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 	u64 reinit_flags = 0;
63 
64 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
65 	 *
66 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
67 	 * also restore HIDs and other SPRs to their original value
68 	 * but it might clobber a bunch.
69 	 */
70 #ifdef __BIG_ENDIAN__
71 	reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
72 #else
73 	reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
74 #endif
75 
76 	/*
77 	 * POWER9 always support running hash:
78 	 *  ie. Host hash  supports  hash guests
79 	 *      Host radix supports  hash/radix guests
80 	 */
81 	if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
82 		reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
83 		if (early_radix_enabled())
84 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
85 	}
86 
87 	opal_reinit_cpus(reinit_flags);
88 
89 	/* Restore some bits */
90 	if (cur_cpu_spec->cpu_restore)
91 		cur_cpu_spec->cpu_restore();
92 }
93 
94 int __init early_init_dt_scan_opal(unsigned long node,
95 				   const char *uname, int depth, void *data)
96 {
97 	const void *basep, *entryp, *sizep;
98 	int basesz, entrysz, runtimesz;
99 
100 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
101 		return 0;
102 
103 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
104 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
105 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
106 
107 	if (!basep || !entryp || !sizep)
108 		return 1;
109 
110 	opal.base = of_read_number(basep, basesz/4);
111 	opal.entry = of_read_number(entryp, entrysz/4);
112 	opal.size = of_read_number(sizep, runtimesz/4);
113 
114 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
115 		 opal.base, basep, basesz);
116 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
117 		 opal.entry, entryp, entrysz);
118 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
119 		 opal.size, sizep, runtimesz);
120 
121 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
122 		powerpc_firmware_features |= FW_FEATURE_OPAL;
123 		pr_info("OPAL detected !\n");
124 	} else {
125 		panic("OPAL != V3 detected, no longer supported.\n");
126 	}
127 
128 	return 1;
129 }
130 
131 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
132 				   const char *uname, int depth, void *data)
133 {
134 	int i, psize, size;
135 	const __be32 *prop;
136 
137 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
138 		return 0;
139 
140 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
141 
142 	if (!prop)
143 		return 1;
144 
145 	pr_debug("Found machine check recoverable ranges.\n");
146 
147 	/*
148 	 * Calculate number of available entries.
149 	 *
150 	 * Each recoverable address range entry is (start address, len,
151 	 * recovery address), 2 cells each for start and recovery address,
152 	 * 1 cell for len, totalling 5 cells per entry.
153 	 */
154 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
155 
156 	/* Sanity check */
157 	if (!mc_recoverable_range_len)
158 		return 1;
159 
160 	/* Size required to hold all the entries. */
161 	size = mc_recoverable_range_len *
162 			sizeof(struct mcheck_recoverable_range);
163 
164 	/*
165 	 * Allocate a buffer to hold the MC recoverable ranges. We would be
166 	 * accessing them in real mode, hence it needs to be within
167 	 * RMO region.
168 	 */
169 	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
170 							ppc64_rma_size));
171 	memset(mc_recoverable_range, 0, size);
172 
173 	for (i = 0; i < mc_recoverable_range_len; i++) {
174 		mc_recoverable_range[i].start_addr =
175 					of_read_number(prop + (i * 5) + 0, 2);
176 		mc_recoverable_range[i].end_addr =
177 					mc_recoverable_range[i].start_addr +
178 					of_read_number(prop + (i * 5) + 2, 1);
179 		mc_recoverable_range[i].recover_addr =
180 					of_read_number(prop + (i * 5) + 3, 2);
181 
182 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
183 				mc_recoverable_range[i].start_addr,
184 				mc_recoverable_range[i].end_addr,
185 				mc_recoverable_range[i].recover_addr);
186 	}
187 	return 1;
188 }
189 
190 static int __init opal_register_exception_handlers(void)
191 {
192 #ifdef __BIG_ENDIAN__
193 	u64 glue;
194 
195 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
196 		return -ENODEV;
197 
198 	/* Hookup some exception handlers except machine check. We use the
199 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
200 	 */
201 	glue = 0x7000;
202 
203 	/*
204 	 * Check if we are running on newer firmware that exports
205 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
206 	 * the HMI interrupt and we catch it directly in Linux.
