1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2001 Dave Engebretsen IBM Corporation
4 */
5
6 #include <linux/sched.h>
7 #include <linux/interrupt.h>
8 #include <linux/irq.h>
9 #include <linux/of.h>
10 #include <linux/fs.h>
11 #include <linux/reboot.h>
12 #include <linux/irq_work.h>
13
14 #include <asm/machdep.h>
15 #include <asm/rtas.h>
16 #include <asm/firmware.h>
17 #include <asm/mce.h>
18
19 #include "pseries.h"
20
21 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
22 static DEFINE_SPINLOCK(ras_log_buf_lock);
23
24 static int ras_check_exception_token;
25
26 static void mce_process_errlog_event(struct irq_work *work);
27 static struct irq_work mce_errlog_process_work = {
28 .func = mce_process_errlog_event,
29 };
30
31 #define EPOW_SENSOR_TOKEN 9
32 #define EPOW_SENSOR_INDEX 0
33
34 /* EPOW events counter variable */
35 static int num_epow_events;
36
37 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id);
38 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
39 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
40
41 /* RTAS pseries MCE errorlog section. */
42 struct pseries_mc_errorlog {
43 __be32 fru_id;
44 __be32 proc_id;
45 u8 error_type;
46 /*
47 * sub_err_type (1 byte). Bit fields depends on error_type
48 *
49 * MSB0
50 * |
51 * V
52 * 01234567
53 * XXXXXXXX
54 *
55 * For error_type == MC_ERROR_TYPE_UE
56 * XXXXXXXX
57 * X 1: Permanent or Transient UE.
58 * X 1: Effective address provided.
59 * X 1: Logical address provided.
60 * XX 2: Reserved.
61 * XXX 3: Type of UE error.
62 *
63 * For error_type != MC_ERROR_TYPE_UE
64 * XXXXXXXX
65 * X 1: Effective address provided.
66 * XXXXX 5: Reserved.
67 * XX 2: Type of SLB/ERAT/TLB error.
68 */
69 u8 sub_err_type;
70 u8 reserved_1[6];
71 __be64 effective_address;
72 __be64 logical_address;
73 } __packed;
74
75 /* RTAS pseries MCE error types */
76 #define MC_ERROR_TYPE_UE 0x00
77 #define MC_ERROR_TYPE_SLB 0x01
78 #define MC_ERROR_TYPE_ERAT 0x02
79 #define MC_ERROR_TYPE_UNKNOWN 0x03
80 #define MC_ERROR_TYPE_TLB 0x04
81 #define MC_ERROR_TYPE_D_CACHE 0x05
82 #define MC_ERROR_TYPE_I_CACHE 0x07
83
84 /* RTAS pseries MCE error sub types */
85 #define MC_ERROR_UE_INDETERMINATE 0
86 #define MC_ERROR_UE_IFETCH 1
87 #define MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH 2
88 #define MC_ERROR_UE_LOAD_STORE 3
89 #define MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE 4
90
91 #define UE_EFFECTIVE_ADDR_PROVIDED 0x40
92 #define UE_LOGICAL_ADDR_PROVIDED 0x20
93
94 #define MC_ERROR_SLB_PARITY 0
95 #define MC_ERROR_SLB_MULTIHIT 1
96 #define MC_ERROR_SLB_INDETERMINATE 2
97
98 #define MC_ERROR_ERAT_PARITY 1
99 #define MC_ERROR_ERAT_MULTIHIT 2
100 #define MC_ERROR_ERAT_INDETERMINATE 3
101
102 #define MC_ERROR_TLB_PARITY 1
103 #define MC_ERROR_TLB_MULTIHIT 2
104 #define MC_ERROR_TLB_INDETERMINATE 3
105
rtas_mc_error_sub_type(const struct pseries_mc_errorlog * mlog)106 static inline u8 rtas_mc_error_sub_type(const struct pseries_mc_errorlog *mlog)
107 {
108 switch (mlog->error_type) {
109 case MC_ERROR_TYPE_UE:
110 return (mlog->sub_err_type & 0x07);
111 case MC_ERROR_TYPE_SLB:
112 case MC_ERROR_TYPE_ERAT:
113 case MC_ERROR_TYPE_TLB:
114 return (mlog->sub_err_type & 0x03);
115 default:
116 return 0;
117 }
118 }
119
120 /*
121 * Enable the hotplug interrupt late because processing them may touch other
122 * devices or systems (e.g. hugepages) that have not been initialized at the
123 * subsys stage.
