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 30 #include <asm/machdep.h> 31 #include <asm/opal.h> 32 #include <asm/firmware.h> 33 #include <asm/mce.h> 34 #include <asm/imc-pmu.h> 35 36 #include "powernv.h" 37 38 /* /sys/firmware/opal */ 39 struct kobject *opal_kobj; 40 41 struct opal { 42 u64 base; 43 u64 entry; 44 u64 size; 45 } opal; 46 47 struct mcheck_recoverable_range { 48 u64 start_addr; 49 u64 end_addr; 50 u64 recover_addr; 51 }; 52 53 static struct mcheck_recoverable_range *mc_recoverable_range; 54 static int mc_recoverable_range_len; 55 56 struct device_node *opal_node; 57 static DEFINE_SPINLOCK(opal_write_lock); 58 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX]; 59 static uint32_t opal_heartbeat; 60 static struct task_struct *kopald_tsk; 61 62 void opal_configure_cores(void) 63 { 64 u64 reinit_flags = 0; 65 66 /* Do the actual re-init, This will clobber all FPRs, VRs, etc... 67 * 68 * It will preserve non volatile GPRs and HSPRG0/1. It will 69 * also restore HIDs and other SPRs to their original value 70 * but it might clobber a bunch. 71 */ 72 #ifdef __BIG_ENDIAN__ 73 reinit_flags |= OPAL_REINIT_CPUS_HILE_BE; 74 #else 75 reinit_flags |= OPAL_REINIT_CPUS_HILE_LE; 76 #endif 77 78 /* 79 * POWER9 always support running hash: 80 * ie. Host hash supports hash guests 81 * Host radix supports hash/radix guests 82 */ 83 if (early_cpu_has_feature(CPU_FTR_ARCH_300)) { 84 reinit_flags |= OPAL_REINIT_CPUS_MMU_HASH; 85 if (early_radix_enabled()) 86 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX; 87 } 88 89 opal_reinit_cpus(reinit_flags); 90 91 /* Restore some bits */ 92 if (cur_cpu_spec->cpu_restore) 93 cur_cpu_spec->cpu_restore(); 94 } 95 96 int __init early_init_dt_scan_opal(unsigned long node, 97 const char *uname, int depth, void *data) 98 { 99 const void *basep, *entryp, *sizep; 100 int basesz, entrysz, runtimesz; 101 102 if (depth != 1 || strcmp(uname, "ibm,opal") != 0) 103 return 0; 104 105 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz); 106 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz); 107 sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz); 108 109 if (!basep || !entryp || !sizep) 110 return 1; 111 112 opal.base = of_read_number(basep, basesz/4); 113 opal.entry = of_read_number(entryp, entrysz/4); 114 opal.size = of_read_number(sizep, runtimesz/4); 115 116 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%d)\n", 117 opal.base, basep, basesz); 118 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n", 119 opal.entry, entryp, entrysz); 120 pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n", 121 opal.size, sizep, runtimesz); 122 123 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) { 124 powerpc_firmware_features |= FW_FEATURE_OPAL; 125 pr_info("OPAL detected !\n"); 126 } else { 127 panic("OPAL != V3 detected, no longer supported.\n"); 128 } 129 130 return 1; 131 } 132 133 int __init early_init_dt_scan_recoverable_ranges(unsigned long node, 134 const char *uname, int depth, void *data) 135 { 136 int i, psize, size; 137 const __be32 *prop; 138 139 if (depth != 1 || strcmp(uname, "ibm,opal") != 0) 140 return 0; 141 142 prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize); 143 144 if (!prop) 145 return 1; 146 147 pr_debug("Found machine check recoverable ranges.\n"); 148 149 /* 150 * Calculate number of available entries. 