treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152
[muen/linux.git] / arch / powerpc / platforms / powernv / opal.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV OPAL high level interfaces
4  *
5  * Copyright 2011 IBM Corp.
6  */
7
8 #define pr_fmt(fmt)     "opal: " fmt
9
10 #include <linux/printk.h>
11 #include <linux/types.h>
12 #include <linux/of.h>
13 #include <linux/of_fdt.h>
14 #include <linux/of_platform.h>
15 #include <linux/of_address.h>
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/slab.h>
19 #include <linux/sched.h>
20 #include <linux/kobject.h>
21 #include <linux/delay.h>
22 #include <linux/memblock.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/kmsg_dump.h>
26 #include <linux/console.h>
27 #include <linux/sched/debug.h>
28
29 #include <asm/machdep.h>
30 #include <asm/opal.h>
31 #include <asm/firmware.h>
32 #include <asm/mce.h>
33 #include <asm/imc-pmu.h>
34 #include <asm/bug.h>
35
36 #include "powernv.h"
37
38 /* /sys/firmware/opal */
39 struct kobject *opal_kobj;
40
41 struct opal {
42         u64 base;
43         u64 entry;
44         u64 size;
45 } opal;
46
47 struct mcheck_recoverable_range {
48         u64 start_addr;
49         u64 end_addr;
50         u64 recover_addr;
51 };
52
53 static struct mcheck_recoverable_range *mc_recoverable_range;
54 static int mc_recoverable_range_len;
55
56 struct device_node *opal_node;
57 static DEFINE_SPINLOCK(opal_write_lock);
58 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
59 static uint32_t opal_heartbeat;
60 static struct task_struct *kopald_tsk;
61
62 void opal_configure_cores(void)
63 {
64         u64 reinit_flags = 0;
65
66         /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
67          *
68          * It will preserve non volatile GPRs and HSPRG0/1. It will
69          * also restore HIDs and other SPRs to their original value
70          * but it might clobber a bunch.
71          */
72 #ifdef __BIG_ENDIAN__
73         reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
74 #else
75         reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
76 #endif
77
78         /*
79          * POWER9 always support running hash:
80          *  ie. Host hash  supports  hash guests
81          *      Host radix supports  hash/radix guests
82          */
83         if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
84                 reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
85                 if (early_radix_enabled())
86                         reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
87         }
88
89         opal_reinit_cpus(reinit_flags);
90
91         /* Restore some bits */
92         if (cur_cpu_spec->cpu_restore)
93                 cur_cpu_spec->cpu_restore();
94 }
95
96 int __init early_init_dt_scan_opal(unsigned long node,
97                                    const char *uname, int depth, void *data)
98 {
99         const void *basep, *entryp, *sizep;
100         int basesz, entrysz, runtimesz;
101
102         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
103                 return 0;
104
105         basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
106         entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
107         sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
108
109         if (!basep || !entryp || !sizep)
110                 return 1;
111
112         opal.base = of_read_number(basep, basesz/4);
113         opal.entry = of_read_number(entryp, entrysz/4);
114         opal.size = of_read_number(sizep, runtimesz/4);
115
116         pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
117                  opal.base, basep, basesz);
118         pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
119                  opal.entry, entryp, entrysz);
120         pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
121                  opal.size, sizep, runtimesz);
122
123         if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
124                 powerpc_firmware_features |= FW_FEATURE_OPAL;
125                 pr_debug("OPAL detected !\n");
126         } else {
127                 panic("OPAL != V3 detected, no longer supported.\n");
128         }
129
130         return 1;
131 }
132
133 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
134                                    const char *uname, int depth, void *data)
135 {
136         int i, psize, size;
137         const __be32 *prop;
138
139         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
140                 return 0;
141
142         prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
143
144         if (!prop)
145                 return 1;
146
147         pr_debug("Found machine check recoverable ranges.\n");
148
149         /*
150          * Calculate number of available entries.
151          *
152          * Each recoverable address range entry is (start address, len,
153          * recovery address), 2 cells each for start and recovery address,
154          * 1 cell for len, totalling 5 cells per entry.
