x86, mce: Add boot options for corrected errors
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kernel / cpu / mcheck / mce_intel_64.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/processor.h>
12 #include <asm/apic.h>
13 #include <asm/msr.h>
14 #include <asm/mce.h>
15 #include <asm/hw_irq.h>
16 #include <asm/idle.h>
17 #include <asm/therm_throt.h>
18 #include <asm/apic.h>
19
20 #include "mce.h"
21
22 asmlinkage void smp_thermal_interrupt(void)
23 {
24 __u64 msr_val;
25
26 ack_APIC_irq();
27
28 exit_idle();
29 irq_enter();
30
31 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
32 if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
33 mce_log_therm_throt_event(msr_val);
34
35 inc_irq_stat(irq_thermal_count);
36 irq_exit();
37 }
38
39 /*
40 * Support for Intel Correct Machine Check Interrupts. This allows
41 * the CPU to raise an interrupt when a corrected machine check happened.
42 * Normally we pick those up using a regular polling timer.
43 * Also supports reliable discovery of shared banks.
44 */
45
46 static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
47
48 /*
49 * cmci_discover_lock protects against parallel discovery attempts
50 * which could race against each other.
51 */
52 static DEFINE_SPINLOCK(cmci_discover_lock);
53
54 #define CMCI_THRESHOLD 1
55
56 static int cmci_supported(int *banks)
57 {
58 u64 cap;
59
60 if (mce_cmci_disabled || mce_ignore_ce)
61 return 0;
62
63 /*
64 * Vendor check is not strictly needed, but the initial
65 * initialization is vendor keyed and this
66 * makes sure none of the backdoors are entered otherwise.
67 */
68 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
69 return 0;
70 if (!cpu_has_apic || lapic_get_maxlvt() < 6)
71 return 0;
72 rdmsrl(MSR_IA32_MCG_CAP, cap);
73 *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
74 return !!(cap & MCG_CMCI_P);
75 }
76
77 /*
78 * The interrupt handler. This is called on every event.
79 * Just call the poller directly to log any events.
80 * This could in theory increase the threshold under high load,
81 * but doesn't for now.
82 */
83 static void intel_threshold_interrupt(void)
84 {
85 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
86 mce_notify_irq();
87 }
88
89 static void print_update(char *type, int *hdr, int num)
90 {
91 if (*hdr == 0)
92 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
93 *hdr = 1;
94 printk(KERN_CONT " %s:%d", type, num);
95 }
96
97 /*
98 * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
99 * on this CPU. Use the algorithm recommended in the SDM to discover shared
100 * banks.
101 */
102 static void cmci_discover(int banks, int boot)
103 {
104 unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
105 unsigned long flags;
106 int hdr = 0;
107 int i;
108
109 spin_lock_irqsave(&cmci_discover_lock, flags);
110 for (i = 0; i < banks; i++) {
111 u64 val;
112
113 if (test_bit(i, owned))
114 continue;
115
116 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
117
118 /* Already owned by someone else? */
119 if (val & CMCI_EN) {
120 if (test_and_clear_bit(i, owned) || boot)
121 print_update("SHD", &hdr, i);
122 __clear_bit(i, __get_cpu_var(mce_poll_banks));
123 continue;
124 }
125
126 val |= CMCI_EN | CMCI_THRESHOLD;
127 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
128 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
129
130 /* Did the enable bit stick? -- the bank supports CMCI */
131 if (val & CMCI_EN) {
132 if (!test_and_set_bit(i, owned) || boot)
133 print_update("CMCI", &hdr, i);
134 __clear_bit(i, __get_cpu_var(mce_poll_banks));
135 } else {
136 WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
137 }
138 }
139 spin_unlock_irqrestore(&cmci_discover_lock, flags);
140 if (hdr)
141 printk(KERN_CONT "\n");
142 }
143
144 /*
145 * Just in case we missed an event during initialization check
146 * all the CMCI owned banks.
147 */
148 void cmci_recheck(void)
149 {
150 unsigned long flags;
151 int banks;
152
153 if (!mce_available(&current_cpu_data) || !cmci_supported(&banks))
154 return;
155 local_irq_save(flags);
156 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
157 local_irq_restore(flags);
158 }
159
160 /*
161 * Disable CMCI on this CPU for all banks it owns when it goes down.
162 * This allows other CPUs to claim the banks on rediscovery.
163 */
164 void cmci_clear(void)
165 {
166 unsigned long flags;
167 int i;
168 int banks;
169 u64 val;
170
171 if (!cmci_supported(&banks))
172 return;
173 spin_lock_irqsave(&cmci_discover_lock, flags);
174 for (i = 0; i < banks; i++) {
175 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
176 continue;
177 /* Disable CMCI */
178 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
179 val &= ~(CMCI_EN|CMCI_THRESHOLD_MASK);
180 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
181 __clear_bit(i, __get_cpu_var(mce_banks_owned));
182 }
183 spin_unlock_irqrestore(&cmci_discover_lock, flags);
184 }
185
186 /*
187 * After a CPU went down cycle through all the others and rediscover
188 * Must run in process context.
189 */
190 void cmci_rediscover(int dying)
191 {
192 int banks;
193 int cpu;
194 cpumask_var_t old;
195
196 if (!cmci_supported(&banks))
197 return;
198 if (!alloc_cpumask_var(&old, GFP_KERNEL))
199 return;
200 cpumask_copy(old, &current->cpus_allowed);
201
202 for_each_online_cpu(cpu) {
203 if (cpu == dying)
204 continue;
205 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
206 continue;
207 /* Recheck banks in case CPUs don't all have the same */
208 if (cmci_supported(&banks))
209 cmci_discover(banks, 0);
210 }
211
212 set_cpus_allowed_ptr(current, old);
213 free_cpumask_var(old);
214 }
215
216 /*
217 * Reenable CMCI on this CPU in case a CPU down failed.
218 */
219 void cmci_reenable(void)
220 {
221 int banks;
222 if (cmci_supported(&banks))
223 cmci_discover(banks, 0);
224 }
225
226 static void intel_init_cmci(void)
227 {
228 int banks;
229
230 if (!cmci_supported(&banks))
231 return;
232
233 mce_threshold_vector = intel_threshold_interrupt;
234 cmci_discover(banks, 1);
235 /*
236 * For CPU #0 this runs with still disabled APIC, but that's
237 * ok because only the vector is set up. We still do another
238 * check for the banks later for CPU #0 just to make sure
239 * to not miss any events.
240 */
241 apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
242 cmci_recheck();
243 }
244
245 void mce_intel_feature_init(struct cpuinfo_x86 *c)
246 {
247 intel_init_thermal(c);
248 intel_init_cmci();
249 }