1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV OPAL high level interfaces
4  *
5  * Copyright 2011 IBM Corp.
6  */
7 
8 #define pr_fmt(fmt)	"opal: " fmt
9 
10 #include <linux/printk.h>
11 #include <linux/types.h>
12 #include <linux/of.h>
13 #include <linux/of_fdt.h>
14 #include <linux/of_platform.h>
15 #include <linux/of_address.h>
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/slab.h>
19 #include <linux/sched.h>
20 #include <linux/kobject.h>
21 #include <linux/delay.h>
22 #include <linux/memblock.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/kmsg_dump.h>
26 #include <linux/console.h>
27 #include <linux/sched/debug.h>
28 
29 #include <asm/machdep.h>
30 #include <asm/opal.h>
31 #include <asm/firmware.h>
32 #include <asm/mce.h>
33 #include <asm/imc-pmu.h>
34 #include <asm/bug.h>
35 
36 #include "powernv.h"
37 
38 #define OPAL_MSG_QUEUE_MAX 16
39 
40 struct opal_msg_node {
41 	struct list_head	list;
42 	struct opal_msg		msg;
43 };
44 
45 static DEFINE_SPINLOCK(msg_list_lock);
46 static LIST_HEAD(msg_list);
47 
48 /* /sys/firmware/opal */
49 struct kobject *opal_kobj;
50 
51 struct opal {
52 	u64 base;
53 	u64 entry;
54 	u64 size;
55 } opal;
56 
57 struct mcheck_recoverable_range {
58 	u64 start_addr;
59 	u64 end_addr;
60 	u64 recover_addr;
61 };
62 
63 static int msg_list_size;
64 
65 static struct mcheck_recoverable_range *mc_recoverable_range;
66 static int mc_recoverable_range_len;
67 
68 struct device_node *opal_node;
69 static DEFINE_SPINLOCK(opal_write_lock);
70 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
71 static uint32_t opal_heartbeat;
72 static struct task_struct *kopald_tsk;
73 static struct opal_msg *opal_msg;
74 static u32 opal_msg_size __ro_after_init;
75 
76 void __init opal_configure_cores(void)
77 {
78 	u64 reinit_flags = 0;
79 
80 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
81 	 *
82 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
83 	 * also restore HIDs and other SPRs to their original value
84 	 * but it might clobber a bunch.
85 	 */
86 #ifdef __BIG_ENDIAN__
87 	reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
88 #else
89 	reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
90 #endif
91 
92 	/*
93 	 * POWER9 always support running hash:
94 	 *  ie. Host hash  supports  hash guests
95 	 *      Host radix supports  hash/radix guests
96 	 */
97 	if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
98 		reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
99 		if (early_radix_enabled())
100 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
101 	}
102 
103 	opal_reinit_cpus(reinit_flags);
104 
105 	/* Restore some bits */
106 	if (cur_cpu_spec->cpu_restore)
107 		cur_cpu_spec->cpu_restore();
108 }
109 
110 int __init early_init_dt_scan_opal(unsigned long node,
111 				   const char *uname, int depth, void *data)
112 {
113 	const void *basep, *entryp, *sizep;
114 	int basesz, entrysz, runtimesz;
115 
116 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
117 		return 0;
118 
119 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
120 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
121 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
122 
123 	if (!basep || !entryp || !sizep)
124 		return 1;
125 
126 	opal.base = of_read_number(basep, basesz/4);
127 	opal.entry = of_read_number(entryp, entrysz/4);
128 	opal.size = of_read_number(sizep, runtimesz/4);
129 
130 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
131 		 opal.base, basep, basesz);
132 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
133 		 opal.entry, entryp, entrysz);
134 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
135 		 opal.size, sizep, runtimesz);
136 
137 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
138 		powerpc_firmware_features |= FW_FEATURE_OPAL;
139 		pr_debug("OPAL detected !\n");
140 	} else {
141 		panic("OPAL != V3 detected, no longer supported.\n");
142 	}
143 
144 	return 1;
145 }
146 
147 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
148 				   const char *uname, int depth, void *data)
149 {
150 	int i, psize, size;
151 	const __be32 *prop;
152 
153 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
154 		return 0;
155 
156 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
157 
158 	if (!prop)
159 		return 1;
160 
161 	pr_debug("Found machine check recoverable ranges.\n");
162 
163 	/*
164 	 * Calculate number of available entries.
165 	 *
166 	 * Each recoverable address range entry is (start address, len,
167 	 * recovery address), 2 cells each for start and recovery address,
168 	 * 1 cell for len, totalling 5 cells per entry.
169 	 */
170 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
171 
172 	/* Sanity check */
173 	if (!mc_recoverable_range_len)
174 		return 1;
175 
176 	/* Size required to hold all the entries. */
177 	size = mc_recoverable_range_len *
178 			sizeof(struct mcheck_recoverable_range);
179 
180 	/*
181 	 * Allocate a buffer to hold the MC recoverable ranges.