155          */
156         mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
157
158         /* Sanity check */
159         if (!mc_recoverable_range_len)
160                 return 1;
161
162         /* Size required to hold all the entries. */
163         size = mc_recoverable_range_len *
164                         sizeof(struct mcheck_recoverable_range);
165
166         /*
167          * Allocate a buffer to hold the MC recoverable ranges.
168          */
169         mc_recoverable_range = memblock_alloc(size, __alignof__(u64));
170         if (!mc_recoverable_range)
171                 panic("%s: Failed to allocate %u bytes align=0x%lx\n",
172                       __func__, size, __alignof__(u64));
173
174         for (i = 0; i < mc_recoverable_range_len; i++) {
175                 mc_recoverable_range[i].start_addr =
176                                         of_read_number(prop + (i * 5) + 0, 2);
177                 mc_recoverable_range[i].end_addr =
178                                         mc_recoverable_range[i].start_addr +
179                                         of_read_number(prop + (i * 5) + 2, 1);
180                 mc_recoverable_range[i].recover_addr =
181                                         of_read_number(prop + (i * 5) + 3, 2);
182
183                 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
184                                 mc_recoverable_range[i].start_addr,
185                                 mc_recoverable_range[i].end_addr,
186                                 mc_recoverable_range[i].recover_addr);
187         }
188         return 1;
189 }
190
191 static int __init opal_register_exception_handlers(void)
192 {
193 #ifdef __BIG_ENDIAN__
194         u64 glue;
195
196         if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
197                 return -ENODEV;
198
199         /* Hookup some exception handlers except machine check. We use the
200          * fwnmi area at 0x7000 to provide the glue space to OPAL
201          */
202         glue = 0x7000;
203
204         /*
205          * Check if we are running on newer firmware that exports
206          * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
207          * the HMI interrupt and we catch it directly in Linux.
208          *
209          * For older firmware (i.e currently released POWER8 System Firmware
210          * as of today <= SV810_087), we fallback to old behavior and let OPAL
211          * patch the HMI vector and handle it inside OPAL firmware.
212          *
213          * For newer firmware (in development/yet to be released) we will
214          * start catching/handling HMI directly in Linux.
215          */
216         if (!opal_check_token(OPAL_HANDLE_HMI)) {
217                 pr_info("Old firmware detected, OPAL handles HMIs.\n");
218                 opal_register_exception_handler(
219                                 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
220                                 0, glue);
221                 glue += 128;
222         }
223
224         opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
225 #endif
226
227         return 0;
228 }
229 machine_early_initcall(powernv, opal_register_exception_handlers);
230
231 /*
232  * Opal message notifier based on message type. Allow subscribers to get
233  * notified for specific messgae type.
234  */
235 int opal_message_notifier_register(enum opal_msg_type msg_type,
236                                         struct notifier_block *nb)
237 {
238         if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
239                 pr_warn("%s: Invalid arguments, msg_type:%d\n",
240                         __func__, msg_type);
241                 return -EINVAL;
242         }
243
244         return atomic_notifier_chain_register(
245                                 &opal_msg_notifier_head[msg_type], nb);
246 }
247 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
248
249 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
250                                      struct notifier_block *nb)
251 {
252         return atomic_notifier_chain_unregister(
253                         &opal_msg_notifier_head[msg_type], nb);
254 }
255 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
256
257 static void opal_message_do_notify(uint32_t msg_type, void *msg)
258 {
259         /* notify subscribers */
260         atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
261                                         msg_type, msg);
262 }
263
264 static void opal_handle_message(void)
265 {
266         s64 ret;
267         /*
268          * TODO: pre-allocate a message buffer depending on opal-msg-size
269          * value in /proc/device-tree.