207 	 *
208 	 * For older firmware (i.e currently released POWER8 System Firmware
209 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
210 	 * patch the HMI vector and handle it inside OPAL firmware.
211 	 *
212 	 * For newer firmware (in development/yet to be released) we will
213 	 * start catching/handling HMI directly in Linux.
214 	 */
215 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
216 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
217 		opal_register_exception_handler(
218 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
219 				0, glue);
220 		glue += 128;
221 	}
222 
223 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
224 #endif
225 
226 	return 0;
227 }
228 machine_early_initcall(powernv, opal_register_exception_handlers);
229 
230 /*
231  * Opal message notifier based on message type. Allow subscribers to get
232  * notified for specific messgae type.
233  */
234 int opal_message_notifier_register(enum opal_msg_type msg_type,
235 					struct notifier_block *nb)
236 {
237 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
238 		pr_warning("%s: Invalid arguments, msg_type:%d\n",
239 			   __func__, msg_type);
240 		return -EINVAL;
241 	}
242 
243 	return atomic_notifier_chain_register(
244 				&opal_msg_notifier_head[msg_type], nb);
245 }
246 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
247 
248 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
249 				     struct notifier_block *nb)
250 {
251 	return atomic_notifier_chain_unregister(
252 			&opal_msg_notifier_head[msg_type], nb);
253 }
254 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
255 
256 static void opal_message_do_notify(uint32_t msg_type, void *msg)
257 {
258 	/* notify subscribers */
259 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
260 					msg_type, msg);
261 }
262 
263 static void opal_handle_message(void)
264 {
265 	s64 ret;
266 	/*
267 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
268 	 * value in /proc/device-tree.
269 	 */
270 	static struct opal_msg msg;
271 	u32 type;
272 
273 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
274 	/* No opal message pending. */
275 	if (ret == OPAL_RESOURCE)
276 		return;
277 
278 	/* check for errors. */
279 	if (ret) {
280 		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
281 				__func__, ret);
282 		return;
283 	}
284 
285 	type = be32_to_cpu(msg.msg_type);
286 
287 	/* Sanity check */
288 	if (type >= OPAL_MSG_TYPE_MAX) {
289 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
290 		return;
291 	}
292 	opal_message_do_notify(type, (void *)&msg);
293 }
294 
295 static irqreturn_t opal_message_notify(int irq, void *data)
296 {
297 	opal_handle_message();
298 	return IRQ_HANDLED;
299 }
300 
301 static int __init opal_message_init(void)
302 {
303 	int ret, i, irq;
304 
305 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
306 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
307 
308 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
309 	if (!irq) {
310 		pr_err("%s: Can't register OPAL event irq (%d)\n",
311 		       __func__, irq);
312 		return irq;
313 	}
314 
315 	ret = request_irq(irq, opal_message_notify,
316 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
317 	if (ret) {
318 		pr_err("%s: Can't request OPAL event irq (%d)\n",
319 		       __func__, ret);
320 		return ret;
321 	}
322 
323 	return 0;
324 }
325 
326 int opal_get_chars(uint32_t vtermno, char *buf, int count)
327 {
328 	s64 rc;
329 	__be64 evt, len;
330 
331 	if (!opal.entry)
332 		return -ENODEV;
333 	opal_poll_events(&evt);
334 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
335 		return 0;
336 	len = cpu_to_be64(count);
337 	rc = opal_console_read(vtermno, &len, buf);
338 	if (rc == OPAL_SUCCESS)
339 		return be64_to_cpu(len);
340 	return 0;
341 }
342 
343 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
344 {
345 	int written = 0;
346 	__be64 olen;
347 	s64 len, rc;
348 	unsigned long flags;
349 	__be64 evt;
350 
351 	if (!opal.entry)
352 		return -ENODEV;
353 
354 	/* We want put_chars to be atomic to avoid mangling of hvsi
355 	 * packets. To do that, we first test for room and return
356 	 * -EAGAIN if there isn't enough.