151 * 152 * Each recoverable address range entry is (start address, len, 153 * recovery address), 2 cells each for start and recovery address, 154 * 1 cell for len, totalling 5 cells per entry. 155 */ 156 mc_recoverable_range_len = psize / (sizeof(*prop) * 5); 157 158 /* Sanity check */ 159 if (!mc_recoverable_range_len) 160 return 1; 161 162 /* Size required to hold all the entries. */ 163 size = mc_recoverable_range_len * 164 sizeof(struct mcheck_recoverable_range); 165 166 /* 167 * Allocate a buffer to hold the MC recoverable ranges. We would be 168 * accessing them in real mode, hence it needs to be within 169 * RMO region. 170 */ 171 mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64), 172 ppc64_rma_size)); 173 memset(mc_recoverable_range, 0, size); 174 175 for (i = 0; i < mc_recoverable_range_len; i++) { 176 mc_recoverable_range[i].start_addr = 177 of_read_number(prop + (i * 5) + 0, 2); 178 mc_recoverable_range[i].end_addr = 179 mc_recoverable_range[i].start_addr + 180 of_read_number(prop + (i * 5) + 2, 1); 181 mc_recoverable_range[i].recover_addr = 182 of_read_number(prop + (i * 5) + 3, 2); 183 184 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n", 185 mc_recoverable_range[i].start_addr, 186 mc_recoverable_range[i].end_addr, 187 mc_recoverable_range[i].recover_addr); 188 } 189 return 1; 190 } 191 192 static int __init opal_register_exception_handlers(void) 193 { 194 #ifdef __BIG_ENDIAN__ 195 u64 glue; 196 197 if (!(powerpc_firmware_features & FW_FEATURE_OPAL)) 198 return -ENODEV; 199 200 /* Hookup some exception handlers except machine check. We use the 201 * fwnmi area at 0x7000 to provide the glue space to OPAL 202 */ 203 glue = 0x7000; 204 205 /* 206 * Check if we are running on newer firmware that exports 207 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch 208 * the HMI interrupt and we catch it directly in Linux. 209 * 210 * For older firmware (i.e currently released POWER8 System Firmware 211 * as of today <= SV810_087), we fallback to old behavior and let OPAL 212 * patch the HMI vector and handle it inside OPAL firmware. 213 * 214 * For newer firmware (in development/yet to be released) we will 215 * start catching/handling HMI directly in Linux. 216 */ 217 if (!opal_check_token(OPAL_HANDLE_HMI)) { 218 pr_info("Old firmware detected, OPAL handles HMIs.\n"); 219 opal_register_exception_handler( 220 OPAL_HYPERVISOR_MAINTENANCE_HANDLER, 221 0, glue); 222 glue += 128; 223 } 224 225 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue); 226 #endif 227 228 return 0; 229 } 230 machine_early_initcall(powernv, opal_register_exception_handlers); 231 232 /* 233 * Opal message notifier based on message type. Allow subscribers to get 234 * notified for specific messgae type. 235 */ 236 int opal_message_notifier_register(enum opal_msg_type msg_type, 237 struct notifier_block *nb) 238 { 239 if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) { 240 pr_warning("%s: Invalid arguments, msg_type:%d\n", 241 __func__, msg_type); 242 return -EINVAL; 243 } 244 245 return atomic_notifier_chain_register( 246 &opal_msg_notifier_head[msg_type], nb); 247 } 248 EXPORT_SYMBOL_GPL(opal_message_notifier_register); 249 250 int opal_message_notifier_unregister(enum opal_msg_type msg_type, 251 struct notifier_block *nb) 252 { 253 return atomic_notifier_chain_unregister( 254 &opal_msg_notifier_head[msg_type], nb); 255 } 256 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister); 257 258 static void