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/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_debug("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_phys_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_warn("%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_warn("%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 static int __opal_put_chars(uint32_t vtermno, const char *data, int total_len, bool atomic)
 348 {
 349         unsigned long flags = 0 /* shut up gcc */;
 350         int written;
 351         __be64 olen;
 352         s64 rc;
 353
 354         if (!opal.entry)
 355                 return -ENODEV;
 356
 357         if (atomic)
 358                 spin_lock_irqsave(&opal_write_lock, flags);
 359         rc = opal_console_write_buffer_space(vtermno, &olen);
 360         if (rc || be64_to_cpu(olen) < total_len) {
 361                 /* Closed -> drop characters */
 362                 if (rc)
 363                         written = total_len;
 364                 else
 365                         written = -EAGAIN;
 366                 goto out;
 367         }
 368
 369         /* Should not get a partial write here because space is available. */
 370         olen = cpu_to_be64(total_len);
 371         rc = opal_console_write(vtermno, &olen, data);
 372         if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 373                 if (rc == OPAL_BUSY_EVENT)
 374                         opal_poll_events(NULL);
 375                 written = -EAGAIN;
 376                 goto out;
 377         }
 378
 379         /* Closed or other error drop */
 380         if (rc != OPAL_SUCCESS) {
 381                 written = opal_error_code(rc);
 382                 goto out;
 383         }
 384
 385         written = be64_to_cpu(olen);
 386         if (written < total_len) {
 387                 if (atomic) {
 388                         /* Should not happen */
 389                         pr_warn("atomic console write returned partial "
 390                                 "len=%d written=%d\n", total_len, written);
 391                 }
 392                 if (!written)
 393                         written = -EAGAIN;
 394         }
 395
 396 out:
 397         if (atomic)
 398                 spin_unlock_irqrestore(&opal_write_lock, flags);
 399
 400         return written;
 401 }
 402
 403 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
 404 {
 405         return __opal_put_chars(vtermno, data, total_len, false);
 406 }
 407
 408 /*
 409  * opal_put_chars_atomic will not perform partial-writes. Data will be
 410  * atomically written to the terminal or not at all. This is not strictly
 411  * true at the moment because console space can race with OPAL's console
 412  * writes.
 413  */
 414 int opal_put_chars_atomic(uint32_t vtermno, const char *data, int total_len)
 415 {
 416         return __opal_put_chars(vtermno, data, total_len, true);
 417 }
 418
 419 static s64 __opal_flush_console(uint32_t vtermno)
 420 {
 421         s64 rc;
 422
 423         if (!opal_check_token(OPAL_CONSOLE_FLUSH)) {
 424                 __be64 evt;
 425
 426                 /*
 427                  * If OPAL_CONSOLE_FLUSH is not implemented in the firmware,
 428                  * the console can still be flushed by calling the polling
 429                  * function while it has OPAL_EVENT_CONSOLE_OUTPUT events.
 430                  */
 431                 WARN_ONCE(1, "opal: OPAL_CONSOLE_FLUSH missing.\n");
 432
 433                 opal_poll_events(&evt);
 434                 if (!(be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT))
 435                         return OPAL_SUCCESS;
 436                 return OPAL_BUSY;
 437
 438         } else {
 439                 rc = opal_console_flush(vtermno);
 440                 if (rc == OPAL_BUSY_EVENT) {
 441                         opal_poll_events(NULL);
 442                         rc = OPAL_BUSY;
 443                 }
 444                 return rc;
 445         }
 446
 447 }
 448
 449 /*
 450  * opal_flush_console spins until the console is flushed
 451  */
 452 int opal_flush_console(uint32_t vtermno)
 453 {
 454         for (;;) {
 455                 s64 rc = __opal_flush_console(vtermno);
 456
 457                 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
 458                         mdelay(1);
 459                         continue;
 460                 }
 461
 462                 return opal_error_code(rc);
 463         }
 464 }
 465
 466 /*
 467  * opal_flush_chars is an hvc interface that sleeps until the console is
 468  * flushed if wait, otherwise it will return -EBUSY if the console has data,
 469  * -EAGAIN if it has data and some of it was flushed.
