arch/powerpc/kernel/fadump.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
   4  * dump with assistance from firmware. This approach does not use kexec,
   5  * instead firmware assists in booting the kdump kernel while preserving
   6  * memory contents. The most of the code implementation has been adapted
   7  * from phyp assisted dump implementation written by Linas Vepstas and
   8  * Manish Ahuja
   9  *
  10  * Copyright 2011 IBM Corporation
  11  * Author: Mahesh Salgaonkar <[email protected]>
  12  */
  13
  14 #undef DEBUG
  15 #define pr_fmt(fmt) "fadump: " fmt
  16
  17 #include <linux/string.h>
  18 #include <linux/memblock.h>
  19 #include <linux/delay.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/crash_dump.h>
  22 #include <linux/kobject.h>
  23 #include <linux/sysfs.h>
  24 #include <linux/slab.h>
  25 #include <linux/cma.h>
  26 #include <linux/hugetlb.h>
  27
  28 #include <asm/debugfs.h>
  29 #include <asm/page.h>
  30 #include <asm/prom.h>
  31 #include <asm/rtas.h>
  32 #include <asm/fadump.h>
  33 #include <asm/setup.h>
  34
  35 static struct fw_dump fw_dump;
  36 static struct fadump_mem_struct fdm;
  37 static const struct fadump_mem_struct *fdm_active;
  38 #ifdef CONFIG_CMA
  39 static struct cma *fadump_cma;
  40 #endif
  41
  42 static DEFINE_MUTEX(fadump_mutex);
  43 struct fad_crash_memory_ranges *crash_memory_ranges;
  44 int crash_memory_ranges_size;
  45 int crash_mem_ranges;
  46 int max_crash_mem_ranges;
  47
  48 #ifdef CONFIG_CMA
  49 /*
  50  * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
  51  *
  52  * This function initializes CMA area from fadump reserved memory.
  53  * The total size of fadump reserved memory covers for boot memory size
  54  * + cpu data size + hpte size and metadata.
  55  * Initialize only the area equivalent to boot memory size for CMA use.
  56  * The reamining portion of fadump reserved memory will be not given
  57  * to CMA and pages for thoes will stay reserved. boot memory size is
  58  * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
  59  * But for some reason even if it fails we still have the memory reservation
  60  * with us and we can still continue doing fadump.
  61  */
  62 int __init fadump_cma_init(void)
  63 {
  64         unsigned long long base, size;
  65         int rc;
  66
  67         if (!fw_dump.fadump_enabled)
  68                 return 0;
  69
  70         /*
  71          * Do not use CMA if user has provided fadump=nocma kernel parameter.
  72          * Return 1 to continue with fadump old behaviour.
  73          */
  74         if (fw_dump.nocma)
  75                 return 1;
  76
  77         base = fw_dump.reserve_dump_area_start;
  78         size = fw_dump.boot_memory_size;
  79
  80         if (!size)
  81                 return 0;
  82
  83         rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
  84         if (rc) {
  85                 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
  86                 /*
  87                  * Though the CMA init has failed we still have memory
  88                  * reservation with us. The reserved memory will be
  89                  * blocked from production system usage.  Hence return 1,
  90                  * so that we can continue with fadump.
  91                  */
  92                 return 1;
  93         }
  94
  95         /*
  96          * So we now have successfully initialized cma area for fadump.
  97          */
  98         pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
  99                 "bytes of memory reserved for firmware-assisted dump\n",
 100                 cma_get_size(fadump_cma),
 101                 (unsigned long)cma_get_base(fadump_cma) >> 20,
 102                 fw_dump.reserve_dump_area_size);
 103         return 1;
 104 }
 105 #else
 106 static int __init fadump_cma_init(void) { return 1; }
 107 #endif /* CONFIG_CMA */
 108
 109 /* Scan the Firmware Assisted dump configuration details. */
 110 int __init early_init_dt_scan_fw_dump(unsigned long node,
 111                         const char *uname, int depth, void *data)
 112 {
 113         const __be32 *sections;
 114         int i, num_sections;
 115         int size;
 116         const __be32 *token;
 117
 118         if (depth != 1 || strcmp(uname, "rtas") != 0)
 119                 return 0;
 120
 121         /*
 122          * Check if Firmware Assisted dump is supported. if yes, check
 123          * if dump has been initiated on last reboot.
 124          */
 125         token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
 126         if (!token)
 127                 return 1;
 128
 129         fw_dump.fadump_supported = 1;
 130         fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
 131
 132         /*
 133          * The 'ibm,kernel-dump' rtas node is present only if there is
 134          * dump data waiting for us.
 135          */
 136         fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
 137         if (fdm_active)
 138                 fw_dump.dump_active = 1;
 139
 140         /* Get the sizes required to store dump data for the firmware provided
 141          * dump sections.
 142          * For each dump section type supported, a 32bit cell which defines
 143          * the ID of a supported section followed by two 32 bit cells which
 144          * gives teh size of the section in bytes.
