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 unsigned int *opal_irqs;
  57 static unsigned int opal_irq_count;
  58 static ATOMIC_NOTIFIER_HEAD(opal_notifier_head);
  59 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
  60 static DEFINE_SPINLOCK(opal_notifier_lock);
  61 static uint64_t last_notified_mask = 0x0ul;
  62 static atomic_t opal_notifier_hold = ATOMIC_INIT(0);
  63 static uint32_t opal_heartbeat;
  64
  65 static void opal_reinit_cores(void)
  66 {
  67         /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
  68          *
  69          * It will preserve non volatile GPRs and HSPRG0/1. It will
  70          * also restore HIDs and other SPRs to their original value
  71          * but it might clobber a bunch.
  72          */
  73 #ifdef __BIG_ENDIAN__
  74         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
  75 #else
  76         opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
  77 #endif
  78 }
  79
  80 int __init early_init_dt_scan_opal(unsigned long node,
  81                                    const char *uname, int depth, void *data)
  82 {
  83         const void *basep, *entryp, *sizep;
  84         int basesz, entrysz, runtimesz;
  85
  86         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
  87                 return 0;
  88
  89         basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
  90         entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
  91         sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
  92
  93         if (!basep || !entryp || !sizep)
  94                 return 1;
  95
  96         opal.base = of_read_number(basep, basesz/4);
  97         opal.entry = of_read_number(entryp, entrysz/4);
  98         opal.size = of_read_number(sizep, runtimesz/4);
  99
 100         pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
 101                  opal.base, basep, basesz);
 102         pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
 103                  opal.entry, entryp, entrysz);
 104         pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
 105                  opal.size, sizep, runtimesz);
 106
 107         powerpc_firmware_features |= FW_FEATURE_OPAL;
 108         if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
 109                 powerpc_firmware_features |= FW_FEATURE_OPALv2;
 110                 powerpc_firmware_features |= FW_FEATURE_OPALv3;
 111                 pr_info("OPAL V3 detected !\n");
 112         } else if (of_flat_dt_is_compatible(node, "ibm,opal-v2")) {
 113                 powerpc_firmware_features |= FW_FEATURE_OPALv2;
 114                 pr_info("OPAL V2 detected !\n");
 115         } else {
 116                 pr_info("OPAL V1 detected !\n");
 117         }
 118
 119         /* Reinit all cores with the right endian */
 120         opal_reinit_cores();
 121
 122         /* Restore some bits */
 123         if (cur_cpu_spec->cpu_restore)
 124                 cur_cpu_spec->cpu_restore();
 125
 126         return 1;
 127 }
 128
 129 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
 130                                    const char *uname, int depth, void *data)
 131 {
 132         int i, psize, size;
 133         const __be32 *prop;
 134
 135         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
 136                 return 0;
 137
 138         prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
 139
 140         if (!prop)
 141                 return 1;
 142
 143         pr_debug("Found machine check recoverable ranges.\n");
 144
 145         /*
 146          * Calculate number of available entries.
 147          *
 148          * Each recoverable address range entry is (start address, len,
 149          * recovery address), 2 cells each for start and recovery address,
 150          * 1 cell for len, totalling 5 cells per entry.
 151          */
 152         mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
 153
 154         /* Sanity check */
 155         if (!mc_recoverable_range_len)
 156                 return 1;
 157
 158         /* Size required to hold all the entries. */
 159         size = mc_recoverable_range_len *
 160                         sizeof(struct mcheck_recoverable_range);
 161
 162         /*
 163          * Allocate a buffer to hold the MC recoverable ranges. We would be
 164          * accessing them in real mode, hence it needs to be within
 165          * RMO region.
