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/of_address.h>
20 #include <linux/interrupt.h>
21 #include <linux/notifier.h>
22 #include <linux/slab.h>
23 #include <linux/sched.h>
24 #include <linux/kobject.h>
25 #include <linux/delay.h>
26 #include <linux/memblock.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/kmsg_dump.h>
30 #include <linux/console.h>
31 #include <linux/sched/debug.h>
32 
33 #include <asm/machdep.h>
34 #include <asm/opal.h>
35 #include <asm/firmware.h>
36 #include <asm/mce.h>
37 #include <asm/imc-pmu.h>
38 #include <asm/bug.h>
39 
40 #include "powernv.h"
41 
42 /* /sys/firmware/opal */
43 struct kobject *opal_kobj;
44 
45 struct opal {
46 	u64 base;
47 	u64 entry;
48 	u64 size;
49 } opal;
50 
51 struct mcheck_recoverable_range {
52 	u64 start_addr;
53 	u64 end_addr;
54 	u64 recover_addr;
55 };
56 
57 static struct mcheck_recoverable_range *mc_recoverable_range;
58 static int mc_recoverable_range_len;
59 
60 struct device_node *opal_node;
61 static DEFINE_SPINLOCK(opal_write_lock);
62 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
63 static uint32_t opal_heartbeat;
64 static struct task_struct *kopald_tsk;
65 
66 void opal_configure_cores(void)
67 {
68 	u64 reinit_flags = 0;
69 
70 	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
71 	 *
72 	 * It will preserve non volatile GPRs and HSPRG0/1. It will
73 	 * also restore HIDs and other SPRs to their original value
74 	 * but it might clobber a bunch.
75 	 */
76 #ifdef __BIG_ENDIAN__
77 	reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
78 #else
79 	reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
80 #endif
81 
82 	/*
83 	 * POWER9 always support running hash:
84 	 *  ie. Host hash  supports  hash guests
85 	 *      Host radix supports  hash/radix guests
86 	 */
87 	if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
88 		reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
89 		if (early_radix_enabled())
90 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
91 	}
92 
93 	opal_reinit_cpus(reinit_flags);
94 
95 	/* Restore some bits */
96 	if (cur_cpu_spec->cpu_restore)
97 		cur_cpu_spec->cpu_restore();
98 }
99 
100 int __init early_init_dt_scan_opal(unsigned long node,
101 				   const char *uname, int depth, void *data)
102 {
103 	const void *basep, *entryp, *sizep;
104 	int basesz, entrysz, runtimesz;
105 
106 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
107 		return 0;
108 
109 	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
110 	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
111 	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
112 
113 	if (!basep || !entryp || !sizep)
114 		return 1;
115 
116 	opal.base = of_read_number(basep, basesz/4);
117 	opal.entry = of_read_number(entryp, entrysz/4);
118 	opal.size = of_read_number(sizep, runtimesz/4);
119 
120 	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
121 		 opal.base, basep, basesz);
122 	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
123 		 opal.entry, entryp, entrysz);
124 	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
125 		 opal.size, sizep, runtimesz);
126 
127 	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
128 		powerpc_firmware_features |= FW_FEATURE_OPAL;
129 		pr_debug("OPAL detected !\n");
130 	} else {
131 		panic("OPAL != V3 detected, no longer supported.\n");
132 	}
133 
134 	return 1;
135 }
136 
137 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
138 				   const char *uname, int depth, void *data)
139 {
140 	int i, psize, size;
141 	const __be32 *prop;
142 
143 	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
144 		return 0;
145 
146 	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
147 
148 	if (!prop)
149 		return 1;
150 
151 	pr_debug("Found machine check recoverable ranges.\n");
152 
153 	/*
154 	 * Calculate number of available entries.
155 	 *
156 	 * Each recoverable address range entry is (start address, len,
157 	 * recovery address), 2 cells each for start and recovery address,
158 	 * 1 cell for len, totalling 5 cells per entry.