 470  */
 471 int opal_flush_chars(uint32_t vtermno, bool wait)
 472 {
 473         for (;;) {
 474                 s64 rc = __opal_flush_console(vtermno);
 475
 476                 if (rc == OPAL_BUSY || rc == OPAL_PARTIAL) {
 477                         if (wait) {
 478                                 msleep(OPAL_BUSY_DELAY_MS);
 479                                 continue;
 480                         }
 481                         if (rc == OPAL_PARTIAL)
 482                                 return -EAGAIN;
 483                 }
 484
 485                 return opal_error_code(rc);
 486         }
 487 }
 488
 489 static int opal_recover_mce(struct pt_regs *regs,
 490                                         struct machine_check_event *evt)
 491 {
 492         int recovered = 0;
 493
 494         if (!(regs->msr & MSR_RI)) {
 495                 /* If MSR_RI isn't set, we cannot recover */
 496                 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
 497                 recovered = 0;
 498         } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
 499                 /* Platform corrected itself */
 500                 recovered = 1;
 501         } else if (evt->severity == MCE_SEV_FATAL) {
 502                 /* Fatal machine check */
 503                 pr_err("Machine check interrupt is fatal\n");
 504                 recovered = 0;
 505         }
 506
 507         if (!recovered && evt->severity == MCE_SEV_ERROR_SYNC) {
 508                 /*
 509                  * Try to kill processes if we get a synchronous machine check
 510                  * (e.g., one caused by execution of this instruction). This
 511                  * will devolve into a panic if we try to kill init or are in
 512                  * an interrupt etc.
 513                  *
 514                  * TODO: Queue up this address for hwpoisioning later.
 515                  * TODO: This is not quite right for d-side machine
 516                  *       checks ->nip is not necessarily the important
 517                  *       address.
 518                  */
 519                 if ((user_mode(regs))) {
 520                         _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 521                         recovered = 1;
 522                 } else if (die_will_crash()) {
 523                         /*
 524                          * die() would kill the kernel, so better to go via
 525                          * the platform reboot code that will log the
 526                          * machine check.
 527                          */
 528                         recovered = 0;
 529                 } else {
 530                         die("Machine check", regs, SIGBUS);
 531                         recovered = 1;
 532                 }
 533         }
 534
 535         return recovered;
 536 }
 537
 538 void __noreturn pnv_platform_error_reboot(struct pt_regs *regs, const char *msg)
 539 {
 540         panic_flush_kmsg_start();
 541
 542         pr_emerg("Hardware platform error: %s\n", msg);
 543         if (regs)
 544                 show_regs(regs);
 545         smp_send_stop();
 546
 547         panic_flush_kmsg_end();
 548
 549         /*
 550          * Don't bother to shut things down because this will
 551          * xstop the system.
 552          */
 553         if (opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, msg)
 554                                                 == OPAL_UNSUPPORTED) {
 555                 pr_emerg("Reboot type %d not supported for %s\n",
 556                                 OPAL_REBOOT_PLATFORM_ERROR, msg);
 557         }
 558
 559         /*
 560          * We reached here. There can be three possibilities:
 561          * 1. We are running on a firmware level that do not support
 562          *    opal_cec_reboot2()
 563          * 2. We are running on a firmware level that do not support
 564          *    OPAL_REBOOT_PLATFORM_ERROR reboot type.
 565          * 3. We are running on FSP based system that does not need
 566          *    opal to trigger checkstop explicitly for error analysis.
 567          *    The FSP PRD component would have already got notified
 568          *    about this error through other channels.
 569          * 4. We are running on a newer skiboot that by default does
 570          *    not cause a checkstop, drops us back to the kernel to
 571          *    extract context and state at the time of the error.
 572          */
 573
 574         panic(msg);
 575 }
 576
 577 int opal_machine_check(struct pt_regs *regs)
 578 {
 579         struct machine_check_event evt;
 580
 581         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 582                 return 0;
 583
 584         /* Print things out */
 585         if (evt.version != MCE_V1) {
 586                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 587                        evt.version);
 588                 return 0;
 589         }
 590         machine_check_print_event_info(&evt, user_mode(regs));
 591
 592         if (opal_recover_mce(regs, &evt))
 593                 return 1;
 594
 595         pnv_platform_error_reboot(regs, "Unrecoverable Machine Check exception");
 596 }
 597
 598 /* Early hmi handler called in real mode. */
 599 int opal_hmi_exception_early(struct pt_regs *regs)
 600 {
 601         s64 rc;
 602
 603         /*
 604          * call opal hmi handler. Pass paca address as token.