357 	 *
358 	 * Unfortunately, opal_console_write_buffer_space() doesn't
359 	 * appear to work on opal v1, so we just assume there is
360 	 * enough room and be done with it
361 	 */
362 	spin_lock_irqsave(&opal_write_lock, flags);
363 	rc = opal_console_write_buffer_space(vtermno, &olen);
364 	len = be64_to_cpu(olen);
365 	if (rc || len < total_len) {
366 		spin_unlock_irqrestore(&opal_write_lock, flags);
367 		/* Closed -> drop characters */
368 		if (rc)
369 			return total_len;
370 		opal_poll_events(NULL);
371 		return -EAGAIN;
372 	}
373 
374 	/* We still try to handle partial completions, though they
375 	 * should no longer happen.
376 	 */
377 	rc = OPAL_BUSY;
378 	while(total_len > 0 && (rc == OPAL_BUSY ||
379 				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
380 		olen = cpu_to_be64(total_len);
381 		rc = opal_console_write(vtermno, &olen, data);
382 		len = be64_to_cpu(olen);
383 
384 		/* Closed or other error drop */
385 		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
386 		    rc != OPAL_BUSY_EVENT) {
387 			written = total_len;
388 			break;
389 		}
390 		if (rc == OPAL_SUCCESS) {
391 			total_len -= len;
392 			data += len;
393 			written += len;
394 		}
395 		/* This is a bit nasty but we need that for the console to
396 		 * flush when there aren't any interrupts. We will clean
397 		 * things a bit later to limit that to synchronous path
398 		 * such as the kernel console and xmon/udbg
399 		 */
400 		do
401 			opal_poll_events(&evt);
402 		while(rc == OPAL_SUCCESS &&
403 			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
404 	}
405 	spin_unlock_irqrestore(&opal_write_lock, flags);
406 	return written;
407 }
408 
409 static int opal_recover_mce(struct pt_regs *regs,
410 					struct machine_check_event *evt)
411 {
412 	int recovered = 0;
413 
414 	if (!(regs->msr & MSR_RI)) {
415 		/* If MSR_RI isn't set, we cannot recover */
416 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
417 		recovered = 0;
418 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
419 		/* Platform corrected itself */
420 		recovered = 1;
421 	} else if (evt->severity == MCE_SEV_FATAL) {
422 		/* Fatal machine check */
423 		pr_err("Machine check interrupt is fatal\n");
424 		recovered = 0;
425 	} else if ((evt->severity == MCE_SEV_ERROR_SYNC) &&
426 			(user_mode(regs) && !is_global_init(current))) {
427 		/*
428 		 * For now, kill the task if we have received exception when
429 		 * in userspace.
430 		 *
431 		 * TODO: Queue up this address for hwpoisioning later.
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 	int ret;
443 
444 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
445 		return 0;
446 
447 	/* Print things out */
448 	if (evt.version != MCE_V1) {
449 		pr_err("Machine Check Exception, Unknown event version %d !\n",
450 		       evt.version);
451 		return 0;
452 	}
453 	machine_check_print_event_info(&evt, user_mode(regs));
454 
455 	if (opal_recover_mce(regs, &evt))
456 		return 1;
457 
458 	/*
459 	 * Unrecovered machine check, we are heading to panic path.
460 	 *
461 	 * We may have hit this MCE in very early stage of kernel
462 	 * initialization even before opal-prd has started running. If
463 	 * this is the case then this MCE error may go un-noticed or
464 	 * un-analyzed if we go down panic path. We need to inform
465 	 * BMC/OCC about this error so that they can collect relevant
466 	 * data for error analysis before rebooting.
467 	 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
468 	 * This function may not return on BMC based system.
