1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV OPAL high level interfaces
4  *
5  * Copyright 2011 IBM Corp.
6  */
7 
8 #define pr_fmt(fmt)	"opal: " fmt
9 
10 #include <linux/printk.h>
11 #include <linux/types.h>
12 #include <linux/of.h>
13 #include <linux/of_fdt.h>
14 #include <linux/of_platform.h>
15 #include <linux/of_address.h>
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/slab.h>
19 #include <linux/sched.h>
20 #include <linux/kobject.h>
21 #include <linux/delay.h>
22 #include <linux/memblock.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/kmsg_dump.h>
26 #include <linux/console.h>
27 #include <linux/sched/debug.h>
28 
29 #include <asm/machdep.h>
30 #include <asm/opal.h>
31 #include <asm/firmware.h>
32 #include <asm/mce.h>
33 #include <asm/imc-pmu.h>
34 #include <asm/bug.h>
35 
36 #include "powernv.h"
37 
38 /* /sys/firmware/opal */
39 struct kobject *opal_kobj;
40 
41 struct opal {
42 	u64 base;
43 	u64 entry;
44 	u64 size;
45 } opal;
46 
47 struct mcheck_recoverable_range {
48 	u64 start_addr;
49 	u64 end_addr;
50 	u64 recover_addr;
51 };
52 
53 static struct mcheck_recoverable_range *mc_recoverable_range;
54 static int mc_recoverable_range_len;
55 
56 struct device_node *opal_node;
57 static DEFINE_SPINLOCK(opal_write_lock);
58 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
59 static uint32_t opal_heartbeat;
60 static struct task_struct *kopald_tsk;
61 static struct opal_msg *opal_msg;
62 static u32 opal_msg_size __ro_after_init;
63 
64 void opal_configure_cores(void)
65 {
66 	u64 reinit_flags = 0;
67 
68 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
69 	 *
70 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
71 	 * also restore HIDs and other SPRs to their original value
72 	 * but it might clobber a bunch.
73 	 */
74 #ifdef __BIG_ENDIAN__
75 	reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
76 #else
77 	reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
78 #endif
79 
80 	/*
81 	 * POWER9 always support running hash:
82 	 *  ie. Host hash  supports  hash guests
83 	 *      Host radix supports  hash/radix guests
84 	 */
85 	if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
86 		reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
87 		if (early_radix_enabled())
88 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
89 	}
90 
91 	opal_reinit_cpus(reinit_flags);
92 
93 	/* Restore some bits */
94 	if (cur_cpu_spec->cpu_restore)
95 		cur_cpu_spec->cpu_restore();
96 }
97 
98 int __init early_init_dt_scan_opal(unsigned long node,
99 				   const char *uname, int depth, void *data)
100 {
101 	const void *basep, *entryp, *sizep;
102 	int basesz, entrysz, runtimesz;
103 
104 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
105 		return 0;
106 
107 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
108 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
109 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
110 
111 	if (!basep || !entryp || !sizep)
112 		return 1;
113 
114 	opal.base = of_read_number(basep, basesz/4);
115 	opal.entry = of_read_number(entryp, entrysz/4);
116 	opal.size = of_read_number(sizep, runtimesz/4);
117 
118 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
119 		 opal.base, basep, basesz);
120 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
121 		 opal.entry, entryp, entrysz);
122 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
123 		 opal.size, sizep, runtimesz);
124 
125 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
126 		powerpc_firmware_features |= FW_FEATURE_OPAL;
127 		pr_debug("OPAL detected !\n");
128 	} else {
129 		panic("OPAL != V3 detected, no longer supported.\n");
130 	}
131 
132 	return 1;
133 }
134 
135 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
136 				   const char *uname, int depth, void *data)
137 {
138 	int i, psize, size;
139 	const __be32 *prop;
140 
141 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
142 		return 0;
143 
144 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
145 
146 	if (!prop)
147 		return 1;
148 
149 	pr_debug("Found machine check recoverable ranges.\n");
150 
151 	/*
152 	 * Calculate number of available entries.
153 	 *
154 	 * Each recoverable address range entry is (start address, len,
155 	 * recovery address), 2 cells each for start and recovery address,
156 	 * 1 cell for len, totalling 5 cells per entry.
157 	 */
158 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
159 
160 	/* Sanity check */
161 	if (!mc_recoverable_range_len)
162 		return 1;
163 
164 	/* Size required to hold all the entries. */
165 	size = mc_recoverable_range_len *
166 			sizeof(struct mcheck_recoverable_range);
167 
168 	/*
169 	 * Allocate a buffer to hold the MC recoverable ranges.
