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 void opal_configure_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 	/* Restore some bits */
75 	if (cur_cpu_spec->cpu_restore)
76 		cur_cpu_spec->cpu_restore();
77 }
78 
79 int __init early_init_dt_scan_opal(unsigned long node,
80 				   const char *uname, int depth, void *data)
81 {
82 	const void *basep, *entryp, *sizep;
83 	int basesz, entrysz, runtimesz;
84 
85 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
86 		return 0;
87 
88 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
89 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
90 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
91 
92 	if (!basep || !entryp || !sizep)
93 		return 1;
94 
95 	opal.base = of_read_number(basep, basesz/4);
96 	opal.entry = of_read_number(entryp, entrysz/4);
97 	opal.size = of_read_number(sizep, runtimesz/4);
98 
99 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
100 		 opal.base, basep, basesz);
101 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
102 		 opal.entry, entryp, entrysz);
103 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
104 		 opal.size, sizep, runtimesz);
105 
106 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
107 		powerpc_firmware_features |= FW_FEATURE_OPAL;
108 		pr_info("OPAL detected !\n");
109 	} else {
110 		panic("OPAL != V3 detected, no longer supported.\n");
111 	}
112 
113 	return 1;
114 }
115 
116 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
117 				   const char *uname, int depth, void *data)
118 {
119 	int i, psize, size;
120 	const __be32 *prop;
121 
122 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
123 		return 0;
124 
125 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
126 
127 	if (!prop)
128 		return 1;
129 
130 	pr_debug("Found machine check recoverable ranges.\n");
131 
132 	/*
133 	 * Calculate number of available entries.
134 	 *
135 	 * Each recoverable address range entry is (start address, len,
136 	 * recovery address), 2 cells each for start and recovery address,
137 	 * 1 cell for len, totalling 5 cells per entry.
138 	 */
139 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
140 
141 	/* Sanity check */
142 	if (!mc_recoverable_range_len)
143 		return 1;
144 
145 	/* Size required to hold all the entries. */
146 	size = mc_recoverable_range_len *
147 			sizeof(struct mcheck_recoverable_range);
148 
149 	/*
150 	 * Allocate a buffer to hold the MC recoverable ranges. We would be
151 	 * accessing them in real mode, hence it needs to be within
152 	 * RMO region.
153 	 */
154 	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
155 							ppc64_rma_size));
156 	memset(mc_recoverable_range, 0, size);
157 
158 	for (i = 0; i < mc_recoverable_range_len; i++) {
159 		mc_recoverable_range[i].start_addr =
160 					of_read_number(prop + (i * 5) + 0, 2);
161 		mc_recoverable_range[i].end_addr =
162 					mc_recoverable_range[i].start_addr +
163 					of_read_number(prop + (i * 5) + 2, 1);
164 		mc_recoverable_range[i].recover_addr =
165 					of_read_number(prop + (i * 5) + 3, 2);
166 
167 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
168 				mc_recoverable_range[i].start_addr,
169 				mc_recoverable_range[i].end_addr,
170 				mc_recoverable_range[i].recover_addr);
171 	}
172 	return 1;
173 }
174 
175 static int __init opal_register_exception_handlers(void)
176 {
177 #ifdef __BIG_ENDIAN__
178 	u64 glue;
179 
180 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
181 		return -ENODEV;
182 
183 	/* Hookup some exception handlers except machine check. We use the
184 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
185 	 */
186 	glue = 0x7000;
187 
188 	/*
189 	 * Check if we are running on newer firmware that exports
190 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
191 	 * the HMI interrupt and we catch it directly in Linux.
192 	 *
193 	 * For older firmware (i.e currently released POWER8 System Firmware
194 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
195 	 * patch the HMI vector and handle it inside OPAL firmware.
196 	 *
197 	 * For newer firmware (in development/yet to be released) we will
198 	 * start catching/handling HMI directly in Linux.
