1 /*
2  * PowerNV OPAL high level interfaces
3  *
4  * Copyright 2011 IBM Corp.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #define pr_fmt(fmt)	"opal: " fmt
13 
14 #include <linux/printk.h>
15 #include <linux/types.h>
16 #include <linux/of.h>
17 #include <linux/of_fdt.h>
18 #include <linux/of_platform.h>
19 #include <linux/interrupt.h>
20 #include <linux/notifier.h>
21 #include <linux/slab.h>
22 #include <linux/sched.h>
23 #include <linux/kobject.h>
24 #include <linux/delay.h>
25 #include <linux/memblock.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
28 
29 #include <asm/machdep.h>
30 #include <asm/opal.h>
31 #include <asm/firmware.h>
32 #include <asm/mce.h>
33 
34 #include "powernv.h"
35 
36 /* /sys/firmware/opal */
37 struct kobject *opal_kobj;
38 
39 struct opal {
40 	u64 base;
41 	u64 entry;
42 	u64 size;
43 } opal;
44 
45 struct mcheck_recoverable_range {
46 	u64 start_addr;
47 	u64 end_addr;
48 	u64 recover_addr;
49 };
50 
51 static struct mcheck_recoverable_range *mc_recoverable_range;
52 static int mc_recoverable_range_len;
53 
54 struct device_node *opal_node;
55 static DEFINE_SPINLOCK(opal_write_lock);
56 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
57 static uint32_t opal_heartbeat;
58 
59 static void opal_reinit_cores(void)
60 {
61 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
62 	 *
63 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
64 	 * also restore HIDs and other SPRs to their original value
65 	 * but it might clobber a bunch.
66 	 */
67 #ifdef __BIG_ENDIAN__
68 	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
69 #else
70 	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
71 #endif
72 }
73 
74 int __init early_init_dt_scan_opal(unsigned long node,
75 				   const char *uname, int depth, void *data)
76 {
77 	const void *basep, *entryp, *sizep;
78 	int basesz, entrysz, runtimesz;
79 
80 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
81 		return 0;
82 
83 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
84 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
85 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
86 
87 	if (!basep || !entryp || !sizep)
88 		return 1;
89 
90 	opal.base = of_read_number(basep, basesz/4);
91 	opal.entry = of_read_number(entryp, entrysz/4);
92 	opal.size = of_read_number(sizep, runtimesz/4);
93 
94 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
95 		 opal.base, basep, basesz);
96 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
97 		 opal.entry, entryp, entrysz);
98 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
99 		 opal.size, sizep, runtimesz);
100 
101 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
102 		powerpc_firmware_features |= FW_FEATURE_OPAL;
103 		pr_info("OPAL detected !\n");
104 	} else {
105 		panic("OPAL != V3 detected, no longer supported.\n");
106 	}
107 
108 	/* Reinit all cores with the right endian */
109 	opal_reinit_cores();
110 
111 	/* Restore some bits */
112 	if (cur_cpu_spec->cpu_restore)
113 		cur_cpu_spec->cpu_restore();
114 
115 	return 1;
116 }
117 
118 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
119 				   const char *uname, int depth, void *data)
120 {
121 	int i, psize, size;
122 	const __be32 *prop;
123 
124 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
125 		return 0;
126 
127 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
128 
129 	if (!prop)
130 		return 1;
131 
132 	pr_debug("Found machine check recoverable ranges.\n");
133 
134 	/*
135 	 * Calculate number of available entries.
136 	 *
137 	 * Each recoverable address range entry is (start address, len,
138 	 * recovery address), 2 cells each for start and recovery address,
139 	 * 1 cell for len, totalling 5 cells per entry.
140 	 */
141 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
142 
143 	/* Sanity check */
144 	if (!mc_recoverable_range_len)
145 		return 1;
146 
147 	/* Size required to hold all the entries. */
148 	size = mc_recoverable_range_len *
149 			sizeof(struct mcheck_recoverable_range);
150 
151 	/*
152 	 * Allocate a buffer to hold the MC recoverable ranges. We would be
153 	 * accessing them in real mode, hence it needs to be within
154 	 * RMO region.
