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 
239 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
240 				     struct notifier_block *nb)
241 {
242 	return atomic_notifier_chain_unregister(
243 			&opal_msg_notifier_head[msg_type], nb);
244 }
245 
246 static void opal_message_do_notify(uint32_t msg_type, void *msg)
247 {
248 	/* notify subscribers */
249 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
250 					msg_type, msg);
251 }
252 
253 static void opal_handle_message(void)
254 {
255 	s64 ret;
256 	/*
257 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
258 	 * value in /proc/device-tree.
259 	 */
260 	static struct opal_msg msg;
261 	u32 type;
262 
263 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
264 	/* No opal message pending. */
265 	if (ret == OPAL_RESOURCE)
266 		return;
267 
268 	/* check for errors. */
269 	if (ret) {
270 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
271 				__func__, ret);
272 		return;
273 	}
274 
275 	type = be32_to_cpu(msg.msg_type);
276 
277 	/* Sanity check */
278 	if (type >= OPAL_MSG_TYPE_MAX) {
279 		pr_warning("%s: Unknown message type: %u\n", __func__, type);
280 		return;
281 	}
282 	opal_message_do_notify(type, (void *)&msg);
283 }
284 
285 static irqreturn_t opal_message_notify(int irq, void *data)
286 {
287 	opal_handle_message();
288 	return IRQ_HANDLED;
289 }
290 
291 static int __init opal_message_init(void)
292 {
293 	int ret, i, irq;
294 
295 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
296 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
297 
298 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
299 	if (!irq) {
300 		pr_err("%s: Can't register OPAL event irq (%d)\n",
301 		       __func__, irq);
302 		return irq;
303 	}
304 
305 	ret = request_irq(irq, opal_message_notify,
306 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
307 	if (ret) {
308 		pr_err("%s: Can't request OPAL event irq (%d)\n",
309 		       __func__, ret);
310 		return ret;
311 	}
312 
313 	return 0;
314 }
315 
316 int opal_get_chars(uint32_t vtermno, char *buf, int count)
317 {
318 	s64 rc;
319 	__be64 evt, len;
320 
321 	if (!opal.entry)
322 		return -ENODEV;
323 	opal_poll_events(&evt);
324 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
325 		return 0;
326 	len = cpu_to_be64(count);
327 	rc = opal_console_read(vtermno, &len, buf);
328 	if (rc == OPAL_SUCCESS)
329 		return be64_to_cpu(len);
330 	return 0;
331 }
332 
333 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
334 {
335 	int written = 0;
336 	__be64 olen;
337 	s64 len, rc;
338 	unsigned long flags;
339 	__be64 evt;
340 
341 	if (!opal.entry)
342 		return -ENODEV;
343 
344 	/* We want put_chars to be atomic to avoid mangling of hvsi
345 	 * packets. To do that, we first test for room and return
346 	 * -EAGAIN if there isn't enough.
347 	 *
348 	 * Unfortunately, opal_console_write_buffer_space() doesn't
349 	 * appear to work on opal v1, so we just assume there is
350 	 * enough room and be done with it
351 	 */
352 	spin_lock_irqsave(&opal_write_lock, flags);
353 	if (firmware_has_feature(FW_FEATURE_OPALv2)) {
354 		rc = opal_console_write_buffer_space(vtermno, &olen);
355 		len = be64_to_cpu(olen);
356 		if (rc || len < total_len) {
357 			spin_unlock_irqrestore(&opal_write_lock, flags);
358 			/* Closed -> drop characters */
359 			if (rc)
360 				return total_len;
361 			opal_poll_events(NULL);
362 			return -EAGAIN;
363 		}
364 	}
365 
366 	/* We still try to handle partial completions, though they
367 	 * should no longer happen.
368 	 */
369 	rc = OPAL_BUSY;
370 	while(total_len > 0 && (rc == OPAL_BUSY ||
371 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
372 		olen = cpu_to_be64(total_len);
373 		rc = opal_console_write(vtermno, &olen, data);
374 		len = be64_to_cpu(olen);
375 
376 		/* Closed or other error drop */
377 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
378 		    rc != OPAL_BUSY_EVENT) {
379 			written = total_len;
380 			break;
381 		}
382 		if (rc == OPAL_SUCCESS) {
383 			total_len -= len;
384 			data += len;
385 			written += len;
386 		}
387 		/* This is a bit nasty but we need that for the console to
388 		 * flush when there aren't any interrupts. We will clean
389 		 * things a bit later to limit that to synchronous path
390 		 * such as the kernel console and xmon/udbg
391 		 */
392 		do
393 			opal_poll_events(&evt);
394 		while(rc == OPAL_SUCCESS &&
395 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
396 	}
397 	spin_unlock_irqrestore(&opal_write_lock, flags);
398 	return written;
399 }
400 
401 static int opal_recover_mce(struct pt_regs *regs,
402 					struct machine_check_event *evt)
403 {
404 	int recovered = 0;
405 	uint64_t ea = get_mce_fault_addr(evt);
406 
407 	if (!(regs->msr & MSR_RI)) {
408 		/* If MSR_RI isn't set, we cannot recover */
409 		recovered = 0;
410 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
411 		/* Platform corrected itself */
412 		recovered = 1;
413 	} else if (ea && !is_kernel_addr(ea)) {
414 		/*
415 		 * Faulting address is not in kernel text. We should be fine.
