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 #define pr_fmt(fmt)     "opal: " fmt
  13
  14 #include <linux/printk.h>
  15 #include <linux/types.h>
  16 #include <linux/of.h>
  17 #include <linux/of_fdt.h>
  18 #include <linux/of_platform.h>
  19 #include <linux/interrupt.h>
  20 #include <linux/notifier.h>
  21 #include <linux/slab.h>
  22 #include <linux/sched.h>
  23 #include <linux/kobject.h>
  24 #include <linux/delay.h>
  25 #include <linux/memblock.h>
  26 #include <linux/kthread.h>
  27 #include <linux/freezer.h>
  28
  29 #include <asm/machdep.h>
  30 #include <asm/opal.h>
  31 #include <asm/firmware.h>
  32 #include <asm/mce.h>
  33
  34 #include "powernv.h"
  35
  36 /* /sys/firmware/opal */
  37 struct kobject *opal_kobj;
  38
  39 struct opal {
  40         u64 base;
  41         u64 entry;
  42         u64 size;
  43 } opal;
  44
  45 struct mcheck_recoverable_range {
  46         u64 start_addr;
  47         u64 end_addr;
  48         u64 recover_addr;
  49 };
  50
  51 static struct mcheck_recoverable_range *mc_recoverable_range;
  52 static int mc_recoverable_range_len;
  53
  54 struct device_node *opal_node;
  55 static DEFINE_SPINLOCK(opal_write_lock);
  56 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
  57 static uint32_t opal_heartbeat;
  58
  59 static void opal_reinit_cores(void)
  60 {
  61         /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
  62          *
  63          * It will preserve non volatile GPRs and HSPRG0/1. It will
  64          * also restore HIDs and other SPRs to their original value
  65          * but it might clobber a bunch.
  66          */
  67 #ifdef __BIG_ENDIAN__
  68         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
  69 #else
  70         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
  71 #endif
  72 }
  73
  74 int __init early_init_dt_scan_opal(unsigned long node,
  75                                    const char *uname, int depth, void *data)
  76 {
  77         const void *basep, *entryp, *sizep;
  78         int basesz, entrysz, runtimesz;
  79
  80         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
  81                 return 0;
  82
  83         basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
  84         entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
  85         sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
  86
  87         if (!basep || !entryp || !sizep)
  88                 return 1;
  89
  90         opal.base = of_read_number(basep, basesz/4);
  91         opal.entry = of_read_number(entryp, entrysz/4);
  92         opal.size = of_read_number(sizep, runtimesz/4);
  93
  94         pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
  95                  opal.base, basep, basesz);
  96         pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
  97                  opal.entry, entryp, entrysz);
  98         pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
  99                  opal.size, sizep, runtimesz);
 100
 101         if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
 102                 powerpc_firmware_features |= FW_FEATURE_OPAL;
 103                 pr_info("OPAL detected !\n");
 104         } else {
 105                 panic("OPAL != V3 detected, no longer supported.\n");
 106         }
 107
 108         /* Reinit all cores with the right endian */
 109         opal_reinit_cores();
 110
 111         /* Restore some bits */
 112         if (cur_cpu_spec->cpu_restore)
 113                 cur_cpu_spec->cpu_restore();
 114
 115         return 1;
 116 }
 117
 118 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
 119                                    const char *uname, int depth, void *data)
 120 {
 121         int i, psize, size;
 122         const __be32 *prop;
 123
 124         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
 125                 return 0;
 126
 127         prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
 128
 129         if (!prop)
 130                 return 1;
 131
 132         pr_debug("Found machine check recoverable ranges.\n");
 133
 134         /*
 135          * Calculate number of available entries.
 136          *
 137          * Each recoverable address range entry is (start address, len,
 138          * recovery address), 2 cells each for start and recovery address,
 139          * 1 cell for len, totalling 5 cells per entry.
 140          */
 141         mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
 142
 143         /* Sanity check */
 144         if (!mc_recoverable_range_len)
 145                 return 1;
 146
 147         /* Size required to hold all the entries. */
 148         size = mc_recoverable_range_len *
 149                         sizeof(struct mcheck_recoverable_range);
 150
 151         /*
 152          * Allocate a buffer to hold the MC recoverable ranges. We would be
 153          * accessing them in real mode, hence it needs to be within
 154          * RMO region.
