ashok Raj <ashok.raj@intel.com>
Date: Tue, 2 Jun 2015 10:49:53 -0700
Subject: [PATCH] x86/mce: Add infrastructure to support Local MCE

Initialize and prepare for handling LMCEs. Add a boot-time option to
disable LMCEs.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/1433267394-10943-3-git-send-email-ashok.raj@intel.com
[ Simplify stuff, align statements for better readability, reflow comments; kill
  unused lmce_clear(); save us an MSR write if LMCE already enabled. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
---
 Documentation/x86/x86_64/boot-options.txt |  3 +++
 arch/x86/include/asm/mce.h                |  5 ++++
 arch/x86/kernel/cpu/mcheck/mce.c          |  3 +++
 arch/x86/kernel/cpu/mcheck/mce_intel.c    | 43 +++++++++++++++++++++++++++++++
 4 files changed, 54 insertions(+)

diff --git a/Documentation/x86/x86_64/boot-options.txt b/Documentation/x86/x86_64/boot-options.txt
index 5223479291a2..68ed3114c363 100644
--- a/Documentation/x86/x86_64/boot-options.txt
+++ b/Documentation/x86/x86_64/boot-options.txt
@@ -31,6 +31,9 @@ Machine check
 		(e.g. BIOS or hardware monitoring applications), conflicting
 		with OS's error handling, and you cannot deactivate the agent,
 		then this option will be a help.
+   mce=no_lmce
+		Do not opt-in to Local MCE delivery. Use legacy method
+		to broadcast MCEs.
    mce=bootlog
 		Enable logging of machine checks left over from booting.
 		Disabled by default on AMD because some BIOS leave bogus ones.
diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
index ae2bfb895994..982dfc3679ad 100644
--- a/arch/x86/include/asm/mce.h
+++ b/arch/x86/include/asm/mce.h
@@ -109,6 +109,7 @@ struct mce_log {
 struct mca_config {
 	bool dont_log_ce;
 	bool cmci_disabled;
+	bool lmce_disabled;
 	bool ignore_ce;
 	bool disabled;
 	bool ser;
@@ -184,12 +185,16 @@ void cmci_clear(void);
 void cmci_reenable(void);
 void cmci_rediscover(void);
 void cmci_recheck(void);
+void lmce_clear(void);
+void lmce_enable(void);
 #else
 static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
 static inline void cmci_clear(void) {}
 static inline void cmci_reenable(void) {}
 static inline void cmci_rediscover(void) {}
 static inline void cmci_recheck(void) {}
+static inline void lmce_clear(void) {}
+static inline void lmce_enable(void) {}
 #endif
 
 #ifdef CONFIG_X86_MCE_AMD
diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index 0cbcd3183acf..c8c6577b4ada 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -1982,6 +1982,7 @@ void mce_disable_bank(int bank)
 /*
  * mce=off Disables machine check
  * mce=no_cmci Disables CMCI
+ * mce=no_lmce Disables LMCE
  * mce=dont_log_ce Clears corrected events silently, no log created for CEs.
  * mce=ignore_ce Disables polling and CMCI, corrected events are not cleared.
  * mce=TOLERANCELEVEL[,monarchtimeout] (number, see above)
@@ -2005,6 +2006,8 @@ static int __init mcheck_enable(char *str)
 		cfg->disabled = true;
 	else if (!strcmp(str, "no_cmci"))
 		cfg->cmci_disabled = true;
+	else if (!strcmp(str, "no_lmce"))
+		cfg->lmce_disabled = true;
 	else if (!strcmp(str, "dont_log_ce"))
 		cfg->dont_log_ce = true;
 	else if (!strcmp(str, "ignore_ce"))
diff --git a/arch/x86/kernel/cpu/mcheck/mce_intel.c b/arch/x86/kernel/cpu/mcheck/mce_intel.c
index b4a41cf030ed..2d872deb2c50 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_intel.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_intel.c
@@ -91,6 +91,36 @@ static int cmci_supported(int *banks)
 	return !!(cap & MCG_CMCI_P);
 }
 
+static bool lmce_supported(void)
+{
+	u64 tmp;
+
+	if (mca_cfg.lmce_disabled)
+		return false;
+
+	rdmsrl(MSR_IA32_MCG_CAP, tmp);
+
+	/*
+	 * LMCE depends on recovery support in the processor. Hence both
+	 * MCG_SER_P and MCG_LMCE_P should be present in MCG_CAP.
+	 */
+	if ((tmp & (MCG_SER_P | MCG_LMCE_P)) !=
+		   (MCG_SER_P | MCG_LMCE_P))
+		return false;
+
+	/*
+	 * BIOS should indicate support for LMCE by setting bit 20 in
+	 * IA32_FEATURE_CONTROL without which touching MCG_EXT_CTL will
+	 * generate a #GP fault.
+	 */
+	rdmsrl(MSR_IA32_FEATURE_CONTROL, tmp);
+	if ((tmp & (FEATURE_CONTROL_LOCKED | FEATURE_CONTROL_LMCE)) ==
+		   (FEATURE_CONTROL_LOCKED | FEATURE_CONTROL_LMCE))
+		return true;
+
+	return false;
+}
+
 bool mce_intel_cmci_poll(void)
 {
 	if (__this_cpu_read(cmci_storm_state) == CMCI_STORM_NONE)
@@ -405,6 +435,19 @@ static void intel_init_cmci(void)
 	cmci_recheck();
 }
 
+void intel_init_lmce(void)
+{
+	u64 val;
+
+	if (!lmce_supported())
+		return;
+
+	rdmsrl(MSR_IA32_MCG_EXT_CTL, val);
+
+	if (!(val & MCG_EXT_CTL_LMCE_EN))
+		wrmsrl(MSR_IA32_MCG_EXT_CTL, val | MCG_EXT_CTL_LMCE_EN);
+}
+
 void mce_intel_feature_init(struct cpuinfo_x86 *c)
 {
 	intel_init_thermal(c);
-- 
2.3.5

-- 
Regards/Gruss,
    Boris.

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