arch/powerpc/platforms/powernv/opal.c

   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 #undef DEBUG
  13
  14 #include <linux/types.h>
  15 #include <linux/of.h>
  16 #include <linux/of_fdt.h>
  17 #include <linux/of_platform.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/notifier.h>
  20 #include <linux/slab.h>
  21 #include <linux/sched.h>
  22 #include <linux/kobject.h>
  23 #include <linux/delay.h>
  24 #include <linux/memblock.h>
  25
  26 #include <asm/machdep.h>
  27 #include <asm/opal.h>
  28 #include <asm/firmware.h>
  29 #include <asm/mce.h>
  30
  31 #include "powernv.h"
  32
  33 /* /sys/firmware/opal */
  34 struct kobject *opal_kobj;
  35
  36 struct opal {
  37         u64 base;
  38         u64 entry;
  39         u64 size;
  40 } opal;
  41
  42 struct mcheck_recoverable_range {
  43         u64 start_addr;
  44         u64 end_addr;
  45         u64 recover_addr;
  46 };
  47
  48 static struct mcheck_recoverable_range *mc_recoverable_range;
  49 static int mc_recoverable_range_len;
  50
  51 struct device_node *opal_node;
  52 static DEFINE_SPINLOCK(opal_write_lock);
  53 extern u64 opal_mc_secondary_handler[];
  54 static unsigned int *opal_irqs;
  55 static unsigned int opal_irq_count;
  56 static ATOMIC_NOTIFIER_HEAD(opal_notifier_head);
  57 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
  58 static DEFINE_SPINLOCK(opal_notifier_lock);
  59 static uint64_t last_notified_mask = 0x0ul;
  60 static atomic_t opal_notifier_hold = ATOMIC_INIT(0);
  61
  62 static void opal_reinit_cores(void)
  63 {
  64         /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
  65          *
  66          * It will preserve non volatile GPRs and HSPRG0/1. It will
  67          * also restore HIDs and other SPRs to their original value
  68          * but it might clobber a bunch.
  69          */
  70 #ifdef __BIG_ENDIAN__
  71         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
  72 #else
  73         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
  74 #endif
  75 }
  76
  77 int __init early_init_dt_scan_opal(unsigned long node,
  78                                    const char *uname, int depth, void *data)
  79 {
  80         const void *basep, *entryp, *sizep;
  81         int basesz, entrysz, runtimesz;
  82
  83         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
  84                 return 0;
  85
  86         basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
  87         entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
  88         sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
  89
  90         if (!basep || !entryp || !sizep)
  91                 return 1;
  92
  93         opal.base = of_read_number(basep, basesz/4);
  94         opal.entry = of_read_number(entryp, entrysz/4);
  95         opal.size = of_read_number(sizep, runtimesz/4);
  96
  97         pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
  98                  opal.base, basep, basesz);
  99         pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
 100                  opal.entry, entryp, entrysz);
 101         pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
 102                  opal.size, sizep, runtimesz);
 103
 104         powerpc_firmware_features |= FW_FEATURE_OPAL;
 105         if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
 106                 powerpc_firmware_features |= FW_FEATURE_OPALv2;
 107                 powerpc_firmware_features |= FW_FEATURE_OPALv3;
 108                 pr_info("OPAL V3 detected !\n");
 109         } else if (of_flat_dt_is_compatible(node, "ibm,opal-v2")) {
 110                 powerpc_firmware_features |= FW_FEATURE_OPALv2;
 111                 pr_info("OPAL V2 detected !\n");
 112         } else {
 113                 pr_info("OPAL V1 detected !\n");
 114         }
 115
 116         /* Reinit all cores with the right endian */
 117         opal_reinit_cores();
 118
 119         /* Restore some bits */
 120         if (cur_cpu_spec->cpu_restore)
 121                 cur_cpu_spec->cpu_restore();
 122
 123         return 1;
 124 }
 125
 126 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
 127                                    const char *uname, int depth, void *data)
 128 {
 129         int i, psize, size;
 130         const __be32 *prop;
 131
 132         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
 133                 return 0;
 134
 135         prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
 136
 137         if (!prop)
 138                 return 1;
 139
 140         pr_debug("Found machine check recoverable ranges.\n");
 141
 142         /*
 143          * Calculate number of available entries.
