Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Intel specific MCE features. | |
3 | * Copyright 2004 Zwane Mwaikambo <zwane@linuxpower.ca> | |
88ccbedd AK |
4 | * Copyright (C) 2008, 2009 Intel Corporation |
5 | * Author: Andi Kleen | |
1da177e4 LT |
6 | */ |
7 | ||
5a0e3ad6 | 8 | #include <linux/gfp.h> |
1da177e4 LT |
9 | #include <linux/interrupt.h> |
10 | #include <linux/percpu.h> | |
d43c36dc | 11 | #include <linux/sched.h> |
1bf7b31e | 12 | #include <asm/apic.h> |
1da177e4 LT |
13 | #include <asm/processor.h> |
14 | #include <asm/msr.h> | |
15 | #include <asm/mce.h> | |
1da177e4 | 16 | |
55babd8f CG |
17 | #include "mce-internal.h" |
18 | ||
88ccbedd AK |
19 | /* |
20 | * Support for Intel Correct Machine Check Interrupts. This allows | |
21 | * the CPU to raise an interrupt when a corrected machine check happened. | |
22 | * Normally we pick those up using a regular polling timer. | |
23 | * Also supports reliable discovery of shared banks. | |
24 | */ | |
25 | ||
0644414e NR |
26 | /* |
27 | * CMCI can be delivered to multiple cpus that share a machine check bank | |
28 | * so we need to designate a single cpu to process errors logged in each bank | |
29 | * in the interrupt handler (otherwise we would have many races and potential | |
30 | * double reporting of the same error). | |
31 | * Note that this can change when a cpu is offlined or brought online since | |
32 | * some MCA banks are shared across cpus. When a cpu is offlined, cmci_clear() | |
33 | * disables CMCI on all banks owned by the cpu and clears this bitfield. At | |
34 | * this point, cmci_rediscover() kicks in and a different cpu may end up | |
35 | * taking ownership of some of the shared MCA banks that were previously | |
36 | * owned by the offlined cpu. | |
37 | */ | |
88ccbedd AK |
38 | static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); |
39 | ||
40 | /* | |
41 | * cmci_discover_lock protects against parallel discovery attempts | |
42 | * which could race against each other. | |
43 | */ | |
59d958d2 | 44 | static DEFINE_RAW_SPINLOCK(cmci_discover_lock); |
88ccbedd | 45 | |
55babd8f CG |
46 | #define CMCI_THRESHOLD 1 |
47 | #define CMCI_POLL_INTERVAL (30 * HZ) | |
48 | #define CMCI_STORM_INTERVAL (1 * HZ) | |
49 | #define CMCI_STORM_THRESHOLD 15 | |
50 | ||
51 | static DEFINE_PER_CPU(unsigned long, cmci_time_stamp); | |
52 | static DEFINE_PER_CPU(unsigned int, cmci_storm_cnt); | |
53 | static DEFINE_PER_CPU(unsigned int, cmci_storm_state); | |
54 | ||
55 | enum { | |
56 | CMCI_STORM_NONE, | |
57 | CMCI_STORM_ACTIVE, | |
58 | CMCI_STORM_SUBSIDED, | |
59 | }; | |
60 | ||
61 | static atomic_t cmci_storm_on_cpus; | |
88ccbedd | 62 | |
df20e2eb | 63 | static int cmci_supported(int *banks) |
88ccbedd AK |
64 | { |
65 | u64 cap; | |
66 | ||
7af19e4a | 67 | if (mca_cfg.cmci_disabled || mca_cfg.ignore_ce) |
62fdac59 HS |
68 | return 0; |
69 | ||
88ccbedd AK |
70 | /* |
71 | * Vendor check is not strictly needed, but the initial | |
72 | * initialization is vendor keyed and this | |
73 | * makes sure none of the backdoors are entered otherwise. | |
74 | */ | |
75 | if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) | |
76 | return 0; | |
77 | if (!cpu_has_apic || lapic_get_maxlvt() < 6) | |
78 | return 0; | |
79 | rdmsrl(MSR_IA32_MCG_CAP, cap); | |
80 | *banks = min_t(unsigned, MAX_NR_BANKS, cap & 0xff); | |
81 | return !!(cap & MCG_CMCI_P); | |
82 | } | |
83 | ||
55babd8f CG |
