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
40 struct opal_msg_node {
41 struct list_head list;
42 struct opal_msg msg;
43 };
44
45 static DEFINE_SPINLOCK(msg_list_lock);
46 static LIST_HEAD(msg_list);
47
48 /* /sys/firmware/opal */
49 struct kobject *opal_kobj;
50
51 struct opal {
52 u64 base;
53 u64 entry;
54 u64 size;
55 } opal;
56
57 struct mcheck_recoverable_range {
58 u64 start_addr;
59 u64 end_addr;
60 u64 recover_addr;
61 };
62
63 static int msg_list_size;
64
65 static struct mcheck_recoverable_range *mc_recoverable_range;
66 static int mc_recoverable_range_len;
67
68 struct device_node *opal_node;
69 static DEFINE_SPINLOCK(opal_write_lock);
70 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
71 static uint32_t opal_heartbeat;
72 static struct task_struct *kopald_tsk;
73 static struct opal_msg *opal_msg;
74 static u32 opal_msg_size __ro_after_init;
75
76 void __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
110 int __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
147 int __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
205 static 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 }
250 machine_early_initcall(powernv, opal_register_exception_handlers);
251
252 static 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
271 static 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 */
293 int 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 }
321 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
322
323 int 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 }
329 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
330
331 static 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
355 static 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
382 static irqreturn_t opal_message_notify(int irq, void *data)
383 {
384 opal_handle_message();
385 return IRQ_HANDLED;
386 }
387
388 static 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
427 ssize_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
444 static 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
494 out:
495 if (atomic)
496 spin_unlock_irqrestore(&opal_write_lock, flags);
497
498 return written;
499 }
500
501 ssize_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 */
512 ssize_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
518 static 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 */
551 int 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 */
570 int 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
588 static 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
637 void __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
676 int 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. */
698 int 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
715 int 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. */
737 int 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
752 static 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
763 bool 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
780 out:
781 return !!recover_addr;
782 }
783
784 static 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
795 static int opal_add_one_export(struct kobject *parent, const char *export_name,
796 struct device_node *np, const char *prop_name)
797 {
798 struct bin_attribute *attr = NULL;
799 const char *name = NULL;
800 u64 vals[2];
801 int rc;
802
803 rc = of_property_read_u64_array(np, prop_name, &vals[0], 2);
804 if (rc)
805 goto out;
806
807 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
808 if (!attr) {
809 rc = -ENOMEM;
810 goto out;
811 }
812 name = kstrdup(export_name, GFP_KERNEL);
813 if (!name) {
814 rc = -ENOMEM;
815 goto out;
816 }
817
818 sysfs_bin_attr_init(attr);
819 attr->attr.name = name;
820 attr->attr.mode = 0400;
821 attr->read = sysfs_bin_attr_simple_read;
822 attr->private = __va(vals[0]);
823 attr->size = vals[1];
824
825 rc = sysfs_create_bin_file(parent, attr);
826 out:
827 if (rc) {
828 kfree(name);
829 kfree(attr);
830 }
831
832 return rc;
833 }
834
835 static void opal_add_exported_attrs(struct device_node *np,
836 struct kobject *kobj)
837 {
838 struct device_node *child;
839 struct property *prop;
840
841 for_each_property_of_node(np, prop) {
842 int rc;
843
844 if (!strcmp(prop->name, "name") ||
845 !strcmp(prop->name, "phandle"))
846 continue;
847
848 rc = opal_add_one_export(kobj, prop->name, np, prop->name);
849 if (rc) {
850 pr_warn("Unable to add export %pOF/%s, rc = %d!\n",
851 np, prop->name, rc);
852 }
853 }
854
855 for_each_child_of_node(np, child) {
856 struct kobject *child_kobj;
857
858 child_kobj = kobject_create_and_add(child->name, kobj);
859 if (!child_kobj) {
860 pr_err("Unable to create export dir for %pOF\n", child);
861 continue;
862 }
863
864 opal_add_exported_attrs(child, child_kobj);
865 }
866 }
867
868 /*
869 * opal_export_attrs: creates a sysfs node for each property listed in
870 * the device-tree under /ibm,opal/firmware/exports/
871 * All new sysfs nodes are created under /opal/exports/.