469 	 */
470 	ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
471 			"Unrecoverable Machine Check exception");
472 	if (ret == OPAL_UNSUPPORTED) {
473 		pr_emerg("Reboot type %d not supported\n",
474 					OPAL_REBOOT_PLATFORM_ERROR);
475 	}
476 
477 	/*
478 	 * We reached here. There can be three possibilities:
479 	 * 1. We are running on a firmware level that do not support
480 	 *    opal_cec_reboot2()
481 	 * 2. We are running on a firmware level that do not support
482 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
483 	 * 3. We are running on FSP based system that does not need opal
484 	 *    to trigger checkstop explicitly for error analysis. The FSP
485 	 *    PRD component would have already got notified about this
486 	 *    error through other channels.
487 	 *
488 	 * If hardware marked this as an unrecoverable MCE, we are
489 	 * going to panic anyway. Even if it didn't, it's not safe to
490 	 * continue at this point, so we should explicitly panic.
491 	 */
492 
493 	panic("PowerNV Unrecovered Machine Check");
494 	return 0;
495 }
496 
497 /* Early hmi handler called in real mode. */
498 int opal_hmi_exception_early(struct pt_regs *regs)
499 {
500 	s64 rc;
501 
502 	/*
503 	 * call opal hmi handler. Pass paca address as token.
504 	 * The return value OPAL_SUCCESS is an indication that there is
505 	 * an HMI event generated waiting to pull by Linux.
506 	 */
507 	rc = opal_handle_hmi();
508 	if (rc == OPAL_SUCCESS) {
509 		local_paca->hmi_event_available = 1;
510 		return 1;
511 	}
512 	return 0;
513 }
514 
515 /* HMI exception handler called in virtual mode during check_irq_replay. */
516 int opal_handle_hmi_exception(struct pt_regs *regs)
517 {
518 	s64 rc;
519 	__be64 evt = 0;
520 
521 	/*
522 	 * Check if HMI event is available.
523 	 * if Yes, then call opal_poll_events to pull opal messages and
524 	 * process them.
525 	 */
526 	if (!local_paca->hmi_event_available)
527 		return 0;
528 
529 	local_paca->hmi_event_available = 0;
530 	rc = opal_poll_events(&evt);
531 	if (rc == OPAL_SUCCESS && evt)
532 		opal_handle_events(be64_to_cpu(evt));
533 
534 	return 1;
535 }
536 
537 static uint64_t find_recovery_address(uint64_t nip)
538 {
539 	int i;
540 
541 	for (i = 0; i < mc_recoverable_range_len; i++)
542 		if ((nip >= mc_recoverable_range[i].start_addr) &&
543 		    (nip < mc_recoverable_range[i].end_addr))
544 		    return mc_recoverable_range[i].recover_addr;
545 	return 0;
546 }
547 
548 bool opal_mce_check_early_recovery(struct pt_regs *regs)
549 {
550 	uint64_t recover_addr = 0;
551 
552 	if (!opal.base || !opal.size)
553 		goto out;
554 
555 	if ((regs->nip >= opal.base) &&
556 			(regs->nip < (opal.base + opal.size)))
557 		recover_addr = find_recovery_address(regs->nip);
558 
559 	/*
560 	 * Setup regs->nip to rfi into fixup address.
561 	 */
562 	if (recover_addr)
563 		regs->nip = recover_addr;
564 
565 out:
566 	return !!recover_addr;
567 }
568 
569 static int opal_sysfs_init(void)
570 {
571 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
572 	if (!opal_kobj) {
573 		pr_warn("kobject_create_and_add opal failed\n");
574 		return -ENOMEM;
575 	}
576 
577 	return 0;
578 }
579 
580 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
581 			       struct bin_attribute *bin_attr,
582 			       char *buf, loff_t off, size_t count)
583 {
584 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
585 				       bin_attr->size);
586 }
587 
588 static BIN_ATTR_RO(symbol_map, 0);
589 
590 static void opal_export_symmap(void)
591 {
592 	const __be64 *syms;
593 	unsigned int size;
594 	struct device_node *fw;
595 	int rc;
596 
597 	fw = of_find_node_by_path("/ibm,opal/firmware");
598 	if (!fw)
599 		return;
600 	syms = of_get_property(fw, "symbol-map", &size);
601 	if (!syms || size != 2 * sizeof(__be64))
602 		return;
603 
604 	/* Setup attributes */
605 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
606 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
607 
608 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
609 	if (rc)
610 		pr_warn("Error %d creating OPAL symbols file\n", rc);
611 }
612 
613 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
614 				struct bin_attribute *bin_attr, char *buf,
615 				loff_t off, size_t count)
616 {
617 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
618 				       bin_attr->size);
619 }
620 
621 /*
622  * opal_export_attrs: creates a sysfs node for each property listed in
623  * the device-tree under /ibm,opal/firmware/exports/
624  * All new sysfs nodes are created under /opal/exports/.