170 	 */
171 	mc_recoverable_range = memblock_alloc(size, __alignof__(u64));
172 	if (!mc_recoverable_range)
173 		panic("%s: Failed to allocate %u bytes align=0x%lx\n",
174 		      __func__, size, __alignof__(u64));
175 
176 	for (i = 0; i < mc_recoverable_range_len; i++) {
177 		mc_recoverable_range[i].start_addr =
178 					of_read_number(prop + (i * 5) + 0, 2);
179 		mc_recoverable_range[i].end_addr =
180 					mc_recoverable_range[i].start_addr +
181 					of_read_number(prop + (i * 5) + 2, 1);
182 		mc_recoverable_range[i].recover_addr =
183 					of_read_number(prop + (i * 5) + 3, 2);
184 
185 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
186 				mc_recoverable_range[i].start_addr,
187 				mc_recoverable_range[i].end_addr,
188 				mc_recoverable_range[i].recover_addr);
189 	}
190 	return 1;
191 }
192 
193 static int __init opal_register_exception_handlers(void)
194 {
195 #ifdef __BIG_ENDIAN__
196 	u64 glue;
197 
198 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
199 		return -ENODEV;
200 
201 	/* Hookup some exception handlers except machine check. We use the
202 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
203 	 */
204 	glue = 0x7000;
205 
206 	/*
207 	 * Only ancient OPAL firmware requires this.
208 	 * Specifically, firmware from FW810.00 (released June 2014)
209 	 * through FW810.20 (Released October 2014).
210 	 *
211 	 * Check if we are running on newer (post Oct 2014) firmware that
212 	 * exports the OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to
213 	 * patch the HMI interrupt and we catch it directly in Linux.
214 	 *
215 	 * For older firmware (i.e < FW810.20), we fallback to old behavior and
216 	 * let OPAL patch the HMI vector and handle it inside OPAL firmware.
217 	 *
218 	 * For newer firmware we catch/handle the HMI directly in Linux.
219 	 */
220 	if (!opal_check_token(OPAL_HANDLE_HMI)) {
221 		pr_info("Old firmware detected, OPAL handles HMIs.\n");
222 		opal_register_exception_handler(
223 				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
224 				0, glue);
225 		glue += 128;
226 	}
227 
228 	/*
229 	 * Only applicable to ancient firmware, all modern
230 	 * (post March 2015/skiboot 5.0) firmware will just return
231 	 * OPAL_UNSUPPORTED.
232 	 */
233 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
234 #endif
235 
236 	return 0;
237 }
238 machine_early_initcall(powernv, opal_register_exception_handlers);
239 
240 /*
241  * Opal message notifier based on message type. Allow subscribers to get
242  * notified for specific messgae type.
243  */
244 int opal_message_notifier_register(enum opal_msg_type msg_type,
245 					struct notifier_block *nb)
246 {
247 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
248 		pr_warn("%s: Invalid arguments, msg_type:%d\n",
249 			__func__, msg_type);
250 		return -EINVAL;
251 	}
252 
253 	return atomic_notifier_chain_register(
254 				&opal_msg_notifier_head[msg_type], nb);
255 }
256 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
257 
258 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
259 				     struct notifier_block *nb)
260 {
261 	return atomic_notifier_chain_unregister(
262 			&opal_msg_notifier_head[msg_type], nb);
263 }
264 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
265 
266 static void opal_message_do_notify(uint32_t msg_type, void *msg)
267 {
268 	/* notify subscribers */
269 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
270 					msg_type, msg);
271 }
272 
273 static void opal_handle_message(void)
274 {
275 	s64 ret;
276 	u32 type;
277 
278 	ret = opal_get_msg(__pa(opal_msg), opal_msg_size);
279 	/* No opal message pending. */
280 	if (ret == OPAL_RESOURCE)
281 		return;
282 
283 	/* check for errors. */
284 	if (ret) {
285 		pr_warn("%s: Failed to retrieve opal message, err=%lld\n",
286 			__func__, ret);
287 		return;
288 	}
