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 */
37struct kobject *opal_kobj;
38
39struct opal {
40	u64 base;
41	u64 entry;
42	u64 size;
43} opal;
44
45struct mcheck_recoverable_range {
46	u64 start_addr;
47	u64 end_addr;
48	u64 recover_addr;
49};
50
51static struct mcheck_recoverable_range *mc_recoverable_range;
52static int mc_recoverable_range_len;
53
54struct device_node *opal_node;
55static DEFINE_SPINLOCK(opal_write_lock);
56static unsigned int *opal_irqs;
57static unsigned int opal_irq_count;
58static ATOMIC_NOTIFIER_HEAD(opal_notifier_head);
59static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
60static DEFINE_SPINLOCK(opal_notifier_lock);
61static uint64_t last_notified_mask = 0x0ul;
62static atomic_t opal_notifier_hold = ATOMIC_INIT(0);
63static uint32_t opal_heartbeat;
64
65static void opal_reinit_cores(void)
66{
67	/* Do the actual re-init, This will clobber all FPRs, VRs, etc...
68	 *
69	 * It will preserve non volatile GPRs and HSPRG0/1. It will
70	 * also restore HIDs and other SPRs to their original value
71	 * but it might clobber a bunch.
72	 */
73#ifdef __BIG_ENDIAN__
74	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
75#else
76	opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
77#endif
78}
79
80int __init early_init_dt_scan_opal(unsigned long node,
81				   const char *uname, int depth, void *data)
82{
83	const void *basep, *entryp, *sizep;
84	int basesz, entrysz, runtimesz;
85
86	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
87		return 0;
88
89	basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
90	entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
91	sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
92
93	if (!basep || !entryp || !sizep)
94		return 1;
95
96	opal.base = of_read_number(basep, basesz/4);
97	opal.entry = of_read_number(entryp, entrysz/4);
98	opal.size = of_read_number(sizep, runtimesz/4);
99
100	pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
101		 opal.base, basep, basesz);
102	pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
103		 opal.entry, entryp, entrysz);
104	pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
105		 opal.size, sizep, runtimesz);
106
107	powerpc_firmware_features |= FW_FEATURE_OPAL;
108	if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
109		powerpc_firmware_features |= FW_FEATURE_OPALv2;
110		powerpc_firmware_features |= FW_FEATURE_OPALv3;
111		pr_info("OPAL V3 detected !\n");
112	} else if (of_flat_dt_is_compatible(node, "ibm,opal-v2")) {
113		powerpc_firmware_features |= FW_FEATURE_OPALv2;
114		pr_info("OPAL V2 detected !\n");
115	} else {
116		pr_info("OPAL V1 detected !\n");
117	}
118
119	/* Reinit all cores with the right endian */
120	opal_reinit_cores();
121
122	/* Restore some bits */
123	if (cur_cpu_spec->cpu_restore)
124		cur_cpu_spec->cpu_restore();
125
126	return 1;
127}
128
129int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
130				   const char *uname, int depth, void *data)
131{
132	int i, psize, size;
133	const __be32 *prop;
134
135	if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
136		return 0;
137
138	prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
139
140	if (!prop)
141		return 1;
142
143	pr_debug("Found machine check recoverable ranges.\n");
144
145	/*
146	 * Calculate number of available entries.
147	 *
148	 * Each recoverable address range entry is (start address, len,
149	 * recovery address), 2 cells each for start and recovery address,
150	 * 1 cell for len, totalling 5 cells per entry.
151	 */
152	mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
153
154	/* Sanity check */
155	if (!mc_recoverable_range_len)
156		return 1;
157
158	/* Size required to hold all the entries. */
159	size = mc_recoverable_range_len *
160			sizeof(struct mcheck_recoverable_range);
161
162	/*
163	 * Allocate a buffer to hold the MC recoverable ranges. We would be
164	 * accessing them in real mode, hence it needs to be within
165	 * RMO region.