124 */
init_ras_hotplug_IRQ(void)125 static int __init init_ras_hotplug_IRQ(void)
126 {
127 struct device_node *np;
128
129 /* Hotplug Events */
130 np = of_find_node_by_path("/event-sources/hot-plug-events");
131 if (np != NULL) {
132 if (dlpar_workqueue_init() == 0)
133 request_event_sources_irqs(np, ras_hotplug_interrupt,
134 "RAS_HOTPLUG");
135 of_node_put(np);
136 }
137
138 return 0;
139 }
140 machine_late_initcall(pseries, init_ras_hotplug_IRQ);
141
142 /*
143 * Initialize handlers for the set of interrupts caused by hardware errors
144 * and power system events.
145 */
init_ras_IRQ(void)146 static int __init init_ras_IRQ(void)
147 {
148 struct device_node *np;
149
150 ras_check_exception_token = rtas_token("check-exception");
151
152 /* Internal Errors */
153 np = of_find_node_by_path("/event-sources/internal-errors");
154 if (np != NULL) {
155 request_event_sources_irqs(np, ras_error_interrupt,
156 "RAS_ERROR");
157 of_node_put(np);
158 }
159
160 /* EPOW Events */
161 np = of_find_node_by_path("/event-sources/epow-events");
162 if (np != NULL) {
163 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW");
164 of_node_put(np);
165 }
166
167 return 0;
168 }
169 machine_subsys_initcall(pseries, init_ras_IRQ);
170
171 #define EPOW_SHUTDOWN_NORMAL 1
172 #define EPOW_SHUTDOWN_ON_UPS 2
173 #define EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS 3
174 #define EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH 4
175
handle_system_shutdown(char event_modifier)176 static void handle_system_shutdown(char event_modifier)
177 {
178 switch (event_modifier) {
179 case EPOW_SHUTDOWN_NORMAL:
180 pr_emerg("Power off requested\n");
181 orderly_poweroff(true);
182 break;
183
184 case EPOW_SHUTDOWN_ON_UPS:
185 pr_emerg("Loss of system power detected. System is running on"
186 " UPS/battery. Check RTAS error log for details\n");
187 break;
188
189 case EPOW_SHUTDOWN_LOSS_OF_CRITICAL_FUNCTIONS:
190 pr_emerg("Loss of system critical functions detected. Check"
191 " RTAS error log for details\n");
192 orderly_poweroff(true);
193 break;
194
195 case EPOW_SHUTDOWN_AMBIENT_TEMPERATURE_TOO_HIGH:
196 pr_emerg("High ambient temperature detected. Check RTAS"
197 " error log for details\n");
198 orderly_poweroff(true);
199 break;
200
201 default:
202 pr_err("Unknown power/cooling shutdown event (modifier = %d)\n",
203 event_modifier);
204 }
205 }
206
207 struct epow_errorlog {
208 unsigned char sensor_value;
209 unsigned char event_modifier;
210 unsigned char extended_modifier;
211 unsigned char reserved;
212 unsigned char platform_reason;
213 };
214
215 #define EPOW_RESET 0
216 #define EPOW_WARN_COOLING 1
217 #define EPOW_WARN_POWER 2
218 #define EPOW_SYSTEM_SHUTDOWN 3
219 #define EPOW_SYSTEM_HALT 4
220 #define EPOW_MAIN_ENCLOSURE 5
221 #define EPOW_POWER_OFF 7
222
rtas_parse_epow_errlog(struct rtas_error_log * log)223 static void rtas_parse_epow_errlog(struct rtas_error_log *log)
224 {
225 struct pseries_errorlog *pseries_log;
226 struct epow_errorlog *epow_log;
227 char action_code;
228 char modifier;
229
230 pseries_log = get_pseries_errorlog(log, PSERIES_ELOG_SECT_ID_EPOW);
231 if (pseries_log == NULL)
232 return;
233
234 epow_log = (struct epow_errorlog *)pseries_log->data;
235 action_code = epow_log->sensor_value & 0xF; /* bottom 4 bits */