 155          */
 156         mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
 157                                                         ppc64_rma_size));
 158         memset(mc_recoverable_range, 0, size);
 159
 160         for (i = 0; i < mc_recoverable_range_len; i++) {
 161                 mc_recoverable_range[i].start_addr =
 162                                         of_read_number(prop + (i * 5) + 0, 2);
 163                 mc_recoverable_range[i].end_addr =
 164                                         mc_recoverable_range[i].start_addr +
 165                                         of_read_number(prop + (i * 5) + 2, 1);
 166                 mc_recoverable_range[i].recover_addr =
 167                                         of_read_number(prop + (i * 5) + 3, 2);
 168
 169                 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
 170                                 mc_recoverable_range[i].start_addr,
 171                                 mc_recoverable_range[i].end_addr,
 172                                 mc_recoverable_range[i].recover_addr);
 173         }
 174         return 1;
 175 }
 176
 177 static int __init opal_register_exception_handlers(void)
 178 {
 179 #ifdef __BIG_ENDIAN__
 180         u64 glue;
 181
 182         if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
 183                 return -ENODEV;
 184
 185         /* Hookup some exception handlers except machine check. We use the
 186          * fwnmi area at 0x7000 to provide the glue space to OPAL
 187          */
 188         glue = 0x7000;
 189
 190         /*
 191          * Check if we are running on newer firmware that exports
 192          * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
 193          * the HMI interrupt and we catch it directly in Linux.
 194          *
 195          * For older firmware (i.e currently released POWER8 System Firmware
 196          * as of today <= SV810_087), we fallback to old behavior and let OPAL
 197          * patch the HMI vector and handle it inside OPAL firmware.
 198          *
 199          * For newer firmware (in development/yet to be released) we will
 200          * start catching/handling HMI directly in Linux.
 201          */
 202         if (!opal_check_token(OPAL_HANDLE_HMI)) {
 203                 pr_info("Old firmware detected, OPAL handles HMIs.\n");
 204                 opal_register_exception_handler(
 205                                 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
 206                                 0, glue);
 207                 glue += 128;
 208         }
 209
 210         opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
 211 #endif
 212
 213         return 0;
 214 }
 215 machine_early_initcall(powernv, opal_register_exception_handlers);
 216
 217 /*
 218  * Opal message notifier based on message type. Allow subscribers to get
 219  * notified for specific messgae type.
 220  */
 221 int opal_message_notifier_register(enum opal_msg_type msg_type,
 222                                         struct notifier_block *nb)
 223 {
 224         if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
 225                 pr_warning("%s: Invalid arguments, msg_type:%d\n",
 226                            __func__, msg_type);
 227                 return -EINVAL;
 228         }
 229
 230         return atomic_notifier_chain_register(
 231                                 &opal_msg_notifier_head[msg_type], nb);
 232 }
 233 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
 234
 235 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
 236                                      struct notifier_block *nb)
 237 {
 238         return atomic_notifier_chain_unregister(
 239                         &opal_msg_notifier_head[msg_type], nb);
 240 }
 241 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
 242
 243 static void opal_message_do_notify(uint32_t msg_type, void *msg)
 244 {
 245         /* notify subscribers */
 246         atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
 247                                         msg_type, msg);
 248 }
 249
 250 static void opal_handle_message(void)
 251 {
 252         s64 ret;
 253         /*
 254          * TODO: pre-allocate a message buffer depending on opal-msg-size
 255          * value in /proc/device-tree.