 166          */
 167         mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
 168                                                         ppc64_rma_size));
 169         memset(mc_recoverable_range, 0, size);
 170
 171         for (i = 0; i < mc_recoverable_range_len; i++) {
 172                 mc_recoverable_range[i].start_addr =
 173                                         of_read_number(prop + (i * 5) + 0, 2);
 174                 mc_recoverable_range[i].end_addr =
 175                                         mc_recoverable_range[i].start_addr +
 176                                         of_read_number(prop + (i * 5) + 2, 1);
 177                 mc_recoverable_range[i].recover_addr =
 178                                         of_read_number(prop + (i * 5) + 3, 2);
 179
 180                 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
 181                                 mc_recoverable_range[i].start_addr,
 182                                 mc_recoverable_range[i].end_addr,
 183                                 mc_recoverable_range[i].recover_addr);
 184         }
 185         return 1;
 186 }
 187
 188 static int __init opal_register_exception_handlers(void)
 189 {
 190 #ifdef __BIG_ENDIAN__
 191         u64 glue;
 192
 193         if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
 194                 return -ENODEV;
 195
 196         /* Hookup some exception handlers except machine check. We use the
 197          * fwnmi area at 0x7000 to provide the glue space to OPAL
 198          */
 199         glue = 0x7000;
 200
 201         /*
 202          * Check if we are running on newer firmware that exports
 203          * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
 204          * the HMI interrupt and we catch it directly in Linux.
 205          *
 206          * For older firmware (i.e currently released POWER8 System Firmware
 207          * as of today <= SV810_087), we fallback to old behavior and let OPAL
 208          * patch the HMI vector and handle it inside OPAL firmware.
 209          *
 210          * For newer firmware (in development/yet to be released) we will
 211          * start catching/handling HMI directly in Linux.
 212          */
 213         if (!opal_check_token(OPAL_HANDLE_HMI)) {
 214                 pr_info("Old firmware detected, OPAL handles HMIs.\n");
 215                 opal_register_exception_handler(
 216                                 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
 217                                 0, glue);
 218                 glue += 128;
 219         }
 220
 221         opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
 222 #endif
 223
 224         return 0;
 225 }
 226 machine_early_initcall(powernv, opal_register_exception_handlers);
 227
 228 int opal_notifier_register(struct notifier_block *nb)
 229 {
 230         if (!nb) {
 231                 pr_warning("%s: Invalid argument (%p)\n",
 232                            __func__, nb);
 233                 return -EINVAL;
 234         }
 235
 236         atomic_notifier_chain_register(&opal_notifier_head, nb);
 237         return 0;
 238 }
 239 EXPORT_SYMBOL_GPL(opal_notifier_register);
 240
 241 int opal_notifier_unregister(struct notifier_block *nb)
 242 {
 243         if (!nb) {
 244                 pr_warning("%s: Invalid argument (%p)\n",
 245                            __func__, nb);
 246                 return -EINVAL;
 247         }
 248
 249         atomic_notifier_chain_unregister(&opal_notifier_head, nb);
 250         return 0;
 251 }
 252 EXPORT_SYMBOL_GPL(opal_notifier_unregister);
 253
 254 static void opal_do_notifier(uint64_t events)
 255 {
 256         unsigned long flags;
 257         uint64_t changed_mask;
 258
 259         if (atomic_read(&opal_notifier_hold))
 260                 return;
 261
 262         spin_lock_irqsave(&opal_notifier_lock, flags);
 263         changed_mask = last_notified_mask ^ events;
 264         last_notified_mask = events;
 265         spin_unlock_irqrestore(&opal_notifier_lock, flags);
 266
 267         /*
 268          * We feed with the event bits and changed bits for
 269          * enough information to the callback.
 270          */
 271         atomic_notifier_call_chain(&opal_notifier_head,
 272                                    events, (void *)changed_mask);
 273 }
 274
 275 void opal_notifier_update_evt(uint64_t evt_mask,
 276                               uint64_t evt_val)
 277 {
 278         unsigned long flags;
 279
 280         spin_lock_irqsave(&opal_notifier_lock, flags);
 281         last_notified_mask &= ~evt_mask;
 282         last_notified_mask |= evt_val;
 283         spin_unlock_irqrestore(&opal_notifier_lock, flags);
 284 }
 285
 286 void opal_notifier_enable(void)
 287 {
 288         int64_t rc;
 289         __be64 evt = 0;
 290
 291         atomic_set(&opal_notifier_hold, 0);
 292
 293         /* Process pending events */
 294         rc = opal_poll_events(&evt);
 295         if (rc == OPAL_SUCCESS && evt)
 296                 opal_do_notifier(be64_to_cpu(evt));
 297 }
 298
 299 void opal_notifier_disable(void)
 300 {
 301         atomic_set(&opal_notifier_hold, 1);
 302 }
 303
 304 /*
 305  * Opal message notifier based on message type. Allow subscribers to get
 306  * notified for specific messgae type.
