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/of_address.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/notifier.h>
  22 #include <linux/slab.h>
  23 #include <linux/sched.h>
  24 #include <linux/kobject.h>
  25 #include <linux/delay.h>
  26 #include <linux/memblock.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/printk.h>
  30 #include <linux/kmsg_dump.h>
  31 #include <linux/console.h>
  32 #include <linux/sched/debug.h>
  33
  34 #include <asm/machdep.h>
  35 #include <asm/opal.h>
  36 #include <asm/firmware.h>
  37 #include <asm/mce.h>
  38 #include <asm/imc-pmu.h>
  39 #include <asm/bug.h>
  40
  41 #include "powernv.h"
  42
  43 /* /sys/firmware/opal */
  44 struct kobject *opal_kobj;
  45
  46 struct opal {
  47         u64 base;
  48         u64 entry;
  49         u64 size;
  50 } opal;
  51
  52 struct mcheck_recoverable_range {
  53         u64 start_addr;
  54         u64 end_addr;
  55         u64 recover_addr;
  56 };
  57
  58 static struct mcheck_recoverable_range *mc_recoverable_range;
  59 static int mc_recoverable_range_len;
  60
  61 struct device_node *opal_node;
  62 static DEFINE_SPINLOCK(opal_write_lock);
  63 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
  64 static uint32_t opal_heartbeat;
  65 static struct task_struct *kopald_tsk;
  66
  67 void opal_configure_cores(void)
  68 {
  69         u64 reinit_flags = 0;
  70
  71         /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
  72          *
  73          * It will preserve non volatile GPRs and HSPRG0/1. It will
  74          * also restore HIDs and other SPRs to their original value
  75          * but it might clobber a bunch.
  76          */
  77 #ifdef __BIG_ENDIAN__
  78         reinit_flags |= OPAL_REINIT_CPUS_HILE_BE;
  79 #else
  80         reinit_flags |= OPAL_REINIT_CPUS_HILE_LE;
  81 #endif
  82
  83         /*
  84          * POWER9 always support running hash:
  85          *  ie. Host hash  supports  hash guests
  86          *      Host radix supports  hash/radix guests
  87          */
  88         if (early_cpu_has_feature(CPU_FTR_ARCH_300)) {
  89                 reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH;
  90                 if (early_radix_enabled())
  91                         reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX;
  92         }
  93
  94         opal_reinit_cpus(reinit_flags);
  95
  96         /* Restore some bits */
  97         if (cur_cpu_spec->cpu_restore)
  98                 cur_cpu_spec->cpu_restore();
  99 }
 100
 101 int __init early_init_dt_scan_opal(unsigned long node,
 102                                    const char *uname, int depth, void *data)
 103 {
 104         const void *basep, *entryp, *sizep;
 105         int basesz, entrysz, runtimesz;
 106
 107         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
 108                 return 0;
 109
 110         basep  = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
 111         entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
 112         sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
 113
 114         if (!basep || !entryp || !sizep)
 115                 return 1;
 116
 117         opal.base = of_read_number(basep, basesz/4);
 118         opal.entry = of_read_number(entryp, entrysz/4);
 119         opal.size = of_read_number(sizep, runtimesz/4);
 120
 121         pr_debug("OPAL Base  = 0x%llx (basep=%p basesz=%d)\n",
 122                  opal.base, basep, basesz);
 123         pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
 124                  opal.entry, entryp, entrysz);
 125         pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
 126                  opal.size, sizep, runtimesz);
 127
 128         if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
 129                 powerpc_firmware_features |= FW_FEATURE_OPAL;
 130                 pr_info("OPAL detected !\n");
 131         } else {
 132                 panic("OPAL != V3 detected, no longer supported.\n");
 133         }
 134
 135         return 1;
 136 }
 137
 138 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
 139                                    const char *uname, int depth, void *data)
 140 {
 141         int i, psize, size;
 142         const __be32 *prop;
 143
 144         if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
 145                 return 0;
 146
 147         prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
 148
 149         if (!prop)
 150                 return 1;
 151
 152         pr_debug("Found machine check recoverable ranges.\n");
 153
 154         /*
 155          * Calculate number of available entries.