199 	 */
200 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
201 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
202 		opal_register_exception_handler(
203 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
204 				0, glue);
205 		glue += 128;
206 	}
207 
208 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
209 #endif
210 
211 	return 0;
212 }
213 machine_early_initcall(powernv, opal_register_exception_handlers);
214 
215 /*
216  * Opal message notifier based on message type. Allow subscribers to get
217  * notified for specific messgae type.
218  */
219 int opal_message_notifier_register(enum opal_msg_type msg_type,
220 					struct notifier_block *nb)
221 {
222 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
223 		pr_warning("%s: Invalid arguments, msg_type:%d\n",
224 			   __func__, msg_type);
225 		return -EINVAL;
226 	}
227 
228 	return atomic_notifier_chain_register(
229 				&opal_msg_notifier_head[msg_type], nb);
230 }
231 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
232 
233 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
234 				     struct notifier_block *nb)
235 {
236 	return atomic_notifier_chain_unregister(
237 			&opal_msg_notifier_head[msg_type], nb);
238 }
239 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
240 
241 static void opal_message_do_notify(uint32_t msg_type, void *msg)
242 {
243 	/* notify subscribers */
244 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
245 					msg_type, msg);
246 }
247 
248 static void opal_handle_message(void)
249 {
250 	s64 ret;
251 	/*
252 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
253 	 * value in /proc/device-tree.
254 	 */
255 	static struct opal_msg msg;
256 	u32 type;
257 
258 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
259 	/* No opal message pending. */
260 	if (ret == OPAL_RESOURCE)
261 		return;
262 
263 	/* check for errors. */
264 	if (ret) {
265 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
266 				__func__, ret);
267 		return;
268 	}
269 
270 	type = be32_to_cpu(msg.msg_type);
271 
272 	/* Sanity check */
273 	if (type >= OPAL_MSG_TYPE_MAX) {
274 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
275 		return;
276 	}
277 	opal_message_do_notify(type, (void *)&msg);
278 }
279 
280 static irqreturn_t opal_message_notify(int irq, void *data)
281 {
282 	opal_handle_message();
283 	return IRQ_HANDLED;
284 }
285 
286 static int __init opal_message_init(void)
287 {
288 	int ret, i, irq;
289 
290 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
291 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
292 
293 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
294 	if (!irq) {
295 		pr_err("%s: Can't register OPAL event irq (%d)\n",
296 		       __func__, irq);
297 		return irq;
298 	}
299 
300 	ret = request_irq(irq, opal_message_notify,
301 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
302 	if (ret) {
303 		pr_err("%s: Can't request OPAL event irq (%d)\n",
304 		       __func__, ret);
305 		return ret;
306 	}
307 
308 	return 0;
309 }
310 
311 int opal_get_chars(uint32_t vtermno, char *buf, int count)
312 {
313 	s64 rc;
314 	__be64 evt, len;
315 
316 	if (!opal.entry)
317 		return -ENODEV;
318 	opal_poll_events(&evt);
319 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
320 		return 0;
321 	len = cpu_to_be64(count);
322 	rc = opal_console_read(vtermno, &len, buf);
323 	if (rc == OPAL_SUCCESS)
324 		return be64_to_cpu(len);
325 	return 0;
326 }
327 
328 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
329 {
330 	int written = 0;
331 	__be64 olen;
332 	s64 len, rc;
333 	unsigned long flags;
334 	__be64 evt;
335 
336 	if (!opal.entry)
337 		return -ENODEV;
338 
339 	/* We want put_chars to be atomic to avoid mangling of hvsi
340 	 * packets. To do that, we first test for room and return
341 	 * -EAGAIN if there isn't enough.