 307  */
 308 int opal_message_notifier_register(enum OpalMessageType msg_type,
 309                                         struct notifier_block *nb)
 310 {
 311         if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
 312                 pr_warning("%s: Invalid arguments, msg_type:%d\n",
 313                            __func__, msg_type);
 314                 return -EINVAL;
 315         }
 316
 317         return atomic_notifier_chain_register(
 318                                 &opal_msg_notifier_head[msg_type], nb);
 319 }
 320
 321 static void opal_message_do_notify(uint32_t msg_type, void *msg)
 322 {
 323         /* notify subscribers */
 324         atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
 325                                         msg_type, msg);
 326 }
 327
 328 static void opal_handle_message(void)
 329 {
 330         s64 ret;
 331         /*
 332          * TODO: pre-allocate a message buffer depending on opal-msg-size
 333          * value in /proc/device-tree.
 334          */
 335         static struct opal_msg msg;
 336         u32 type;
 337
 338         ret = opal_get_msg(__pa(&msg), sizeof(msg));
 339         /* No opal message pending. */
 340         if (ret == OPAL_RESOURCE)
 341                 return;
 342
 343         /* check for errors. */
 344         if (ret) {
 345                 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
 346                                 __func__, ret);
 347                 return;
 348         }
 349
 350         type = be32_to_cpu(msg.msg_type);
 351
 352         /* Sanity check */
 353         if (type >= OPAL_MSG_TYPE_MAX) {
 354                 pr_warning("%s: Unknown message type: %u\n", __func__, type);
 355                 return;
 356         }
 357         opal_message_do_notify(type, (void *)&msg);
 358 }
 359
 360 static int opal_message_notify(struct notifier_block *nb,
 361                           unsigned long events, void *change)
 362 {
 363         if (events & OPAL_EVENT_MSG_PENDING)
 364                 opal_handle_message();
 365         return 0;
 366 }
 367
 368 static struct notifier_block opal_message_nb = {
 369         .notifier_call  = opal_message_notify,
 370         .next           = NULL,
 371         .priority       = 0,
 372 };
 373
 374 static int __init opal_message_init(void)
 375 {
 376         int ret, i;
 377
 378         for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
 379                 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
 380
 381         ret = opal_notifier_register(&opal_message_nb);
 382         if (ret) {
 383                 pr_err("%s: Can't register OPAL event notifier (%d)\n",
 384                        __func__, ret);
 385                 return ret;
 386         }
 387         return 0;
 388 }
 389 machine_early_initcall(powernv, opal_message_init);
 390
 391 int opal_get_chars(uint32_t vtermno, char *buf, int count)
 392 {
 393         s64 rc;
 394         __be64 evt, len;
 395
 396         if (!opal.entry)
 397                 return -ENODEV;
 398         opal_poll_events(&evt);
 399         if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
 400                 return 0;
 401         len = cpu_to_be64(count);
 402         rc = opal_console_read(vtermno, &len, buf);
 403         if (rc == OPAL_SUCCESS)
 404                 return be64_to_cpu(len);
 405         return 0;
 406 }
 407
 408 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
 409 {
 410         int written = 0;
 411         __be64 olen;
 412         s64 len, rc;
 413         unsigned long flags;
 414         __be64 evt;
 415
 416         if (!opal.entry)
 417                 return -ENODEV;
 418
 419         /* We want put_chars to be atomic to avoid mangling of hvsi
 420          * packets. To do that, we first test for room and return
 421          * -EAGAIN if there isn't enough.