 145          */
 146         sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
 147                                         &size);
 148
 149         if (!sections)
 150                 return 1;
 151
 152         num_sections = size / (3 * sizeof(u32));
 153
 154         for (i = 0; i < num_sections; i++, sections += 3) {
 155                 u32 type = (u32)of_read_number(sections, 1);
 156
 157                 switch (type) {
 158                 case FADUMP_CPU_STATE_DATA:
 159                         fw_dump.cpu_state_data_size =
 160                                         of_read_ulong(&sections[1], 2);
 161                         break;
 162                 case FADUMP_HPTE_REGION:
 163                         fw_dump.hpte_region_size =
 164                                         of_read_ulong(&sections[1], 2);
 165                         break;
 166                 }
 167         }
 168
 169         return 1;
 170 }
 171
 172 /*
 173  * If fadump is registered, check if the memory provided
 174  * falls within boot memory area and reserved memory area.
 175  */
 176 int is_fadump_memory_area(u64 addr, ulong size)
 177 {
 178         u64 d_start = fw_dump.reserve_dump_area_start;
 179         u64 d_end = d_start + fw_dump.reserve_dump_area_size;
 180
 181         if (!fw_dump.dump_registered)
 182                 return 0;
 183
 184         if (((addr + size) > d_start) && (addr <= d_end))
 185                 return 1;
 186
 187         return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size;
 188 }
 189
 190 int should_fadump_crash(void)
 191 {
 192         if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
 193                 return 0;
 194         return 1;
 195 }
 196
 197 int is_fadump_active(void)
 198 {
 199         return fw_dump.dump_active;
 200 }
 201
 202 /*
 203  * Returns 1, if there are no holes in boot memory area,
 204  * 0 otherwise.
 205  */
 206 static int is_boot_memory_area_contiguous(void)
 207 {
 208         struct memblock_region *reg;
 209         unsigned long tstart, tend;
 210         unsigned long start_pfn = PHYS_PFN(RMA_START);
 211         unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size);
 212         unsigned int ret = 0;
 213
 214         for_each_memblock(memory, reg) {
 215                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 216                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 217                 if (tstart < tend) {
 218                         /* Memory hole from start_pfn to tstart */
 219                         if (tstart > start_pfn)
 220                                 break;
 221
 222                         if (tend == end_pfn) {
 223                                 ret = 1;
 224                                 break;
 225                         }
 226
 227                         start_pfn = tend + 1;
 228                 }
 229         }
 230
 231         return ret;
 232 }
 233
 234 /*
 235  * Returns true, if there are no holes in reserved memory area,
 236  * false otherwise.
 237  */
 238 static bool is_reserved_memory_area_contiguous(void)
 239 {
 240         struct memblock_region *reg;
 241         unsigned long start, end;
 242         unsigned long d_start = fw_dump.reserve_dump_area_start;
 243         unsigned long d_end = d_start + fw_dump.reserve_dump_area_size;
 244
 245         for_each_memblock(memory, reg) {
 246                 start = max(d_start, (unsigned long)reg->base);
 247                 end = min(d_end, (unsigned long)(reg->base + reg->size));
 248                 if (d_start < end) {
 249                         /* Memory hole from d_start to start */
 250                         if (start > d_start)
 251                                 break;
 252
 253                         if (end == d_end)
 254                                 return true;
 255
 256                         d_start = end + 1;
 257                 }
 258         }
 259
 260         return false;
 261 }
 262
 263 /* Print firmware assisted dump configurations for debugging purpose. */
 264 static void fadump_show_config(void)
 265 {
 266         pr_debug("Support for firmware-assisted dump (fadump): %s\n",
 267                         (fw_dump.fadump_supported ? "present" : "no support"));
 268
 269         if (!fw_dump.fadump_supported)
 270                 return;
 271
 272         pr_debug("Fadump enabled    : %s\n",
 273                                 (fw_dump.fadump_enabled ? "yes" : "no"));
 274         pr_debug("Dump Active       : %s\n",
 275                                 (fw_dump.dump_active ? "yes" : "no"));
 276         pr_debug("Dump section sizes:\n");
 277         pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
 278         pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
 279         pr_debug("Boot memory size  : %lx\n", fw_dump.boot_memory_size);
 280 }
 281
 282 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
 283                                 unsigned long addr)
 284 {
 285         if (!fdm)
 286                 return 0;
 287
 288         memset(fdm, 0, sizeof(struct fadump_mem_struct));
 289         addr = addr & PAGE_MASK;
 290
 291         fdm->header.dump_format_version = cpu_to_be32(0x00000001);
 292         fdm->header.dump_num_sections = cpu_to_be16(3);
 293         fdm->header.dump_status_flag = 0;
 294         fdm->header.offset_first_dump_section =
 295                 cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
 296
 297         /*
 298          * Fields for disk dump option.
 299          * We are not using disk dump option, hence set these fields to 0.