159 	 */
160 	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
161 
162 	/* Sanity check */
163 	if (!mc_recoverable_range_len)
164 		return 1;
165 
166 	/* Size required to hold all the entries. */
167 	size = mc_recoverable_range_len *
168 			sizeof(struct mcheck_recoverable_range);
169 
170 	/*
171 	 * Allocate a buffer to hold the MC recoverable ranges.
172 	 */
173 	mc_recoverable_range = memblock_alloc(size, __alignof__(u64));
174 	if (!mc_recoverable_range)
175 		panic("%s: Failed to allocate %u bytes align=0x%lx\n",
176 		      __func__, size, __alignof__(u64));
177 
178 	for (i = 0; i < mc_recoverable_range_len; i++) {
179 		mc_recoverable_range[i].start_addr =
180 					of_read_number(prop + (i * 5) + 0, 2);
181 		mc_recoverable_range[i].end_addr =
182 					mc_recoverable_range[i].start_addr +
183 					of_read_number(prop + (i * 5) + 2, 1);
184 		mc_recoverable_range[i].recover_addr =
185 					of_read_number(prop + (i * 5) + 3, 2);
186 
187 		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
188 				mc_recoverable_range[i].start_addr,
189 				mc_recoverable_range[i].end_addr,
190 				mc_recoverable_range[i].recover_addr);
191 	}
192 	return 1;
193 }
194 
195 static int __init opal_register_exception_handlers(void)
196 {
197 #ifdef __BIG_ENDIAN__
198 	u64 glue;
199 
200 	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
201 		return -ENODEV;
202 
203 	/* Hookup some exception handlers except machine check. We use the
204 	 * fwnmi area at 0x7000 to provide the glue space to OPAL
205 	 */
206 	glue = 0x7000;
207 
208 	/*
209 	 * Check if we are running on newer firmware that exports
210 	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
211 	 * the HMI interrupt and we catch it directly in Linux.
212 	 *
213 	 * For older firmware (i.e currently released POWER8 System Firmware
214 	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
215 	 * patch the HMI vector and handle it inside OPAL firmware.
216 	 *
217 	 * For newer firmware (in development/yet to be released) we will
218 	 * start catching/handling 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 	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
229 #endif
230 
231 	return 0;
232 }
233 machine_early_initcall(powernv, opal_register_exception_handlers);
234 
235 /*
236  * Opal message notifier based on message type. Allow subscribers to get
237  * notified for specific messgae type.
238  */
239 int opal_message_notifier_register(enum opal_msg_type msg_type,
240 					struct notifier_block *nb)
241 {
242 	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
243 		pr_warn("%s: Invalid arguments, msg_type:%d\n",
244 			__func__, msg_type);
245 		return -EINVAL;
246 	}
247 
248 	return atomic_notifier_chain_register(
249 				&opal_msg_notifier_head[msg_type], nb);
250 }
251 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
252 
253 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
254 				     struct notifier_block *nb)
255 {
256 	return atomic_notifier_chain_unregister(
257 			&opal_msg_notifier_head[msg_type], nb);
258 }
259 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
260 
261 static void opal_message_do_notify(uint32_t msg_type, void *msg)
262 {
263 	/* notify subscribers */
264 	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
265 					msg_type, msg);
266 }
267 
268 static void opal_handle_message(void)
269 {
270 	s64 ret;
271 	/*
272 	 * TODO: pre-allocate a message buffer depending on opal-msg-size
273 	 * value in /proc/device-tree.