 156          *
 157          * Each recoverable address range entry is (start address, len,
 158          * recovery address), 2 cells each for start and recovery address,
 159          * 1 cell for len, totalling 5 cells per entry.
 160          */
 161         mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
 162
 163         /* Sanity check */
 164         if (!mc_recoverable_range_len)
 165                 return 1;
 166
 167         /* Size required to hold all the entries. */
 168         size = mc_recoverable_range_len *
 169                         sizeof(struct mcheck_recoverable_range);
 170
 171         /*
 172          * Allocate a buffer to hold the MC recoverable ranges.
 173          */
 174         mc_recoverable_range =__va(memblock_alloc(size, __alignof__(u64)));
 175         memset(mc_recoverable_range, 0, size);
 176
 177         for (i = 0; i < mc_recoverable_range_len; i++) {
 178                 mc_recoverable_range[i].start_addr =
 179                                         of_read_number(prop + (i * 5) + 0, 2);
 180                 mc_recoverable_range[i].end_addr =
 181                                         mc_recoverable_range[i].start_addr +
 182                                         of_read_number(prop + (i * 5) + 2, 1);
 183                 mc_recoverable_range[i].recover_addr =
 184                                         of_read_number(prop + (i * 5) + 3, 2);
 185
 186                 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
 187                                 mc_recoverable_range[i].start_addr,
 188                                 mc_recoverable_range[i].end_addr,
 189                                 mc_recoverable_range[i].recover_addr);
 190         }
 191         return 1;
 192 }
 193
 194 static int __init opal_register_exception_handlers(void)
 195 {
 196 #ifdef __BIG_ENDIAN__
 197         u64 glue;
 198
 199         if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
 200                 return -ENODEV;
 201
 202         /* Hookup some exception handlers except machine check. We use the
 203          * fwnmi area at 0x7000 to provide the glue space to OPAL
 204          */
 205         glue = 0x7000;
 206
 207         /*
 208          * Check if we are running on newer firmware that exports
 209          * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
 210          * the HMI interrupt and we catch it directly in Linux.
 211          *
 212          * For older firmware (i.e currently released POWER8 System Firmware
 213          * as of today <= SV810_087), we fallback to old behavior and let OPAL
 214          * patch the HMI vector and handle it inside OPAL firmware.
 215          *
 216          * For newer firmware (in development/yet to be released) we will
 217          * start catching/handling HMI directly in Linux.
 218          */
 219         if (!opal_check_token(OPAL_HANDLE_HMI)) {
 220                 pr_info("Old firmware detected, OPAL handles HMIs.\n");
 221                 opal_register_exception_handler(
 222                                 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
 223                                 0, glue);
 224                 glue += 128;
 225         }
 226
 227         opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
 228 #endif
 229
 230         return 0;
 231 }
 232 machine_early_initcall(powernv, opal_register_exception_handlers);
 233
 234 /*
 235  * Opal message notifier based on message type. Allow subscribers to get
 236  * notified for specific messgae type.
 237  */
 238 int opal_message_notifier_register(enum opal_msg_type msg_type,
 239                                         struct notifier_block *nb)
 240 {
 241         if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
 242                 pr_warning("%s: Invalid arguments, msg_type:%d\n",
 243                            __func__, msg_type);
 244                 return -EINVAL;
 245         }
 246
 247         return atomic_notifier_chain_register(
 248                                 &opal_msg_notifier_head[msg_type], nb);
 249 }
 250 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
 251
 252 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
 253                                      struct notifier_block *nb)
 254 {
 255         return atomic_notifier_chain_unregister(
 256                         &opal_msg_notifier_head[msg_type], nb);
 257 }
 258 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
 259
 260 static void opal_message_do_notify(uint32_t msg_type, void *msg)
 261 {
 262         /* notify subscribers */
 263         atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
 264                                         msg_type, msg);
 265 }
 266
 267 static void opal_handle_message(void)
 268 {
 269         s64 ret;
 270         /*
 271          * TODO: pre-allocate a message buffer depending on opal-msg-size
 272          * value in /proc/device-tree.