166	 */
167	mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
168							ppc64_rma_size));
169	memset(mc_recoverable_range, 0, size);
170
171	for (i = 0; i < mc_recoverable_range_len; i++) {
172		mc_recoverable_range[i].start_addr =
173					of_read_number(prop + (i * 5) + 0, 2);
174		mc_recoverable_range[i].end_addr =
175					mc_recoverable_range[i].start_addr +
176					of_read_number(prop + (i * 5) + 2, 1);
177		mc_recoverable_range[i].recover_addr =
178					of_read_number(prop + (i * 5) + 3, 2);
179
180		pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
181				mc_recoverable_range[i].start_addr,
182				mc_recoverable_range[i].end_addr,
183				mc_recoverable_range[i].recover_addr);
184	}
185	return 1;
186}
187
188static int __init opal_register_exception_handlers(void)
189{
190#ifdef __BIG_ENDIAN__
191	u64 glue;
192
193	if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
194		return -ENODEV;
195
196	/* Hookup some exception handlers except machine check. We use the
197	 * fwnmi area at 0x7000 to provide the glue space to OPAL
198	 */
199	glue = 0x7000;
200
201	/*
202	 * Check if we are running on newer firmware that exports
203	 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
204	 * the HMI interrupt and we catch it directly in Linux.
205	 *
206	 * For older firmware (i.e currently released POWER8 System Firmware
207	 * as of today <= SV810_087), we fallback to old behavior and let OPAL
208	 * patch the HMI vector and handle it inside OPAL firmware.
209	 *
210	 * For newer firmware (in development/yet to be released) we will
211	 * start catching/handling HMI directly in Linux.
212	 */
213	if (!opal_check_token(OPAL_HANDLE_HMI)) {
214		pr_info("Old firmware detected, OPAL handles HMIs.\n");
215		opal_register_exception_handler(
216				OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
217				0, glue);
218		glue += 128;
219	}
220
221	opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
222#endif
223
224	return 0;
225}
226machine_early_initcall(powernv, opal_register_exception_handlers);
227
228int opal_notifier_register(struct notifier_block *nb)
229{
230	if (!nb) {
231		pr_warning("%s: Invalid argument (%p)\n",
232			   __func__, nb);
233		return -EINVAL;
234	}
235
236	atomic_notifier_chain_register(&opal_notifier_head, nb);
237	return 0;
238}
239EXPORT_SYMBOL_GPL(opal_notifier_register);
240
241int opal_notifier_unregister(struct notifier_block *nb)
242{
243	if (!nb) {
244		pr_warning("%s: Invalid argument (%p)\n",
245			   __func__, nb);
246		return -EINVAL;
247	}
248
249	atomic_notifier_chain_unregister(&opal_notifier_head, nb);
250	return 0;
251}
252EXPORT_SYMBOL_GPL(opal_notifier_unregister);
253
254static void opal_do_notifier(uint64_t events)
255{
256	unsigned long flags;
257	uint64_t changed_mask;
258
259	if (atomic_read(&opal_notifier_hold))
260		return;
261
262	spin_lock_irqsave(&opal_notifier_lock, flags);
263	changed_mask = last_notified_mask ^ events;
264	last_notified_mask = events;
265	spin_unlock_irqrestore(&opal_notifier_lock, flags);
266
267	/*
268	 * We feed with the event bits and changed bits for
269	 * enough information to the callback.
270	 */
271	atomic_notifier_call_chain(&opal_notifier_head,
272				   events, (void *)changed_mask);
273}
274
275void opal_notifier_update_evt(uint64_t evt_mask,
276			      uint64_t evt_val)
277{
278	unsigned long flags;
279
280	spin_lock_irqsave(&opal_notifier_lock, flags);
281	last_notified_mask &= ~evt_mask;
282	last_notified_mask |= evt_val;
283	spin_unlock_irqrestore(&opal_notifier_lock, flags);
284}
285
286void opal_notifier_enable(void)
287{
288	int64_t rc;
289	__be64 evt = 0;
290
291	atomic_set(&opal_notifier_hold, 0);
292
293	/* Process pending events */
294	rc = opal_poll_events(&evt);
295	if (rc == OPAL_SUCCESS && evt)
296		opal_do_notifier(be64_to_cpu(evt));
297}
298
299void opal_notifier_disable(void)
300{
301	atomic_set(&opal_notifier_hold, 1);
302}
303
304/*
305 * Opal message notifier based on message type. Allow subscribers to get
306 * notified for specific messgae type.
