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 		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 
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 
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. */
316 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
317 {
318 	int status;
319 	int state;
320 	int critical;
321 
322 	status = rtas_get_sensor_fast(EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX,
323 				      &state);
324 
325 	if (state > 3)
326 		critical = 1;		/* Time Critical */
327 	else
328 		critical = 0;
329 
330 	spin_lock(&ras_log_buf_lock);
331 
332 	status = rtas_call(ras_check_exception_token, 6, 1, NULL,
333 			   RTAS_VECTOR_EXTERNAL_INTERRUPT,
334 			   virq_to_hw(irq),
335 			   RTAS_EPOW_WARNING,
336 			   critical, __pa(&ras_log_buf),
337 				rtas_get_error_log_max());
338 
339 	log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
340 
341 	rtas_parse_epow_errlog((struct rtas_error_log *)ras_log_buf);
342 
343 	spin_unlock(&ras_log_buf_lock);
344 	return IRQ_HANDLED;
345 }
346 
347 /*
348  * Handle hardware error interrupts.
349  *
350  * RTAS check-exception is called to collect data on the exception.  If
351  * the error is deemed recoverable, we log a warning and return.
352  * For nonrecoverable errors, an error is logged and we stop all processing
353  * as quickly as possible in order to prevent propagation of the failure.
354  */
355 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
356 {
357 	struct rtas_error_log *rtas_elog;
358 	int status;
359 	int fatal;
360 
361 	spin_lock(&ras_log_buf_lock);
362 
363 	status = rtas_call(ras_check_exception_token, 6, 1, NULL,
364 			   RTAS_VECTOR_EXTERNAL_INTERRUPT,
365 			   virq_to_hw(irq),
366 			   RTAS_INTERNAL_ERROR, 1 /* Time Critical */,
367 			   __pa(&ras_log_buf),
368 				rtas_get_error_log_max());
369 
370 	rtas_elog = (struct rtas_error_log *)ras_log_buf;
371 
372 	if (status == 0 &&
373 	    rtas_error_severity(rtas_elog) >= RTAS_SEVERITY_ERROR_SYNC)
374 		fatal = 1;
375 	else
376 		fatal = 0;
377 
378 	/* format and print the extended information */
379 	log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
380 
381 	if (fatal) {
382 		pr_emerg("Fatal hardware error detected. Check RTAS error"
383 			 " log for details. Powering off immediately\n");
384 		emergency_sync();
385 		kernel_power_off();
386 	} else {
387 		pr_err("Recoverable hardware error detected\n");
388 	}
389 
390 	spin_unlock(&ras_log_buf_lock);
391 	return IRQ_HANDLED;
392 }
393 
394 /*
395  * Some versions of FWNMI place the buffer inside the 4kB page starting at
396  * 0x7000. Other versions place it inside the rtas buffer. We check both.
397  * Minimum size of the buffer is 16 bytes.
398  */
399 #define VALID_FWNMI_BUFFER(A) \
400 	((((A) >= 0x7000) && ((A) <= 0x8000 - 16)) || \
401 	(((A) >= rtas.base) && ((A) <= (rtas.base + rtas.size - 16))))
402 
403 static inline struct rtas_error_log *fwnmi_get_errlog(void)
404 {
405 	return (struct rtas_error_log *)local_paca->mce_data_buf;
406 }
407 
408 static __be64 *fwnmi_get_savep(struct pt_regs *regs)
409 {
410 	unsigned long savep_ra;
411 
412 	/* Mask top two bits */
413 	savep_ra = regs->gpr[3] & ~(0x3UL << 62);
414 	if (!VALID_FWNMI_BUFFER(savep_ra)) {
415 		printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]);
416 		return NULL;
417 	}
418 
419 	return __va(savep_ra);
420 }
421 
422 /*
423  * Get the error information for errors coming through the
424  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
425  * the actual r3 if possible, and a ptr to the error log entry
426  * will be returned if found.
427  *
428  * Use one buffer mce_data_buf per cpu to store RTAS error.
429  *
430  * The mce_data_buf does not have any locks or protection around it,
431  * if a second machine check comes in, or a system reset is done
432  * before we have logged the error, then we will get corruption in the
433  * error log.  This is preferable over holding off on calling
434  * ibm,nmi-interlock which would result in us checkstopping if a
435  * second machine check did come in.