182 	 */
183 	mc_recoverable_range = memblock_alloc(size, __alignof__(u64));
184 	if (!mc_recoverable_range)
185 		panic("%s: Failed to allocate %u bytes align=0x%lx\n",
186 		      __func__, size, __alignof__(u64));
187 
188 	for (i = 0; i < mc_recoverable_range_len; i++) {
189 		mc_recoverable_range[i].start_addr =
190 					of_read_number(prop + (i * 5) + 0, 2);
191 		mc_recoverable_range[i].end_addr =
192 					mc_recoverable_range[i].start_addr +
193 					of_read_number(prop + (i * 5) + 2, 1);
194 		mc_recoverable_range[i].recover_addr =
195 					of_read_number(prop + (i * 5) + 3, 2);
196 
197 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
198 				mc_recoverable_range[i].start_addr,
199 				mc_recoverable_range[i].end_addr,
200 				mc_recoverable_range[i].recover_addr);
201 	}
202 	return 1;
203 }
204 
205 static int __init opal_register_exception_handlers(void)
206 {
207 #ifdef __BIG_ENDIAN__
208 	u64 glue;
209 
210 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
211 		return -ENODEV;
212 
213 	/* Hookup some exception handlers except machine check. We use the
214 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
215 	 */
216 	glue = 0x7000;
217 
218 	/*
219 	 * Only ancient OPAL firmware requires this.
220 	 * Specifically, firmware from FW810.00 (released June 2014)
221 	 * through FW810.20 (Released October 2014).
222 	 *
223 	 * Check if we are running on newer (post Oct 2014) firmware that
224 	 * exports the OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to
225 	 * patch the HMI interrupt and we catch it directly in Linux.
226 	 *
227 	 * For older firmware (i.e < FW810.20), we fallback to old behavior and
228 	 * let OPAL patch the HMI vector and handle it inside OPAL firmware.
229 	 *
230 	 * For newer firmware we catch/handle the HMI directly in Linux.
231 	 */
232 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
233 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
234 		opal_register_exception_handler(
235 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
236 				0, glue);
237 		glue += 128;
238 	}
239 
240 	/*
241 	 * Only applicable to ancient firmware, all modern
242 	 * (post March 2015/skiboot 5.0) firmware will just return
243 	 * OPAL_UNSUPPORTED.
244 	 */
245 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
246 #endif
247 
248 	return 0;
249 }
250 machine_early_initcall(powernv, opal_register_exception_handlers);
251 
252 static void queue_replay_msg(void *msg)
253 {
254 	struct opal_msg_node *msg_node;
255 
256 	if (msg_list_size < OPAL_MSG_QUEUE_MAX) {
257 		msg_node = kzalloc(sizeof(*msg_node), GFP_ATOMIC);
258 		if (msg_node) {
259 			INIT_LIST_HEAD(&msg_node->list);
260 			memcpy(&msg_node->msg, msg, sizeof(struct opal_msg));
261 			list_add_tail(&msg_node->list, &msg_list);
262 			msg_list_size++;
263 		} else
264 			pr_warn_once("message queue no memory\n");
265 
266 		if (msg_list_size >= OPAL_MSG_QUEUE_MAX)
267 			pr_warn_once("message queue full\n");
268 	}
269 }
270 
271 static void dequeue_replay_msg(enum opal_msg_type msg_type)
272 {
273 	struct opal_msg_node *msg_node, *tmp;
274 
275 	list_for_each_entry_safe(msg_node, tmp, &msg_list, list) {
276 		if (be32_to_cpu(msg_node->msg.msg_type) != msg_type)
277 			continue;
278 
279 		atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
280 					msg_type,
281 					&msg_node->msg);
282 
283 		list_del(&msg_node->list);
284 		kfree(msg_node);
285 		msg_list_size--;
286 	}
287 }
288 
289 /*
290  * Opal message notifier based on message type. Allow subscribers to get
291  * notified for specific messgae type.
292  */
293 int opal_message_notifier_register(enum opal_msg_type msg_type,
294 					struct notifier_block *nb)
295 {
296 	int ret;
297 	unsigned long flags;
298 
299 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
300 		pr_warn("%s: Invalid arguments, msg_type:%d\n",
301 			__func__, msg_type);
302 		return -EINVAL;
303 	}
304 
305 	spin_lock_irqsave(&msg_list_lock, flags);
306 	ret = atomic_notifier_chain_register(
307 		&opal_msg_notifier_head[msg_type], nb);
308 
309 	/*
310 	 * If the registration succeeded, replay any queued messages that came
311 	 * in prior to the notifier chain registration. msg_list_lock held here
312 	 * to ensure they're delivered prior to any subsequent messages.