307 */
308int opal_message_notifier_register(enum opal_msg_type msg_type,
309					struct notifier_block *nb)
310{
311	if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
312		pr_warning("%s: Invalid arguments, msg_type:%d\n",
313			   __func__, msg_type);
314		return -EINVAL;
315	}
316
317	return atomic_notifier_chain_register(
318				&opal_msg_notifier_head[msg_type], nb);
319}
320
321int opal_message_notifier_unregister(enum opal_msg_type msg_type,
322				     struct notifier_block *nb)
323{
324	return atomic_notifier_chain_unregister(
325			&opal_msg_notifier_head[msg_type], nb);
326}
327
328static void opal_message_do_notify(uint32_t msg_type, void *msg)
329{
330	/* notify subscribers */
331	atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
332					msg_type, msg);
333}
334
335static void opal_handle_message(void)
336{
337	s64 ret;
338	/*
339	 * TODO: pre-allocate a message buffer depending on opal-msg-size
340	 * value in /proc/device-tree.
341	 */
342	static struct opal_msg msg;
343	u32 type;
344
345	ret = opal_get_msg(__pa(&msg), sizeof(msg));
346	/* No opal message pending. */
347	if (ret == OPAL_RESOURCE)
348		return;
349
350	/* check for errors. */
351	if (ret) {
352		pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
353				__func__, ret);
354		return;
355	}
356
357	type = be32_to_cpu(msg.msg_type);
358
359	/* Sanity check */
360	if (type >= OPAL_MSG_TYPE_MAX) {
361		pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
362		return;
363	}
364	opal_message_do_notify(type, (void *)&msg);
365}
366
367static int opal_message_notify(struct notifier_block *nb,
368			  unsigned long events, void *change)
369{
370	if (events & OPAL_EVENT_MSG_PENDING)
371		opal_handle_message();
372	return 0;
373}
374
375static struct notifier_block opal_message_nb = {
376	.notifier_call	= opal_message_notify,
377	.next		= NULL,
378	.priority	= 0,
379};
380
381static int __init opal_message_init(void)
382{
383	int ret, i;
384
385	for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
386		ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
387
388	ret = opal_notifier_register(&opal_message_nb);
389	if (ret) {
390		pr_err("%s: Can't register OPAL event notifier (%d)\n",
391		       __func__, ret);
392		return ret;
393	}
394	return 0;
395}
396machine_early_initcall(powernv, opal_message_init);
397
398int opal_get_chars(uint32_t vtermno, char *buf, int count)
399{
400	s64 rc;
401	__be64 evt, len;
402
403	if (!opal.entry)
404		return -ENODEV;
405	opal_poll_events(&evt);
406	if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
407		return 0;
408	len = cpu_to_be64(count);
409	rc = opal_console_read(vtermno, &len, buf);
410	if (rc == OPAL_SUCCESS)
411		return be64_to_cpu(len);
412	return 0;
413}
414
415int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
416{
417	int written = 0;
418	__be64 olen;
419	s64 len, rc;
420	unsigned long flags;
421	__be64 evt;
422
423	if (!opal.entry)
424		return -ENODEV;
425
426	/* We want put_chars to be atomic to avoid mangling of hvsi
427	 * packets. To do that, we first test for room and return
428	 * -EAGAIN if there isn't enough.