872 * This allows for reserved memory regions (e.g. HDAT) to be read.
873 * The new sysfs nodes are only readable by root.
874 */
875 static void opal_export_attrs(void)
876 {
877 struct device_node *np;
878 struct kobject *kobj;
879 int rc;
880
881 np = of_find_node_by_path("/ibm,opal/firmware/exports");
882 if (!np)
883 return;
884
885 /* Create new 'exports' directory - /sys/firmware/opal/exports */
886 kobj = kobject_create_and_add("exports", opal_kobj);
887 if (!kobj) {
888 pr_warn("kobject_create_and_add() of exports failed\n");
889 of_node_put(np);
890 return;
891 }
892
893 opal_add_exported_attrs(np, kobj);
894
895 /*
896 * NB: symbol_map existed before the generic export interface so it
897 * lives under the top level opal_kobj.
898 */
899 rc = opal_add_one_export(opal_kobj, "symbol_map",
900 np->parent, "symbol-map");
901 if (rc)
902 pr_warn("Error %d creating OPAL symbols file\n", rc);
903
904 of_node_put(np);
905 }
906
907 static void __init opal_dump_region_init(void)
908 {
909 void *addr;
910 uint64_t size;
911 int rc;
912
913 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
914 return;
915
916 /* Register kernel log buffer */
917 addr = log_buf_addr_get();
918 if (addr == NULL)
919 return;
920
921 size = log_buf_len_get();
922 if (size == 0)
923 return;
924
925 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
926 __pa(addr), size);
927 /* Don't warn if this is just an older OPAL that doesn't
928 * know about that call
929 */
930 if (rc && rc != OPAL_UNSUPPORTED)
931 pr_warn("DUMP: Failed to register kernel log buffer. "
932 "rc = %d\n", rc);
933 }
934
935 static void __init opal_pdev_init(const char *compatible)
936 {
937 struct device_node *np;
938
939 for_each_compatible_node(np, NULL, compatible)
940 of_platform_device_create(np, NULL, NULL);
941 }
942
943 static void __init opal_imc_init_dev(void)
944 {
945 struct device_node *np;
946
947 np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
948 if (np)
949 of_platform_device_create(np, NULL, NULL);
950
951 of_node_put(np);
952 }
953
954 static int kopald(void *unused)
955 {
956 unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
957
958 set_freezable();
959 do {
960 try_to_freeze();
961
962 opal_handle_events();
963
964 set_current_state(TASK_INTERRUPTIBLE);
965 if (opal_have_pending_events())
966 __set_current_state(TASK_RUNNING);
967 else
968 schedule_timeout(timeout);
969
970 } while (!kthread_should_stop());
971
972 return 0;
973 }
974
975 void opal_wake_poller(void)
976 {
977 if (kopald_tsk)
978 wake_up_process(kopald_tsk);
979 }
980
981 static void __init opal_init_heartbeat(void)
982 {
983 /* Old firwmware, we assume the HVC heartbeat is sufficient */
984 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
985 &opal_heartbeat) != 0)
986 opal_heartbeat = 0;
987
988 if (opal_heartbeat)
989 kopald_tsk = kthread_run(kopald, NULL, "kopald");
990 }
991
992 static int __init opal_init(void)
993 {
994 struct device_node *np, *consoles, *leds;
995 int rc;
996
997 opal_node = of_find_node_by_path("/ibm,opal");
998 if (!opal_node) {
999 pr_warn("Device node not found\n");
1000 return -ENODEV;
1001 }
1002
1003 /* Register OPAL consoles if any ports */
1004 consoles = of_find_node_by_path("/ibm,opal/consoles");
1005 if (consoles) {
1006 for_each_child_of_node(consoles, np) {
1007 if (!of_node_name_eq(np, "serial"))
1008 continue;
1009 of_platform_device_create(np, NULL, NULL);
1010 }
1011 of_node_put(consoles);
