Loading...
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#define OPAL_MSG_QUEUE_MAX 16
39
40struct opal_msg_node {
41 struct list_head list;
42 struct opal_msg msg;
43};
44
45static DEFINE_SPINLOCK(msg_list_lock);
46static LIST_HEAD(msg_list);
47
48/* /sys/firmware/opal */
49struct kobject *opal_kobj;
50
51struct opal {
52 u64 base;
53 u64 entry;
54 u64 size;
55} opal;
56
57struct mcheck_recoverable_range {
58 u64 start_addr;
59 u64 end_addr;
60 u64 recover_addr;
61};
62
63static int msg_list_size;
64
65static struct mcheck_recoverable_range *mc_recoverable_range;
66static int mc_recoverable_range_len;
67
68struct device_node *opal_node;
69static DEFINE_SPINLOCK(opal_write_lock);
70static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
71static uint32_t opal_heartbeat;
72static struct task_struct *kopald_tsk;
73static struct opal_msg *opal_msg;
74static u32 opal_msg_size __ro_after_init;
75
76void __init opal_configure_cores(void)
77{
78 u64 reinit_flags = 0;
79
80 /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
81 *
82 * It will preserve non volatile GPRs and HSPRG0/1. It will
83 * also restore HIDs and other SPRs to their original value
84 * but it might clobber a bunch.
85 */
86#ifdef __BIG_ENDIAN__
87 reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
88#else
89 reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
90#endif
91
92 /*
93 * POWER9 always support running hash:
94 * ie. Host hash supports hash guests
95 * Host radix supports hash/radix guests
96 */
97 if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
98 reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
99 if (early_radix_enabled())
100 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
101 }
102
103 opal_reinit_cpus(reinit_flags);
104
105 /* Restore some bits */
106 if (cur_cpu_spec->cpu_restore)
107 cur_cpu_spec->cpu_restore();
108}
109
110int __init early_init_dt_scan_opal(unsigned long node,
111 const char *uname, int depth, void *data)
112{
113 const void *basep, *entryp, *sizep;
114 int basesz, entrysz, runtimesz;
115
116 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
117 return 0;
118
119 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
120 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
121 sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
122
123 if (!basep || !entryp || !sizep)
124 return 1;
125
126 opal.base = of_read_number(basep, basesz/4);
127 opal.entry = of_read_number(entryp, entrysz/4);
128 opal.size = of_read_number(sizep, runtimesz/4);
129
130 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%d)\n",
131 opal.base, basep, basesz);
132 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
133 opal.entry, entryp, entrysz);
134 pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
135 opal.size, sizep, runtimesz);
136
137 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
138 powerpc_firmware_features |= FW_FEATURE_OPAL;
139 pr_debug("OPAL detected !\n");
140 } else {
141 panic("OPAL != V3 detected, no longer supported.\n");
142 }
143
144 return 1;
145}
146
147int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
148 const char *uname, int depth, void *data)
149{
150 int i, psize, size;
151 const __be32 *prop;
152
153 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
154 return 0;
155
156 prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
157
158 if (!prop)
159 return 1;
160
161 pr_debug("Found machine check recoverable ranges.\n");
162
163 /*
164 * Calculate number of available entries.
165 *
166 * Each recoverable address range entry is (start address, len,
167 * recovery address), 2 cells each for start and recovery address,
168 * 1 cell for len, totalling 5 cells per entry.
169 */
170 mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
171
172 /* Sanity check */
173 if (!mc_recoverable_range_len)
174 return 1;
175
176 /* Size required to hold all the entries. */
177 size = mc_recoverable_range_len *
178 sizeof(struct mcheck_recoverable_range);
179
180 /*
181 * Allocate a buffer to hold the MC recoverable ranges.
182 */
183 mc_recoverable_range = memblock_alloc(size, __alignof__(u64));
184 if (!mc_recoverable_range)
185 panic("%s: Failed to allocate %u bytes align=0x%lx\n",
186 __func__, size, __alignof__(u64));
187
188 for (i = 0; i < mc_recoverable_range_len; i++) {
189 mc_recoverable_range[i].start_addr =
190 of_read_number(prop + (i * 5) + 0, 2);
191 mc_recoverable_range[i].end_addr =
192 mc_recoverable_range[i].start_addr +
193 of_read_number(prop + (i * 5) + 2, 1);
194 mc_recoverable_range[i].recover_addr =
195 of_read_number(prop + (i * 5) + 3, 2);
196
197 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
198 mc_recoverable_range[i].start_addr,
199 mc_recoverable_range[i].end_addr,
200 mc_recoverable_range[i].recover_addr);
201 }
202 return 1;
203}
204
205static int __init opal_register_exception_handlers(void)
206{
207#ifdef __BIG_ENDIAN__
208 u64 glue;
209
210 if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
211 return -ENODEV;
212
213 /* Hookup some exception handlers except machine check. We use the
214 * fwnmi area at 0x7000 to provide the glue space to OPAL
215 */
216 glue = 0x7000;
217
218 /*
219 * Only ancient OPAL firmware requires this.
220 * Specifically, firmware from FW810.00 (released June 2014)
221 * through FW810.20 (Released October 2014).
222 *
223 * Check if we are running on newer (post Oct 2014) firmware that
224 * exports the OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to
225 * patch the HMI interrupt and we catch it directly in Linux.
226 *
227 * For older firmware (i.e < FW810.20), we fallback to old behavior and
228 * let OPAL patch the HMI vector and handle it inside OPAL firmware.
229 *
230 * For newer firmware we catch/handle the HMI directly in Linux.
231 */
232 if (!opal_check_token(OPAL_HANDLE_HMI)) {
233 pr_info("Old firmware detected, OPAL handles HMIs.\n");
234 opal_register_exception_handler(
235 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
236 0, glue);
237 glue += 128;
238 }
239
240 /*
241 * Only applicable to ancient firmware, all modern
242 * (post March 2015/skiboot 5.0) firmware will just return
243 * OPAL_UNSUPPORTED.
244 */
245 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
246#endif
247
248 return 0;
249}
250machine_early_initcall(powernv, opal_register_exception_handlers);
251
252static void queue_replay_msg(void *msg)
253{
254 struct opal_msg_node *msg_node;
255
256 if (msg_list_size < OPAL_MSG_QUEUE_MAX) {
257 msg_node = kzalloc(sizeof(*msg_node), GFP_ATOMIC);
258 if (msg_node) {
259 INIT_LIST_HEAD(&msg_node->list);
260 memcpy(&msg_node->msg, msg, sizeof(struct opal_msg));
261 list_add_tail(&msg_node->list, &msg_list);
262 msg_list_size++;
263 } else
264 pr_warn_once("message queue no memory\n");
265
266 if (msg_list_size >= OPAL_MSG_QUEUE_MAX)
267 pr_warn_once("message queue full\n");
268 }
269}
270
271static void dequeue_replay_msg(enum opal_msg_type msg_type)
272{
273 struct opal_msg_node *msg_node, *tmp;
274
275 list_for_each_entry_safe(msg_node, tmp, &msg_list, list) {
276 if (be32_to_cpu(msg_node->msg.msg_type) != msg_type)
277 continue;
278
279 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
280 msg_type,
281 &msg_node->msg);
282
283 list_del(&msg_node->list);
284 kfree(msg_node);
285 msg_list_size--;
286 }
287}
288
289/*
290 * Opal message notifier based on message type. Allow subscribers to get
291 * notified for specific messgae type.