 300          */
 301         fdm->header.dd_block_size = 0;
 302         fdm->header.dd_block_offset = 0;
 303         fdm->header.dd_num_blocks = 0;
 304         fdm->header.dd_offset_disk_path = 0;
 305
 306         /* set 0 to disable an automatic dump-reboot. */
 307         fdm->header.max_time_auto = 0;
 308
 309         /* Kernel dump sections */
 310         /* cpu state data section. */
 311         fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 312         fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
 313         fdm->cpu_state_data.source_address = 0;
 314         fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
 315         fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
 316         addr += fw_dump.cpu_state_data_size;
 317
 318         /* hpte region section */
 319         fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 320         fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
 321         fdm->hpte_region.source_address = 0;
 322         fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
 323         fdm->hpte_region.destination_address = cpu_to_be64(addr);
 324         addr += fw_dump.hpte_region_size;
 325
 326         /* RMA region section */
 327         fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 328         fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
 329         fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
 330         fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
 331         fdm->rmr_region.destination_address = cpu_to_be64(addr);
 332         addr += fw_dump.boot_memory_size;
 333
 334         return addr;
 335 }
 336
 337 /**
 338  * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
 339  *
 340  * Function to find the largest memory size we need to reserve during early
 341  * boot process. This will be the size of the memory that is required for a
 342  * kernel to boot successfully.
 343  *
 344  * This function has been taken from phyp-assisted dump feature implementation.
 345  *
 346  * returns larger of 256MB or 5% rounded down to multiples of 256MB.
 347  *
 348  * TODO: Come up with better approach to find out more accurate memory size
 349  * that is required for a kernel to boot successfully.
 350  *
 351  */
 352 static inline unsigned long fadump_calculate_reserve_size(void)
 353 {
 354         int ret;
 355         unsigned long long base, size;
 356
 357         if (fw_dump.reserve_bootvar)
 358                 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
 359
 360         /*
 361          * Check if the size is specified through crashkernel= cmdline
 362          * option. If yes, then use that but ignore base as fadump reserves
 363          * memory at a predefined offset.
 364          */
 365         ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
 366                                 &size, &base);
 367         if (ret == 0 && size > 0) {
 368                 unsigned long max_size;
 369
 370                 if (fw_dump.reserve_bootvar)
 371                         pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
 372
 373                 fw_dump.reserve_bootvar = (unsigned long)size;
 374
 375                 /*
 376                  * Adjust if the boot memory size specified is above
 377                  * the upper limit.
 378                  */
 379                 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
 380                 if (fw_dump.reserve_bootvar > max_size) {
 381                         fw_dump.reserve_bootvar = max_size;
 382                         pr_info("Adjusted boot memory size to %luMB\n",
 383                                 (fw_dump.reserve_bootvar >> 20));
 384                 }
 385
 386                 return fw_dump.reserve_bootvar;
 387         } else if (fw_dump.reserve_bootvar) {
 388                 /*
 389                  * 'fadump_reserve_mem=' is being used to reserve memory
 390                  * for firmware-assisted dump.
 391                  */
 392                 return fw_dump.reserve_bootvar;
 393         }
 394
 395         /* divide by 20 to get 5% of value */
 396         size = memblock_phys_mem_size() / 20;
 397
 398         /* round it down in multiples of 256 */
 399         size = size & ~0x0FFFFFFFUL;
 400
 401         /* Truncate to memory_limit. We don't want to over reserve the memory.*/
 402         if (memory_limit && size > memory_limit)
 403                 size = memory_limit;
 404
 405         return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
 406 }
 407
 408 /*
 409  * Calculate the total memory size required to be reserved for
 410  * firmware-assisted dump registration.
 411  */
 412 static unsigned long get_fadump_area_size(void)
 413 {
 414         unsigned long size = 0;
 415
 416         size += fw_dump.cpu_state_data_size;
 417         size += fw_dump.hpte_region_size;
 418         size += fw_dump.boot_memory_size;
 419         size += sizeof(struct fadump_crash_info_header);
 420         size += sizeof(struct elfhdr); /* ELF core header.*/
 421         size += sizeof(struct elf_phdr); /* place holder for cpu notes */
 422         /* Program headers for crash memory regions. */
 423         size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
 424
 425         size = PAGE_ALIGN(size);
 426         return size;
 427 }
 428
 429 static void __init fadump_reserve_crash_area(unsigned long base,
 430                                              unsigned long size)
 431 {
 432         struct memblock_region *reg;
 433         unsigned long mstart, mend, msize;
 434
 435         for_each_memblock(memory, reg) {
 436                 mstart = max_t(unsigned long, base, reg->base);
 437                 mend = reg->base + reg->size;
 438                 mend = min(base + size, mend);
 439
 440                 if (mstart < mend) {
 441                         msize = mend - mstart;
 442                         memblock_reserve(mstart, msize);
 443                         pr_info("Reserved %ldMB of memory at %#016lx for saving crash dump\n",
 444                                 (msize >> 20), mstart);
 445                 }
 446         }
 447 }
 448
 449 int __init fadump_reserve_mem(void)
 450 {
 451         unsigned long base, size, memory_boundary;
 452
 453         if (!fw_dump.fadump_enabled)
 454                 return 0;
 455
 456         if (!fw_dump.fadump_supported) {
 457                 printk(KERN_INFO "Firmware-assisted dump is not supported on"
 458                                 " this hardware\n");
 459                 fw_dump.fadump_enabled = 0;
 460                 return 0;
 461         }
 462         /*
 463          * Initialize boot memory size
 464          * If dump is active then we have already calculated the size during
 465          * first kernel.