270          */
271         static struct opal_msg msg;
272         u32 type;
273
274         ret = opal_get_msg(__pa(&msg), sizeof(msg));
275         /* No opal message pending. */
276         if (ret == OPAL_RESOURCE)
277                 return;
278
279         /* check for errors. */
280         if (ret) {
281                 pr_warn("%s: Failed to retrieve opal message, err=%lld\n",
282                         __func__, ret);
283                 return;
284         }
285
286         type = be32_to_cpu(msg.msg_type);
287
288         /* Sanity check */
289         if (type >= OPAL_MSG_TYPE_MAX) {
290                 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
291                 return;
292         }
293         opal_message_do_notify(type, (void *)&msg);
294 }
295
296 static irqreturn_t opal_message_notify(int irq, void *data)
297 {
298         opal_handle_message();
299         return IRQ_HANDLED;
300 }
301
302 static int __init opal_message_init(void)
303 {
304         int ret, i, irq;
305
306         for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
307                 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
308
309         irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
310         if (!irq) {
311                 pr_err("%s: Can't register OPAL event irq (%d)\n",
312                        __func__, irq);
313                 return irq;
314         }
315
316         ret = request_irq(irq, opal_message_notify,
317                         IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
318         if (ret) {
319                 pr_err("%s: Can't request OPAL event irq (%d)\n",
320                        __func__, ret);
321                 return ret;
322         }
323
324         return 0;
325 }
326
327 int opal_get_chars(uint32_t vtermno, char *buf, int count)
328 {
329         s64 rc;
330         __be64 evt, len;
331
332         if (!opal.entry)
333                 return -ENODEV;
334         opal_poll_events(&evt);
335         if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
336                 return 0;
337         len = cpu_to_be64(count);
338         rc = opal_console_read(vtermno, &len, buf);
339         if (rc == OPAL_SUCCESS)
340                 return be64_to_cpu(len);
341         return 0;
342 }
343
344 static int __opal_put_chars(uint32_t vtermno, const char *data, int total_len, bool atomic)
345 {
346         unsigned long flags = 0 /* shut up gcc */;
347         int written;
348         __be64 olen;
349         s64 rc;
350
351         if (!opal.entry)
352                 return -ENODEV;
353
354         if (atomic)
355                 spin_lock_irqsave(&opal_write_lock, flags);
356         rc = opal_console_write_buffer_space(vtermno, &olen);
357         if (rc || be64_to_cpu(olen) < total_len) {
358                 /* Closed -> drop characters */
359                 if (rc)
360                         written = total_len;
361                 else
362                         written = -EAGAIN;
363                 goto out;
364         }
365
366         /* Should not get a partial write here because space is available. */
367         olen = cpu_to_be64(total_len);
368         rc = opal_console_write(vtermno, &olen, data);
369         if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
370                 if (rc == OPAL_BUSY_EVENT)
371                         opal_poll_events(NULL);
372                 written = -EAGAIN;
373                 goto out;
374         }
375
376         /* Closed or other error drop */
377         if (rc != OPAL_SUCCESS) {
378                 written = opal_error_code(rc);
379                 goto out;
380         }
381
382         written = be64_to_cpu(olen);
383         if (written < total_len) {
384                 if (atomic) {
385                         /* Should not happen */
386                         pr_warn("atomic console write returned partial "
387                                 "len=%d written=%d\n", total_len, written);
388                 }
389                 if (!written)
390                         written = -EAGAIN;
391         }
392
393 out:
394         if (atomic)
395                 spin_unlock_irqrestore(&opal_write_lock, flags);
396
397         return written;
398 }
399
400 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
401 {
402         return __opal_put_chars(vtermno, data, total_len, false);
403 }
404
405 /*
406  * opal_put_chars_atomic will not perform partial-writes. Data will be
407  * atomically written to the terminal or not at all. This is not strictly
408  * true at the moment because console space can race with OPAL's console
409  * writes.
410  */
411 int opal_put_chars_atomic(uint32_t vtermno, const char *data, int total_len)
412 {
413         return __opal_put_chars(vtermno, data, total_len, true);
414 }
415
416 static s64 __opal_flush_console(uint32_t vtermno)
417 {
418         s64 rc;
419
420         if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
421                 __be64 evt;
422
423                 /*
424                  * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
425                  * the console can still be flushed by calling the polling
426                  * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
427                  */
428                 WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
429
430                 opal_poll_events(&evt);
431                 if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
432                         return OPAL_SUCCESS;
433                 return OPAL_BUSY;
434
435         } else {
436                 rc = opal_console_flush(vtermno);
437                 if (rc == OPAL_BUSY_EVENT) {
438                         opal_poll_events(NULL);
439                         rc = OPAL_BUSY;
440                 }
441                 return rc;
442         }
443
444 }
445
446 /*
447  * opal_flush_console spins until the console is flushed
448  */
449 int opal_flush_console(uint32_t vtermno)
450 {
451         for (;;) {
452                 s64 rc = __opal_flush_console(vtermno);
453
454                 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
455                         mdelay(1);
456                         continue;
457                 }
458
459                 return opal_error_code(rc);
460         }
461 }
462
463 /*
464  * opal_flush_chars is an hvc interface that sleeps until the console is
465  * flushed if wait, otherwise it will return -EBUSY if the console has data,
466  * -EAGAIN if it has data and some of it was flushed.