416 		 * We need to find which process uses this address.
417 		 * For now, kill the task if we have received exception when
418 		 * in userspace.
419 		 *
420 		 * TODO: Queue up this address for hwpoisioning later.
421 		 */
422 		if (user_mode(regs) && !is_global_init(current)) {
423 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
424 			recovered = 1;
425 		} else
426 			recovered = 0;
427 	} else if (user_mode(regs) && !is_global_init(current) &&
428 		evt->severity == MCE_SEV_ERROR_SYNC) {
429 		/*
430 		 * If we have received a synchronous error when in userspace
431 		 * kill the task.
432 		 */
433 		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
434 		recovered = 1;
435 	}
436 	return recovered;
437 }
438 
439 int opal_machine_check(struct pt_regs *regs)
440 {
441 	struct machine_check_event evt;
442 
443 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
444 		return 0;
445 
446 	/* Print things out */
447 	if (evt.version != MCE_V1) {
448 		pr_err("Machine Check Exception, Unknown event version %d !\n",
449 		       evt.version);
450 		return 0;
451 	}
452 	machine_check_print_event_info(&evt);
453 
454 	if (opal_recover_mce(regs, &evt))
455 		return 1;
456 	return 0;
457 }
458 
459 /* Early hmi handler called in real mode. */
460 int opal_hmi_exception_early(struct pt_regs *regs)
461 {
462 	s64 rc;
463 
464 	/*
465 	 * call opal hmi handler. Pass paca address as token.
466 	 * The return value OPAL_SUCCESS is an indication that there is
467 	 * an HMI event generated waiting to pull by Linux.
468 	 */
469 	rc = opal_handle_hmi();
470 	if (rc == OPAL_SUCCESS) {
471 		local_paca->hmi_event_available = 1;
472 		return 1;
473 	}
474 	return 0;
475 }
476 
477 /* HMI exception handler called in virtual mode during check_irq_replay. */
478 int opal_handle_hmi_exception(struct pt_regs *regs)
479 {
480 	s64 rc;
481 	__be64 evt = 0;
482 
483 	/*
484 	 * Check if HMI event is available.
485 	 * if Yes, then call opal_poll_events to pull opal messages and
486 	 * process them.
487 	 */
488 	if (!local_paca->hmi_event_available)
489 		return 0;
490 
491 	local_paca->hmi_event_available = 0;
492 	rc = opal_poll_events(&evt);
493 	if (rc == OPAL_SUCCESS && evt)
494 		opal_handle_events(be64_to_cpu(evt));
495 
496 	return 1;
497 }
498 
499 static uint64_t find_recovery_address(uint64_t nip)
500 {
501 	int i;
502 
503 	for (i = 0; i < mc_recoverable_range_len; i++)
504 		if ((nip >= mc_recoverable_range[i].start_addr) &&
505 		    (nip < mc_recoverable_range[i].end_addr))
506 		    return mc_recoverable_range[i].recover_addr;
507 	return 0;
508 }
509 
510 bool opal_mce_check_early_recovery(struct pt_regs *regs)
511 {
512 	uint64_t recover_addr = 0;
513 
514 	if (!opal.base || !opal.size)
515 		goto out;
516 
517 	if ((regs->nip >= opal.base) &&
518 			(regs->nip <= (opal.base + opal.size)))
519 		recover_addr = find_recovery_address(regs->nip);
520 
521 	/*
522 	 * Setup regs->nip to rfi into fixup address.