236 modifier = epow_log->event_modifier & 0xF; /* bottom 4 bits */
237
238 switch (action_code) {
239 case EPOW_RESET:
240 if (num_epow_events) {
241 pr_info("Non critical power/cooling issue cleared\n");
242 num_epow_events--;
243 }
244 break;
245
246 case EPOW_WARN_COOLING:
247 pr_info("Non-critical cooling issue detected. Check RTAS error"
248 " log for details\n");
249 break;
250
251 case EPOW_WARN_POWER:
252 pr_info("Non-critical power issue detected. Check RTAS error"
253 " log for details\n");
254 break;
255
256 case EPOW_SYSTEM_SHUTDOWN:
257 handle_system_shutdown(modifier);
258 break;
259
260 case EPOW_SYSTEM_HALT:
261 pr_emerg("Critical power/cooling issue detected. Check RTAS"
262 " error log for details. Powering off.\n");
263 orderly_poweroff(true);
264 break;
265
266 case EPOW_MAIN_ENCLOSURE:
267 case EPOW_POWER_OFF:
268 pr_emerg("System about to lose power. Check RTAS error log "
269 " for details. Powering off immediately.\n");
270 emergency_sync();
271 kernel_power_off();
272 break;
273
274 default:
275 pr_err("Unknown power/cooling event (action code = %d)\n",
276 action_code);
277 }
278
279 /* Increment epow events counter variable */
280 if (action_code != EPOW_RESET)
281 num_epow_events++;
282 }
283
ras_hotplug_interrupt(int irq,void * dev_id)284 static irqreturn_t ras_hotplug_interrupt(int irq, void *dev_id)
285 {
286 struct pseries_errorlog *pseries_log;
287 struct pseries_hp_errorlog *hp_elog;
288
289 spin_lock(&ras_log_buf_lock);
290
291 rtas_call(ras_check_exception_token, 6, 1, NULL,
292 RTAS_VECTOR_EXTERNAL_INTERRUPT, virq_to_hw(irq),
293 RTAS_HOTPLUG_EVENTS, 0, __pa(&ras_log_buf),
294 rtas_get_error_log_max());
295
296 pseries_log = get_pseries_errorlog((struct rtas_error_log *)ras_log_buf,
297 PSERIES_ELOG_SECT_ID_HOTPLUG);
298 hp_elog = (struct pseries_hp_errorlog *)pseries_log->data;
299
300 /*
301 * Since PCI hotplug is not currently supported on pseries, put PCI
302 * hotplug events on the ras_log_buf to be handled by rtas_errd.
303 */
304 if (hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_MEM ||
305 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_CPU ||
306 hp_elog->resource == PSERIES_HP_ELOG_RESOURCE_PMEM)
307 queue_hotplug_event(hp_elog);
308 else
309 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
310
311 spin_unlock(&ras_log_buf_lock);
312 return IRQ_HANDLED;
313 }
314
315 /* Handle environmental and power warning (EPOW) interrupts. */
ras_epow_interrupt(int irq,void * dev_id)316 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
317 {
318 int state;
319 int critical;
320
321 rtas_get_sensor_fast(EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX, &state);
322
323 if (state > 3)
324 critical = 1; /* Time Critical */
325 else
326 critical = 0;
327
328 spin_lock(&ras_log_buf_lock);
329
330 rtas_call(ras_check_exception_token, 6, 1, NULL, RTAS_VECTOR_EXTERNAL_INTERRUPT,
331 virq_to_hw(irq), RTAS_EPOW_WARNING, critical, __pa(&ras_log_buf),
332 rtas_get_error_log_max());
333
334 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
335
336 rtas_parse_epow_errlog((struct rtas_error_log *)ras_log_buf);
337
338 spin_unlock(&ras_log_buf_lock);
339 return IRQ_HANDLED;
340 }
341
342 /*
343 * Handle hardware error interrupts.