292 */
293int opal_message_notifier_register(enum opal_msg_type msg_type,
294 struct notifier_block *nb)
295{
296 int ret;
297 unsigned long flags;
298
299 if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
300 pr_warn("%s: Invalid arguments, msg_type:%d\n",
301 __func__, msg_type);
302 return -EINVAL;
303 }
304
305 spin_lock_irqsave(&msg_list_lock, flags);
306 ret = atomic_notifier_chain_register(
307 &opal_msg_notifier_head[msg_type], nb);
308
309 /*
310 * If the registration succeeded, replay any queued messages that came
311 * in prior to the notifier chain registration. msg_list_lock held here
312 * to ensure they're delivered prior to any subsequent messages.
313 */
314 if (ret == 0)
315 dequeue_replay_msg(msg_type);
316
317 spin_unlock_irqrestore(&msg_list_lock, flags);
318
319 return ret;
320}
321EXPORT_SYMBOL_GPL(opal_message_notifier_register);
322
323int opal_message_notifier_unregister(enum opal_msg_type msg_type,
324 struct notifier_block *nb)
325{
326 return atomic_notifier_chain_unregister(
327 &opal_msg_notifier_head[msg_type], nb);
328}
329EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
330
331static void opal_message_do_notify(uint32_t msg_type, void *msg)
332{
333 unsigned long flags;
334 bool queued = false;
335
336 spin_lock_irqsave(&msg_list_lock, flags);
337 if (opal_msg_notifier_head[msg_type].head == NULL) {
338 /*
339 * Queue up the msg since no notifiers have registered
340 * yet for this msg_type.
341 */
342 queue_replay_msg(msg);
343 queued = true;
344 }
345 spin_unlock_irqrestore(&msg_list_lock, flags);
346
347 if (queued)
348 return;
349
350 /* notify subscribers */
351 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
352 msg_type, msg);
353}
354
355static void opal_handle_message(void)
356{
357 s64 ret;
358 u32 type;
359
360 ret = opal_get_msg(__pa(opal_msg), opal_msg_size);
361 /* No opal message pending. */
362 if (ret == OPAL_RESOURCE)
363 return;
364
365 /* check for errors. */
366 if (ret) {
367 pr_warn("%s: Failed to retrieve opal message, err=%lld\n",
368 __func__, ret);
369 return;
370 }
371
372 type = be32_to_cpu(opal_msg->msg_type);
373
374 /* Sanity check */
375 if (type >= OPAL_MSG_TYPE_MAX) {
376 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
377 return;
378 }
379 opal_message_do_notify(type, (void *)opal_msg);
380}
381
382static irqreturn_t opal_message_notify(int irq, void *data)
383{
384 opal_handle_message();
385 return IRQ_HANDLED;
386}
387
388static int __init opal_message_init(struct device_node *opal_node)
389{
390 int ret, i, irq;
391
392 ret = of_property_read_u32(opal_node, "opal-msg-size", &opal_msg_size);
393 if (ret) {
394 pr_notice("Failed to read opal-msg-size property\n");
395 opal_msg_size = sizeof(struct opal_msg);
396 }
397
398 opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
399 if (!opal_msg) {
400 opal_msg_size = sizeof(struct opal_msg);
401 /* Try to allocate fixed message size */
402 opal_msg = kmalloc(opal_msg_size, GFP_KERNEL);
403 BUG_ON(opal_msg == NULL);
404 }
405
406 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
407 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
408
409 irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
410 if (!irq) {
411 pr_err("%s: Can't register OPAL event irq (%d)\n",
412 __func__, irq);
413 return irq;
414 }
415
416 ret = request_irq(irq, opal_message_notify,
417 IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
418 if (ret) {
419 pr_err("%s: Can't request OPAL event irq (%d)\n",
420 __func__, ret);
421 return ret;
422 }
423
424 return 0;
425}
426
427ssize_t opal_get_chars(uint32_t vtermno, u8 *buf, size_t count)
428{
429 s64 rc;
430 __be64 evt, len;
431
432 if (!opal.entry)
433 return -ENODEV;
434 opal_poll_events(&evt);
435 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
436 return 0;
437 len = cpu_to_be64(count);
438 rc = opal_console_read(vtermno, &len, buf);
439 if (rc == OPAL_SUCCESS)
440 return be64_to_cpu(len);
441 return 0;
442}
443
444static ssize_t __opal_put_chars(uint32_t vtermno, const u8 *data,
445 size_t total_len, bool atomic)
446{
447 unsigned long flags = 0 /* shut up gcc */;
448 ssize_t written;
449 __be64 olen;
450 s64 rc;
451
452 if (!opal.entry)
453 return -ENODEV;
454
455 if (atomic)
456 spin_lock_irqsave(&opal_write_lock, flags);
457 rc = opal_console_write_buffer_space(vtermno, &olen);
458 if (rc || be64_to_cpu(olen) < total_len) {
459 /* Closed -> drop characters */
460 if (rc)
461 written = total_len;
462 else
463 written = -EAGAIN;
464 goto out;
465 }
466
467 /* Should not get a partial write here because space is available. */
468 olen = cpu_to_be64(total_len);
469 rc = opal_console_write(vtermno, &olen, data);
470 if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
471 if (rc == OPAL_BUSY_EVENT)
472 opal_poll_events(NULL);
473 written = -EAGAIN;
474 goto out;
475 }
476
477 /* Closed or other error drop */
478 if (rc != OPAL_SUCCESS) {
479 written = opal_error_code(rc);
480 goto out;
481 }
482
483 written = be64_to_cpu(olen);
484 if (written < total_len) {
485 if (atomic) {
486 /* Should not happen */
487 pr_warn("atomic console write returned partial "
488 "len=%zu written=%zd\n", total_len, written);
489 }
490 if (!written)
491 written = -EAGAIN;
492 }
493
494out:
495 if (atomic)
496 spin_unlock_irqrestore(&opal_write_lock, flags);
497
498 return written;
499}
500
501ssize_t opal_put_chars(uint32_t vtermno, const u8 *data, size_t total_len)
502{
503 return __opal_put_chars(vtermno, data, total_len, false);
504}
505
506/*
507 * opal_put_chars_atomic will not perform partial-writes. Data will be
508 * atomically written to the terminal or not at all. This is not strictly
509 * true at the moment because console space can race with OPAL's console
510 * writes.
511 */
512ssize_t opal_put_chars_atomic(uint32_t vtermno, const u8 *data,
513 size_t total_len)
514{
515 return __opal_put_chars(vtermno, data, total_len, true);
516}
517
518static s64 __opal_flush_console(uint32_t vtermno)
519{
520 s64 rc;
521
522 if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
523 __be64 evt;
524
525 /*
526 * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
527 * the console can still be flushed by calling the polling
528 * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
529 */
530 WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
531
532 opal_poll_events(&evt);
533 if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
534 return OPAL_SUCCESS;
535 return OPAL_BUSY;
536
537 } else {
538 rc = opal_console_flush(vtermno);
539 if (rc == OPAL_BUSY_EVENT) {
540 opal_poll_events(NULL);
541 rc = OPAL_BUSY;
542 }
543 return rc;
544 }
545
546}
547
548/*
549 * opal_flush_console spins until the console is flushed
550 */
551int opal_flush_console(uint32_t vtermno)
552{
553 for (;;) {
554 s64 rc = __opal_flush_console(vtermno);
555
556 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
557 mdelay(1);
558 continue;
559 }
560
561 return opal_error_code(rc);
562 }
563}
564
565/*
566 * opal_flush_chars is an hvc interface that sleeps until the console is
567 * flushed if wait, otherwise it will return -EBUSY if the console has data,
568 * -EAGAIN if it has data and some of it was flushed.