 256          */
 257         static struct opal_msg msg;
 258         u32 type;
 259
 260         ret = opal_get_msg(__pa(&msg), sizeof(msg));
 261         /* No opal message pending. */
 262         if (ret == OPAL_RESOURCE)
 263                 return;
 264
 265         /* check for errors. */
 266         if (ret) {
 267                 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
 268                                 __func__, ret);
 269                 return;
 270         }
 271
 272         type = be32_to_cpu(msg.msg_type);
 273
 274         /* Sanity check */
 275         if (type >= OPAL_MSG_TYPE_MAX) {
 276                 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
 277                 return;
 278         }
 279         opal_message_do_notify(type, (void *)&msg);
 280 }
 281
 282 static irqreturn_t opal_message_notify(int irq, void *data)
 283 {
 284         opal_handle_message();
 285         return IRQ_HANDLED;
 286 }
 287
 288 static int __init opal_message_init(void)
 289 {
 290         int ret, i, irq;
 291
 292         for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
 293                 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
 294
 295         irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
 296         if (!irq) {
 297                 pr_err("%s: Can't register OPAL event irq (%d)\n",
 298                        __func__, irq);
 299                 return irq;
 300         }
 301
 302         ret = request_irq(irq, opal_message_notify,
 303                         IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
 304         if (ret) {
 305                 pr_err("%s: Can't request OPAL event irq (%d)\n",
 306                        __func__, ret);
 307                 return ret;
 308         }
 309
 310         return 0;
 311 }
 312
 313 int opal_get_chars(uint32_t vtermno, char *buf, int count)
 314 {
 315         s64 rc;
 316         __be64 evt, len;
 317
 318         if (!opal.entry)
 319                 return -ENODEV;
 320         opal_poll_events(&evt);
 321         if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
 322                 return 0;
 323         len = cpu_to_be64(count);
 324         rc = opal_console_read(vtermno, &len, buf);
 325         if (rc == OPAL_SUCCESS)
 326                 return be64_to_cpu(len);
 327         return 0;
 328 }
 329
 330 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
 331 {
 332         int written = 0;
 333         __be64 olen;
 334         s64 len, rc;
 335         unsigned long flags;
 336         __be64 evt;
 337
 338         if (!opal.entry)
 339                 return -ENODEV;
 340
 341         /* We want put_chars to be atomic to avoid mangling of hvsi
 342          * packets. To do that, we first test for room and return
 343          * -EAGAIN if there isn't enough.
 344          *
 345          * Unfortunately, opal_console_write_buffer_space() doesn't
 346          * appear to work on opal v1, so we just assume there is
 347          * enough room and be done with it
 348          */
 349         spin_lock_irqsave(&opal_write_lock, flags);
 350         rc = opal_console_write_buffer_space(vtermno, &olen);
 351         len = be64_to_cpu(olen);
 352         if (rc || len < total_len) {
 353                 spin_unlock_irqrestore(&opal_write_lock, flags);
 354                 /* Closed -> drop characters */
 355                 if (rc)
 356                         return total_len;
 357                 opal_poll_events(NULL);
 358                 return -EAGAIN;
 359         }
 360
 361         /* We still try to handle partial completions, though they
 362          * should no longer happen.
 363          */
 364         rc = OPAL_BUSY;
 365         while(total_len > 0 && (rc == OPAL_BUSY ||
 366                                 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
 367                 olen = cpu_to_be64(total_len);
 368                 rc = opal_console_write(vtermno, &olen, data);
 369                 len = be64_to_cpu(olen);
 370
 371                 /* Closed or other error drop */
 372                 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
 373                     rc != OPAL_BUSY_EVENT) {
 374                         written += total_len;
 375                         break;
 376                 }
 377                 if (rc == OPAL_SUCCESS) {
 378                         total_len -= len;
 379                         data += len;
 380                         written += len;
 381                 }
 382                 /* This is a bit nasty but we need that for the console to
 383                  * flush when there aren't any interrupts. We will clean
 384                  * things a bit later to limit that to synchronous path
 385                  * such as the kernel console and xmon/udbg
 386                  */
 387                 do
 388                         opal_poll_events(&evt);
 389                 while(rc == OPAL_SUCCESS &&
 390                         (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
 391         }
 392         spin_unlock_irqrestore(&opal_write_lock, flags);
 393         return written;
 394 }
 395
 396 static int opal_recover_mce(struct pt_regs *regs,
 397                                         struct machine_check_event *evt)
 398 {
 399         int recovered = 0;
 400         uint64_t ea = get_mce_fault_addr(evt);
 401
 402         if (!(regs->msr & MSR_RI)) {
 403                 /* If MSR_RI isn't set, we cannot recover */
 404                 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
 405                 recovered = 0;
 406         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
 407                 /* Platform corrected itself */
 408                 recovered = 1;
 409         } else if (ea && !is_kernel_addr(ea)) {
 410                 /*
 411                  * Faulting address is not in kernel text. We should be fine.