344 *
345 * RTAS check-exception is called to collect data on the exception. If
346 * the error is deemed recoverable, we log a warning and return.
347 * For nonrecoverable errors, an error is logged and we stop all processing
348 * as quickly as possible in order to prevent propagation of the failure.
349 */
ras_error_interrupt(int irq,void * dev_id)350 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
351 {
352 struct rtas_error_log *rtas_elog;
353 int status;
354 int fatal;
355
356 spin_lock(&ras_log_buf_lock);
357
358 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
359 RTAS_VECTOR_EXTERNAL_INTERRUPT,
360 virq_to_hw(irq),
361 RTAS_INTERNAL_ERROR, 1 /* Time Critical */,
362 __pa(&ras_log_buf),
363 rtas_get_error_log_max());
364
365 rtas_elog = (struct rtas_error_log *)ras_log_buf;
366
367 if (status == 0 &&
368 rtas_error_severity(rtas_elog) >= RTAS_SEVERITY_ERROR_SYNC)
369 fatal = 1;
370 else
371 fatal = 0;
372
373 /* format and print the extended information */
374 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
375
376 if (fatal) {
377 pr_emerg("Fatal hardware error detected. Check RTAS error"
378 " log for details. Powering off immediately\n");
379 emergency_sync();
380 kernel_power_off();
381 } else {
382 pr_err("Recoverable hardware error detected\n");
383 }
384
385 spin_unlock(&ras_log_buf_lock);
386 return IRQ_HANDLED;
387 }
388
389 /*
390 * Some versions of FWNMI place the buffer inside the 4kB page starting at
391 * 0x7000. Other versions place it inside the rtas buffer. We check both.
392 * Minimum size of the buffer is 16 bytes.
393 */
394 #define VALID_FWNMI_BUFFER(A) \
395 ((((A) >= 0x7000) && ((A) <= 0x8000 - 16)) || \
396 (((A) >= rtas.base) && ((A) <= (rtas.base + rtas.size - 16))))
397
fwnmi_get_errlog(void)398 static inline struct rtas_error_log *fwnmi_get_errlog(void)
399 {
400 return (struct rtas_error_log *)local_paca->mce_data_buf;
401 }
402
fwnmi_get_savep(struct pt_regs * regs)403 static __be64 *fwnmi_get_savep(struct pt_regs *regs)
404 {
405 unsigned long savep_ra;
406
407 /* Mask top two bits */
408 savep_ra = regs->gpr[3] & ~(0x3UL << 62);
409 if (!VALID_FWNMI_BUFFER(savep_ra)) {
410 printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]);
411 return NULL;
412 }
413
414 return __va(savep_ra);
415 }
416
417 /*
418 * Get the error information for errors coming through the
419 * FWNMI vectors. The pt_regs' r3 will be updated to reflect
420 * the actual r3 if possible, and a ptr to the error log entry
421 * will be returned if found.
422 *
423 * Use one buffer mce_data_buf per cpu to store RTAS error.
424 *
425 * The mce_data_buf does not have any locks or protection around it,
426 * if a second machine check comes in, or a system reset is done
427 * before we have logged the error, then we will get corruption in the
428 * error log. This is preferable over holding off on calling
429 * ibm,nmi-interlock which would result in us checkstopping if a
430 * second machine check did come in.
431 */
fwnmi_get_errinfo(struct pt_regs * regs)432 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
433 {
434 struct rtas_error_log *h;
435 __be64 *savep;
436
437 savep = fwnmi_get_savep(regs);
438 if (!savep)
439 return NULL;
440
441 regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */
442
443 h = (struct rtas_error_log *)&savep[1];
444 /* Use the per cpu buffer from paca to store rtas error log */
445 memset(local_paca->mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
446 if (!rtas_error_extended(h)) {
447 memcpy(local_paca->mce_data_buf, h, sizeof(__u64));
448 } else {
449 int len, error_log_length;
450
451 error_log_length = 8 + rtas_error_extended_log_length(h);
452 len = min_t(int, error_log_length, RTAS_ERROR_LOG_MAX);
453 memcpy(local_paca->mce_data_buf, h, len);
454 }
455
456 return (struct rtas_error_log *)local_paca->mce_data_buf;
457 }
458
459 /* Call this when done with the data returned by FWNMI_get_errinfo.