 273          */
 274         static struct opal_msg msg;
 275         u32 type;
 276
 277         ret = opal_get_msg(__pa(&msg), sizeof(msg));
 278         /* No opal message pending. */
 279         if (ret == OPAL_RESOURCE)
 280                 return;
 281
 282         /* check for errors. */
 283         if (ret) {
 284                 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
 285                                 __func__, ret);
 286                 return;
 287         }
 288
 289         type = be32_to_cpu(msg.msg_type);
 290
 291         /* Sanity check */
 292         if (type >= OPAL_MSG_TYPE_MAX) {
 293                 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
 294                 return;
 295         }
 296         opal_message_do_notify(type, (void *)&msg);
 297 }
 298
 299 static irqreturn_t opal_message_notify(int irq, void *data)
 300 {
 301         opal_handle_message();
 302         return IRQ_HANDLED;
 303 }
 304
 305 static int __init opal_message_init(void)
 306 {
 307         int ret, i, irq;
 308
 309         for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
 310                 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
 311
 312         irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
 313         if (!irq) {
 314                 pr_err("%s: Can't register OPAL event irq (%d)\n",
 315                        __func__, irq);
 316                 return irq;
 317         }
 318
 319         ret = request_irq(irq, opal_message_notify,
 320                         IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
 321         if (ret) {
 322                 pr_err("%s: Can't request OPAL event irq (%d)\n",
 323                        __func__, ret);
 324                 return ret;
 325         }
 326
 327         return 0;
 328 }
 329
 330 int opal_get_chars(uint32_t vtermno, char *buf, int count)
 331 {
 332         s64 rc;
 333         __be64 evt, len;
 334
 335         if (!opal.entry)
 336                 return -ENODEV;
 337         opal_poll_events(&evt);
 338         if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
 339                 return 0;
 340         len = cpu_to_be64(count);
 341         rc = opal_console_read(vtermno, &len, buf);
 342         if (rc == OPAL_SUCCESS)
 343                 return be64_to_cpu(len);
 344         return 0;
 345 }
 346
 347 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
 348 {
 349         int written = 0;
 350         __be64 olen;
 351         s64 len, rc;
 352         unsigned long flags;
 353         __be64 evt;
 354
 355         if (!opal.entry)
 356                 return -ENODEV;
 357
 358         /* We want put_chars to be atomic to avoid mangling of hvsi
 359          * packets. To do that, we first test for room and return
 360          * -EAGAIN if there isn't enough.
 361          *
 362          * Unfortunately, opal_console_write_buffer_space() doesn't
 363          * appear to work on opal v1, so we just assume there is
 364          * enough room and be done with it
 365          */
 366         spin_lock_irqsave(&opal_write_lock, flags);
 367         rc = opal_console_write_buffer_space(vtermno, &olen);
 368         len = be64_to_cpu(olen);
 369         if (rc || len < total_len) {
 370                 spin_unlock_irqrestore(&opal_write_lock, flags);
 371                 /* Closed -> drop characters */
 372                 if (rc)
 373                         return total_len;
 374                 opal_poll_events(NULL);
 375                 return -EAGAIN;
 376         }
 377
 378         /* We still try to handle partial completions, though they
 379          * should no longer happen.
