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 	powerpc_firmware_features |= FW_FEATURE_OPAL;
102 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
103 		powerpc_firmware_features |= FW_FEATURE_OPALv2;
104 		powerpc_firmware_features |= FW_FEATURE_OPALv3;
105 		pr_info("OPAL V3 detected !\n");
106 	} else if (of_flat_dt_is_compatible(node, "ibm,opal-v2")) {
107 		powerpc_firmware_features |= FW_FEATURE_OPALv2;
108 		pr_info("OPAL V2 detected !\n");
109 	} else {
110 		pr_info("OPAL V1 detected !\n");
111 	}
112 
113 	/* Reinit all cores with the right endian */
114 	opal_reinit_cores();
115 
116 	/* Restore some bits */
117 	if (cur_cpu_spec->cpu_restore)
118 		cur_cpu_spec->cpu_restore();
119 
120 	return 1;
121 }
122 
123 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
124 				   const char *uname, int depth, void *data)
125 {
126 	int i, psize, size;
127 	const __be32 *prop;
128 
129 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
130 		return 0;
131 
132 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
133 
134 	if (!prop)
135 		return 1;
136 
137 	pr_debug("Found machine check recoverable ranges.\n");
138 
139 	/*
140 	 * Calculate number of available entries.
141 	 *
142 	 * Each recoverable address range entry is (start address, len,
143 	 * recovery address), 2 cells each for start and recovery address,
144 	 * 1 cell for len, totalling 5 cells per entry.
145 	 */
146 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
147 
148 	/* Sanity check */
149 	if (!mc_recoverable_range_len)
150 		return 1;
151 
152 	/* Size required to hold all the entries. */
153 	size = mc_recoverable_range_len *
154 			sizeof(struct mcheck_recoverable_range);
155 
156 	/*
157 	 * Allocate a buffer to hold the MC recoverable ranges. We would be
158 	 * accessing them in real mode, hence it needs to be within
159 	 * RMO region.
160 	 */
161 	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
162 							ppc64_rma_size));
163 	memset(mc_recoverable_range, 0, size);
164 
165 	for (i = 0; i < mc_recoverable_range_len; i++) {
166 		mc_recoverable_range[i].start_addr =
167 					of_read_number(prop + (i * 5) + 0, 2);
168 		mc_recoverable_range[i].end_addr =
169 					mc_recoverable_range[i].start_addr +
170 					of_read_number(prop + (i * 5) + 2, 1);
171 		mc_recoverable_range[i].recover_addr =
172 					of_read_number(prop + (i * 5) + 3, 2);
173 
174 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
175 				mc_recoverable_range[i].start_addr,
176 				mc_recoverable_range[i].end_addr,
177 				mc_recoverable_range[i].recover_addr);
178 	}
179 	return 1;
180 }
181 
182 static int __init opal_register_exception_handlers(void)
183 {
184 #ifdef __BIG_ENDIAN__
185 	u64 glue;
186 
187 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
188 		return -ENODEV;
189 
190 	/* Hookup some exception handlers except machine check. We use the
191 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
192 	 */
193 	glue = 0x7000;
194 
195 	/*
196 	 * Check if we are running on newer firmware that exports
197 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
198 	 * the HMI interrupt and we catch it directly in Linux.
199 	 *
200 	 * For older firmware (i.e currently released POWER8 System Firmware
201 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
202 	 * patch the HMI vector and handle it inside OPAL firmware.
203 	 *
204 	 * For newer firmware (in development/yet to be released) we will
205 	 * start catching/handling HMI directly in Linux.
206 	 */
207 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
208 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
209 		opal_register_exception_handler(
210 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
211 				0, glue);
212 		glue += 128;
213 	}
214 
215 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
216 #endif
217 
218 	return 0;
219 }
220 machine_early_initcall(powernv, opal_register_exception_handlers);
221 
222 /*
223  * Opal message notifier based on message type. Allow subscribers to get
224  * notified for specific messgae type.