 144          *
 145          * Each recoverable address range entry is (start address, len,
 146          * recovery address), 2 cells each for start and recovery address,
 147          * 1 cell for len, totalling 5 cells per entry.
 148          */
 149         mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
 150
 151         /* Sanity check */
 152         if (!mc_recoverable_range_len)
 153                 return 1;
 154
 155         /* Size required to hold all the entries. */
 156         size = mc_recoverable_range_len *
 157                         sizeof(struct mcheck_recoverable_range);
 158
 159         /*
 160          * Allocate a buffer to hold the MC recoverable ranges. We would be
 161          * accessing them in real mode, hence it needs to be within
 162          * RMO region.
 163          */
 164         mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
 165                                                         ppc64_rma_size));
 166         memset(mc_recoverable_range, 0, size);
 167
 168         for (i = 0; i < mc_recoverable_range_len; i++) {
 169                 mc_recoverable_range[i].start_addr =
 170                                         of_read_number(prop + (i * 5) + 0, 2);
 171                 mc_recoverable_range[i].end_addr =
 172                                         mc_recoverable_range[i].start_addr +
 173                                         of_read_number(prop + (i * 5) + 2, 1);
 174                 mc_recoverable_range[i].recover_addr =
 175                                         of_read_number(prop + (i * 5) + 3, 2);
 176
 177                 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
 178                                 mc_recoverable_range[i].start_addr,
 179                                 mc_recoverable_range[i].end_addr,
 180                                 mc_recoverable_range[i].recover_addr);
 181         }
 182         return 1;
 183 }
 184
 185 static int __init opal_register_exception_handlers(void)
 186 {
 187 #ifdef __BIG_ENDIAN__
 188         u64 glue;
 189
 190         if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
 191                 return -ENODEV;
 192
 193         /* Hookup some exception handlers except machine check. We use the
 194          * fwnmi area at 0x7000 to provide the glue space to OPAL
 195          */
 196         glue = 0x7000;
 197         opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
 198 #endif
 199
 200         return 0;
 201 }
 202 machine_early_initcall(powernv, opal_register_exception_handlers);
 203
 204 int opal_notifier_register(struct notifier_block *nb)
 205 {
 206         if (!nb) {
 207                 pr_warning("%s: Invalid argument (%p)\n",
 208                            __func__, nb);
 209                 return -EINVAL;
 210         }
 211
 212         atomic_notifier_chain_register(&opal_notifier_head, nb);
 213         return 0;
 214 }
 215 EXPORT_SYMBOL_GPL(opal_notifier_register);
 216
 217 int opal_notifier_unregister(struct notifier_block *nb)
 218 {
 219         if (!nb) {
 220                 pr_warning("%s: Invalid argument (%p)\n",
 221                            __func__, nb);
 222                 return -EINVAL;
 223         }
 224
 225         atomic_notifier_chain_unregister(&opal_notifier_head, nb);
 226         return 0;
 227 }
 228 EXPORT_SYMBOL_GPL(opal_notifier_unregister);
 229
 230 static void opal_do_notifier(uint64_t events)
 231 {
 232         unsigned long flags;
 233         uint64_t changed_mask;
 234
 235         if (atomic_read(&opal_notifier_hold))
 236                 return;
 237
 238         spin_lock_irqsave(&opal_notifier_lock, flags);
 239         changed_mask = last_notified_mask ^ events;
 240         last_notified_mask = events;
 241         spin_unlock_irqrestore(&opal_notifier_lock, flags);
 242
 243         /*
 244          * We feed with the event bits and changed bits for
 245          * enough information to the callback.
 246          */
 247         atomic_notifier_call_chain(&opal_notifier_head,
 248                                    events, (void *)changed_mask);
 249 }
 250
 251 void opal_notifier_update_evt(uint64_t evt_mask,
 252                               uint64_t evt_val)
 253 {
 254         unsigned long flags;
 255
 256         spin_lock_irqsave(&opal_notifier_lock, flags);
 257         last_notified_mask &= ~evt_mask;
 258         last_notified_mask |= evt_val;
 259         spin_unlock_irqrestore(&opal_notifier_lock, flags);
 260 }
 261
 262 void opal_notifier_enable(void)
 263 {
 264         int64_t rc;
 265         __be64 evt = 0;
 266
 267         atomic_set(&opal_notifier_hold, 0);
 268
 269         /* Process pending events */
 270         rc = opal_poll_events(&evt);
 271         if (rc == OPAL_SUCCESS && evt)
 272                 opal_do_notifier(be64_to_cpu(evt));
 273 }
 274
 275 void opal_notifier_disable(void)
 276 {
 277         atomic_set(&opal_notifier_hold, 1);
 278 }
 279
 280 /*
 281  * Opal message notifier based on message type. Allow subscribers to get
 282  * notified for specific messgae type.