289 
290 	type = be32_to_cpu(opal_msg->msg_type);
291 
292 	/* Sanity check */
293 	if (type >= OPAL_MSG_TYPE_MAX) {
294 		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
295 		return;
296 	}
297 	opal_message_do_notify(type, (void *)opal_msg);
298 }
299 
300 static irqreturn_t opal_message_notify(int irq, void *data)
301 {
302 	opal_handle_message();
303 	return IRQ_HANDLED;
304 }
305 
306 static int __init opal_message_init(struct device_node *opal_node)
307 {
308 	int ret, i, irq;
309 
310 	ret = of_property_read_u32(opal_node, "opal-msg-size", &opal_msg_size);
311 	if (ret) {
312 		pr_notice("Failed to read opal-msg-size property\n");
313 		opal_msg_size = sizeof(struct opal_msg);
314 	}
315 
316 	opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
317 	if (!opal_msg) {
318 		opal_msg_size = sizeof(struct opal_msg);
319 		/* Try to allocate fixed message size */
320 		opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
321 		BUG_ON(opal_msg == NULL);
322 	}
323 
324 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
325 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
326 
327 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
328 	if (!irq) {
329 		pr_err("%s: Can't register OPAL event irq (%d)\n",
330 		       __func__, irq);
331 		return irq;
332 	}
333 
334 	ret = request_irq(irq, opal_message_notify,
335 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
336 	if (ret) {
337 		pr_err("%s: Can't request OPAL event irq (%d)\n",
338 		       __func__, ret);
339 		return ret;
340 	}
341 
342 	return 0;
343 }
344 
345 int opal_get_chars(uint32_t vtermno, char *buf, int count)
346 {
347 	s64 rc;
348 	__be64 evt, len;
349 
350 	if (!opal.entry)
351 		return -ENODEV;
352 	opal_poll_events(&evt);
353 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
354 		return 0;
355 	len = cpu_to_be64(count);
356 	rc = opal_console_read(vtermno, &len, buf);
357 	if (rc == OPAL_SUCCESS)
358 		return be64_to_cpu(len);
359 	return 0;
360 }
361 
362 static int __opal_put_chars(uint32_t vtermno, const char *data, int total_len, bool atomic)
363 {
364 	unsigned long flags = 0 /* shut up gcc */;
365 	int written;
366 	__be64 olen;
367 	s64 rc;
368 
369 	if (!opal.entry)
370 		return -ENODEV;
371 
372 	if (atomic)
373 		spin_lock_irqsave(&opal_write_lock, flags);
374 	rc = opal_console_write_buffer_space(vtermno, &olen);
375 	if (rc || be64_to_cpu(olen) < total_len) {
376 		/* Closed -> drop characters */
377 		if (rc)
378 			written = total_len;
379 		else
380 			written = -EAGAIN;
381 		goto out;
382 	}
383 
384 	/* Should not get a partial write here because space is available. */
385 	olen = cpu_to_be64(total_len);
386 	rc = opal_console_write(vtermno, &olen, data);
387 	if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
388 		if (rc == OPAL_BUSY_EVENT)
389 			opal_poll_events(NULL);
390 		written = -EAGAIN;
391 		goto out;
392 	}
393 
394 	/* Closed or other error drop */
395 	if (rc != OPAL_SUCCESS) {
396 		written = opal_error_code(rc);
397 		goto out;
398 	}
399 
400 	written = be64_to_cpu(olen);
401 	if (written < total_len) {
402 		if (atomic) {
403 			/* Should not happen */
404 			pr_warn("atomic console write returned partial "
405 				"len=%d written=%d\n", total_len, written);
406 		}
407 		if (!written)
408 			written = -EAGAIN;
409 	}
410 
411 out:
412 	if (atomic)
413 		spin_unlock_irqrestore(&opal_write_lock, flags);
414 
415 	return written;
416 }
417 
418 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
419 {
420 	return __opal_put_chars(vtermno, data, total_len, false);
421 }
422 
423 /*
424  * opal_put_chars_atomic will not perform partial-writes. Data will be
425  * atomically written to the terminal or not at all. This is not strictly
426  * true at the moment because console space can race with OPAL's console
427  * writes.