225  */
226 int opal_message_notifier_register(enum opal_msg_type msg_type,
227 					struct notifier_block *nb)
228 {
229 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
230 		pr_warning("%s: Invalid arguments, msg_type:%d\n",
231 			   __func__, msg_type);
232 		return -EINVAL;
233 	}
234 
235 	return atomic_notifier_chain_register(
236 				&opal_msg_notifier_head[msg_type], nb);
237 }
238 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
239 
240 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
241 				     struct notifier_block *nb)
242 {
243 	return atomic_notifier_chain_unregister(
244 			&opal_msg_notifier_head[msg_type], nb);
245 }
246 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
247 
248 static void opal_message_do_notify(uint32_t msg_type, void *msg)
249 {
250 	/* notify subscribers */
251 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
252 					msg_type, msg);
253 }
254 
255 static void opal_handle_message(void)
256 {
257 	s64 ret;
258 	/*
259 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
260 	 * value in /proc/device-tree.
261 	 */
262 	static struct opal_msg msg;
263 	u32 type;
264 
265 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
266 	/* No opal message pending. */
267 	if (ret == OPAL_RESOURCE)
268 		return;
269 
270 	/* check for errors. */
271 	if (ret) {
272 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
273 				__func__, ret);
274 		return;
275 	}
276 
277 	type = be32_to_cpu(msg.msg_type);
278 
279 	/* Sanity check */
280 	if (type >= OPAL_MSG_TYPE_MAX) {
281 		pr_warning("%s: Unknown message type: %u\n", __func__, type);
282 		return;
283 	}
284 	opal_message_do_notify(type, (void *)&msg);
285 }
286 
287 static irqreturn_t opal_message_notify(int irq, void *data)
288 {
289 	opal_handle_message();
290 	return IRQ_HANDLED;
291 }
292 
293 static int __init opal_message_init(void)
294 {
295 	int ret, i, irq;
296 
297 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
298 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
299 
300 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
301 	if (!irq) {
302 		pr_err("%s: Can't register OPAL event irq (%d)\n",
303 		       __func__, irq);
304 		return irq;
305 	}
306 
307 	ret = request_irq(irq, opal_message_notify,
308 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
309 	if (ret) {
310 		pr_err("%s: Can't request OPAL event irq (%d)\n",
311 		       __func__, ret);
312 		return ret;
313 	}
314 
315 	return 0;
316 }
317 
318 int opal_get_chars(uint32_t vtermno, char *buf, int count)
319 {
320 	s64 rc;
321 	__be64 evt, len;
322 
323 	if (!opal.entry)
324 		return -ENODEV;
325 	opal_poll_events(&evt);
326 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
327 		return 0;
328 	len = cpu_to_be64(count);
329 	rc = opal_console_read(vtermno, &len, buf);
330 	if (rc == OPAL_SUCCESS)
331 		return be64_to_cpu(len);
332 	return 0;
333 }
334 
335 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
336 {
337 	int written = 0;
338 	__be64 olen;
339 	s64 len, rc;
340 	unsigned long flags;
341 	__be64 evt;
342 
343 	if (!opal.entry)
344 		return -ENODEV;
345 
346 	/* We want put_chars to be atomic to avoid mangling of hvsi
347 	 * packets. To do that, we first test for room and return
348 	 * -EAGAIN if there isn't enough.
349 	 *
350 	 * Unfortunately, opal_console_write_buffer_space() doesn't
351 	 * appear to work on opal v1, so we just assume there is
352 	 * enough room and be done with it
353 	 */
354 	spin_lock_irqsave(&opal_write_lock, flags);
355 	if (firmware_has_feature(FW_FEATURE_OPALv2)) {
356 		rc = opal_console_write_buffer_space(vtermno, &olen);
357 		len = be64_to_cpu(olen);
358 		if (rc || len < total_len) {
359 			spin_unlock_irqrestore(&opal_write_lock, flags);
360 			/* Closed -> drop characters */
361 			if (rc)
362 				return total_len;
363 			opal_poll_events(NULL);
364 			return -EAGAIN;
365 		}
366 	}
367 
368 	/* We still try to handle partial completions, though they
369 	 * should no longer happen.