 605          * The return value OPAL_SUCCESS is an indication that there is
 606          * an HMI event generated waiting to pull by Linux.
 607          */
 608         rc = opal_handle_hmi();
 609         if (rc == OPAL_SUCCESS) {
 610                 local_paca->hmi_event_available = 1;
 611                 return 1;
 612         }
 613         return 0;
 614 }
 615
 616 /* HMI exception handler called in virtual mode during check_irq_replay. */
 617 int opal_handle_hmi_exception(struct pt_regs *regs)
 618 {
 619         /*
 620          * Check if HMI event is available.
 621          * if Yes, then wake kopald to process them.
 622          */
 623         if (!local_paca->hmi_event_available)
 624                 return 0;
 625
 626         local_paca->hmi_event_available = 0;
 627         opal_wake_poller();
 628
 629         return 1;
 630 }
 631
 632 static uint64_t find_recovery_address(uint64_t nip)
 633 {
 634         int i;
 635
 636         for (i = 0; i < mc_recoverable_range_len; i++)
 637                 if ((nip >= mc_recoverable_range[i].start_addr) &&
 638                     (nip < mc_recoverable_range[i].end_addr))
 639                     return mc_recoverable_range[i].recover_addr;
 640         return 0;
 641 }
 642
 643 bool opal_mce_check_early_recovery(struct pt_regs *regs)
 644 {
 645         uint64_t recover_addr = 0;
 646
 647         if (!opal.base || !opal.size)
 648                 goto out;
 649
 650         if ((regs->nip >= opal.base) &&
 651                         (regs->nip < (opal.base + opal.size)))
 652                 recover_addr = find_recovery_address(regs->nip);
 653
 654         /*
 655          * Setup regs->nip to rfi into fixup address.
 656          */
 657         if (recover_addr)
 658                 regs->nip = recover_addr;
 659
 660 out:
 661         return !!recover_addr;
 662 }
 663
 664 static int opal_sysfs_init(void)
 665 {
 666         opal_kobj = kobject_create_and_add("opal", firmware_kobj);
 667         if (!opal_kobj) {
 668                 pr_warn("kobject_create_and_add opal failed\n");
 669                 return -ENOMEM;
 670         }
 671
 672         return 0;
 673 }
 674
 675 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
 676                                struct bin_attribute *bin_attr,
 677                                char *buf, loff_t off, size_t count)
 678 {
 679         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 680                                        bin_attr->size);
 681 }
 682
 683 static BIN_ATTR_RO(symbol_map, 0);
 684
 685 static void opal_export_symmap(void)
 686 {
 687         const __be64 *syms;
 688         unsigned int size;
 689         struct device_node *fw;
 690         int rc;
 691
 692         fw = of_find_node_by_path("/ibm,opal/firmware");
 693         if (!fw)
 694                 return;
 695         syms = of_get_property(fw, "symbol-map", &size);
 696         if (!syms || size != 2 * sizeof(__be64))
 697                 return;
 698
 699         /* Setup attributes */
 700         bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
 701         bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
 702
 703         rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
 704         if (rc)
 705                 pr_warn("Error %d creating OPAL symbols file\n", rc);
 706 }
 707
 708 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj,
 709                                 struct bin_attribute *bin_attr, char *buf,
 710                                 loff_t off, size_t count)
 711 {
 712         return memory_read_from_buffer(buf, count, &off, bin_attr->private,
 713                                        bin_attr->size);
 714 }
 715
 716 /*
 717  * opal_export_attrs: creates a sysfs node for each property listed in
 718  * the device-tree under /ibm,opal/firmware/exports/
 719  * All new sysfs nodes are created under /opal/exports/.
 720  * This allows for reserved memory regions (e.g. HDAT) to be read.
 721  * The new sysfs nodes are only readable by root.