429	 *
430	 * Unfortunately, opal_console_write_buffer_space() doesn't
431	 * appear to work on opal v1, so we just assume there is
432	 * enough room and be done with it
433	 */
434	spin_lock_irqsave(&opal_write_lock, flags);
435	if (firmware_has_feature(FW_FEATURE_OPALv2)) {
436		rc = opal_console_write_buffer_space(vtermno, &olen);
437		len = be64_to_cpu(olen);
438		if (rc || len < total_len) {
439			spin_unlock_irqrestore(&opal_write_lock, flags);
440			/* Closed -> drop characters */
441			if (rc)
442				return total_len;
443			opal_poll_events(NULL);
444			return -EAGAIN;
445		}
446	}
447
448	/* We still try to handle partial completions, though they
449	 * should no longer happen.
450	 */
451	rc = OPAL_BUSY;
452	while(total_len > 0 && (rc == OPAL_BUSY ||
453				rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
454		olen = cpu_to_be64(total_len);
455		rc = opal_console_write(vtermno, &olen, data);
456		len = be64_to_cpu(olen);
457
458		/* Closed or other error drop */
459		if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
460		    rc != OPAL_BUSY_EVENT) {
461			written = total_len;
462			break;
463		}
464		if (rc == OPAL_SUCCESS) {
465			total_len -= len;
466			data += len;
467			written += len;
468		}
469		/* This is a bit nasty but we need that for the console to
470		 * flush when there aren't any interrupts. We will clean
471		 * things a bit later to limit that to synchronous path
472		 * such as the kernel console and xmon/udbg
473		 */
474		do
475			opal_poll_events(&evt);
476		while(rc == OPAL_SUCCESS &&
477			(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
478	}
479	spin_unlock_irqrestore(&opal_write_lock, flags);
480	return written;
481}
482
483static int opal_recover_mce(struct pt_regs *regs,
484					struct machine_check_event *evt)
485{
486	int recovered = 0;
487	uint64_t ea = get_mce_fault_addr(evt);
488
489	if (!(regs->msr & MSR_RI)) {
490		/* If MSR_RI isn't set, we cannot recover */
491		recovered = 0;
492	} else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
493		/* Platform corrected itself */
494		recovered = 1;
495	} else if (ea && !is_kernel_addr(ea)) {
496		/*
497		 * Faulting address is not in kernel text. We should be fine.
498		 * We need to find which process uses this address.
499		 * For now, kill the task if we have received exception when
500		 * in userspace.
501		 *
502		 * TODO: Queue up this address for hwpoisioning later.
503		 */
504		if (user_mode(regs) && !is_global_init(current)) {
505			_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
506			recovered = 1;
507		} else
508			recovered = 0;
509	} else if (user_mode(regs) && !is_global_init(current) &&
510		evt->severity == MCE_SEV_ERROR_SYNC) {
511		/*
512		 * If we have received a synchronous error when in userspace
513		 * kill the task.
514		 */
515		_exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
516		recovered = 1;
517	}
518	return recovered;
519}
520
521int opal_machine_check(struct pt_regs *regs)
522{
523	struct machine_check_event evt;
524
525	if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
526		return 0;
527
528	/* Print things out */
529	if (evt.version != MCE_V1) {
530		pr_err("Machine Check Exception, Unknown event version %d !\n",
531		       evt.version);
532		return 0;
533	}
534	machine_check_print_event_info(&evt);
535
536	if (opal_recover_mce(regs, &evt))
537		return 1;
538	return 0;
539}
540
541/* Early hmi handler called in real mode. */
542int opal_hmi_exception_early(struct pt_regs *regs)
543{
544	s64 rc;
545
546	/*
547	 * call opal hmi handler. Pass paca address as token.
548	 * The return value OPAL_SUCCESS is an indication that there is
549	 * an HMI event generated waiting to pull by Linux.
550	 */
551	rc = opal_handle_hmi();
552	if (rc == OPAL_SUCCESS) {
553		local_paca->hmi_event_available = 1;
554		return 1;
555	}
556	return 0;
557}
558
559/* HMI exception handler called in virtual mode during check_irq_replay. */
560int opal_handle_hmi_exception(struct pt_regs *regs)
561{
562	s64 rc;
563	__be64 evt = 0;
564
565	/*
566	 * Check if HMI event is available.
567	 * if Yes, then call opal_poll_events to pull opal messages and
568	 * process them.