370 	 */
371 	rc = OPAL_BUSY;
372 	while(total_len > 0 && (rc == OPAL_BUSY ||
373 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
374 		olen = cpu_to_be64(total_len);
375 		rc = opal_console_write(vtermno, &olen, data);
376 		len = be64_to_cpu(olen);
377 
378 		/* Closed or other error drop */
379 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
380 		    rc != OPAL_BUSY_EVENT) {
381 			written = total_len;
382 			break;
383 		}
384 		if (rc == OPAL_SUCCESS) {
385 			total_len -= len;
386 			data += len;
387 			written += len;
388 		}
389 		/* This is a bit nasty but we need that for the console to
390 		 * flush when there aren't any interrupts. We will clean
391 		 * things a bit later to limit that to synchronous path
392 		 * such as the kernel console and xmon/udbg
393 		 */
394 		do
395 			opal_poll_events(&evt);
396 		while(rc == OPAL_SUCCESS &&
397 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
398 	}
399 	spin_unlock_irqrestore(&opal_write_lock, flags);
400 	return written;
401 }
402 
403 static int opal_recover_mce(struct pt_regs *regs,
404 					struct machine_check_event *evt)
405 {
406 	int recovered = 0;
407 	uint64_t ea = get_mce_fault_addr(evt);
408 
409 	if (!(regs->msr & MSR_RI)) {
410 		/* If MSR_RI isn't set, we cannot recover */
411 		recovered = 0;
412 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
413 		/* Platform corrected itself */
414 		recovered = 1;
415 	} else if (ea && !is_kernel_addr(ea)) {
416 		/*
417 		 * Faulting address is not in kernel text. We should be fine.
418 		 * We need to find which process uses this address.
419 		 * For now, kill the task if we have received exception when
420 		 * in userspace.
421 		 *
422 		 * TODO: Queue up this address for hwpoisioning later.
423 		 */
424 		if (user_mode(regs) && !is_global_init(current)) {
425 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
426 			recovered = 1;
427 		} else
428 			recovered = 0;
429 	} else if (user_mode(regs) && !is_global_init(current) &&
430 		evt->severity == MCE_SEV_ERROR_SYNC) {
431 		/*
432 		 * If we have received a synchronous error when in userspace
433 		 * kill the task.
434 		 */
435 		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
436 		recovered = 1;
437 	}
438 	return recovered;
439 }
440 
441 int opal_machine_check(struct pt_regs *regs)
442 {
443 	struct machine_check_event evt;
444 	int ret;
445 
446 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
447 		return 0;
448 
449 	/* Print things out */
450 	if (evt.version != MCE_V1) {
451 		pr_err("Machine Check Exception, Unknown event version %d !\n",
452 		       evt.version);
453 		return 0;
454 	}
455 	machine_check_print_event_info(&evt);
456 
457 	if (opal_recover_mce(regs, &evt))
458 		return 1;
459 
460 	/*
461 	 * Unrecovered machine check, we are heading to panic path.
462 	 *
463 	 * We may have hit this MCE in very early stage of kernel
464 	 * initialization even before opal-prd has started running. If
465 	 * this is the case then this MCE error may go un-noticed or
466 	 * un-analyzed if we go down panic path. We need to inform
467 	 * BMC/OCC about this error so that they can collect relevant
468 	 * data for error analysis before rebooting.
469 	 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
470 	 * This function may not return on BMC based system.
471 	 */
472 	ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
473 			"Unrecoverable Machine Check exception");
474 	if (ret == OPAL_UNSUPPORTED) {
475 		pr_emerg("Reboot type %d not supported\n",
476 					OPAL_REBOOT_PLATFORM_ERROR);
477 	}
478 
479 	/*
480 	 * We reached here. There can be three possibilities:
481 	 * 1. We are running on a firmware level that do not support
482 	 *    opal_cec_reboot2()
483 	 * 2. We are running on a firmware level that do not support
484 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
485 	 * 3. We are running on FSP based system that does not need opal
486 	 *    to trigger checkstop explicitly for error analysis. The FSP
487 	 *    PRD component would have already got notified about this
488 	 *    error through other channels.
489 	 *
490 	 * If hardware marked this as an unrecoverable MCE, we are
491 	 * going to panic anyway. Even if it didn't, it's not safe to
492 	 * continue at this point, so we should explicitly panic.
