arch/x86/kernel/cpu/mcheck/mce_intel.c

   1 /*
   2  * Intel specific MCE features.
   3  * Copyright 2004 Zwane Mwaikambo <zwane@linuxpower.ca>
   4  * Copyright (C) 2008, 2009 Intel Corporation
   5  * Author: Andi Kleen
   6  */
   7
   8 #include <linux/init.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/percpu.h>
  11 #include <asm/apic.h>
  12 #include <asm/processor.h>
  13 #include <asm/msr.h>
  14 #include <asm/mce.h>
  15
  16 /*
  17  * Support for Intel Correct Machine Check Interrupts. This allows
  18  * the CPU to raise an interrupt when a corrected machine check happened.
  19  * Normally we pick those up using a regular polling timer.
  20  * Also supports reliable discovery of shared banks.
  21  */
  22
  23 static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
  24
  25 /*
  26  * cmci_discover_lock protects against parallel discovery attempts
  27  * which could race against each other.
  28  */
  29 static DEFINE_SPINLOCK(cmci_discover_lock);
  30
  31 #define CMCI_THRESHOLD 1
  32
  33 static int cmci_supported(int *banks)
  34 {
  35         u64 cap;
  36
  37         if (mce_cmci_disabled || mce_ignore_ce)
  38                 return 0;
  39
  40         /*
  41          * Vendor check is not strictly needed, but the initial
  42          * initialization is vendor keyed and this
  43          * makes sure none of the backdoors are entered otherwise.
  44          */
  45         if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
  46                 return 0;
  47         if (!cpu_has_apic || lapic_get_maxlvt() < 6)
  48                 return 0;
  49         rdmsrl(MSR_IA32_MCG_CAP, cap);
  50         *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
  51         return !!(cap & MCG_CMCI_P);
  52 }
  53
  54 /*
  55  * The interrupt handler. This is called on every event.
  56  * Just call the poller directly to log any events.
  57  * This could in theory increase the threshold under high load,
  58  * but doesn't for now.
  59  */
  60 static void intel_threshold_interrupt(void)
  61 {
  62         machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
  63         mce_notify_irq();
  64 }
  65
  66 static void print_update(char *type, int *hdr, int num)
  67 {
  68         if (*hdr == 0)
  69                 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
  70         *hdr = 1;
  71         printk(KERN_CONT " %s:%d", type, num);
  72 }
  73
  74 /*
  75  * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
  76  * on this CPU. Use the algorithm recommended in the SDM to discover shared
  77  * banks.
  78  */
  79 static void cmci_discover(int banks, int boot)
  80 {
  81         unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
  82         unsigned long flags;
  83         int hdr = 0;
  84         int i;
  85
  86         spin_lock_irqsave(&cmci_discover_lock, flags);
  87         for (i = 0; i < banks; i++) {
  88                 u64 val;
  89
  90                 if (test_bit(i, owned))
  91                         continue;
  92
  93                 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
  94
  95                 /* Already owned by someone else? */
  96                 if (val & CMCI_EN) {
  97                         if (test_and_clear_bit(i, owned) || boot)
  98                                 print_update("SHD", &hdr, i);
  99                         __clear_bit(i, __get_cpu_var(mce_poll_banks));
 100                         continue;
 101                 }
 102
 103                 val |= CMCI_EN | CMCI_THRESHOLD;
 104                 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
 105                 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
 106
 107                 /* Did the enable bit stick? -- the bank supports CMCI */
 108                 if (val & CMCI_EN) {
 109                         if (!test_and_set_bit(i, owned) || boot)
 110                                 print_update("CMCI", &hdr, i);
 111                         __clear_bit(i, __get_cpu_var(mce_poll_banks));
 112                 } else {
 113                         WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
 114                 }
 115         }
 116         spin_unlock_irqrestore(&cmci_discover_lock, flags);
 117         if (hdr)
 118                 printk(KERN_CONT "\n");
 119 }
 120
 121 /*
 122  * Just in case we missed an event during initialization check
 123  * all the CMCI owned banks.
 124  */
 125 void cmci_recheck(void)
 126 {
 127         unsigned long flags;
 128         int banks;
 129
 130         if (!mce_available(&current_cpu_data) || !cmci_supported(&banks))
 131                 return;
 132         local_irq_save(flags);
 133         machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
 134         local_irq_restore(flags);
 135 }
 136
 137 /*
 138  * Disable CMCI on this CPU for all banks it owns when it goes down.
 139  * This allows other CPUs to claim the banks on rediscovery.
 140  */
 141 void cmci_clear(void)
 142 {
 143         unsigned long flags;
 144         int i;
 145         int banks;
 146         u64 val;
 147
 148         if (!cmci_supported(&banks))
 149                 return;
 150         spin_lock_irqsave(&cmci_discover_lock, flags);
 151         for (i = 0; i < banks; i++) {
 152                 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
 153                         continue;
 154                 /* Disable CMCI */
 155                 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
 156                 val &= ~(CMCI_EN|CMCI_THRESHOLD_MASK);
 157                 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
 158                 __clear_bit(i, __get_cpu_var(mce_banks_owned));
 159         }
 160         spin_unlock_irqrestore(&cmci_discover_lock, flags);
 161 }
 162
 163 /*
 164  * After a CPU went down cycle through all the others and rediscover
 165  * Must run in process context.
 166  */
 167 void cmci_rediscover(int dying)
 168 {
 169         int banks;
 170         int cpu;
 171         cpumask_var_t old;
 172
 173         if (!cmci_supported(&banks))
 174                 return;
 175         if (!alloc_cpumask_var(&old, GFP_KERNEL))
 176                 return;
 177         cpumask_copy(old, &current->cpus_allowed);
 178
 179         for_each_online_cpu(cpu) {
 180                 if (cpu == dying)
 181                         continue;
 182                 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
 183                         continue;
 184                 /* Recheck banks in case CPUs don't all have the same */
 185                 if (cmci_supported(&banks))
 186                         cmci_discover(banks, 0);
 187         }
 188
 189         set_cpus_allowed_ptr(current, old);
 190         free_cpumask_var(old);
 191 }
 192
 193 /*
 194  * Reenable CMCI on this CPU in case a CPU down failed.
 195  */
 196 void cmci_reenable(void)
 197 {
 198         int banks;
 199         if (cmci_supported(&banks))
 200                 cmci_discover(banks, 0);
 201 }
 202
 203 static void intel_init_cmci(void)
 204 {
 205         int banks;
 206
 207         if (!cmci_supported(&banks))
 208                 return;
 209
 210         mce_threshold_vector = intel_threshold_interrupt;
 211         cmci_discover(banks, 1);
 212         /*
 213          * For CPU #0 this runs with still disabled APIC, but that's
 214          * ok because only the vector is set up. We still do another
 215          * check for the banks later for CPU #0 just to make sure
 216          * to not miss any events.
 217          */
 218         apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
 219         cmci_recheck();
 220 }
 221
 222 void mce_intel_feature_init(struct cpuinfo_x86 *c)
 223 {
 224         intel_init_thermal(c);
 225         intel_init_cmci();
 226 }