569 */
570int opal_flush_chars(uint32_t vtermno, bool wait)
571{
572 for (;;) {
573 s64 rc = __opal_flush_console(vtermno);
574
575 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
576 if (wait) {
577 msleep(OPAL_BUSY_DELAY_MS);
578 continue;
579 }
580 if (rc == OPAL_PARTIAL)
581 return -EAGAIN;
582 }
583
584 return opal_error_code(rc);
585 }
586}
587
588static int opal_recover_mce(struct pt_regs *regs,
589 struct machine_check_event *evt)
590{
591 int recovered = 0;
592
593 if (regs_is_unrecoverable(regs)) {
594 /* If MSR_RI isn't set, we cannot recover */
595 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
596 recovered = 0;
597 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
598 /* Platform corrected itself */
599 recovered = 1;
600 } else if (evt->severity == MCE_SEV_FATAL) {
601 /* Fatal machine check */
602 pr_err("Machine check interrupt is fatal\n");
603 recovered = 0;
604 }
605
606 if (!recovered && evt->sync_error) {
607 /*
608 * Try to kill processes if we get a synchronous machine check
609 * (e.g., one caused by execution of this instruction). This
610 * will devolve into a panic if we try to kill init or are in
611 * an interrupt etc.
612 *
613 * TODO: Queue up this address for hwpoisioning later.
614 * TODO: This is not quite right for d-side machine
615 * checks ->nip is not necessarily the important
616 * address.
617 */
618 if ((user_mode(regs))) {
619 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
620 recovered = 1;
621 } else if (die_will_crash()) {
622 /*
623 * die() would kill the kernel, so better to go via
624 * the platform reboot code that will log the
625 * machine check.
626 */
627 recovered = 0;
628 } else {
629 die_mce("Machine check", regs, SIGBUS);
630 recovered = 1;
631 }
632 }
633
634 return recovered;
635}
636
637void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
638{
639 panic_flush_kmsg_start();
640
641 pr_emerg("Hardware platform error: %s\n", msg);
642 if (regs)
643 show_regs(regs);
644 smp_send_stop();
645
646 panic_flush_kmsg_end();
647
648 /*
649 * Don't bother to shut things down because this will
650 * xstop the system.
651 */
652 if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
653 == OPAL_UNSUPPORTED) {
654 pr_emerg("Reboot type %d not supported for %s\n",
655 OPAL_REBOOT_PLATFORM_ERROR, msg);
656 }
657
658 /*
659 * We reached here. There can be three possibilities:
660 * 1. We are running on a firmware level that do not support
661 * opal_cec_reboot2()
662 * 2. We are running on a firmware level that do not support
663 * OPAL_REBOOT_PLATFORM_ERROR reboot type.
664 * 3. We are running on FSP based system that does not need
665 * opal to trigger checkstop explicitly for error analysis.
666 * The FSP PRD component would have already got notified
667 * about this error through other channels.
668 * 4. We are running on a newer skiboot that by default does
669 * not cause a checkstop, drops us back to the kernel to
670 * extract context and state at the time of the error.
671 */
672
673 panic(msg);
674}
675
676int opal_machine_check(struct pt_regs *regs)
677{
678 struct machine_check_event evt;
679
680 if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
681 return 0;
682
683 /* Print things out */
684 if (evt.version != MCE_V1) {
685 pr_err("Machine Check Exception, Unknown event version %d !\n",
686 evt.version);
687 return 0;
688 }
689 machine_check_print_event_info(&evt, user_mode(regs), false);
690
691 if (opal_recover_mce(regs, &evt))
692 return 1;
693
694 pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
695}
696
697/* Early hmi handler called in real mode. */
698int opal_hmi_exception_early(struct pt_regs *regs)
699{
700 s64 rc;
701
702 /*
703 * call opal hmi handler. Pass paca address as token.
704 * The return value OPAL_SUCCESS is an indication that there is
705 * an HMI event generated waiting to pull by Linux.
706 */
707 rc = opal_handle_hmi();
708 if (rc == OPAL_SUCCESS) {
709 local_paca->hmi_event_available = 1;
710 return 1;
711 }
712 return 0;
713}
714
715int opal_hmi_exception_early2(struct pt_regs *regs)
716{
717 s64 rc;
718 __be64 out_flags;
719
720 /*
721 * call opal hmi handler.
722 * Check 64-bit flag mask to find out if an event was generated,
723 * and whether TB is still valid or not etc.
724 */
725 rc = opal_handle_hmi2(&out_flags);
726 if (rc != OPAL_SUCCESS)
727 return 0;
728
729 if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_NEW_EVENT)
730 local_paca->hmi_event_available = 1;
731 if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_TOD_TB_FAIL)
732 tb_invalid = true;
733 return 1;
734}
735
736/* HMI exception handler called in virtual mode when irqs are next enabled. */
737int opal_handle_hmi_exception(struct pt_regs *regs)
738{
739 /*
740 * Check if HMI event is available.
741 * if Yes, then wake kopald to process them.
742 */
743 if (!local_paca->hmi_event_available)
744 return 0;
745
746 local_paca->hmi_event_available = 0;
747 opal_wake_poller();
748
749 return 1;
750}
751
752static uint64_t find_recovery_address(uint64_t nip)
753{
754 int i;
755
756 for (i = 0; i < mc_recoverable_range_len; i++)
757 if ((nip >= mc_recoverable_range[i].start_addr) &&
758 (nip < mc_recoverable_range[i].end_addr))
759 return mc_recoverable_range[i].recover_addr;
760 return 0;
761}
762
763bool opal_mce_check_early_recovery(struct pt_regs *regs)
764{
765 uint64_t recover_addr = 0;
766
767 if (!opal.base || !opal.size)
768 goto out;
769
770 if ((regs->nip >= opal.base) &&
771 (regs->nip < (opal.base + opal.size)))
772 recover_addr = find_recovery_address(regs->nip);
773
774 /*
775 * Setup regs->nip to rfi into fixup address.