 412                  * We need to find which process uses this address.
 413                  * For now, kill the task if we have received exception when
 414                  * in userspace.
 415                  *
 416                  * TODO: Queue up this address for hwpoisioning later.
 417                  */
 418                 if (user_mode(regs) && !is_global_init(current)) {
 419                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 420                         recovered = 1;
 421                 } else
 422                         recovered = 0;
 423         } else if (user_mode(regs) && !is_global_init(current) &&
 424                 evt->severity == MCE_SEV_ERROR_SYNC) {
 425                 /*
 426                  * If we have received a synchronous error when in userspace
 427                  * kill the task.
 428                  */
 429                 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 430                 recovered = 1;
 431         }
 432         return recovered;
 433 }
 434
 435 int opal_machine_check(struct pt_regs *regs)
 436 {
 437         struct machine_check_event evt;
 438         int ret;
 439
 440         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 441                 return 0;
 442
 443         /* Print things out */
 444         if (evt.version != MCE_V1) {
 445                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 446                        evt.version);
 447                 return 0;
 448         }
 449         machine_check_print_event_info(&evt);
 450
 451         if (opal_recover_mce(regs, &evt))
 452                 return 1;
 453
 454         /*
 455          * Unrecovered machine check, we are heading to panic path.
 456          *
 457          * We may have hit this MCE in very early stage of kernel
 458          * initialization even before opal-prd has started running. If
 459          * this is the case then this MCE error may go un-noticed or
 460          * un-analyzed if we go down panic path. We need to inform
 461          * BMC/OCC about this error so that they can collect relevant
 462          * data for error analysis before rebooting.
 463          * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
 464          * This function may not return on BMC based system.
 465          */
 466         ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
 467                         "Unrecoverable Machine Check exception");
 468         if (ret == OPAL_UNSUPPORTED) {
 469                 pr_emerg("Reboot type %d not supported\n",
 470                                         OPAL_REBOOT_PLATFORM_ERROR);
 471         }
 472
 473         /*
 474          * We reached here. There can be three possibilities:
 475          * 1. We are running on a firmware level that do not support
 476          *    opal_cec_reboot2()
 477          * 2. We are running on a firmware level that do not support
 478          *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
 479          * 3. We are running on FSP based system that does not need opal
 480          *    to trigger checkstop explicitly for error analysis. The FSP
 481          *    PRD component would have already got notified about this
 482          *    error through other channels.
 483          *
 484          * If hardware marked this as an unrecoverable MCE, we are
 485          * going to panic anyway. Even if it didn't, it's not safe to
 486          * continue at this point, so we should explicitly panic.
 487          */
 488
 489         panic("PowerNV Unrecovered Machine Check");
 490         return 0;
 491 }
 492
 493 /* Early hmi handler called in real mode. */
 494 int opal_hmi_exception_early(struct pt_regs *regs)
 495 {
 496         s64 rc;
 497
 498         /*
 499          * call opal hmi handler. Pass paca address as token.
 500          * The return value OPAL_SUCCESS is an indication that there is
 501          * an HMI event generated waiting to pull by Linux.
 502          */
 503         rc = opal_handle_hmi();
 504         if (rc == OPAL_SUCCESS) {
 505                 local_paca->hmi_event_available = 1;
 506                 return 1;
 507         }
 508         return 0;
 509 }
 510
 511 /* HMI exception handler called in virtual mode during check_irq_replay. */
 512 int opal_handle_hmi_exception(struct pt_regs *regs)
 513 {
 514         s64 rc;
 515         __be64 evt = 0;
 516
 517         /*
 518          * Check if HMI event is available.
 519          * if Yes, then call opal_poll_events to pull opal messages and
 520          * process them.