460 * It will release the saved data area for other CPUs in the
461 * partition to receive FWNMI errors.
462 */
fwnmi_release_errinfo(void)463 static void fwnmi_release_errinfo(void)
464 {
465 struct rtas_args rtas_args;
466 int ret;
467
468 /*
469 * On pseries, the machine check stack is limited to under 4GB, so
470 * args can be on-stack.
471 */
472 rtas_call_unlocked(&rtas_args, ibm_nmi_interlock_token, 0, 1, NULL);
473 ret = be32_to_cpu(rtas_args.rets[0]);
474 if (ret != 0)
475 printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret);
476 }
477
pSeries_system_reset_exception(struct pt_regs * regs)478 int pSeries_system_reset_exception(struct pt_regs *regs)
479 {
480 #ifdef __LITTLE_ENDIAN__
481 /*
482 * Some firmware byteswaps SRR registers and gives incorrect SRR1. Try
483 * to detect the bad SRR1 pattern here. Flip the NIP back to correct
484 * endian for reporting purposes. Unfortunately the MSR can't be fixed,
485 * so clear it. It will be missing MSR_RI so we won't try to recover.
486 */
487 if ((be64_to_cpu(regs->msr) &
488 (MSR_LE|MSR_RI|MSR_DR|MSR_IR|MSR_ME|MSR_PR|
489 MSR_ILE|MSR_HV|MSR_SF)) == (MSR_DR|MSR_SF)) {
490 regs_set_return_ip(regs, be64_to_cpu((__be64)regs->nip));
491 regs_set_return_msr(regs, 0);
492 }
493 #endif
494
495 if (fwnmi_active) {
496 __be64 *savep;
497
498 /*
499 * Firmware (PowerVM and KVM) saves r3 to a save area like
500 * machine check, which is not exactly what PAPR (2.9)
501 * suggests but there is no way to detect otherwise, so this
502 * is the interface now.
503 *
504 * System resets do not save any error log or require an
505 * "ibm,nmi-interlock" rtas call to release.
506 */
507
508 savep = fwnmi_get_savep(regs);
509 if (savep)
510 regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */
511 }
512
513 if (smp_handle_nmi_ipi(regs))
514 return 1;
515
516 return 0; /* need to perform reset */
517 }
518
mce_handle_err_realmode(int disposition,u8 error_type)519 static int mce_handle_err_realmode(int disposition, u8 error_type)
520 {
521 #ifdef CONFIG_PPC_BOOK3S_64
522 if (disposition == RTAS_DISP_NOT_RECOVERED) {
523 switch (error_type) {
524 case MC_ERROR_TYPE_ERAT:
525 flush_erat();
526 disposition = RTAS_DISP_FULLY_RECOVERED;
527 break;
528 case MC_ERROR_TYPE_SLB:
529 /*
530 * Store the old slb content in paca before flushing.
531 * Print this when we go to virtual mode.
532 * There are chances that we may hit MCE again if there
533 * is a parity error on the SLB entry we trying to read
534 * for saving. Hence limit the slb saving to single
535 * level of recursion.