569	 */
570	if (!local_paca->hmi_event_available)
571		return 0;
572
573	local_paca->hmi_event_available = 0;
574	rc = opal_poll_events(&evt);
575	if (rc == OPAL_SUCCESS && evt)
576		opal_do_notifier(be64_to_cpu(evt));
577
578	return 1;
579}
580
581static uint64_t find_recovery_address(uint64_t nip)
582{
583	int i;
584
585	for (i = 0; i < mc_recoverable_range_len; i++)
586		if ((nip >= mc_recoverable_range[i].start_addr) &&
587		    (nip < mc_recoverable_range[i].end_addr))
588		    return mc_recoverable_range[i].recover_addr;
589	return 0;
590}
591
592bool opal_mce_check_early_recovery(struct pt_regs *regs)
593{
594	uint64_t recover_addr = 0;
595
596	if (!opal.base || !opal.size)
597		goto out;
598
599	if ((regs->nip >= opal.base) &&
600			(regs->nip <= (opal.base + opal.size)))
601		recover_addr = find_recovery_address(regs->nip);
602
603	/*
604	 * Setup regs->nip to rfi into fixup address.
605	 */
606	if (recover_addr)
607		regs->nip = recover_addr;
608
609out:
610	return !!recover_addr;
611}
612
613static irqreturn_t opal_interrupt(int irq, void *data)
614{
615	__be64 events;
616
617	opal_handle_interrupt(virq_to_hw(irq), &events);
618
619	opal_do_notifier(be64_to_cpu(events));
620
621	return IRQ_HANDLED;
622}
623
624static int opal_sysfs_init(void)
625{
626	opal_kobj = kobject_create_and_add("opal", firmware_kobj);
627	if (!opal_kobj) {
628		pr_warn("kobject_create_and_add opal failed\n");
629		return -ENOMEM;
630	}
631
632	return 0;
633}
634
635static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
636			       struct bin_attribute *bin_attr,
637			       char *buf, loff_t off, size_t count)
638{
639	return memory_read_from_buffer(buf, count, &off, bin_attr->private,
640				       bin_attr->size);
641}
642
643static BIN_ATTR_RO(symbol_map, 0);
644
645static void opal_export_symmap(void)
646{
647	const __be64 *syms;
648	unsigned int size;
649	struct device_node *fw;
650	int rc;
651
652	fw = of_find_node_by_path("/ibm,opal/firmware");
653	if (!fw)
654		return;
655	syms = of_get_property(fw, "symbol-map", &size);
656	if (!syms || size != 2 * sizeof(__be64))
657		return;
658
659	/* Setup attributes */
660	bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
661	bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
662
663	rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
664	if (rc)
665		pr_warn("Error %d creating OPAL symbols file\n", rc);
666}
667
668static void __init opal_dump_region_init(void)
669{
670	void *addr;
671	uint64_t size;
672	int rc;
673
674	if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
675		return;
676
677	/* Register kernel log buffer */
678	addr = log_buf_addr_get();
679	if (addr == NULL)
680		return;
681
682	size = log_buf_len_get();
683	if (size == 0)
684		return;
685
686	rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
687				       __pa(addr), size);
688	/* Don't warn if this is just an older OPAL that doesn't
689	 * know about that call
690	 */
691	if (rc && rc != OPAL_UNSUPPORTED)
692		pr_warn("DUMP: Failed to register kernel log buffer. "
693			"rc = %d\n", rc);
694}
695
696static void opal_flash_init(struct device_node *opal_node)
697{
698	struct device_node *np;
699
700	for_each_child_of_node(opal_node, np)
701		if (of_device_is_compatible(np, "ibm,opal-flash"))
702			of_platform_device_create(np, NULL, NULL);
703}
704
705static void opal_ipmi_init(struct device_node *opal_node)
706{
707	struct device_node *np;
708