 521          */
 522         if (!local_paca->hmi_event_available)
 523                 return 0;
 524
 525         local_paca->hmi_event_available = 0;
 526         rc = opal_poll_events(&evt);
 527         if (rc == OPAL_SUCCESS && evt)
 528                 opal_handle_events(be64_to_cpu(evt));
 529
 530         return 1;
 531 }
 532
 533 static uint64_t find_recovery_address(uint64_t nip)
 534 {
 535         int i;
 536
 537         for (i = 0; i < mc_recoverable_range_len; i++)
 538                 if ((nip >= mc_recoverable_range[i].start_addr) &&
 539                     (nip < mc_recoverable_range[i].end_addr))
 540                     return mc_recoverable_range[i].recover_addr;
 541         return 0;
 542 }
 543
 544 bool opal_mce_check_early_recovery(struct pt_regs *regs)
 545 {
 546         uint64_t recover_addr = 0;
 547
 548         if (!opal.base || !opal.size)
 549                 goto out;
 550
 551         if ((regs->nip >= opal.base) &&
 552                         (regs->nip <= (opal.base + opal.size)))
 553                 recover_addr = find_recovery_address(regs->nip);
 554
 555         /*
 556          * Setup regs->nip to rfi into fixup address.
 557          */
 558         if (recover_addr)
 559                 regs->nip = recover_addr;
 560
 561 out:
 562         return !!recover_addr;
 563 }
 564
 565 static int opal_sysfs_init(void)
 566 {
 567         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
 568         if (!opal_kobj) {
 569                 pr_warn("kobject_create_and_add opal failed\n");
 570                 return -ENOMEM;
 571         }
 572
 573         return 0;
 574 }
 575
 576 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
 577                                struct bin_attribute *bin_attr,
 578                                char *buf, loff_t off, size_t count)
 579 {
 580         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 581                                        bin_attr->size);
 582 }
 583
 584 static struct bin_attribute symbol_map_attr = {
 585         .attr = {.name = "symbol_map", .mode = 0400},
 586         .read = symbol_map_read
 587 };
 588
 589 static void opal_export_symmap(void)
 590 {
 591         const __be64 *syms;
 592         unsigned int size;
 593         struct device_node *fw;
 594         int rc;
 595
 596         fw = of_find_node_by_path("/ibm,opal/firmware");
 597         if (!fw)
 598                 return;
 599         syms = of_get_property(fw, "symbol-map", &size);
 600         if (!syms || size != 2 * sizeof(__be64))
 601                 return;
 602
 603         /* Setup attributes */
 604         symbol_map_attr.private = __va(be64_to_cpu(syms[0]));
 605         symbol_map_attr.size = be64_to_cpu(syms[1]);
 606
 607         rc = sysfs_create_bin_file(opal_kobj, &symbol_map_attr);
 608         if (rc)
 609                 pr_warn("Error %d creating OPAL symbols file\n", rc);
 610 }
 611
 612 static void __init opal_dump_region_init(void)
 613 {
 614         void *addr;
 615         uint64_t size;
 616         int rc;
 617
 618         if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
 619                 return;
 620
 621         /* Register kernel log buffer */
 622         addr = log_buf_addr_get();
 623         if (addr == NULL)
 624                 return;
 625
 626         size = log_buf_len_get();
 627         if (size == 0)
 628                 return;
 629
 630         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
 631                                        __pa(addr), size);
 632         /* Don't warn if this is just an older OPAL that doesn't
 633          * know about that call
 634          */
 635         if (rc && rc != OPAL_UNSUPPORTED)
 636                 pr_warn("DUMP: Failed to register kernel log buffer. "
 637                         "rc = %d\n", rc);
 638 }
 639
 640 static void opal_pdev_init(struct device_node *opal_node,
 641                 const char *compatible)
 642 {
 643         struct device_node *np;
 644
 645         for_each_child_of_node(opal_node, np)
 646                 if (of_device_is_compatible(np, compatible))
 647                         of_platform_device_create(np, NULL, NULL);
 648 }
 649
 650 static void opal_i2c_create_devs(void)
 651 {
