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 static struct task_struct *kopald_tsk;
59 
60 static void opal_reinit_cores(void)
61 {
62 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
63 	 *
64 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
65 	 * also restore HIDs and other SPRs to their original value
66 	 * but it might clobber a bunch.
67 	 */
68 #ifdef __BIG_ENDIAN__
69 	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
70 #else
71 	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
72 #endif
73 }
74 
75 int __init early_init_dt_scan_opal(unsigned long node,
76 				   const char *uname, int depth, void *data)
77 {
78 	const void *basep, *entryp, *sizep;
79 	int basesz, entrysz, runtimesz;
80 
81 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
82 		return 0;
83 
84 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
85 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
86 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
87 
88 	if (!basep || !entryp || !sizep)
89 		return 1;
90 
91 	opal.base = of_read_number(basep, basesz/4);
92 	opal.entry = of_read_number(entryp, entrysz/4);
93 	opal.size = of_read_number(sizep, runtimesz/4);
94 
95 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
96 		 opal.base, basep, basesz);
97 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
98 		 opal.entry, entryp, entrysz);
99 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
100 		 opal.size, sizep, runtimesz);
101 
102 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
103 		powerpc_firmware_features |= FW_FEATURE_OPAL;
104 		pr_info("OPAL detected !\n");
105 	} else {
106 		panic("OPAL != V3 detected, no longer supported.\n");
107 	}
108 
109 	/* Reinit all cores with the right endian */
110 	opal_reinit_cores();
111 
112 	/* Restore some bits */
113 	if (cur_cpu_spec->cpu_restore)
114 		cur_cpu_spec->cpu_restore();
115 
116 	return 1;
117 }
118 
119 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
120 				   const char *uname, int depth, void *data)
121 {
122 	int i, psize, size;
123 	const __be32 *prop;
124 
125 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
126 		return 0;
127 
128 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
129 
130 	if (!prop)
131 		return 1;
132 
133 	pr_debug("Found machine check recoverable ranges.\n");
134 
135 	/*
136 	 * Calculate number of available entries.
137 	 *
138 	 * Each recoverable address range entry is (start address, len,
139 	 * recovery address), 2 cells each for start and recovery address,
140 	 * 1 cell for len, totalling 5 cells per entry.
141 	 */
142 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
143 
144 	/* Sanity check */
145 	if (!mc_recoverable_range_len)
146 		return 1;
147 
148 	/* Size required to hold all the entries. */
149 	size = mc_recoverable_range_len *
150 			sizeof(struct mcheck_recoverable_range);
151 
152 	/*
153 	 * Allocate a buffer to hold the MC recoverable ranges. We would be
154 	 * accessing them in real mode, hence it needs to be within
155 	 * RMO region.
156 	 */
157 	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
158 							ppc64_rma_size));
159 	memset(mc_recoverable_range, 0, size);
160 
161 	for (i = 0; i < mc_recoverable_range_len; i++) {
162 		mc_recoverable_range[i].start_addr =
163 					of_read_number(prop + (i * 5) + 0, 2);
164 		mc_recoverable_range[i].end_addr =
165 					mc_recoverable_range[i].start_addr +
166 					of_read_number(prop + (i * 5) + 2, 1);
167 		mc_recoverable_range[i].recover_addr =
168 					of_read_number(prop + (i * 5) + 3, 2);
169 
170 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
171 				mc_recoverable_range[i].start_addr,
172 				mc_recoverable_range[i].end_addr,
173 				mc_recoverable_range[i].recover_addr);
174 	}
175 	return 1;
176 }
177 
178 static int __init opal_register_exception_handlers(void)
179 {
180 #ifdef __BIG_ENDIAN__
181 	u64 glue;
182 
183 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
184 		return -ENODEV;
185 
186 	/* Hookup some exception handlers except machine check. We use the
187 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
188 	 */
189 	glue = 0x7000;
190 
191 	/*
192 	 * Check if we are running on newer firmware that exports
193 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
194 	 * the HMI interrupt and we catch it directly in Linux.
195 	 *
196 	 * For older firmware (i.e currently released POWER8 System Firmware
197 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
198 	 * patch the HMI vector and handle it inside OPAL firmware.
199 	 *
200 	 * For newer firmware (in development/yet to be released) we will
201 	 * start catching/handling HMI directly in Linux.