1012 }
1013
1014 /* Initialise OPAL messaging system */
1015 opal_message_init(opal_node);
1016
1017 /* Initialise OPAL asynchronous completion interface */
1018 opal_async_comp_init();
1019
1020 /* Initialise OPAL sensor interface */
1021 opal_sensor_init();
1022
1023 /* Initialise OPAL hypervisor maintainence interrupt handling */
1024 opal_hmi_handler_init();
1025
1026 /* Create i2c platform devices */
1027 opal_pdev_init("ibm,opal-i2c");
1028
1029 /* Handle non-volatile memory devices */
1030 opal_pdev_init("pmem-region");
1031
1032 /* Setup a heatbeat thread if requested by OPAL */
1033 opal_init_heartbeat();
1034
1035 /* Detect In-Memory Collection counters and create devices*/
1036 opal_imc_init_dev();
1037
1038 /* Create leds platform devices */
1039 leds = of_find_node_by_path("/ibm,opal/leds");
1040 if (leds) {
1041 of_platform_device_create(leds, "opal_leds", NULL);
1042 of_node_put(leds);
1043 }
1044
1045 /* Initialise OPAL message log interface */
1046 opal_msglog_init();
1047
1048 /* Create "opal" kobject under /sys/firmware */
1049 rc = opal_sysfs_init();
1050 if (rc == 0) {
1051 /* Setup dump region interface */
1052 opal_dump_region_init();
1053 /* Setup error log interface */
1054 rc = opal_elog_init();
1055 /* Setup code update interface */
1056 opal_flash_update_init();
1057 /* Setup platform dump extract interface */
1058 opal_platform_dump_init();
1059 /* Setup system parameters interface */
1060 opal_sys_param_init();
1061 /* Setup message log sysfs interface. */
1062 opal_msglog_sysfs_init();
1063 /* Add all export properties*/
1064 opal_export_attrs();
1065 }
1066
1067 /* Initialize platform devices: IPMI backend, PRD & flash interface */
1068 opal_pdev_init("ibm,opal-ipmi");
1069 opal_pdev_init("ibm,opal-flash");
1070 opal_pdev_init("ibm,opal-prd");
1071
1072 /* Initialise platform device: oppanel interface */
1073 opal_pdev_init("ibm,opal-oppanel");
1074
1075 /* Initialise OPAL kmsg dumper for flushing console on panic */
1076 opal_kmsg_init();
1077
1078 /* Initialise OPAL powercap interface */
1079 opal_powercap_init();
1080
1081 /* Initialise OPAL Power-Shifting-Ratio interface */
1082 opal_psr_init();
1083
1084 /* Initialise OPAL sensor groups */
1085 opal_sensor_groups_init();
1086
1087 /* Initialise OPAL Power control interface */
1088 opal_power_control_init();
1089
1090 /* Initialize OPAL secure variables */
1091 opal_pdev_init("ibm,secvar-backend");
1092
1093 return 0;
1094 }
1095 machine_subsys_initcall(powernv, opal_init);
1096
1097 void opal_shutdown(void)
1098 {
1099 long rc = OPAL_BUSY;
1100
1101 opal_event_shutdown();
1102
1103 /*
1104 * Then sync with OPAL which ensure anything that can
1105 * potentially write to our memory has completed such
1106 * as an ongoing dump retrieval
1107 */
1108 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
1109 rc = opal_sync_host_reboot();
1110 if (rc == OPAL_BUSY)
1111 opal_poll_events(NULL);
1112 else
1113 mdelay(10);
1114 }
1115
1116 /* Unregister memory dump region */
1117 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
1118 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
1119 }
1120
1121 /* Export this so that test modules can use it */
1122 EXPORT_SYMBOL_GPL(opal_invalid_call);
1123 EXPORT_SYMBOL_GPL(opal_xscom_read);
1124 EXPORT_SYMBOL_GPL(opal_xscom_write);
1125 EXPORT_SYMBOL_GPL(opal_ipmi_send);
1126 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1127 EXPORT_SYMBOL_GPL(opal_flash_read);