 722  */
 723 static void opal_export_attrs(void)
 724 {
 725         struct bin_attribute *attr;
 726         struct device_node *np;
 727         struct property *prop;
 728         struct kobject *kobj;
 729         u64 vals[2];
 730         int rc;
 731
 732         np = of_find_node_by_path("/ibm,opal/firmware/exports");
 733         if (!np)
 734                 return;
 735
 736         /* Create new 'exports' directory - /sys/firmware/opal/exports */
 737         kobj = kobject_create_and_add("exports", opal_kobj);
 738         if (!kobj) {
 739                 pr_warn("kobject_create_and_add() of exports failed\n");
 740                 return;
 741         }
 742
 743         for_each_property_of_node(np, prop) {
 744                 if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle"))
 745                         continue;
 746
 747                 if (of_property_read_u64_array(np, prop->name, &vals[0], 2))
 748                         continue;
 749
 750                 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 751
 752                 if (attr == NULL) {
 753                         pr_warn("Failed kmalloc for bin_attribute!");
 754                         continue;
 755                 }
 756
 757                 sysfs_bin_attr_init(attr);
 758                 attr->attr.name = kstrdup(prop->name, GFP_KERNEL);
 759                 attr->attr.mode = 0400;
 760                 attr->read = export_attr_read;
 761                 attr->private = __va(vals[0]);
 762                 attr->size = vals[1];
 763
 764                 if (attr->attr.name == NULL) {
 765                         pr_warn("Failed kstrdup for bin_attribute attr.name");
 766                         kfree(attr);
 767                         continue;
 768                 }
 769
 770                 rc = sysfs_create_bin_file(kobj, attr);
 771                 if (rc) {
 772                         pr_warn("Error %d creating OPAL sysfs exports/%s file\n",
 773                                  rc, prop->name);
 774                         kfree(attr->attr.name);
 775                         kfree(attr);
 776                 }
 777         }
 778
 779         of_node_put(np);
 780 }
 781
 782 static void __init opal_dump_region_init(void)
 783 {
 784         void *addr;
 785         uint64_t size;
 786         int rc;
 787
 788         if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
 789                 return;
 790
 791         /* Register kernel log buffer */
 792         addr = log_buf_addr_get();
 793         if (addr == NULL)
 794                 return;
 795
 796         size = log_buf_len_get();
 797         if (size == 0)
 798                 return;
 799
 800         rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
 801                                        __pa(addr), size);
 802         /* Don't warn if this is just an older OPAL that doesn't
 803          * know about that call
 804          */
 805         if (rc && rc != OPAL_UNSUPPORTED)
 806                 pr_warn("DUMP: Failed to register kernel log buffer. "
 807                         "rc = %d\n", rc);
 808 }
 809
 810 static void opal_pdev_init(const char *compatible)
 811 {
 812         struct device_node *np;
 813
 814         for_each_compatible_node(np, NULL, compatible)
 815                 of_platform_device_create(np, NULL, NULL);
 816 }
 817
 818 static void __init opal_imc_init_dev(void)
 819 {
 820         struct device_node *np;
 821
 822         np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT);
 823         if (np)
 824                 of_platform_device_create(np, NULL, NULL);
 825 }
 826
 827 static int kopald(void *unused)
 828 {
 829         unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
 830
 831         set_freezable();
 832         do {
 833                 try_to_freeze();
 834
 835                 opal_handle_events();
 836
 837                 set_current_state(TASK_INTERRUPTIBLE);
 838                 if (opal_have_pending_events())
 839                         __set_current_state(TASK_RUNNING);
 840                 else
 841                         schedule_timeout(timeout);
 842
 843         } while (!kthread_should_stop());
 844
 845         return 0;
 846 }
 847
 848 void opal_wake_poller(void)
 849 {
 850         if (kopald_tsk)
 851                 wake_up_process(kopald_tsk);
 852 }
 853
 854 static void opal_init_heartbeat(void)
 855 {
 856         /* Old firwmware, we assume the HVC heartbeat is sufficient */
 857         if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
 858                                  &opal_heartbeat) != 0)
 859                 opal_heartbeat = 0;
 860