202 	 */
203 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
204 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
205 		opal_register_exception_handler(
206 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
207 				0, glue);
208 		glue += 128;
209 	}
210 
211 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
212 #endif
213 
214 	return 0;
215 }
216 machine_early_initcall(powernv, opal_register_exception_handlers);
217 
218 /*
219  * Opal message notifier based on message type. Allow subscribers to get
220  * notified for specific messgae type.
221  */
222 int opal_message_notifier_register(enum opal_msg_type msg_type,
223 					struct notifier_block *nb)
224 {
225 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
226 		pr_warning("%s: Invalid arguments, msg_type:%d\n",
227 			   __func__, msg_type);
228 		return -EINVAL;
229 	}
230 
231 	return atomic_notifier_chain_register(
232 				&opal_msg_notifier_head[msg_type], nb);
233 }
234 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
235 
236 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
237 				     struct notifier_block *nb)
238 {
239 	return atomic_notifier_chain_unregister(
240 			&opal_msg_notifier_head[msg_type], nb);
241 }
242 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
243 
244 static void opal_message_do_notify(uint32_t msg_type, void *msg)
245 {
246 	/* notify subscribers */
247 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
248 					msg_type, msg);
249 }
250 
251 static void opal_handle_message(void)
252 {
253 	s64 ret;
254 	/*
255 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
256 	 * value in /proc/device-tree.
257 	 */
258 	static struct opal_msg msg;
259 	u32 type;
260 
261 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
262 	/* No opal message pending. */
263 	if (ret == OPAL_RESOURCE)
264 		return;
265 
266 	/* check for errors. */
267 	if (ret) {
268 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
269 				__func__, ret);
270 		return;
271 	}
272 
273 	type = be32_to_cpu(msg.msg_type);
274 
275 	/* Sanity check */
276 	if (type >= OPAL_MSG_TYPE_MAX) {
277 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
278 		return;
279 	}
280 	opal_message_do_notify(type, (void *)&msg);
281 }
282 
283 static irqreturn_t opal_message_notify(int irq, void *data)
284 {
285 	opal_handle_message();
286 	return IRQ_HANDLED;
287 }
288 
289 static int __init opal_message_init(void)
290 {
291 	int ret, i, irq;
292 
293 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
294 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
295 
296 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
297 	if (!irq) {
298 		pr_err("%s: Can't register OPAL event irq (%d)\n",
299 		       __func__, irq);
300 		return irq;
301 	}
302 
303 	ret = request_irq(irq, opal_message_notify,
304 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
305 	if (ret) {
306 		pr_err("%s: Can't request OPAL event irq (%d)\n",
307 		       __func__, ret);
308 		return ret;
309 	}
310 
311 	return 0;
312 }
313 
314 int opal_get_chars(uint32_t vtermno, char *buf, int count)
315 {
316 	s64 rc;
317 	__be64 evt, len;
318 
319 	if (!opal.entry)
320 		return -ENODEV;
321 	opal_poll_events(&evt);
322 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
323 		return 0;
324 	len = cpu_to_be64(count);
325 	rc = opal_console_read(vtermno, &len, buf);
326 	if (rc == OPAL_SUCCESS)
327 		return be64_to_cpu(len);
328 	return 0;
329 }
330 
331 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
332 {
333 	int written = 0;
334 	__be64 olen;
335 	s64 len, rc;
336 	unsigned long flags;
337 	__be64 evt;
338 
339 	if (!opal.entry)
340 		return -ENODEV;
341 
342 	/* We want put_chars to be atomic to avoid mangling of hvsi
343 	 * packets. To do that, we first test for room and return
344 	 * -EAGAIN if there isn't enough.
345 	 *
346 	 * Unfortunately, opal_console_write_buffer_space() doesn't
347 	 * appear to work on opal v1, so we just assume there is
348 	 * enough room and be done with it
349 	 */
350 	spin_lock_irqsave(&opal_write_lock, flags);
351 	rc = opal_console_write_buffer_space(vtermno, &olen);
352 	len = be64_to_cpu(olen);
353 	if (rc || len < total_len) {
354 		spin_unlock_irqrestore(&opal_write_lock, flags);
355 		/* Closed -> drop characters */
356 		if (rc)
357 			return total_len;
358 		opal_poll_events(NULL);
359 		return -EAGAIN;
360 	}
361 
362 	/* We still try to handle partial completions, though they
363 	 * should no longer happen.