709	for_each_child_of_node(opal_node, np)
710		if (of_device_is_compatible(np, "ibm,opal-ipmi"))
711			of_platform_device_create(np, NULL, NULL);
712}
713
714static void opal_i2c_create_devs(void)
715{
716	struct device_node *np;
717
718	for_each_compatible_node(np, NULL, "ibm,opal-i2c")
719		of_platform_device_create(np, NULL, NULL);
720}
721
722static void __init opal_irq_init(struct device_node *dn)
723{
724	const __be32 *irqs;
725	int i, irqlen;
726
727	/* Get interrupt property */
728	irqs = of_get_property(opal_node, "opal-interrupts", &irqlen);
729	opal_irq_count = irqs ? (irqlen / 4) : 0;
730	pr_debug("Found %d interrupts reserved for OPAL\n", opal_irq_count);
731	if (!opal_irq_count)
732		return;
733
734	/* Install interrupt handlers */
735	opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL);
736	for (i = 0; irqs && i < opal_irq_count; i++, irqs++) {
737		unsigned int irq, virq;
738		int rc;
739
740		/* Get hardware and virtual IRQ */
741		irq = be32_to_cpup(irqs);
742		virq = irq_create_mapping(NULL, irq);
743		if (virq == NO_IRQ) {
744			pr_warn("Failed to map irq 0x%x\n", irq);
745			continue;
746		}
747
748		/* Install interrupt handler */
749		rc = request_irq(virq, opal_interrupt, 0, "opal", NULL);
750		if (rc) {
751			irq_dispose_mapping(virq);
752			pr_warn("Error %d requesting irq %d (0x%x)\n",
753				 rc, virq, irq);
754			continue;
755		}
756
757		/* Cache IRQ */
758		opal_irqs[i] = virq;
759	}
760}
761
762static int kopald(void *unused)
763{
764	set_freezable();
765	do {
766		try_to_freeze();
767		opal_poll_events(NULL);
768		msleep_interruptible(opal_heartbeat);
769	} while (!kthread_should_stop());
770
771	return 0;
772}
773
774static void opal_init_heartbeat(void)
775{
776	/* Old firwmware, we assume the HVC heartbeat is sufficient */
777	if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
778				 &opal_heartbeat) != 0)
779		opal_heartbeat = 0;
780
781	if (opal_heartbeat)
782		kthread_run(kopald, NULL, "kopald");
783}
784
785static int __init opal_init(void)
786{
787	struct device_node *np, *consoles;
788	int rc;
789
790	opal_node = of_find_node_by_path("/ibm,opal");
791	if (!opal_node) {
792		pr_warn("Device node not found\n");
793		return -ENODEV;
794	}
795
796	/* Register OPAL consoles if any ports */
797	if (firmware_has_feature(FW_FEATURE_OPALv2))
798		consoles = of_find_node_by_path("/ibm,opal/consoles");
799	else
800		consoles = of_node_get(opal_node);
801	if (consoles) {
802		for_each_child_of_node(consoles, np) {
803			if (strcmp(np->name, "serial"))
804				continue;
805			of_platform_device_create(np, NULL, NULL);
806		}
807		of_node_put(consoles);
808	}
809
810	/* Create i2c platform devices */
811	opal_i2c_create_devs();
812
813	/* Setup a heatbeat thread if requested by OPAL */
814	opal_init_heartbeat();
815
816	/* Find all OPAL interrupts and request them */
817	opal_irq_init(opal_node);
818
819	/* Create "opal" kobject under /sys/firmware */
820	rc = opal_sysfs_init();
821	if (rc == 0) {
822		/* Export symbol map to userspace */
823		opal_export_symmap();
824		/* Setup dump region interface */
825		opal_dump_region_init();
826		/* Setup error log interface */
827		rc = opal_elog_init();
828		/* Setup code update interface */
829		opal_flash_update_init();
830		/* Setup platform dump extract interface */
831		opal_platform_dump_init();
832		/* Setup system parameters interface */
833		opal_sys_param_init();
834		/* Setup message log interface. */
835		opal_msglog_init();
