Skip to contentSkip to conversation

x86: mce: Bugfixes, cleanups and a new CMCI poll version

PATCH In review Thomas Gleixner 2 participants 10 messages Jun 6, 2012 x86

lore
Share & tools
lore
#0 lore raw

cover x86: mce: Bugfixes, cleanups and a new CMCI poll version

The following series fixes a few interesting bugs (found by review in context of the CMCI poll effort) and a cleanup to the timer/hotplug code followed by a consolidated version of the CMCI poll implementation. This series is based on

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git

which contains the bugfix for the dropped timer interval init.

Thanks,

tglx

#1 lore raw

PATCH 1/6 x86: mce: Create devices in CPU_UP_PREPARE

This makes the treshold_cpu_callback for AMD actually work and makes the code symetric vs. teardown.

---
 arch/x86/kernel/cpu/mcheck/mce.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/mcheck/mce.c +2 -2
Index: tip/arch/x86/kernel/cpu/mcheck/mce.c
===================================================================
@@ -2256,8 +2256,8 @@ mce_cpu_callback(struct notifier_block *
 	struct timer_list *t = &per_cpu(mce_timer, cpu);
 
 	switch (action) {
-	case CPU_ONLINE:
-	case CPU_ONLINE_FROZEN:
+	case CPU_UP_PREPARE:
+	case CPU_UP_PREPARE_FROZEN:
 		mce_device_create(cpu);
 		if (threshold_cpu_callback)
 			threshold_cpu_callback(action, cpu);
#2 lore raw

Re: [patch 1/6] x86: mce: Create devices in CPU_UP_PREPARE

Quoted 2 lines · Thomas Gleixner

This makes the treshold_cpu_callback for AMD actually work and makes the code symetric vs. teardown.

Ignore that. I sent out the wrong version of this lot. Sorry for the noise

#3 lore raw

Re: [patch 1/6] x86: mce: Create devices in CPU_UP_PREPARE

Quoted 6 lines · Thomas Gleixner
Quoted 2 lines · Thomas Gleixner

This makes the treshold_cpu_callback for AMD actually work and makes the code symetric vs. teardown.

Ignore that. I sent out the wrong version of this lot. Sorry for the noise

By this you mean the chunk at the end of 5/6, right?

There's a chunk touching amd_64_threshold_cpu_callback...

Signature · 8 lines
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551
#4 lore raw

Re: [patch 1/6] x86: mce: Create devices in CPU_UP_PREPARE

Quoted 11 lines · Borislav Petkov
Quoted 6 lines · Thomas Gleixner
Quoted 2 lines · Thomas Gleixner

This makes the treshold_cpu_callback for AMD actually work and makes the code symetric vs. teardown.

Ignore that. I sent out the wrong version of this lot. Sorry for the noise

By this you mean the chunk at the end of 5/6, right?

There's a chunk touching amd_64_threshold_cpu_callback...

Both 1/6 and 5/6 are wrong. That's a leftover from a previous iteration. About to send the fixed version.

#5 lore raw

PATCH 2/6 x86: mce: Disable preemption when calling raise_local()

In reply to Thomas Gleixner

raise_mce() has a code path which does not disable preemption when the raise_local() is called. The per cpu variable access in raise_local() depends on preemption being disabled to be functional. So that code path was either never tested or never tested with CONFIG_DEBUG_PREEMPT enabled.

Add the missing preempt_disable/enable() pair around the call.

---
 arch/x86/kernel/cpu/mcheck/mce-inject.c |    4 ++++
 1 file changed, 4 insertions(+)
arch/x86/kernel/cpu/mcheck/mce-inject.c +4 -0
Index: tip/arch/x86/kernel/cpu/mcheck/mce-inject.c
===================================================================
@@ -194,7 +194,11 @@ static void raise_mce(struct mce *m)
 		put_online_cpus();
 	} else
 #endif
+	{
+		preempt_disable();
 		raise_local();
+		preempt_enable();
+	}
 }
 
 /* Error injection interface */
#6 lore raw

PATCH 3/6 x86: mce: Serialize mce injection

In reply to Thomas Gleixner

raise_mce() fiddles with global state, but lacks any kind of serialization.