625  * This allows for reserved memory regions (e.g. HDAT) to be read.
626  * The new sysfs nodes are only readable by root.
627  */
628 static void opal_export_attrs(void)
629 {
630 	struct bin_attribute *attr;
631 	struct device_node *np;
632 	struct property *prop;
633 	struct kobject *kobj;
634 	u64 vals[2];
635 	int rc;
636 
637 	np = of_find_node_by_path("/ibm,opal/firmware/exports");
638 	if (!np)
639 		return;
640 
641 	/* Create new 'exports' directory - /sys/firmware/opal/exports */
642 	kobj = kobject_create_and_add("exports", opal_kobj);
643 	if (!kobj) {
644 		pr_warn("kobject_create_and_add() of exports failed\n");
645 		return;
646 	}
647 
648 	for_each_property_of_node(np, prop) {
649 		if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
650 			continue;
651 
652 		if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
653 			continue;
654 
655 		attr = kzalloc(sizeof(*attr), GFP_KERNEL);
656 
657 		if (attr == NULL) {
658 			pr_warn("Failed kmalloc for bin_attribute!");
659 			continue;
660 		}
661 
662 		sysfs_bin_attr_init(attr);
663 		attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
664 		attr->attr.mode = 0400;
665 		attr->read = export_attr_read;
666 		attr->private = __va(vals[0]);
667 		attr->size = vals[1];
668 
669 		if (attr->attr.name == NULL) {
670 			pr_warn("Failed kstrdup for bin_attribute attr.name");
671 			kfree(attr);
672 			continue;
673 		}
674 
675 		rc = sysfs_create_bin_file(kobj, attr);
676 		if (rc) {
677 			pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
678 				 rc, prop->name);
679 			kfree(attr->attr.name);
680 			kfree(attr);
681 		}
682 	}
683 
684 	of_node_put(np);
685 }
686 
687 static void __init opal_dump_region_init(void)
688 {
689 	void *addr;
690 	uint64_t size;
691 	int rc;
692 
693 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
694 		return;
695 
696 	/* Register kernel log buffer */
697 	addr = log_buf_addr_get();
698 	if (addr == NULL)
699 		return;
700 
701 	size = log_buf_len_get();
702 	if (size == 0)
703 		return;
704 
705 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
706 				       __pa(addr), size);
707 	/* Don't warn if this is just an older OPAL that doesn't
708 	 * know about that call
709 	 */
710 	if (rc && rc != OPAL_UNSUPPORTED)
711 		pr_warn("DUMP: Failed to register kernel log buffer. "
712 			"rc = %d\n", rc);
713 }
714 
715 static void opal_pdev_init(const char *compatible)
716 {
717 	struct device_node *np;
718 
719 	for_each_compatible_node(np, NULL, compatible)
720 		of_platform_device_create(np, NULL, NULL);
721 }
722 
723 static int kopald(void *unused)
724 {
725 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
726 	__be64 events;
727 
728 	set_freezable();
729 	do {
730 		try_to_freeze();
731 		opal_poll_events(&events);
732 		opal_handle_events(be64_to_cpu(events));
733 		schedule_timeout_interruptible(timeout);
734 	} while (!kthread_should_stop());
735 
736 	return 0;
737 }
738 
739 void opal_wake_poller(void)
740 {
741 	if (kopald_tsk)
742 		wake_up_process(kopald_tsk);
743 }
744 
745 static void opal_init_heartbeat(void)
746 {
747 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
748 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
749 				 &opal_heartbeat) != 0)
750 		opal_heartbeat = 0;
751 
752 	if (opal_heartbeat)
753 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