776 */
777 if (recover_addr)
778 regs_set_return_ip(regs, recover_addr);
779
780out:
781 return !!recover_addr;
782}
783
784static int __init opal_sysfs_init(void)
785{
786 opal_kobj = kobject_create_and_add("opal", firmware_kobj);
787 if (!opal_kobj) {
788 pr_warn("kobject_create_and_add opal failed\n");
789 return -ENOMEM;
790 }
791
792 return 0;
793}
794
795static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
796 struct bin_attribute *bin_attr, char *buf,
797 loff_t off, size_t count)
798{
799 return memory_read_from_buffer(buf, count, &off, bin_attr->private,
800 bin_attr->size);
801}
802
803static int opal_add_one_export(struct kobject *parent, const char *export_name,
804 struct device_node *np, const char *prop_name)
805{
806 struct bin_attribute *attr = NULL;
807 const char *name = NULL;
808 u64 vals[2];
809 int rc;
810
811 rc = of_property_read_u64_array(np, prop_name, &vals[0], 2);
812 if (rc)
813 goto out;
814
815 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
816 if (!attr) {
817 rc = -ENOMEM;
818 goto out;
819 }
820 name = kstrdup(export_name, GFP_KERNEL);
821 if (!name) {
822 rc = -ENOMEM;
823 goto out;
824 }
825
826 sysfs_bin_attr_init(attr);
827 attr->attr.name = name;
828 attr->attr.mode = 0400;
829 attr->read = export_attr_read;
830 attr->private = __va(vals[0]);
831 attr->size = vals[1];
832
833 rc = sysfs_create_bin_file(parent, attr);
834out:
835 if (rc) {
836 kfree(name);
837 kfree(attr);
838 }
839
840 return rc;
841}
842
843static void opal_add_exported_attrs(struct device_node *np,
844 struct kobject *kobj)
845{
846 struct device_node *child;
847 struct property *prop;
848
849 for_each_property_of_node(np, prop) {
850 int rc;
851
852 if (!strcmp(prop->name, "name") ||
853 !strcmp(prop->name, "phandle"))
854 continue;
855
856 rc = opal_add_one_export(kobj, prop->name, np, prop->name);
857 if (rc) {
858 pr_warn("Unable to add export %pOF/%s, rc = %d!\n",
859 np, prop->name, rc);
860 }
861 }
862
863 for_each_child_of_node(np, child) {
864 struct kobject *child_kobj;
865
866 child_kobj = kobject_create_and_add(child->name, kobj);
867 if (!child_kobj) {
868 pr_err("Unable to create export dir for %pOF\n", child);
869 continue;
870 }
871
872 opal_add_exported_attrs(child, child_kobj);
873 }
874}
875
876/*
877 * opal_export_attrs: creates a sysfs node for each property listed in
878 * the device-tree under /ibm,opal/firmware/exports/
879 * All new sysfs nodes are created under /opal/exports/.
880 * This allows for reserved memory regions (e.g. HDAT) to be read.
881 * The new sysfs nodes are only readable by root.
882 */
883static void opal_export_attrs(void)
884{
885 struct device_node *np;
886 struct kobject *kobj;
887 int rc;
888
889 np = of_find_node_by_path("/ibm,opal/firmware/exports");
890 if (!np)
891 return;
892
893 /* Create new 'exports' directory - /sys/firmware/opal/exports */
894 kobj = kobject_create_and_add("exports", opal_kobj);
895 if (!kobj) {
896 pr_warn("kobject_create_and_add() of exports failed\n");
897 of_node_put(np);
898 return;
899 }
900
901 opal_add_exported_attrs(np, kobj);
902
903 /*
904 * NB: symbol_map existed before the generic export interface so it
905 * lives under the top level opal_kobj.
906 */
907 rc = opal_add_one_export(opal_kobj, "symbol_map",
908 np->parent, "symbol-map");
909 if (rc)
910 pr_warn("Error %d creating OPAL symbols file\n", rc);
911
912 of_node_put(np);
913}
914
915static void __init opal_dump_region_init(void)
916{
917 void *addr;
918 uint64_t size;
919 int rc;
920
921 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
922 return;
923
924 /* Register kernel log buffer */
925 addr = log_buf_addr_get();
926 if (addr == NULL)
927 return;
928
929 size = log_buf_len_get();
930 if (size == 0)
931 return;
932
933 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
934 __pa(addr), size);
935 /* Don't warn if this is just an older OPAL that doesn't
936 * know about that call
937 */
938 if (rc && rc != OPAL_UNSUPPORTED)
939 pr_warn("DUMP: Failed to register kernel log buffer. "
940 "rc = %d\n", rc);
941}
942
943static void __init opal_pdev_init(const char *compatible)
944{
945 struct device_node *np;
946
947 for_each_compatible_node(np, NULL, compatible)
948 of_platform_device_create(np, NULL, NULL);
949}
950
951static void __init opal_imc_init_dev(void)
952{
953 struct device_node *np;
954
955 np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
956 if (np)
957 of_platform_device_create(np, NULL, NULL);
958
959 of_node_put(np);
960}
961
962static int kopald(void *unused)
963{
964 unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
965
966 set_freezable();
967 do {
968 try_to_freeze();
969
970 opal_handle_events();
971
972 set_current_state(TASK_INTERRUPTIBLE);
973 if (opal_have_pending_events())
974 __set_current_state(TASK_RUNNING);
975 else
976 schedule_timeout(timeout);
977
978 } while (!kthread_should_stop());
979
980 return 0;
981}
982
983void opal_wake_poller(void)
984{
985 if (kopald_tsk)
986 wake_up_process(kopald_tsk);
987}
988
989static void __init opal_init_heartbeat(void)
990{
991 /* Old firwmware, we assume the HVC heartbeat is sufficient */
992 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
993 &opal_heartbeat) != 0)
994 opal_heartbeat = 0;
995
996 if (opal_heartbeat)
997 kopald_tsk = kthread_run(kopald, NULL, "kopald");
998}
999
1000static int __init opal_init(void)
1001{
1002 struct device_node *np, *consoles, *leds;
1003 int rc;
1004
1005 opal_node = of_find_node_by_path("/ibm,opal");
1006 if (!opal_node) {
1007 pr_warn("Device node not found\n");
1008 return -ENODEV;
1009 }
1010
1011 /* Register OPAL consoles if any ports */
1012 consoles = of_find_node_by_path("/ibm,opal/consoles");
1013 if (consoles) {
1014 for_each_child_of_node(consoles, np) {
1015 if (!of_node_name_eq(np, "serial"))
1016 continue;
1017 of_platform_device_create(np, NULL, NULL);
1018 }
1019 of_node_put(consoles);
1020 }
1021
1022 /* Initialise OPAL messaging system */
1023 opal_message_init(opal_node);
1024
1025 /* Initialise OPAL asynchronous completion interface */
1026 opal_async_comp_init();
1027
1028 /* Initialise OPAL sensor interface */
1029 opal_sensor_init();
1030
1031 /* Initialise OPAL hypervisor maintainence interrupt handling */
1032 opal_hmi_handler_init();
1033
1034 /* Create i2c platform devices */
1035 opal_pdev_init("ibm,opal-i2c");
1036
1037 /* Handle non-volatile memory devices */
1038 opal_pdev_init("pmem-region");
1039
1040 /* Setup a heatbeat thread if requested by OPAL */
1041 opal_init_heartbeat();
1042
1043 /* Detect In-Memory Collection counters and create devices*/
1044 opal_imc_init_dev();
1045
1046 /* Create leds platform devices */
1047 leds = of_find_node_by_path("/ibm,opal/leds");
1048 if (leds) {
1049 of_platform_device_create(leds, "opal_leds", NULL);
1050 of_node_put(leds);
1051 }
1052
1053 /* Initialise OPAL message log interface */
1054 opal_msglog_init();
1055
1056 /* Create "opal" kobject under /sys/firmware */
1057 rc = opal_sysfs_init();
1058 if (rc == 0) {
1059 /* Setup dump region interface */
1060 opal_dump_region_init();
1061 /* Setup error log interface */
1062 rc = opal_elog_init();
1063 /* Setup code update interface */
1064 opal_flash_update_init();
1065 /* Setup platform dump extract interface */
1066 opal_platform_dump_init();
1067 /* Setup system parameters interface */
1068 opal_sys_param_init();
1069 /* Setup message log sysfs interface. */
1070 opal_msglog_sysfs_init();
1071 /* Add all export properties*/
1072 opal_export_attrs();
1073 }
1074