 380          */
 381         rc = OPAL_BUSY;
 382         while(total_len > 0 && (rc == OPAL_BUSY ||
 383                                 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
 384                 olen = cpu_to_be64(total_len);
 385                 rc = opal_console_write(vtermno, &olen, data);
 386                 len = be64_to_cpu(olen);
 387
 388                 /* Closed or other error drop */
 389                 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
 390                     rc != OPAL_BUSY_EVENT) {
 391                         written += total_len;
 392                         break;
 393                 }
 394                 if (rc == OPAL_SUCCESS) {
 395                         total_len -= len;
 396                         data += len;
 397                         written += len;
 398                 }
 399                 /* This is a bit nasty but we need that for the console to
 400                  * flush when there aren't any interrupts. We will clean
 401                  * things a bit later to limit that to synchronous path
 402                  * such as the kernel console and xmon/udbg
 403                  */
 404                 do
 405                         opal_poll_events(&evt);
 406                 while(rc == OPAL_SUCCESS &&
 407                         (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
 408         }
 409         spin_unlock_irqrestore(&opal_write_lock, flags);
 410         return written;
 411 }
 412
 413 static int opal_recover_mce(struct pt_regs *regs,
 414                                         struct machine_check_event *evt)
 415 {
 416         int recovered = 0;
 417
 418         if (!(regs->msr & MSR_RI)) {
 419                 /* If MSR_RI isn't set, we cannot recover */
 420                 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
 421                 recovered = 0;
 422         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
 423                 /* Platform corrected itself */
 424                 recovered = 1;
 425         } else if (evt->severity == MCE_SEV_FATAL) {
 426                 /* Fatal machine check */
 427                 pr_err("Machine check interrupt is fatal\n");
 428                 recovered = 0;
 429         }
 430
 431         if (!recovered && evt->severity == MCE_SEV_ERROR_SYNC) {
 432                 /*
 433                  * Try to kill processes if we get a synchronous machine check
 434                  * (e.g., one caused by execution of this instruction). This
 435                  * will devolve into a panic if we try to kill init or are in
 436                  * an interrupt etc.
 437                  *
 438                  * TODO: Queue up this address for hwpoisioning later.
 439                  * TODO: This is not quite right for d-side machine
 440                  *       checks ->nip is not necessarily the important
 441                  *       address.
 442                  */
 443                 if ((user_mode(regs))) {
 444                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 445                         recovered = 1;
 446                 } else if (die_will_crash()) {
 447                         /*
 448                          * die() would kill the kernel, so better to go via
 449                          * the platform reboot code that will log the
 450                          * machine check.
 451                          */
 452                         recovered = 0;
 453                 } else {
 454                         die("Machine check", regs, SIGBUS);
 455                         recovered = 1;
 456                 }
 457         }
 458
 459         return recovered;
 460 }
 461
 462 void pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
 463 {
 464         /*
 465          * This is mostly taken from kernel/panic.c, but tries to do
 466          * relatively minimal work. Don't use delay functions (TB may
 467          * be broken), don't crash dump (need to set a firmware log),
 468          * don't run notifiers. We do want to get some information to
 469          * Linux console.
 470          */
 471         console_verbose();
 472         bust_spinlocks(1);
 473         pr_emerg("Hardware platform error: %s\n", msg);
 474         if (regs)
 475                 show_regs(regs);
 476         smp_send_stop();
 477         printk_safe_flush_on_panic();
 478         kmsg_dump(KMSG_DUMP_PANIC);
 479         bust_spinlocks(0);
 480         debug_locks_off();
 481         console_flush_on_panic();
 482
 483         /*
 484          * Don't bother to shut things down because this will
 485          * xstop the system.
 486          */
 487         if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
 488                                                 == OPAL_UNSUPPORTED) {
 489                 pr_emerg("Reboot type %d not supported for %s\n",
 490                                 OPAL_REBOOT_PLATFORM_ERROR, msg);
 491         }
 492
 493         /*
 494          * We reached here. There can be three possibilities:
 495          * 1. We are running on a firmware level that do not support
 496          *    opal_cec_reboot2()
 497          * 2. We are running on a firmware level that do not support
 498          *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
 499          * 3. We are running on FSP based system that does not need
 500          *    opal to trigger checkstop explicitly for error analysis.
 501          *    The FSP PRD component would have already got notified
 502          *    about this error through other channels.
 503          */
 504
 505         ppc_md.restart(NULL);
 506 }
 507
 508 int opal_machine_check(struct pt_regs *regs)
 509 {
 510         struct machine_check_event evt;
 511
 512         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 513                 return 0;
 514
 515         /* Print things out */
 516         if (evt.version != MCE_V1) {
 517                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 518                        evt.version);
 519                 return 0;
 520         }
 521         machine_check_print_event_info(&evt, user_mode(regs));
 522
 523         if (opal_recover_mce(regs, &evt))
 524                 return 1;
 525
 526         pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
 527 }
 528
 529 /* Early hmi handler called in real mode. */
 530 int opal_hmi_exception_early(struct pt_regs *regs)
 531 {
 532         s64 rc;
 533
 534         /*
 535          * call opal hmi handler. Pass paca address as token.