 422          *
 423          * Unfortunately, opal_console_write_buffer_space() doesn't
 424          * appear to work on opal v1, so we just assume there is
 425          * enough room and be done with it
 426          */
 427         spin_lock_irqsave(&opal_write_lock, flags);
 428         if (firmware_has_feature(FW_FEATURE_OPALv2)) {
 429                 rc = opal_console_write_buffer_space(vtermno, &olen);
 430                 len = be64_to_cpu(olen);
 431                 if (rc || len < total_len) {
 432                         spin_unlock_irqrestore(&opal_write_lock, flags);
 433                         /* Closed -> drop characters */
 434                         if (rc)
 435                                 return total_len;
 436                         opal_poll_events(NULL);
 437                         return -EAGAIN;
 438                 }
 439         }
 440
 441         /* We still try to handle partial completions, though they
 442          * should no longer happen.
 443          */
 444         rc = OPAL_BUSY;
 445         while(total_len > 0 && (rc == OPAL_BUSY ||
 446                                 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
 447                 olen = cpu_to_be64(total_len);
 448                 rc = opal_console_write(vtermno, &olen, data);
 449                 len = be64_to_cpu(olen);
 450
 451                 /* Closed or other error drop */
 452                 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
 453                     rc != OPAL_BUSY_EVENT) {
 454                         written = total_len;
 455                         break;
 456                 }
 457                 if (rc == OPAL_SUCCESS) {
 458                         total_len -= len;
 459                         data += len;
 460                         written += len;
 461                 }
 462                 /* This is a bit nasty but we need that for the console to
 463                  * flush when there aren't any interrupts. We will clean
 464                  * things a bit later to limit that to synchronous path
 465                  * such as the kernel console and xmon/udbg
 466                  */
 467                 do
 468                         opal_poll_events(&evt);
 469                 while(rc == OPAL_SUCCESS &&
 470                         (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
 471         }
 472         spin_unlock_irqrestore(&opal_write_lock, flags);
 473         return written;
 474 }
 475
 476 static int opal_recover_mce(struct pt_regs *regs,
 477                                         struct machine_check_event *evt)
 478 {
 479         int recovered = 0;
 480         uint64_t ea = get_mce_fault_addr(evt);
 481
 482         if (!(regs->msr & MSR_RI)) {
 483                 /* If MSR_RI isn't set, we cannot recover */
 484                 recovered = 0;
 485         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
 486                 /* Platform corrected itself */
 487                 recovered = 1;
 488         } else if (ea && !is_kernel_addr(ea)) {
 489                 /*
 490                  * Faulting address is not in kernel text. We should be fine.
 491                  * We need to find which process uses this address.
 492                  * For now, kill the task if we have received exception when
 493                  * in userspace.
 494                  *
 495                  * TODO: Queue up this address for hwpoisioning later.
 496                  */
 497                 if (user_mode(regs) && !is_global_init(current)) {
 498                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 499                         recovered = 1;
 500                 } else
 501                         recovered = 0;
 502         } else if (user_mode(regs) && !is_global_init(current) &&
 503                 evt->severity == MCE_SEV_ERROR_SYNC) {
 504                 /*
 505                  * If we have received a synchronous error when in userspace
 506                  * kill the task.
 507                  */
 508                 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 509                 recovered = 1;
 510         }
 511         return recovered;
 512 }
 513
 514 int opal_machine_check(struct pt_regs *regs)
 515 {
 516         struct machine_check_event evt;
 517
 518         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 519                 return 0;
 520
 521         /* Print things out */
 522         if (evt.version != MCE_V1) {
 523                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 524                        evt.version);
 525                 return 0;
 526         }
 527         machine_check_print_event_info(&evt);
 528
 529         if (opal_recover_mce(regs, &evt))
 530                 return 1;
 531         return 0;
 532 }
 533
 534 /* Early hmi handler called in real mode. */
 535 int opal_hmi_exception_early(struct pt_regs *regs)
 536 {
 537         s64 rc;
 538
 539         /*
 540          * call opal hmi handler. Pass paca address as token.
 541          * The return value OPAL_SUCCESS is an indication that there is
 542          * an HMI event generated waiting to pull by Linux.