313 	 */
314 	if (ret == 0)
315 		dequeue_replay_msg(msg_type);
316 
317 	spin_unlock_irqrestore(&msg_list_lock, flags);
318 
319 	return ret;
320 }
321 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
322 
323 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
324 				     struct notifier_block *nb)
325 {
326 	return atomic_notifier_chain_unregister(
327 			&opal_msg_notifier_head[msg_type], nb);
328 }
329 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
330 
331 static void opal_message_do_notify(uint32_t msg_type, void *msg)
332 {
333 	unsigned long flags;
334 	bool queued = false;
335 
336 	spin_lock_irqsave(&msg_list_lock, flags);
337 	if (opal_msg_notifier_head[msg_type].head == NULL) {
338 		/*
339 		 * Queue up the msg since no notifiers have registered
340 		 * yet for this msg_type.
341 		 */
342 		queue_replay_msg(msg);
343 		queued = true;
344 	}
345 	spin_unlock_irqrestore(&msg_list_lock, flags);
346 
347 	if (queued)
348 		return;
349 
350 	/* notify subscribers */
351 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
352 					msg_type, msg);
353 }
354 
355 static void opal_handle_message(void)
356 {
357 	s64 ret;
358 	u32 type;
359 
360 	ret = opal_get_msg(__pa(opal_msg), opal_msg_size);
361 	/* No opal message pending. */
362 	if (ret == OPAL_RESOURCE)
363 		return;
364 
365 	/* check for errors. */
366 	if (ret) {
367 		pr_warn("%s: Failed to retrieve opal message, err=%lld\n",
368 			__func__, ret);
369 		return;
370 	}
371 
372 	type = be32_to_cpu(opal_msg->msg_type);
373 
374 	/* Sanity check */
375 	if (type >= OPAL_MSG_TYPE_MAX) {
376 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
377 		return;
378 	}
379 	opal_message_do_notify(type, (void *)opal_msg);
380 }
381 
382 static irqreturn_t opal_message_notify(int irq, void *data)
383 {
384 	opal_handle_message();
385 	return IRQ_HANDLED;
386 }
387 
388 static int __init opal_message_init(struct device_node *opal_node)
389 {
390 	int ret, i, irq;
391 
392 	ret = of_property_read_u32(opal_node, "opal-msg-size", &opal_msg_size);
393 	if (ret) {
394 		pr_notice("Failed to read opal-msg-size property\n");
395 		opal_msg_size = sizeof(struct opal_msg);
396 	}
397 
398 	opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
399 	if (!opal_msg) {
400 		opal_msg_size = sizeof(struct opal_msg);
401 		/* Try to allocate fixed message size */
402 		opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
403 		BUG_ON(opal_msg == NULL);
404 	}
405 
406 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
407 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
408 
409 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
410 	if (!irq) {
411 		pr_err("%s: Can't register OPAL event irq (%d)\n",
412 		       __func__, irq);
413 		return irq;
414 	}
415 
416 	ret = request_irq(irq, opal_message_notify,
417 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
418 	if (ret) {
419 		pr_err("%s: Can't request OPAL event irq (%d)\n",
420 		       __func__, ret);
421 		return ret;
422 	}
423 
424 	return 0;
425 }
426 
427 int opal_get_chars(uint32_t vtermno, char *buf, int count)
428 {
429 	s64 rc;
430 	__be64 evt, len;
431 
432 	if (!opal.entry)
433 		return -ENODEV;
434 	opal_poll_events(&evt);
435 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
436 		return 0;
437 	len = cpu_to_be64(count);
438 	rc = opal_console_read(vtermno, &len, buf);
439 	if (rc == OPAL_SUCCESS)
440 		return be64_to_cpu(len);
441 	return 0;
442 }
443 
444 static int __opal_put_chars(uint32_t vtermno, const char *data, int total_len, bool atomic)
445 {
446 	unsigned long flags = 0 /* shut up gcc */;
447 	int written;
448 	__be64 olen;
449 	s64 rc;
450 
451 	if (!opal.entry)
452 		return -ENODEV;
453 
454 	if (atomic)
455 		spin_lock_irqsave(&opal_write_lock, flags);
456 	rc = opal_console_write_buffer_space(vtermno, &olen);
457 	if (rc || be64_to_cpu(olen) < total_len) {
458 		/* Closed -> drop characters */
459 		if (rc)
460 			written = total_len;
461 		else
462 			written = -EAGAIN;
463 		goto out;
464 	}
465 
466 	/* Should not get a partial write here because space is available. */
467 	olen = cpu_to_be64(total_len);
468 	rc = opal_console_write(vtermno, &olen, data);
469 	if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
470 		if (rc == OPAL_BUSY_EVENT)
471 			opal_poll_events(NULL);
472 		written = -EAGAIN;
473 		goto out;
474 	}
475 
476 	/* Closed or other error drop */
477 	if (rc != OPAL_SUCCESS) {
478 		written = opal_error_code(rc);
479 		goto out;
480 	}
481 
482 	written = be64_to_cpu(olen);
483 	if (written < total_len) {
484 		if (atomic) {
485 			/* Should not happen */
486 			pr_warn("atomic console write returned partial "
487 				"len=%d written=%d\n", total_len, written);
488 		}
489 		if (!written)
490 			written = -EAGAIN;
491 	}
492 
493 out:
494 	if (atomic)
495 		spin_unlock_irqrestore(&opal_write_lock, flags);
496 
497 	return written;
498 }
499 
500 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
501 {
502 	return __opal_put_chars(vtermno, data, total_len, false);
503 }
504 
505 /*
506  * opal_put_chars_atomic will not perform partial-writes. Data will be
507  * atomically written to the terminal or not at all. This is not strictly
508  * true at the moment because console space can race with OPAL's console
509  * writes.