1075 /* Initialize platform devices: IPMI backend, PRD & flash interface */
1076 opal_pdev_init("ibm,opal-ipmi");
1077 opal_pdev_init("ibm,opal-flash");
1078 opal_pdev_init("ibm,opal-prd");
1079
1080 /* Initialise platform device: oppanel interface */
1081 opal_pdev_init("ibm,opal-oppanel");
1082
1083 /* Initialise OPAL kmsg dumper for flushing console on panic */
1084 opal_kmsg_init();
1085
1086 /* Initialise OPAL powercap interface */
1087 opal_powercap_init();
1088
1089 /* Initialise OPAL Power-Shifting-Ratio interface */
1090 opal_psr_init();
1091
1092 /* Initialise OPAL sensor groups */
1093 opal_sensor_groups_init();
1094
1095 /* Initialise OPAL Power control interface */
1096 opal_power_control_init();
1097
1098 /* Initialize OPAL secure variables */
1099 opal_pdev_init("ibm,secvar-backend");
1100
1101 return 0;
1102}
1103machine_subsys_initcall(powernv, opal_init);
1104
1105void opal_shutdown(void)
1106{
1107 long rc = OPAL_BUSY;
1108
1109 opal_event_shutdown();
1110
1111 /*
1112 * Then sync with OPAL which ensure anything that can
1113 * potentially write to our memory has completed such
1114 * as an ongoing dump retrieval
1115 */
1116 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
1117 rc = opal_sync_host_reboot();
1118 if (rc == OPAL_BUSY)
1119 opal_poll_events(NULL);
1120 else
1121 mdelay(10);
1122 }
1123
1124 /* Unregister memory dump region */
1125 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
1126 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
1127}
1128
1129/* Export this so that test modules can use it */
1130EXPORT_SYMBOL_GPL(opal_invalid_call);
1131EXPORT_SYMBOL_GPL(opal_xscom_read);
1132EXPORT_SYMBOL_GPL(opal_xscom_write);
1133EXPORT_SYMBOL_GPL(opal_ipmi_send);
1134EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1135EXPORT_SYMBOL_GPL(opal_flash_read);
1136EXPORT_SYMBOL_GPL(opal_flash_write);
1137EXPORT_SYMBOL_GPL(opal_flash_erase);
1138EXPORT_SYMBOL_GPL(opal_prd_msg);
1139EXPORT_SYMBOL_GPL(opal_check_token);
1140
1141/* Convert a region of vmalloc memory to an opal sg list */
1142struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1143 unsigned long vmalloc_size)
1144{
1145 struct opal_sg_list *sg, *first = NULL;
1146 unsigned long i = 0;
1147
1148 sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1149 if (!sg)
1150 goto nomem;
1151
1152 first = sg;
1153
1154 while (vmalloc_size > 0) {
1155 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1156 uint64_t length = min(vmalloc_size, PAGE_SIZE);
1157
1158 sg->entry[i].data = cpu_to_be64(data);
1159 sg->entry[i].length = cpu_to_be64(length);
1160 i++;
1161
1162 if (i >= SG_ENTRIES_PER_NODE) {
1163 struct opal_sg_list *next;
1164
1165 next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1166 if (!next)
1167 goto nomem;
1168
1169 sg->length = cpu_to_be64(
1170 i * sizeof(struct opal_sg_entry) + 16);
1171 i = 0;
1172 sg->next = cpu_to_be64(__pa(next));
1173 sg = next;
1174 }
1175
1176 vmalloc_addr += length;
1177 vmalloc_size -= length;
1178 }
1179
1180 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1181
1182 return first;
1183
1184nomem:
1185 pr_err("%s : Failed to allocate memory\n", __func__);
1186 opal_free_sg_list(first);
1187 return NULL;
1188}
1189
1190void opal_free_sg_list(struct opal_sg_list *sg)
1191{
1192 while (sg) {
1193 uint64_t next = be64_to_cpu(sg->next);
1194
1195 kfree(sg);
1196
1197 if (next)
1198 sg = __va(next);
1199 else
1200 sg = NULL;
1201 }
1202}
1203
1204int opal_error_code(int rc)
1205{
1206 switch (rc) {
1207 case OPAL_SUCCESS: return 0;
1208
1209 case OPAL_PARAMETER: return -EINVAL;
1210 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS;
1211 case OPAL_BUSY:
1212 case OPAL_BUSY_EVENT: return -EBUSY;
1213 case OPAL_NO_MEM: return -ENOMEM;
1214 case OPAL_PERMISSION: return -EPERM;
1215
1216 case OPAL_UNSUPPORTED: return -EIO;
1217 case OPAL_HARDWARE: return -EIO;
1218 case OPAL_INTERNAL_ERROR: return -EIO;
1219 case OPAL_TIMEOUT: return -ETIMEDOUT;
1220 default:
1221 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1222 return -EIO;
1223 }
1224}
1225
1226void powernv_set_nmmu_ptcr(unsigned long ptcr)
1227{
1228 int rc;
1229
1230 if (firmware_has_feature(FW_FEATURE_OPAL)) {
1231 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1232 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1233 pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1234 }
1235}
1236
1237EXPORT_SYMBOL_GPL(opal_poll_events);
1238EXPORT_SYMBOL_GPL(opal_rtc_read);
1239EXPORT_SYMBOL_GPL(opal_rtc_write);
1240EXPORT_SYMBOL_GPL(opal_tpo_read);
1241EXPORT_SYMBOL_GPL(opal_tpo_write);
1242EXPORT_SYMBOL_GPL(opal_i2c_request);
1243/* Export these symbols for PowerNV LED class driver */
1244EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1245EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1246/* Export this symbol for PowerNV Operator Panel class driver */
1247EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1248/* Export this for KVM */
1249EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1250EXPORT_SYMBOL_GPL(opal_int_eoi);
1251EXPORT_SYMBOL_GPL(opal_error_code);
1252/* Export the below symbol for NX compression */
1253EXPORT_SYMBOL(opal_nx_coproc_init);
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 struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
57static uint32_t opal_heartbeat;
58
59static void opal_reinit_cores(void)
60{
61 /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
62 *
63 * It will preserve non volatile GPRs and HSPRG0/1. It will
64 * also restore HIDs and other SPRs to their original value
65 * but it might clobber a bunch.
66 */
67#ifdef __BIG_ENDIAN__
68 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
69#else
70 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
71#endif
72}
73
74int __init early_init_dt_scan_opal(unsigned long node,
75 const char *uname, int depth, void *data)
76{
77 const void *basep, *entryp, *sizep;
78 int basesz, entrysz, runtimesz;
79
80 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
81 return 0;
82
83 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
84 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
85 sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
86
87 if (!basep || !entryp || !sizep)
88 return 1;
89
90 opal.base = of_read_number(basep, basesz/4);
91 opal.entry = of_read_number(entryp, entrysz/4);
92 opal.size = of_read_number(sizep, runtimesz/4);
93
94 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%d)\n",
95 opal.base, basep, basesz);
96 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
97 opal.entry, entryp, entrysz);
98 pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
99 opal.size, sizep, runtimesz);
100
101 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
102 powerpc_firmware_features |= FW_FEATURE_OPAL;
103 pr_info("OPAL detected !\n");
104 } else {
105 panic("OPAL != V3 detected, no longer supported.\n");
106 }
107
108 /* Reinit all cores with the right endian */
109 opal_reinit_cores();
110
111 /* Restore some bits */
112 if (cur_cpu_spec->cpu_restore)
113 cur_cpu_spec->cpu_restore();
114
115 return 1;
116}
117
118int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
119 const char *uname, int depth, void *data)
120{
121 int i, psize, size;
122 const __be32 *prop;
123
124 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
125 return 0;
126
127 prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
128
129 if (!prop)
130 return 1;
131
132 pr_debug("Found machine check recoverable ranges.\n");
133
134 /*
135 * Calculate number of available entries.