 861         if (opal_heartbeat)
 862                 kopald_tsk = kthread_run(kopald, NULL, "kopald");
 863 }
 864
 865 static int __init opal_init(void)
 866 {
 867         struct device_node *np, *consoles, *leds;
 868         int rc;
 869
 870         opal_node = of_find_node_by_path("/ibm,opal");
 871         if (!opal_node) {
 872                 pr_warn("Device node not found\n");
 873                 return -ENODEV;
 874         }
 875
 876         /* Register OPAL consoles if any ports */
 877         consoles = of_find_node_by_path("/ibm,opal/consoles");
 878         if (consoles) {
 879                 for_each_child_of_node(consoles, np) {
 880                         if (strcmp(np->name, "serial"))
 881                                 continue;
 882                         of_platform_device_create(np, NULL, NULL);
 883                 }
 884                 of_node_put(consoles);
 885         }
 886
 887         /* Initialise OPAL messaging system */
 888         opal_message_init();
 889
 890         /* Initialise OPAL asynchronous completion interface */
 891         opal_async_comp_init();
 892
 893         /* Initialise OPAL sensor interface */
 894         opal_sensor_init();
 895
 896         /* Initialise OPAL hypervisor maintainence interrupt handling */
 897         opal_hmi_handler_init();
 898
 899         /* Create i2c platform devices */
 900         opal_pdev_init("ibm,opal-i2c");
 901
 902         /* Handle non-volatile memory devices */
 903         opal_pdev_init("pmem-region");
 904
 905         /* Setup a heatbeat thread if requested by OPAL */
 906         opal_init_heartbeat();
 907
 908         /* Detect In-Memory Collection counters and create devices*/
 909         opal_imc_init_dev();
 910
 911         /* Create leds platform devices */
 912         leds = of_find_node_by_path("/ibm,opal/leds");
 913         if (leds) {
 914                 of_platform_device_create(leds, "opal_leds", NULL);
 915                 of_node_put(leds);
 916         }
 917
 918         /* Initialise OPAL message log interface */
 919         opal_msglog_init();
 920
 921         /* Create "opal" kobject under /sys/firmware */
 922         rc = opal_sysfs_init();
 923         if (rc == 0) {
 924                 /* Export symbol map to userspace */
 925                 opal_export_symmap();
 926                 /* Setup dump region interface */
 927                 opal_dump_region_init();
 928                 /* Setup error log interface */
 929                 rc = opal_elog_init();
 930                 /* Setup code update interface */
 931                 opal_flash_update_init();
 932                 /* Setup platform dump extract interface */
 933                 opal_platform_dump_init();
 934                 /* Setup system parameters interface */
 935                 opal_sys_param_init();
 936                 /* Setup message log sysfs interface. */
 937                 opal_msglog_sysfs_init();
 938         }
 939
 940         /* Export all properties */
 941         opal_export_attrs();
 942
 943         /* Initialize platform devices: IPMI backend, PRD & flash interface */
 944         opal_pdev_init("ibm,opal-ipmi");
 945         opal_pdev_init("ibm,opal-flash");
 946         opal_pdev_init("ibm,opal-prd");
 947
 948         /* Initialise platform device: oppanel interface */
 949         opal_pdev_init("ibm,opal-oppanel");
 950
 951         /* Initialise OPAL kmsg dumper for flushing console on panic */
 952         opal_kmsg_init();
 953
 954         /* Initialise OPAL powercap interface */
 955         opal_powercap_init();
 956
 957         /* Initialise OPAL Power-Shifting-Ratio interface */
 958         opal_psr_init();
 959
 960         /* Initialise OPAL sensor groups */
 961         opal_sensor_groups_init();
 962
 963         return 0;
 964 }
 965 machine_subsys_initcall(powernv, opal_init);
 966
 967 void opal_shutdown(void)
 968 {
 969         long rc = OPAL_BUSY;
 970
 971         opal_event_shutdown();
 972
 973         /*
 974          * Then sync with OPAL which ensure anything that can
 975          * potentially write to our memory has completed such
 976          * as an ongoing dump retrieval
 977          */
 978         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 979                 rc = opal_sync_host_reboot();
 980                 if (rc == OPAL_BUSY)
 981                         opal_poll_events(NULL);
 982                 else
 983                         mdelay(10);
 984         }
 985
 986         /* Unregister memory dump region */
 987         if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
 988                 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