342 	 *
343 	 * Unfortunately, opal_console_write_buffer_space() doesn't
344 	 * appear to work on opal v1, so we just assume there is
345 	 * enough room and be done with it
346 	 */
347 	spin_lock_irqsave(&opal_write_lock, flags);
348 	rc = opal_console_write_buffer_space(vtermno, &olen);
349 	len = be64_to_cpu(olen);
350 	if (rc || len < total_len) {
351 		spin_unlock_irqrestore(&opal_write_lock, flags);
352 		/* Closed -> drop characters */
353 		if (rc)
354 			return total_len;
355 		opal_poll_events(NULL);
356 		return -EAGAIN;
357 	}
358 
359 	/* We still try to handle partial completions, though they
360 	 * should no longer happen.
361 	 */
362 	rc = OPAL_BUSY;
363 	while(total_len > 0 && (rc == OPAL_BUSY ||
364 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
365 		olen = cpu_to_be64(total_len);
366 		rc = opal_console_write(vtermno, &olen, data);
367 		len = be64_to_cpu(olen);
368 
369 		/* Closed or other error drop */
370 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
371 		    rc != OPAL_BUSY_EVENT) {
372 			written = total_len;
373 			break;
374 		}
375 		if (rc == OPAL_SUCCESS) {
376 			total_len -= len;
377 			data += len;
378 			written += len;
379 		}
380 		/* This is a bit nasty but we need that for the console to
381 		 * flush when there aren't any interrupts. We will clean
382 		 * things a bit later to limit that to synchronous path
383 		 * such as the kernel console and xmon/udbg
384 		 */
385 		do
386 			opal_poll_events(&evt);
387 		while(rc == OPAL_SUCCESS &&
388 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
389 	}
390 	spin_unlock_irqrestore(&opal_write_lock, flags);
391 	return written;
392 }
393 
394 static int opal_recover_mce(struct pt_regs *regs,
395 					struct machine_check_event *evt)
396 {
397 	int recovered = 0;
398 
399 	if (!(regs->msr & MSR_RI)) {
400 		/* If MSR_RI isn't set, we cannot recover */
401 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
402 		recovered = 0;
403 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
404 		/* Platform corrected itself */
405 		recovered = 1;
406 	} else if (evt->severity == MCE_SEV_FATAL) {
407 		/* Fatal machine check */
408 		pr_err("Machine check interrupt is fatal\n");
409 		recovered = 0;
410 	} else if ((evt->severity == MCE_SEV_ERROR_SYNC) &&
411 			(user_mode(regs) && !is_global_init(current))) {
412 		/*
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 		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
419 		recovered = 1;
420 	}
421 	return recovered;
422 }
423 
424 int opal_machine_check(struct pt_regs *regs)
425 {
426 	struct machine_check_event evt;
427 	int ret;
428 
429 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
430 		return 0;
431 
432 	/* Print things out */
433 	if (evt.version != MCE_V1) {
434 		pr_err("Machine Check Exception, Unknown event version %d !\n",
435 		       evt.version);
436 		return 0;
437 	}
438 	machine_check_print_event_info(&evt, user_mode(regs));
439 
440 	if (opal_recover_mce(regs, &evt))
441 		return 1;
442 
443 	/*
444 	 * Unrecovered machine check, we are heading to panic path.
445 	 *
446 	 * We may have hit this MCE in very early stage of kernel
447 	 * initialization even before opal-prd has started running. If
448 	 * this is the case then this MCE error may go un-noticed or
449 	 * un-analyzed if we go down panic path. We need to inform
450 	 * BMC/OCC about this error so that they can collect relevant
451 	 * data for error analysis before rebooting.
452 	 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
453 	 * This function may not return on BMC based system.
454 	 */
455 	ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
456 			"Unrecoverable Machine Check exception");
457 	if (ret == OPAL_UNSUPPORTED) {
458 		pr_emerg("Reboot type %d not supported\n",
459 					OPAL_REBOOT_PLATFORM_ERROR);
460 	}
461 
462 	/*
463 	 * We reached here. There can be three possibilities:
464 	 * 1. We are running on a firmware level that do not support
465 	 *    opal_cec_reboot2()
466 	 * 2. We are running on a firmware level that do not support
467 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
468 	 * 3. We are running on FSP based system that does not need opal
469 	 *    to trigger checkstop explicitly for error analysis. The FSP
470 	 *    PRD component would have already got notified about this
471 	 *    error through other channels.