136 *
137 * Each recoverable address range entry is (start address, len,
138 * recovery address), 2 cells each for start and recovery address,
139 * 1 cell for len, totalling 5 cells per entry.
140 */
141 mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
142
143 /* Sanity check */
144 if (!mc_recoverable_range_len)
145 return 1;
146
147 /* Size required to hold all the entries. */
148 size = mc_recoverable_range_len *
149 sizeof(struct mcheck_recoverable_range);
150
151 /*
152 * Allocate a buffer to hold the MC recoverable ranges. We would be
153 * accessing them in real mode, hence it needs to be within
154 * RMO region.
155 */
156 mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
157 ppc64_rma_size));
158 memset(mc_recoverable_range, 0, size);
159
160 for (i = 0; i < mc_recoverable_range_len; i++) {
161 mc_recoverable_range[i].start_addr =
162 of_read_number(prop + (i * 5) + 0, 2);
163 mc_recoverable_range[i].end_addr =
164 mc_recoverable_range[i].start_addr +
165 of_read_number(prop + (i * 5) + 2, 1);
166 mc_recoverable_range[i].recover_addr =
167 of_read_number(prop + (i * 5) + 3, 2);
168
169 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
170 mc_recoverable_range[i].start_addr,
171 mc_recoverable_range[i].end_addr,
172 mc_recoverable_range[i].recover_addr);
173 }
174 return 1;
175}
176
177static int __init opal_register_exception_handlers(void)
178{
179#ifdef __BIG_ENDIAN__
180 u64 glue;
181
182 if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
183 return -ENODEV;
184
185 /* Hookup some exception handlers except machine check. We use the
186 * fwnmi area at 0x7000 to provide the glue space to OPAL
187 */
188 glue = 0x7000;
189
190 /*
191 * Check if we are running on newer firmware that exports
192 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
193 * the HMI interrupt and we catch it directly in Linux.
194 *
195 * For older firmware (i.e currently released POWER8 System Firmware
196 * as of today <= SV810_087), we fallback to old behavior and let OPAL
197 * patch the HMI vector and handle it inside OPAL firmware.
198 *
199 * For newer firmware (in development/yet to be released) we will
200 * start catching/handling HMI directly in Linux.
201 */
202 if (!opal_check_token(OPAL_HANDLE_HMI)) {
203 pr_info("Old firmware detected, OPAL handles HMIs.\n");
204 opal_register_exception_handler(
205 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
206 0, glue);
207 glue += 128;
208 }
209
210 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
211#endif
212
213 return 0;
214}
215machine_early_initcall(powernv, opal_register_exception_handlers);
216
217/*
218 * Opal message notifier based on message type. Allow subscribers to get
219 * notified for specific messgae type.
220 */
221int opal_message_notifier_register(enum opal_msg_type msg_type,
222 struct notifier_block *nb)
223{
224 if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
225 pr_warning("%s: Invalid arguments, msg_type:%d\n",
226 __func__, msg_type);
227 return -EINVAL;
228 }
229
230 return atomic_notifier_chain_register(
231 &opal_msg_notifier_head[msg_type], nb);
232}
233EXPORT_SYMBOL_GPL(opal_message_notifier_register);
234
235int opal_message_notifier_unregister(enum opal_msg_type msg_type,
236 struct notifier_block *nb)
237{
238 return atomic_notifier_chain_unregister(
239 &opal_msg_notifier_head[msg_type], nb);
240}
241EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
242
243static void opal_message_do_notify(uint32_t msg_type, void *msg)
244{
245 /* notify subscribers */
246 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
247 msg_type, msg);
248}
249
250static void opal_handle_message(void)
251{
252 s64 ret;
253 /*
254 * TODO: pre-allocate a message buffer depending on opal-msg-size
255 * value in /proc/device-tree.
256 */
257 static struct opal_msg msg;
258 u32 type;
259
260 ret = opal_get_msg(__pa(&msg), sizeof(msg));
261 /* No opal message pending. */
262 if (ret == OPAL_RESOURCE)
263 return;
264
265 /* check for errors. */
266 if (ret) {
267 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
268 __func__, ret);
269 return;
270 }
271
272 type = be32_to_cpu(msg.msg_type);
273
274 /* Sanity check */
275 if (type >= OPAL_MSG_TYPE_MAX) {
276 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
277 return;
278 }
279 opal_message_do_notify(type, (void *)&msg);
280}
281
282static irqreturn_t opal_message_notify(int irq, void *data)
283{
284 opal_handle_message();
285 return IRQ_HANDLED;
286}
287
288static int __init opal_message_init(void)
289{
290 int ret, i, irq;
291
292 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
293 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
294
295 irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
296 if (!irq) {
297 pr_err("%s: Can't register OPAL event irq (%d)\n",
298 __func__, irq);
299 return irq;
300 }
301
302 ret = request_irq(irq, opal_message_notify,
303 IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
304 if (ret) {
305 pr_err("%s: Can't request OPAL event irq (%d)\n",
306 __func__, ret);
307 return ret;
308 }
309
310 return 0;
311}
312
313int opal_get_chars(uint32_t vtermno, char *buf, int count)
314{
315 s64 rc;
316 __be64 evt, len;
317
318 if (!opal.entry)
319 return -ENODEV;
320 opal_poll_events(&evt);
321 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
322 return 0;
323 len = cpu_to_be64(count);
324 rc = opal_console_read(vtermno, &len, buf);
325 if (rc == OPAL_SUCCESS)
326 return be64_to_cpu(len);
327 return 0;
328}
329
330int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
331{
332 int written = 0;
333 __be64 olen;
334 s64 len, rc;
335 unsigned long flags;
336 __be64 evt;
337
338 if (!opal.entry)
339 return -ENODEV;
340
341 /* We want put_chars to be atomic to avoid mangling of hvsi
342 * packets. To do that, we first test for room and return
343 * -EAGAIN if there isn't enough.
344 *
345 * Unfortunately, opal_console_write_buffer_space() doesn't
346 * appear to work on opal v1, so we just assume there is
347 * enough room and be done with it
348 */
349 spin_lock_irqsave(&opal_write_lock, flags);
350 rc = opal_console_write_buffer_space(vtermno, &olen);
351 len = be64_to_cpu(olen);
352 if (rc || len < total_len) {
353 spin_unlock_irqrestore(&opal_write_lock, flags);
354 /* Closed -> drop characters */
355 if (rc)
356 return total_len;
357 opal_poll_events(NULL);
358 return -EAGAIN;
359 }
360
361 /* We still try to handle partial completions, though they
362 * should no longer happen.
363 */
364 rc = OPAL_BUSY;
365 while(total_len > 0 && (rc == OPAL_BUSY ||
366 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
367 olen = cpu_to_be64(total_len);
368 rc = opal_console_write(vtermno, &olen, data);
369 len = be64_to_cpu(olen);
370
371 /* Closed or other error drop */
372 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
373 rc != OPAL_BUSY_EVENT) {
374 written = total_len;
375 break;
376 }
377 if (rc == OPAL_SUCCESS) {
378 total_len -= len;
379 data += len;
380 written += len;
381 }
382 /* This is a bit nasty but we need that for the console to
383 * flush when there aren't any interrupts. We will clean
384 * things a bit later to limit that to synchronous path
385 * such as the kernel console and xmon/udbg
386 */
387 do
388 opal_poll_events(&evt);
389 while(rc == OPAL_SUCCESS &&
390 (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
391 }
392 spin_unlock_irqrestore(&opal_write_lock, flags);
393 return written;
394}
395
396static int opal_recover_mce(struct pt_regs *regs,
397 struct machine_check_event *evt)
398{
399 int recovered = 0;
400 uint64_t ea = get_mce_fault_addr(evt);
401
402 if (!(regs->msr & MSR_RI)) {
403 /* If MSR_RI isn't set, we cannot recover */
404 recovered = 0;
405 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
406 /* Platform corrected itself */
407 recovered = 1;
408 } else if (ea && !is_kernel_addr(ea)) {
409 /*
410 * Faulting address is not in kernel text. We should be fine.