467  */
468 int opal_flush_chars(uint32_t vtermno, bool wait)
469 {
470         for (;;) {
471                 s64 rc = __opal_flush_console(vtermno);
472
473                 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
474                         if (wait) {
475                                 msleep(OPAL_BUSY_DELAY_MS);
476                                 continue;
477                         }
478                         if (rc == OPAL_PARTIAL)
479                                 return -EAGAIN;
480                 }
481
482                 return opal_error_code(rc);
483         }
484 }
485
486 static int opal_recover_mce(struct pt_regs *regs,
487                                         struct machine_check_event *evt)
488 {
489         int recovered = 0;
490
491         if (!(regs->msr & MSR_RI)) {
492                 /* If MSR_RI isn't set, we cannot recover */
493                 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
494                 recovered = 0;
495         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
496                 /* Platform corrected itself */
497                 recovered = 1;
498         } else if (evt->severity == MCE_SEV_FATAL) {
499                 /* Fatal machine check */
500                 pr_err("Machine check interrupt is fatal\n");
501                 recovered = 0;
502         }
503
504         if (!recovered && evt->sync_error) {
505                 /*
506                  * Try to kill processes if we get a synchronous machine check
507                  * (e.g., one caused by execution of this instruction). This
508                  * will devolve into a panic if we try to kill init or are in
509                  * an interrupt etc.
510                  *
511                  * TODO: Queue up this address for hwpoisioning later.
512                  * TODO: This is not quite right for d-side machine
513                  *       checks ->nip is not necessarily the important
514                  *       address.
515                  */
516                 if ((user_mode(regs))) {
517                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
518                         recovered = 1;
519                 } else if (die_will_crash()) {
520                         /*
521                          * die() would kill the kernel, so better to go via
522                          * the platform reboot code that will log the
523                          * machine check.
524                          */
525                         recovered = 0;
526                 } else {
527                         die("Machine check", regs, SIGBUS);
528                         recovered = 1;
529                 }
530         }
531
532         return recovered;
533 }
534
535 void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
536 {
537         panic_flush_kmsg_start();
538
539         pr_emerg("Hardware platform error: %s\n", msg);
540         if (regs)
541                 show_regs(regs);
542         smp_send_stop();
543
544         panic_flush_kmsg_end();
545
546         /*
547          * Don't bother to shut things down because this will
548          * xstop the system.
549          */
550         if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
551                                                 == OPAL_UNSUPPORTED) {
552                 pr_emerg("Reboot type %d not supported for %s\n",
553                                 OPAL_REBOOT_PLATFORM_ERROR, msg);
554         }
555
556         /*
557          * We reached here. There can be three possibilities:
558          * 1. We are running on a firmware level that do not support
559          *    opal_cec_reboot2()
560          * 2. We are running on a firmware level that do not support
561          *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
562          * 3. We are running on FSP based system that does not need
563          *    opal to trigger checkstop explicitly for error analysis.
564          *    The FSP PRD component would have already got notified
565          *    about this error through other channels.
566          * 4. We are running on a newer skiboot that by default does
567          *    not cause a checkstop, drops us back to the kernel to
568          *    extract context and state at the time of the error.