Add a mutex around the raise_mce() call, so concurrent writers do not stomp on each other toes.

---
 arch/x86/kernel/cpu/mcheck/mce-inject.c |    4 ++++
 1 file changed, 4 insertions(+)
arch/x86/kernel/cpu/mcheck/mce-inject.c +4 -0
Index: tip/arch/x86/kernel/cpu/mcheck/mce-inject.c
===================================================================
@@ -78,6 +78,7 @@ static void raise_exception(struct mce *
 }
 
 static cpumask_var_t mce_inject_cpumask;
+static DEFINE_MUTEX(mce_inject_mutex);
 
 static int mce_raise_notify(unsigned int cmd, struct pt_regs *regs)
 {
@@ -229,7 +230,10 @@ static ssize_t mce_write(struct file *fi
 	 * so do it a jiffie or two later everywhere.
 	 */
 	schedule_timeout(2);
+
+	mutex_lock(&mce_inject_mutex);
 	raise_mce(&m);
+	mutex_unlock(&mce_inject_mutex);
 	return usize;
 }
 
#7 lore raw

PATCH 4/6 x86: mce: Split timer init

In reply to Thomas Gleixner

Split timer init function into the init and the start part, so the start part can replace the open coded version in CPU_DOWN_FAILED.

---
 arch/x86/kernel/cpu/mcheck/mce.c |   25 +++++++++++++------------
 1 file changed, 13 insertions(+), 12 deletions(-)
arch/x86/kernel/cpu/mcheck/mce.c +13 -12
Index: tip/arch/x86/kernel/cpu/mcheck/mce.c
===================================================================
@@ -1554,23 +1554,28 @@ static void __mcheck_cpu_init_vendor(str
 	}
 }
 
-static void __mcheck_cpu_init_timer(void)
+static void mce_start_timer(unsigned int cpu, struct timer_list *t)
 {
-	struct timer_list *t = &__get_cpu_var(mce_timer);
 	unsigned long iv = check_interval * HZ;
 
-	setup_timer(t, mce_timer_fn, smp_processor_id());
+	__this_cpu_write(mce_next_interval, iv);
 
-	if (mce_ignore_ce)
+	if (mce_ignore_ce || !iv)
 		return;
 
-	__this_cpu_write(mce_next_interval, iv);
-	if (!iv)
-		return;
 	t->expires = round_jiffies(jiffies + iv);
 	add_timer_on(t, smp_processor_id());
 }
 
+static void __mcheck_cpu_init_timer(void)
+{
+	struct timer_list *t = &__get_cpu_var(mce_timer);
+	unsigned int cpu = smp_processor_id();
+
+	setup_timer(t, mce_timer_fn, cpu);
+	mce_start_timer(cpu, t);
+}
+
 /* Handle unconfigured int18 (should never happen) */
 static void unexpected_machine_check(struct pt_regs *regs, long error_code)
 {
@@ -2275,12 +2280,8 @@ mce_cpu_callback(struct notifier_block *
 		break;
 	case CPU_DOWN_FAILED:
 	case CPU_DOWN_FAILED_FROZEN:
-		if (!mce_ignore_ce && check_interval) {
-			t->expires = round_jiffies(jiffies +
-					per_cpu(mce_next_interval, cpu));
-			add_timer_on(t, cpu);
-		}
 		smp_call_function_single(cpu, mce_reenable_cpu, &action, 1);
+		mce_start_timer(cpu, t);
 		break;
 	case CPU_POST_DEAD:
 		/* intentionally ignoring frozen here */
#8 lore raw

PATCH 5/6 x86: mce: Remove the frozen cases in the hotplug code

In reply to Thomas Gleixner

No point in having double cases if we can simply mask the FROZEN bit out.