472 	 *
473 	 * If hardware marked this as an unrecoverable MCE, we are
474 	 * going to panic anyway. Even if it didn't, it's not safe to
475 	 * continue at this point, so we should explicitly panic.
476 	 */
477 
478 	panic("PowerNV Unrecovered Machine Check");
479 	return 0;
480 }
481 
482 /* Early hmi handler called in real mode. */
483 int opal_hmi_exception_early(struct pt_regs *regs)
484 {
485 	s64 rc;
486 
487 	/*
488 	 * call opal hmi handler. Pass paca address as token.
489 	 * The return value OPAL_SUCCESS is an indication that there is
490 	 * an HMI event generated waiting to pull by Linux.
491 	 */
492 	rc = opal_handle_hmi();
493 	if (rc == OPAL_SUCCESS) {
494 		local_paca->hmi_event_available = 1;
495 		return 1;
496 	}
497 	return 0;
498 }
499 
500 /* HMI exception handler called in virtual mode during check_irq_replay. */
501 int opal_handle_hmi_exception(struct pt_regs *regs)
502 {
503 	s64 rc;
504 	__be64 evt = 0;
505 
506 	/*
507 	 * Check if HMI event is available.
508 	 * if Yes, then call opal_poll_events to pull opal messages and
509 	 * process them.
510 	 */
511 	if (!local_paca->hmi_event_available)
512 		return 0;
513 
514 	local_paca->hmi_event_available = 0;
515 	rc = opal_poll_events(&evt);
516 	if (rc == OPAL_SUCCESS && evt)
517 		opal_handle_events(be64_to_cpu(evt));
518 
519 	return 1;
520 }
521 
522 static uint64_t find_recovery_address(uint64_t nip)
523 {
524 	int i;
525 
526 	for (i = 0; i < mc_recoverable_range_len; i++)
527 		if ((nip >= mc_recoverable_range[i].start_addr) &&
528 		    (nip < mc_recoverable_range[i].end_addr))
529 		    return mc_recoverable_range[i].recover_addr;
530 	return 0;
531 }
532 
533 bool opal_mce_check_early_recovery(struct pt_regs *regs)
534 {
535 	uint64_t recover_addr = 0;
536 
537 	if (!opal.base || !opal.size)
538 		goto out;
539 
540 	if ((regs->nip >= opal.base) &&
541 			(regs->nip < (opal.base + opal.size)))
542 		recover_addr = find_recovery_address(regs->nip);
543 
544 	/*
545 	 * Setup regs->nip to rfi into fixup address.
546 	 */
547 	if (recover_addr)
548 		regs->nip = recover_addr;
549 
550 out:
551 	return !!recover_addr;
552 }
553 
554 static int opal_sysfs_init(void)
555 {
556 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
557 	if (!opal_kobj) {
558 		pr_warn("kobject_create_and_add opal failed\n");
559 		return -ENOMEM;
560 	}
561 
562 	return 0;
563 }
564 
565 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
566 			       struct bin_attribute *bin_attr,
567 			       char *buf, loff_t off, size_t count)
568 {
569 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
570 				       bin_attr->size);
571 }
572 
573 static BIN_ATTR_RO(symbol_map, 0);
574 
575 static void opal_export_symmap(void)
576 {
577 	const __be64 *syms;
578 	unsigned int size;
579 	struct device_node *fw;
580 	int rc;
581 
582 	fw = of_find_node_by_path("/ibm,opal/firmware");
583 	if (!fw)
584 		return;
585 	syms = of_get_property(fw, "symbol-map", &size);
586 	if (!syms || size != 2 * sizeof(__be64))
587 		return;
588 
589 	/* Setup attributes */
590 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
591 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
592 
593 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
594 	if (rc)
595 		pr_warn("Error %d creating OPAL symbols file\n", rc);
596 }
597 
598 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
599 				struct bin_attribute *bin_attr, char *buf,
600 				loff_t off, size_t count)
601 {
602 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
603 				       bin_attr->size);
604 }
605 
606 /*
607  * opal_export_attrs: creates a sysfs node for each property listed in
608  * the device-tree under /ibm,opal/firmware/exports/
609  * All new sysfs nodes are created under /opal/exports/.