523 	 */
524 	if (recover_addr)
525 		regs->nip = recover_addr;
526 
527 out:
528 	return !!recover_addr;
529 }
530 
531 static int opal_sysfs_init(void)
532 {
533 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
534 	if (!opal_kobj) {
535 		pr_warn("kobject_create_and_add opal failed\n");
536 		return -ENOMEM;
537 	}
538 
539 	return 0;
540 }
541 
542 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
543 			       struct bin_attribute *bin_attr,
544 			       char *buf, loff_t off, size_t count)
545 {
546 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
547 				       bin_attr->size);
548 }
549 
550 static BIN_ATTR_RO(symbol_map, 0);
551 
552 static void opal_export_symmap(void)
553 {
554 	const __be64 *syms;
555 	unsigned int size;
556 	struct device_node *fw;
557 	int rc;
558 
559 	fw = of_find_node_by_path("/ibm,opal/firmware");
560 	if (!fw)
561 		return;
562 	syms = of_get_property(fw, "symbol-map", &size);
563 	if (!syms || size != 2 * sizeof(__be64))
564 		return;
565 
566 	/* Setup attributes */
567 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
568 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
569 
570 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
571 	if (rc)
572 		pr_warn("Error %d creating OPAL symbols file\n", rc);
573 }
574 
575 static void __init opal_dump_region_init(void)
576 {
577 	void *addr;
578 	uint64_t size;
579 	int rc;
580 
581 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
582 		return;
583 
584 	/* Register kernel log buffer */
585 	addr = log_buf_addr_get();
586 	if (addr == NULL)
587 		return;
588 
589 	size = log_buf_len_get();
590 	if (size == 0)
591 		return;
592 
593 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
594 				       __pa(addr), size);
595 	/* Don't warn if this is just an older OPAL that doesn't
596 	 * know about that call
597 	 */
598 	if (rc && rc != OPAL_UNSUPPORTED)
599 		pr_warn("DUMP: Failed to register kernel log buffer. "
600 			"rc = %d\n", rc);
601 }
602 
603 static void opal_flash_init(struct device_node *opal_node)
604 {
605 	struct device_node *np;
606 
607 	for_each_child_of_node(opal_node, np)
608 		if (of_device_is_compatible(np, "ibm,opal-flash"))
609 			of_platform_device_create(np, NULL, NULL);
610 }
611 
612 static void opal_ipmi_init(struct device_node *opal_node)
613 {
614 	struct device_node *np;
615 
616 	for_each_child_of_node(opal_node, np)
617 		if (of_device_is_compatible(np, "ibm,opal-ipmi"))
618 			of_platform_device_create(np, NULL, NULL);
619 }
620 
621 static void opal_i2c_create_devs(void)
622 {
623 	struct device_node *np;
624 
625 	for_each_compatible_node(np, NULL, "ibm,opal-i2c")
626 		of_platform_device_create(np, NULL, NULL);
627 }
628 
629 static int kopald(void *unused)
630 {
631 	__be64 events;
632 
633 	set_freezable();
634 	do {
635 		try_to_freeze();
636 		opal_poll_events(&events);
637 		opal_handle_events(be64_to_cpu(events));
638 		msleep_interruptible(opal_heartbeat);
639 	} while (!kthread_should_stop());
640 
641 	return 0;
642 }
643 
644 static void opal_init_heartbeat(void)
645 {
646 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
647 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
648 				 &opal_heartbeat) != 0)
649 		opal_heartbeat = 0;
650 
651 	if (opal_heartbeat)
652 		kthread_run(kopald, NULL, "kopald");
653 }
654 
655 static int __init opal_init(void)
656 {
657 	struct device_node *np, *consoles;
658 	int rc;
659 
660 	opal_node = of_find_node_by_path("/ibm,opal");
661 	if (!opal_node) {
662 		pr_warn("Device node not found\n");
663 		return -ENODEV;
664 	}
665 
666 	/* Initialise OPAL events */
667 	opal_event_init();
668 
669 	/* Register OPAL consoles if any ports */
670 	if (firmware_has_feature(FW_FEATURE_OPALv2))
671 		consoles = of_find_node_by_path("/ibm,opal/consoles");
672 	else
673 		consoles = of_node_get(opal_node);
674 	if (consoles) {
675 		for_each_child_of_node(consoles, np) {
676 			if (strcmp(np->name, "serial"))
677 				continue;
678 			of_platform_device_create(np, NULL, NULL);
679 		}
680 		of_node_put(consoles);
681 	}
682 
683 	/* Initialise OPAL messaging system */
684 	opal_message_init();
685 
686 	/* Initialise OPAL asynchronous completion interface */
687 	opal_async_comp_init();
688 
689 	/* Initialise OPAL sensor interface */
690 	opal_sensor_init();
691 