364 	 */
365 	rc = OPAL_BUSY;
366 	while(total_len > 0 && (rc == OPAL_BUSY ||
367 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
368 		olen = cpu_to_be64(total_len);
369 		rc = opal_console_write(vtermno, &olen, data);
370 		len = be64_to_cpu(olen);
371 
372 		/* Closed or other error drop */
373 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
374 		    rc != OPAL_BUSY_EVENT) {
375 			written = total_len;
376 			break;
377 		}
378 		if (rc == OPAL_SUCCESS) {
379 			total_len -= len;
380 			data += len;
381 			written += len;
382 		}
383 		/* This is a bit nasty but we need that for the console to
384 		 * flush when there aren't any interrupts. We will clean
385 		 * things a bit later to limit that to synchronous path
386 		 * such as the kernel console and xmon/udbg
387 		 */
388 		do
389 			opal_poll_events(&evt);
390 		while(rc == OPAL_SUCCESS &&
391 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
392 	}
393 	spin_unlock_irqrestore(&opal_write_lock, flags);
394 	return written;
395 }
396 
397 static int opal_recover_mce(struct pt_regs *regs,
398 					struct machine_check_event *evt)
399 {
400 	int recovered = 0;
401 	uint64_t ea = get_mce_fault_addr(evt);
402 
403 	if (!(regs->msr & MSR_RI)) {
404 		/* If MSR_RI isn't set, we cannot recover */
405 		recovered = 0;
406 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
407 		/* Platform corrected itself */
408 		recovered = 1;
409 	} else if (ea && !is_kernel_addr(ea)) {
410 		/*
411 		 * Faulting address is not in kernel text. We should be fine.
412 		 * We need to find which process uses this address.
413 		 * For now, kill the task if we have received exception when
414 		 * in userspace.
415 		 *
416 		 * TODO: Queue up this address for hwpoisioning later.
417 		 */
418 		if (user_mode(regs) && !is_global_init(current)) {
419 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
420 			recovered = 1;
421 		} else
422 			recovered = 0;
423 	} else if (user_mode(regs) && !is_global_init(current) &&
424 		evt->severity == MCE_SEV_ERROR_SYNC) {
425 		/*
426 		 * If we have received a synchronous error when in userspace
427 		 * kill the task.
428 		 */
429 		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
430 		recovered = 1;
431 	}
432 	return recovered;
433 }
434 
435 int opal_machine_check(struct pt_regs *regs)
436 {
437 	struct machine_check_event evt;
438 	int ret;
439 
440 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
441 		return 0;
442 
443 	/* Print things out */
444 	if (evt.version != MCE_V1) {
445 		pr_err("Machine Check Exception, Unknown event version %d !\n",
446 		       evt.version);
447 		return 0;
448 	}
449 	machine_check_print_event_info(&evt);
450 
451 	if (opal_recover_mce(regs, &evt))
452 		return 1;
453 
454 	/*
455 	 * Unrecovered machine check, we are heading to panic path.
456 	 *
457 	 * We may have hit this MCE in very early stage of kernel
458 	 * initialization even before opal-prd has started running. If
459 	 * this is the case then this MCE error may go un-noticed or
460 	 * un-analyzed if we go down panic path. We need to inform
461 	 * BMC/OCC about this error so that they can collect relevant
462 	 * data for error analysis before rebooting.
463 	 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
464 	 * This function may not return on BMC based system.
465 	 */
466 	ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
467 			"Unrecoverable Machine Check exception");
468 	if (ret == OPAL_UNSUPPORTED) {
469 		pr_emerg("Reboot type %d not supported\n",
470 					OPAL_REBOOT_PLATFORM_ERROR);
471 	}
472 
473 	/*
474 	 * We reached here. There can be three possibilities:
475 	 * 1. We are running on a firmware level that do not support
476 	 *    opal_cec_reboot2()
477 	 * 2. We are running on a firmware level that do not support
478 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
479 	 * 3. We are running on FSP based system that does not need opal
480 	 *    to trigger checkstop explicitly for error analysis. The FSP
481 	 *    PRD component would have already got notified about this
482 	 *    error through other channels.