610  * This allows for reserved memory regions (e.g. HDAT) to be read.
611  * The new sysfs nodes are only readable by root.
612  */
613 static void opal_export_attrs(void)
614 {
615 	struct bin_attribute *attr;
616 	struct device_node *np;
617 	struct property *prop;
618 	struct kobject *kobj;
619 	u64 vals[2];
620 	int rc;
621 
622 	np = of_find_node_by_path("/ibm,opal/firmware/exports");
623 	if (!np)
624 		return;
625 
626 	/* Create new 'exports' directory - /sys/firmware/opal/exports */
627 	kobj = kobject_create_and_add("exports", opal_kobj);
628 	if (!kobj) {
629 		pr_warn("kobject_create_and_add() of exports failed\n");
630 		return;
631 	}
632 
633 	for_each_property_of_node(np, prop) {
634 		if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
635 			continue;
636 
637 		if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
638 			continue;
639 
640 		attr = kzalloc(sizeof(*attr), GFP_KERNEL);
641 
642 		if (attr == NULL) {
643 			pr_warn("Failed kmalloc for bin_attribute!");
644 			continue;
645 		}
646 
647 		sysfs_bin_attr_init(attr);
648 		attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
649 		attr->attr.mode = 0400;
650 		attr->read = export_attr_read;
651 		attr->private = __va(vals[0]);
652 		attr->size = vals[1];
653 
654 		if (attr->attr.name == NULL) {
655 			pr_warn("Failed kstrdup for bin_attribute attr.name");
656 			kfree(attr);
657 			continue;
658 		}
659 
660 		rc = sysfs_create_bin_file(kobj, attr);
661 		if (rc) {
662 			pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
663 				 rc, prop->name);
664 			kfree(attr->attr.name);
665 			kfree(attr);
666 		}
667 	}
668 
669 	of_node_put(np);
670 }
671 
672 static void __init opal_dump_region_init(void)
673 {
674 	void *addr;
675 	uint64_t size;
676 	int rc;
677 
678 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
679 		return;
680 
681 	/* Register kernel log buffer */
682 	addr = log_buf_addr_get();
683 	if (addr == NULL)
684 		return;
685 
686 	size = log_buf_len_get();
687 	if (size == 0)
688 		return;
689 
690 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
691 				       __pa(addr), size);
692 	/* Don't warn if this is just an older OPAL that doesn't
693 	 * know about that call
694 	 */
695 	if (rc && rc != OPAL_UNSUPPORTED)
696 		pr_warn("DUMP: Failed to register kernel log buffer. "
697 			"rc = %d\n", rc);
698 }
699 
700 static void opal_pdev_init(const char *compatible)
701 {
702 	struct device_node *np;
703 
704 	for_each_compatible_node(np, NULL, compatible)
705 		of_platform_device_create(np, NULL, NULL);
706 }
707 
708 static int kopald(void *unused)
709 {
710 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
711 	__be64 events;
712 
713 	set_freezable();
714 	do {
715 		try_to_freeze();
716 		opal_poll_events(&events);
717 		opal_handle_events(be64_to_cpu(events));
718 		schedule_timeout_interruptible(timeout);
719 	} while (!kthread_should_stop());
720 
721 	return 0;
722 }
723 
724 void opal_wake_poller(void)
725 {
726 	if (kopald_tsk)
727 		wake_up_process(kopald_tsk);
728 }
729 
730 static void opal_init_heartbeat(void)
731 {
732 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
733 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
734 				 &opal_heartbeat) != 0)
735 		opal_heartbeat = 0;
736 
737 	if (opal_heartbeat)
738 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