 466          */
 467         if (fdm_active)
 468                 fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
 469         else {
 470                 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
 471 #ifdef CONFIG_CMA
 472                 if (!fw_dump.nocma)
 473                         fw_dump.boot_memory_size =
 474                                 ALIGN(fw_dump.boot_memory_size,
 475                                                         FADUMP_CMA_ALIGNMENT);
 476 #endif
 477         }
 478
 479         /*
 480          * Calculate the memory boundary.
 481          * If memory_limit is less than actual memory boundary then reserve
 482          * the memory for fadump beyond the memory_limit and adjust the
 483          * memory_limit accordingly, so that the running kernel can run with
 484          * specified memory_limit.
 485          */
 486         if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
 487                 size = get_fadump_area_size();
 488                 if ((memory_limit + size) < memblock_end_of_DRAM())
 489                         memory_limit += size;
 490                 else
 491                         memory_limit = memblock_end_of_DRAM();
 492                 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
 493                                 " dump, now %#016llx\n", memory_limit);
 494         }
 495         if (memory_limit)
 496                 memory_boundary = memory_limit;
 497         else
 498                 memory_boundary = memblock_end_of_DRAM();
 499
 500         if (fw_dump.dump_active) {
 501                 pr_info("Firmware-assisted dump is active.\n");
 502
 503 #ifdef CONFIG_HUGETLB_PAGE
 504                 /*
 505                  * FADump capture kernel doesn't care much about hugepages.
 506                  * In fact, handling hugepages in capture kernel is asking for
 507                  * trouble. So, disable HugeTLB support when fadump is active.
 508                  */
 509                 hugetlb_disabled = true;
 510 #endif
 511                 /*
 512                  * If last boot has crashed then reserve all the memory
 513                  * above boot_memory_size so that we don't touch it until
 514                  * dump is written to disk by userspace tool. This memory
 515                  * will be released for general use once the dump is saved.
 516                  */
 517                 base = fw_dump.boot_memory_size;
 518                 size = memory_boundary - base;
 519                 fadump_reserve_crash_area(base, size);
 520
 521                 fw_dump.fadumphdr_addr =
 522                                 be64_to_cpu(fdm_active->rmr_region.destination_address) +
 523                                 be64_to_cpu(fdm_active->rmr_region.source_len);
 524                 pr_debug("fadumphdr_addr = %pa\n", &fw_dump.fadumphdr_addr);
 525                 fw_dump.reserve_dump_area_start = base;
 526                 fw_dump.reserve_dump_area_size = size;
 527         } else {
 528                 size = get_fadump_area_size();
 529
 530                 /*
 531                  * Reserve memory at an offset closer to bottom of the RAM to
 532                  * minimize the impact of memory hot-remove operation. We can't
 533                  * use memblock_find_in_range() here since it doesn't allocate
 534                  * from bottom to top.
 535                  */
 536                 for (base = fw_dump.boot_memory_size;
 537                      base <= (memory_boundary - size);
 538                      base += size) {
 539                         if (memblock_is_region_memory(base, size) &&
 540                             !memblock_is_region_reserved(base, size))
 541                                 break;
 542                 }
 543                 if ((base > (memory_boundary - size)) ||
 544                     memblock_reserve(base, size)) {
 545                         pr_err("Failed to reserve memory\n");
 546                         return 0;
 547                 }
 548
 549                 pr_info("Reserved %ldMB of memory at %ldMB for firmware-"
 550                         "assisted dump (System RAM: %ldMB)\n",
 551                         (unsigned long)(size >> 20),
 552                         (unsigned long)(base >> 20),
 553                         (unsigned long)(memblock_phys_mem_size() >> 20));
 554
 555                 fw_dump.reserve_dump_area_start = base;
 556                 fw_dump.reserve_dump_area_size = size;
 557                 return fadump_cma_init();
 558         }
 559         return 1;
 560 }
 561
 562 unsigned long __init arch_reserved_kernel_pages(void)
 563 {
 564         return memblock_reserved_size() / PAGE_SIZE;
 565 }
 566
 567 /* Look for fadump= cmdline option. */
 568 static int __init early_fadump_param(char *p)
 569 {
 570         if (!p)
 571                 return 1;
 572
 573         if (strncmp(p, "on", 2) == 0)
 574                 fw_dump.fadump_enabled = 1;
 575         else if (strncmp(p, "off", 3) == 0)
 576                 fw_dump.fadump_enabled = 0;
 577         else if (strncmp(p, "nocma", 5) == 0) {
 578                 fw_dump.fadump_enabled = 1;
 579                 fw_dump.nocma = 1;
 580         }
 581
 582         return 0;
 583 }
 584 early_param("fadump", early_fadump_param);
 585
 586 /*
 587  * Look for fadump_reserve_mem= cmdline option
 588  * TODO: Remove references to 'fadump_reserve_mem=' parameter,
 589  *       the sooner 'crashkernel=' parameter is accustomed to.