483 	 *
484 	 * If hardware marked this as an unrecoverable MCE, we are
485 	 * going to panic anyway. Even if it didn't, it's not safe to
486 	 * continue at this point, so we should explicitly panic.
487 	 */
488 
489 	panic("PowerNV Unrecovered Machine Check");
490 	return 0;
491 }
492 
493 /* Early hmi handler called in real mode. */
494 int opal_hmi_exception_early(struct pt_regs *regs)
495 {
496 	s64 rc;
497 
498 	/*
499 	 * call opal hmi handler. Pass paca address as token.
500 	 * The return value OPAL_SUCCESS is an indication that there is
501 	 * an HMI event generated waiting to pull by Linux.
502 	 */
503 	rc = opal_handle_hmi();
504 	if (rc == OPAL_SUCCESS) {
505 		local_paca->hmi_event_available = 1;
506 		return 1;
507 	}
508 	return 0;
509 }
510 
511 /* HMI exception handler called in virtual mode during check_irq_replay. */
512 int opal_handle_hmi_exception(struct pt_regs *regs)
513 {
514 	s64 rc;
515 	__be64 evt = 0;
516 
517 	/*
518 	 * Check if HMI event is available.
519 	 * if Yes, then call opal_poll_events to pull opal messages and
520 	 * process them.
521 	 */
522 	if (!local_paca->hmi_event_available)
523 		return 0;
524 
525 	local_paca->hmi_event_available = 0;
526 	rc = opal_poll_events(&evt);
527 	if (rc == OPAL_SUCCESS && evt)
528 		opal_handle_events(be64_to_cpu(evt));
529 
530 	return 1;
531 }
532 
533 static uint64_t find_recovery_address(uint64_t nip)
534 {
535 	int i;
536 
537 	for (i = 0; i < mc_recoverable_range_len; i++)
538 		if ((nip >= mc_recoverable_range[i].start_addr) &&
539 		    (nip < mc_recoverable_range[i].end_addr))
540 		    return mc_recoverable_range[i].recover_addr;
541 	return 0;
542 }
543 
544 bool opal_mce_check_early_recovery(struct pt_regs *regs)
545 {
546 	uint64_t recover_addr = 0;
547 
548 	if (!opal.base || !opal.size)
549 		goto out;
550 
551 	if ((regs->nip >= opal.base) &&
552 			(regs->nip < (opal.base + opal.size)))
553 		recover_addr = find_recovery_address(regs->nip);
554 
555 	/*
556 	 * Setup regs->nip to rfi into fixup address.
557 	 */
558 	if (recover_addr)
559 		regs->nip = recover_addr;
560 
561 out:
562 	return !!recover_addr;
563 }
564 
565 static int opal_sysfs_init(void)
566 {
567 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
568 	if (!opal_kobj) {
569 		pr_warn("kobject_create_and_add opal failed\n");
570 		return -ENOMEM;
571 	}
572 
573 	return 0;
574 }
575 
576 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
577 			       struct bin_attribute *bin_attr,
578 			       char *buf, loff_t off, size_t count)
579 {
580 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
581 				       bin_attr->size);
582 }
583 
584 static BIN_ATTR_RO(symbol_map, 0);
585 
586 static void opal_export_symmap(void)
587 {
588 	const __be64 *syms;
589 	unsigned int size;
590 	struct device_node *fw;
591 	int rc;
592 
593 	fw = of_find_node_by_path("/ibm,opal/firmware");
594 	if (!fw)
595 		return;
596 	syms = of_get_property(fw, "symbol-map", &size);
597 	if (!syms || size != 2 * sizeof(__be64))
598 		return;
599 
600 	/* Setup attributes */
601 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
602 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
603 
604 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
605 	if (rc)
606 		pr_warn("Error %d creating OPAL symbols file\n", rc);
607 }
608 
609 static void __init opal_dump_region_init(void)
610 {
611 	void *addr;
612 	uint64_t size;
613 	int rc;
614 
615 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
616 		return;
617 
618 	/* Register kernel log buffer */
619 	addr = log_buf_addr_get();
620 	if (addr == NULL)
621 		return;
622 
623 	size = log_buf_len_get();
624 	if (size == 0)
625 		return;
626 
627 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
628 				       __pa(addr), size);
629 	/* Don't warn if this is just an older OPAL that doesn't