 543          */
 544         rc = opal_handle_hmi();
 545         if (rc == OPAL_SUCCESS) {
 546                 local_paca->hmi_event_available = 1;
 547                 return 1;
 548         }
 549         return 0;
 550 }
 551
 552 /* HMI exception handler called in virtual mode during check_irq_replay. */
 553 int opal_handle_hmi_exception(struct pt_regs *regs)
 554 {
 555         s64 rc;
 556         __be64 evt = 0;
 557
 558         /*
 559          * Check if HMI event is available.
 560          * if Yes, then call opal_poll_events to pull opal messages and
 561          * process them.
 562          */
 563         if (!local_paca->hmi_event_available)
 564                 return 0;
 565
 566         local_paca->hmi_event_available = 0;
 567         rc = opal_poll_events(&evt);
 568         if (rc == OPAL_SUCCESS && evt)
 569                 opal_do_notifier(be64_to_cpu(evt));
 570
 571         return 1;
 572 }
 573
 574 static uint64_t find_recovery_address(uint64_t nip)
 575 {
 576         int i;
 577
 578         for (i = 0; i < mc_recoverable_range_len; i++)
 579                 if ((nip >= mc_recoverable_range[i].start_addr) &&
 580                     (nip < mc_recoverable_range[i].end_addr))
 581                     return mc_recoverable_range[i].recover_addr;
 582         return 0;
 583 }
 584
 585 bool opal_mce_check_early_recovery(struct pt_regs *regs)
 586 {
 587         uint64_t recover_addr = 0;
 588
 589         if (!opal.base || !opal.size)
 590                 goto out;
 591
 592         if ((regs->nip >= opal.base) &&
 593                         (regs->nip <= (opal.base + opal.size)))
 594                 recover_addr = find_recovery_address(regs->nip);
 595
 596         /*
 597          * Setup regs->nip to rfi into fixup address.
 598          */
 599         if (recover_addr)
 600                 regs->nip = recover_addr;
 601
 602 out:
 603         return !!recover_addr;
 604 }
 605
 606 static irqreturn_t opal_interrupt(int irq, void *data)
 607 {
 608         __be64 events;
 609
 610         opal_handle_interrupt(virq_to_hw(irq), &events);
 611
 612         opal_do_notifier(be64_to_cpu(events));
 613
 614         return IRQ_HANDLED;
 615 }
 616
 617 static int opal_sysfs_init(void)
 618 {
 619         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
 620         if (!opal_kobj) {
 621                 pr_warn("kobject_create_and_add opal failed\n");
 622                 return -ENOMEM;
 623         }
 624
 625         return 0;
 626 }
 627
 628 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
 629                                struct bin_attribute *bin_attr,
 630                                char *buf, loff_t off, size_t count)
 631 {
 632         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 633                                        bin_attr->size);
 634 }
 635
 636 static BIN_ATTR_RO(symbol_map, 0);
 637
 638 static void opal_export_symmap(void)
 639 {
 640         const __be64 *syms;
 641         unsigned int size;
 642         struct device_node *fw;
 643         int rc;
 644
 645         fw = of_find_node_by_path("/ibm,opal/firmware");
 646         if (!fw)
 647                 return;
 648         syms = of_get_property(fw, "symbol-map", &size);
 649         if (!syms || size != 2 * sizeof(__be64))
 650                 return;
 651
 652         /* Setup attributes */
 653         bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
 654         bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
 655
 656         rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
 657         if (rc)
 658                 pr_warn("Error %d creating OPAL symbols file\n", rc);
 659 }
 660
 661 static void __init opal_dump_region_init(void)
 662 {
 663         void *addr;
 664         uint64_t size;
 665         int rc;
 666
 667         /* Register kernel log buffer */
 668         addr = log_buf_addr_get();
 669         if (addr == NULL)
 670                 return;
 671
 672         size = log_buf_len_get();
 673         if (size == 0)
 674                 return;
 675
 676         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
 677                                        __pa(addr), size);
 678         /* Don't warn if this is just an older OPAL that doesn't
 679          * know about that call
 680          */
 681         if (rc && rc != OPAL_UNSUPPORTED)