opal_message_do_notify(uint32_t msg_type, void *msg) 259 { 260 /* notify subscribers */ 261 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type], 262 msg_type, msg); 263 } 264 265 static void opal_handle_message(void) 266 { 267 s64 ret; 268 /* 269 * TODO: pre-allocate a message buffer depending on opal-msg-size 270 * value in /proc/device-tree. 271 */ 272 static struct opal_msg msg; 273 u32 type; 274 275 ret = opal_get_msg(__pa(&msg), sizeof(msg)); 276 /* No opal message pending. */ 277 if (ret == OPAL_RESOURCE) 278 return; 279 280 /* check for errors. */ 281 if (ret) { 282 pr_warning("%s: Failed to retrieve opal message, err=%lld\n", 283 __func__, ret); 284 return; 285 } 286 287 type = be32_to_cpu(msg.msg_type); 288 289 /* Sanity check */ 290 if (type >= OPAL_MSG_TYPE_MAX) { 291 pr_warn_once("%s: Unknown message type: %u\n", __func__, type); 292 return; 293 } 294 opal_message_do_notify(type, (void *)&msg); 295 } 296 297 static irqreturn_t opal_message_notify(int irq, void *data) 298 { 299 opal_handle_message(); 300 return IRQ_HANDLED; 301 } 302 303 static int __init opal_message_init(void) 304 { 305 int ret, i, irq; 306 307 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++) 308 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]); 309 310 irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING)); 311 if (!irq) { 312 pr_err("%s: Can't register OPAL event irq (%d)\n", 313 __func__, irq); 314 return irq; 315 } 316 317 ret = request_irq(irq, opal_message_notify, 318 IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL); 319 if (ret) { 320 pr_err("%s: Can't request OPAL event irq (%d)\n", 321 __func__, ret); 322 return ret; 323 } 324 325 return 0; 326 } 327 328 int opal_get_chars(uint32_t vtermno, char *buf, int count) 329 { 330 s64 rc; 331 __be64 evt, len; 332 333 if (!opal.entry) 334 return -ENODEV; 335 opal_poll_events(&evt); 336 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0) 337 return 0; 338 len = cpu_to_be64(count); 339 rc = opal_console_read(vtermno, &len, buf); 340 if (rc == OPAL_SUCCESS) 341 return be64_to_cpu(len); 342 return 0; 343 } 344 345 int opal_put_chars(uint32_t vtermno, const char *data, int total_len) 346 { 347 int written = 0; 348 __be64 olen; 349 s64 len, rc; 350 unsigned long flags; 351 __be64 evt; 352 353 if (!opal.entry) 354 return -ENODEV; 355 356 /* We want put_chars to be atomic to avoid mangling of hvsi 357 * packets. To do that, we first test for room and return 358 * -EAGAIN if there isn't enough. 359 * 360 * Unfortunately, opal_console_write_buffer_space() doesn't 361 * appear to work on opal v1, so we just assume there is 362 * enough room and be done with it 363 */ 364 spin_lock_irqsave(&opal_write_lock, flags); 365 rc = opal_console_write_buffer_space(vtermno, &olen); 366 len = be64_to_cpu(olen); 367 if (rc || len < total_len) { 368 spin_unlock_irqrestore(&opal_write_lock, flags); 369 /* Closed -> drop characters */ 370 if (rc) 371 return total_len; 372 opal_poll_events(NULL); 373 return -EAGAIN; 374 } 375 376 /* We still try to handle partial completions, though they 377 * should no longer happen. 