510  */
511 int opal_put_chars_atomic(uint32_t vtermno, const char *data, int total_len)
512 {
513 	return __opal_put_chars(vtermno, data, total_len, true);
514 }
515 
516 static s64 __opal_flush_console(uint32_t vtermno)
517 {
518 	s64 rc;
519 
520 	if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
521 		__be64 evt;
522 
523 		/*
524 		 * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
525 		 * the console can still be flushed by calling the polling
526 		 * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
527 		 */
528 		WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
529 
530 		opal_poll_events(&evt);
531 		if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
532 			return OPAL_SUCCESS;
533 		return OPAL_BUSY;
534 
535 	} else {
536 		rc = opal_console_flush(vtermno);
537 		if (rc == OPAL_BUSY_EVENT) {
538 			opal_poll_events(NULL);
539 			rc = OPAL_BUSY;
540 		}
541 		return rc;
542 	}
543 
544 }
545 
546 /*
547  * opal_flush_console spins until the console is flushed
548  */
549 int opal_flush_console(uint32_t vtermno)
550 {
551 	for (;;) {
552 		s64 rc = __opal_flush_console(vtermno);
553 
554 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
555 			mdelay(1);
556 			continue;
557 		}
558 
559 		return opal_error_code(rc);
560 	}
561 }
562 
563 /*
564  * opal_flush_chars is an hvc interface that sleeps until the console is
565  * flushed if wait, otherwise it will return -EBUSY if the console has data,
566  * -EAGAIN if it has data and some of it was flushed.
567  */
568 int opal_flush_chars(uint32_t vtermno, bool wait)
569 {
570 	for (;;) {
571 		s64 rc = __opal_flush_console(vtermno);
572 
573 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
574 			if (wait) {
575 				msleep(OPAL_BUSY_DELAY_MS);
576 				continue;
577 			}
578 			if (rc == OPAL_PARTIAL)
579 				return -EAGAIN;
580 		}
581 
582 		return opal_error_code(rc);
583 	}
584 }
585 
586 static int opal_recover_mce(struct pt_regs *regs,
587 					struct machine_check_event *evt)
588 {
589 	int recovered = 0;
590 
591 	if (regs_is_unrecoverable(regs)) {
592 		/* If MSR_RI isn't set, we cannot recover */
593 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
594 		recovered = 0;
595 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
596 		/* Platform corrected itself */
597 		recovered = 1;
598 	} else if (evt->severity == MCE_SEV_FATAL) {
599 		/* Fatal machine check */
600 		pr_err("Machine check interrupt is fatal\n");
601 		recovered = 0;
602 	}
603 
604 	if (!recovered && evt->sync_error) {
605 		/*
606 		 * Try to kill processes if we get a synchronous machine check
607 		 * (e.g., one caused by execution of this instruction). This
608 		 * will devolve into a panic if we try to kill init or are in
609 		 * an interrupt etc.
610 		 *
611 		 * TODO: Queue up this address for hwpoisioning later.
612 		 * TODO: This is not quite right for d-side machine
613 		 *       checks ->nip is not necessarily the important
614 		 *       address.
615 		 */
616 		if ((user_mode(regs))) {
617 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
618 			recovered = 1;
619 		} else if (die_will_crash()) {
620 			/*
621 			 * die() would kill the kernel, so better to go via
622 			 * the platform reboot code that will log the
623 			 * machine check.
624 			 */
625 			recovered = 0;
626 		} else {
627 			die_mce("Machine check", regs, SIGBUS);
628 			recovered = 1;
629 		}
630 	}
631 
632 	return recovered;
633 }
634 
635 void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
636 {
637 	panic_flush_kmsg_start();
638 
639 	pr_emerg("Hardware platform error: %s\n", msg);
640 	if (regs)
641 		show_regs(regs);
642 	smp_send_stop();
643 
644 	panic_flush_kmsg_end();
645 
646 	/*
647 	 * Don't bother to shut things down because this will
648 	 * xstop the system.