428  */
429 int opal_put_chars_atomic(uint32_t vtermno, const char *data, int total_len)
430 {
431 	return __opal_put_chars(vtermno, data, total_len, true);
432 }
433 
434 static s64 __opal_flush_console(uint32_t vtermno)
435 {
436 	s64 rc;
437 
438 	if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
439 		__be64 evt;
440 
441 		/*
442 		 * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
443 		 * the console can still be flushed by calling the polling
444 		 * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
445 		 */
446 		WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
447 
448 		opal_poll_events(&evt);
449 		if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
450 			return OPAL_SUCCESS;
451 		return OPAL_BUSY;
452 
453 	} else {
454 		rc = opal_console_flush(vtermno);
455 		if (rc == OPAL_BUSY_EVENT) {
456 			opal_poll_events(NULL);
457 			rc = OPAL_BUSY;
458 		}
459 		return rc;
460 	}
461 
462 }
463 
464 /*
465  * opal_flush_console spins until the console is flushed
466  */
467 int opal_flush_console(uint32_t vtermno)
468 {
469 	for (;;) {
470 		s64 rc = __opal_flush_console(vtermno);
471 
472 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
473 			mdelay(1);
474 			continue;
475 		}
476 
477 		return opal_error_code(rc);
478 	}
479 }
480 
481 /*
482  * opal_flush_chars is an hvc interface that sleeps until the console is
483  * flushed if wait, otherwise it will return -EBUSY if the console has data,
484  * -EAGAIN if it has data and some of it was flushed.
485  */
486 int opal_flush_chars(uint32_t vtermno, bool wait)
487 {
488 	for (;;) {
489 		s64 rc = __opal_flush_console(vtermno);
490 
491 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
492 			if (wait) {
493 				msleep(OPAL_BUSY_DELAY_MS);
494 				continue;
495 			}
496 			if (rc == OPAL_PARTIAL)
497 				return -EAGAIN;
498 		}
499 
500 		return opal_error_code(rc);
501 	}
502 }
503 
504 static int opal_recover_mce(struct pt_regs *regs,
505 					struct machine_check_event *evt)
506 {
507 	int recovered = 0;
508 
509 	if (!(regs->msr & MSR_RI)) {
510 		/* If MSR_RI isn't set, we cannot recover */
511 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
512 		recovered = 0;
513 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
514 		/* Platform corrected itself */
515 		recovered = 1;
516 	} else if (evt->severity == MCE_SEV_FATAL) {
517 		/* Fatal machine check */
518 		pr_err("Machine check interrupt is fatal\n");
519 		recovered = 0;
520 	}
521 
522 	if (!recovered && evt->sync_error) {
523 		/*
524 		 * Try to kill processes if we get a synchronous machine check
525 		 * (e.g., one caused by execution of this instruction). This
526 		 * will devolve into a panic if we try to kill init or are in
527 		 * an interrupt etc.
528 		 *
529 		 * TODO: Queue up this address for hwpoisioning later.
530 		 * TODO: This is not quite right for d-side machine
531 		 *       checks ->nip is not necessarily the important
532 		 *       address.
533 		 */
534 		if ((user_mode(regs))) {
535 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
536 			recovered = 1;
537 		} else if (die_will_crash()) {
538 			/*
539 			 * die() would kill the kernel, so better to go via
540 			 * the platform reboot code that will log the
541 			 * machine check.
542 			 */
543 			recovered = 0;
544 		} else {
545 			die("Machine check", regs, SIGBUS);
546 			recovered = 1;
547 		}
548 	}
549 
550 	return recovered;
551 }
552 
553 void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
554 {
555 	panic_flush_kmsg_start();
556 
557 	pr_emerg("Hardware platform error: %s\n", msg);
558 	if (regs)
559 		show_regs(regs);
560 	smp_send_stop();
561 
562 	panic_flush_kmsg_end();
563 
564 	/*
565 	 * Don't bother to shut things down because this will
566 	 * xstop the system.
567 	 */
568 	if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
569 						== OPAL_UNSUPPORTED) {
570 		pr_emerg("Reboot type %d not supported for %s\n",
571 				OPAL_REBOOT_PLATFORM_ERROR, msg);
572 	}
573 
574 	/*
575 	 * We reached here. There can be three possibilities:
576 	 * 1. We are running on a firmware level that do not support
577 	 *    opal_cec_reboot2()
578 	 * 2. We are running on a firmware level that do not support
579 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
580 	 * 3. We are running on FSP based system that does not need
581 	 *    opal to trigger checkstop explicitly for error analysis.
582 	 *    The FSP PRD component would have already got notified
583 	 *    about this error through other channels.
584 	 * 4. We are running on a newer skiboot that by default does
585 	 *    not cause a checkstop, drops us back to the kernel to
586 	 *    extract context and state at the time of the error.