436  */
437 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
438 {
439 	struct rtas_error_log *h;
440 	__be64 *savep;
441 
442 	savep = fwnmi_get_savep(regs);
443 	if (!savep)
444 		return NULL;
445 
446 	regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */
447 
448 	h = (struct rtas_error_log *)&savep[1];
449 	/* Use the per cpu buffer from paca to store rtas error log */
450 	memset(local_paca->mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
451 	if (!rtas_error_extended(h)) {
452 		memcpy(local_paca->mce_data_buf, h, sizeof(__u64));
453 	} else {
454 		int len, error_log_length;
455 
456 		error_log_length = 8 + rtas_error_extended_log_length(h);
457 		len = min_t(int, error_log_length, RTAS_ERROR_LOG_MAX);
458 		memcpy(local_paca->mce_data_buf, h, len);
459 	}
460 
461 	return (struct rtas_error_log *)local_paca->mce_data_buf;
462 }
463 
464 /* Call this when done with the data returned by FWNMI_get_errinfo.
465  * It will release the saved data area for other CPUs in the
466  * partition to receive FWNMI errors.
467  */
468 static void fwnmi_release_errinfo(void)
469 {
470 	struct rtas_args rtas_args;
471 	int ret;
472 
473 	/*
474 	 * On pseries, the machine check stack is limited to under 4GB, so
475 	 * args can be on-stack.
476 	 */
477 	rtas_call_unlocked(&rtas_args, ibm_nmi_interlock_token, 0, 1, NULL);
478 	ret = be32_to_cpu(rtas_args.rets[0]);
479 	if (ret != 0)
480 		printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret);
481 }
482 
483 int pSeries_system_reset_exception(struct pt_regs *regs)
484 {
485 #ifdef __LITTLE_ENDIAN__
486 	/*
487 	 * Some firmware byteswaps SRR registers and gives incorrect SRR1. Try
488 	 * to detect the bad SRR1 pattern here. Flip the NIP back to correct
489 	 * endian for reporting purposes. Unfortunately the MSR can't be fixed,
490 	 * so clear it. It will be missing MSR_RI so we won't try to recover.
491 	 */
492 	if ((be64_to_cpu(regs->msr) &
493 			(MSR_LE|MSR_RI|MSR_DR|MSR_IR|MSR_ME|MSR_PR|
494 			 MSR_ILE|MSR_HV|MSR_SF)) == (MSR_DR|MSR_SF)) {
495 		regs->nip = be64_to_cpu((__be64)regs->nip);
496 		regs->msr = 0;
497 	}
498 #endif
499 
500 	if (fwnmi_active) {
501 		__be64 *savep;
502 
503 		/*
504 		 * Firmware (PowerVM and KVM) saves r3 to a save area like
505 		 * machine check, which is not exactly what PAPR (2.9)
506 		 * suggests but there is no way to detect otherwise, so this
507 		 * is the interface now.
508 		 *
509 		 * System resets do not save any error log or require an
510 		 * "ibm,nmi-interlock" rtas call to release.
511 		 */
512 
513 		savep = fwnmi_get_savep(regs);
514 		if (savep)
515 			regs->gpr[3] = be64_to_cpu(savep[0]); /* restore original r3 */
516 	}
517 
518 	if (smp_handle_nmi_ipi(regs))
519 		return 1;
520 
521 	return 0; /* need to perform reset */
522 }
523 
524 static int mce_handle_err_realmode(int disposition, u8 error_type)
525 {
526 #ifdef CONFIG_PPC_BOOK3S_64
527 	if (disposition == RTAS_DISP_NOT_RECOVERED) {
528 		switch (error_type) {
529 		case	MC_ERROR_TYPE_SLB:
530 		case	MC_ERROR_TYPE_ERAT:
531 			/*
532 			 * Store the old slb content in paca before flushing.
533 			 * Print this when we go to virtual mode.
534 			 * There are chances that we may hit MCE again if there
535 			 * is a parity error on the SLB entry we trying to read
536 			 * for saving. Hence limit the slb saving to single
537 			 * level of recursion.