411 * We need to find which process uses this address.
412 * For now, kill the task if we have received exception when
413 * in userspace.
414 *
415 * TODO: Queue up this address for hwpoisioning later.
416 */
417 if (user_mode(regs) && !is_global_init(current)) {
418 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
419 recovered = 1;
420 } else
421 recovered = 0;
422 } else if (user_mode(regs) && !is_global_init(current) &&
423 evt->severity == MCE_SEV_ERROR_SYNC) {
424 /*
425 * If we have received a synchronous error when in userspace
426 * kill the task.
427 */
428 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
429 recovered = 1;
430 }
431 return recovered;
432}
433
434int opal_machine_check(struct pt_regs *regs)
435{
436 struct machine_check_event evt;
437 int ret;
438
439 if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
440 return 0;
441
442 /* Print things out */
443 if (evt.version != MCE_V1) {
444 pr_err("Machine Check Exception, Unknown event version %d !\n",
445 evt.version);
446 return 0;
447 }
448 machine_check_print_event_info(&evt);
449
450 if (opal_recover_mce(regs, &evt))
451 return 1;
452
453 /*
454 * Unrecovered machine check, we are heading to panic path.
455 *
456 * We may have hit this MCE in very early stage of kernel
457 * initialization even before opal-prd has started running. If
458 * this is the case then this MCE error may go un-noticed or
459 * un-analyzed if we go down panic path. We need to inform
460 * BMC/OCC about this error so that they can collect relevant
461 * data for error analysis before rebooting.
462 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
463 * This function may not return on BMC based system.
464 */
465 ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
466 "Unrecoverable Machine Check exception");
467 if (ret == OPAL_UNSUPPORTED) {
468 pr_emerg("Reboot type %d not supported\n",
469 OPAL_REBOOT_PLATFORM_ERROR);
470 }
471
472 /*
473 * We reached here. There can be three possibilities:
474 * 1. We are running on a firmware level that do not support
475 * opal_cec_reboot2()
476 * 2. We are running on a firmware level that do not support
477 * OPAL_REBOOT_PLATFORM_ERROR reboot type.
478 * 3. We are running on FSP based system that does not need opal
479 * to trigger checkstop explicitly for error analysis. The FSP
480 * PRD component would have already got notified about this
481 * error through other channels.
482 *
483 * If hardware marked this as an unrecoverable MCE, we are
484 * going to panic anyway. Even if it didn't, it's not safe to
485 * continue at this point, so we should explicitly panic.
486 */
487
488 panic("PowerNV Unrecovered Machine Check");
489 return 0;
490}
491
492/* Early hmi handler called in real mode. */
493int opal_hmi_exception_early(struct pt_regs *regs)
494{
495 s64 rc;
496
497 /*
498 * call opal hmi handler. Pass paca address as token.
499 * The return value OPAL_SUCCESS is an indication that there is
500 * an HMI event generated waiting to pull by Linux.
501 */
502 rc = opal_handle_hmi();
503 if (rc == OPAL_SUCCESS) {
504 local_paca->hmi_event_available = 1;
505 return 1;
506 }
507 return 0;
508}
509
510/* HMI exception handler called in virtual mode during check_irq_replay. */
511int opal_handle_hmi_exception(struct pt_regs *regs)
512{
513 s64 rc;
514 __be64 evt = 0;
515
516 /*
517 * Check if HMI event is available.
518 * if Yes, then call opal_poll_events to pull opal messages and
519 * process them.
520 */
521 if (!local_paca->hmi_event_available)
522 return 0;
523
524 local_paca->hmi_event_available = 0;
525 rc = opal_poll_events(&evt);
526 if (rc == OPAL_SUCCESS && evt)
527 opal_handle_events(be64_to_cpu(evt));
528
529 return 1;
530}
531
532static uint64_t find_recovery_address(uint64_t nip)
533{
534 int i;
535
536 for (i = 0; i < mc_recoverable_range_len; i++)
537 if ((nip >= mc_recoverable_range[i].start_addr) &&
538 (nip < mc_recoverable_range[i].end_addr))
539 return mc_recoverable_range[i].recover_addr;
540 return 0;
541}
542
543bool opal_mce_check_early_recovery(struct pt_regs *regs)
544{
545 uint64_t recover_addr = 0;
546
547 if (!opal.base || !opal.size)
548 goto out;
549
550 if ((regs->nip >= opal.base) &&
551 (regs->nip < (opal.base + opal.size)))
552 recover_addr = find_recovery_address(regs->nip);
553
554 /*
555 * Setup regs->nip to rfi into fixup address.
556 */
557 if (recover_addr)
558 regs->nip = recover_addr;
559
560out:
561 return !!recover_addr;
562}
563
564static int opal_sysfs_init(void)
565{
566 opal_kobj = kobject_create_and_add("opal", firmware_kobj);
567 if (!opal_kobj) {
568 pr_warn("kobject_create_and_add opal failed\n");
569 return -ENOMEM;
570 }
571
572 return 0;
573}
574
575static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
576 struct bin_attribute *bin_attr,
577 char *buf, loff_t off, size_t count)
578{
579 return memory_read_from_buffer(buf, count, &off, bin_attr->private,
580 bin_attr->size);
581}
582
583static BIN_ATTR_RO(symbol_map, 0);
584
585static void opal_export_symmap(void)
586{
587 const __be64 *syms;
588 unsigned int size;
589 struct device_node *fw;
590 int rc;
591
592 fw = of_find_node_by_path("/ibm,opal/firmware");
593 if (!fw)
594 return;
595 syms = of_get_property(fw, "symbol-map", &size);
596 if (!syms || size != 2 * sizeof(__be64))
597 return;
598
599 /* Setup attributes */
600 bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
601 bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
602
603 rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
604 if (rc)
605 pr_warn("Error %d creating OPAL symbols file\n", rc);
606}
607
608static void __init opal_dump_region_init(void)
609{
610 void *addr;
611 uint64_t size;
612 int rc;
613
614 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
615 return;
616
617 /* Register kernel log buffer */
618 addr = log_buf_addr_get();
619 if (addr == NULL)
620 return;
621
622 size = log_buf_len_get();
623 if (size == 0)
624 return;
625
626 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
627 __pa(addr), size);
628 /* Don't warn if this is just an older OPAL that doesn't
629 * know about that call
630 */
631 if (rc && rc != OPAL_UNSUPPORTED)
632 pr_warn("DUMP: Failed to register kernel log buffer. "
633 "rc = %d\n", rc);
634}
635
636static void opal_pdev_init(struct device_node *opal_node,
637 const char *compatible)
638{
639 struct device_node *np;
640
641 for_each_child_of_node(opal_node, np)
642 if (of_device_is_compatible(np, compatible))