 536          * The return value OPAL_SUCCESS is an indication that there is
 537          * an HMI event generated waiting to pull by Linux.
 538          */
 539         rc = opal_handle_hmi();
 540         if (rc == OPAL_SUCCESS) {
 541                 local_paca->hmi_event_available = 1;
 542                 return 1;
 543         }
 544         return 0;
 545 }
 546
 547 /* HMI exception handler called in virtual mode during check_irq_replay. */
 548 int opal_handle_hmi_exception(struct pt_regs *regs)
 549 {
 550         s64 rc;
 551         __be64 evt = 0;
 552
 553         /*
 554          * Check if HMI event is available.
 555          * if Yes, then call opal_poll_events to pull opal messages and
 556          * process them.
 557          */
 558         if (!local_paca->hmi_event_available)
 559                 return 0;
 560
 561         local_paca->hmi_event_available = 0;
 562         rc = opal_poll_events(&evt);
 563         if (rc == OPAL_SUCCESS && evt)
 564                 opal_handle_events(be64_to_cpu(evt));
 565
 566         return 1;
 567 }
 568
 569 static uint64_t find_recovery_address(uint64_t nip)
 570 {
 571         int i;
 572
 573         for (i = 0; i < mc_recoverable_range_len; i++)
 574                 if ((nip >= mc_recoverable_range[i].start_addr) &&
 575                     (nip < mc_recoverable_range[i].end_addr))
 576                     return mc_recoverable_range[i].recover_addr;
 577         return 0;
 578 }
 579
 580 bool opal_mce_check_early_recovery(struct pt_regs *regs)
 581 {
 582         uint64_t recover_addr = 0;
 583
 584         if (!opal.base || !opal.size)
 585                 goto out;
 586
 587         if ((regs->nip >= opal.base) &&
 588                         (regs->nip < (opal.base + opal.size)))
 589                 recover_addr = find_recovery_address(regs->nip);
 590
 591         /*
 592          * Setup regs->nip to rfi into fixup address.
 593          */
 594         if (recover_addr)
 595                 regs->nip = recover_addr;
 596
 597 out:
 598         return !!recover_addr;
 599 }
 600
 601 static int opal_sysfs_init(void)
 602 {
 603         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
 604         if (!opal_kobj) {
 605                 pr_warn("kobject_create_and_add opal failed\n");
 606                 return -ENOMEM;
 607         }
 608
 609         return 0;
 610 }
 611
 612 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
 613                                struct bin_attribute *bin_attr,
 614                                char *buf, loff_t off, size_t count)
 615 {
 616         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 617                                        bin_attr->size);
 618 }
 619
 620 static struct bin_attribute symbol_map_attr = {
 621         .attr = {.name = "symbol_map", .mode = 0400},
 622         .read = symbol_map_read
 623 };
 624
 625 static void opal_export_symmap(void)
 626 {
 627         const __be64 *syms;
 628         unsigned int size;
 629         struct device_node *fw;
 630         int rc;
 631
 632         fw = of_find_node_by_path("/ibm,opal/firmware");
 633         if (!fw)
 634                 return;
 635         syms = of_get_property(fw, "symbol-map", &size);
 636         if (!syms || size != 2 * sizeof(__be64))
 637                 return;
 638
 639         /* Setup attributes */
 640         symbol_map_attr.private = __va(be64_to_cpu(syms[0]));
 641         symbol_map_attr.size = be64_to_cpu(syms[1]);
 642
 643         rc = sysfs_create_bin_file(opal_kobj, &symbol_map_attr);
 644         if (rc)
 645                 pr_warn("Error %d creating OPAL symbols file\n", rc);
 646 }
 647
 648 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
 649                                 struct bin_attribute *bin_attr, char *buf,
 650                                 loff_t off, size_t count)
 651 {
 652         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 653                                        bin_attr->size);
 654 }
 655
 656 /*
 657  * opal_export_attrs: creates a sysfs node for each property listed in
 658  * the device-tree under /ibm,opal/firmware/exports/
 659  * All new sysfs nodes are created under /opal/exports/.