274 	 */
275 	static struct opal_msg msg;
276 	u32 type;
277 
278 	ret = opal_get_msg(__pa(&msg), sizeof(msg));
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(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 *)&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(void)
307 {
308 	int ret, i, irq;
309 
310 	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
311 		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
312 
313 	irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
314 	if (!irq) {
315 		pr_err("%s: Can't register OPAL event irq (%d)\n",
316 		       __func__, irq);
317 		return irq;
318 	}
319 
320 	ret = request_irq(irq, opal_message_notify,
321 			IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
322 	if (ret) {
323 		pr_err("%s: Can't request OPAL event irq (%d)\n",
324 		       __func__, ret);
325 		return ret;
326 	}
327 
328 	return 0;
329 }
330 
331 int opal_get_chars(uint32_t vtermno, char *buf, int count)
332 {
333 	s64 rc;
334 	__be64 evt, len;
335 
336 	if (!opal.entry)
337 		return -ENODEV;
338 	opal_poll_events(&evt);
339 	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
340 		return 0;
341 	len = cpu_to_be64(count);
342 	rc = opal_console_read(vtermno, &len, buf);
343 	if (rc == OPAL_SUCCESS)
344 		return be64_to_cpu(len);
345 	return 0;
346 }
347 
348 static int __opal_put_chars(uint32_t vtermno, const char *data, int total_len, bool atomic)
349 {
350 	unsigned long flags = 0 /* shut up gcc */;
351 	int written;
352 	__be64 olen;
353 	s64 rc;
354 
355 	if (!opal.entry)
356 		return -ENODEV;
357 
358 	if (atomic)
359 		spin_lock_irqsave(&opal_write_lock, flags);
360 	rc = opal_console_write_buffer_space(vtermno, &olen);
361 	if (rc || be64_to_cpu(olen) < total_len) {
362 		/* Closed -> drop characters */
363 		if (rc)
364 			written = total_len;
365 		else
366 			written = -EAGAIN;
367 		goto out;
368 	}
369 
370 	/* Should not get a partial write here because space is available. */
371 	olen = cpu_to_be64(total_len);
372 	rc = opal_console_write(vtermno, &olen, data);
373 	if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
374 		if (rc == OPAL_BUSY_EVENT)
375 			opal_poll_events(NULL);
376 		written = -EAGAIN;
377 		goto out;
378 	}
379 
380 	/* Closed or other error drop */
381 	if (rc != OPAL_SUCCESS) {
382 		written = opal_error_code(rc);
383 		goto out;
384 	}
385 
386 	written = be64_to_cpu(olen);
387 	if (written < total_len) {
388 		if (atomic) {
389 			/* Should not happen */
390 			pr_warn("atomic console write returned partial "
391 				"len=%d written=%d\n", total_len, written);
392 		}
393 		if (!written)
394 			written = -EAGAIN;
395 	}
396 
397 out:
398 	if (atomic)
399 		spin_unlock_irqrestore(&opal_write_lock, flags);
400 
401 	return written;
402 }
403 
404 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
405 {
406 	return __opal_put_chars(vtermno, data, total_len, false);
407 }
408 
409 /*
410  * opal_put_chars_atomic will not perform partial-writes. Data will be
411  * atomically written to the terminal or not at all. This is not strictly
412  * true at the moment because console space can race with OPAL's console
413  * writes.
414  */
415 int opal_put_chars_atomic(uint32_t vtermno, const char *data, int total_len)
416 {
417 	return __opal_put_chars(vtermno, data, total_len, true);
418 }
419 
420 static s64 __opal_flush_console(uint32_t vtermno)
421 {
422 	s64 rc;
423 
424 	if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
425 		__be64 evt;
426 
427 		/*
428 		 * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
429 		 * the console can still be flushed by calling the polling
430 		 * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
431 		 */
432 		WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
433 
434 		opal_poll_events(&evt);
435 		if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
436 			return OPAL_SUCCESS;
437 		return OPAL_BUSY;
438 
439 	} else {
440 		rc = opal_console_flush(vtermno);
441 		if (rc == OPAL_BUSY_EVENT) {
442 			opal_poll_events(NULL);
443 			rc = OPAL_BUSY;
444 		}
445 		return rc;
446 	}
447 
448 }
449 
450 /*
451  * opal_flush_console spins until the console is flushed
452  */
453 int opal_flush_console(uint32_t vtermno)
454 {
455 	for (;;) {
456 		s64 rc = __opal_flush_console(vtermno);
457 
458 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
459 			mdelay(1);
460 			continue;
461 		}
462 
463 		return opal_error_code(rc);
464 	}
465 }
466 
467 /*
468  * opal_flush_chars is an hvc interface that sleeps until the console is
469  * flushed if wait, otherwise it will return -EBUSY if the console has data,
470  * -EAGAIN if it has data and some of it was flushed.