155 	 */
156 	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
157 							ppc64_rma_size));
158 	memset(mc_recoverable_range, 0, size);
159 
160 	for (i = 0; i < mc_recoverable_range_len; i++) {
161 		mc_recoverable_range[i].start_addr =
162 					of_read_number(prop + (i * 5) + 0, 2);
163 		mc_recoverable_range[i].end_addr =
164 					mc_recoverable_range[i].start_addr +
165 					of_read_number(prop + (i * 5) + 2, 1);
166 		mc_recoverable_range[i].recover_addr =
167 					of_read_number(prop + (i * 5) + 3, 2);
168 
169 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
170 				mc_recoverable_range[i].start_addr,
171 				mc_recoverable_range[i].end_addr,
172 				mc_recoverable_range[i].recover_addr);
173 	}
174 	return 1;
175 }
176 
177 static int __init opal_register_exception_handlers(void)
178 {
179 #ifdef __BIG_ENDIAN__
180 	u64 glue;
181 
182 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
183 		return -ENODEV;
184 
185 	/* Hookup some exception handlers except machine check. We use the
186 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
187 	 */
188 	glue = 0x7000;
189 
190 	/*
191 	 * Check if we are running on newer firmware that exports
192 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
193 	 * the HMI interrupt and we catch it directly in Linux.
194 	 *
195 	 * For older firmware (i.e currently released POWER8 System Firmware
196 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
197 	 * patch the HMI vector and handle it inside OPAL firmware.
198 	 *
199 	 * For newer firmware (in development/yet to be released) we will
200 	 * start catching/handling HMI directly in Linux.
201 	 */
202 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
203 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
204 		opal_register_exception_handler(
205 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
206 				0, glue);
207 		glue += 128;
208 	}
209 
210 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
211 #endif
212 
213 	return 0;
214 }
215 machine_early_initcall(powernv, opal_register_exception_handlers);
216 
217 /*
218  * Opal message notifier based on message type. Allow subscribers to get
219  * notified for specific messgae type.
220  */
221 int opal_message_notifier_register(enum opal_msg_type msg_type,
222 					struct notifier_block *nb)
223 {
224 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
225 		pr_warning("%s: Invalid arguments, msg_type:%d\n",
226 			   __func__, msg_type);
227 		return -EINVAL;
228 	}
229 
230 	return atomic_notifier_chain_register(
231 				&opal_msg_notifier_head[msg_type], nb);
232 }
233 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
234 
235 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
236 				     struct notifier_block *nb)
237 {
238 	return atomic_notifier_chain_unregister(
239 			&opal_msg_notifier_head[msg_type], nb);
240 }
241 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
242 
243 static void opal_message_do_notify(uint32_t msg_type, void *msg)
244 {
245 	/* notify subscribers */
246 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
247 					msg_type, msg);
248 }
249 
250 static void opal_handle_message(void)
251 {
252 	s64 ret;
253 	/*
254 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
255 	 * value in /proc/device-tree.
256 	 */
257 	static struct opal_msg msg;
258 	u32 type;
259 
260 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
261 	/* No opal message pending. */
262 	if (ret == OPAL_RESOURCE)
263 		return;
264 
265 	/* check for errors. */
266 	if (ret) {
267 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
268 				__func__, ret);
269 		return;
270 	}
271 
272 	type = be32_to_cpu(msg.msg_type);
273 
274 	/* Sanity check */
275 	if (type >= OPAL_MSG_TYPE_MAX) {
276 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
277 		return;
278 	}
279 	opal_message_do_notify(type, (void *)&msg);
280 }
281 
282 static irqreturn_t opal_message_notify(int irq, void *data)
283 {
284 	opal_handle_message();
285 	return IRQ_HANDLED;
286 }
287 
288 static int __init opal_message_init(void)
289 {
290 	int ret, i, irq;
291 
292 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
293 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
294 
295 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
296 	if (!irq) {
297 		pr_err("%s: Can't register OPAL event irq (%d)\n",
298 		       __func__, irq);
299 		return irq;
300 	}
301 
302 	ret = request_irq(irq, opal_message_notify,
303 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
304 	if (ret) {
305 		pr_err("%s: Can't request OPAL event irq (%d)\n",
306 		       __func__, ret);
307 		return ret;
308 	}
309 
310 	return 0;
311 }
312 
313 int opal_get_chars(uint32_t vtermno, char *buf, int count)
314 {
315 	s64 rc;
316 	__be64 evt, len;
317 
318 	if (!opal.entry)
319 		return -ENODEV;
320 	opal_poll_events(&evt);
321 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
322 		return 0;
323 	len = cpu_to_be64(count);
324 	rc = opal_console_read(vtermno, &len, buf);
325 	if (rc == OPAL_SUCCESS)
326 		return be64_to_cpu(len);
327 	return 0;
328 }
329 
330 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
331 {
332 	int written = 0;
333 	__be64 olen;
334 	s64 len, rc;
335 	unsigned long flags;
336 	__be64 evt;
337 
338 	if (!opal.entry)
339 		return -ENODEV;
340 
341 	/* We want put_chars to be atomic to avoid mangling of hvsi
342 	 * packets. To do that, we first test for room and return
343 	 * -EAGAIN if there isn't enough.
