Merge branch 'irq/numa' into x86/mce3
[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 /*
61 * Vendor check is not strictly needed, but the initial
62 * initialization is vendor keyed and this
63 * makes sure none of the backdoors are entered otherwise.
64 */
65 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
66 return 0;
67 if (!cpu_has_apic || lapic_get_maxlvt() < 6)
68 return 0;
69 rdmsrl(MSR_IA32_MCG_CAP, cap);
70 *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
71 return !!(cap & MCG_CMCI_P);
72 }
73
74 /*
75 * The interrupt handler. This is called on every event.
76 * Just call the poller directly to log any events.
77 * This could in theory increase the threshold under high load,
78 * but doesn't for now.
79 */
80 static void intel_threshold_interrupt(void)
81 {
82 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
83 mce_notify_user();
84 }
85
86 static void print_update(char *type, int *hdr, int num)
87 {
88 if (*hdr == 0)
89 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
90 *hdr = 1;
91 printk(KERN_CONT " %s:%d", type, num);
92 }
93
94 /*
95 * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
96 * on this CPU. Use the algorithm recommended in the SDM to discover shared
97 * banks.
98 */
99 static void cmci_discover(int banks, int boot)
100 {
101 unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
102 unsigned long flags;
103 int hdr = 0;
104 int i;
105
106 spin_lock_irqsave(&cmci_discover_lock, flags);
107 for (i = 0; i < banks; i++) {
108 u64 val;
109
110 if (test_bit(i, owned))
111 continue;
112
113 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
114
115 /* Already owned by someone else? */
116 if (val & CMCI_EN) {
117 if (test_and_clear_bit(i, owned) || boot)
118 print_update("SHD", &hdr, i);
119 __clear_bit(i, __get_cpu_var(mce_poll_banks));
120 continue;
121 }
122
123 val |= CMCI_EN | CMCI_THRESHOLD;
124 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
125 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
126
127 /* Did the enable bit stick? -- the bank supports CMCI */
128 if (val & CMCI_EN) {
129 if (!test_and_set_bit(i, owned) || boot)
130 print_update("CMCI", &hdr, i);
131 __clear_bit(i, __get_cpu_var(mce_poll_banks));
132 } else {
133 WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
134 }
135 }
136 spin_unlock_irqrestore(&cmci_discover_lock, flags);
137 if (hdr)
138 printk(KERN_CONT "\n");
139 }
140
141 /*
142 * Just in case we missed an event during initialization check
143 * all the CMCI owned banks.
144 */
145 void cmci_recheck(void)
146 {
147 unsigned long flags;
148 int banks;
149
150 if (!mce_available(&current_cpu_data) || !cmci_supported(&banks))
151 return;
152 local_irq_save(flags);
153 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
154 local_irq_restore(flags);
155 }
156
157 /*
158 * Disable CMCI on this CPU for all banks it owns when it goes down.
159 * This allows other CPUs to claim the banks on rediscovery.
160 */
161 void cmci_clear(void)
162 {
163 unsigned long flags;
164 int i;
165 int banks;
166 u64 val;
167
168 if (!cmci_supported(&banks))
169 return;
170 spin_lock_irqsave(&cmci_discover_lock, flags);
171 for (i = 0; i < banks; i++) {
172 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
173 continue;
174 /* Disable CMCI */
175 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
176 val &= ~(CMCI_EN|CMCI_THRESHOLD_MASK);
177 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
178 __clear_bit(i, __get_cpu_var(mce_banks_owned));
179 }
180 spin_unlock_irqrestore(&cmci_discover_lock, flags);
181 }
182
183 /*
184 * After a CPU went down cycle through all the others and rediscover
185 * Must run in process context.
186 */
187 void cmci_rediscover(int dying)
188 {
189 int banks;
190 int cpu;
191 cpumask_var_t old;
192
193 if (!cmci_supported(&banks))
194 return;
195 if (!alloc_cpumask_var(&old, GFP_KERNEL))
196 return;
197 cpumask_copy(old, &current->cpus_allowed);
198
199 for_each_online_cpu(cpu) {
200 if (cpu == dying)
201 continue;
202 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
203 continue;
204 /* Recheck banks in case CPUs don't all have the same */
205 if (cmci_supported(&banks))
206 cmci_discover(banks, 0);
207 }
208
209 set_cpus_allowed_ptr(current, old);
210 free_cpumask_var(old);
211 }
212
213 /*
214 * Reenable CMCI on this CPU in case a CPU down failed.
215 */
216 void cmci_reenable(void)
217 {
218 int banks;
219 if (cmci_supported(&banks))
220 cmci_discover(banks, 0);
221 }
222
223 static void intel_init_cmci(void)
224 {
225 int banks;
226
227 if (!cmci_supported(&banks))
228 return;
229
230 mce_threshold_vector = intel_threshold_interrupt;
231 cmci_discover(banks, 1);
232 /*
233 * For CPU #0 this runs with still disabled APIC, but that's
234 * ok because only the vector is set up. We still do another
235 * check for the banks later for CPU #0 just to make sure
236 * to not miss any events.
237 */
238 apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
239 cmci_recheck();
240 }
241
242 void mce_intel_feature_init(struct cpuinfo_x86 *c)
243 {
244 intel_init_thermal(c);
245 intel_init_cmci();
246 }