471  */
472 int opal_flush_chars(uint32_t vtermno, bool wait)
473 {
474 	for (;;) {
475 		s64 rc = __opal_flush_console(vtermno);
476 
477 		if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
478 			if (wait) {
479 				msleep(OPAL_BUSY_DELAY_MS);
480 				continue;
481 			}
482 			if (rc == OPAL_PARTIAL)
483 				return -EAGAIN;
484 		}
485 
486 		return opal_error_code(rc);
487 	}
488 }
489 
490 static int opal_recover_mce(struct pt_regs *regs,
491 					struct machine_check_event *evt)
492 {
493 	int recovered = 0;
494 
495 	if (!(regs->msr & MSR_RI)) {
496 		/* If MSR_RI isn't set, we cannot recover */
497 		pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
498 		recovered = 0;
499 	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
500 		/* Platform corrected itself */
501 		recovered = 1;
502 	} else if (evt->severity == MCE_SEV_FATAL) {
503 		/* Fatal machine check */
504 		pr_err("Machine check interrupt is fatal\n");
505 		recovered = 0;
506 	}
507 
508 	if (!recovered && evt->severity == MCE_SEV_ERROR_SYNC) {
509 		/*
510 		 * Try to kill processes if we get a synchronous machine check
511 		 * (e.g., one caused by execution of this instruction). This
512 		 * will devolve into a panic if we try to kill init or are in
513 		 * an interrupt etc.
514 		 *
515 		 * TODO: Queue up this address for hwpoisioning later.
516 		 * TODO: This is not quite right for d-side machine
517 		 *       checks ->nip is not necessarily the important
518 		 *       address.
519 		 */
520 		if ((user_mode(regs))) {
521 			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
522 			recovered = 1;
523 		} else if (die_will_crash()) {
524 			/*
525 			 * die() would kill the kernel, so better to go via
526 			 * the platform reboot code that will log the
527 			 * machine check.
528 			 */
529 			recovered = 0;
530 		} else {
531 			die("Machine check", regs, SIGBUS);
532 			recovered = 1;
533 		}
534 	}
535 
536 	return recovered;
537 }
538 
539 void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
540 {
541 	panic_flush_kmsg_start();
542 
543 	pr_emerg("Hardware platform error: %s\n", msg);
544 	if (regs)
545 		show_regs(regs);
546 	smp_send_stop();
547 
548 	panic_flush_kmsg_end();
549 
550 	/*
551 	 * Don't bother to shut things down because this will
552 	 * xstop the system.
553 	 */
554 	if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
555 						== OPAL_UNSUPPORTED) {
556 		pr_emerg("Reboot type %d not supported for %s\n",
557 				OPAL_REBOOT_PLATFORM_ERROR, msg);
558 	}
559 
560 	/*
561 	 * We reached here. There can be three possibilities:
562 	 * 1. We are running on a firmware level that do not support
563 	 *    opal_cec_reboot2()
564 	 * 2. We are running on a firmware level that do not support
565 	 *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
566 	 * 3. We are running on FSP based system that does not need
567 	 *    opal to trigger checkstop explicitly for error analysis.
568 	 *    The FSP PRD component would have already got notified
569 	 *    about this error through other channels.
570 	 * 4. We are running on a newer skiboot that by default does
571 	 *    not cause a checkstop, drops us back to the kernel to
572 	 *    extract context and state at the time of the error.
573 	 */
574 
575 	panic(msg);
576 }
577 
578 int opal_machine_check(struct pt_regs *regs)
579 {
580 	struct machine_check_event evt;
581 
582 	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
583 		return 0;
584 
585 	/* Print things out */
586 	if (evt.version != MCE_V1) {
587 		pr_err("Machine Check Exception, Unknown event version %d !\n",
588 		       evt.version);
589 		return 0;
590 	}
591 	machine_check_print_event_info(&evt, user_mode(regs), false);
592 
593 	if (opal_recover_mce(regs, &evt))
594 		return 1;
595 
596 	pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
597 }
598 
599 /* Early hmi handler called in real mode. */
600 int opal_hmi_exception_early(struct pt_regs *regs)
601 {
602 	s64 rc;
603 
604 	/*
605 	 * call opal hmi handler. Pass paca address as token.