630 	 * know about that call
631 	 */
632 	if (rc && rc != OPAL_UNSUPPORTED)
633 		pr_warn("DUMP: Failed to register kernel log buffer. "
634 			"rc = %d\n", rc);
635 }
636 
637 static void opal_pdev_init(struct device_node *opal_node,
638 		const char *compatible)
639 {
640 	struct device_node *np;
641 
642 	for_each_child_of_node(opal_node, np)
643 		if (of_device_is_compatible(np, compatible))
644 			of_platform_device_create(np, NULL, NULL);
645 }
646 
647 static void opal_i2c_create_devs(void)
648 {
649 	struct device_node *np;
650 
651 	for_each_compatible_node(np, NULL, "ibm,opal-i2c")
652 		of_platform_device_create(np, NULL, NULL);
653 }
654 
655 static int kopald(void *unused)
656 {
657 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
658 	__be64 events;
659 
660 	set_freezable();
661 	do {
662 		try_to_freeze();
663 		opal_poll_events(&events);
664 		opal_handle_events(be64_to_cpu(events));
665 		schedule_timeout_interruptible(timeout);
666 	} while (!kthread_should_stop());
667 
668 	return 0;
669 }
670 
671 void opal_wake_poller(void)
672 {
673 	if (kopald_tsk)
674 		wake_up_process(kopald_tsk);
675 }
676 
677 static void opal_init_heartbeat(void)
678 {
679 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
680 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
681 				 &opal_heartbeat) != 0)
682 		opal_heartbeat = 0;
683 
684 	if (opal_heartbeat)
685 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
686 }
687 
688 static int __init opal_init(void)
689 {
690 	struct device_node *np, *consoles, *leds;
691 	int rc;
692 
693 	opal_node = of_find_node_by_path("/ibm,opal");
694 	if (!opal_node) {
695 		pr_warn("Device node not found\n");
696 		return -ENODEV;
697 	}
698 
699 	/* Register OPAL consoles if any ports */
700 	consoles = of_find_node_by_path("/ibm,opal/consoles");
701 	if (consoles) {
702 		for_each_child_of_node(consoles, np) {
703 			if (strcmp(np->name, "serial"))
704 				continue;
705 			of_platform_device_create(np, NULL, NULL);
706 		}
707 		of_node_put(consoles);
708 	}
709 
710 	/* Initialise OPAL messaging system */
711 	opal_message_init();
712 
713 	/* Initialise OPAL asynchronous completion interface */
714 	opal_async_comp_init();
715 
716 	/* Initialise OPAL sensor interface */
717 	opal_sensor_init();
718 
719 	/* Initialise OPAL hypervisor maintainence interrupt handling */
720 	opal_hmi_handler_init();
721 
722 	/* Create i2c platform devices */
723 	opal_i2c_create_devs();
724 
725 	/* Setup a heatbeat thread if requested by OPAL */
726 	opal_init_heartbeat();
727 
728 	/* Create leds platform devices */
729 	leds = of_find_node_by_path("/ibm,opal/leds");
730 	if (leds) {
731 		of_platform_device_create(leds, "opal_leds", NULL);
732 		of_node_put(leds);
733 	}
734 
735 	/* Initialise OPAL message log interface */
736 	opal_msglog_init();
737 
738 	/* Create "opal" kobject under /sys/firmware */
739 	rc = opal_sysfs_init();
740 	if (rc == 0) {
741 		/* Export symbol map to userspace */
742 		opal_export_symmap();
743 		/* Setup dump region interface */
744 		opal_dump_region_init();
745 		/* Setup error log interface */
746 		rc = opal_elog_init();
747 		/* Setup code update interface */
748 		opal_flash_update_init();
749 		/* Setup platform dump extract interface */
750 		opal_platform_dump_init();
751 		/* Setup system parameters interface */
752 		opal_sys_param_init();
753 		/* Setup message log sysfs interface. */
754 		opal_msglog_sysfs_init();
755 	}
756 
757 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
758 	opal_pdev_init(opal_node, "ibm,opal-ipmi");
759 	opal_pdev_init(opal_node, "ibm,opal-flash");
760 	opal_pdev_init(opal_node, "ibm,opal-prd");
761 
762 	/* Initialise platform device: oppanel interface */
763 	opal_pdev_init(opal_node, "ibm,opal-oppanel");
764 