344 	 *
345 	 * Unfortunately, opal_console_write_buffer_space() doesn't
346 	 * appear to work on opal v1, so we just assume there is
347 	 * enough room and be done with it
348 	 */
349 	spin_lock_irqsave(&opal_write_lock, flags);
350 	rc = opal_console_write_buffer_space(vtermno, &olen);
351 	len = be64_to_cpu(olen);
352 	if (rc || len < total_len) {
353 		spin_unlock_irqrestore(&opal_write_lock, flags);
354 		/* Closed -> drop characters */
355 		if (rc)
356 			return total_len;
357 		opal_poll_events(NULL);
358 		return -EAGAIN;
359 	}
360 
361 	/* We still try to handle partial completions, though they
362 	 * should no longer happen.
363 	 */
364 	rc = OPAL_BUSY;
365 	while(total_len > 0 && (rc == OPAL_BUSY ||
366 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
367 		olen = cpu_to_be64(total_len);
368 		rc = opal_console_write(vtermno, &olen, data);
369 		len = be64_to_cpu(olen);
370 
371 		/* Closed or other error drop */
372 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
373 		    rc != OPAL_BUSY_EVENT) {
374 			written = total_len;
375 			break;
376 		}
377 		if (rc == OPAL_SUCCESS) {
378 			total_len -= len;
379 			data += len;
380 			written += len;
381 		}
382 		/* This is a bit nasty but we need that for the console to
383 		 * flush when there aren't any interrupts. We will clean
384 		 * things a bit later to limit that to synchronous path
385 		 * such as the kernel console and xmon/udbg
386 		 */
387 		do
388 			opal_poll_events(&evt);
389 		while(rc == OPAL_SUCCESS &&
390 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
391 	}
392 	spin_unlock_irqrestore(&opal_write_lock, flags);
393 	return written;
394 }
395 
396 static int opal_recover_mce(struct pt_regs *regs,
397 					struct machine_check_event *evt)
398 {
399 	int recovered = 0;
400 	uint64_t ea = get_mce_fault_addr(evt);
401 
402 	if (!(regs->msr & MSR_RI)) {
403 		/* If MSR_RI isn't set, we cannot recover */
404 		recovered = 0;
405 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
406 		/* Platform corrected itself */
407 		recovered = 1;
408 	} else if (ea && !is_kernel_addr(ea)) {
409 		/*
410 		 * Faulting address is not in kernel text. We should be fine.
411 		 * We need to find which process uses this address.
412 		 * For now, kill the task if we have received exception when
413 		 * in userspace.
414 		 *
415 		 * TODO: Queue up this address for hwpoisioning later.
416 		 */
417 		if (user_mode(regs) && !is_global_init(current)) {
418 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
419 			recovered = 1;
420 		} else
421 			recovered = 0;
422 	} else if (user_mode(regs) && !is_global_init(current) &&
423 		evt->severity == MCE_SEV_ERROR_SYNC) {
424 		/*
425 		 * If we have received a synchronous error when in userspace
426 		 * kill the task.
427 		 */
428 		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
429 		recovered = 1;
430 	}
431 	return recovered;
432 }
433 
434 int opal_machine_check(struct pt_regs *regs)
435 {
436 	struct machine_check_event evt;
437 	int ret;
438 
439 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
440 		return 0;
441 
442 	/* Print things out */
443 	if (evt.version != MCE_V1) {
444 		pr_err("Machine Check Exception, Unknown event version %d !\n",
445 		       evt.version);
446 		return 0;
447 	}
448 	machine_check_print_event_info(&evt);
449 
450 	if (opal_recover_mce(regs, &evt))
451 		return 1;
452 
453 	/*
454 	 * Unrecovered machine check, we are heading to panic path.
455 	 *
456 	 * We may have hit this MCE in very early stage of kernel
457 	 * initialization even before opal-prd has started running. If
458 	 * this is the case then this MCE error may go un-noticed or
459 	 * un-analyzed if we go down panic path. We need to inform
460 	 * BMC/OCC about this error so that they can collect relevant
461 	 * data for error analysis before rebooting.