 283  */
 284 int opal_message_notifier_register(enum OpalMessageType msg_type,
 285                                         struct notifier_block *nb)
 286 {
 287         if (!nb) {
 288                 pr_warning("%s: Invalid argument (%p)\n",
 289                            __func__, nb);
 290                 return -EINVAL;
 291         }
 292         if (msg_type > OPAL_MSG_TYPE_MAX) {
 293                 pr_warning("%s: Invalid message type argument (%d)\n",
 294                            __func__, msg_type);
 295                 return -EINVAL;
 296         }
 297         return atomic_notifier_chain_register(
 298                                 &opal_msg_notifier_head[msg_type], nb);
 299 }
 300
 301 static void opal_message_do_notify(uint32_t msg_type, void *msg)
 302 {
 303         /* notify subscribers */
 304         atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
 305                                         msg_type, msg);
 306 }
 307
 308 static void opal_handle_message(void)
 309 {
 310         s64 ret;
 311         /*
 312          * TODO: pre-allocate a message buffer depending on opal-msg-size
 313          * value in /proc/device-tree.
 314          */
 315         static struct opal_msg msg;
 316         u32 type;
 317
 318         ret = opal_get_msg(__pa(&msg), sizeof(msg));
 319         /* No opal message pending. */
 320         if (ret == OPAL_RESOURCE)
 321                 return;
 322
 323         /* check for errors. */
 324         if (ret) {
 325                 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
 326                                 __func__, ret);
 327                 return;
 328         }
 329
 330         type = be32_to_cpu(msg.msg_type);
 331
 332         /* Sanity check */
 333         if (type > OPAL_MSG_TYPE_MAX) {
 334                 pr_warning("%s: Unknown message type: %u\n", __func__, type);
 335                 return;
 336         }
 337         opal_message_do_notify(type, (void *)&msg);
 338 }
 339
 340 static int opal_message_notify(struct notifier_block *nb,
 341                           unsigned long events, void *change)
 342 {
 343         if (events & OPAL_EVENT_MSG_PENDING)
 344                 opal_handle_message();
 345         return 0;
 346 }
 347
 348 static struct notifier_block opal_message_nb = {
 349         .notifier_call  = opal_message_notify,
 350         .next           = NULL,
 351         .priority       = 0,
 352 };
 353
 354 static int __init opal_message_init(void)
 355 {
 356         int ret, i;
 357
 358         for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
 359                 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
 360
 361         ret = opal_notifier_register(&opal_message_nb);
 362         if (ret) {
 363                 pr_err("%s: Can't register OPAL event notifier (%d)\n",
 364                        __func__, ret);
 365                 return ret;
 366         }
 367         return 0;
 368 }
 369 machine_early_initcall(powernv, opal_message_init);
 370
 371 int opal_get_chars(uint32_t vtermno, char *buf, int count)
 372 {
 373         s64 rc;
 374         __be64 evt, len;
 375
 376         if (!opal.entry)
 377                 return -ENODEV;
 378         opal_poll_events(&evt);
 379         if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
 380                 return 0;
 381         len = cpu_to_be64(count);
 382         rc = opal_console_read(vtermno, &len, buf);
 383         if (rc == OPAL_SUCCESS)
 384                 return be64_to_cpu(len);
 385         return 0;
 386 }
 387
 388 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
 389 {
 390         int written = 0;
 391         __be64 olen;
 392         s64 len, rc;
 393         unsigned long flags;
 394         __be64 evt;
 395
 396         if (!opal.entry)
 397                 return -ENODEV;
 398
 399         /* We want put_chars to be atomic to avoid mangling of hvsi
 400          * packets. To do that, we first test for room and return
 401          * -EAGAIN if there isn't enough.