739 }
740 
741 static int __init opal_init(void)
742 {
743 	struct device_node *np, *consoles, *leds;
744 	int rc;
745 
746 	opal_node = of_find_node_by_path("/ibm,opal");
747 	if (!opal_node) {
748 		pr_warn("Device node not found\n");
749 		return -ENODEV;
750 	}
751 
752 	/* Register OPAL consoles if any ports */
753 	consoles = of_find_node_by_path("/ibm,opal/consoles");
754 	if (consoles) {
755 		for_each_child_of_node(consoles, np) {
756 			if (strcmp(np->name, "serial"))
757 				continue;
758 			of_platform_device_create(np, NULL, NULL);
759 		}
760 		of_node_put(consoles);
761 	}
762 
763 	/* Initialise OPAL messaging system */
764 	opal_message_init();
765 
766 	/* Initialise OPAL asynchronous completion interface */
767 	opal_async_comp_init();
768 
769 	/* Initialise OPAL sensor interface */
770 	opal_sensor_init();
771 
772 	/* Initialise OPAL hypervisor maintainence interrupt handling */
773 	opal_hmi_handler_init();
774 
775 	/* Create i2c platform devices */
776 	opal_pdev_init("ibm,opal-i2c");
777 
778 	/* Setup a heatbeat thread if requested by OPAL */
779 	opal_init_heartbeat();
780 
781 	/* Create leds platform devices */
782 	leds = of_find_node_by_path("/ibm,opal/leds");
783 	if (leds) {
784 		of_platform_device_create(leds, "opal_leds", NULL);
785 		of_node_put(leds);
786 	}
787 
788 	/* Initialise OPAL message log interface */
789 	opal_msglog_init();
790 
791 	/* Create "opal" kobject under /sys/firmware */
792 	rc = opal_sysfs_init();
793 	if (rc == 0) {
794 		/* Export symbol map to userspace */
795 		opal_export_symmap();
796 		/* Setup dump region interface */
797 		opal_dump_region_init();
798 		/* Setup error log interface */
799 		rc = opal_elog_init();
800 		/* Setup code update interface */
801 		opal_flash_update_init();
802 		/* Setup platform dump extract interface */
803 		opal_platform_dump_init();
804 		/* Setup system parameters interface */
805 		opal_sys_param_init();
806 		/* Setup message log sysfs interface. */
807 		opal_msglog_sysfs_init();
808 	}
809 
810 	/* Export all properties */
811 	opal_export_attrs();
812 
813 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
814 	opal_pdev_init("ibm,opal-ipmi");
815 	opal_pdev_init("ibm,opal-flash");
816 	opal_pdev_init("ibm,opal-prd");
817 
818 	/* Initialise platform device: oppanel interface */
819 	opal_pdev_init("ibm,opal-oppanel");
820 
821 	/* Initialise OPAL kmsg dumper for flushing console on panic */
822 	opal_kmsg_init();
823 
824 	return 0;
825 }
826 machine_subsys_initcall(powernv, opal_init);
827 
828 void opal_shutdown(void)
829 {
830 	long rc = OPAL_BUSY;
831 
832 	opal_event_shutdown();
833 
834 	/*
835 	 * Then sync with OPAL which ensure anything that can
836 	 * potentially write to our memory has completed such
837 	 * as an ongoing dump retrieval
838 	 */
839 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
840 		rc = opal_sync_host_reboot();
841 		if (rc == OPAL_BUSY)
842 			opal_poll_events(NULL);
843 		else
844 			mdelay(10);
845 	}
846 
847 	/* Unregister memory dump region */
848 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
849 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
850 }
851 
852 /* Export this so that test modules can use it */
853 EXPORT_SYMBOL_GPL(opal_invalid_call);
854 EXPORT_SYMBOL_GPL(opal_xscom_read);