462 	 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
463 	 * This function may not return on BMC based system.
464 	 */
465 	ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
466 			"Unrecoverable Machine Check exception");
467 	if (ret == OPAL_UNSUPPORTED) {
468 		pr_emerg("Reboot type %d not supported\n",
469 					OPAL_REBOOT_PLATFORM_ERROR);
470 	}
471 
472 	/*
473 	 * We reached here. There can be three possibilities:
474 	 * 1. We are running on a firmware level that do not support
475 	 *    opal_cec_reboot2()
476 	 * 2. We are running on a firmware level that do not support
477 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
478 	 * 3. We are running on FSP based system that does not need opal
479 	 *    to trigger checkstop explicitly for error analysis. The FSP
480 	 *    PRD component would have already got notified about this
481 	 *    error through other channels.
482 	 *
483 	 * If hardware marked this as an unrecoverable MCE, we are
484 	 * going to panic anyway. Even if it didn't, it's not safe to
485 	 * continue at this point, so we should explicitly panic.
486 	 */
487 
488 	panic("PowerNV Unrecovered Machine Check");
489 	return 0;
490 }
491 
492 /* Early hmi handler called in real mode. */
493 int opal_hmi_exception_early(struct pt_regs *regs)
494 {
495 	s64 rc;
496 
497 	/*
498 	 * call opal hmi handler. Pass paca address as token.
499 	 * The return value OPAL_SUCCESS is an indication that there is
500 	 * an HMI event generated waiting to pull by Linux.
501 	 */
502 	rc = opal_handle_hmi();
503 	if (rc == OPAL_SUCCESS) {
504 		local_paca->hmi_event_available = 1;
505 		return 1;
506 	}
507 	return 0;
508 }
509 
510 /* HMI exception handler called in virtual mode during check_irq_replay. */
511 int opal_handle_hmi_exception(struct pt_regs *regs)
512 {
513 	s64 rc;
514 	__be64 evt = 0;
515 
516 	/*
517 	 * Check if HMI event is available.
518 	 * if Yes, then call opal_poll_events to pull opal messages and
519 	 * process them.
520 	 */
521 	if (!local_paca->hmi_event_available)
522 		return 0;
523 
524 	local_paca->hmi_event_available = 0;
525 	rc = opal_poll_events(&evt);
526 	if (rc == OPAL_SUCCESS && evt)
527 		opal_handle_events(be64_to_cpu(evt));
528 
529 	return 1;
530 }
531 
532 static uint64_t find_recovery_address(uint64_t nip)
533 {
534 	int i;
535 
536 	for (i = 0; i < mc_recoverable_range_len; i++)
537 		if ((nip >= mc_recoverable_range[i].start_addr) &&
538 		    (nip < mc_recoverable_range[i].end_addr))
539 		    return mc_recoverable_range[i].recover_addr;
540 	return 0;
541 }
542 
543 bool opal_mce_check_early_recovery(struct pt_regs *regs)
544 {
545 	uint64_t recover_addr = 0;
546 
547 	if (!opal.base || !opal.size)
548 		goto out;
549 
550 	if ((regs->nip >= opal.base) &&
551 			(regs->nip < (opal.base + opal.size)))
552 		recover_addr = find_recovery_address(regs->nip);
553 
554 	/*
555 	 * Setup regs->nip to rfi into fixup address.