 682                 pr_warn("DUMP: Failed to register kernel log buffer. "
 683                         "rc = %d\n", rc);
 684 }
 685
 686 static void opal_ipmi_init(struct device_node *opal_node)
 687 {
 688         struct device_node *np;
 689
 690         for_each_child_of_node(opal_node, np)
 691                 if (of_device_is_compatible(np, "ibm,opal-ipmi"))
 692                         of_platform_device_create(np, NULL, NULL);
 693 }
 694
 695 static void opal_i2c_create_devs(void)
 696 {
 697         struct device_node *np;
 698
 699         for_each_compatible_node(np, NULL, "ibm,opal-i2c")
 700                 of_platform_device_create(np, NULL, NULL);
 701 }
 702
 703 static void __init opal_irq_init(struct device_node *dn)
 704 {
 705         const __be32 *irqs;
 706         int i, irqlen;
 707
 708         /* Get interrupt property */
 709         irqs = of_get_property(opal_node, "opal-interrupts", &irqlen);
 710         opal_irq_count = irqs ? (irqlen / 4) : 0;
 711         pr_debug("Found %d interrupts reserved for OPAL\n", opal_irq_count);
 712         if (!opal_irq_count)
 713                 return;
 714
 715         /* Install interrupt handlers */
 716         opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL);
 717         for (i = 0; irqs && i < opal_irq_count; i++, irqs++) {
 718                 unsigned int irq, virq;
 719                 int rc;
 720
 721                 /* Get hardware and virtual IRQ */
 722                 irq = be32_to_cpup(irqs);
 723                 virq = irq_create_mapping(NULL, irq);
 724                 if (virq == NO_IRQ) {
 725                         pr_warn("Failed to map irq 0x%x\n", irq);
 726                         continue;
 727                 }
 728
 729                 /* Install interrupt handler */
 730                 rc = request_irq(virq, opal_interrupt, 0, "opal", NULL);
 731                 if (rc) {
 732                         irq_dispose_mapping(virq);
 733                         pr_warn("Error %d requesting irq %d (0x%x)\n",
 734                                  rc, virq, irq);
 735                         continue;
 736                 }
 737
 738                 /* Cache IRQ */
 739                 opal_irqs[i] = virq;
 740         }
 741 }
 742
 743 static int kopald(void *unused)
 744 {
 745         set_freezable();
 746         do {
 747                 try_to_freeze();
 748                 opal_poll_events(NULL);
 749                 msleep_interruptible(opal_heartbeat);
 750         } while (!kthread_should_stop());
 751
 752         return 0;
 753 }
 754
 755 static void opal_init_heartbeat(void)
 756 {
 757         /* Old firwmware, we assume the HVC heartbeat is sufficient */
 758         if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
 759                                  &opal_heartbeat) != 0)
 760                 opal_heartbeat = 0;
 761
 762         if (opal_heartbeat)
 763                 kthread_run(kopald, NULL, "kopald");
 764 }
 765
 766 static int __init opal_init(void)
 767 {
 768         struct device_node *np, *consoles;
 769         int rc;
 770
 771         opal_node = of_find_node_by_path("/ibm,opal");
 772         if (!opal_node) {
 773                 pr_warn("Device node not found\n");
 774                 return -ENODEV;
 775         }
 776
 777         /* Register OPAL consoles if any ports */
 778         if (firmware_has_feature(FW_FEATURE_OPALv2))
 779                 consoles = of_find_node_by_path("/ibm,opal/consoles");
 780         else
 781                 consoles = of_node_get(opal_node);
 782         if (consoles) {
 783                 for_each_child_of_node(consoles, np) {
 784                         if (strcmp(np->name, "serial"))
 785                                 continue;
 786                         of_platform_device_create(np, NULL, NULL);
 787                 }
 788                 of_node_put(consoles);
 789         }
 790
 791         /* Create i2c platform devices */
 792         opal_i2c_create_devs();
 793
 794         /* Setup a heatbeat thread if requested by OPAL */
 795         opal_init_heartbeat();
 796
 797         /* Find all OPAL interrupts and request them */
 798         opal_irq_init(opal_node);
 799
 800         /* Create "opal" kobject under /sys/firmware */
 801         rc = opal_sysfs_init();
 802         if (rc == 0) {
 803                 /* Export symbol map to userspace */