649 	 */
650 	if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
651 						== OPAL_UNSUPPORTED) {
652 		pr_emerg("Reboot type %d not supported for %s\n",
653 				OPAL_REBOOT_PLATFORM_ERROR, msg);
654 	}
655 
656 	/*
657 	 * We reached here. There can be three possibilities:
658 	 * 1. We are running on a firmware level that do not support
659 	 *    opal_cec_reboot2()
660 	 * 2. We are running on a firmware level that do not support
661 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
662 	 * 3. We are running on FSP based system that does not need
663 	 *    opal to trigger checkstop explicitly for error analysis.
664 	 *    The FSP PRD component would have already got notified
665 	 *    about this error through other channels.
666 	 * 4. We are running on a newer skiboot that by default does
667 	 *    not cause a checkstop, drops us back to the kernel to
668 	 *    extract context and state at the time of the error.
669 	 */
670 
671 	panic(msg);
672 }
673 
674 int opal_machine_check(struct pt_regs *regs)
675 {
676 	struct machine_check_event evt;
677 
678 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
679 		return 0;
680 
681 	/* Print things out */
682 	if (evt.version != MCE_V1) {
683 		pr_err("Machine Check Exception, Unknown event version %d !\n",
684 		       evt.version);
685 		return 0;
686 	}
687 	machine_check_print_event_info(&evt, user_mode(regs), false);
688 
689 	if (opal_recover_mce(regs, &evt))
690 		return 1;
691 
692 	pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
693 }
694 
695 /* Early hmi handler called in real mode. */
696 int opal_hmi_exception_early(struct pt_regs *regs)
697 {
698 	s64 rc;
699 
700 	/*
701 	 * call opal hmi handler. Pass paca address as token.
702 	 * The return value OPAL_SUCCESS is an indication that there is
703 	 * an HMI event generated waiting to pull by Linux.
704 	 */
705 	rc = opal_handle_hmi();
706 	if (rc == OPAL_SUCCESS) {
707 		local_paca->hmi_event_available = 1;
708 		return 1;
709 	}
710 	return 0;
711 }
712 
713 int opal_hmi_exception_early2(struct pt_regs *regs)
714 {
715 	s64 rc;
716 	__be64 out_flags;
717 
718 	/*
719 	 * call opal hmi handler.
720 	 * Check 64-bit flag mask to find out if an event was generated,
721 	 * and whether TB is still valid or not etc.
722 	 */
723 	rc = opal_handle_hmi2(&out_flags);
724 	if (rc != OPAL_SUCCESS)
725 		return 0;
726 
727 	if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_NEW_EVENT)
728 		local_paca->hmi_event_available = 1;
729 	if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_TOD_TB_FAIL)
730 		tb_invalid = true;
731 	return 1;
732 }
733 
734 /* HMI exception handler called in virtual mode when irqs are next enabled. */
735 int opal_handle_hmi_exception(struct pt_regs *regs)
736 {
737 	/*
738 	 * Check if HMI event is available.
739 	 * if Yes, then wake kopald to process them.
740 	 */
741 	if (!local_paca->hmi_event_available)
742 		return 0;
743 
744 	local_paca->hmi_event_available = 0;
745 	opal_wake_poller();
746 
747 	return 1;
748 }
749 
750 static uint64_t find_recovery_address(uint64_t nip)
751 {
752 	int i;
753 
754 	for (i = 0; i < mc_recoverable_range_len; i++)
755 		if ((nip >= mc_recoverable_range[i].start_addr) &&
756 		    (nip < mc_recoverable_range[i].end_addr))
757 		    return mc_recoverable_range[i].recover_addr;
758 	return 0;
759 }
760 
761 bool opal_mce_check_early_recovery(struct pt_regs *regs)
762 {
763 	uint64_t recover_addr = 0;
764 
765 	if (!opal.base || !opal.size)
766 		goto out;
767 
768 	if ((regs->nip >= opal.base) &&
769 			(regs->nip < (opal.base + opal.size)))
770 		recover_addr = find_recovery_address(regs->nip);
771 
772 	/*
773 	 * Setup regs->nip to rfi into fixup address.