 652         struct device_node *np;
 653
 654         for_each_compatible_node(np, NULL, "ibm,opal-i2c")
 655                 of_platform_device_create(np, NULL, NULL);
 656 }
 657
 658 static int kopald(void *unused)
 659 {
 660         __be64 events;
 661
 662         set_freezable();
 663         do {
 664                 try_to_freeze();
 665                 opal_poll_events(&events);
 666                 opal_handle_events(be64_to_cpu(events));
 667                 msleep_interruptible(opal_heartbeat);
 668         } while (!kthread_should_stop());
 669
 670         return 0;
 671 }
 672
 673 static void opal_init_heartbeat(void)
 674 {
 675         /* Old firwmware, we assume the HVC heartbeat is sufficient */
 676         if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
 677                                  &opal_heartbeat) != 0)
 678                 opal_heartbeat = 0;
 679
 680         if (opal_heartbeat)
 681                 kthread_run(kopald, NULL, "kopald");
 682 }
 683
 684 static int __init opal_init(void)
 685 {
 686         struct device_node *np, *consoles, *leds;
 687         int rc;
 688
 689         opal_node = of_find_node_by_path("/ibm,opal");
 690         if (!opal_node) {
 691                 pr_warn("Device node not found\n");
 692                 return -ENODEV;
 693         }
 694
 695         /* Register OPAL consoles if any ports */
 696         consoles = of_find_node_by_path("/ibm,opal/consoles");
 697         if (consoles) {
 698                 for_each_child_of_node(consoles, np) {
 699                         if (strcmp(np->name, "serial"))
 700                                 continue;
 701                         of_platform_device_create(np, NULL, NULL);
 702                 }
 703                 of_node_put(consoles);
 704         }
 705
 706         /* Initialise OPAL messaging system */
 707         opal_message_init();
 708
 709         /* Initialise OPAL asynchronous completion interface */
 710         opal_async_comp_init();
 711
 712         /* Initialise OPAL sensor interface */
 713         opal_sensor_init();
 714
 715         /* Initialise OPAL hypervisor maintainence interrupt handling */
 716         opal_hmi_handler_init();
 717
 718         /* Create i2c platform devices */
 719         opal_i2c_create_devs();
 720
 721         /* Setup a heatbeat thread if requested by OPAL */
 722         opal_init_heartbeat();
 723
 724         /* Create leds platform devices */
 725         leds = of_find_node_by_path("/ibm,opal/leds");
 726         if (leds) {
 727                 of_platform_device_create(leds, "opal_leds", NULL);
 728                 of_node_put(leds);
 729         }
 730
 731         /* Create "opal" kobject under /sys/firmware */
 732         rc = opal_sysfs_init();
 733         if (rc == 0) {
 734                 /* Export symbol map to userspace */
 735                 opal_export_symmap();
 736                 /* Setup dump region interface */
 737                 opal_dump_region_init();
 738                 /* Setup error log interface */
 739                 rc = opal_elog_init();
 740                 /* Setup code update interface */
 741                 opal_flash_update_init();
 742                 /* Setup platform dump extract interface */
 743                 opal_platform_dump_init();
 744                 /* Setup system parameters interface */
 745                 opal_sys_param_init();
 746                 /* Setup message log interface. */
 747                 opal_msglog_init();
 748         }
 749
 750         /* Initialize platform devices: IPMI backend, PRD & flash interface */
 751         opal_pdev_init(opal_node, "ibm,opal-ipmi");
 752         opal_pdev_init(opal_node, "ibm,opal-flash");
 753         opal_pdev_init(opal_node, "ibm,opal-prd");
 754
 755         /* Initialise OPAL kmsg dumper for flushing console on panic */
 756         opal_kmsg_init();
 757
 758         return 0;
 759 }
 760 machine_subsys_initcall(powernv, opal_init);
 761
 762 void opal_shutdown(void)
 763 {
 764         long rc = OPAL_BUSY;
 765
 766         opal_event_shutdown();
 767
 768         /*
 769          * Then sync with OPAL which ensure anything that can
 770          * potentially write to our memory has completed such
 771          * as an ongoing dump retrieval