855 EXPORT_SYMBOL_GPL(opal_xscom_write);
856 EXPORT_SYMBOL_GPL(opal_ipmi_send);
857 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
858 EXPORT_SYMBOL_GPL(opal_flash_read);
859 EXPORT_SYMBOL_GPL(opal_flash_write);
860 EXPORT_SYMBOL_GPL(opal_flash_erase);
861 EXPORT_SYMBOL_GPL(opal_prd_msg);
862 
863 /* Convert a region of vmalloc memory to an opal sg list */
864 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
865 					     unsigned long vmalloc_size)
866 {
867 	struct opal_sg_list *sg, *first = NULL;
868 	unsigned long i = 0;
869 
870 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
871 	if (!sg)
872 		goto nomem;
873 
874 	first = sg;
875 
876 	while (vmalloc_size > 0) {
877 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
878 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
879 
880 		sg->entry[i].data = cpu_to_be64(data);
881 		sg->entry[i].length = cpu_to_be64(length);
882 		i++;
883 
884 		if (i >= SG_ENTRIES_PER_NODE) {
885 			struct opal_sg_list *next;
886 
887 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
888 			if (!next)
889 				goto nomem;
890 
891 			sg->length = cpu_to_be64(
892 					i * sizeof(struct opal_sg_entry) + 16);
893 			i = 0;
894 			sg->next = cpu_to_be64(__pa(next));
895 			sg = next;
896 		}
897 
898 		vmalloc_addr += length;
899 		vmalloc_size -= length;
900 	}
901 
902 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
903 
904 	return first;
905 
906 nomem:
907 	pr_err("%s : Failed to allocate memory\n", __func__);
908 	opal_free_sg_list(first);
909 	return NULL;
910 }
911 
912 void opal_free_sg_list(struct opal_sg_list *sg)
913 {
914 	while (sg) {
915 		uint64_t next = be64_to_cpu(sg->next);
916 
917 		kfree(sg);
918 
919 		if (next)
920 			sg = __va(next);
921 		else
922 			sg = NULL;
923 	}
924 }
925 
926 int opal_error_code(int rc)
927 {
928 	switch (rc) {
929 	case OPAL_SUCCESS:		return 0;
930 
931 	case OPAL_PARAMETER:		return -EINVAL;
932 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
933 	case OPAL_BUSY_EVENT:		return -EBUSY;
934 	case OPAL_NO_MEM:		return -ENOMEM;
935 	case OPAL_PERMISSION:		return -EPERM;
936 
937 	case OPAL_UNSUPPORTED:		return -EIO;
938 	case OPAL_HARDWARE:		return -EIO;
939 	case OPAL_INTERNAL_ERROR:	return -EIO;
940 	default:
941 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
942 		return -EIO;
943 	}
944 }
945 
946 void powernv_set_nmmu_ptcr(unsigned long ptcr)
947 {
948 	int rc;
949 
950 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
951 		rc = opal_nmmu_set_ptcr(-1UL, ptcr);
952 		if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
953 			pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
954 	}
955 }
956 
957 EXPORT_SYMBOL_GPL(opal_poll_events);
958 EXPORT_SYMBOL_GPL(opal_rtc_read);
959 EXPORT_SYMBOL_GPL(opal_rtc_write);
960 EXPORT_SYMBOL_GPL(opal_tpo_read);
961 EXPORT_SYMBOL_GPL(opal_tpo_write);
962 EXPORT_SYMBOL_GPL(opal_i2c_request);
963 /* Export these symbols for PowerNV LED class driver */
964 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
965 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
966 /* Export this symbol for PowerNV Operator Panel class driver */
967 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
968 /* Export this for KVM */
969 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
970 EXPORT_SYMBOL_GPL(opal_int_eoi);
971