774 	 */
775 	if (recover_addr)
776 		regs_set_return_ip(regs, recover_addr);
777 
778 out:
779 	return !!recover_addr;
780 }
781 
782 static int __init opal_sysfs_init(void)
783 {
784 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
785 	if (!opal_kobj) {
786 		pr_warn("kobject_create_and_add opal failed\n");
787 		return -ENOMEM;
788 	}
789 
790 	return 0;
791 }
792 
793 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
794 				struct bin_attribute *bin_attr, char *buf,
795 				loff_t off, size_t count)
796 {
797 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
798 				       bin_attr->size);
799 }
800 
801 static int opal_add_one_export(struct kobject *parent, const char *export_name,
802 			       struct device_node *np, const char *prop_name)
803 {
804 	struct bin_attribute *attr = NULL;
805 	const char *name = NULL;
806 	u64 vals[2];
807 	int rc;
808 
809 	rc = of_property_read_u64_array(np, prop_name, &vals[0], 2);
810 	if (rc)
811 		goto out;
812 
813 	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
814 	if (!attr) {
815 		rc = -ENOMEM;
816 		goto out;
817 	}
818 	name = kstrdup(export_name, GFP_KERNEL);
819 	if (!name) {
820 		rc = -ENOMEM;
821 		goto out;
822 	}
823 
824 	sysfs_bin_attr_init(attr);
825 	attr->attr.name = name;
826 	attr->attr.mode = 0400;
827 	attr->read = export_attr_read;
828 	attr->private = __va(vals[0]);
829 	attr->size = vals[1];
830 
831 	rc = sysfs_create_bin_file(parent, attr);
832 out:
833 	if (rc) {
834 		kfree(name);
835 		kfree(attr);
836 	}
837 
838 	return rc;
839 }
840 
841 static void opal_add_exported_attrs(struct device_node *np,
842 				    struct kobject *kobj)
843 {
844 	struct device_node *child;
845 	struct property *prop;
846 
847 	for_each_property_of_node(np, prop) {
848 		int rc;
849 
850 		if (!strcmp(prop->name, "name") ||
851 		    !strcmp(prop->name, "phandle"))
852 			continue;
853 
854 		rc = opal_add_one_export(kobj, prop->name, np, prop->name);
855 		if (rc) {
856 			pr_warn("Unable to add export %pOF/%s, rc = %d!\n",
857 				np, prop->name, rc);
858 		}
859 	}
860 
861 	for_each_child_of_node(np, child) {
862 		struct kobject *child_kobj;
863 
864 		child_kobj = kobject_create_and_add(child->name, kobj);
865 		if (!child_kobj) {
866 			pr_err("Unable to create export dir for %pOF\n", child);
867 			continue;
868 		}
869 
870 		opal_add_exported_attrs(child, child_kobj);
871 	}
872 }
873 
874 /*
875  * opal_export_attrs: creates a sysfs node for each property listed in
876  * the device-tree under /ibm,opal/firmware/exports/
877  * All new sysfs nodes are created under /opal/exports/.
878  * This allows for reserved memory regions (e.g. HDAT) to be read.
879  * The new sysfs nodes are only readable by root.
880  */
881 static void opal_export_attrs(void)
882 {
883 	struct device_node *np;
884 	struct kobject *kobj;
885 	int rc;
886 
887 	np = of_find_node_by_path("/ibm,opal/firmware/exports");
888 	if (!np)
889 		return;
890 
891 	/* Create new 'exports' directory - /sys/firmware/opal/exports */
892 	kobj = kobject_create_and_add("exports", opal_kobj);
893 	if (!kobj) {
894 		pr_warn("kobject_create_and_add() of exports failed\n");
895 		return;
896 	}
897 
898 	opal_add_exported_attrs(np, kobj);
899 
900 	/*
901 	 * NB: symbol_map existed before the generic export interface so it
902 	 * lives under the top level opal_kobj.
903 	 */
904 	rc = opal_add_one_export(opal_kobj, "symbol_map",
905 				 np->parent, "symbol-map");
906 	if (rc)
907 		pr_warn("Error %d creating OPAL symbols file\n", rc);
908 
909 	of_node_put(np);
910 }
911 
912 static void __init opal_dump_region_init(void)
913 {
914 	void *addr;
915 	uint64_t size;
916 	int rc;
917 
918 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
919 		return;
920 
921 	/* Register kernel log buffer */
922 	addr = log_buf_addr_get();
923 	if (addr == NULL)
924 		return;
925 
926 	size = log_buf_len_get();
927 	if (size == 0)
928 		return;
929 
930 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
931 				       __pa(addr), size);
932 	/* Don't warn if this is just an older OPAL that doesn't
933 	 * know about that call
934 	 */
935 	if (rc && rc != OPAL_UNSUPPORTED)
936 		pr_warn("DUMP: Failed to register kernel log buffer. "
937 			"rc = %d\n", rc);
938 }
939 
940 static void __init opal_pdev_init(const char *compatible)
941 {
942 	struct device_node *np;
943 
944 	for_each_compatible_node(np, NULL, compatible)