536 */
537 if (local_paca->in_mce == 1)
538 slb_save_contents(local_paca->mce_faulty_slbs);
539 flush_and_reload_slb();
540 disposition = RTAS_DISP_FULLY_RECOVERED;
541 break;
542 default:
543 break;
544 }
545 } else if (disposition == RTAS_DISP_LIMITED_RECOVERY) {
546 /* Platform corrected itself but could be degraded */
547 pr_err("MCE: limited recovery, system may be degraded\n");
548 disposition = RTAS_DISP_FULLY_RECOVERED;
549 }
550 #endif
551 return disposition;
552 }
553
mce_handle_err_virtmode(struct pt_regs * regs,struct rtas_error_log * errp,struct pseries_mc_errorlog * mce_log,int disposition)554 static int mce_handle_err_virtmode(struct pt_regs *regs,
555 struct rtas_error_log *errp,
556 struct pseries_mc_errorlog *mce_log,
557 int disposition)
558 {
559 struct mce_error_info mce_err = { 0 };
560 int initiator = rtas_error_initiator(errp);
561 int severity = rtas_error_severity(errp);
562 unsigned long eaddr = 0, paddr = 0;
563 u8 error_type, err_sub_type;
564
565 if (!mce_log)
566 goto out;
567
568 error_type = mce_log->error_type;
569 err_sub_type = rtas_mc_error_sub_type(mce_log);
570
571 if (initiator == RTAS_INITIATOR_UNKNOWN)
572 mce_err.initiator = MCE_INITIATOR_UNKNOWN;
573 else if (initiator == RTAS_INITIATOR_CPU)
574 mce_err.initiator = MCE_INITIATOR_CPU;
575 else if (initiator == RTAS_INITIATOR_PCI)
576 mce_err.initiator = MCE_INITIATOR_PCI;
577 else if (initiator == RTAS_INITIATOR_ISA)
578 mce_err.initiator = MCE_INITIATOR_ISA;
579 else if (initiator == RTAS_INITIATOR_MEMORY)
580 mce_err.initiator = MCE_INITIATOR_MEMORY;
581 else if (initiator == RTAS_INITIATOR_POWERMGM)
582 mce_err.initiator = MCE_INITIATOR_POWERMGM;
583 else
584 mce_err.initiator = MCE_INITIATOR_UNKNOWN;
585
586 if (severity == RTAS_SEVERITY_NO_ERROR)
587 mce_err.severity = MCE_SEV_NO_ERROR;
588 else if (severity == RTAS_SEVERITY_EVENT)
589 mce_err.severity = MCE_SEV_WARNING;
590 else if (severity == RTAS_SEVERITY_WARNING)
591 mce_err.severity = MCE_SEV_WARNING;
592 else if (severity == RTAS_SEVERITY_ERROR_SYNC)
593 mce_err.severity = MCE_SEV_SEVERE;
594 else if (severity == RTAS_SEVERITY_ERROR)
595 mce_err.severity = MCE_SEV_SEVERE;
596 else
597 mce_err.severity = MCE_SEV_FATAL;
598
599 if (severity <= RTAS_SEVERITY_ERROR_SYNC)
600 mce_err.sync_error = true;
601 else
602 mce_err.sync_error = false;
603
604 mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN;
605 mce_err.error_class = MCE_ECLASS_UNKNOWN;
606
607 switch (error_type) {
608 case MC_ERROR_TYPE_UE:
609 mce_err.error_type = MCE_ERROR_TYPE_UE;
610 mce_common_process_ue(regs, &mce_err);
611 if (mce_err.ignore_event)
612 disposition = RTAS_DISP_FULLY_RECOVERED;
613 switch (err_sub_type) {
614 case MC_ERROR_UE_IFETCH:
615 mce_err.u.ue_error_type = MCE_UE_ERROR_IFETCH;
616 break;
617 case MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH:
618 mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH;
619 break;
620 case MC_ERROR_UE_LOAD_STORE:
621 mce_err.u.ue_error_type = MCE_UE_ERROR_LOAD_STORE;
622 break;
623 case MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE:
624 mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE;
625 break;
626 case MC_ERROR_UE_INDETERMINATE:
627 default:
628 mce_err.u.ue_error_type = MCE_UE_ERROR_INDETERMINATE;
629 break;
630 }
631 if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED)
632 eaddr = be64_to_cpu(mce_log->effective_address);
633
634 if (mce_log->sub_err_type & UE_LOGICAL_ADDR_PROVIDED) {
635 paddr = be64_to_cpu(mce_log->logical_address);
636 } else if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED) {