84 | void mce_intel_cmci_poll(void) |
85 | { | |
86 | if (__this_cpu_read(cmci_storm_state) == CMCI_STORM_NONE) | |
87 | return; | |
88 | machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned)); | |
89 | } | |
90 | ||
91 | void mce_intel_hcpu_update(unsigned long cpu) | |
92 | { | |
93 | if (per_cpu(cmci_storm_state, cpu) == CMCI_STORM_ACTIVE) | |
94 | atomic_dec(&cmci_storm_on_cpus); | |
95 | ||
96 | per_cpu(cmci_storm_state, cpu) = CMCI_STORM_NONE; | |
97 | } | |
98 | ||
99 | unsigned long mce_intel_adjust_timer(unsigned long interval) | |
100 | { | |
101 | int r; | |
102 | ||
103 | if (interval < CMCI_POLL_INTERVAL) | |
104 | return interval; | |
105 | ||
106 | switch (__this_cpu_read(cmci_storm_state)) { | |
107 | case CMCI_STORM_ACTIVE: | |
108 | /* | |
109 | * We switch back to interrupt mode once the poll timer has | |
110 | * silenced itself. That means no events recorded and the | |
111 | * timer interval is back to our poll interval. | |
112 | */ | |
113 | __this_cpu_write(cmci_storm_state, CMCI_STORM_SUBSIDED); | |
114 | r = atomic_sub_return(1, &cmci_storm_on_cpus); | |
115 | if (r == 0) | |
116 | pr_notice("CMCI storm subsided: switching to interrupt mode\n"); | |
117 | /* FALLTHROUGH */ | |
118 | ||
119 | case CMCI_STORM_SUBSIDED: | |
120 | /* | |
121 | * We wait for all cpus to go back to SUBSIDED | |
122 | * state. When that happens we switch back to | |
123 | * interrupt mode. | |
124 | */ | |
125 | if (!atomic_read(&cmci_storm_on_cpus)) { | |
126 | __this_cpu_write(cmci_storm_state, CMCI_STORM_NONE); | |
127 | cmci_reenable(); | |
128 | cmci_recheck(); | |
129 | } | |
130 | return CMCI_POLL_INTERVAL; | |
131 | default: | |
132 | /* | |
133 | * We have shiny weather. Let the poll do whatever it | |
134 | * thinks. | |
135 | */ | |
136 | return interval; | |
137 | } | |
138 | } | |
139 | ||
140 | static bool cmci_storm_detect(void) | |
141 | { | |
142 | unsigned int cnt = __this_cpu_read(cmci_storm_cnt); | |
143 | unsigned long ts = __this_cpu_read(cmci_time_stamp); | |
144 | unsigned long now = jiffies; | |
145 | int r; | |
146 | ||
147 | if (__this_cpu_read(cmci_storm_state) != CMCI_STORM_NONE) | |
148 | return true; | |
149 | ||
150 | if (time_before_eq(now, ts + CMCI_STORM_INTERVAL)) { | |
151 | cnt++; | |
152 | } else { | |
153 | cnt = 1; | |
154 | __this_cpu_write(cmci_time_stamp, now); | |
155 | } | |
156 | __this_cpu_write(cmci_storm_cnt, cnt); | |
157 | ||
158 | if (cnt <= CMCI_STORM_THRESHOLD) | |
159 | return false; | |
160 | ||
161 | cmci_clear(); | |
162 | __this_cpu_write(cmci_storm_state, CMCI_STORM_ACTIVE); | |
163 | r = atomic_add_return(1, &cmci_storm_on_cpus); | |
164 | mce_timer_kick(CMCI_POLL_INTERVAL); | |
165 | ||
166 | if (r == 1) | |
167 | pr_notice("CMCI storm detected: switching to poll mode\n"); | |
168 | return true; | |
169 | } | |
170 | ||
88ccbedd AK |
171 | /* |
172 | * The interrupt handler. This is called on every event. | |
173 | * Just call the poller directly to log any events. | |
174 | * This could in theory increase the threshold under high load, | |
175 | * but doesn't for now. | |
176 | */ | |
177 | static void intel_threshold_interrupt(void) | |
178 | { | |
55babd8f CG |
179 | if (cmci_storm_detect()) |
180 | return; | |
88ccbedd | 181 | machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned)); |
9ff36ee9 | 182 | mce_notify_irq(); |
88ccbedd AK |
183 | } |
184 | ||
88ccbedd AK |
185 | /* |