587 	 */
588 
589 	panic(msg);
590 }
591 
592 int opal_machine_check(struct pt_regs *regs)
593 {
594 	struct machine_check_event evt;
595 
596 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
597 		return 0;
598 
599 	/* Print things out */
600 	if (evt.version != MCE_V1) {
601 		pr_err("Machine Check Exception, Unknown event version %d !\n",
602 		       evt.version);
603 		return 0;
604 	}
605 	machine_check_print_event_info(&evt, user_mode(regs), false);
606 
607 	if (opal_recover_mce(regs, &evt))
608 		return 1;
609 
610 	pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
611 }
612 
613 /* Early hmi handler called in real mode. */
614 int opal_hmi_exception_early(struct pt_regs *regs)
615 {
616 	s64 rc;
617 
618 	/*
619 	 * call opal hmi handler. Pass paca address as token.
620 	 * The return value OPAL_SUCCESS is an indication that there is
621 	 * an HMI event generated waiting to pull by Linux.
622 	 */
623 	rc = opal_handle_hmi();
624 	if (rc == OPAL_SUCCESS) {
625 		local_paca->hmi_event_available = 1;
626 		return 1;
627 	}
628 	return 0;
629 }
630 
631 int opal_hmi_exception_early2(struct pt_regs *regs)
632 {
633 	s64 rc;
634 	__be64 out_flags;
635 
636 	/*
637 	 * call opal hmi handler.
638 	 * Check 64-bit flag mask to find out if an event was generated,
639 	 * and whether TB is still valid or not etc.
640 	 */
641 	rc = opal_handle_hmi2(&out_flags);
642 	if (rc != OPAL_SUCCESS)
643 		return 0;
644 
645 	if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_NEW_EVENT)
646 		local_paca->hmi_event_available = 1;
647 	if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_TOD_TB_FAIL)
648 		tb_invalid = true;
649 	return 1;
650 }
651 
652 /* HMI exception handler called in virtual mode during check_irq_replay. */
653 int opal_handle_hmi_exception(struct pt_regs *regs)
654 {
655 	/*
656 	 * Check if HMI event is available.
657 	 * if Yes, then wake kopald to process them.
658 	 */
659 	if (!local_paca->hmi_event_available)
660 		return 0;
661 
662 	local_paca->hmi_event_available = 0;
663 	opal_wake_poller();
664 
665 	return 1;
666 }
667 
668 static uint64_t find_recovery_address(uint64_t nip)
669 {
670 	int i;
671 
672 	for (i = 0; i < mc_recoverable_range_len; i++)
673 		if ((nip >= mc_recoverable_range[i].start_addr) &&
674 		    (nip < mc_recoverable_range[i].end_addr))
675 		    return mc_recoverable_range[i].recover_addr;
676 	return 0;
677 }
678 
679 bool opal_mce_check_early_recovery(struct pt_regs *regs)
680 {
681 	uint64_t recover_addr = 0;
682 
683 	if (!opal.base || !opal.size)
684 		goto out;
685 
686 	if ((regs->nip >= opal.base) &&
687 			(regs->nip < (opal.base + opal.size)))
688 		recover_addr = find_recovery_address(regs->nip);
689 
690 	/*
691 	 * Setup regs->nip to rfi into fixup address.
692 	 */
693 	if (recover_addr)
694 		regs->nip = recover_addr;
695 
696 out:
697 	return !!recover_addr;
698 }
699 
700 static int opal_sysfs_init(void)
701 {
702 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
703 	if (!opal_kobj) {
704 		pr_warn("kobject_create_and_add opal failed\n");
705 		return -ENOMEM;
706 	}
707 
708 	return 0;
709 }
710 
711 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
712 			       struct bin_attribute *bin_attr,
713 			       char *buf, loff_t off, size_t count)
714 {
715 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
716 				       bin_attr->size);
717 }
718 
719 static struct bin_attribute symbol_map_attr = {
720 	.attr = {.name = "symbol_map", .mode = 0400},
721 	.read = symbol_map_read
722 };
723 
724 static void opal_export_symmap(void)
725 {
726 	const __be64 *syms;
727 	unsigned int size;
728 	struct device_node *fw;
729 	int rc;
730 
731 	fw = of_find_node_by_path("/ibm,opal/firmware");
732 	if (!fw)
733 		return;
734 	syms = of_get_property(fw, "symbol-map", &size);
735 	if (!syms || size != 2 * sizeof(__be64))
736 		return;
737 
738 	/* Setup attributes */
739 	symbol_map_attr.private = __va(be64_to_cpu(syms[0]));
740 	symbol_map_attr.size = be64_to_cpu(syms[1]);
741 
742 	rc = sysfs_create_bin_file(opal_kobj, &symbol_map_attr);
743 	if (rc)
744 		pr_warn("Error %d creating OPAL symbols file\n", rc);
745 }
746 
747 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
748 				struct bin_attribute *bin_attr, char *buf,
749 				loff_t off, size_t count)
750 {
751 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
752 				       bin_attr->size);
753 }
754 
755 /*
756  * opal_export_attrs: creates a sysfs node for each property listed in
757  * the device-tree under /ibm,opal/firmware/exports/
758  * All new sysfs nodes are created under /opal/exports/.