606 	 * The return value OPAL_SUCCESS is an indication that there is
607 	 * an HMI event generated waiting to pull by Linux.
608 	 */
609 	rc = opal_handle_hmi();
610 	if (rc == OPAL_SUCCESS) {
611 		local_paca->hmi_event_available = 1;
612 		return 1;
613 	}
614 	return 0;
615 }
616 
617 /* HMI exception handler called in virtual mode during check_irq_replay. */
618 int opal_handle_hmi_exception(struct pt_regs *regs)
619 {
620 	/*
621 	 * Check if HMI event is available.
622 	 * if Yes, then wake kopald to process them.
623 	 */
624 	if (!local_paca->hmi_event_available)
625 		return 0;
626 
627 	local_paca->hmi_event_available = 0;
628 	opal_wake_poller();
629 
630 	return 1;
631 }
632 
633 static uint64_t find_recovery_address(uint64_t nip)
634 {
635 	int i;
636 
637 	for (i = 0; i < mc_recoverable_range_len; i++)
638 		if ((nip >= mc_recoverable_range[i].start_addr) &&
639 		    (nip < mc_recoverable_range[i].end_addr))
640 		    return mc_recoverable_range[i].recover_addr;
641 	return 0;
642 }
643 
644 bool opal_mce_check_early_recovery(struct pt_regs *regs)
645 {
646 	uint64_t recover_addr = 0;
647 
648 	if (!opal.base || !opal.size)
649 		goto out;
650 
651 	if ((regs->nip >= opal.base) &&
652 			(regs->nip < (opal.base + opal.size)))
653 		recover_addr = find_recovery_address(regs->nip);
654 
655 	/*
656 	 * Setup regs->nip to rfi into fixup address.
657 	 */
658 	if (recover_addr)
659 		regs->nip = recover_addr;
660 
661 out:
662 	return !!recover_addr;
663 }
664 
665 static int opal_sysfs_init(void)
666 {
667 	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
668 	if (!opal_kobj) {
669 		pr_warn("kobject_create_and_add opal failed\n");
670 		return -ENOMEM;
671 	}
672 
673 	return 0;
674 }
675 
676 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
677 			       struct bin_attribute *bin_attr,
678 			       char *buf, loff_t off, size_t count)
679 {
680 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
681 				       bin_attr->size);
682 }
683 
684 static BIN_ATTR_RO(symbol_map, 0);
685 
686 static void opal_export_symmap(void)
687 {
688 	const __be64 *syms;
689 	unsigned int size;
690 	struct device_node *fw;
691 	int rc;
692 
693 	fw = of_find_node_by_path("/ibm,opal/firmware");
694 	if (!fw)
695 		return;
696 	syms = of_get_property(fw, "symbol-map", &size);
697 	if (!syms || size != 2 * sizeof(__be64))
698 		return;
699 
700 	/* Setup attributes */
701 	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
702 	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
703 
704 	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
705 	if (rc)
706 		pr_warn("Error %d creating OPAL symbols file\n", rc);
707 }
708 
709 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
710 				struct bin_attribute *bin_attr, char *buf,
711 				loff_t off, size_t count)
712 {
713 	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
714 				       bin_attr->size);
715 }
716 
717 /*
718  * opal_export_attrs: creates a sysfs node for each property listed in
719  * the device-tree under /ibm,opal/firmware/exports/
720  * All new sysfs nodes are created under /opal/exports/.
721  * This allows for reserved memory regions (e.g. HDAT) to be read.
722  * The new sysfs nodes are only readable by root.