186 | * Enable CMCI (Corrected Machine Check Interrupt) for available MCE banks | |
187 | * on this CPU. Use the algorithm recommended in the SDM to discover shared | |
188 | * banks. | |
189 | */ | |
4670a300 | 190 | static void cmci_discover(int banks) |
88ccbedd AK |
191 | { |
192 | unsigned long *owned = (void *)&__get_cpu_var(mce_banks_owned); | |
e5299926 | 193 | unsigned long flags; |
88ccbedd | 194 | int i; |
450cc201 | 195 | int bios_wrong_thresh = 0; |
88ccbedd | 196 | |
59d958d2 | 197 | raw_spin_lock_irqsave(&cmci_discover_lock, flags); |
88ccbedd AK |
198 | for (i = 0; i < banks; i++) { |
199 | u64 val; | |
450cc201 | 200 | int bios_zero_thresh = 0; |
88ccbedd AK |
201 | |
202 | if (test_bit(i, owned)) | |
203 | continue; | |
204 | ||
c3d1fb56 NR |
205 | /* Skip banks in firmware first mode */ |
206 | if (test_bit(i, mce_banks_ce_disabled)) | |
207 | continue; | |
208 | ||
a2d32bcb | 209 | rdmsrl(MSR_IA32_MCx_CTL2(i), val); |
88ccbedd AK |
210 | |
211 | /* Already owned by someone else? */ | |
1f9a0bd4 | 212 | if (val & MCI_CTL2_CMCI_EN) { |
4670a300 | 213 | clear_bit(i, owned); |
88ccbedd AK |
214 | __clear_bit(i, __get_cpu_var(mce_poll_banks)); |
215 | continue; | |
216 | } | |
217 | ||
1462594b | 218 | if (!mca_cfg.bios_cmci_threshold) { |
450cc201 NR |
219 | val &= ~MCI_CTL2_CMCI_THRESHOLD_MASK; |
220 | val |= CMCI_THRESHOLD; | |
221 | } else if (!(val & MCI_CTL2_CMCI_THRESHOLD_MASK)) { | |
222 | /* | |
223 | * If bios_cmci_threshold boot option was specified | |
224 | * but the threshold is zero, we'll try to initialize | |
225 | * it to 1. | |
226 | */ | |
227 | bios_zero_thresh = 1; | |
228 | val |= CMCI_THRESHOLD; | |
229 | } | |
230 | ||
231 | val |= MCI_CTL2_CMCI_EN; | |
a2d32bcb AK |
232 | wrmsrl(MSR_IA32_MCx_CTL2(i), val); |
233 | rdmsrl(MSR_IA32_MCx_CTL2(i), val); | |
88ccbedd AK |
234 | |
235 | /* Did the enable bit stick? -- the bank supports CMCI */ | |
1f9a0bd4 | 236 | if (val & MCI_CTL2_CMCI_EN) { |
4670a300 | 237 | set_bit(i, owned); |
88ccbedd | 238 | __clear_bit(i, __get_cpu_var(mce_poll_banks)); |
450cc201 NR |
239 | /* |
240 | * We are able to set thresholds for some banks that | |
241 | * had a threshold of 0. This means the BIOS has not | |
242 | * set the thresholds properly or does not work with | |
243 | * this boot option. Note down now and report later. | |
244 | */ | |
1462594b | 245 | if (mca_cfg.bios_cmci_threshold && bios_zero_thresh && |
450cc201 NR |
246 | (val & MCI_CTL2_CMCI_THRESHOLD_MASK)) |
247 | bios_wrong_thresh = 1; | |
88ccbedd AK |
248 | } else { |
249 | WARN_ON(!test_bit(i, __get_cpu_var(mce_poll_banks))); | |
250 | } | |
251 | } | |
59d958d2 | 252 | raw_spin_unlock_irqrestore(&cmci_discover_lock, flags); |
1462594b | 253 | if (mca_cfg.bios_cmci_threshold && bios_wrong_thresh) { |
450cc201 NR |
254 | pr_info_once( |
255 | "bios_cmci_threshold: Some banks do not have valid thresholds set\n"); | |
256 | pr_info_once( | |
257 | "bios_cmci_threshold: Make sure your BIOS supports this boot option\n"); | |
258 | } | |
88ccbedd AK |
259 | } |
260 | ||
261 | /* | |
262 | * Just in case we missed an event during initialization check | |
263 | * all the CMCI owned banks. | |
264 | */ | |
df20e2eb | 265 | void cmci_recheck(void) |
88ccbedd AK |
266 | { |
267 | unsigned long flags; | |
268 | int banks; | |
269 | ||
7b543a53 | 270 | if (!mce_available(__this_cpu_ptr(&cpu_info)) || !cmci_supported(&banks)) |
88ccbedd AK |
271 | return; |
272 | local_irq_save(flags); | |