759  * This allows for reserved memory regions (e.g. HDAT) to be read.
760  * The new sysfs nodes are only readable by root.
761  */
762 static void opal_export_attrs(void)
763 {
764 	struct bin_attribute *attr;
765 	struct device_node *np;
766 	struct property *prop;
767 	struct kobject *kobj;
768 	u64 vals[2];
769 	int rc;
770 
771 	np = of_find_node_by_path("/ibm,opal/firmware/exports");
772 	if (!np)
773 		return;
774 
775 	/* Create new 'exports' directory - /sys/firmware/opal/exports */
776 	kobj = kobject_create_and_add("exports", opal_kobj);
777 	if (!kobj) {
778 		pr_warn("kobject_create_and_add() of exports failed\n");
779 		return;
780 	}
781 
782 	for_each_property_of_node(np, prop) {
783 		if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
784 			continue;
785 
786 		if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
787 			continue;
788 
789 		attr = kzalloc(sizeof(*attr), GFP_KERNEL);
790 
791 		if (attr == NULL) {
792 			pr_warn("Failed kmalloc for bin_attribute!");
793 			continue;
794 		}
795 
796 		sysfs_bin_attr_init(attr);
797 		attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
798 		attr->attr.mode = 0400;
799 		attr->read = export_attr_read;
800 		attr->private = __va(vals[0]);
801 		attr->size = vals[1];
802 
803 		if (attr->attr.name == NULL) {
804 			pr_warn("Failed kstrdup for bin_attribute attr.name");
805 			kfree(attr);
806 			continue;
807 		}
808 
809 		rc = sysfs_create_bin_file(kobj, attr);
810 		if (rc) {
811 			pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
812 				 rc, prop->name);
813 			kfree(attr->attr.name);
814 			kfree(attr);
815 		}
816 	}
817 
818 	of_node_put(np);
819 }
820 
821 static void __init opal_dump_region_init(void)
822 {
823 	void *addr;
824 	uint64_t size;
825 	int rc;
826 
827 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
828 		return;
829 
830 	/* Register kernel log buffer */
831 	addr = log_buf_addr_get();
832 	if (addr == NULL)
833 		return;
834 
835 	size = log_buf_len_get();
836 	if (size == 0)
837 		return;
838 
839 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
840 				       __pa(addr), size);
841 	/* Don't warn if this is just an older OPAL that doesn't
842 	 * know about that call
843 	 */
844 	if (rc && rc != OPAL_UNSUPPORTED)
845 		pr_warn("DUMP: Failed to register kernel log buffer. "
846 			"rc = %d\n", rc);
847 }
848 
849 static void opal_pdev_init(const char *compatible)
850 {
851 	struct device_node *np;
852 
853 	for_each_compatible_node(np, NULL, compatible)
854 		of_platform_device_create(np, NULL, NULL);
855 }
856 
857 static void __init opal_imc_init_dev(void)
858 {
859 	struct device_node *np;
860 
861 	np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
862 	if (np)
863 		of_platform_device_create(np, NULL, NULL);
864 }
865 
866 static int kopald(void *unused)
867 {
868 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
869 
870 	set_freezable();
871 	do {
872 		try_to_freeze();
873 
874 		opal_handle_events();
875 
876 		set_current_state(TASK_INTERRUPTIBLE);
877 		if (opal_have_pending_events())
878 			__set_current_state(TASK_RUNNING);
879 		else
880 			schedule_timeout(timeout);
881 
882 	} while (!kthread_should_stop());
883 
884 	return 0;
885 }
886 
887 void opal_wake_poller(void)
888 {
889 	if (kopald_tsk)
890 		wake_up_process(kopald_tsk);
891 }
892 
893 static void opal_init_heartbeat(void)
894 {
895 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
896 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
897 				 &opal_heartbeat) != 0)
898 		opal_heartbeat = 0;
899 
900 	if (opal_heartbeat)
901 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
902 }
903 
904 static int __init opal_init(void)
905 {
906 	struct device_node *np, *consoles, *leds;