637 unsigned long pfn;
638
639 pfn = addr_to_pfn(regs, eaddr);
640 if (pfn != ULONG_MAX)
641 paddr = pfn << PAGE_SHIFT;
642 }
643
644 break;
645 case MC_ERROR_TYPE_SLB:
646 mce_err.error_type = MCE_ERROR_TYPE_SLB;
647 switch (err_sub_type) {
648 case MC_ERROR_SLB_PARITY:
649 mce_err.u.slb_error_type = MCE_SLB_ERROR_PARITY;
650 break;
651 case MC_ERROR_SLB_MULTIHIT:
652 mce_err.u.slb_error_type = MCE_SLB_ERROR_MULTIHIT;
653 break;
654 case MC_ERROR_SLB_INDETERMINATE:
655 default:
656 mce_err.u.slb_error_type = MCE_SLB_ERROR_INDETERMINATE;
657 break;
658 }
659 if (mce_log->sub_err_type & 0x80)
660 eaddr = be64_to_cpu(mce_log->effective_address);
661 break;
662 case MC_ERROR_TYPE_ERAT:
663 mce_err.error_type = MCE_ERROR_TYPE_ERAT;
664 switch (err_sub_type) {
665 case MC_ERROR_ERAT_PARITY:
666 mce_err.u.erat_error_type = MCE_ERAT_ERROR_PARITY;
667 break;
668 case MC_ERROR_ERAT_MULTIHIT:
669 mce_err.u.erat_error_type = MCE_ERAT_ERROR_MULTIHIT;
670 break;
671 case MC_ERROR_ERAT_INDETERMINATE:
672 default:
673 mce_err.u.erat_error_type = MCE_ERAT_ERROR_INDETERMINATE;
674 break;
675 }
676 if (mce_log->sub_err_type & 0x80)
677 eaddr = be64_to_cpu(mce_log->effective_address);
678 break;
679 case MC_ERROR_TYPE_TLB:
680 mce_err.error_type = MCE_ERROR_TYPE_TLB;
681 switch (err_sub_type) {
682 case MC_ERROR_TLB_PARITY:
683 mce_err.u.tlb_error_type = MCE_TLB_ERROR_PARITY;
684 break;
685 case MC_ERROR_TLB_MULTIHIT:
686 mce_err.u.tlb_error_type = MCE_TLB_ERROR_MULTIHIT;
687 break;
688 case MC_ERROR_TLB_INDETERMINATE:
689 default:
690 mce_err.u.tlb_error_type = MCE_TLB_ERROR_INDETERMINATE;
691 break;
692 }
693 if (mce_log->sub_err_type & 0x80)
694 eaddr = be64_to_cpu(mce_log->effective_address);
695 break;
696 case MC_ERROR_TYPE_D_CACHE:
697 mce_err.error_type = MCE_ERROR_TYPE_DCACHE;
698 break;
699 case MC_ERROR_TYPE_I_CACHE:
700 mce_err.error_type = MCE_ERROR_TYPE_ICACHE;
701 break;
702 case MC_ERROR_TYPE_UNKNOWN:
703 default:
704 mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN;
705 break;
706 }
707 out:
708 save_mce_event(regs, disposition == RTAS_DISP_FULLY_RECOVERED,
709 &mce_err, regs->nip, eaddr, paddr);
710 return disposition;
711 }
712
mce_handle_error(struct pt_regs * regs,struct rtas_error_log * errp)713 static int mce_handle_error(struct pt_regs *regs, struct rtas_error_log *errp)
714 {
715 struct pseries_errorlog *pseries_log;
716 struct pseries_mc_errorlog *mce_log = NULL;
717 int disposition = rtas_error_disposition(errp);
718 unsigned long msr;
719 u8 error_type;
720
721 if (!rtas_error_extended(errp))
722 goto out;
723
724 pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE);
725 if (!pseries_log)
726 goto out;
727
728 mce_log = (struct pseries_mc_errorlog *)pseries_log->data;
729 error_type = mce_log->error_type;
730
731 disposition = mce_handle_err_realmode(disposition, error_type);
732
733 /*
734 * Enable translation as we will be accessing per-cpu variables
735 * in save_mce_event() which may fall outside RMO region, also
736 * leave it enabled because subsequently we will be queuing work
737 * to workqueues where again per-cpu variables accessed, besides
738 * fwnmi_release_errinfo() crashes when called in realmode on
739 * pseries.
740 * Note: All the realmode handling like flushing SLB entries for
741 * SLB multihit is done by now.
742 */
743 out:
744 msr = mfmsr();
745 mtmsr(msr | MSR_IR | MSR_DR);
746
747 disposition = mce_handle_err_virtmode(regs, errp, mce_log,
748 disposition);
749
750 /*
751 * Queue irq work to log this rtas event later.