378 */ 379 rc = OPAL_BUSY; 380 while(total_len > 0 && (rc == OPAL_BUSY || 381 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) { 382 olen = cpu_to_be64(total_len); 383 rc = opal_console_write(vtermno, &olen, data); 384 len = be64_to_cpu(olen); 385 386 /* Closed or other error drop */ 387 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY && 388 rc != OPAL_BUSY_EVENT) { 389 written = total_len; 390 break; 391 } 392 if (rc == OPAL_SUCCESS) { 393 total_len -= len; 394 data += len; 395 written += len; 396 } 397 /* This is a bit nasty but we need that for the console to 398 * flush when there aren't any interrupts. We will clean 399 * things a bit later to limit that to synchronous path 400 * such as the kernel console and xmon/udbg 401 */ 402 do 403 opal_poll_events(&evt); 404 while(rc == OPAL_SUCCESS && 405 (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT)); 406 } 407 spin_unlock_irqrestore(&opal_write_lock, flags); 408 return written; 409 } 410 411 static int opal_recover_mce(struct pt_regs *regs, 412 struct machine_check_event *evt) 413 { 414 int recovered = 0; 415 416 if (!(regs->msr & MSR_RI)) { 417 /* If MSR_RI isn't set, we cannot recover */ 418 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n"); 419 recovered = 0; 420 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) { 421 /* Platform corrected itself */ 422 recovered = 1; 423 } else if (evt->severity == MCE_SEV_FATAL) { 424 /* Fatal machine check */ 425 pr_err("Machine check interrupt is fatal\n"); 426 recovered = 0; 427 } else if ((evt->severity == MCE_SEV_ERROR_SYNC) && 428 (user_mode(regs) && !is_global_init(current))) { 429 /* 430 * For now, kill the task if we have received exception when 431 * in userspace. 432 * 433 * TODO: Queue up this address for hwpoisioning later. 434 */ 435 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 436 recovered = 1; 437 } 438 return recovered; 439 } 440 441 int opal_machine_check(struct pt_regs *regs) 442 { 443 struct machine_check_event evt; 444 int ret; 445 446 if (!get_mce_event(&evt, MCE_EVENT_RELEASE)) 447 return 0; 448 449 /* Print things out */ 450 if (evt.version != MCE_V1) { 451 pr_err("Machine Check Exception, Unknown event version %d !\n", 452 evt.version); 453 return 0; 454 } 455 machine_check_print_event_info(&evt, user_mode(regs)); 456 457 if (opal_recover_mce(regs, &evt)) 458 return 1; 459 460 /* 461 * Unrecovered machine check, we are heading to panic path. 462 * 463 * We may have hit this MCE in very early stage of kernel 464 * initialization even before opal-prd has started running. If 465 * this is the case then this MCE error may go un-noticed or 466 * un-analyzed if we go down panic path. We need to inform 467 * BMC/OCC about this error so that they can collect relevant 468 * data for error analysis before rebooting. 469 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so. 470 * This function may not return on BMC based system. 471 */ 472 ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, 473 "Unrecoverable Machine Check exception"); 474 if (ret == OPAL_UNSUPPORTED) { 475 pr_emerg("Reboot type %d not supported\n", 476 OPAL_REBOOT_PLATFORM_ERROR); 477 } 478 479 /* 480 * We reached here. There can be three possibilities: 481 * 1. We are running on a firmware level that do not support 482 * opal_cec_reboot2() 483 * 2. We are running on a firmware level that do not support 484 * OPAL_REBOOT_PLATFORM_ERROR reboot type. 485 * 3. We are running on FSP based system that does not need opal 486 * to trigger checkstop explicitly for error analysis. The FSP 487 * PRD component would have already got notified about this 488 * error through other channels. 