907 	int rc;
908 
909 	opal_node = of_find_node_by_path("/ibm,opal");
910 	if (!opal_node) {
911 		pr_warn("Device node not found\n");
912 		return -ENODEV;
913 	}
914 
915 	/* Register OPAL consoles if any ports */
916 	consoles = of_find_node_by_path("/ibm,opal/consoles");
917 	if (consoles) {
918 		for_each_child_of_node(consoles, np) {
919 			if (!of_node_name_eq(np, "serial"))
920 				continue;
921 			of_platform_device_create(np, NULL, NULL);
922 		}
923 		of_node_put(consoles);
924 	}
925 
926 	/* Initialise OPAL messaging system */
927 	opal_message_init(opal_node);
928 
929 	/* Initialise OPAL asynchronous completion interface */
930 	opal_async_comp_init();
931 
932 	/* Initialise OPAL sensor interface */
933 	opal_sensor_init();
934 
935 	/* Initialise OPAL hypervisor maintainence interrupt handling */
936 	opal_hmi_handler_init();
937 
938 	/* Create i2c platform devices */
939 	opal_pdev_init("ibm,opal-i2c");
940 
941 	/* Handle non-volatile memory devices */
942 	opal_pdev_init("pmem-region");
943 
944 	/* Setup a heatbeat thread if requested by OPAL */
945 	opal_init_heartbeat();
946 
947 	/* Detect In-Memory Collection counters and create devices*/
948 	opal_imc_init_dev();
949 
950 	/* Create leds platform devices */
951 	leds = of_find_node_by_path("/ibm,opal/leds");
952 	if (leds) {
953 		of_platform_device_create(leds, "opal_leds", NULL);
954 		of_node_put(leds);
955 	}
956 
957 	/* Initialise OPAL message log interface */
958 	opal_msglog_init();
959 
960 	/* Create "opal" kobject under /sys/firmware */
961 	rc = opal_sysfs_init();
962 	if (rc == 0) {
963 		/* Export symbol map to userspace */
964 		opal_export_symmap();
965 		/* Setup dump region interface */
966 		opal_dump_region_init();
967 		/* Setup error log interface */
968 		rc = opal_elog_init();
969 		/* Setup code update interface */
970 		opal_flash_update_init();
971 		/* Setup platform dump extract interface */
972 		opal_platform_dump_init();
973 		/* Setup system parameters interface */
974 		opal_sys_param_init();
975 		/* Setup message log sysfs interface. */
976 		opal_msglog_sysfs_init();
977 	}
978 
979 	/* Export all properties */
980 	opal_export_attrs();
981 
982 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
983 	opal_pdev_init("ibm,opal-ipmi");
984 	opal_pdev_init("ibm,opal-flash");
985 	opal_pdev_init("ibm,opal-prd");
986 
987 	/* Initialise platform device: oppanel interface */
988 	opal_pdev_init("ibm,opal-oppanel");
989 
990 	/* Initialise OPAL kmsg dumper for flushing console on panic */
991 	opal_kmsg_init();
992 
993 	/* Initialise OPAL powercap interface */
994 	opal_powercap_init();
995 
996 	/* Initialise OPAL Power-Shifting-Ratio interface */
997 	opal_psr_init();
998 
999 	/* Initialise OPAL sensor groups */
1000 	opal_sensor_groups_init();
1001 
1002 	/* Initialise OPAL Power control interface */
1003 	opal_power_control_init();
1004 
1005 	return 0;
1006 }
1007 machine_subsys_initcall(powernv, opal_init);
1008 
1009 void opal_shutdown(void)
1010 {
1011 	long rc = OPAL_BUSY;
1012 
1013 	opal_event_shutdown();
1014 
1015 	/*
1016 	 * Then sync with OPAL which ensure anything that can
1017 	 * potentially write to our memory has completed such
1018 	 * as an ongoing dump retrieval
1019 	 */
1020 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
1021 		rc = opal_sync_host_reboot();
1022 		if (rc == OPAL_BUSY)
1023 			opal_poll_events(NULL);
1024 		else
1025 			mdelay(10);
1026 	}
1027 
1028 	/* Unregister memory dump region */
1029 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
1030 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
1031 }
1032 
1033 /* Export this so that test modules can use it */
1034 EXPORT_SYMBOL_GPL(opal_invalid_call);