765 	/* Initialise OPAL kmsg dumper for flushing console on panic */
766 	opal_kmsg_init();
767 
768 	return 0;
769 }
770 machine_subsys_initcall(powernv, opal_init);
771 
772 void opal_shutdown(void)
773 {
774 	long rc = OPAL_BUSY;
775 
776 	opal_event_shutdown();
777 
778 	/*
779 	 * Then sync with OPAL which ensure anything that can
780 	 * potentially write to our memory has completed such
781 	 * as an ongoing dump retrieval
782 	 */
783 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
784 		rc = opal_sync_host_reboot();
785 		if (rc == OPAL_BUSY)
786 			opal_poll_events(NULL);
787 		else
788 			mdelay(10);
789 	}
790 
791 	/* Unregister memory dump region */
792 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
793 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
794 }
795 
796 /* Export this so that test modules can use it */
797 EXPORT_SYMBOL_GPL(opal_invalid_call);
798 EXPORT_SYMBOL_GPL(opal_xscom_read);
799 EXPORT_SYMBOL_GPL(opal_xscom_write);
800 EXPORT_SYMBOL_GPL(opal_ipmi_send);
801 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
802 EXPORT_SYMBOL_GPL(opal_flash_read);
803 EXPORT_SYMBOL_GPL(opal_flash_write);
804 EXPORT_SYMBOL_GPL(opal_flash_erase);
805 EXPORT_SYMBOL_GPL(opal_prd_msg);
806 
807 /* Convert a region of vmalloc memory to an opal sg list */
808 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
809 					     unsigned long vmalloc_size)
810 {
811 	struct opal_sg_list *sg, *first = NULL;
812 	unsigned long i = 0;
813 
814 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
815 	if (!sg)
816 		goto nomem;
817 
818 	first = sg;
819 
820 	while (vmalloc_size > 0) {
821 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
822 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
823 
824 		sg->entry[i].data = cpu_to_be64(data);
825 		sg->entry[i].length = cpu_to_be64(length);
826 		i++;
827 
828 		if (i >= SG_ENTRIES_PER_NODE) {
829 			struct opal_sg_list *next;
830 
831 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
832 			if (!next)
833 				goto nomem;
834 
835 			sg->length = cpu_to_be64(
836 					i * sizeof(struct opal_sg_entry) + 16);
837 			i = 0;
838 			sg->next = cpu_to_be64(__pa(next));
839 			sg = next;
840 		}
841 
842 		vmalloc_addr += length;
843 		vmalloc_size -= length;
844 	}
845 
846 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
847 
848 	return first;
849 
850 nomem:
851 	pr_err("%s : Failed to allocate memory\n", __func__);
852 	opal_free_sg_list(first);
853 	return NULL;
854 }
855 
856 void opal_free_sg_list(struct opal_sg_list *sg)
857 {
858 	while (sg) {
859 		uint64_t next = be64_to_cpu(sg->next);
860 
861 		kfree(sg);
862 
863 		if (next)
864 			sg = __va(next);
865 		else
866 			sg = NULL;
867 	}
868 }
869 
870 int opal_error_code(int rc)
871 {
872 	switch (rc) {
873 	case OPAL_SUCCESS:		return 0;
874 
875 	case OPAL_PARAMETER:		return -EINVAL;
876 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
877 	case OPAL_BUSY_EVENT:		return -EBUSY;
878 	case OPAL_NO_MEM:		return -ENOMEM;
879 	case OPAL_PERMISSION:		return -EPERM;
880 
881 	case OPAL_UNSUPPORTED:		return -EIO;
882 	case OPAL_HARDWARE:		return -EIO;
883 	case OPAL_INTERNAL_ERROR:	return -EIO;
884 	default:
885 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
886 		return -EIO;
887 	}
888 }
889 
890 EXPORT_SYMBOL_GPL(opal_poll_events);
891 EXPORT_SYMBOL_GPL(opal_rtc_read);
892 EXPORT_SYMBOL_GPL(opal_rtc_write);
893 EXPORT_SYMBOL_GPL(opal_tpo_read);
894 EXPORT_SYMBOL_GPL(opal_tpo_write);
895 EXPORT_SYMBOL_GPL(opal_i2c_request);
896 /* Export these symbols for PowerNV LED class driver */
897 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
898 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
899 /* Export this symbol for PowerNV Operator Panel class driver */
900 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
901