 660  * This allows for reserved memory regions (e.g. HDAT) to be read.
 661  * The new sysfs nodes are only readable by root.
 662  */
 663 static void opal_export_attrs(void)
 664 {
 665         struct bin_attribute *attr;
 666         struct device_node *np;
 667         struct property *prop;
 668         struct kobject *kobj;
 669         u64 vals[2];
 670         int rc;
 671
 672         np = of_find_node_by_path("/ibm,opal/firmware/exports");
 673         if (!np)
 674                 return;
 675
 676         /* Create new 'exports' directory - /sys/firmware/opal/exports */
 677         kobj = kobject_create_and_add("exports", opal_kobj);
 678         if (!kobj) {
 679                 pr_warn("kobject_create_and_add() of exports failed\n");
 680                 of_node_put(np);
 681                 return;
 682         }
 683
 684         for_each_property_of_node(np, prop) {
 685                 if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
 686                         continue;
 687
 688                 if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
 689                         continue;
 690
 691                 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 692
 693                 if (attr == NULL) {
 694                         pr_warn("Failed kmalloc for bin_attribute!");
 695                         continue;
 696                 }
 697
 698                 sysfs_bin_attr_init(attr);
 699                 attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
 700                 attr->attr.mode = 0400;
 701                 attr->read = export_attr_read;
 702                 attr->private = __va(vals[0]);
 703                 attr->size = vals[1];
 704
 705                 if (attr->attr.name == NULL) {
 706                         pr_warn("Failed kstrdup for bin_attribute attr.name");
 707                         kfree(attr);
 708                         continue;
 709                 }
 710
 711                 rc = sysfs_create_bin_file(kobj, attr);
 712                 if (rc) {
 713                         pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
 714                                  rc, prop->name);
 715                         kfree(attr->attr.name);
 716                         kfree(attr);
 717                 }
 718         }
 719
 720         of_node_put(np);
 721 }
 722
 723 static void __init opal_dump_region_init(void)
 724 {
 725         void *addr;
 726         uint64_t size;
 727         int rc;
 728
 729         if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
 730                 return;
 731
 732         /* Register kernel log buffer */
 733         addr = log_buf_addr_get();
 734         if (addr == NULL)
 735                 return;
 736
 737         size = log_buf_len_get();
 738         if (size == 0)
 739                 return;
 740
 741         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
 742                                        __pa(addr), size);
 743         /* Don't warn if this is just an older OPAL that doesn't
 744          * know about that call
 745          */
 746         if (rc && rc != OPAL_UNSUPPORTED)
 747                 pr_warn("DUMP: Failed to register kernel log buffer. "
 748                         "rc = %d\n", rc);
 749 }
 750
 751 static void opal_pdev_init(const char *compatible)
 752 {
 753         struct device_node *np;
 754
 755         for_each_compatible_node(np, NULL, compatible)
 756                 of_platform_device_create(np, NULL, NULL);
 757 }
 758
 759 static void __init opal_imc_init_dev(void)
 760 {
 761         struct device_node *np;
 762
 763         np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
 764         if (np)
 765                 of_platform_device_create(np, NULL, NULL);
 766 }
 767
 768 static int kopald(void *unused)
 769 {
 770         unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
 771         __be64 events;
 772
 773         set_freezable();
 774         do {
 775                 try_to_freeze();
 776                 opal_poll_events(&events);
 777                 opal_handle_events(be64_to_cpu(events));
 778                 schedule_timeout_interruptible(timeout);
 779         } while (!kthread_should_stop());
 780
 781         return 0;
 782 }
 783
 784 void opal_wake_poller(void)
 785 {
 786         if (kopald_tsk)
 787                 wake_up_process(kopald_tsk);
 788 }
 789
 790 static void opal_init_heartbeat(void)
 791 {
 792         /* Old firwmware, we assume the HVC heartbeat is sufficient */
 793         if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