643 of_platform_device_create(np, NULL, NULL);
644}
645
646static void opal_i2c_create_devs(void)
647{
648 struct device_node *np;
649
650 for_each_compatible_node(np, NULL, "ibm,opal-i2c")
651 of_platform_device_create(np, NULL, NULL);
652}
653
654static int kopald(void *unused)
655{
656 __be64 events;
657
658 set_freezable();
659 do {
660 try_to_freeze();
661 opal_poll_events(&events);
662 opal_handle_events(be64_to_cpu(events));
663 msleep_interruptible(opal_heartbeat);
664 } while (!kthread_should_stop());
665
666 return 0;
667}
668
669static void opal_init_heartbeat(void)
670{
671 /* Old firwmware, we assume the HVC heartbeat is sufficient */
672 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
673 &opal_heartbeat) != 0)
674 opal_heartbeat = 0;
675
676 if (opal_heartbeat)
677 kthread_run(kopald, NULL, "kopald");
678}
679
680static int __init opal_init(void)
681{
682 struct device_node *np, *consoles, *leds;
683 int rc;
684
685 opal_node = of_find_node_by_path("/ibm,opal");
686 if (!opal_node) {
687 pr_warn("Device node not found\n");
688 return -ENODEV;
689 }
690
691 /* Register OPAL consoles if any ports */
692 consoles = of_find_node_by_path("/ibm,opal/consoles");
693 if (consoles) {
694 for_each_child_of_node(consoles, np) {
695 if (strcmp(np->name, "serial"))
696 continue;
697 of_platform_device_create(np, NULL, NULL);
698 }
699 of_node_put(consoles);
700 }
701
702 /* Initialise OPAL messaging system */
703 opal_message_init();
704
705 /* Initialise OPAL asynchronous completion interface */
706 opal_async_comp_init();
707
708 /* Initialise OPAL sensor interface */
709 opal_sensor_init();
710
711 /* Initialise OPAL hypervisor maintainence interrupt handling */
712 opal_hmi_handler_init();
713
714 /* Create i2c platform devices */
715 opal_i2c_create_devs();
716
717 /* Setup a heatbeat thread if requested by OPAL */
718 opal_init_heartbeat();
719
720 /* Create leds platform devices */
721 leds = of_find_node_by_path("/ibm,opal/leds");
722 if (leds) {
723 of_platform_device_create(leds, "opal_leds", NULL);
724 of_node_put(leds);
725 }
726
727 /* Initialise OPAL message log interface */
728 opal_msglog_init();
729
730 /* Create "opal" kobject under /sys/firmware */
731 rc = opal_sysfs_init();
732 if (rc == 0) {
733 /* Export symbol map to userspace */
734 opal_export_symmap();
735 /* Setup dump region interface */
736 opal_dump_region_init();
737 /* Setup error log interface */
738 rc = opal_elog_init();
739 /* Setup code update interface */
740 opal_flash_update_init();
741 /* Setup platform dump extract interface */
742 opal_platform_dump_init();
743 /* Setup system parameters interface */
744 opal_sys_param_init();
745 /* Setup message log sysfs interface. */
746 opal_msglog_sysfs_init();
747 }
748
749 /* Initialize platform devices: IPMI backend, PRD & flash interface */
750 opal_pdev_init(opal_node, "ibm,opal-ipmi");
751 opal_pdev_init(opal_node, "ibm,opal-flash");
752 opal_pdev_init(opal_node, "ibm,opal-prd");
753
754 /* Initialise OPAL kmsg dumper for flushing console on panic */
755 opal_kmsg_init();
756
757 return 0;
758}
759machine_subsys_initcall(powernv, opal_init);
760
761void opal_shutdown(void)
762{
763 long rc = OPAL_BUSY;
764
765 opal_event_shutdown();
766
767 /*
768 * Then sync with OPAL which ensure anything that can
769 * potentially write to our memory has completed such
770 * as an ongoing dump retrieval
771 */
772 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
773 rc = opal_sync_host_reboot();
774 if (rc == OPAL_BUSY)
775 opal_poll_events(NULL);
776 else
777 mdelay(10);
778 }
779
780 /* Unregister memory dump region */
781 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
782 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
783}
784
785/* Export this so that test modules can use it */
786EXPORT_SYMBOL_GPL(opal_invalid_call);
787EXPORT_SYMBOL_GPL(opal_xscom_read);
788EXPORT_SYMBOL_GPL(opal_xscom_write);
789EXPORT_SYMBOL_GPL(opal_ipmi_send);
790EXPORT_SYMBOL_GPL(opal_ipmi_recv);
791EXPORT_SYMBOL_GPL(opal_flash_read);
792EXPORT_SYMBOL_GPL(opal_flash_write);
793EXPORT_SYMBOL_GPL(opal_flash_erase);
794EXPORT_SYMBOL_GPL(opal_prd_msg);
795
796/* Convert a region of vmalloc memory to an opal sg list */
797struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
798 unsigned long vmalloc_size)
799{
800 struct opal_sg_list *sg, *first = NULL;
801 unsigned long i = 0;
802
803 sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
804 if (!sg)
805 goto nomem;
806
807 first = sg;
808
809 while (vmalloc_size > 0) {
810 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
811 uint64_t length = min(vmalloc_size, PAGE_SIZE);
812
813 sg->entry[i].data = cpu_to_be64(data);
814 sg->entry[i].length = cpu_to_be64(length);
815 i++;
816
817 if (i >= SG_ENTRIES_PER_NODE) {
818 struct opal_sg_list *next;
819
820 next = kzalloc(PAGE_SIZE, GFP_KERNEL);
821 if (!next)
822 goto nomem;
823
824 sg->length = cpu_to_be64(
825 i * sizeof(struct opal_sg_entry) + 16);
826 i = 0;
827 sg->next = cpu_to_be64(__pa(next));
828 sg = next;
829 }
830
831 vmalloc_addr += length;
832 vmalloc_size -= length;
833 }
834
835 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
836
837 return first;
838
839nomem:
840 pr_err("%s : Failed to allocate memory\n", __func__);
841 opal_free_sg_list(first);
842 return NULL;
843}
844
845void opal_free_sg_list(struct opal_sg_list *sg)
846{
847 while (sg) {
848 uint64_t next = be64_to_cpu(sg->next);
849
850 kfree(sg);
851
852 if (next)
853 sg = __va(next);
854 else
855 sg = NULL;
856 }
857}
858
859int opal_error_code(int rc)
860{
861 switch (rc) {
862 case OPAL_SUCCESS: return 0;
863
864 case OPAL_PARAMETER: return -EINVAL;
865 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS;
866 case OPAL_BUSY_EVENT: return -EBUSY;
867 case OPAL_NO_MEM: return -ENOMEM;
868 case OPAL_PERMISSION: return -EPERM;
869
870 case OPAL_UNSUPPORTED: return -EIO;
871 case OPAL_HARDWARE: return -EIO;
872 case OPAL_INTERNAL_ERROR: return -EIO;
873 default:
874 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
875 return -EIO;
876 }
877}
878
879EXPORT_SYMBOL_GPL(opal_poll_events);
880EXPORT_SYMBOL_GPL(opal_rtc_read);
881EXPORT_SYMBOL_GPL(opal_rtc_write);
882EXPORT_SYMBOL_GPL(opal_tpo_read);
883EXPORT_SYMBOL_GPL(opal_tpo_write);
884EXPORT_SYMBOL_GPL(opal_i2c_request);
885/* Export these symbols for PowerNV LED class driver */
886EXPORT_SYMBOL_GPL(opal_leds_get_ind);
887EXPORT_SYMBOL_GPL(opal_leds_set_ind);