945 		of_platform_device_create(np, NULL, NULL);
946 }
947 
948 static void __init opal_imc_init_dev(void)
949 {
950 	struct device_node *np;
951 
952 	np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
953 	if (np)
954 		of_platform_device_create(np, NULL, NULL);
955 }
956 
957 static int kopald(void *unused)
958 {
959 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
960 
961 	set_freezable();
962 	do {
963 		try_to_freeze();
964 
965 		opal_handle_events();
966 
967 		set_current_state(TASK_INTERRUPTIBLE);
968 		if (opal_have_pending_events())
969 			__set_current_state(TASK_RUNNING);
970 		else
971 			schedule_timeout(timeout);
972 
973 	} while (!kthread_should_stop());
974 
975 	return 0;
976 }
977 
978 void opal_wake_poller(void)
979 {
980 	if (kopald_tsk)
981 		wake_up_process(kopald_tsk);
982 }
983 
984 static void __init opal_init_heartbeat(void)
985 {
986 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
987 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
988 				 &opal_heartbeat) != 0)
989 		opal_heartbeat = 0;
990 
991 	if (opal_heartbeat)
992 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
993 }
994 
995 static int __init opal_init(void)
996 {
997 	struct device_node *np, *consoles, *leds;
998 	int rc;
999 
1000 	opal_node = of_find_node_by_path("/ibm,opal");
1001 	if (!opal_node) {
1002 		pr_warn("Device node not found\n");
1003 		return -ENODEV;
1004 	}
1005 
1006 	/* Register OPAL consoles if any ports */
1007 	consoles = of_find_node_by_path("/ibm,opal/consoles");
1008 	if (consoles) {
1009 		for_each_child_of_node(consoles, np) {
1010 			if (!of_node_name_eq(np, "serial"))
1011 				continue;
1012 			of_platform_device_create(np, NULL, NULL);
1013 		}
1014 		of_node_put(consoles);
1015 	}
1016 
1017 	/* Initialise OPAL messaging system */
1018 	opal_message_init(opal_node);
1019 
1020 	/* Initialise OPAL asynchronous completion interface */
1021 	opal_async_comp_init();
1022 
1023 	/* Initialise OPAL sensor interface */
1024 	opal_sensor_init();
1025 
1026 	/* Initialise OPAL hypervisor maintainence interrupt handling */
1027 	opal_hmi_handler_init();
1028 
1029 	/* Create i2c platform devices */
1030 	opal_pdev_init("ibm,opal-i2c");
1031 
1032 	/* Handle non-volatile memory devices */
1033 	opal_pdev_init("pmem-region");
1034 
1035 	/* Setup a heatbeat thread if requested by OPAL */
1036 	opal_init_heartbeat();
1037 
1038 	/* Detect In-Memory Collection counters and create devices*/
1039 	opal_imc_init_dev();
1040 
1041 	/* Create leds platform devices */
1042 	leds = of_find_node_by_path("/ibm,opal/leds");
1043 	if (leds) {
1044 		of_platform_device_create(leds, "opal_leds", NULL);
1045 		of_node_put(leds);
1046 	}
1047 
1048 	/* Initialise OPAL message log interface */
1049 	opal_msglog_init();
1050 
1051 	/* Create "opal" kobject under /sys/firmware */
1052 	rc = opal_sysfs_init();
1053 	if (rc == 0) {
1054 		/* Setup dump region interface */
1055 		opal_dump_region_init();
1056 		/* Setup error log interface */
1057 		rc = opal_elog_init();
1058 		/* Setup code update interface */
1059 		opal_flash_update_init();
1060 		/* Setup platform dump extract interface */
1061 		opal_platform_dump_init();
1062 		/* Setup system parameters interface */
1063 		opal_sys_param_init();
1064 		/* Setup message log sysfs interface. */
1065 		opal_msglog_sysfs_init();
1066 		/* Add all export properties*/
1067 		opal_export_attrs();
1068 	}
1069 
1070 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
1071 	opal_pdev_init("ibm,opal-ipmi");
1072 	opal_pdev_init("ibm,opal-flash");
1073 	opal_pdev_init("ibm,opal-prd");
1074 
1075 	/* Initialise platform device: oppanel interface */
1076 	opal_pdev_init("ibm,opal-oppanel");
1077 
1078 	/* Initialise OPAL kmsg dumper for flushing console on panic */
1079 	opal_kmsg_init();
1080 
1081 	/* Initialise OPAL powercap interface */
1082 	opal_powercap_init();
1083 
1084 	/* Initialise OPAL Power-Shifting-Ratio interface */
1085 	opal_psr_init();
1086 
1087 	/* Initialise OPAL sensor groups */
1088 	opal_sensor_groups_init();
1089 
1090 	/* Initialise OPAL Power control interface */
1091 	opal_power_control_init();
1092 
1093 	/* Initialize OPAL secure variables */
1094 	opal_pdev_init("ibm,secvar-backend");
1095 
1096 	return 0;
1097 }
1098 machine_subsys_initcall(powernv, opal_init);
1099 
1100 void opal_shutdown(void)
1101 {
1102 	long rc = OPAL_BUSY;
1103 
1104 	opal_event_shutdown();