569          */
570
571         panic(msg);
572 }
573
574 int opal_machine_check(struct pt_regs *regs)
575 {
576         struct machine_check_event evt;
577
578         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
579                 return 0;
580
581         /* Print things out */
582         if (evt.version != MCE_V1) {
583                 pr_err("Machine Check Exception, Unknown event version %d !\n",
584                        evt.version);
585                 return 0;
586         }
587         machine_check_print_event_info(&evt, user_mode(regs), false);
588
589         if (opal_recover_mce(regs, &evt))
590                 return 1;
591
592         pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
593 }
594
595 /* Early hmi handler called in real mode. */
596 int opal_hmi_exception_early(struct pt_regs *regs)
597 {
598         s64 rc;
599
600         /*
601          * call opal hmi handler. Pass paca address as token.
602          * The return value OPAL_SUCCESS is an indication that there is
603          * an HMI event generated waiting to pull by Linux.
604          */
605         rc = opal_handle_hmi();
606         if (rc == OPAL_SUCCESS) {
607                 local_paca->hmi_event_available = 1;
608                 return 1;
609         }
610         return 0;
611 }
612
613 int opal_hmi_exception_early2(struct pt_regs *regs)
614 {
615         s64 rc;
616         __be64 out_flags;
617
618         /*
619          * call opal hmi handler.
620          * Check 64-bit flag mask to find out if an event was generated,
621          * and whether TB is still valid or not etc.
622          */
623         rc = opal_handle_hmi2(&out_flags);
624         if (rc != OPAL_SUCCESS)
625                 return 0;
626
627         if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_NEW_EVENT)
628                 local_paca->hmi_event_available = 1;
629         if (be64_to_cpu(out_flags) & OPAL_HMI_FLAGS_TOD_TB_FAIL)
630                 tb_invalid = true;
631         return 1;
632 }
633
634 /* HMI exception handler called in virtual mode during check_irq_replay. */
635 int opal_handle_hmi_exception(struct pt_regs *regs)
636 {
637         /*
638          * Check if HMI event is available.
639          * if Yes, then wake kopald to process them.
640          */
641         if (!local_paca->hmi_event_available)
642                 return 0;
643
644         local_paca->hmi_event_available = 0;
645         opal_wake_poller();
646
647         return 1;
648 }
649
650 static uint64_t find_recovery_address(uint64_t nip)
651 {
652         int i;
653
654         for (i = 0; i < mc_recoverable_range_len; i++)
655                 if ((nip >= mc_recoverable_range[i].start_addr) &&
656                     (nip < mc_recoverable_range[i].end_addr))
657                     return mc_recoverable_range[i].recover_addr;
658         return 0;
659 }
660
661 bool opal_mce_check_early_recovery(struct pt_regs *regs)
662 {
663         uint64_t recover_addr = 0;
664
665         if (!opal.base || !opal.size)
666                 goto out;
667
668         if ((regs->nip >= opal.base) &&
669                         (regs->nip < (opal.base + opal.size)))
670                 recover_addr = find_recovery_address(regs->nip);
671
672         /*
673          * Setup regs->nip to rfi into fixup address.
674          */
675         if (recover_addr)
676                 regs->nip = recover_addr;
677
678 out:
679         return !!recover_addr;
680 }
681
682 static int opal_sysfs_init(void)
683 {
684         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
685         if (!opal_kobj) {
686                 pr_warn("kobject_create_and_add opal failed\n");
687                 return -ENOMEM;
688         }
689
690         return 0;
691 }
692
693 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
694                                struct bin_attribute *bin_attr,
695                                char *buf, loff_t off, size_t count)
696 {
697         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
698                                        bin_attr->size);
699 }
700
701 static BIN_ATTR_RO(symbol_map, 0);
702
703 static void opal_export_symmap(void)
704 {
705         const __be64 *syms;
706         unsigned int size;
707         struct device_node *fw;
708         int rc;
709
710         fw = of_find_node_by_path("/ibm,opal/firmware");
711         if (!fw)
712                 return;
713         syms = of_get_property(fw, "symbol-map", &size);
714         if (!syms || size != 2 * sizeof(__be64))
715                 return;
716
717         /* Setup attributes */
718         bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
719         bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
720
721         rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
722         if (rc)
723                 pr_warn("Error %d creating OPAL symbols file\n", rc);
724 }
725
726 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
727                                 struct bin_attribute *bin_attr, char *buf,
728                                 loff_t off, size_t count)
729 {
730         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
731                                        bin_attr->size);
732 }
733
734 /*
735  * opal_export_attrs: creates a sysfs node for each property listed in
736  * the device-tree under /ibm,opal/firmware/exports/
737  * All new sysfs nodes are created under /opal/exports/.