538 			 */
539 			if (local_paca->in_mce == 1)
540 				slb_save_contents(local_paca->mce_faulty_slbs);
541 			flush_and_reload_slb();
542 			disposition = RTAS_DISP_FULLY_RECOVERED;
543 			break;
544 		default:
545 			break;
546 		}
547 	} else if (disposition == RTAS_DISP_LIMITED_RECOVERY) {
548 		/* Platform corrected itself but could be degraded */
549 		pr_err("MCE: limited recovery, system may be degraded\n");
550 		disposition = RTAS_DISP_FULLY_RECOVERED;
551 	}
552 #endif
553 	return disposition;
554 }
555 
556 static int mce_handle_err_virtmode(struct pt_regs *regs,
557 				   struct rtas_error_log *errp,
558 				   struct pseries_mc_errorlog *mce_log,
559 				   int disposition)
560 {
561 	struct mce_error_info mce_err = { 0 };
562 	int initiator = rtas_error_initiator(errp);
563 	int severity = rtas_error_severity(errp);
564 	unsigned long eaddr = 0, paddr = 0;
565 	u8 error_type, err_sub_type;
566 
567 	if (!mce_log)
568 		goto out;
569 
570 	error_type = mce_log->error_type;
571 	err_sub_type = rtas_mc_error_sub_type(mce_log);
572 
573 	if (initiator == RTAS_INITIATOR_UNKNOWN)
574 		mce_err.initiator = MCE_INITIATOR_UNKNOWN;
575 	else if (initiator == RTAS_INITIATOR_CPU)
576 		mce_err.initiator = MCE_INITIATOR_CPU;
577 	else if (initiator == RTAS_INITIATOR_PCI)
578 		mce_err.initiator = MCE_INITIATOR_PCI;
579 	else if (initiator == RTAS_INITIATOR_ISA)
580 		mce_err.initiator = MCE_INITIATOR_ISA;
581 	else if (initiator == RTAS_INITIATOR_MEMORY)
582 		mce_err.initiator = MCE_INITIATOR_MEMORY;
583 	else if (initiator == RTAS_INITIATOR_POWERMGM)
584 		mce_err.initiator = MCE_INITIATOR_POWERMGM;
585 	else
586 		mce_err.initiator = MCE_INITIATOR_UNKNOWN;
587 
588 	if (severity == RTAS_SEVERITY_NO_ERROR)
589 		mce_err.severity = MCE_SEV_NO_ERROR;
590 	else if (severity == RTAS_SEVERITY_EVENT)
591 		mce_err.severity = MCE_SEV_WARNING;
592 	else if (severity == RTAS_SEVERITY_WARNING)
593 		mce_err.severity = MCE_SEV_WARNING;
594 	else if (severity == RTAS_SEVERITY_ERROR_SYNC)
595 		mce_err.severity = MCE_SEV_SEVERE;
596 	else if (severity == RTAS_SEVERITY_ERROR)
597 		mce_err.severity = MCE_SEV_SEVERE;
598 	else if (severity == RTAS_SEVERITY_FATAL)
599 		mce_err.severity = MCE_SEV_FATAL;
600 	else
601 		mce_err.severity = MCE_SEV_FATAL;
602 
603 	if (severity <= RTAS_SEVERITY_ERROR_SYNC)
604 		mce_err.sync_error = true;
605 	else
606 		mce_err.sync_error = false;
607 
608 	mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN;
609 	mce_err.error_class = MCE_ECLASS_UNKNOWN;
610 
611 	switch (error_type) {
612 	case MC_ERROR_TYPE_UE:
613 		mce_err.error_type = MCE_ERROR_TYPE_UE;
614 		mce_common_process_ue(regs, &mce_err);
615 		if (mce_err.ignore_event)
616 			disposition = RTAS_DISP_FULLY_RECOVERED;
617 		switch (err_sub_type) {
618 		case MC_ERROR_UE_IFETCH:
619 			mce_err.u.ue_error_type = MCE_UE_ERROR_IFETCH;
620 			break;
621 		case MC_ERROR_UE_PAGE_TABLE_WALK_IFETCH:
622 			mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH;
623 			break;
624 		case MC_ERROR_UE_LOAD_STORE:
625 			mce_err.u.ue_error_type = MCE_UE_ERROR_LOAD_STORE;
626 			break;
627 		case MC_ERROR_UE_PAGE_TABLE_WALK_LOAD_STORE:
628 			mce_err.u.ue_error_type = MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE;
629 			break;
630 		case MC_ERROR_UE_INDETERMINATE:
631 		default:
632 			mce_err.u.ue_error_type = MCE_UE_ERROR_INDETERMINATE;
633 			break;
634 		}
635 		if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED)
636 			eaddr = be64_to_cpu(mce_log->effective_address);