556 	 */
557 	if (recover_addr)
558 		regs->nip = recover_addr;
559 
560 out:
561 	return !!recover_addr;
562 }
563 
564 static int opal_sysfs_init(void)
565 {
566 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
567 	if (!opal_kobj) {
568 		pr_warn("kobject_create_and_add opal failed\n");
569 		return -ENOMEM;
570 	}
571 
572 	return 0;
573 }
574 
575 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
576 			       struct bin_attribute *bin_attr,
577 			       char *buf, loff_t off, size_t count)
578 {
579 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
580 				       bin_attr->size);
581 }
582 
583 static BIN_ATTR_RO(symbol_map, 0);
584 
585 static void opal_export_symmap(void)
586 {
587 	const __be64 *syms;
588 	unsigned int size;
589 	struct device_node *fw;
590 	int rc;
591 
592 	fw = of_find_node_by_path("/ibm,opal/firmware");
593 	if (!fw)
594 		return;
595 	syms = of_get_property(fw, "symbol-map", &size);
596 	if (!syms || size != 2 * sizeof(__be64))
597 		return;
598 
599 	/* Setup attributes */
600 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
601 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
602 
603 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
604 	if (rc)
605 		pr_warn("Error %d creating OPAL symbols file\n", rc);
606 }
607 
608 static void __init opal_dump_region_init(void)
609 {
610 	void *addr;
611 	uint64_t size;
612 	int rc;
613 
614 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
615 		return;
616 
617 	/* Register kernel log buffer */
618 	addr = log_buf_addr_get();
619 	if (addr == NULL)
620 		return;
621 
622 	size = log_buf_len_get();
623 	if (size == 0)
624 		return;
625 
626 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
627 				       __pa(addr), size);
628 	/* Don't warn if this is just an older OPAL that doesn't
629 	 * know about that call
630 	 */
631 	if (rc && rc != OPAL_UNSUPPORTED)
632 		pr_warn("DUMP: Failed to register kernel log buffer. "
633 			"rc = %d\n", rc);
634 }
635 
636 static void opal_pdev_init(struct device_node *opal_node,
637 		const char *compatible)
638 {
639 	struct device_node *np;
640 
641 	for_each_child_of_node(opal_node, np)
642 		if (of_device_is_compatible(np, compatible))
643 			of_platform_device_create(np, NULL, NULL);
644 }
645 
646 static void opal_i2c_create_devs(void)
647 {
648 	struct device_node *np;
649 
650 	for_each_compatible_node(np, NULL, "ibm,opal-i2c")
651 		of_platform_device_create(np, NULL, NULL);
652 }
653 
654 static int kopald(void *unused)
655 {
656 	__be64 events;
657 
658 	set_freezable();
659 	do {
660 		try_to_freeze();
661 		opal_poll_events(&events);
662 		opal_handle_events(be64_to_cpu(events));
663 		msleep_interruptible(opal_heartbeat);
664 	} while (!kthread_should_stop());
665 
666 	return 0;
667 }
668 
669 static void opal_init_heartbeat(void)
670 {
671 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
672 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
673 				 &opal_heartbeat) != 0)
674 		opal_heartbeat = 0;
675 
676 	if (opal_heartbeat)
677 		kthread_run(kopald, NULL, "kopald");
678 }
679 
680 static int __init opal_init(void)
681 {
682 	struct device_node *np, *consoles, *leds;
683 	int rc;
684 
685 	opal_node = of_find_node_by_path("/ibm,opal");
686 	if (!opal_node) {
687 		pr_warn("Device node not found\n");
688 		return -ENODEV;
689 	}
690 
691 	/* Register OPAL consoles if any ports */
692 	consoles = of_find_node_by_path("/ibm,opal/consoles");
693 	if (consoles) {
694 		for_each_child_of_node(consoles, np) {
695 			if (strcmp(np->name, "serial"))
696 				continue;
697 			of_platform_device_create(np, NULL, NULL);
698 		}
699 		of_node_put(consoles);
700 	}
701 
702 	/* Initialise OPAL messaging system */
703 	opal_message_init();
704 
705 	/* Initialise OPAL asynchronous completion interface */
706 	opal_async_comp_init();
707 
708 	/* Initialise OPAL sensor interface */
709 	opal_sensor_init();
710 
711 	/* Initialise OPAL hypervisor maintainence interrupt handling */
712 	opal_hmi_handler_init();
713 
714 	/* Create i2c platform devices */
715 	opal_i2c_create_devs();
716 
717 	/* Setup a heatbeat thread if requested by OPAL */
718 	opal_init_heartbeat();
719 
720 	/* Create leds platform devices */
721 	leds = of_find_node_by_path("/ibm,opal/leds");
722 	if (leds) {
723 		of_platform_device_create(leds, "opal_leds", NULL);
724 		of_node_put(leds);
725 	}
726 
727 	/* Initialise OPAL message log interface */
728 	opal_msglog_init();
729 
730 	/* Create "opal" kobject under /sys/firmware */