754 }
755 
756 static int __init opal_init(void)
757 {
758 	struct device_node *np, *consoles, *leds;
759 	int rc;
760 
761 	opal_node = of_find_node_by_path("/ibm,opal");
762 	if (!opal_node) {
763 		pr_warn("Device node not found\n");
764 		return -ENODEV;
765 	}
766 
767 	/* Register OPAL consoles if any ports */
768 	consoles = of_find_node_by_path("/ibm,opal/consoles");
769 	if (consoles) {
770 		for_each_child_of_node(consoles, np) {
771 			if (strcmp(np->name, "serial"))
772 				continue;
773 			of_platform_device_create(np, NULL, NULL);
774 		}
775 		of_node_put(consoles);
776 	}
777 
778 	/* Initialise OPAL messaging system */
779 	opal_message_init();
780 
781 	/* Initialise OPAL asynchronous completion interface */
782 	opal_async_comp_init();
783 
784 	/* Initialise OPAL sensor interface */
785 	opal_sensor_init();
786 
787 	/* Initialise OPAL hypervisor maintainence interrupt handling */
788 	opal_hmi_handler_init();
789 
790 	/* Create i2c platform devices */
791 	opal_pdev_init("ibm,opal-i2c");
792 
793 	/* Setup a heatbeat thread if requested by OPAL */
794 	opal_init_heartbeat();
795 
796 	/* Create leds platform devices */
797 	leds = of_find_node_by_path("/ibm,opal/leds");
798 	if (leds) {
799 		of_platform_device_create(leds, "opal_leds", NULL);
800 		of_node_put(leds);
801 	}
802 
803 	/* Initialise OPAL message log interface */
804 	opal_msglog_init();
805 
806 	/* Create "opal" kobject under /sys/firmware */
807 	rc = opal_sysfs_init();
808 	if (rc == 0) {
809 		/* Export symbol map to userspace */
810 		opal_export_symmap();
811 		/* Setup dump region interface */
812 		opal_dump_region_init();
813 		/* Setup error log interface */
814 		rc = opal_elog_init();
815 		/* Setup code update interface */
816 		opal_flash_update_init();
817 		/* Setup platform dump extract interface */
818 		opal_platform_dump_init();
819 		/* Setup system parameters interface */
820 		opal_sys_param_init();
821 		/* Setup message log sysfs interface. */
822 		opal_msglog_sysfs_init();
823 	}
824 
825 	/* Export all properties */
826 	opal_export_attrs();
827 
828 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
829 	opal_pdev_init("ibm,opal-ipmi");
830 	opal_pdev_init("ibm,opal-flash");
831 	opal_pdev_init("ibm,opal-prd");
832 
833 	/* Initialise platform device: oppanel interface */
834 	opal_pdev_init("ibm,opal-oppanel");
835 
836 	/* Initialise OPAL kmsg dumper for flushing console on panic */
837 	opal_kmsg_init();
838 
839 	return 0;
840 }
841 machine_subsys_initcall(powernv, opal_init);
842 
843 void opal_shutdown(void)
844 {
845 	long rc = OPAL_BUSY;
846 
847 	opal_event_shutdown();
848 
849 	/*
850 	 * Then sync with OPAL which ensure anything that can
851 	 * potentially write to our memory has completed such
852 	 * as an ongoing dump retrieval
853 	 */
854 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
855 		rc = opal_sync_host_reboot();
856 		if (rc == OPAL_BUSY)
857 			opal_poll_events(NULL);
858 		else
859 			mdelay(10);
860 	}
861 
862 	/* Unregister memory dump region */
863 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
864 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
865 }
866 
867 /* Export this so that test modules can use it */
868 EXPORT_SYMBOL_GPL(opal_invalid_call);
869 EXPORT_SYMBOL_GPL(opal_xscom_read);