 590  */
 591 static int __init early_fadump_reserve_mem(char *p)
 592 {
 593         if (p)
 594                 fw_dump.reserve_bootvar = memparse(p, &p);
 595         return 0;
 596 }
 597 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
 598
 599 static int register_fw_dump(struct fadump_mem_struct *fdm)
 600 {
 601         int rc, err;
 602         unsigned int wait_time;
 603
 604         pr_debug("Registering for firmware-assisted kernel dump...\n");
 605
 606         /* TODO: Add upper time limit for the delay */
 607         do {
 608                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
 609                         FADUMP_REGISTER, fdm,
 610                         sizeof(struct fadump_mem_struct));
 611
 612                 wait_time = rtas_busy_delay_time(rc);
 613                 if (wait_time)
 614                         mdelay(wait_time);
 615
 616         } while (wait_time);
 617
 618         err = -EIO;
 619         switch (rc) {
 620         default:
 621                 pr_err("Failed to register. Unknown Error(%d).\n", rc);
 622                 break;
 623         case -1:
 624                 printk(KERN_ERR "Failed to register firmware-assisted kernel"
 625                         " dump. Hardware Error(%d).\n", rc);
 626                 break;
 627         case -3:
 628                 if (!is_boot_memory_area_contiguous())
 629                         pr_err("Can't have holes in boot memory area while registering fadump\n");
 630                 else if (!is_reserved_memory_area_contiguous())
 631                         pr_err("Can't have holes in reserved memory area while"
 632                                " registering fadump\n");
 633
 634                 printk(KERN_ERR "Failed to register firmware-assisted kernel"
 635                         " dump. Parameter Error(%d).\n", rc);
 636                 err = -EINVAL;
 637                 break;
 638         case -9:
 639                 printk(KERN_ERR "firmware-assisted kernel dump is already "
 640                         " registered.");
 641                 fw_dump.dump_registered = 1;
 642                 err = -EEXIST;
 643                 break;
 644         case 0:
 645                 printk(KERN_INFO "firmware-assisted kernel dump registration"
 646                         " is successful\n");
 647                 fw_dump.dump_registered = 1;
 648                 err = 0;
 649                 break;
 650         }
 651         return err;
 652 }
 653
 654 void crash_fadump(struct pt_regs *regs, const char *str)
 655 {
 656         struct fadump_crash_info_header *fdh = NULL;
 657         int old_cpu, this_cpu;
 658
 659         if (!should_fadump_crash())
 660                 return;
 661
 662         /*
 663          * old_cpu == -1 means this is the first CPU which has come here,
 664          * go ahead and trigger fadump.
 665          *
 666          * old_cpu != -1 means some other CPU has already on it's way
 667          * to trigger fadump, just keep looping here.
 668          */
 669         this_cpu = smp_processor_id();
 670         old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
 671
 672         if (old_cpu != -1) {
 673                 /*
 674                  * We can't loop here indefinitely. Wait as long as fadump
 675                  * is in force. If we race with fadump un-registration this
 676                  * loop will break and then we go down to normal panic path
 677                  * and reboot. If fadump is in force the first crashing
 678                  * cpu will definitely trigger fadump.
 679                  */
 680                 while (fw_dump.dump_registered)
 681                         cpu_relax();
 682                 return;
 683         }
 684
 685         fdh = __va(fw_dump.fadumphdr_addr);
 686         fdh->crashing_cpu = crashing_cpu;
 687         crash_save_vmcoreinfo();
 688
 689         if (regs)
 690                 fdh->regs = *regs;
 691         else
 692                 ppc_save_regs(&fdh->regs);
 693
 694         fdh->online_mask = *cpu_online_mask;
 695
 696         /* Call ibm,os-term rtas call to trigger firmware assisted dump */
 697         rtas_os_term((char *)str);
 698 }
 699
 700 #define GPR_MASK        0xffffff0000000000
 701 static inline int fadump_gpr_index(u64 id)
 702 {
 703         int i = -1;
 704         char str[3];
 705
 706         if ((id & GPR_MASK) == REG_ID("GPR")) {
 707                 /* get the digits at the end */
 708                 id &= ~GPR_MASK;
 709                 id >>= 24;
 710                 str[2] = '\0';
 711                 str[1] = id & 0xff;
 712                 str[0] = (id >> 8) & 0xff;
 713                 sscanf(str, "%d", &i);
 714                 if (i > 31)
 715                         i = -1;
 716         }
 717         return i;
 718 }
 719
 720 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
 721                                                                 u64 reg_val)
 722 {
 723         int i;
 724
 725         i = fadump_gpr_index(reg_id);
 726         if (i >= 0)
 727                 regs->gpr[i] = (unsigned long)reg_val;
 728         else if (reg_id == REG_ID("NIA"))
 729                 regs->nip = (unsigned long)reg_val;
 730         else if (reg_id == REG_ID("MSR"))
 731                 regs->msr = (unsigned long)reg_val;
 732         else if (reg_id == REG_ID("CTR"))
 733                 regs->ctr = (unsigned long)reg_val;
 734         else if (reg_id == REG_ID("LR"))
 735                 regs->link = (unsigned long)reg_val;
 736         else if (reg_id == REG_ID("XER"))
 737                 regs->xer = (unsigned long)reg_val;
 738         else if (reg_id == REG_ID("CR"))
 739                 regs->ccr = (unsigned long)reg_val;
 740         else if (reg_id == REG_ID("DAR"))
 741                 regs->dar = (unsigned long)reg_val;
 742         else if (reg_id == REG_ID("DSISR"))
 743                 regs->dsisr = (unsigned long)reg_val;
 744 }
 745
 746 static struct fadump_reg_entry*
 747 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
 748 {
 749         memset(regs, 0, sizeof(struct pt_regs));
 750
 751         while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
 752                 fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
 753                                         be64_to_cpu(reg_entry->reg_value));
 754                 reg_entry++;
 755         }
 756         reg_entry++;
 757         return reg_entry;
 758 }
 759
 760 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
 761 {
 762         struct elf_prstatus prstatus;
 763
 764         memset(&prstatus, 0, sizeof(prstatus));
 765         /*
 766          * FIXME: How do i get PID? Do I really need it?