 804                 opal_export_symmap();
 805                 /* Setup dump region interface */
 806                 opal_dump_region_init();
 807                 /* Setup error log interface */
 808                 rc = opal_elog_init();
 809                 /* Setup code update interface */
 810                 opal_flash_init();
 811                 /* Setup platform dump extract interface */
 812                 opal_platform_dump_init();
 813                 /* Setup system parameters interface */
 814                 opal_sys_param_init();
 815                 /* Setup message log interface. */
 816                 opal_msglog_init();
 817         }
 818
 819         /* Initialize OPAL IPMI backend */
 820         opal_ipmi_init(opal_node);
 821
 822         return 0;
 823 }
 824 machine_subsys_initcall(powernv, opal_init);
 825
 826 void opal_shutdown(void)
 827 {
 828         unsigned int i;
 829         long rc = OPAL_BUSY;
 830
 831         /* First free interrupts, which will also mask them */
 832         for (i = 0; i < opal_irq_count; i++) {
 833                 if (opal_irqs[i])
 834                         free_irq(opal_irqs[i], NULL);
 835                 opal_irqs[i] = 0;
 836         }
 837
 838         /*
 839          * Then sync with OPAL which ensure anything that can
 840          * potentially write to our memory has completed such
 841          * as an ongoing dump retrieval
 842          */
 843         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 844                 rc = opal_sync_host_reboot();
 845                 if (rc == OPAL_BUSY)
 846                         opal_poll_events(NULL);
 847                 else
 848                         mdelay(10);
 849         }
 850
 851         /* Unregister memory dump region */
 852         opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
 853 }
 854
 855 /* Export this so that test modules can use it */
 856 EXPORT_SYMBOL_GPL(opal_invalid_call);
 857 EXPORT_SYMBOL_GPL(opal_ipmi_send);
 858 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
 859
 860 /* Convert a region of vmalloc memory to an opal sg list */
 861 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
 862                                              unsigned long vmalloc_size)
 863 {
 864         struct opal_sg_list *sg, *first = NULL;
 865         unsigned long i = 0;
 866
 867         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
 868         if (!sg)
 869                 goto nomem;
 870
 871         first = sg;
 872
 873         while (vmalloc_size > 0) {
 874                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
 875                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
 876
 877                 sg->entry[i].data = cpu_to_be64(data);
 878                 sg->entry[i].length = cpu_to_be64(length);
 879                 i++;
 880
 881                 if (i >= SG_ENTRIES_PER_NODE) {
 882                         struct opal_sg_list *next;
 883
 884                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
 885                         if (!next)
 886                                 goto nomem;
 887
 888                         sg->length = cpu_to_be64(
 889                                         i * sizeof(struct opal_sg_entry) + 16);
 890                         i = 0;
 891                         sg->next = cpu_to_be64(__pa(next));
 892                         sg = next;
 893                 }
 894
 895                 vmalloc_addr += length;
 896                 vmalloc_size -= length;
 897         }
 898
 899         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
 900
 901         return first;
 902
 903 nomem:
 904         pr_err("%s : Failed to allocate memory\n", __func__);
 905         opal_free_sg_list(first);
 906         return NULL;
 907 }
 908
 909 void opal_free_sg_list(struct opal_sg_list *sg)
 910 {
 911         while (sg) {
 912                 uint64_t next = be64_to_cpu(sg->next);
 913
 914                 kfree(sg);
 915
 916                 if (next)
 917                         sg = __va(next);
 918                 else
 919                         sg = NULL;
 920         }
 921 }
 922
 923 EXPORT_SYMBOL_GPL(opal_poll_events);
 924 EXPORT_SYMBOL_GPL(opal_rtc_read);
 925 EXPORT_SYMBOL_GPL(opal_rtc_write);
 926 EXPORT_SYMBOL_GPL(opal_tpo_read);
 927 EXPORT_SYMBOL_GPL(opal_tpo_write);
 928 EXPORT_SYMBOL_GPL(opal_i2c_request);