---
 arch/x86/kernel/cpu/mcheck/mce.c     |   12 +++++-------
 arch/x86/kernel/cpu/mcheck/mce_amd.c |    6 ++----
 2 files changed, 7 insertions(+), 11 deletions(-)
arch/x86/kernel/cpu/mcheck/mce.c +5 -7
Index: tip/arch/x86/kernel/cpu/mcheck/mce.c
===================================================================
@@ -2260,34 +2260,32 @@ mce_cpu_callback(struct notifier_block *
 	unsigned int cpu = (unsigned long)hcpu;
 	struct timer_list *t = &per_cpu(mce_timer, cpu);
 
-	switch (action) {
+	switch (action & ~CPU_TASKS_FROZEN) {
 	case CPU_UP_PREPARE:
-	case CPU_UP_PREPARE_FROZEN:
 		mce_device_create(cpu);
 		if (threshold_cpu_callback)
 			threshold_cpu_callback(action, cpu);
 		break;
 	case CPU_DEAD:
-	case CPU_DEAD_FROZEN:
 		if (threshold_cpu_callback)
 			threshold_cpu_callback(action, cpu);
 		mce_device_remove(cpu);
 		break;
 	case CPU_DOWN_PREPARE:
-	case CPU_DOWN_PREPARE_FROZEN:
 		del_timer_sync(t);
 		smp_call_function_single(cpu, mce_disable_cpu, &action, 1);
 		break;
 	case CPU_DOWN_FAILED:
-	case CPU_DOWN_FAILED_FROZEN:
 		smp_call_function_single(cpu, mce_reenable_cpu, &action, 1);
 		mce_start_timer(cpu, t);
 		break;
-	case CPU_POST_DEAD:
+	}
+
+	if (action == CPU_POST_DEAD) {
 		/* intentionally ignoring frozen here */
 		cmci_rediscover(cpu);
-		break;
 	}
+
 	return NOTIFY_OK;
 }
 
arch/x86/kernel/cpu/mcheck/mce_amd.c +2 -4
Index: tip/arch/x86/kernel/cpu/mcheck/mce_amd.c
===================================================================
@@ -748,13 +748,11 @@ static void threshold_remove_device(unsi
 static void __cpuinit
 amd_64_threshold_cpu_callback(unsigned long action, unsigned int cpu)
 {
-	switch (action) {
-	case CPU_ONLINE:
-	case CPU_ONLINE_FROZEN:
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_UP_PREPARE:
 		threshold_create_device(cpu);
 		break;
 	case CPU_DEAD:
-	case CPU_DEAD_FROZEN:
 		threshold_remove_device(cpu);
 		break;
 	default:
#9 lore raw

PATCH 6/6 x86: mce: Add cmci poll mode

In reply to Thomas Gleixner

Still waits for explanation :)

---
 arch/x86/kernel/cpu/mcheck/mce-internal.h |   10 +++
 arch/x86/kernel/cpu/mcheck/mce.c          |   46 +++++++++++++--
 arch/x86/kernel/cpu/mcheck/mce_intel.c    |   88 +++++++++++++++++++++++++++++-
 3 files changed, 137 insertions(+), 7 deletions(-)
arch/x86/kernel/cpu/mcheck/mce-internal.h +10 -0
Index: tip/arch/x86/kernel/cpu/mcheck/mce-internal.h
===================================================================
@@ -28,6 +28,16 @@ extern int mce_ser;
 
 extern struct mce_bank *mce_banks;
 
+#ifdef CONFIG_X86_MCE_INTEL
+unsigned long mce_intel_adjust_timer(unsigned long interval);
+void mce_intel_cmci_poll(void);
+#else
+# define mce_intel_adjust_timer mce_adjust_timer_default
+static inline void mce_intel_cmci_poll(void) { }
+#endif
+
+void mce_timer_kick(unsigned long interval);
+
 #ifdef CONFIG_ACPI_APEI
 int apei_write_mce(struct mce *m);
 ssize_t apei_read_mce(struct mce *m, u64 *record_id);
arch/x86/kernel/cpu/mcheck/mce.c +40 -6

87 lines not on this page · raw message

arch/x86/kernel/cpu/mcheck/mce_intel.c +87 -1

112 lines not on this page · raw message