x86, mce: fix reporting of Thermal Monitoring mechanism enabled
[linux-2.6.git] / arch / x86 / kernel / cpu / mcheck / therm_throt.c
index 1203dc5..15f2bc0 100644 (file)
@@ -1,8 +1,7 @@
 /*
- * linux/arch/i386/kernel/cpu/mcheck/therm_throt.c
- *
  * Thermal throttle event support code (such as syslog messaging and rate
  * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
+ *
  * This allows consistent reporting of CPU thermal throttle events.
  *
  * Maintains a counter in /sys that keeps track of the number of thermal
  * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
  *          Inspired by Ross Biro's and Al Borchers' counter code.
  */
-
+#include <linux/interrupt.h>
+#include <linux/notifier.h>
+#include <linux/jiffies.h>
+#include <linux/kernel.h>
 #include <linux/percpu.h>
 #include <linux/sysdev.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/smp.h>
 #include <linux/cpu.h>
-#include <asm/cpu.h>
-#include <linux/notifier.h>
-#include <linux/jiffies.h>
-#include <asm/therm_throt.h>
+
+#include <asm/processor.h>
+#include <asm/system.h>
+#include <asm/apic.h>
+#include <asm/idle.h>
+#include <asm/mce.h>
+#include <asm/msr.h>
 
 /* How long to wait between reporting thermal events */
-#define CHECK_INTERVAL              (300 * HZ)
+#define CHECK_INTERVAL         (300 * HZ)
 
 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
-atomic_t therm_throt_en = ATOMIC_INIT(0);
+
+static atomic_t therm_throt_en         = ATOMIC_INIT(0);
 
 #ifdef CONFIG_SYSFS
-#define define_therm_throt_sysdev_one_ro(_name)                              \
-        static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
-
-#define define_therm_throt_sysdev_show_func(name)                            \
-static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,        \
-                                              char *buf)                     \
-{                                                                            \
-       unsigned int cpu = dev->id;                                          \
-       ssize_t ret;                                                         \
-                                                                             \
-       preempt_disable();              /* CPU hotplug */                    \
-       if (cpu_online(cpu))                                                 \
-               ret = sprintf(buf, "%lu\n",                                  \
-                             per_cpu(thermal_throttle_##name, cpu));        \
-       else                                                                 \
-               ret = 0;                                                     \
-       preempt_enable();                                                    \
-                                                                             \
-       return ret;                                                          \
+#define define_therm_throt_sysdev_one_ro(_name)                                \
+       static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
+
+#define define_therm_throt_sysdev_show_func(name)                      \
+static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,  \
+                                       struct sysdev_attribute *attr,  \
+                                             char *buf)                \
+{                                                                      \
+       unsigned int cpu = dev->id;                                     \
+       ssize_t ret;                                                    \
+                                                                       \
+       preempt_disable();      /* CPU hotplug */                       \
+       if (cpu_online(cpu))                                            \
+               ret = sprintf(buf, "%lu\n",                             \
+                             per_cpu(thermal_throttle_##name, cpu));   \
+       else                                                            \
+               ret = 0;                                                \
+       preempt_enable();                                               \
+                                                                       \
+       return ret;                                                     \
 }
 
 define_therm_throt_sysdev_show_func(count);
@@ -61,8 +71,8 @@ static struct attribute *thermal_throttle_attrs[] = {
 };
 
 static struct attribute_group thermal_throttle_attr_group = {
-       .attrs = thermal_throttle_attrs,
-       .name = "thermal_throttle"
+       .attrs  = thermal_throttle_attrs,
+       .name   = "thermal_throttle"
 };
 #endif /* CONFIG_SYSFS */
 
@@ -82,7 +92,7 @@ static struct attribute_group thermal_throttle_attr_group = {
  *          1 : Event should be logged further, and a message has been
  *              printed to the syslog.
  */
-int therm_throt_process(int curr)
+static int therm_throt_process(int curr)
 {
        unsigned int cpu = smp_processor_id();
        __u64 tmp_jiffs = get_jiffies_64();
@@ -110,38 +120,43 @@ int therm_throt_process(int curr)
 }
 
 #ifdef CONFIG_SYSFS
-/* Add/Remove thermal_throttle interface for CPU device */
+/* Add/Remove thermal_throttle interface for CPU device: */
 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 {
-       return sysfs_create_group(&sys_dev->kobj, &thermal_throttle_attr_group);
+       return sysfs_create_group(&sys_dev->kobj,
+                                 &thermal_throttle_attr_group);
 }
 
 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 {
-       return sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
+       sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 }
 
-/* Mutex protecting device creation against CPU hotplug */
+/* Mutex protecting device creation against CPU hotplug: */
 static DEFINE_MUTEX(therm_cpu_lock);
 