489 * 490 * If hardware marked this as an unrecoverable MCE, we are 491 * going to panic anyway. Even if it didn't, it's not safe to 492 * continue at this point, so we should explicitly panic. 493 */ 494 495 panic("PowerNV Unrecovered Machine Check"); 496 return 0; 497 } 498 499 /* Early hmi handler called in real mode. */ 500 int opal_hmi_exception_early(struct pt_regs *regs) 501 { 502 s64 rc; 503 504 /* 505 * call opal hmi handler. Pass paca address as token. 506 * The return value OPAL_SUCCESS is an indication that there is 507 * an HMI event generated waiting to pull by Linux. 508 */ 509 rc = opal_handle_hmi(); 510 if (rc == OPAL_SUCCESS) { 511 local_paca->hmi_event_available = 1; 512 return 1; 513 } 514 return 0; 515 } 516 517 /* HMI exception handler called in virtual mode during check_irq_replay. */ 518 int opal_handle_hmi_exception(struct pt_regs *regs) 519 { 520 s64 rc; 521 __be64 evt = 0; 522 523 /* 524 * Check if HMI event is available. 525 * if Yes, then call opal_poll_events to pull opal messages and 526 * process them. 527 */ 528 if (!local_paca->hmi_event_available) 529 return 0; 530 531 local_paca->hmi_event_available = 0; 532 rc = opal_poll_events(&evt); 533 if (rc == OPAL_SUCCESS && evt) 534 opal_handle_events(be64_to_cpu(evt)); 535 536 return 1; 537 } 538 539 static uint64_t find_recovery_address(uint64_t nip) 540 { 541 int i; 542 543 for (i = 0; i < mc_recoverable_range_len; i++) 544 if ((nip >= mc_recoverable_range[i].start_addr) && 545 (nip < mc_recoverable_range[i].end_addr)) 546 return mc_recoverable_range[i].recover_addr; 547 return 0; 548 } 549 550 bool opal_mce_check_early_recovery(struct pt_regs *regs) 551 { 552 uint64_t recover_addr = 0; 553 554 if (!opal.base || !opal.size) 555 goto out; 556 557 if ((regs->nip >= opal.base) && 558 (regs->nip < (opal.base + opal.size))) 559 recover_addr = find_recovery_address(regs->nip); 560 561 /* 562 * Setup regs->nip to rfi into fixup address. 563 */ 564 if (recover_addr) 565 regs->nip = recover_addr; 566 567 out: 568 return !!recover_addr; 569 } 570 571 static int opal_sysfs_init(void) 572 { 573 opal_kobj = kobject_create_and_add("opal", firmware_kobj); 574 if (!opal_kobj) { 575 pr_warn("kobject_create_and_add opal failed\n"); 576 return -ENOMEM; 577 } 578 579 return 0; 580 } 581 582 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj, 583 struct bin_attribute *bin_attr, 584 char *buf, loff_t off, size_t count) 585 { 586 return memory_read_from_buffer(buf, count, &off, bin_attr->private, 587 bin_attr->size); 588 } 589 590 static BIN_ATTR_RO(symbol_map, 0); 591 592 static void opal_export_symmap(void) 593 { 594 const __be64 *syms; 595 unsigned int size; 596 struct device_node *fw; 597 int rc; 598 599 fw = of_find_node_by_path("/ibm,opal/firmware"); 600 if (!fw) 601 return; 602 syms = of_get_property(fw, "symbol-map", &size); 603 if (!syms || size != 2 * sizeof(__be64)) 604 return; 605 606 /* Setup attributes */ 607 bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0])); 608 bin_attr_symbol_map.size = be64_to_cpu(syms[1]); 609 610 rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map); 611 if (rc) 612 pr_warn("Error %d creating OPAL symbols file\n", rc); 613 } 614 615 static ssize_t export_attr_read(struct file *fp, struct kobject *kobj, 616 struct bin_attribute *bin_attr, char *buf, 617 loff_t off, size_t count) 618 { 619 return memory_read_from_buffer(buf, count, &off, bin_attr->private, 620 bin_attr->size); 621 } 622 623 /* 624 * opal_export_attrs: creates a sysfs node for each property listed in 625 * the device-tree under /ibm,opal/firmware/exports/ 626 * All new sysfs nodes are created under /opal/exports/. 