493 	 */
494 
495 	panic("PowerNV Unrecovered Machine Check");
496 	return 0;
497 }
498 
499 /* Early hmi handler called in real mode. */
500 int opal_hmi_exception_early(struct pt_regs *regs)
501 {
502 	s64 rc;
503 
504 	/*
505 	 * call opal hmi handler. Pass paca address as token.
506 	 * The return value OPAL_SUCCESS is an indication that there is
507 	 * an HMI event generated waiting to pull by Linux.
508 	 */
509 	rc = opal_handle_hmi();
510 	if (rc == OPAL_SUCCESS) {
511 		local_paca->hmi_event_available = 1;
512 		return 1;
513 	}
514 	return 0;
515 }
516 
517 /* HMI exception handler called in virtual mode during check_irq_replay. */
518 int opal_handle_hmi_exception(struct pt_regs *regs)
519 {
520 	s64 rc;
521 	__be64 evt = 0;
522 
523 	/*
524 	 * Check if HMI event is available.
525 	 * if Yes, then call opal_poll_events to pull opal messages and
526 	 * process them.
527 	 */
528 	if (!local_paca->hmi_event_available)
529 		return 0;
530 
531 	local_paca->hmi_event_available = 0;
532 	rc = opal_poll_events(&evt);
533 	if (rc == OPAL_SUCCESS && evt)
534 		opal_handle_events(be64_to_cpu(evt));
535 
536 	return 1;
537 }
538 
539 static uint64_t find_recovery_address(uint64_t nip)
540 {
541 	int i;
542 
543 	for (i = 0; i < mc_recoverable_range_len; i++)
544 		if ((nip >= mc_recoverable_range[i].start_addr) &&
545 		    (nip < mc_recoverable_range[i].end_addr))
546 		    return mc_recoverable_range[i].recover_addr;
547 	return 0;
548 }
549 
550 bool opal_mce_check_early_recovery(struct pt_regs *regs)
551 {
552 	uint64_t recover_addr = 0;
553 
554 	if (!opal.base || !opal.size)
555 		goto out;
556 
557 	if ((regs->nip >= opal.base) &&
558 			(regs->nip <= (opal.base + opal.size)))
559 		recover_addr = find_recovery_address(regs->nip);
560 
561 	/*
562 	 * Setup regs->nip to rfi into fixup address.
563 	 */
564 	if (recover_addr)
565 		regs->nip = recover_addr;
566 
567 out:
568 	return !!recover_addr;
569 }
570 
571 static int opal_sysfs_init(void)
572 {
573 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
574 	if (!opal_kobj) {
575 		pr_warn("kobject_create_and_add opal failed\n");
576 		return -ENOMEM;
577 	}
578 
579 	return 0;
580 }
581 
582 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
583 			       struct bin_attribute *bin_attr,
584 			       char *buf, loff_t off, size_t count)
585 {
586 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
587 				       bin_attr->size);
588 }
589 
590 static BIN_ATTR_RO(symbol_map, 0);
591 
592 static void opal_export_symmap(void)
593 {
594 	const __be64 *syms;
595 	unsigned int size;
596 	struct device_node *fw;
597 	int rc;
598 
599 	fw = of_find_node_by_path("/ibm,opal/firmware");
600 	if (!fw)
601 		return;
602 	syms = of_get_property(fw, "symbol-map", &size);
603 	if (!syms || size != 2 * sizeof(__be64))
604 		return;
605 
606 	/* Setup attributes */
607 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
608 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
609 
610 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
611 	if (rc)
612 		pr_warn("Error %d creating OPAL symbols file\n", rc);
613 }
614 
615 static void __init opal_dump_region_init(void)
616 {
617 	void *addr;
618 	uint64_t size;
619 	int rc;
620 
621 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
622 		return;
623 
624 	/* Register kernel log buffer */
625 	addr = log_buf_addr_get();
626 	if (addr == NULL)
627 		return;
628 
629 	size = log_buf_len_get();
630 	if (size == 0)
631 		return;
632 
633 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
634 				       __pa(addr), size);