 767          * prstatus.pr_pid = ????
 768          */
 769         elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
 770         buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
 771                               &prstatus, sizeof(prstatus));
 772         return buf;
 773 }
 774
 775 static void fadump_update_elfcore_header(char *bufp)
 776 {
 777         struct elfhdr *elf;
 778         struct elf_phdr *phdr;
 779
 780         elf = (struct elfhdr *)bufp;
 781         bufp += sizeof(struct elfhdr);
 782
 783         /* First note is a place holder for cpu notes info. */
 784         phdr = (struct elf_phdr *)bufp;
 785
 786         if (phdr->p_type == PT_NOTE) {
 787                 phdr->p_paddr = fw_dump.cpu_notes_buf;
 788                 phdr->p_offset  = phdr->p_paddr;
 789                 phdr->p_filesz  = fw_dump.cpu_notes_buf_size;
 790                 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
 791         }
 792         return;
 793 }
 794
 795 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
 796 {
 797         void *vaddr;
 798         struct page *page;
 799         unsigned long order, count, i;
 800
 801         order = get_order(size);
 802         vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
 803         if (!vaddr)
 804                 return NULL;
 805
 806         count = 1 << order;
 807         page = virt_to_page(vaddr);
 808         for (i = 0; i < count; i++)
 809                 SetPageReserved(page + i);
 810         return vaddr;
 811 }
 812
 813 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
 814 {
 815         struct page *page;
 816         unsigned long order, count, i;
 817
 818         order = get_order(size);
 819         count = 1 << order;
 820         page = virt_to_page(vaddr);
 821         for (i = 0; i < count; i++)
 822                 ClearPageReserved(page + i);
 823         __free_pages(page, order);
 824 }
 825
 826 /*
 827  * Read CPU state dump data and convert it into ELF notes.
 828  * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
 829  * used to access the data to allow for additional fields to be added without
 830  * affecting compatibility. Each list of registers for a CPU starts with
 831  * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
 832  * 8 Byte ASCII identifier and 8 Byte register value. The register entry
 833  * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
 834  * of register value. For more details refer to PAPR document.
 835  *
 836  * Only for the crashing cpu we ignore the CPU dump data and get exact
 837  * state from fadump crash info structure populated by first kernel at the
 838  * time of crash.