 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
-static __cpuinit int thermal_throttle_cpu_callback(struct notifier_block *nfb,
-                                                  unsigned long action,
-                                                  void *hcpu)
+static __cpuinit int
+thermal_throttle_cpu_callback(struct notifier_block *nfb,
+                             unsigned long action,
+                             void *hcpu)
 {
        unsigned int cpu = (unsigned long)hcpu;
        struct sys_device *sys_dev;
-       int err;
+       int err = 0;
 
        sys_dev = get_cpu_sysdev(cpu);
+
        switch (action) {
-       case CPU_ONLINE:
-       case CPU_ONLINE_FROZEN:
+       case CPU_UP_PREPARE:
+       case CPU_UP_PREPARE_FROZEN:
                mutex_lock(&therm_cpu_lock);
                err = thermal_throttle_add_dev(sys_dev);
                mutex_unlock(&therm_cpu_lock);
                WARN_ON(err);
                break;
+       case CPU_UP_CANCELED:
+       case CPU_UP_CANCELED_FROZEN:
        case CPU_DEAD:
        case CPU_DEAD_FROZEN:
                mutex_lock(&therm_cpu_lock);
@@ -149,10 +164,10 @@ static __cpuinit int thermal_throttle_cpu_callback(struct notifier_block *nfb,
                mutex_unlock(&therm_cpu_lock);
                break;
        }
-       return NOTIFY_OK;
+       return err ? NOTIFY_BAD : NOTIFY_OK;
 }
 
-static struct notifier_block thermal_throttle_cpu_notifier =
+static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 {
        .notifier_call = thermal_throttle_cpu_callback,
 };
@@ -181,6 +196,101 @@ static __init int thermal_throttle_init_device(void)
 
        return 0;
 }
-
 device_initcall(thermal_throttle_init_device);
+
 #endif /* CONFIG_SYSFS */
+
+/* Thermal transition interrupt handler */
+static void intel_thermal_interrupt(void)
+{
+       __u64 msr_val;
+
+       rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
+       if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
+               mce_log_therm_throt_event(msr_val);
+}
+
+static void unexpected_thermal_interrupt(void)
+{
+       printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
+                       smp_processor_id());
+       add_taint(TAINT_MACHINE_CHECK);
+}
+
+static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
+
+asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
+{
+       exit_idle();
+       irq_enter();
+       inc_irq_stat(irq_thermal_count);
+       smp_thermal_vector();
+       irq_exit();
+       /* Ack only at the end to avoid potential reentry */
+       ack_APIC_irq();
+}
+
+void intel_init_thermal(struct cpuinfo_x86 *c)
+{
+       unsigned int cpu = smp_processor_id();
+       int tm2 = 0;
+       u32 l, h;
+
+       /* Thermal monitoring depends on ACPI and clock modulation*/
+       if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
+               return;
+
+       /*
+        * First check if its enabled already, in which case there might
+        * be some SMM goo which handles it, so we can't even put a handler
+        * since it might be delivered via SMI already:
+        */
+       rdmsr(MSR_IA32_MISC_ENABLE, l, h);
+       h = apic_read(APIC_LVTTHMR);
+       if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
+               printk(KERN_DEBUG
+                      "CPU%d: Thermal monitoring handled by SMI\n", cpu);
+               return;
+       }
+
+       /* Check whether a vector already exists */
+       if (h & APIC_VECTOR_MASK) {
+               printk(KERN_DEBUG
+                      "CPU%d: Thermal LVT vector (%#x) already installed\n",
+                      cpu, (h & APIC_VECTOR_MASK));
+               return;
+       }
+
+       /* early Pentium M models use different method for enabling TM2 */
+       if (cpu_has(c, X86_FEATURE_TM2)) {
+               if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
+                       rdmsr(MSR_THERM2_CTL, l, h);
+                       if (l & MSR_THERM2_CTL_TM_SELECT)
+                               tm2 = 1;
+               } else if (l & MSR_IA32_MISC_ENABLE_TM2)
+                       tm2 = 1;
+       }
+
+       /* We'll mask the thermal vector in the lapic till we're ready: */
+       h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
+       apic_write(APIC_LVTTHMR, h);
+
+       rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
+       wrmsr(MSR_IA32_THERM_INTERRUPT,
+               l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
+
+       smp_thermal_vector = intel_thermal_interrupt;
+
+       rdmsr(MSR_IA32_MISC_ENABLE, l, h);
+       wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
+
+       /* Unmask the thermal vector: */
+       l = apic_read(APIC_LVTTHMR);
+       apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
+
+       printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
+              cpu, tm2 ? "TM2" : "TM1");
+
+       /* enable thermal throttle processing */
+       atomic_set(&therm_throt_en, 1);
+}