723  */
724 static void opal_export_attrs(void)
725 {
726 	struct bin_attribute *attr;
727 	struct device_node *np;
728 	struct property *prop;
729 	struct kobject *kobj;
730 	u64 vals[2];
731 	int rc;
732 
733 	np = of_find_node_by_path("/ibm,opal/firmware/exports");
734 	if (!np)
735 		return;
736 
737 	/* Create new 'exports' directory - /sys/firmware/opal/exports */
738 	kobj = kobject_create_and_add("exports", opal_kobj);
739 	if (!kobj) {
740 		pr_warn("kobject_create_and_add() of exports failed\n");
741 		return;
742 	}
743 
744 	for_each_property_of_node(np, prop) {
745 		if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
746 			continue;
747 
748 		if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
749 			continue;
750 
751 		attr = kzalloc(sizeof(*attr), GFP_KERNEL);
752 
753 		if (attr == NULL) {
754 			pr_warn("Failed kmalloc for bin_attribute!");
755 			continue;
756 		}
757 
758 		sysfs_bin_attr_init(attr);
759 		attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
760 		attr->attr.mode = 0400;
761 		attr->read = export_attr_read;
762 		attr->private = __va(vals[0]);
763 		attr->size = vals[1];
764 
765 		if (attr->attr.name == NULL) {
766 			pr_warn("Failed kstrdup for bin_attribute attr.name");
767 			kfree(attr);
768 			continue;
769 		}
770 
771 		rc = sysfs_create_bin_file(kobj, attr);
772 		if (rc) {
773 			pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
774 				 rc, prop->name);
775 			kfree(attr->attr.name);
776 			kfree(attr);
777 		}
778 	}
779 
780 	of_node_put(np);
781 }
782 
783 static void __init opal_dump_region_init(void)
784 {
785 	void *addr;
786 	uint64_t size;
787 	int rc;
788 
789 	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
790 		return;
791 
792 	/* Register kernel log buffer */
793 	addr = log_buf_addr_get();
794 	if (addr == NULL)
795 		return;
796 
797 	size = log_buf_len_get();
798 	if (size == 0)
799 		return;
800 
801 	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
802 				       __pa(addr), size);
803 	/* Don't warn if this is just an older OPAL that doesn't
804 	 * know about that call
805 	 */
806 	if (rc && rc != OPAL_UNSUPPORTED)
807 		pr_warn("DUMP: Failed to register kernel log buffer. "
808 			"rc = %d\n", rc);
809 }
810 
811 static void opal_pdev_init(const char *compatible)
812 {
813 	struct device_node *np;
814 
815 	for_each_compatible_node(np, NULL, compatible)
816 		of_platform_device_create(np, NULL, NULL);
817 }
818 
819 static void __init opal_imc_init_dev(void)
820 {
821 	struct device_node *np;
822 
823 	np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
824 	if (np)
825 		of_platform_device_create(np, NULL, NULL);
826 }
827 
828 static int kopald(void *unused)
829 {
830 	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
831 
832 	set_freezable();
833 	do {
834 		try_to_freeze();
835 
836 		opal_handle_events();
837 
838 		set_current_state(TASK_INTERRUPTIBLE);
839 		if (opal_have_pending_events())
840 			__set_current_state(TASK_RUNNING);
841 		else
842 			schedule_timeout(timeout);
843 
844 	} while (!kthread_should_stop());
845 
846 	return 0;
847 }
848 
849 void opal_wake_poller(void)
850 {
851 	if (kopald_tsk)
852 		wake_up_process(kopald_tsk);
853 }
854 
855 static void opal_init_heartbeat(void)
856 {
857 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
858 	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
859 				 &opal_heartbeat) != 0)
860 		opal_heartbeat = 0;
861 
862 	if (opal_heartbeat)
863 		kopald_tsk = kthread_run(kopald, NULL, "kopald");
864 }
865 
866 static int __init opal_init(void)
867 {
868 	struct device_node *np, *consoles, *leds;
869 	int rc;
870 
871 	opal_node = of_find_node_by_path("/ibm,opal");