752 * irq_work_queue uses per-cpu variables, so do this in virt
753 * mode as well.
754 */
755 irq_work_queue(&mce_errlog_process_work);
756
757 mtmsr(msr);
758
759 return disposition;
760 }
761
762 /*
763 * Process MCE rtas errlog event.
764 */
mce_process_errlog_event(struct irq_work * work)765 static void mce_process_errlog_event(struct irq_work *work)
766 {
767 struct rtas_error_log *err;
768
769 err = fwnmi_get_errlog();
770 log_error((char *)err, ERR_TYPE_RTAS_LOG, 0);
771 }
772
773 /*
774 * See if we can recover from a machine check exception.
775 * This is only called on power4 (or above) and only via
776 * the Firmware Non-Maskable Interrupts (fwnmi) handler
777 * which provides the error analysis for us.
778 *
779 * Return 1 if corrected (or delivered a signal).
780 * Return 0 if there is nothing we can do.
781 */
recover_mce(struct pt_regs * regs,struct machine_check_event * evt)782 static int recover_mce(struct pt_regs *regs, struct machine_check_event *evt)
783 {
784 int recovered = 0;
785
786 if (regs_is_unrecoverable(regs)) {
787 /* If MSR_RI isn't set, we cannot recover */
788 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
789 recovered = 0;
790 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
791 /* Platform corrected itself */
792 recovered = 1;
793 } else if (evt->severity == MCE_SEV_FATAL) {
794 /* Fatal machine check */
795 pr_err("Machine check interrupt is fatal\n");
796 recovered = 0;
797 }
798
799 if (!recovered && evt->sync_error) {
800 /*
801 * Try to kill processes if we get a synchronous machine check
802 * (e.g., one caused by execution of this instruction). This
803 * will devolve into a panic if we try to kill init or are in
804 * an interrupt etc.
805 *
806 * TODO: Queue up this address for hwpoisioning later.
807 * TODO: This is not quite right for d-side machine
808 * checks ->nip is not necessarily the important
809 * address.
810 */
811 if ((user_mode(regs))) {
812 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
813 recovered = 1;
814 } else if (die_will_crash()) {
815 /*
816 * die() would kill the kernel, so better to go via
817 * the platform reboot code that will log the
818 * machine check.
819 */
820 recovered = 0;
821 } else {
822 die_mce("Machine check", regs, SIGBUS);
823 recovered = 1;
824 }
825 }
826
827 return recovered;
828 }
829
830 /*
831 * Handle a machine check.
832 *
833 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
834 * should be present. If so the handler which called us tells us if the
835 * error was recovered (never true if RI=0).
836 *
837 * On hardware prior to Power 4 these exceptions were asynchronous which
838 * means we can't tell exactly where it occurred and so we can't recover.
839 */
pSeries_machine_check_exception(struct pt_regs * regs)840 int pSeries_machine_check_exception(struct pt_regs *regs)
841 {
842 struct machine_check_event evt;
843
844 if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
845 return 0;
846
847 /* Print things out */
848 if (evt.version != MCE_V1) {
849 pr_err("Machine Check Exception, Unknown event version %d !\n",
850 evt.version);
851 return 0;
852 }
853 machine_check_print_event_info(&evt, user_mode(regs), false);
854
855 if (recover_mce(regs, &evt))
856 return 1;
857
858 return 0;
859 }
860
pseries_machine_check_realmode(struct pt_regs * regs)861 long pseries_machine_check_realmode(struct pt_regs *regs)
862 {
863 struct rtas_error_log *errp;
864 int disposition;
865
866 if (fwnmi_active) {
867 errp = fwnmi_get_errinfo(regs);
868 /*
869 * Call to fwnmi_release_errinfo() in real mode causes kernel
870 * to panic. Hence we will call it as soon as we go into
871 * virtual mode.
872 */
873 disposition = mce_handle_error(regs, errp);
874
875 fwnmi_release_errinfo();
876
877 if (disposition == RTAS_DISP_FULLY_RECOVERED)
878 return 1;
879 }
880
881 return 0;
882 }
883