 839  */
 840 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
 841 {
 842         struct fadump_reg_save_area_header *reg_header;
 843         struct fadump_reg_entry *reg_entry;
 844         struct fadump_crash_info_header *fdh = NULL;
 845         void *vaddr;
 846         unsigned long addr;
 847         u32 num_cpus, *note_buf;
 848         struct pt_regs regs;
 849         int i, rc = 0, cpu = 0;
 850
 851         if (!fdm->cpu_state_data.bytes_dumped)
 852                 return -EINVAL;
 853
 854         addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
 855         vaddr = __va(addr);
 856
 857         reg_header = vaddr;
 858         if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
 859                 printk(KERN_ERR "Unable to read register save area.\n");
 860                 return -ENOENT;
 861         }
 862         pr_debug("--------CPU State Data------------\n");
 863         pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
 864         pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
 865
 866         vaddr += be32_to_cpu(reg_header->num_cpu_offset);
 867         num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
 868         pr_debug("NumCpus     : %u\n", num_cpus);
 869         vaddr += sizeof(u32);
 870         reg_entry = (struct fadump_reg_entry *)vaddr;
 871
 872         /* Allocate buffer to hold cpu crash notes. */
 873         fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
 874         fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
 875         note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
 876         if (!note_buf) {
 877                 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
 878                         "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
 879                 return -ENOMEM;
 880         }
 881         fw_dump.cpu_notes_buf = __pa(note_buf);
 882
 883         pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
 884                         (num_cpus * sizeof(note_buf_t)), note_buf);
 885
 886         if (fw_dump.fadumphdr_addr)
 887                 fdh = __va(fw_dump.fadumphdr_addr);
 888
 889         for (i = 0; i < num_cpus; i++) {
 890                 if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
 891                         printk(KERN_ERR "Unable to read CPU state data\n");
 892                         rc = -ENOENT;
 893                         goto error_out;
 894                 }
 895                 /* Lower 4 bytes of reg_value contains logical cpu id */
 896                 cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
 897                 if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
 898                         SKIP_TO_NEXT_CPU(reg_entry);
 899                         continue;
 900                 }
 901                 pr_debug("Reading register data for cpu %d...\n", cpu);
 902                 if (fdh && fdh->crashing_cpu == cpu) {
 903                         regs = fdh->regs;
 904                         note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 905                         SKIP_TO_NEXT_CPU(reg_entry);
 906                 } else {
 907                         reg_entry++;
 908                         reg_entry = fadump_read_registers(reg_entry, &regs);
 909                         note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 910                 }
 911         }
 912         final_note(note_buf);
 913
 914         if (fdh) {
 915                 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
 916                                                         fdh->elfcorehdr_addr);
 917                 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
 918         }
 919         return 0;
 920
 921 error_out:
 922         fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
 923                                         fw_dump.cpu_notes_buf_size);
 924         fw_dump.cpu_notes_buf = 0;
 925         fw_dump.cpu_notes_buf_size = 0;
 926         return rc;
 927
 928 }
 929
 930 /*
 931  * Validate and process the dump data stored by firmware before exporting
 932  * it through '/proc/vmcore'.
 933  */
 934 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
 935 {
 936         struct fadump_crash_info_header *fdh;
 937         int rc = 0;
 938
 939         if (!fdm_active || !fw_dump.fadumphdr_addr)
 940                 return -EINVAL;
 941
 942         /* Check if the dump data is valid. */
 943         if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
 944                         (fdm_active->cpu_state_data.error_flags != 0) ||
 945                         (fdm_active->rmr_region.error_flags != 0)) {
 946                 printk(KERN_ERR "Dump taken by platform is not valid\n");
 947                 return -EINVAL;
 948         }
 949         if ((fdm_active->rmr_region.bytes_dumped !=
 950                         fdm_active->rmr_region.source_len) ||
 951                         !fdm_active->cpu_state_data.bytes_dumped) {
 952                 printk(KERN_ERR "Dump taken by platform is incomplete\n");
 953                 return -EINVAL;
 954         }
 955
 956         /* Validate the fadump crash info header */
 957         fdh = __va(fw_dump.fadumphdr_addr);
 958         if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
 959                 printk(KERN_ERR "Crash info header is not valid.\n");
 960                 return -EINVAL;
 961         }
 962
 963         rc = fadump_build_cpu_notes(fdm_active);
 964         if (rc)
 965                 return rc;
 966
 967         /*
 968          * We are done validating dump info and elfcore header is now ready
 969          * to be exported. set elfcorehdr_addr so that vmcore module will
 970          * export the elfcore header through '/proc/vmcore'.
 971          */
 972         elfcorehdr_addr = fdh->elfcorehdr_addr;
 973
 974         return 0;
 975 }
 976
 977 static void free_crash_memory_ranges(void)
 978 {
 979         kfree(crash_memory_ranges);
 980         crash_memory_ranges = NULL;
 981         crash_memory_ranges_size = 0;
 982         max_crash_mem_ranges = 0;
 983 }
 984
 985 /*
 986  * Allocate or reallocate crash memory ranges array in incremental units
 987  * of PAGE_SIZE.
 988  */
 989 static int allocate_crash_memory_ranges(void)
 990 {
 991         struct fad_crash_memory_ranges *new_array;
 992         u64 new_size;
 993
 994         new_size = crash_memory_ranges_size + PAGE_SIZE;
 995         pr_debug("Allocating %llu bytes of memory for crash memory ranges\n",
 996                  new_size);
 997
 998         new_array = krealloc(crash_memory_ranges, new_size, GFP_KERNEL);
 999         if (new_array == NULL) {
1000                 pr_err("Insufficient memory for setting up crash memory ranges\n");
1001                 free_crash_memory_ranges();
1002                 return -ENOMEM;
1003         }
1004
1005         crash_memory_ranges = new_array;
1006         crash_memory_ranges_size = new_size;
1007         max_crash_mem_ranges = (new_size /
1008                                 sizeof(struct fad_crash_memory_ranges));
1009         return 0;
1010 }
1011
1012 static inline int fadump_add_crash_memory(unsigned long long base,
1013                                           unsigned long long end)
1014 {
1015         u64  start, size;
1016         bool is_adjacent = false;
1017
1018         if (base == end)
1019                 return 0;
1020
1021         /*
1022          * Fold adjacent memory ranges to bring down the memory ranges/
1023          * PT_LOAD segments count.