836	}
837
838	/* Initialize OPAL IPMI backend */
839	opal_ipmi_init(opal_node);
840
841	opal_flash_init(opal_node);
842
843	return 0;
844}
845machine_subsys_initcall(powernv, opal_init);
846
847void opal_shutdown(void)
848{
849	unsigned int i;
850	long rc = OPAL_BUSY;
851
852	/* First free interrupts, which will also mask them */
853	for (i = 0; i < opal_irq_count; i++) {
854		if (opal_irqs[i])
855			free_irq(opal_irqs[i], NULL);
856		opal_irqs[i] = 0;
857	}
858
859	/*
860	 * Then sync with OPAL which ensure anything that can
861	 * potentially write to our memory has completed such
862	 * as an ongoing dump retrieval
863	 */
864	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
865		rc = opal_sync_host_reboot();
866		if (rc == OPAL_BUSY)
867			opal_poll_events(NULL);
868		else
869			mdelay(10);
870	}
871
872	/* Unregister memory dump region */
873	if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
874		opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
875}
876
877/* Export this so that test modules can use it */
878EXPORT_SYMBOL_GPL(opal_invalid_call);
879EXPORT_SYMBOL_GPL(opal_ipmi_send);
880EXPORT_SYMBOL_GPL(opal_ipmi_recv);
881EXPORT_SYMBOL_GPL(opal_flash_read);
882EXPORT_SYMBOL_GPL(opal_flash_write);
883EXPORT_SYMBOL_GPL(opal_flash_erase);
884
885/* Convert a region of vmalloc memory to an opal sg list */
886struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
887					     unsigned long vmalloc_size)
888{
889	struct opal_sg_list *sg, *first = NULL;
890	unsigned long i = 0;
891
892	sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
893	if (!sg)
894		goto nomem;
895
896	first = sg;
897
898	while (vmalloc_size > 0) {
899		uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
900		uint64_t length = min(vmalloc_size, PAGE_SIZE);
901
902		sg->entry[i].data = cpu_to_be64(data);
903		sg->entry[i].length = cpu_to_be64(length);
904		i++;
905
906		if (i >= SG_ENTRIES_PER_NODE) {
907			struct opal_sg_list *next;
908
909			next = kzalloc(PAGE_SIZE, GFP_KERNEL);
910			if (!next)
911				goto nomem;
912
913			sg->length = cpu_to_be64(
914					i * sizeof(struct opal_sg_entry) + 16);
915			i = 0;
916			sg->next = cpu_to_be64(__pa(next));
917			sg = next;
918		}
919
920		vmalloc_addr += length;
921		vmalloc_size -= length;
922	}
923
924	sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
925
926	return first;
927
928nomem:
929	pr_err("%s : Failed to allocate memory\n", __func__);
930	opal_free_sg_list(first);
931	return NULL;
932}
933
934void opal_free_sg_list(struct opal_sg_list *sg)
935{
936	while (sg) {
937		uint64_t next = be64_to_cpu(sg->next);
938
939		kfree(sg);
940
941		if (next)
942			sg = __va(next);
943		else
944			sg = NULL;
945	}
946}
947
948int opal_error_code(int rc)
949{
950	switch (rc) {
951	case OPAL_SUCCESS:		return 0;
952
953	case OPAL_PARAMETER:		return -EINVAL;
954	case OPAL_ASYNC_COMPLETION:	return -EINPROGRESS;
955	case OPAL_BUSY_EVENT:		return -EBUSY;
956	case OPAL_NO_MEM:		return -ENOMEM;
957
958	case OPAL_UNSUPPORTED:		return -EIO;
959	case OPAL_HARDWARE:		return -EIO;
960	case OPAL_INTERNAL_ERROR:	return -EIO;
961	default:
962		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
963		return -EIO;
964	}
965}
966
967EXPORT_SYMBOL_GPL(opal_poll_events);
968EXPORT_SYMBOL_GPL(opal_rtc_read);
969EXPORT_SYMBOL_GPL(opal_rtc_write);
970EXPORT_SYMBOL_GPL(opal_tpo_read);
971EXPORT_SYMBOL_GPL(opal_tpo_write);
972EXPORT_SYMBOL_GPL(opal_i2c_request);
973