 402          *
 403          * Unfortunately, opal_console_write_buffer_space() doesn't
 404          * appear to work on opal v1, so we just assume there is
 405          * enough room and be done with it
 406          */
 407         spin_lock_irqsave(&opal_write_lock, flags);
 408         if (firmware_has_feature(FW_FEATURE_OPALv2)) {
 409                 rc = opal_console_write_buffer_space(vtermno, &olen);
 410                 len = be64_to_cpu(olen);
 411                 if (rc || len < total_len) {
 412                         spin_unlock_irqrestore(&opal_write_lock, flags);
 413                         /* Closed -> drop characters */
 414                         if (rc)
 415                                 return total_len;
 416                         opal_poll_events(NULL);
 417                         return -EAGAIN;
 418                 }
 419         }
 420
 421         /* We still try to handle partial completions, though they
 422          * should no longer happen.
 423          */
 424         rc = OPAL_BUSY;
 425         while(total_len > 0 && (rc == OPAL_BUSY ||
 426                                 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
 427                 olen = cpu_to_be64(total_len);
 428                 rc = opal_console_write(vtermno, &olen, data);
 429                 len = be64_to_cpu(olen);
 430
 431                 /* Closed or other error drop */
 432                 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
 433                     rc != OPAL_BUSY_EVENT) {
 434                         written = total_len;
 435                         break;
 436                 }
 437                 if (rc == OPAL_SUCCESS) {
 438                         total_len -= len;
 439                         data += len;
 440                         written += len;
 441                 }
 442                 /* This is a bit nasty but we need that for the console to
 443                  * flush when there aren't any interrupts. We will clean
 444                  * things a bit later to limit that to synchronous path
 445                  * such as the kernel console and xmon/udbg
 446                  */
 447                 do
 448                         opal_poll_events(&evt);
 449                 while(rc == OPAL_SUCCESS &&
 450                         (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
 451         }
 452         spin_unlock_irqrestore(&opal_write_lock, flags);
 453         return written;
 454 }
 455
 456 static int opal_recover_mce(struct pt_regs *regs,
 457                                         struct machine_check_event *evt)
 458 {
 459         int recovered = 0;
 460         uint64_t ea = get_mce_fault_addr(evt);
 461
 462         if (!(regs->msr & MSR_RI)) {
 463                 /* If MSR_RI isn't set, we cannot recover */
 464                 recovered = 0;
 465         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
 466                 /* Platform corrected itself */
 467                 recovered = 1;
 468         } else if (ea && !is_kernel_addr(ea)) {
 469                 /*
 470                  * Faulting address is not in kernel text. We should be fine.
 471                  * We need to find which process uses this address.
 472                  * For now, kill the task if we have received exception when
 473                  * in userspace.
 474                  *
 475                  * TODO: Queue up this address for hwpoisioning later.
 476                  */
 477                 if (user_mode(regs) && !is_global_init(current)) {
 478                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 479                         recovered = 1;
 480                 } else
 481                         recovered = 0;
 482         } else if (user_mode(regs) && !is_global_init(current) &&
 483                 evt->severity == MCE_SEV_ERROR_SYNC) {
 484                 /*
 485                  * If we have received a synchronous error when in userspace
 486                  * kill the task.
 487                  */
 488                 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 489                 recovered = 1;
 490         }
 491         return recovered;
 492 }
 493
 494 int opal_machine_check(struct pt_regs *regs)
 495 {
 496         struct machine_check_event evt;
 497
 498         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 499                 return 0;
 500
 501         /* Print things out */
 502         if (evt.version != MCE_V1) {
 503                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 504                        evt.version);
 505                 return 0;
 506         }
 507         machine_check_print_event_info(&evt);
 508
 509         if (opal_recover_mce(regs, &evt))
 510                 return 1;
 511         return 0;
 512 }
 513
 514 /* Early hmi handler called in real mode. */
 515 int opal_hmi_exception_early(struct pt_regs *regs)
 516 {
 517         s64 rc;
 518
 519         /*
 520          * call opal hmi handler. Pass paca address as token.
 521          * The return value OPAL_SUCCESS is an indication that there is
 522          * an HMI event generated waiting to pull by Linux.
