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/gfp.h>
   9 #include <linux/init.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/percpu.h>
  12 #include <linux/sched.h>
  13 #include <asm/apic.h>
  14 #include <asm/processor.h>
  15 #include <asm/msr.h>
  16 #include <asm/mce.h>
  17
  18 /*
  19  * Support for Intel Correct Machine Check Interrupts. This allows
  20  * the CPU to raise an interrupt when a corrected machine check happened.
  21  * Normally we pick those up using a regular polling timer.
  22  * Also supports reliable discovery of shared banks.
  23  */
  24
  25 static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
  26
  27 /*
  28  * cmci_discover_lock protects against parallel discovery attempts
  29  * which could race against each other.
  30  */
  31 static DEFINE_SPINLOCK(cmci_discover_lock);
  32
  33 #define CMCI_THRESHOLD 1
  34
  35 static int cmci_supported(int *banks)
  36 {
  37         u64 cap;
  38
  39         if (mce_cmci_disabled || mce_ignore_ce)
  40                 return 0;
  41
  42         /*
  43          * Vendor check is not strictly needed, but the initial
  44          * initialization is vendor keyed and this
  45          * makes sure none of the backdoors are entered otherwise.
  46          */
  47         if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
  48                 return 0;
  49         if (!cpu_has_apic || lapic_get_maxlvt() < 6)
  50                 return 0;
  51         rdmsrl(MSR_IA32_MCG_CAP, cap);
  52         *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
  53         return !!(cap & MCG_CMCI_P);
  54 }
  55
  56 /*
  57  * The interrupt handler. This is called on every event.
  58  * Just call the poller directly to log any events.
  59  * This could in theory increase the threshold under high load,
  60  * but doesn't for now.
  61  */
  62 static void intel_threshold_interrupt(void)
  63 {
  64         machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
  65         mce_notify_irq();
  66 }
  67
  68 static void print_update(char *type, int *hdr, int num)
  69 {
  70         if (*hdr == 0)
  71                 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
  72         *hdr = 1;
  73         printk(KERN_CONT " %s:%d", type, num);
  74 }
  75
  76 /*
  77  * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
  78  * on this CPU. Use the algorithm recommended in the SDM to discover shared
  79  * banks.
  80  */
  81 static void cmci_discover(int banks, int boot)
  82 {
  83         unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
  84         unsigned long flags;
  85         int hdr = 0;
  86         int i;
  87
  88         spin_lock_irqsave(&cmci_discover_lock, flags);
  89         for (i = 0; i < banks; i++) {
  90                 u64 val;
  91
  92                 if (test_bit(i, owned))
  93                         continue;
  94
  95                 rdmsrl(MSR_IA32_MCx_CTL2(i), val);
  96
  97                 /* Already owned by someone else? */
  98                 if (val & MCI_CTL2_CMCI_EN) {
  99                         if (test_and_clear_bit(i, owned) && !boot)
 100                                 print_update("SHD", &hdr, i);
 101                         __clear_bit(i, __get_cpu_var(mce_poll_banks));
 102                         continue;
 103                 }
 104
 105                 val &= ~MCI_CTL2_CMCI_THRESHOLD_MASK;
 106                 val |= MCI_CTL2_CMCI_EN | CMCI_THRESHOLD;
 107                 wrmsrl(MSR_IA32_MCx_CTL2(i), val);
 108                 rdmsrl(MSR_IA32_MCx_CTL2(i), val);
 109
 110                 /* Did the enable bit stick? -- the bank supports CMCI */
 111                 if (val & MCI_CTL2_CMCI_EN) {
 112                         if (!test_and_set_bit(i, owned) && !boot)
 113                                 print_update("CMCI", &hdr, i);
 114                         __clear_bit(i, __get_cpu_var(mce_poll_banks));
 115                 } else {
 116                         WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
 117                 }
 118         }
 119         spin_unlock_irqrestore(&cmci_discover_lock, flags);
 120         if (hdr)
 121                 printk(KERN_CONT "\n");
 122 }
 123
 124 /*
 125  * Just in case we missed an event during initialization check
 126  * all the CMCI owned banks.
 127  */
 128 void cmci_recheck(void)
 129 {
 130         unsigned long flags;
 131         int banks;
 132
 133         if (!mce_available(&current_cpu_data) || !cmci_supported(&banks))
 134                 return;
 135         local_irq_save(flags);
 136         machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
 137         local_irq_restore(flags);
 138 }
 139
 140 /*
 141  * Disable CMCI on this CPU for all banks it owns when it goes down.
 142  * This allows other CPUs to claim the banks on rediscovery.
 143  */
 144 void cmci_clear(void)
 145 {
 146         unsigned long flags;
 147         int i;
 148         int banks;
 149         u64 val;
 150
 151         if (!cmci_supported(&banks))
 152                 return;
 153         spin_lock_irqsave(&cmci_discover_lock, flags);
 154         for (i = 0; i < banks; i++) {
 155                 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
 156                         continue;
 157                 /* Disable CMCI */
 158                 rdmsrl(MSR_IA32_MCx_CTL2(i), val);
 159                 val &= ~(MCI_CTL2_CMCI_EN|MCI_CTL2_CMCI_THRESHOLD_MASK);
 160                 wrmsrl(MSR_IA32_MCx_CTL2(i), val);
 161                 __clear_bit(i, __get_cpu_var(mce_banks_owned));
 162         }
 163         spin_unlock_irqrestore(&cmci_discover_lock, flags);
 164 }
 165
 166 /*
 167  * After a CPU went down cycle through all the others and rediscover
 168  * Must run in process context.
 169  */
 170 void cmci_rediscover(int dying)
 171 {
 172         int banks;
 173         int cpu;
 174         cpumask_var_t old;
 175
 176         if (!cmci_supported(&banks))
 177                 return;
 178         if (!alloc_cpumask_var(&old, GFP_KERNEL))
 179                 return;
 180         cpumask_copy(old, &current->cpus_allowed);
 181
 182         for_each_online_cpu(cpu) {
 183                 if (cpu == dying)
 184                         continue;
 185                 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
 186                         continue;
 187                 /* Recheck banks in case CPUs don't all have the same */
 188                 if (cmci_supported(&banks))
 189                         cmci_discover(banks, 0);
 190         }
 191
 192         set_cpus_allowed_ptr(current, old);
 193         free_cpumask_var(old);
 194 }
 195
 196 /*
 197  * Reenable CMCI on this CPU in case a CPU down failed.
 198  */
 199 void cmci_reenable(void)
 200 {
 201         int banks;
 202         if (cmci_supported(&banks))
 203                 cmci_discover(banks, 0);
 204 }
 205
 206 static void intel_init_cmci(void)
 207 {
 208         int banks;
 209
 210         if (!cmci_supported(&banks))
 211                 return;
 212
 213         mce_threshold_vector = intel_threshold_interrupt;
 214         cmci_discover(banks, 1);
 215         /*
 216          * For CPU #0 this runs with still disabled APIC, but that's
 217          * ok because only the vector is set up. We still do another
 218          * check for the banks later for CPU #0 just to make sure
 219          * to not miss any events.
 220          */
 221         apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
 222         cmci_recheck();
 223 }
 224
 225 void mce_intel_feature_init(struct cpuinfo_x86 *c)
 226 {
 227         intel_init_thermal(c);
 228         intel_init_cmci();
 229 }