731 	rc = opal_sysfs_init();
732 	if (rc == 0) {
733 		/* Export symbol map to userspace */
734 		opal_export_symmap();
735 		/* Setup dump region interface */
736 		opal_dump_region_init();
737 		/* Setup error log interface */
738 		rc = opal_elog_init();
739 		/* Setup code update interface */
740 		opal_flash_update_init();
741 		/* Setup platform dump extract interface */
742 		opal_platform_dump_init();
743 		/* Setup system parameters interface */
744 		opal_sys_param_init();
745 		/* Setup message log sysfs interface. */
746 		opal_msglog_sysfs_init();
747 	}
748 
749 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
750 	opal_pdev_init(opal_node, "ibm,opal-ipmi");
751 	opal_pdev_init(opal_node, "ibm,opal-flash");
752 	opal_pdev_init(opal_node, "ibm,opal-prd");
753 
754 	/* Initialise OPAL kmsg dumper for flushing console on panic */
755 	opal_kmsg_init();
756 
757 	return 0;
758 }
759 machine_subsys_initcall(powernv, opal_init);
760 
761 void opal_shutdown(void)
762 {
763 	long rc = OPAL_BUSY;
764 
765 	opal_event_shutdown();
766 
767 	/*
768 	 * Then sync with OPAL which ensure anything that can
769 	 * potentially write to our memory has completed such
770 	 * as an ongoing dump retrieval
771 	 */
772 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
773 		rc = opal_sync_host_reboot();
774 		if (rc == OPAL_BUSY)
775 			opal_poll_events(NULL);
776 		else
777 			mdelay(10);
778 	}
779 
780 	/* Unregister memory dump region */
781 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
782 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
783 }
784 
785 /* Export this so that test modules can use it */
786 EXPORT_SYMBOL_GPL(opal_invalid_call);
787 EXPORT_SYMBOL_GPL(opal_xscom_read);
788 EXPORT_SYMBOL_GPL(opal_xscom_write);
789 EXPORT_SYMBOL_GPL(opal_ipmi_send);
790 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
791 EXPORT_SYMBOL_GPL(opal_flash_read);
792 EXPORT_SYMBOL_GPL(opal_flash_write);
793 EXPORT_SYMBOL_GPL(opal_flash_erase);
794 EXPORT_SYMBOL_GPL(opal_prd_msg);
795 
796 /* Convert a region of vmalloc memory to an opal sg list */
797 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
798 					     unsigned long vmalloc_size)
799 {
800 	struct opal_sg_list *sg, *first = NULL;
801 	unsigned long i = 0;
802 
803 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
804 	if (!sg)
805 		goto nomem;
806 
807 	first = sg;
808 
809 	while (vmalloc_size > 0) {
810 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
811 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
812 
813 		sg->entry[i].data = cpu_to_be64(data);
814 		sg->entry[i].length = cpu_to_be64(length);
815 		i++;
816 
817 		if (i >= SG_ENTRIES_PER_NODE) {
818 			struct opal_sg_list *next;
819 
820 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
821 			if (!next)
822 				goto nomem;
823 
824 			sg->length = cpu_to_be64(
825 					i * sizeof(struct opal_sg_entry) + 16);
826 			i = 0;
827 			sg->next = cpu_to_be64(__pa(next));
828 			sg = next;
829 		}
830 
831 		vmalloc_addr += length;
832 		vmalloc_size -= length;
833 	}
834 
835 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
836 
837 	return first;
838 
839 nomem:
840 	pr_err("%s : Failed to allocate memory\n", __func__);
841 	opal_free_sg_list(first);
842 	return NULL;
843 }
844 
845 void opal_free_sg_list(struct opal_sg_list *sg)
846 {
847 	while (sg) {
848 		uint64_t next = be64_to_cpu(sg->next);
849 
850 		kfree(sg);
851 
852 		if (next)
853 			sg = __va(next);
854 		else
855 			sg = NULL;
856 	}
857 }
858 
859 int opal_error_code(int rc)
860 {
861 	switch (rc) {
862 	case OPAL_SUCCESS:		return 0;
863 
864 	case OPAL_PARAMETER:		return -EINVAL;
865 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
866 	case OPAL_BUSY_EVENT:		return -EBUSY;
867 	case OPAL_NO_MEM:		return -ENOMEM;
868 	case OPAL_PERMISSION:		return -EPERM;
869 
870 	case OPAL_UNSUPPORTED:		return -EIO;
871 	case OPAL_HARDWARE:		return -EIO;
872 	case OPAL_INTERNAL_ERROR:	return -EIO;
873 	default:
874 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
875 		return -EIO;
876 	}
877 }
878 
879 EXPORT_SYMBOL_GPL(opal_poll_events);
880 EXPORT_SYMBOL_GPL(opal_rtc_read);
881 EXPORT_SYMBOL_GPL(opal_rtc_write);
882 EXPORT_SYMBOL_GPL(opal_tpo_read);
883 EXPORT_SYMBOL_GPL(opal_tpo_write);
884 EXPORT_SYMBOL_GPL(opal_i2c_request);
885 /* Export these symbols for PowerNV LED class driver */
886 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
887 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
888