637 
638 		if (mce_log->sub_err_type & UE_LOGICAL_ADDR_PROVIDED) {
639 			paddr = be64_to_cpu(mce_log->logical_address);
640 		} else if (mce_log->sub_err_type & UE_EFFECTIVE_ADDR_PROVIDED) {
641 			unsigned long pfn;
642 
643 			pfn = addr_to_pfn(regs, eaddr);
644 			if (pfn != ULONG_MAX)
645 				paddr = pfn << PAGE_SHIFT;
646 		}
647 
648 		break;
649 	case MC_ERROR_TYPE_SLB:
650 		mce_err.error_type = MCE_ERROR_TYPE_SLB;
651 		switch (err_sub_type) {
652 		case MC_ERROR_SLB_PARITY:
653 			mce_err.u.slb_error_type = MCE_SLB_ERROR_PARITY;
654 			break;
655 		case MC_ERROR_SLB_MULTIHIT:
656 			mce_err.u.slb_error_type = MCE_SLB_ERROR_MULTIHIT;
657 			break;
658 		case MC_ERROR_SLB_INDETERMINATE:
659 		default:
660 			mce_err.u.slb_error_type = MCE_SLB_ERROR_INDETERMINATE;
661 			break;
662 		}
663 		if (mce_log->sub_err_type & 0x80)
664 			eaddr = be64_to_cpu(mce_log->effective_address);
665 		break;
666 	case MC_ERROR_TYPE_ERAT:
667 		mce_err.error_type = MCE_ERROR_TYPE_ERAT;
668 		switch (err_sub_type) {
669 		case MC_ERROR_ERAT_PARITY:
670 			mce_err.u.erat_error_type = MCE_ERAT_ERROR_PARITY;
671 			break;
672 		case MC_ERROR_ERAT_MULTIHIT:
673 			mce_err.u.erat_error_type = MCE_ERAT_ERROR_MULTIHIT;
674 			break;
675 		case MC_ERROR_ERAT_INDETERMINATE:
676 		default:
677 			mce_err.u.erat_error_type = MCE_ERAT_ERROR_INDETERMINATE;
678 			break;
679 		}
680 		if (mce_log->sub_err_type & 0x80)
681 			eaddr = be64_to_cpu(mce_log->effective_address);
682 		break;
683 	case MC_ERROR_TYPE_TLB:
684 		mce_err.error_type = MCE_ERROR_TYPE_TLB;
685 		switch (err_sub_type) {
686 		case MC_ERROR_TLB_PARITY:
687 			mce_err.u.tlb_error_type = MCE_TLB_ERROR_PARITY;
688 			break;
689 		case MC_ERROR_TLB_MULTIHIT:
690 			mce_err.u.tlb_error_type = MCE_TLB_ERROR_MULTIHIT;
691 			break;
692 		case MC_ERROR_TLB_INDETERMINATE:
693 		default:
694 			mce_err.u.tlb_error_type = MCE_TLB_ERROR_INDETERMINATE;
695 			break;
696 		}
697 		if (mce_log->sub_err_type & 0x80)
698 			eaddr = be64_to_cpu(mce_log->effective_address);
699 		break;
700 	case MC_ERROR_TYPE_D_CACHE:
701 		mce_err.error_type = MCE_ERROR_TYPE_DCACHE;
702 		break;
703 	case MC_ERROR_TYPE_I_CACHE:
704 		mce_err.error_type = MCE_ERROR_TYPE_DCACHE;
705 		break;
706 	case MC_ERROR_TYPE_UNKNOWN:
707 	default:
708 		mce_err.error_type = MCE_ERROR_TYPE_UNKNOWN;
709 		break;
710 	}
711 out:
712 	save_mce_event(regs, disposition == RTAS_DISP_FULLY_RECOVERED,
713 		       &mce_err, regs->nip, eaddr, paddr);
714 	return disposition;
715 }
716 
717 static int mce_handle_error(struct pt_regs *regs, struct rtas_error_log *errp)
718 {
719 	struct pseries_errorlog *pseries_log;
720 	struct pseries_mc_errorlog *mce_log = NULL;
721 	int disposition = rtas_error_disposition(errp);
722 	u8 error_type;
723 
724 	if (!rtas_error_extended(errp))
725 		goto out;
726 
727 	pseries_log = get_pseries_errorlog(errp, PSERIES_ELOG_SECT_ID_MCE);
728 	if (!pseries_log)
729 		goto out;
730 
731 	mce_log = (struct pseries_mc_errorlog *)pseries_log->data;
732 	error_type = mce_log->error_type;
733 
734 	disposition = mce_handle_err_realmode(disposition, error_type);
735 
736 	/*
737 	 * Enable translation as we will be accessing per-cpu variables
738 	 * in save_mce_event() which may fall outside RMO region, also
739 	 * leave it enabled because subsequently we will be queuing work
740 	 * to workqueues where again per-cpu variables accessed, besides
741 	 * fwnmi_release_errinfo() crashes when called in realmode on
742 	 * pseries.