738  * This allows for reserved memory regions (e.g. HDAT) to be read.
739  * The new sysfs nodes are only readable by root.
740  */
741 static void opal_export_attrs(void)
742 {
743         struct bin_attribute *attr;
744         struct device_node *np;
745         struct property *prop;
746         struct kobject *kobj;
747         u64 vals[2];
748         int rc;
749
750         np = of_find_node_by_path("/ibm,opal/firmware/exports");
751         if (!np)
752                 return;
753
754         /* Create new 'exports' directory - /sys/firmware/opal/exports */
755         kobj = kobject_create_and_add("exports", opal_kobj);
756         if (!kobj) {
757                 pr_warn("kobject_create_and_add() of exports failed\n");
758                 return;
759         }
760
761         for_each_property_of_node(np, prop) {
762                 if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
763                         continue;
764
765                 if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
766                         continue;
767
768                 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
769
770                 if (attr == NULL) {
771                         pr_warn("Failed kmalloc for bin_attribute!");
772                         continue;
773                 }
774
775                 sysfs_bin_attr_init(attr);
776                 attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
777                 attr->attr.mode = 0400;
778                 attr->read = export_attr_read;
779                 attr->private = __va(vals[0]);
780                 attr->size = vals[1];
781
782                 if (attr->attr.name == NULL) {
783                         pr_warn("Failed kstrdup for bin_attribute attr.name");
784                         kfree(attr);
785                         continue;
786                 }
787
788                 rc = sysfs_create_bin_file(kobj, attr);
789                 if (rc) {
790                         pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
791                                  rc, prop->name);
792                         kfree(attr->attr.name);
793                         kfree(attr);
794                 }
795         }
796
797         of_node_put(np);
798 }
799
800 static void __init opal_dump_region_init(void)
801 {
802         void *addr;
803         uint64_t size;
804         int rc;
805
806         if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
807                 return;
808
809         /* Register kernel log buffer */
810         addr = log_buf_addr_get();
811         if (addr == NULL)
812                 return;
813
814         size = log_buf_len_get();
815         if (size == 0)
816                 return;
817
818         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
819                                        __pa(addr), size);
820         /* Don't warn if this is just an older OPAL that doesn't
821          * know about that call
822          */
823         if (rc && rc != OPAL_UNSUPPORTED)
824                 pr_warn("DUMP: Failed to register kernel log buffer. "
825                         "rc = %d\n", rc);
826 }
827
828 static void opal_pdev_init(const char *compatible)
829 {
830         struct device_node *np;
831
832         for_each_compatible_node(np, NULL, compatible)
833                 of_platform_device_create(np, NULL, NULL);
834 }
835
836 static void __init opal_imc_init_dev(void)
837 {
838         struct device_node *np;
839
840         np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
841         if (np)
842                 of_platform_device_create(np, NULL, NULL);
843 }
844
845 static int kopald(void *unused)
846 {
847         unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
848
849         set_freezable();
850         do {
851                 try_to_freeze();
852
853                 opal_handle_events();
854
855                 set_current_state(TASK_INTERRUPTIBLE);
856                 if (opal_have_pending_events())
857                         __set_current_state(TASK_RUNNING);
858                 else
859                         schedule_timeout(timeout);
860
861         } while (!kthread_should_stop());
862
863         return 0;
864 }
865
866 void opal_wake_poller(void)
867 {
868         if (kopald_tsk)
869                 wake_up_process(kopald_tsk);
870 }
871
872 static void opal_init_heartbeat(void)
873 {
874         /* Old firwmware, we assume the HVC heartbeat is sufficient */
875         if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
876                                  &opal_heartbeat) != 0)
877                 opal_heartbeat = 0;
878
879         if (opal_heartbeat)
