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