1035 EXPORT_SYMBOL_GPL(opal_xscom_read);
1036 EXPORT_SYMBOL_GPL(opal_xscom_write);
1037 EXPORT_SYMBOL_GPL(opal_ipmi_send);
1038 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1039 EXPORT_SYMBOL_GPL(opal_flash_read);
1040 EXPORT_SYMBOL_GPL(opal_flash_write);
1041 EXPORT_SYMBOL_GPL(opal_flash_erase);
1042 EXPORT_SYMBOL_GPL(opal_prd_msg);
1043 EXPORT_SYMBOL_GPL(opal_check_token);
1044 
1045 /* Convert a region of vmalloc memory to an opal sg list */
1046 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1047 					     unsigned long vmalloc_size)
1048 {
1049 	struct opal_sg_list *sg, *first = NULL;
1050 	unsigned long i = 0;
1051 
1052 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1053 	if (!sg)
1054 		goto nomem;
1055 
1056 	first = sg;
1057 
1058 	while (vmalloc_size > 0) {
1059 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1060 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
1061 
1062 		sg->entry[i].data = cpu_to_be64(data);
1063 		sg->entry[i].length = cpu_to_be64(length);
1064 		i++;
1065 
1066 		if (i >= SG_ENTRIES_PER_NODE) {
1067 			struct opal_sg_list *next;
1068 
1069 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1070 			if (!next)
1071 				goto nomem;
1072 
1073 			sg->length = cpu_to_be64(
1074 					i * sizeof(struct opal_sg_entry) + 16);
1075 			i = 0;
1076 			sg->next = cpu_to_be64(__pa(next));
1077 			sg = next;
1078 		}
1079 
1080 		vmalloc_addr += length;
1081 		vmalloc_size -= length;
1082 	}
1083 
1084 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1085 
1086 	return first;
1087 
1088 nomem:
1089 	pr_err("%s : Failed to allocate memory\n", __func__);
1090 	opal_free_sg_list(first);
1091 	return NULL;
1092 }
1093 
1094 void opal_free_sg_list(struct opal_sg_list *sg)
1095 {
1096 	while (sg) {
1097 		uint64_t next = be64_to_cpu(sg->next);
1098 
1099 		kfree(sg);
1100 
1101 		if (next)
1102 			sg = __va(next);
1103 		else
1104 			sg = NULL;
1105 	}
1106 }
1107 
1108 int opal_error_code(int rc)
1109 {
1110 	switch (rc) {
1111 	case OPAL_SUCCESS:		return 0;
1112 
1113 	case OPAL_PARAMETER:		return -EINVAL;
1114 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
1115 	case OPAL_BUSY:
1116 	case OPAL_BUSY_EVENT:		return -EBUSY;
1117 	case OPAL_NO_MEM:		return -ENOMEM;
1118 	case OPAL_PERMISSION:		return -EPERM;
1119 
1120 	case OPAL_UNSUPPORTED:		return -EIO;
1121 	case OPAL_HARDWARE:		return -EIO;
1122 	case OPAL_INTERNAL_ERROR:	return -EIO;
1123 	case OPAL_TIMEOUT:		return -ETIMEDOUT;
1124 	default:
1125 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1126 		return -EIO;
1127 	}
1128 }
1129 
1130 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1131 {
1132 	int rc;
1133 
1134 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
1135 		rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1136 		if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1137 			pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1138 	}
1139 }
1140 
1141 EXPORT_SYMBOL_GPL(opal_poll_events);
1142 EXPORT_SYMBOL_GPL(opal_rtc_read);
1143 EXPORT_SYMBOL_GPL(opal_rtc_write);
1144 EXPORT_SYMBOL_GPL(opal_tpo_read);
1145 EXPORT_SYMBOL_GPL(opal_tpo_write);
1146 EXPORT_SYMBOL_GPL(opal_i2c_request);
1147 /* Export these symbols for PowerNV LED class driver */
1148 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1149 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1150 /* Export this symbol for PowerNV Operator Panel class driver */
1151 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1152 /* Export this for KVM */
1153 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1154 EXPORT_SYMBOL_GPL(opal_int_eoi);
1155 EXPORT_SYMBOL_GPL(opal_error_code);
1156 /* Export the below symbol for NX compression */
1157 EXPORT_SYMBOL(opal_nx_coproc_init);
1158