 794                                  &opal_heartbeat) != 0)
 795                 opal_heartbeat = 0;
 796
 797         if (opal_heartbeat)
 798                 kopald_tsk = kthread_run(kopald, NULL, "kopald");
 799 }
 800
 801 static int __init opal_init(void)
 802 {
 803         struct device_node *np, *consoles, *leds;
 804         int rc;
 805
 806         opal_node = of_find_node_by_path("/ibm,opal");
 807         if (!opal_node) {
 808                 pr_warn("Device node not found\n");
 809                 return -ENODEV;
 810         }
 811
 812         /* Register OPAL consoles if any ports */
 813         consoles = of_find_node_by_path("/ibm,opal/consoles");
 814         if (consoles) {
 815                 for_each_child_of_node(consoles, np) {
 816                         if (strcmp(np->name, "serial"))
 817                                 continue;
 818                         of_platform_device_create(np, NULL, NULL);
 819                 }
 820                 of_node_put(consoles);
 821         }
 822
 823         /* Initialise OPAL messaging system */
 824         opal_message_init();
 825
 826         /* Initialise OPAL asynchronous completion interface */
 827         opal_async_comp_init();
 828
 829         /* Initialise OPAL sensor interface */
 830         opal_sensor_init();
 831
 832         /* Initialise OPAL hypervisor maintainence interrupt handling */
 833         opal_hmi_handler_init();
 834
 835         /* Create i2c platform devices */
 836         opal_pdev_init("ibm,opal-i2c");
 837
 838         /* Setup a heatbeat thread if requested by OPAL */
 839         opal_init_heartbeat();
 840
 841         /* Detect In-Memory Collection counters and create devices*/
 842         opal_imc_init_dev();
 843
 844         /* Create leds platform devices */
 845         leds = of_find_node_by_path("/ibm,opal/leds");
 846         if (leds) {
 847                 of_platform_device_create(leds, "opal_leds", NULL);
 848                 of_node_put(leds);
 849         }
 850
 851         /* Initialise OPAL message log interface */
 852         opal_msglog_init();
 853
 854         /* Create "opal" kobject under /sys/firmware */
 855         rc = opal_sysfs_init();
 856         if (rc == 0) {
 857                 /* Export symbol map to userspace */
 858                 opal_export_symmap();
 859                 /* Setup dump region interface */
 860                 opal_dump_region_init();
 861                 /* Setup error log interface */
 862                 rc = opal_elog_init();
 863                 /* Setup code update interface */
 864                 opal_flash_update_init();
 865                 /* Setup platform dump extract interface */
 866                 opal_platform_dump_init();
 867                 /* Setup system parameters interface */
 868                 opal_sys_param_init();
 869                 /* Setup message log sysfs interface. */
 870                 opal_msglog_sysfs_init();
 871         }
 872
 873         /* Export all properties */
 874         opal_export_attrs();
 875
 876         /* Initialize platform devices: IPMI backend, PRD & flash interface */
 877         opal_pdev_init("ibm,opal-ipmi");
 878         opal_pdev_init("ibm,opal-flash");
 879         opal_pdev_init("ibm,opal-prd");
 880
 881         /* Initialise platform device: oppanel interface */
 882         opal_pdev_init("ibm,opal-oppanel");
 883
 884         /* Initialise OPAL kmsg dumper for flushing console on panic */
 885         opal_kmsg_init();
 886
 887         /* Initialise OPAL powercap interface */
 888         opal_powercap_init();
 889
 890         /* Initialise OPAL Power-Shifting-Ratio interface */
 891         opal_psr_init();
 892
 893         /* Initialise OPAL sensor groups */
 894         opal_sensor_groups_init();
 895
 896         return 0;
 897 }
 898 machine_subsys_initcall(powernv, opal_init);
 899
 900 void opal_shutdown(void)
 901 {
 902         long rc = OPAL_BUSY;
 903
 904         opal_event_shutdown();
 905
 906         /*
 907          * Then sync with OPAL which ensure anything that can
 908          * potentially write to our memory has completed such
 909          * as an ongoing dump retrieval
 910          */
 911         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 912                 rc = opal_sync_host_reboot();
 913                 if (rc == OPAL_BUSY)
 914                         opal_poll_events(NULL);
 915                 else
 916                         mdelay(10);
 917         }
 918
 919         /* Unregister memory dump region */