692 	/* Initialise OPAL hypervisor maintainence interrupt handling */
693 	opal_hmi_handler_init();
694 
695 	/* Create i2c platform devices */
696 	opal_i2c_create_devs();
697 
698 	/* Setup a heatbeat thread if requested by OPAL */
699 	opal_init_heartbeat();
700 
701 	/* Create "opal" kobject under /sys/firmware */
702 	rc = opal_sysfs_init();
703 	if (rc == 0) {
704 		/* Export symbol map to userspace */
705 		opal_export_symmap();
706 		/* Setup dump region interface */
707 		opal_dump_region_init();
708 		/* Setup error log interface */
709 		rc = opal_elog_init();
710 		/* Setup code update interface */
711 		opal_flash_update_init();
712 		/* Setup platform dump extract interface */
713 		opal_platform_dump_init();
714 		/* Setup system parameters interface */
715 		opal_sys_param_init();
716 		/* Setup message log interface. */
717 		opal_msglog_init();
718 	}
719 
720 	/* Initialize OPAL IPMI backend */
721 	opal_ipmi_init(opal_node);
722 
723 	opal_flash_init(opal_node);
724 
725 	return 0;
726 }
727 machine_subsys_initcall(powernv, opal_init);
728 
729 void opal_shutdown(void)
730 {
731 	long rc = OPAL_BUSY;
732 
733 	opal_event_shutdown();
734 
735 	/*
736 	 * Then sync with OPAL which ensure anything that can
737 	 * potentially write to our memory has completed such
738 	 * as an ongoing dump retrieval
739 	 */
740 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
741 		rc = opal_sync_host_reboot();
742 		if (rc == OPAL_BUSY)
743 			opal_poll_events(NULL);
744 		else
745 			mdelay(10);
746 	}
747 
748 	/* Unregister memory dump region */
749 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
750 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
751 }
752 
753 /* Export this so that test modules can use it */
754 EXPORT_SYMBOL_GPL(opal_invalid_call);
755 EXPORT_SYMBOL_GPL(opal_ipmi_send);
756 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
757 EXPORT_SYMBOL_GPL(opal_flash_read);
758 EXPORT_SYMBOL_GPL(opal_flash_write);
759 EXPORT_SYMBOL_GPL(opal_flash_erase);
760 
761 /* Convert a region of vmalloc memory to an opal sg list */
762 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
763 					     unsigned long vmalloc_size)
764 {
765 	struct opal_sg_list *sg, *first = NULL;
766 	unsigned long i = 0;
767 
768 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
769 	if (!sg)
770 		goto nomem;
771 
772 	first = sg;
773 
774 	while (vmalloc_size > 0) {
775 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
776 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
777 
778 		sg->entry[i].data = cpu_to_be64(data);
779 		sg->entry[i].length = cpu_to_be64(length);
780 		i++;
781 
782 		if (i >= SG_ENTRIES_PER_NODE) {
783 			struct opal_sg_list *next;
784 
785 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
786 			if (!next)
787 				goto nomem;
788 
789 			sg->length = cpu_to_be64(
790 					i * sizeof(struct opal_sg_entry) + 16);
791 			i = 0;
792 			sg->next = cpu_to_be64(__pa(next));
793 			sg = next;
794 		}
795 
796 		vmalloc_addr += length;
797 		vmalloc_size -= length;
798 	}
799 
800 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
801 
802 	return first;
803 
804 nomem:
805 	pr_err("%s : Failed to allocate memory\n", __func__);
806 	opal_free_sg_list(first);
807 	return NULL;
808 }
809 
810 void opal_free_sg_list(struct opal_sg_list *sg)
811 {
812 	while (sg) {
813 		uint64_t next = be64_to_cpu(sg->next);
814 
815 		kfree(sg);
816 
817 		if (next)
818 			sg = __va(next);
819 		else
820 			sg = NULL;
821 	}
822 }
823 
824 int opal_error_code(int rc)
825 {
826 	switch (rc) {
827 	case OPAL_SUCCESS:		return 0;
828 
829 	case OPAL_PARAMETER:		return -EINVAL;
830 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
831 	case OPAL_BUSY_EVENT:		return -EBUSY;
832 	case OPAL_NO_MEM:		return -ENOMEM;
833 	case OPAL_PERMISSION:		return -EPERM;
834 
835 	case OPAL_UNSUPPORTED:		return -EIO;
836 	case OPAL_HARDWARE:		return -EIO;
837 	case OPAL_INTERNAL_ERROR:	return -EIO;
838 	default:
839 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
840 		return -EIO;
841 	}
842 }
843 
844 EXPORT_SYMBOL_GPL(opal_poll_events);
845 EXPORT_SYMBOL_GPL(opal_rtc_read);
846 EXPORT_SYMBOL_GPL(opal_rtc_write);
847 EXPORT_SYMBOL_GPL(opal_tpo_read);
848 EXPORT_SYMBOL_GPL(opal_tpo_write);
849 EXPORT_SYMBOL_GPL(opal_i2c_request);
850