2 * Intel specific MCE features.
3 * Copyright 2004 Zwane Mwaikambo <zwane@linuxpower.ca>
4 * Copyright (C) 2008, 2009 Intel Corporation
8 #include <linux/init.h>
9 #include <linux/interrupt.h>
10 #include <linux/percpu.h>
11 #include <asm/processor.h>
15 #include <asm/hw_irq.h>
17 #include <asm/therm_throt.h>
20 asmlinkage void smp_thermal_interrupt(void)
29 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
30 if (therm_throt_process(msr_val & 1))
31 mce_log_therm_throt_event(msr_val);
33 inc_irq_stat(irq_thermal_count);
37 static void intel_init_thermal(struct cpuinfo_x86 *c)
41 unsigned int cpu = smp_processor_id();
43 if (!cpu_has(c, X86_FEATURE_ACPI))
46 if (!cpu_has(c, X86_FEATURE_ACC))
49 /* first check if TM1 is already enabled by the BIOS, in which
50 * case there might be some SMM goo which handles it, so we can't even
51 * put a handler since it might be delivered via SMI already.
53 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
54 h = apic_read(APIC_LVTTHMR);
55 if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
57 "CPU%d: Thermal monitoring handled by SMI\n", cpu);
61 if (cpu_has(c, X86_FEATURE_TM2) && (l & MSR_IA32_MISC_ENABLE_TM2))
64 if (h & APIC_VECTOR_MASK) {
66 "CPU%d: Thermal LVT vector (%#x) already "
67 "installed\n", cpu, (h & APIC_VECTOR_MASK));
71 h = THERMAL_APIC_VECTOR;
72 h |= (APIC_DM_FIXED | APIC_LVT_MASKED);
73 apic_write(APIC_LVTTHMR, h);
75 rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
76 wrmsr(MSR_IA32_THERM_INTERRUPT, l | 0x03, h);
78 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
79 wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
81 l = apic_read(APIC_LVTTHMR);
82 apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
83 printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
84 cpu, tm2 ? "TM2" : "TM1");
86 /* enable thermal throttle processing */
87 atomic_set(&therm_throt_en, 1);
92 * Support for Intel Correct Machine Check Interrupts. This allows
93 * the CPU to raise an interrupt when a corrected machine check happened.
94 * Normally we pick those up using a regular polling timer.
95 * Also supports reliable discovery of shared banks.
98 static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
101 * cmci_discover_lock protects against parallel discovery attempts
102 * which could race against each other.
104 static DEFINE_SPINLOCK(cmci_discover_lock);
106 #define CMCI_THRESHOLD 1
108 static int cmci_supported(int *banks)
113 * Vendor check is not strictly needed, but the initial
114 * initialization is vendor keyed and this
115 * makes sure none of the backdoors are entered otherwise.
117 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
119 if (!cpu_has_apic || lapic_get_maxlvt() < 6)
121 rdmsrl(MSR_IA32_MCG_CAP, cap);
122 *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
123 return !!(cap & MCG_CMCI_P);
127 * The interrupt handler. This is called on every event.
128 * Just call the poller directly to log any events.
129 * This could in theory increase the threshold under high load,
130 * but doesn't for now.
132 static void intel_threshold_interrupt(void)
134 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
138 static void print_update(char *type, int *hdr, int num)
141 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
143 printk(KERN_CONT " %s:%d", type, num);
147 * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
148 * on this CPU. Use the algorithm recommended in the SDM to discover shared
151 static void cmci_discover(int banks, int boot)
153 unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
157 spin_lock(&cmci_discover_lock);
158 for (i = 0; i < banks; i++) {
161 if (test_bit(i, owned))
164 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
166 /* Already owned by someone else? */
168 if (test_and_clear_bit(i, owned) || boot)
169 print_update("SHD", &hdr, i);
170 __clear_bit(i, __get_cpu_var(mce_poll_banks));
174 val |= CMCI_EN | CMCI_THRESHOLD;
175 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
176 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
178 /* Did the enable bit stick? -- the bank supports CMCI */
180 if (!test_and_set_bit(i, owned) || boot)
181 print_update("CMCI", &hdr, i);
182 __clear_bit(i, __get_cpu_var(mce_poll_banks));
184 WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
187 spin_unlock(&cmci_discover_lock);
189 printk(KERN_CONT "\n");
193 * Just in case we missed an event during initialization check
194 * all the CMCI owned banks.
196 void cmci_recheck(void)
201 if (!mce_available(¤t_cpu_data) || !cmci_supported(&banks))
203 local_irq_save(flags);
204 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
205 local_irq_restore(flags);
209 * Disable CMCI on this CPU for all banks it owns when it goes down.
210 * This allows other CPUs to claim the banks on rediscovery.
212 void cmci_clear(void)
218 if (!cmci_supported(&banks))
220 spin_lock(&cmci_discover_lock);
221 for (i = 0; i < banks; i++) {
222 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
225 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
226 val &= ~(CMCI_EN|CMCI_THRESHOLD_MASK);
227 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
228 __clear_bit(i, __get_cpu_var(mce_banks_owned));
230 spin_unlock(&cmci_discover_lock);
234 * After a CPU went down cycle through all the others and rediscover
235 * Must run in process context.
237 void cmci_rediscover(int dying)
243 if (!cmci_supported(&banks))
245 if (!alloc_cpumask_var(&old, GFP_KERNEL))
247 cpumask_copy(old, ¤t->cpus_allowed);
249 for_each_online_cpu (cpu) {
252 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
254 /* Recheck banks in case CPUs don't all have the same */
255 if (cmci_supported(&banks))
256 cmci_discover(banks, 0);
259 set_cpus_allowed_ptr(current, old);
260 free_cpumask_var(old);
264 * Reenable CMCI on this CPU in case a CPU down failed.
266 void cmci_reenable(void)
269 if (cmci_supported(&banks))
270 cmci_discover(banks, 0);
273 static void intel_init_cmci(void)
277 if (!cmci_supported(&banks))
280 mce_threshold_vector = intel_threshold_interrupt;
281 cmci_discover(banks, 1);
283 * For CPU #0 this runs with still disabled APIC, but that's
284 * ok because only the vector is set up. We still do another
285 * check for the banks later for CPU #0 just to make sure
286 * to not miss any events.
288 apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
292 void mce_intel_feature_init(struct cpuinfo_x86 *c)
294 intel_init_thermal(c);