880                 kopald_tsk = kthread_run(kopald, NULL, "kopald");
881 }
882
883 static int __init opal_init(void)
884 {
885         struct device_node *np, *consoles, *leds;
886         int rc;
887
888         opal_node = of_find_node_by_path("/ibm,opal");
889         if (!opal_node) {
890                 pr_warn("Device node not found\n");
891                 return -ENODEV;
892         }
893
894         /* Register OPAL consoles if any ports */
895         consoles = of_find_node_by_path("/ibm,opal/consoles");
896         if (consoles) {
897                 for_each_child_of_node(consoles, np) {
898                         if (!of_node_name_eq(np, "serial"))
899                                 continue;
900                         of_platform_device_create(np, NULL, NULL);
901                 }
902                 of_node_put(consoles);
903         }
904
905         /* Initialise OPAL messaging system */
906         opal_message_init();
907
908         /* Initialise OPAL asynchronous completion interface */
909         opal_async_comp_init();
910
911         /* Initialise OPAL sensor interface */
912         opal_sensor_init();
913
914         /* Initialise OPAL hypervisor maintainence interrupt handling */
915         opal_hmi_handler_init();
916
917         /* Create i2c platform devices */
918         opal_pdev_init("ibm,opal-i2c");
919
920         /* Handle non-volatile memory devices */
921         opal_pdev_init("pmem-region");
922
923         /* Setup a heatbeat thread if requested by OPAL */
924         opal_init_heartbeat();
925
926         /* Detect In-Memory Collection counters and create devices*/
927         opal_imc_init_dev();
928
929         /* Create leds platform devices */
930         leds = of_find_node_by_path("/ibm,opal/leds");
931         if (leds) {
932                 of_platform_device_create(leds, "opal_leds", NULL);
933                 of_node_put(leds);
934         }
935
936         /* Initialise OPAL message log interface */
937         opal_msglog_init();
938
939         /* Create "opal" kobject under /sys/firmware */
940         rc = opal_sysfs_init();
941         if (rc == 0) {
942                 /* Export symbol map to userspace */
943                 opal_export_symmap();
944                 /* Setup dump region interface */
945                 opal_dump_region_init();
946                 /* Setup error log interface */
947                 rc = opal_elog_init();
948                 /* Setup code update interface */
949                 opal_flash_update_init();
950                 /* Setup platform dump extract interface */
951                 opal_platform_dump_init();
952                 /* Setup system parameters interface */
953                 opal_sys_param_init();
954                 /* Setup message log sysfs interface. */
955                 opal_msglog_sysfs_init();
956         }
957
958         /* Export all properties */
959         opal_export_attrs();
960
961         /* Initialize platform devices: IPMI backend, PRD & flash interface */
962         opal_pdev_init("ibm,opal-ipmi");
963         opal_pdev_init("ibm,opal-flash");
964         opal_pdev_init("ibm,opal-prd");
965
966         /* Initialise platform device: oppanel interface */
967         opal_pdev_init("ibm,opal-oppanel");
968
969         /* Initialise OPAL kmsg dumper for flushing console on panic */
970         opal_kmsg_init();
971
972         /* Initialise OPAL powercap interface */
973         opal_powercap_init();
974
975         /* Initialise OPAL Power-Shifting-Ratio interface */
976         opal_psr_init();
977
978         /* Initialise OPAL sensor groups */
979         opal_sensor_groups_init();
980
981         /* Initialise OPAL Power control interface */
982         opal_power_control_init();
983
984         return 0;
985 }
986 machine_subsys_initcall(powernv, opal_init);
987
988 void opal_shutdown(void)
989 {
990         long rc = OPAL_BUSY;
991
992         opal_event_shutdown();
993
994         /*
995          * Then sync with OPAL which ensure anything that can
996          * potentially write to our memory has completed such
997          * as an ongoing dump retrieval
998          */
999         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
1000                 rc = opal_sync_host_reboot();
1001                 if (rc == OPAL_BUSY)
1002                         opal_poll_events(NULL);
1003                 else
1004                         mdelay(10);
1005         }
1006
1007         /* Unregister memory dump region */
1008         if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
1009                 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
1010 }