 523          */
 524         rc = opal_handle_hmi();
 525         if (rc == OPAL_SUCCESS) {
 526                 local_paca->hmi_event_available = 1;
 527                 return 1;
 528         }
 529         return 0;
 530 }
 531
 532 /* HMI exception handler called in virtual mode during check_irq_replay. */
 533 int opal_handle_hmi_exception(struct pt_regs *regs)
 534 {
 535         s64 rc;
 536         __be64 evt = 0;
 537
 538         /*
 539          * Check if HMI event is available.
 540          * if Yes, then call opal_poll_events to pull opal messages and
 541          * process them.
 542          */
 543         if (!local_paca->hmi_event_available)
 544                 return 0;
 545
 546         local_paca->hmi_event_available = 0;
 547         rc = opal_poll_events(&evt);
 548         if (rc == OPAL_SUCCESS && evt)
 549                 opal_do_notifier(be64_to_cpu(evt));
 550
 551         return 1;
 552 }
 553
 554 static uint64_t find_recovery_address(uint64_t nip)
 555 {
 556         int i;
 557
 558         for (i = 0; i < mc_recoverable_range_len; i++)
 559                 if ((nip >= mc_recoverable_range[i].start_addr) &&
 560                     (nip < mc_recoverable_range[i].end_addr))
 561                     return mc_recoverable_range[i].recover_addr;
 562         return 0;
 563 }
 564
 565 bool opal_mce_check_early_recovery(struct pt_regs *regs)
 566 {
 567         uint64_t recover_addr = 0;
 568
 569         if (!opal.base || !opal.size)
 570                 goto out;
 571
 572         if ((regs->nip >= opal.base) &&
 573                         (regs->nip <= (opal.base + opal.size)))
 574                 recover_addr = find_recovery_address(regs->nip);
 575
 576         /*
 577          * Setup regs->nip to rfi into fixup address.
 578          */
 579         if (recover_addr)
 580                 regs->nip = recover_addr;
 581
 582 out:
 583         return !!recover_addr;
 584 }
 585
 586 static irqreturn_t opal_interrupt(int irq, void *data)
 587 {
 588         __be64 events;
 589
 590         opal_handle_interrupt(virq_to_hw(irq), &events);
 591
 592         opal_do_notifier(be64_to_cpu(events));
 593
 594         return IRQ_HANDLED;
 595 }
 596
 597 static int opal_sysfs_init(void)
 598 {
 599         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
 600         if (!opal_kobj) {
 601                 pr_warn("kobject_create_and_add opal failed\n");
 602                 return -ENOMEM;
 603         }
 604
 605         return 0;
 606 }
 607
 608 static void __init opal_dump_region_init(void)
 609 {
 610         void *addr;
 611         uint64_t size;
 612         int rc;
 613
 614         /* Register kernel log buffer */
 615         addr = log_buf_addr_get();
 616         size = log_buf_len_get();
 617         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
 618                                        __pa(addr), size);
 619         /* Don't warn if this is just an older OPAL that doesn't
 620          * know about that call
 621          */
 622         if (rc && rc != OPAL_UNSUPPORTED)
 623                 pr_warn("DUMP: Failed to register kernel log buffer. "
 624                         "rc = %d\n", rc);
 625 }
 626 static int __init opal_init(void)
 627 {
 628         struct device_node *np, *consoles;
 629         const __be32 *irqs;
 630         int rc, i, irqlen;
 631
 632         opal_node = of_find_node_by_path("/ibm,opal");
 633         if (!opal_node) {
 634                 pr_warn("opal: Node not found\n");
 635                 return -ENODEV;
 636         }
 637
 638         /* Register OPAL consoles if any ports */
 639         if (firmware_has_feature(FW_FEATURE_OPALv2))
 640                 consoles = of_find_node_by_path("/ibm,opal/consoles");
 641         else
 642                 consoles = of_node_get(opal_node);
 643         if (consoles) {
 644                 for_each_child_of_node(consoles, np) {
 645                         if (strcmp(np->name, "serial"))
 646                                 continue;
 647                         of_platform_device_create(np, NULL, NULL);
 648                 }
 649                 of_node_put(consoles);
 650         }
 651
 652         /* Find all OPAL interrupts and request them */
 653         irqs = of_get_property(opal_node, "opal-interrupts", &irqlen);
 654         pr_debug("opal: Found %d interrupts reserved for OPAL\n",
 655                  irqs ? (irqlen / 4) : 0);
 656         opal_irq_count = irqlen / 4;
 657         opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL);
 658         for (i = 0; irqs && i < (irqlen / 4); i++, irqs++) {
 659                 unsigned int hwirq = be32_to_cpup(irqs);