627 * This allows for reserved memory regions (e.g. HDAT) to be read. 628 * The new sysfs nodes are only readable by root. 629 */ 630 static void opal_export_attrs(void) 631 { 632 struct bin_attribute *attr; 633 struct device_node *np; 634 struct property *prop; 635 struct kobject *kobj; 636 u64 vals[2]; 637 int rc; 638 639 np = of_find_node_by_path("/ibm,opal/firmware/exports"); 640 if (!np) 641 return; 642 643 /* Create new 'exports' directory - /sys/firmware/opal/exports */ 644 kobj = kobject_create_and_add("exports", opal_kobj); 645 if (!kobj) { 646 pr_warn("kobject_create_and_add() of exports failed\n"); 647 return; 648 } 649 650 for_each_property_of_node(np, prop) { 651 if (!strcmp(prop->name, "name") || !strcmp(prop->name, "phandle")) 652 continue; 653 654 if (of_property_read_u64_array(np, prop->name, &vals[0], 2)) 655 continue; 656 657 attr = kzalloc(sizeof(*attr), GFP_KERNEL); 658 659 if (attr == NULL) { 660 pr_warn("Failed kmalloc for bin_attribute!"); 661 continue; 662 } 663 664 sysfs_bin_attr_init(attr); 665 attr->attr.name = kstrdup(prop->name, GFP_KERNEL); 666 attr->attr.mode = 0400; 667 attr->read = export_attr_read; 668 attr->private = __va(vals[0]); 669 attr->size = vals[1]; 670 671 if (attr->attr.name == NULL) { 672 pr_warn("Failed kstrdup for bin_attribute attr.name"); 673 kfree(attr); 674 continue; 675 } 676 677 rc = sysfs_create_bin_file(kobj, attr); 678 if (rc) { 679 pr_warn("Error %d creating OPAL sysfs exports/%s file\n", 680 rc, prop->name); 681 kfree(attr->attr.name); 682 kfree(attr); 683 } 684 } 685 686 of_node_put(np); 687 } 688 689 static void __init opal_dump_region_init(void) 690 { 691 void *addr; 692 uint64_t size; 693 int rc; 694 695 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION)) 696 return; 697 698 /* Register kernel log buffer */ 699 addr = log_buf_addr_get(); 700 if (addr == NULL) 701 return; 702 703 size = log_buf_len_get(); 704 if (size == 0) 705 return; 706 707 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF, 708 __pa(addr), size); 709 /* Don't warn if this is just an older OPAL that doesn't 710 * know about that call 711 */ 712 if (rc && rc != OPAL_UNSUPPORTED) 713 pr_warn("DUMP: Failed to register kernel log buffer. " 714 "rc = %d\n", rc); 715 } 716 717 static void opal_pdev_init(const char *compatible) 718 { 719 struct device_node *np; 720 721 for_each_compatible_node(np, NULL, compatible) 722 of_platform_device_create(np, NULL, NULL); 723 } 724 725 static void __init opal_imc_init_dev(void) 726 { 727 struct device_node *np; 728 729 np = of_find_compatible_node(NULL, NULL, IMC_DTB_COMPAT); 730 if (np) 731 of_platform_device_create(np, NULL, NULL); 732 } 733 734 static int kopald(void *unused) 735 { 736 unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1; 737 __be64 events; 738 739 set_freezable(); 740 do { 741 try_to_freeze(); 742 opal_poll_events(&events); 743 opal_handle_events(be64_to_cpu(events)); 744 schedule_timeout_interruptible(timeout); 745 } while (!kthread_should_stop()); 746 747 return 0; 748 } 749 750 void opal_wake_poller(void) 751 { 752 if (kopald_tsk) 753 wake_up_process(kopald_tsk); 754 } 755 756 static void opal_init_heartbeat(void) 757 { 758 /* Old firwmware, we assume the HVC heartbeat