870 EXPORT_SYMBOL_GPL(opal_xscom_write);
871 EXPORT_SYMBOL_GPL(opal_ipmi_send);
872 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
873 EXPORT_SYMBOL_GPL(opal_flash_read);
874 EXPORT_SYMBOL_GPL(opal_flash_write);
875 EXPORT_SYMBOL_GPL(opal_flash_erase);
876 EXPORT_SYMBOL_GPL(opal_prd_msg);
877 
878 /* Convert a region of vmalloc memory to an opal sg list */
879 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
880 					     unsigned long vmalloc_size)
881 {
882 	struct opal_sg_list *sg, *first = NULL;
883 	unsigned long i = 0;
884 
885 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
886 	if (!sg)
887 		goto nomem;
888 
889 	first = sg;
890 
891 	while (vmalloc_size > 0) {
892 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
893 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
894 
895 		sg->entry[i].data = cpu_to_be64(data);
896 		sg->entry[i].length = cpu_to_be64(length);
897 		i++;
898 
899 		if (i >= SG_ENTRIES_PER_NODE) {
900 			struct opal_sg_list *next;
901 
902 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
903 			if (!next)
904 				goto nomem;
905 
906 			sg->length = cpu_to_be64(
907 					i * sizeof(struct opal_sg_entry) + 16);
908 			i = 0;
909 			sg->next = cpu_to_be64(__pa(next));
910 			sg = next;
911 		}
912 
913 		vmalloc_addr += length;
914 		vmalloc_size -= length;
915 	}
916 
917 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
918 
919 	return first;
920 
921 nomem:
922 	pr_err("%s : Failed to allocate memory\n", __func__);
923 	opal_free_sg_list(first);
924 	return NULL;
925 }
926 
927 void opal_free_sg_list(struct opal_sg_list *sg)
928 {
929 	while (sg) {
930 		uint64_t next = be64_to_cpu(sg->next);
931 
932 		kfree(sg);
933 
934 		if (next)
935 			sg = __va(next);
936 		else
937 			sg = NULL;
938 	}
939 }
940 
941 int opal_error_code(int rc)
942 {
943 	switch (rc) {
944 	case OPAL_SUCCESS:		return 0;
945 
946 	case OPAL_PARAMETER:		return -EINVAL;
947 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
948 	case OPAL_BUSY_EVENT:		return -EBUSY;
949 	case OPAL_NO_MEM:		return -ENOMEM;
950 	case OPAL_PERMISSION:		return -EPERM;
951 
952 	case OPAL_UNSUPPORTED:		return -EIO;
953 	case OPAL_HARDWARE:		return -EIO;
954 	case OPAL_INTERNAL_ERROR:	return -EIO;
955 	default:
956 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
957 		return -EIO;
958 	}
959 }
960 
961 void powernv_set_nmmu_ptcr(unsigned long ptcr)
962 {
963 	int rc;
964 
965 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
966 		rc = opal_nmmu_set_ptcr(-1UL, ptcr);
967 		if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
968 			pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
969 	}
970 }
971 
972 EXPORT_SYMBOL_GPL(opal_poll_events);
973 EXPORT_SYMBOL_GPL(opal_rtc_read);
974 EXPORT_SYMBOL_GPL(opal_rtc_write);
975 EXPORT_SYMBOL_GPL(opal_tpo_read);
976 EXPORT_SYMBOL_GPL(opal_tpo_write);
977 EXPORT_SYMBOL_GPL(opal_i2c_request);
978 /* Export these symbols for PowerNV LED class driver */
979 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
980 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
981 /* Export this symbol for PowerNV Operator Panel class driver */
982 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
983 /* Export this for KVM */
984 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
985 EXPORT_SYMBOL_GPL(opal_int_eoi);
986