1011
1012 /* Export this so that test modules can use it */
1013 EXPORT_SYMBOL_GPL(opal_invalid_call);
1014 EXPORT_SYMBOL_GPL(opal_xscom_read);
1015 EXPORT_SYMBOL_GPL(opal_xscom_write);
1016 EXPORT_SYMBOL_GPL(opal_ipmi_send);
1017 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
1018 EXPORT_SYMBOL_GPL(opal_flash_read);
1019 EXPORT_SYMBOL_GPL(opal_flash_write);
1020 EXPORT_SYMBOL_GPL(opal_flash_erase);
1021 EXPORT_SYMBOL_GPL(opal_prd_msg);
1022 EXPORT_SYMBOL_GPL(opal_check_token);
1023
1024 /* Convert a region of vmalloc memory to an opal sg list */
1025 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1026                                              unsigned long vmalloc_size)
1027 {
1028         struct opal_sg_list *sg, *first = NULL;
1029         unsigned long i = 0;
1030
1031         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1032         if (!sg)
1033                 goto nomem;
1034
1035         first = sg;
1036
1037         while (vmalloc_size > 0) {
1038                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1039                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
1040
1041                 sg->entry[i].data = cpu_to_be64(data);
1042                 sg->entry[i].length = cpu_to_be64(length);
1043                 i++;
1044
1045                 if (i >= SG_ENTRIES_PER_NODE) {
1046                         struct opal_sg_list *next;
1047
1048                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1049                         if (!next)
1050                                 goto nomem;
1051
1052                         sg->length = cpu_to_be64(
1053                                         i * sizeof(struct opal_sg_entry) + 16);
1054                         i = 0;
1055                         sg->next = cpu_to_be64(__pa(next));
1056                         sg = next;
1057                 }
1058
1059                 vmalloc_addr += length;
1060                 vmalloc_size -= length;
1061         }
1062
1063         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1064
1065         return first;
1066
1067 nomem:
1068         pr_err("%s : Failed to allocate memory\n", __func__);
1069         opal_free_sg_list(first);
1070         return NULL;
1071 }
1072
1073 void opal_free_sg_list(struct opal_sg_list *sg)
1074 {
1075         while (sg) {
1076                 uint64_t next = be64_to_cpu(sg->next);
1077
1078                 kfree(sg);
1079
1080                 if (next)
1081                         sg = __va(next);
1082                 else
1083                         sg = NULL;
1084         }
1085 }
1086
1087 int opal_error_code(int rc)
1088 {
1089         switch (rc) {
1090         case OPAL_SUCCESS:              return 0;
1091
1092         case OPAL_PARAMETER:            return -EINVAL;
1093         case OPAL_ASYNC_COMPLETION:     return -EINPROGRESS;
1094         case OPAL_BUSY:
1095         case OPAL_BUSY_EVENT:           return -EBUSY;
1096         case OPAL_NO_MEM:               return -ENOMEM;
1097         case OPAL_PERMISSION:           return -EPERM;
1098
1099         case OPAL_UNSUPPORTED:          return -EIO;
1100         case OPAL_HARDWARE:             return -EIO;
1101         case OPAL_INTERNAL_ERROR:       return -EIO;
1102         case OPAL_TIMEOUT:              return -ETIMEDOUT;
1103         default:
1104                 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1105                 return -EIO;
1106         }
1107 }
1108
1109 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1110 {
1111         int rc;
1112
1113         if (firmware_has_feature(FW_FEATURE_OPAL)) {
1114                 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1115                 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1116                         pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1117         }
1118 }
1119
1120 EXPORT_SYMBOL_GPL(opal_poll_events);
1121 EXPORT_SYMBOL_GPL(opal_rtc_read);
1122 EXPORT_SYMBOL_GPL(opal_rtc_write);
1123 EXPORT_SYMBOL_GPL(opal_tpo_read);
1124 EXPORT_SYMBOL_GPL(opal_tpo_write);
1125 EXPORT_SYMBOL_GPL(opal_i2c_request);
1126 /* Export these symbols for PowerNV LED class driver */
1127 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1128 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1129 /* Export this symbol for PowerNV Operator Panel class driver */
1130 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1131 /* Export this for KVM */
1132 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1133 EXPORT_SYMBOL_GPL(opal_int_eoi);
1134 EXPORT_SYMBOL_GPL(opal_error_code);
1135 /* Export the below symbol for NX compression */
1136 EXPORT_SYMBOL(opal_nx_coproc_init);