 660                 unsigned int irq = irq_create_mapping(NULL, hwirq);
 661                 if (irq == NO_IRQ) {
 662                         pr_warning("opal: Failed to map irq 0x%x\n", hwirq);
 663                         continue;
 664                 }
 665                 rc = request_irq(irq, opal_interrupt, 0, "opal", NULL);
 666                 if (rc)
 667                         pr_warning("opal: Error %d requesting irq %d"
 668                                    " (0x%x)\n", rc, irq, hwirq);
 669                 opal_irqs[i] = irq;
 670         }
 671
 672         /* Create "opal" kobject under /sys/firmware */
 673         rc = opal_sysfs_init();
 674         if (rc == 0) {
 675                 /* Setup dump region interface */
 676                 opal_dump_region_init();
 677                 /* Setup error log interface */
 678                 rc = opal_elog_init();
 679                 /* Setup code update interface */
 680                 opal_flash_init();
 681                 /* Setup platform dump extract interface */
 682                 opal_platform_dump_init();
 683                 /* Setup system parameters interface */
 684                 opal_sys_param_init();
 685                 /* Setup message log interface. */
 686                 opal_msglog_init();
 687         }
 688
 689         return 0;
 690 }
 691 machine_subsys_initcall(powernv, opal_init);
 692
 693 void opal_shutdown(void)
 694 {
 695         unsigned int i;
 696         long rc = OPAL_BUSY;
 697
 698         /* First free interrupts, which will also mask them */
 699         for (i = 0; i < opal_irq_count; i++) {
 700                 if (opal_irqs[i])
 701                         free_irq(opal_irqs[i], NULL);
 702                 opal_irqs[i] = 0;
 703         }
 704
 705         /*
 706          * Then sync with OPAL which ensure anything that can
 707          * potentially write to our memory has completed such
 708          * as an ongoing dump retrieval
 709          */
 710         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 711                 rc = opal_sync_host_reboot();
 712                 if (rc == OPAL_BUSY)
 713                         opal_poll_events(NULL);
 714                 else
 715                         mdelay(10);
 716         }
 717
 718         /* Unregister memory dump region */
 719         opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
 720 }
 721
 722 /* Export this so that test modules can use it */
 723 EXPORT_SYMBOL_GPL(opal_invalid_call);
 724
 725 /* Convert a region of vmalloc memory to an opal sg list */
 726 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
 727                                              unsigned long vmalloc_size)
 728 {
 729         struct opal_sg_list *sg, *first = NULL;
 730         unsigned long i = 0;
 731
 732         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
 733         if (!sg)
 734                 goto nomem;
 735
 736         first = sg;
 737
 738         while (vmalloc_size > 0) {
 739                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
 740                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
 741
 742                 sg->entry[i].data = cpu_to_be64(data);
 743                 sg->entry[i].length = cpu_to_be64(length);
 744                 i++;
 745
 746                 if (i >= SG_ENTRIES_PER_NODE) {
 747                         struct opal_sg_list *next;
 748
 749                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
 750                         if (!next)
 751                                 goto nomem;
 752
 753                         sg->length = cpu_to_be64(
 754                                         i * sizeof(struct opal_sg_entry) + 16);
 755                         i = 0;
 756                         sg->next = cpu_to_be64(__pa(next));
 757                         sg = next;
 758                 }
 759
 760                 vmalloc_addr += length;
 761                 vmalloc_size -= length;
 762         }
 763
 764         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
 765
 766         return first;
 767
 768 nomem:
 769         pr_err("%s : Failed to allocate memory\n", __func__);
 770         opal_free_sg_list(first);
 771         return NULL;
 772 }
 773
 774 void opal_free_sg_list(struct opal_sg_list *sg)
 775 {
 776         while (sg) {
 777                 uint64_t next = be64_to_cpu(sg->next);
 778
 779                 kfree(sg);
 780
 781                 if (next)
 782                         sg = __va(next);
 783                 else
 784                         sg = NULL;
 785         }
 786 }