1105 
1106 	/*
1107 	 * Then sync with OPAL which ensure anything that can
1108 	 * potentially write to our memory has completed such
1109 	 * as an ongoing dump retrieval
1110 	 */
1111 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
1112 		rc = opal_sync_host_reboot();
1113 		if (rc == OPAL_BUSY)
1114 			opal_poll_events(NULL);
1115 		else
1116 			mdelay(10);
1117 	}
1118 
1119 	/* Unregister memory dump region */
1120 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
1121 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
1122 }
1123 
1124 /* Export this so that test modules can use it */
1125 EXPORT_SYMBOL_GPL(opal_invalid_call);
1126 EXPORT_SYMBOL_GPL(opal_xscom_read);
1127 EXPORT_SYMBOL_GPL(opal_xscom_write);
1128 EXPORT_SYMBOL_GPL(opal_ipmi_send);
1129 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1130 EXPORT_SYMBOL_GPL(opal_flash_read);
1131 EXPORT_SYMBOL_GPL(opal_flash_write);
1132 EXPORT_SYMBOL_GPL(opal_flash_erase);
1133 EXPORT_SYMBOL_GPL(opal_prd_msg);
1134 EXPORT_SYMBOL_GPL(opal_check_token);
1135 
1136 /* Convert a region of vmalloc memory to an opal sg list */
1137 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1138 					     unsigned long vmalloc_size)
1139 {
1140 	struct opal_sg_list *sg, *first = NULL;
1141 	unsigned long i = 0;
1142 
1143 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1144 	if (!sg)
1145 		goto nomem;
1146 
1147 	first = sg;
1148 
1149 	while (vmalloc_size > 0) {
1150 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1151 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
1152 
1153 		sg->entry[i].data = cpu_to_be64(data);
1154 		sg->entry[i].length = cpu_to_be64(length);
1155 		i++;
1156 
1157 		if (i >= SG_ENTRIES_PER_NODE) {
1158 			struct opal_sg_list *next;
1159 
1160 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1161 			if (!next)
1162 				goto nomem;
1163 
1164 			sg->length = cpu_to_be64(
1165 					i * sizeof(struct opal_sg_entry) + 16);
1166 			i = 0;
1167 			sg->next = cpu_to_be64(__pa(next));
1168 			sg = next;
1169 		}
1170 
1171 		vmalloc_addr += length;
1172 		vmalloc_size -= length;
1173 	}
1174 
1175 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1176 
1177 	return first;
1178 
1179 nomem:
1180 	pr_err("%s : Failed to allocate memory\n", __func__);
1181 	opal_free_sg_list(first);
1182 	return NULL;
1183 }
1184 
1185 void opal_free_sg_list(struct opal_sg_list *sg)
1186 {
1187 	while (sg) {
1188 		uint64_t next = be64_to_cpu(sg->next);
1189 
1190 		kfree(sg);
1191 
1192 		if (next)
1193 			sg = __va(next);
1194 		else
1195 			sg = NULL;
1196 	}
1197 }
1198 
1199 int opal_error_code(int rc)
1200 {
1201 	switch (rc) {
1202 	case OPAL_SUCCESS:		return 0;
1203 
1204 	case OPAL_PARAMETER:		return -EINVAL;
1205 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
1206 	case OPAL_BUSY:
1207 	case OPAL_BUSY_EVENT:		return -EBUSY;
1208 	case OPAL_NO_MEM:		return -ENOMEM;
1209 	case OPAL_PERMISSION:		return -EPERM;
1210 
1211 	case OPAL_UNSUPPORTED:		return -EIO;
1212 	case OPAL_HARDWARE:		return -EIO;
1213 	case OPAL_INTERNAL_ERROR:	return -EIO;
1214 	case OPAL_TIMEOUT:		return -ETIMEDOUT;
1215 	default:
1216 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1217 		return -EIO;
1218 	}
1219 }
1220 
1221 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1222 {
1223 	int rc;
1224 
1225 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
1226 		rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1227 		if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1228 			pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1229 	}
1230 }
1231 
1232 EXPORT_SYMBOL_GPL(opal_poll_events);
1233 EXPORT_SYMBOL_GPL(opal_rtc_read);
1234 EXPORT_SYMBOL_GPL(opal_rtc_write);
1235 EXPORT_SYMBOL_GPL(opal_tpo_read);
1236 EXPORT_SYMBOL_GPL(opal_tpo_write);
1237 EXPORT_SYMBOL_GPL(opal_i2c_request);
1238 /* Export these symbols for PowerNV LED class driver */
1239 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1240 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1241 /* Export this symbol for PowerNV Operator Panel class driver */
1242 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1243 /* Export this for KVM */
1244 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1245 EXPORT_SYMBOL_GPL(opal_int_eoi);
1246 EXPORT_SYMBOL_GPL(opal_error_code);
1247 /* Export the below symbol for NX compression */
1248 EXPORT_SYMBOL(opal_nx_coproc_init);
1249