 989 }
 990
 991 /* Export this so that test modules can use it */
 992 EXPORT_SYMBOL_GPL(opal_invalid_call);
 993 EXPORT_SYMBOL_GPL(opal_xscom_read);
 994 EXPORT_SYMBOL_GPL(opal_xscom_write);
 995 EXPORT_SYMBOL_GPL(opal_ipmi_send);
 996 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
 997 EXPORT_SYMBOL_GPL(opal_flash_read);
 998 EXPORT_SYMBOL_GPL(opal_flash_write);
 999 EXPORT_SYMBOL_GPL(opal_flash_erase);
1000 EXPORT_SYMBOL_GPL(opal_prd_msg);
1001 EXPORT_SYMBOL_GPL(opal_check_token);
1002
1003 /* Convert a region of vmalloc memory to an opal sg list */
1004 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
1005                                              unsigned long vmalloc_size)
1006 {
1007         struct opal_sg_list *sg, *first = NULL;
1008         unsigned long i = 0;
1009
1010         sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
1011         if (!sg)
1012                 goto nomem;
1013
1014         first = sg;
1015
1016         while (vmalloc_size > 0) {
1017                 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
1018                 uint64_t length = min(vmalloc_size, PAGE_SIZE);
1019
1020                 sg->entry[i].data = cpu_to_be64(data);
1021                 sg->entry[i].length = cpu_to_be64(length);
1022                 i++;
1023
1024                 if (i >= SG_ENTRIES_PER_NODE) {
1025                         struct opal_sg_list *next;
1026
1027                         next = kzalloc(PAGE_SIZE, GFP_KERNEL);
1028                         if (!next)
1029                                 goto nomem;
1030
1031                         sg->length = cpu_to_be64(
1032                                         i * sizeof(struct opal_sg_entry) + 16);
1033                         i = 0;
1034                         sg->next = cpu_to_be64(__pa(next));
1035                         sg = next;
1036                 }
1037
1038                 vmalloc_addr += length;
1039                 vmalloc_size -= length;
1040         }
1041
1042         sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
1043
1044         return first;
1045
1046 nomem:
1047         pr_err("%s : Failed to allocate memory\n", __func__);
1048         opal_free_sg_list(first);
1049         return NULL;
1050 }
1051
1052 void opal_free_sg_list(struct opal_sg_list *sg)
1053 {
1054         while (sg) {
1055                 uint64_t next = be64_to_cpu(sg->next);
1056
1057                 kfree(sg);
1058
1059                 if (next)
1060                         sg = __va(next);
1061                 else
1062                         sg = NULL;
1063         }
1064 }
1065
1066 int opal_error_code(int rc)
1067 {
1068         switch (rc) {
1069         case OPAL_SUCCESS:              return 0;
1070
1071         case OPAL_PARAMETER:            return -EINVAL;
1072         case OPAL_ASYNC_COMPLETION:     return -EINPROGRESS;
1073         case OPAL_BUSY:
1074         case OPAL_BUSY_EVENT:           return -EBUSY;
1075         case OPAL_NO_MEM:               return -ENOMEM;
1076         case OPAL_PERMISSION:           return -EPERM;
1077
1078         case OPAL_UNSUPPORTED:          return -EIO;
1079         case OPAL_HARDWARE:             return -EIO;
1080         case OPAL_INTERNAL_ERROR:       return -EIO;
1081         case OPAL_TIMEOUT:              return -ETIMEDOUT;
1082         default:
1083                 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
1084                 return -EIO;
1085         }
1086 }
1087
1088 void powernv_set_nmmu_ptcr(unsigned long ptcr)
1089 {
1090         int rc;
1091
1092         if (firmware_has_feature(FW_FEATURE_OPAL)) {
1093                 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
1094                 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
1095                         pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
1096         }
1097 }
1098
1099 EXPORT_SYMBOL_GPL(opal_poll_events);
1100 EXPORT_SYMBOL_GPL(opal_rtc_read);
1101 EXPORT_SYMBOL_GPL(opal_rtc_write);
1102 EXPORT_SYMBOL_GPL(opal_tpo_read);
1103 EXPORT_SYMBOL_GPL(opal_tpo_write);
1104 EXPORT_SYMBOL_GPL(opal_i2c_request);
1105 /* Export these symbols for PowerNV LED class driver */
1106 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
1107 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
1108 /* Export this symbol for PowerNV Operator Panel class driver */
1109 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
1110 /* Export this for KVM */
1111 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);
1112 EXPORT_SYMBOL_GPL(opal_int_eoi);
1113 EXPORT_SYMBOL_GPL(opal_error_code);
1114 /* Export the below symbol for NX compression */
1115 EXPORT_SYMBOL(opal_nx_coproc_init);