 772          */
 773         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 774                 rc = opal_sync_host_reboot();
 775                 if (rc == OPAL_BUSY)
 776                         opal_poll_events(NULL);
 777                 else
 778                         mdelay(10);
 779         }
 780
 781         /* Unregister memory dump region */
 782         if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
 783                 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
 784 }
 785
 786 /* Export this so that test modules can use it */
 787 EXPORT_SYMBOL_GPL(opal_invalid_call);
 788 EXPORT_SYMBOL_GPL(opal_xscom_read);
 789 EXPORT_SYMBOL_GPL(opal_xscom_write);
 790 EXPORT_SYMBOL_GPL(opal_ipmi_send);
 791 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
 792 EXPORT_SYMBOL_GPL(opal_flash_read);
 793 EXPORT_SYMBOL_GPL(opal_flash_write);
 794 EXPORT_SYMBOL_GPL(opal_flash_erase);
 795 EXPORT_SYMBOL_GPL(opal_prd_msg);
 796
 797 /* Convert a region of vmalloc memory to an opal sg list */
 798 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
 799                                              unsigned long vmalloc_size)
 800 {
 801         struct opal_sg_list *sg, *first = NULL;
 802         unsigned long i = 0;
 803
 804         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
 805         if (!sg)
 806                 goto nomem;
 807
 808         first = sg;
 809
 810         while (vmalloc_size > 0) {
 811                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
 812                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
 813
 814                 sg->entry[i].data = cpu_to_be64(data);
 815                 sg->entry[i].length = cpu_to_be64(length);
 816                 i++;
 817
 818                 if (i >= SG_ENTRIES_PER_NODE) {
 819                         struct opal_sg_list *next;
 820
 821                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
 822                         if (!next)
 823                                 goto nomem;
 824
 825                         sg->length = cpu_to_be64(
 826                                         i * sizeof(struct opal_sg_entry) + 16);
 827                         i = 0;
 828                         sg->next = cpu_to_be64(__pa(next));
 829                         sg = next;
 830                 }
 831
 832                 vmalloc_addr += length;
 833                 vmalloc_size -= length;
 834         }
 835
 836         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
 837
 838         return first;
 839
 840 nomem:
 841         pr_err("%s : Failed to allocate memory\n", __func__);
 842         opal_free_sg_list(first);
 843         return NULL;
 844 }
 845
 846 void opal_free_sg_list(struct opal_sg_list *sg)
 847 {
 848         while (sg) {
 849                 uint64_t next = be64_to_cpu(sg->next);
 850
 851                 kfree(sg);
 852
 853                 if (next)
 854                         sg = __va(next);
 855                 else
 856                         sg = NULL;
 857         }
 858 }
 859
 860 int opal_error_code(int rc)
 861 {
 862         switch (rc) {
 863         case OPAL_SUCCESS:              return 0;
 864
 865         case OPAL_PARAMETER:            return -EINVAL;
 866         case OPAL_ASYNC_COMPLETION:     return -EINPROGRESS;
 867         case OPAL_BUSY_EVENT:           return -EBUSY;
 868         case OPAL_NO_MEM:               return -ENOMEM;
 869         case OPAL_PERMISSION:           return -EPERM;
 870
 871         case OPAL_UNSUPPORTED:          return -EIO;
 872         case OPAL_HARDWARE:             return -EIO;
 873         case OPAL_INTERNAL_ERROR:       return -EIO;
 874         default:
 875                 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
 876                 return -EIO;
 877         }
 878 }
 879
 880 EXPORT_SYMBOL_GPL(opal_poll_events);
 881 EXPORT_SYMBOL_GPL(opal_rtc_read);
 882 EXPORT_SYMBOL_GPL(opal_rtc_write);
 883 EXPORT_SYMBOL_GPL(opal_tpo_read);
 884 EXPORT_SYMBOL_GPL(opal_tpo_write);
 885 EXPORT_SYMBOL_GPL(opal_i2c_request);
 886 /* Export these symbols for PowerNV LED class driver */
 887 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
 888 EXPORT_SYMBOL_GPL(opal_leds_set_ind);