635 	/* Don't warn if this is just an older OPAL that doesn't
636 	 * know about that call
637 	 */
638 	if (rc && rc != OPAL_UNSUPPORTED)
639 		pr_warn("DUMP: Failed to register kernel log buffer. "
640 			"rc = %d\n", rc);
641 }
642 
643 static void opal_pdev_init(struct device_node *opal_node,
644 		const char *compatible)
645 {
646 	struct device_node *np;
647 
648 	for_each_child_of_node(opal_node, np)
649 		if (of_device_is_compatible(np, compatible))
650 			of_platform_device_create(np, NULL, NULL);
651 }
652 
653 static void opal_i2c_create_devs(void)
654 {
655 	struct device_node *np;
656 
657 	for_each_compatible_node(np, NULL, "ibm,opal-i2c")
658 		of_platform_device_create(np, NULL, NULL);
659 }
660 
661 static int kopald(void *unused)
662 {
663 	__be64 events;
664 
665 	set_freezable();
666 	do {
667 		try_to_freeze();
668 		opal_poll_events(&events);
669 		opal_handle_events(be64_to_cpu(events));
670 		msleep_interruptible(opal_heartbeat);
671 	} while (!kthread_should_stop());
672 
673 	return 0;
674 }
675 
676 static void opal_init_heartbeat(void)
677 {
678 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
679 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
680 				 &opal_heartbeat) != 0)
681 		opal_heartbeat = 0;
682 
683 	if (opal_heartbeat)
684 		kthread_run(kopald, NULL, "kopald");
685 }
686 
687 static int __init opal_init(void)
688 {
689 	struct device_node *np, *consoles, *leds;
690 	int rc;
691 
692 	opal_node = of_find_node_by_path("/ibm,opal");
693 	if (!opal_node) {
694 		pr_warn("Device node not found\n");
695 		return -ENODEV;
696 	}
697 
698 	/* Register OPAL consoles if any ports */
699 	if (firmware_has_feature(FW_FEATURE_OPALv2))
700 		consoles = of_find_node_by_path("/ibm,opal/consoles");
701 	else
702 		consoles = of_node_get(opal_node);
703 	if (consoles) {
704 		for_each_child_of_node(consoles, np) {
705 			if (strcmp(np->name, "serial"))
706 				continue;
707 			of_platform_device_create(np, NULL, NULL);
708 		}
709 		of_node_put(consoles);
710 	}
711 
712 	/* Initialise OPAL messaging system */
713 	opal_message_init();
714 
715 	/* Initialise OPAL asynchronous completion interface */
716 	opal_async_comp_init();
717 
718 	/* Initialise OPAL sensor interface */
719 	opal_sensor_init();
720 
721 	/* Initialise OPAL hypervisor maintainence interrupt handling */
722 	opal_hmi_handler_init();
723 
724 	/* Create i2c platform devices */
725 	opal_i2c_create_devs();
726 
727 	/* Setup a heatbeat thread if requested by OPAL */
728 	opal_init_heartbeat();
729 
730 	/* Create leds platform devices */
731 	leds = of_find_node_by_path("/ibm,opal/leds");
732 	if (leds) {
733 		of_platform_device_create(leds, "opal_leds", NULL);
734 		of_node_put(leds);
735 	}
736 
737 	/* Create "opal" kobject under /sys/firmware */
738 	rc = opal_sysfs_init();
739 	if (rc == 0) {
740 		/* Export symbol map to userspace */
741 		opal_export_symmap();
742 		/* Setup dump region interface */
743 		opal_dump_region_init();
744 		/* Setup error log interface */
745 		rc = opal_elog_init();
746 		/* Setup code update interface */
747 		opal_flash_update_init();
748 		/* Setup platform dump extract interface */
749 		opal_platform_dump_init();
750 		/* Setup system parameters interface */
751 		opal_sys_param_init();
752 		/* Setup message log interface. */
753 		opal_msglog_init();
754 	}
755 
756 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
757 	opal_pdev_init(opal_node, "ibm,opal-ipmi");
758 	opal_pdev_init(opal_node, "ibm,opal-flash");
759 	opal_pdev_init(opal_node, "ibm,opal-prd");
760 
761 	return 0;
762 }