1128 EXPORT_SYMBOL_GPL(opal_flash_write);
1129 EXPORT_SYMBOL_GPL(opal_flash_erase);
1130 EXPORT_SYMBOL_GPL(opal_prd_msg);
1131 EXPORT_SYMBOL_GPL(opal_check_token);
1132
1133 /* Convert a region of vmalloc memory to an opal sg list */
1134 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1135 unsigned long vmalloc_size)
1136 {
1137 struct opal_sg_list *sg, *first = NULL;
1138 unsigned long i = 0;
1139
1140 sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1141 if (!sg)
1142 goto nomem;
1143
1144 first = sg;
1145
1146 while (vmalloc_size > 0) {
1147 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1148 uint64_t length = min(vmalloc_size, PAGE_SIZE);
1149
1150 sg->entry[i].data = cpu_to_be64(data);
1151 sg->entry[i].length = cpu_to_be64(length);
1152 i++;
1153
1154 if (i >= SG_ENTRIES_PER_NODE) {
1155 struct opal_sg_list *next;
1156
1157 next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1158 if (!next)
1159 goto nomem;
1160
1161 sg->length = cpu_to_be64(
1162 i * sizeof(struct opal_sg_entry) + 16);
1163 i = 0;
1164 sg->next = cpu_to_be64(__pa(next));
1165 sg = next;
1166 }
1167
1168 vmalloc_addr += length;
1169 vmalloc_size -= length;
1170 }
1171
1172 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1173
1174 return first;
1175
1176 nomem:
1177 pr_err("%s : Failed to allocate memory\n", __func__);
1178 opal_free_sg_list(first);
1179 return NULL;
1180 }
1181
1182 void opal_free_sg_list(struct opal_sg_list *sg)
1183 {
1184 while (sg) {
1185 uint64_t next = be64_to_cpu(sg->next);
1186
1187 kfree(sg);
1188
1189 if (next)
1190 sg = __va(next);
1191 else
1192 sg = NULL;
1193 }
1194 }
1195
1196 int opal_error_code(int rc)
1197 {
1198 switch (rc) {
1199 case OPAL_SUCCESS: return 0;
1200
1201 case OPAL_PARAMETER: return -EINVAL;
1202 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS;
1203 case OPAL_BUSY:
1204 case OPAL_BUSY_EVENT: return -EBUSY;
1205 case OPAL_NO_MEM: return -ENOMEM;
1206 case OPAL_PERMISSION: return -EPERM;
1207
1208 case OPAL_UNSUPPORTED: return -EIO;
1209 case OPAL_HARDWARE: return -EIO;
1210 case OPAL_INTERNAL_ERROR: return -EIO;
1211 case OPAL_TIMEOUT: return -ETIMEDOUT;
1212 default:
1213 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1214 return -EIO;
1215 }
1216 }
1217
1218 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1219 {
1220 int rc;
1221
1222 if (firmware_has_feature(FW_FEATURE_OPAL)) {
1223 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1224 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1225 pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1226 }
1227 }
1228
1229 EXPORT_SYMBOL_GPL(opal_poll_events);
1230 EXPORT_SYMBOL_GPL(opal_rtc_read);
1231 EXPORT_SYMBOL_GPL(opal_rtc_write);
1232 EXPORT_SYMBOL_GPL(opal_tpo_read);
1233 EXPORT_SYMBOL_GPL(opal_tpo_write);
1234 EXPORT_SYMBOL_GPL(opal_i2c_request);
1235 /* Export these symbols for PowerNV LED class driver */
1236 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1237 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1238 /* Export this symbol for PowerNV Operator Panel class driver */
1239 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1240 /* Export this for KVM */
1241 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1242 EXPORT_SYMBOL_GPL(opal_int_eoi);
1243 EXPORT_SYMBOL_GPL(opal_error_code);
1244 /* Export the below symbol for NX compression */
1245 EXPORT_SYMBOL(opal_nx_coproc_init);
1246
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.