872 	if (!opal_node) {
873 		pr_warn("Device node not found\n");
874 		return -ENODEV;
875 	}
876 
877 	/* Register OPAL consoles if any ports */
878 	consoles = of_find_node_by_path("/ibm,opal/consoles");
879 	if (consoles) {
880 		for_each_child_of_node(consoles, np) {
881 			if (!of_node_name_eq(np, "serial"))
882 				continue;
883 			of_platform_device_create(np, NULL, NULL);
884 		}
885 		of_node_put(consoles);
886 	}
887 
888 	/* Initialise OPAL messaging system */
889 	opal_message_init();
890 
891 	/* Initialise OPAL asynchronous completion interface */
892 	opal_async_comp_init();
893 
894 	/* Initialise OPAL sensor interface */
895 	opal_sensor_init();
896 
897 	/* Initialise OPAL hypervisor maintainence interrupt handling */
898 	opal_hmi_handler_init();
899 
900 	/* Create i2c platform devices */
901 	opal_pdev_init("ibm,opal-i2c");
902 
903 	/* Handle non-volatile memory devices */
904 	opal_pdev_init("pmem-region");
905 
906 	/* Setup a heatbeat thread if requested by OPAL */
907 	opal_init_heartbeat();
908 
909 	/* Detect In-Memory Collection counters and create devices*/
910 	opal_imc_init_dev();
911 
912 	/* Create leds platform devices */
913 	leds = of_find_node_by_path("/ibm,opal/leds");
914 	if (leds) {
915 		of_platform_device_create(leds, "opal_leds", NULL);
916 		of_node_put(leds);
917 	}
918 
919 	/* Initialise OPAL message log interface */
920 	opal_msglog_init();
921 
922 	/* Create "opal" kobject under /sys/firmware */
923 	rc = opal_sysfs_init();
924 	if (rc == 0) {
925 		/* Export symbol map to userspace */
926 		opal_export_symmap();
927 		/* Setup dump region interface */
928 		opal_dump_region_init();
929 		/* Setup error log interface */
930 		rc = opal_elog_init();
931 		/* Setup code update interface */
932 		opal_flash_update_init();
933 		/* Setup platform dump extract interface */
934 		opal_platform_dump_init();
935 		/* Setup system parameters interface */
936 		opal_sys_param_init();
937 		/* Setup message log sysfs interface. */
938 		opal_msglog_sysfs_init();
939 	}
940 
941 	/* Export all properties */
942 	opal_export_attrs();
943 
944 	/* Initialize platform devices: IPMI backend, PRD & flash interface */
945 	opal_pdev_init("ibm,opal-ipmi");
946 	opal_pdev_init("ibm,opal-flash");
947 	opal_pdev_init("ibm,opal-prd");
948 
949 	/* Initialise platform device: oppanel interface */
950 	opal_pdev_init("ibm,opal-oppanel");
951 
952 	/* Initialise OPAL kmsg dumper for flushing console on panic */
953 	opal_kmsg_init();
954 
955 	/* Initialise OPAL powercap interface */
956 	opal_powercap_init();
957 
958 	/* Initialise OPAL Power-Shifting-Ratio interface */
959 	opal_psr_init();
960 
961 	/* Initialise OPAL sensor groups */
962 	opal_sensor_groups_init();
963 
964 	/* Initialise OPAL Power control interface */
965 	opal_power_control_init();
966 
967 	return 0;
968 }
969 machine_subsys_initcall(powernv, opal_init);
970 
971 void opal_shutdown(void)
972 {
973 	long rc = OPAL_BUSY;
974 
975 	opal_event_shutdown();
976 
977 	/*
978 	 * Then sync with OPAL which ensure anything that can
979 	 * potentially write to our memory has completed such
980 	 * as an ongoing dump retrieval
981 	 */
982 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
983 		rc = opal_sync_host_reboot();
984 		if (rc == OPAL_BUSY)
985 			opal_poll_events(NULL);
986 		else
987 			mdelay(10);
988 	}
989 
990 	/* Unregister memory dump region */
991 	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
992 		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
993 }
994 
995 /* Export this so that test modules can use it */
996 EXPORT_SYMBOL_GPL(opal_invalid_call);
997 EXPORT_SYMBOL_GPL(opal_xscom_read);