273 | machine_check_poll(MCP_TIMESTAMP, &__get_cpu_var(mce_banks_owned)); | |
274 | local_irq_restore(flags); | |
275 | } | |
276 | ||
c3d1fb56 NR |
277 | /* Caller must hold the lock on cmci_discover_lock */ |
278 | static void __cmci_disable_bank(int bank) | |
279 | { | |
280 | u64 val; | |
281 | ||
282 | if (!test_bit(bank, __get_cpu_var(mce_banks_owned))) | |
283 | return; | |
284 | rdmsrl(MSR_IA32_MCx_CTL2(bank), val); | |
285 | val &= ~MCI_CTL2_CMCI_EN; | |
286 | wrmsrl(MSR_IA32_MCx_CTL2(bank), val); | |
287 | __clear_bit(bank, __get_cpu_var(mce_banks_owned)); | |
288 | } | |
289 | ||
88ccbedd AK |
290 | /* |
291 | * Disable CMCI on this CPU for all banks it owns when it goes down. | |
292 | * This allows other CPUs to claim the banks on rediscovery. | |
293 | */ | |
df20e2eb | 294 | void cmci_clear(void) |
88ccbedd | 295 | { |
e5299926 | 296 | unsigned long flags; |
88ccbedd AK |
297 | int i; |
298 | int banks; | |
88ccbedd AK |
299 | |
300 | if (!cmci_supported(&banks)) | |
301 | return; | |
59d958d2 | 302 | raw_spin_lock_irqsave(&cmci_discover_lock, flags); |
c3d1fb56 NR |
303 | for (i = 0; i < banks; i++) |
304 | __cmci_disable_bank(i); | |
59d958d2 | 305 | raw_spin_unlock_irqrestore(&cmci_discover_lock, flags); |
88ccbedd AK |
306 | } |
307 | ||
7a0c819d | 308 | static void cmci_rediscover_work_func(void *arg) |
85b97637 TC |
309 | { |
310 | int banks; | |
311 | ||
312 | /* Recheck banks in case CPUs don't all have the same */ | |
313 | if (cmci_supported(&banks)) | |
314 | cmci_discover(banks); | |
85b97637 TC |
315 | } |
316 | ||
7a0c819d SB |
317 | /* After a CPU went down cycle through all the others and rediscover */ |
318 | void cmci_rediscover(void) | |
88ccbedd | 319 | { |
7a0c819d | 320 | int banks; |
88ccbedd AK |
321 | |
322 | if (!cmci_supported(&banks)) | |
323 | return; | |
88ccbedd | 324 | |
7a0c819d | 325 | on_each_cpu(cmci_rediscover_work_func, NULL, 1); |
88ccbedd AK |
326 | } |
327 | ||
328 | /* | |
329 | * Reenable CMCI on this CPU in case a CPU down failed. | |
330 | */ | |
331 | void cmci_reenable(void) | |
332 | { | |
333 | int banks; | |
334 | if (cmci_supported(&banks)) | |
4670a300 | 335 | cmci_discover(banks); |
88ccbedd AK |
336 | } |
337 | ||
c3d1fb56 NR |
338 | void cmci_disable_bank(int bank) |
339 | { | |
340 | int banks; | |
341 | unsigned long flags; | |
342 | ||
343 | if (!cmci_supported(&banks)) | |
344 | return; | |
345 | ||
346 | raw_spin_lock_irqsave(&cmci_discover_lock, flags); | |
347 | __cmci_disable_bank(bank); | |
348 | raw_spin_unlock_irqrestore(&cmci_discover_lock, flags); | |
349 | } | |
350 | ||
514ec49a | 351 | static void intel_init_cmci(void) |
88ccbedd AK |
352 | { |
353 | int banks; | |
354 | ||
355 | if (!cmci_supported(&banks)) | |
356 | return; | |
357 | ||
358 | mce_threshold_vector = intel_threshold_interrupt; | |
4670a300 | 359 | cmci_discover(banks); |
88ccbedd AK |
360 | /* |
361 | * For CPU #0 this runs with still disabled APIC, but that's | |
362 | * ok because only the vector is set up. We still do another | |
363 | * check for the banks later for CPU #0 just to make sure | |
364 | * to not miss any events. | |
365 | */ | |
366 | apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED); | |
367 | cmci_recheck(); | |
368 | } | |
369 | ||
cc3ca220 | 370 | void mce_intel_feature_init(struct cpuinfo_x86 *c) |
1da177e4 LT |
371 | { |
372 | intel_init_thermal(c); | |
88ccbedd | 373 | intel_init_cmci(); |
1da177e4 | 374 | } |