is sufficient */ 759 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms", 760 &opal_heartbeat) != 0) 761 opal_heartbeat = 0; 762 763 if (opal_heartbeat) 764 kopald_tsk = kthread_run(kopald, NULL, "kopald"); 765 } 766 767 static int __init opal_init(void) 768 { 769 struct device_node *np, *consoles, *leds; 770 int rc; 771 772 opal_node = of_find_node_by_path("/ibm,opal"); 773 if (!opal_node) { 774 pr_warn("Device node not found\n"); 775 return -ENODEV; 776 } 777 778 /* Register OPAL consoles if any ports */ 779 consoles = of_find_node_by_path("/ibm,opal/consoles"); 780 if (consoles) { 781 for_each_child_of_node(consoles, np) { 782 if (strcmp(np->name, "serial")) 783 continue; 784 of_platform_device_create(np, NULL, NULL); 785 } 786 of_node_put(consoles); 787 } 788 789 /* Initialise OPAL messaging system */ 790 opal_message_init(); 791 792 /* Initialise OPAL asynchronous completion interface */ 793 opal_async_comp_init(); 794 795 /* Initialise OPAL sensor interface */ 796 opal_sensor_init(); 797 798 /* Initialise OPAL hypervisor maintainence interrupt handling */ 799 opal_hmi_handler_init(); 800 801 /* Create i2c platform devices */ 802 opal_pdev_init("ibm,opal-i2c"); 803 804 /* Setup a heatbeat thread if requested by OPAL */ 805 opal_init_heartbeat(); 806 807 /* Detect In-Memory Collection counters and create devices*/ 808 opal_imc_init_dev(); 809 810 /* Create leds platform devices */ 811 leds = of_find_node_by_path("/ibm,opal/leds"); 812 if (leds) { 813 of_platform_device_create(leds, "opal_leds", NULL); 814 of_node_put(leds); 815 } 816 817 /* Initialise OPAL message log interface */ 818 opal_msglog_init(); 819 820 /* Create "opal" kobject under /sys/firmware */ 821 rc = opal_sysfs_init(); 822 if (rc == 0) { 823 /* Export symbol map to userspace */ 824 opal_export_symmap(); 825 /* Setup dump region interface */ 826 opal_dump_region_init(); 827 /* Setup error log interface */ 828 rc = opal_elog_init(); 829 /* Setup code update interface */ 830 opal_flash_update_init(); 831 /* Setup platform dump extract interface */ 832 opal_platform_dump_init(); 833 /* Setup system parameters interface */ 834 opal_sys_param_init(); 835 /* Setup message log sysfs interface. */ 836 opal_msglog_sysfs_init(); 837 } 838 839 /* Export all properties */ 840 opal_export_attrs(); 841 842 /* Initialize platform devices: IPMI backend, PRD & flash interface */ 843 opal_pdev_init("ibm,opal-ipmi"); 844 opal_pdev_init("ibm,opal-flash"); 845 opal_pdev_init("ibm,opal-prd"); 846 847 /* Initialise platform device: oppanel interface */ 848 opal_pdev_init("ibm,opal-oppanel"); 849 850 /* Initialise OPAL kmsg dumper for flushing console on panic */ 851 opal_kmsg_init(); 852 853 /* Initialise OPAL powercap interface */ 854 opal_powercap_init(); 855 856 /* Initialise OPAL Power-Shifting-Ratio interface */ 857 opal_psr_init(); 858 859 return 0; 860 } 861 machine_subsys_initcall(powernv, opal_init); 862 863 void opal_shutdown(void) 864 { 865 long rc = OPAL_BUSY; 866 867 opal_event_shutdown(); 868 869 /* 870 * Then sync with OPAL which ensure anything that can 871 * potentially write to our memory has completed such 872 * as an ongoing dump retrieval 873 */ 874 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) { 875 rc = opal_sync_host_reboot(); 876 if (rc == OPAL_BUSY) 877 opal_poll_events(NULL); 878 else 879 mdelay(10); 880 } 881 882 /* Unregister memory dump region */ 883 