763 machine_subsys_initcall(powernv, opal_init);
764 
765 void opal_shutdown(void)
766 {
767 	long rc = OPAL_BUSY;
768 
769 	opal_event_shutdown();
770 
771 	/*
772 	 * Then sync with OPAL which ensure anything that can
773 	 * potentially write to our memory has completed such
774 	 * as an ongoing dump retrieval
775 	 */
776 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
777 		rc = opal_sync_host_reboot();
778 		if (rc == OPAL_BUSY)
779 			opal_poll_events(NULL);
780 		else
781 			mdelay(10);
782 	}
783 
784 	/* Unregister memory dump region */
785 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
786 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
787 }
788 
789 /* Export this so that test modules can use it */
790 EXPORT_SYMBOL_GPL(opal_invalid_call);
791 EXPORT_SYMBOL_GPL(opal_xscom_read);
792 EXPORT_SYMBOL_GPL(opal_xscom_write);
793 EXPORT_SYMBOL_GPL(opal_ipmi_send);
794 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
795 EXPORT_SYMBOL_GPL(opal_flash_read);
796 EXPORT_SYMBOL_GPL(opal_flash_write);
797 EXPORT_SYMBOL_GPL(opal_flash_erase);
798 EXPORT_SYMBOL_GPL(opal_prd_msg);
799 
800 /* Convert a region of vmalloc memory to an opal sg list */
801 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
802 					     unsigned long vmalloc_size)
803 {
804 	struct opal_sg_list *sg, *first = NULL;
805 	unsigned long i = 0;
806 
807 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
808 	if (!sg)
809 		goto nomem;
810 
811 	first = sg;
812 
813 	while (vmalloc_size > 0) {
814 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
815 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
816 
817 		sg->entry[i].data = cpu_to_be64(data);
818 		sg->entry[i].length = cpu_to_be64(length);
819 		i++;
820 
821 		if (i >= SG_ENTRIES_PER_NODE) {
822 			struct opal_sg_list *next;
823 
824 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
825 			if (!next)
826 				goto nomem;
827 
828 			sg->length = cpu_to_be64(
829 					i * sizeof(struct opal_sg_entry) + 16);
830 			i = 0;
831 			sg->next = cpu_to_be64(__pa(next));
832 			sg = next;
833 		}
834 
835 		vmalloc_addr += length;
836 		vmalloc_size -= length;
837 	}
838 
839 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
840 
841 	return first;
842 
843 nomem:
844 	pr_err("%s : Failed to allocate memory\n", __func__);
845 	opal_free_sg_list(first);
846 	return NULL;
847 }
848 
849 void opal_free_sg_list(struct opal_sg_list *sg)
850 {
851 	while (sg) {
852 		uint64_t next = be64_to_cpu(sg->next);
853 
854 		kfree(sg);
855 
856 		if (next)
857 			sg = __va(next);
858 		else
859 			sg = NULL;
860 	}
861 }
862 
863 int opal_error_code(int rc)
864 {
865 	switch (rc) {
866 	case OPAL_SUCCESS:		return 0;
867 
868 	case OPAL_PARAMETER:		return -EINVAL;
869 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
870 	case OPAL_BUSY_EVENT:		return -EBUSY;
871 	case OPAL_NO_MEM:		return -ENOMEM;
872 	case OPAL_PERMISSION:		return -EPERM;
873 
874 	case OPAL_UNSUPPORTED:		return -EIO;
875 	case OPAL_HARDWARE:		return -EIO;
876 	case OPAL_INTERNAL_ERROR:	return -EIO;
877 	default:
878 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
879 		return -EIO;
880 	}
881 }
882 
883 EXPORT_SYMBOL_GPL(opal_poll_events);
884 EXPORT_SYMBOL_GPL(opal_rtc_read);
885 EXPORT_SYMBOL_GPL(opal_rtc_write);
886 EXPORT_SYMBOL_GPL(opal_tpo_read);
887 EXPORT_SYMBOL_GPL(opal_tpo_write);
888 EXPORT_SYMBOL_GPL(opal_i2c_request);
889 /* Export these symbols for PowerNV LED class driver */
890 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
891 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
892