743 	 * Note: All the realmode handling like flushing SLB entries for
744 	 *       SLB multihit is done by now.
745 	 */
746 out:
747 	mtmsr(mfmsr() | MSR_IR | MSR_DR);
748 	disposition = mce_handle_err_virtmode(regs, errp, mce_log,
749 					      disposition);
750 	return disposition;
751 }
752 
753 /*
754  * Process MCE rtas errlog event.
755  */
756 static void mce_process_errlog_event(struct irq_work *work)
757 {
758 	struct rtas_error_log *err;
759 
760 	err = fwnmi_get_errlog();
761 	log_error((char *)err, ERR_TYPE_RTAS_LOG, 0);
762 }
763 
764 /*
765  * See if we can recover from a machine check exception.
766  * This is only called on power4 (or above) and only via
767  * the Firmware Non-Maskable Interrupts (fwnmi) handler
768  * which provides the error analysis for us.
769  *
770  * Return 1 if corrected (or delivered a signal).
771  * Return 0 if there is nothing we can do.
772  */
773 static int recover_mce(struct pt_regs *regs, struct machine_check_event *evt)
774 {
775 	int recovered = 0;
776 
777 	if (!(regs->msr & MSR_RI)) {
778 		/* If MSR_RI isn't set, we cannot recover */
779 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
780 		recovered = 0;
781 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
782 		/* Platform corrected itself */
783 		recovered = 1;
784 	} else if (evt->severity == MCE_SEV_FATAL) {
785 		/* Fatal machine check */
786 		pr_err("Machine check interrupt is fatal\n");
787 		recovered = 0;
788 	}
789 
790 	if (!recovered && evt->sync_error) {
791 		/*
792 		 * Try to kill processes if we get a synchronous machine check
793 		 * (e.g., one caused by execution of this instruction). This
794 		 * will devolve into a panic if we try to kill init or are in
795 		 * an interrupt etc.
796 		 *
797 		 * TODO: Queue up this address for hwpoisioning later.
798 		 * TODO: This is not quite right for d-side machine
799 		 *       checks ->nip is not necessarily the important
800 		 *       address.
801 		 */
802 		if ((user_mode(regs))) {
803 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
804 			recovered = 1;
805 		} else if (die_will_crash()) {
806 			/*
807 			 * die() would kill the kernel, so better to go via
808 			 * the platform reboot code that will log the
809 			 * machine check.
810 			 */
811 			recovered = 0;
812 		} else {
813 			die("Machine check", regs, SIGBUS);
814 			recovered = 1;
815 		}
816 	}
817 
818 	return recovered;
819 }
820 
821 /*
822  * Handle a machine check.
823  *
824  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
825  * should be present.  If so the handler which called us tells us if the
826  * error was recovered (never true if RI=0).
827  *
828  * On hardware prior to Power 4 these exceptions were asynchronous which
829  * means we can't tell exactly where it occurred and so we can't recover.
830  */
831 int pSeries_machine_check_exception(struct pt_regs *regs)
832 {
833 	struct machine_check_event evt;
834 
835 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
836 		return 0;
837 
838 	/* Print things out */
839 	if (evt.version != MCE_V1) {
840 		pr_err("Machine Check Exception, Unknown event version %d !\n",
841 		       evt.version);
842 		return 0;
843 	}
844 	machine_check_print_event_info(&evt, user_mode(regs), false);
845 
846 	if (recover_mce(regs, &evt))
847 		return 1;
848 
849 	return 0;
850 }
851 
852 long pseries_machine_check_realmode(struct pt_regs *regs)
853 {
854 	struct rtas_error_log *errp;
855 	int disposition;
856 
857 	if (fwnmi_active) {
858 		errp = fwnmi_get_errinfo(regs);
859 		/*
860 		 * Call to fwnmi_release_errinfo() in real mode causes kernel
861 		 * to panic. Hence we will call it as soon as we go into
862 		 * virtual mode.
863 		 */
864 		disposition = mce_handle_error(regs, errp);
865 		fwnmi_release_errinfo();
866 
867 		/* Queue irq work to log this rtas event later. */
868 		irq_work_queue(&mce_errlog_process_work);
869 
870 		if (disposition == RTAS_DISP_FULLY_RECOVERED)
871 			return 1;
872 	}
873 
874 	return 0;
875 }
876