998 EXPORT_SYMBOL_GPL(opal_xscom_write);
999 EXPORT_SYMBOL_GPL(opal_ipmi_send);
1000 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1001 EXPORT_SYMBOL_GPL(opal_flash_read);
1002 EXPORT_SYMBOL_GPL(opal_flash_write);
1003 EXPORT_SYMBOL_GPL(opal_flash_erase);
1004 EXPORT_SYMBOL_GPL(opal_prd_msg);
1005 EXPORT_SYMBOL_GPL(opal_check_token);
1006 
1007 /* Convert a region of vmalloc memory to an opal sg list */
1008 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1009 					     unsigned long vmalloc_size)
1010 {
1011 	struct opal_sg_list *sg, *first = NULL;
1012 	unsigned long i = 0;
1013 
1014 	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1015 	if (!sg)
1016 		goto nomem;
1017 
1018 	first = sg;
1019 
1020 	while (vmalloc_size > 0) {
1021 		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1022 		uint64_t length = min(vmalloc_size, PAGE_SIZE);
1023 
1024 		sg->entry[i].data = cpu_to_be64(data);
1025 		sg->entry[i].length = cpu_to_be64(length);
1026 		i++;
1027 
1028 		if (i >= SG_ENTRIES_PER_NODE) {
1029 			struct opal_sg_list *next;
1030 
1031 			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1032 			if (!next)
1033 				goto nomem;
1034 
1035 			sg->length = cpu_to_be64(
1036 					i * sizeof(struct opal_sg_entry) + 16);
1037 			i = 0;
1038 			sg->next = cpu_to_be64(__pa(next));
1039 			sg = next;
1040 		}
1041 
1042 		vmalloc_addr += length;
1043 		vmalloc_size -= length;
1044 	}
1045 
1046 	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1047 
1048 	return first;
1049 
1050 nomem:
1051 	pr_err("%s : Failed to allocate memory\n", __func__);
1052 	opal_free_sg_list(first);
1053 	return NULL;
1054 }
1055 
1056 void opal_free_sg_list(struct opal_sg_list *sg)
1057 {
1058 	while (sg) {
1059 		uint64_t next = be64_to_cpu(sg->next);
1060 
1061 		kfree(sg);
1062 
1063 		if (next)
1064 			sg = __va(next);
1065 		else
1066 			sg = NULL;
1067 	}
1068 }
1069 
1070 int opal_error_code(int rc)
1071 {
1072 	switch (rc) {
1073 	case OPAL_SUCCESS:		return 0;
1074 
1075 	case OPAL_PARAMETER:		return -EINVAL;
1076 	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
1077 	case OPAL_BUSY:
1078 	case OPAL_BUSY_EVENT:		return -EBUSY;
1079 	case OPAL_NO_MEM:		return -ENOMEM;
1080 	case OPAL_PERMISSION:		return -EPERM;
1081 
1082 	case OPAL_UNSUPPORTED:		return -EIO;
1083 	case OPAL_HARDWARE:		return -EIO;
1084 	case OPAL_INTERNAL_ERROR:	return -EIO;
1085 	case OPAL_TIMEOUT:		return -ETIMEDOUT;
1086 	default:
1087 		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1088 		return -EIO;
1089 	}
1090 }
1091 
1092 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1093 {
1094 	int rc;
1095 
1096 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
1097 		rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1098 		if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1099 			pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1100 	}
1101 }
1102 
1103 EXPORT_SYMBOL_GPL(opal_poll_events);
1104 EXPORT_SYMBOL_GPL(opal_rtc_read);
1105 EXPORT_SYMBOL_GPL(opal_rtc_write);
1106 EXPORT_SYMBOL_GPL(opal_tpo_read);
1107 EXPORT_SYMBOL_GPL(opal_tpo_write);
1108 EXPORT_SYMBOL_GPL(opal_i2c_request);
1109 /* Export these symbols for PowerNV LED class driver */
1110 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1111 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1112 /* Export this symbol for PowerNV Operator Panel class driver */
1113 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1114 /* Export this for KVM */
1115 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1116 EXPORT_SYMBOL_GPL(opal_int_eoi);
1117 EXPORT_SYMBOL_GPL(opal_error_code);
1118 /* Export the below symbol for NX compression */
1119 EXPORT_SYMBOL(opal_nx_coproc_init);
1120