 920         if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
 921                 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
 922 }
 923
 924 /* Export this so that test modules can use it */
 925 EXPORT_SYMBOL_GPL(opal_invalid_call);
 926 EXPORT_SYMBOL_GPL(opal_xscom_read);
 927 EXPORT_SYMBOL_GPL(opal_xscom_write);
 928 EXPORT_SYMBOL_GPL(opal_ipmi_send);
 929 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
 930 EXPORT_SYMBOL_GPL(opal_flash_read);
 931 EXPORT_SYMBOL_GPL(opal_flash_write);
 932 EXPORT_SYMBOL_GPL(opal_flash_erase);
 933 EXPORT_SYMBOL_GPL(opal_prd_msg);
 934
 935 /* Convert a region of vmalloc memory to an opal sg list */
 936 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
 937                                              unsigned long vmalloc_size)
 938 {
 939         struct opal_sg_list *sg, *first = NULL;
 940         unsigned long i = 0;
 941
 942         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
 943         if (!sg)
 944                 goto nomem;
 945
 946         first = sg;
 947
 948         while (vmalloc_size > 0) {
 949                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
 950                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
 951
 952                 sg->entry[i].data = cpu_to_be64(data);
 953                 sg->entry[i].length = cpu_to_be64(length);
 954                 i++;
 955
 956                 if (i >= SG_ENTRIES_PER_NODE) {
 957                         struct opal_sg_list *next;
 958
 959                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
 960                         if (!next)
 961                                 goto nomem;
 962
 963                         sg->length = cpu_to_be64(
 964                                         i * sizeof(struct opal_sg_entry) + 16);
 965                         i = 0;
 966                         sg->next = cpu_to_be64(__pa(next));
 967                         sg = next;
 968                 }
 969
 970                 vmalloc_addr += length;
 971                 vmalloc_size -= length;
 972         }
 973
 974         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
 975
 976         return first;
 977
 978 nomem:
 979         pr_err("%s : Failed to allocate memory\n", __func__);
 980         opal_free_sg_list(first);
 981         return NULL;
 982 }
 983
 984 void opal_free_sg_list(struct opal_sg_list *sg)
 985 {
 986         while (sg) {
 987                 uint64_t next = be64_to_cpu(sg->next);
 988
 989                 kfree(sg);
 990
 991                 if (next)
 992                         sg = __va(next);
 993                 else
 994                         sg = NULL;
 995         }
 996 }
 997
 998 int opal_error_code(int rc)
 999 {
1000         switch (rc) {
1001         case OPAL_SUCCESS:              return 0;
1002
1003         case OPAL_PARAMETER:            return -EINVAL;
1004         case OPAL_ASYNC_COMPLETION:     return -EINPROGRESS;
1005         case OPAL_BUSY_EVENT:           return -EBUSY;
1006         case OPAL_NO_MEM:               return -ENOMEM;
1007         case OPAL_PERMISSION:           return -EPERM;
1008
1009         case OPAL_UNSUPPORTED:          return -EIO;
1010         case OPAL_HARDWARE:             return -EIO;
1011         case OPAL_INTERNAL_ERROR:       return -EIO;
1012         case OPAL_TIMEOUT:              return -ETIMEDOUT;
1013         default:
1014                 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1015                 return -EIO;
1016         }
1017 }
1018
1019 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1020 {
1021         int rc;
1022
1023         if (firmware_has_feature(FW_FEATURE_OPAL)) {
1024                 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1025                 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1026                         pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1027         }
1028 }
1029
1030 EXPORT_SYMBOL_GPL(opal_poll_events);
1031 EXPORT_SYMBOL_GPL(opal_rtc_read);
1032 EXPORT_SYMBOL_GPL(opal_rtc_write);
1033 EXPORT_SYMBOL_GPL(opal_tpo_read);
1034 EXPORT_SYMBOL_GPL(opal_tpo_write);
1035 EXPORT_SYMBOL_GPL(opal_i2c_request);
1036 /* Export these symbols for PowerNV LED class driver */
1037 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1038 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1039 /* Export this symbol for PowerNV Operator Panel class driver */
1040 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1041 /* Export this for KVM */
1042 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1043 EXPORT_SYMBOL_GPL(opal_int_eoi);