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION)) 884 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF); 885 } 886 887 /* Export this so that test modules can use it */ 888 EXPORT_SYMBOL_GPL(opal_invalid_call); 889 EXPORT_SYMBOL_GPL(opal_xscom_read); 890 EXPORT_SYMBOL_GPL(opal_xscom_write); 891 EXPORT_SYMBOL_GPL(opal_ipmi_send); 892 EXPORT_SYMBOL_GPL(opal_ipmi_recv); 893 EXPORT_SYMBOL_GPL(opal_flash_read); 894 EXPORT_SYMBOL_GPL(opal_flash_write); 895 EXPORT_SYMBOL_GPL(opal_flash_erase); 896 EXPORT_SYMBOL_GPL(opal_prd_msg); 897 898 /* Convert a region of vmalloc memory to an opal sg list */ 899 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr, 900 unsigned long vmalloc_size) 901 { 902 struct opal_sg_list *sg, *first = NULL; 903 unsigned long i = 0; 904 905 sg = kzalloc(PAGE_SIZE, GFP_KERNEL); 906 if (!sg) 907 goto nomem; 908 909 first = sg; 910 911 while (vmalloc_size > 0) { 912 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT; 913 uint64_t length = min(vmalloc_size, PAGE_SIZE); 914 915 sg->entry[i].data = cpu_to_be64(data); 916 sg->entry[i].length = cpu_to_be64(length); 917 i++; 918 919 if (i >= SG_ENTRIES_PER_NODE) { 920 struct opal_sg_list *next; 921 922 next = kzalloc(PAGE_SIZE, GFP_KERNEL); 923 if (!next) 924 goto nomem; 925 926 sg->length = cpu_to_be64( 927 i * sizeof(struct opal_sg_entry) + 16); 928 i = 0; 929 sg->next = cpu_to_be64(__pa(next)); 930 sg = next; 931 } 932 933 vmalloc_addr += length; 934 vmalloc_size -= length; 935 } 936 937 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16); 938 939 return first; 940 941 nomem: 942 pr_err("%s : Failed to allocate memory\n", __func__); 943 opal_free_sg_list(first); 944 return NULL; 945 } 946 947 void opal_free_sg_list(struct opal_sg_list *sg) 948 { 949 while (sg) { 950 uint64_t next = be64_to_cpu(sg->next); 951 952 kfree(sg); 953 954 if (next) 955 sg = __va(next); 956 else 957 sg = NULL; 958 } 959 } 960 961 int opal_error_code(int rc) 962 { 963 switch (rc) { 964 case OPAL_SUCCESS: return 0; 965 966 case OPAL_PARAMETER: return -EINVAL; 967 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS; 968 case OPAL_BUSY_EVENT: return -EBUSY; 969 case OPAL_NO_MEM: return -ENOMEM; 970 case OPAL_PERMISSION: return -EPERM; 971 972 case OPAL_UNSUPPORTED: return -EIO; 973 case OPAL_HARDWARE: return -EIO; 974 case OPAL_INTERNAL_ERROR: return -EIO; 975 case OPAL_TIMEOUT: return -ETIMEDOUT; 976 default: 977 pr_err("%s: unexpected OPAL error %d\n", __func__, rc); 978 return -EIO; 979 } 980 } 981 982 void powernv_set_nmmu_ptcr(unsigned long ptcr) 983 { 984 int rc; 985 986 if (firmware_has_feature(FW_FEATURE_OPAL)) { 987 rc = opal_nmmu_set_ptcr(-1UL, ptcr); 988 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED) 989 pr_warn("%s: Unable to set nest mmu ptcr\n", __func__); 990 } 991 } 992 993 EXPORT_SYMBOL_GPL(opal_poll_events); 994 EXPORT_SYMBOL_GPL(opal_rtc_read); 995 EXPORT_SYMBOL_GPL(opal_rtc_write); 996 EXPORT_SYMBOL_GPL(opal_tpo_read); 997 EXPORT_SYMBOL_GPL(opal_tpo_write); 998 EXPORT_SYMBOL_GPL(opal_i2c_request); 999 /* Export these symbols for PowerNV LED class driver */ 1000 EXPORT_SYMBOL_GPL(opal_leds_get_ind); 1001 EXPORT_SYMBOL_GPL(opal_leds_set_ind); 1002 /* Export this symbol for PowerNV Operator Panel class driver */ 1003 EXPORT_SYMBOL_GPL(opal_write_oppanel_async); 1004 /* Export this for KVM */ 1005 EXPORT_SYMBOL_GPL(opal_int_set_mfrr); 1006 EXPORT_SYMBOL_GPL(opal_int_eoi); 1007