Merge branch 'for-linus' of master.kernel.org:/home/rmk/linux-2.6-arm
[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
19 #include "mce.h"
20
21 asmlinkage void smp_thermal_interrupt(void)
22 {
23 __u64 msr_val;
24
25 ack_APIC_irq();
26
27 exit_idle();
28 irq_enter();
29
30 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
31 if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
32 mce_log_therm_throt_event(msr_val);
33
34 inc_irq_stat(irq_thermal_count);
35 irq_exit();
36 }
37
38 /*
39 * Support for Intel Correct Machine Check Interrupts. This allows
40 * the CPU to raise an interrupt when a corrected machine check happened.
41 * Normally we pick those up using a regular polling timer.
42 * Also supports reliable discovery of shared banks.
43 */
44
45 static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
46
47 /*
48 * cmci_discover_lock protects against parallel discovery attempts
49 * which could race against each other.
50 */
51 static DEFINE_SPINLOCK(cmci_discover_lock);
52
53 #define CMCI_THRESHOLD 1
54
55 static int cmci_supported(int *banks)
56 {
57 u64 cap;
58
59 if (mce_cmci_disabled || mce_ignore_ce)
60 return 0;
61
62 /*
63 * Vendor check is not strictly needed, but the initial
64 * initialization is vendor keyed and this
65 * makes sure none of the backdoors are entered otherwise.
66 */
67 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
68 return 0;
69 if (!cpu_has_apic || lapic_get_maxlvt() < 6)
70 return 0;
71 rdmsrl(MSR_IA32_MCG_CAP, cap);
72 *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff);
73 return !!(cap & MCG_CMCI_P);
74 }
75
76 /*
77 * The interrupt handler. This is called on every event.
78 * Just call the poller directly to log any events.
79 * This could in theory increase the threshold under high load,
80 * but doesn't for now.
81 */
82 static void intel_threshold_interrupt(void)
83 {
84 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
85 mce_notify_irq();
86 }
87
88 static void print_update(char *type, int *hdr, int num)
89 {
90 if (*hdr == 0)
91 printk(KERN_INFO "CPU %d MCA banks", smp_processor_id());
92 *hdr = 1;
93 printk(KERN_CONT " %s:%d", type, num);
94 }
95
96 /*
97 * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks
98 * on this CPU. Use the algorithm recommended in the SDM to discover shared
99 * banks.
100 */
101 static void cmci_discover(int banks, int boot)
102 {
103 unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned);
104 unsigned long flags;
105 int hdr = 0;
106 int i;
107
108 spin_lock_irqsave(&cmci_discover_lock, flags);
109 for (i = 0; i < banks; i++) {
110 u64 val;
111
112 if (test_bit(i, owned))
113 continue;
114
115 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
116
117 /* Already owned by someone else? */
118 if (val & CMCI_EN) {
119 if (test_and_clear_bit(i, owned) || boot)
120 print_update("SHD", &hdr, i);
121 __clear_bit(i, __get_cpu_var(mce_poll_banks));
122 continue;
123 }
124
125 val |= CMCI_EN | CMCI_THRESHOLD;
126 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
127 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
128
129 /* Did the enable bit stick? -- the bank supports CMCI */
130 if (val & CMCI_EN) {
131 if (!test_and_set_bit(i, owned) || boot)
132 print_update("CMCI", &hdr, i);
133 __clear_bit(i, __get_cpu_var(mce_poll_banks));
134 } else {
135 WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks)));
136 }
137 }
138 spin_unlock_irqrestore(&cmci_discover_lock, flags);
139 if (hdr)
140 printk(KERN_CONT "\n");
141 }
142
143 /*
144 * Just in case we missed an event during initialization check
145 * all the CMCI owned banks.
146 */
147 void cmci_recheck(void)
148 {
149 unsigned long flags;
150 int banks;
151
152 if (!mce_available(&current_cpu_data) || !cmci_supported(&banks))
153 return;
154 local_irq_save(flags);
155 machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned));
156 local_irq_restore(flags);
157 }
158
159 /*
160 * Disable CMCI on this CPU for all banks it owns when it goes down.
161 * This allows other CPUs to claim the banks on rediscovery.
162 */
163 void cmci_clear(void)
164 {
165 unsigned long flags;
166 int i;
167 int banks;
168 u64 val;
169
170 if (!cmci_supported(&banks))
171 return;
172 spin_lock_irqsave(&cmci_discover_lock, flags);
173 for (i = 0; i < banks; i++) {
174 if (!test_bit(i, __get_cpu_var(mce_banks_owned)))
175 continue;
176 /* Disable CMCI */
177 rdmsrl(MSR_IA32_MC0_CTL2 + i, val);
178 val &= ~(CMCI_EN|CMCI_THRESHOLD_MASK);
179 wrmsrl(MSR_IA32_MC0_CTL2 + i, val);
180 __clear_bit(i, __get_cpu_var(mce_banks_owned));
181 }
182 spin_unlock_irqrestore(&cmci_discover_lock, flags);
183 }
184
185 /*
186 * After a CPU went down cycle through all the others and rediscover
187 * Must run in process context.
188 */
189 void cmci_rediscover(int dying)
190 {
191 int banks;
192 int cpu;
193 cpumask_var_t old;
194
195 if (!cmci_supported(&banks))
196 return;
197 if (!alloc_cpumask_var(&old, GFP_KERNEL))
198 return;
199 cpumask_copy(old, &current->cpus_allowed);
200
201 for_each_online_cpu(cpu) {
202 if (cpu == dying)
203 continue;
204 if (set_cpus_allowed_ptr(current, cpumask_of(cpu)))
205 continue;
206 /* Recheck banks in case CPUs don't all have the same */
207 if (cmci_supported(&banks))
208 cmci_discover(banks, 0);
209 }
210
211 set_cpus_allowed_ptr(current, old);
212 free_cpumask_var(old);
213 }
214
215 /*
216 * Reenable CMCI on this CPU in case a CPU down failed.
217 */
218 void cmci_reenable(void)
219 {
220 int banks;
221 if (cmci_supported(&banks))
222 cmci_discover(banks, 0);
223 }
224
225 static void intel_init_cmci(void)
226 {
227 int banks;
228
229 if (!cmci_supported(&banks))
230 return;
231
232 mce_threshold_vector = intel_threshold_interrupt;
233 cmci_discover(banks, 1);
234 /*
235 * For CPU #0 this runs with still disabled APIC, but that's
236 * ok because only the vector is set up. We still do another
237 * check for the banks later for CPU #0 just to make sure
238 * to not miss any events.
239 */
240 apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);
241 cmci_recheck();
242 }
243
244 void mce_intel_feature_init(struct cpuinfo_x86 *c)
245 {
246 intel_init_thermal(c);
247 intel_init_cmci();
248 }