1024          */
1025         if (crash_mem_ranges) {
1026                 start = crash_memory_ranges[crash_mem_ranges - 1].base;
1027                 size = crash_memory_ranges[crash_mem_ranges - 1].size;
1028
1029                 if ((start + size) == base)
1030                         is_adjacent = true;
1031         }
1032         if (!is_adjacent) {
1033                 /* resize the array on reaching the limit */
1034                 if (crash_mem_ranges == max_crash_mem_ranges) {
1035                         int ret;
1036
1037                         ret = allocate_crash_memory_ranges();
1038                         if (ret)
1039                                 return ret;
1040                 }
1041
1042                 start = base;
1043                 crash_memory_ranges[crash_mem_ranges].base = start;
1044                 crash_mem_ranges++;
1045         }
1046
1047         crash_memory_ranges[crash_mem_ranges - 1].size = (end - start);
1048         pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
1049                 (crash_mem_ranges - 1), start, end - 1, (end - start));
1050         return 0;
1051 }
1052
1053 static int fadump_exclude_reserved_area(unsigned long long start,
1054                                         unsigned long long end)
1055 {
1056         unsigned long long ra_start, ra_end;
1057         int ret = 0;
1058
1059         ra_start = fw_dump.reserve_dump_area_start;
1060         ra_end = ra_start + fw_dump.reserve_dump_area_size;
1061
1062         if ((ra_start < end) && (ra_end > start)) {
1063                 if ((start < ra_start) && (end > ra_end)) {
1064                         ret = fadump_add_crash_memory(start, ra_start);
1065                         if (ret)
1066                                 return ret;
1067
1068                         ret = fadump_add_crash_memory(ra_end, end);
1069                 } else if (start < ra_start) {
1070                         ret = fadump_add_crash_memory(start, ra_start);
1071                 } else if (ra_end < end) {
1072                         ret = fadump_add_crash_memory(ra_end, end);
1073                 }
1074         } else
1075                 ret = fadump_add_crash_memory(start, end);
1076
1077         return ret;
1078 }
1079
1080 static int fadump_init_elfcore_header(char *bufp)
1081 {
1082         struct elfhdr *elf;
1083
1084         elf = (struct elfhdr *) bufp;
1085         bufp += sizeof(struct elfhdr);
1086         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1087         elf->e_ident[EI_CLASS] = ELF_CLASS;
1088         elf->e_ident[EI_DATA] = ELF_DATA;
1089         elf->e_ident[EI_VERSION] = EV_CURRENT;
1090         elf->e_ident[EI_OSABI] = ELF_OSABI;
1091         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1092         elf->e_type = ET_CORE;
1093         elf->e_machine = ELF_ARCH;
1094         elf->e_version = EV_CURRENT;
1095         elf->e_entry = 0;
1096         elf->e_phoff = sizeof(struct elfhdr);
1097         elf->e_shoff = 0;
1098 #if defined(_CALL_ELF)
1099         elf->e_flags = _CALL_ELF;
1100 #else
1101         elf->e_flags = 0;
1102 #endif
1103         elf->e_ehsize = sizeof(struct elfhdr);
1104         elf->e_phentsize = sizeof(struct elf_phdr);
1105         elf->e_phnum = 0;
1106         elf->e_shentsize = 0;
1107         elf->e_shnum = 0;
1108         elf->e_shstrndx = 0;
1109
1110         return 0;
1111 }
1112
1113 /*
1114  * Traverse through memblock structure and setup crash memory ranges. These
1115  * ranges will be used create PT_LOAD program headers in elfcore header.
1116  */
1117 static int fadump_setup_crash_memory_ranges(void)
1118 {
1119         struct memblock_region *reg;
1120         unsigned long long start, end;
1121         int ret;
1122
1123         pr_debug("Setup crash memory ranges.\n");
1124         crash_mem_ranges = 0;
1125
1126         /*
1127          * add the first memory chunk (RMA_START through boot_memory_size) as
1128          * a separate memory chunk. The reason is, at the time crash firmware
1129          * will move the content of this memory chunk to different location
1130          * specified during fadump registration. We need to create a separate
1131          * program header for this chunk with the correct offset.
1132          */
1133         ret = fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
1134         if (ret)
1135                 return ret;
1136
1137         for_each_memblock(memory, reg) {
1138                 start = (unsigned long long)reg->base;
1139                 end = start + (unsigned long long)reg->size;
1140
1141                 /*
1142                  * skip the first memory chunk that is already added (RMA_START
1143                  * through boot_memory_size). This logic needs a relook if and
1144                  * when RMA_START changes to a non-zero value.
1145                  */
1146                 BUILD_BUG_ON(RMA_START != 0);
1147                 if (start < fw_dump.boot_memory_size) {
1148                         if (end > fw_dump.boot_memory_size)
1149                                 start = fw_dump.boot_memory_size;
1150                         else
1151                                 continue;
1152                 }
1153
1154                 /* add this range excluding the reserved dump area. */
1155                 ret = fadump_exclude_reserved_area(start, end);
1156                 if (ret)
1157                         return ret;
1158         }
1159
1160         return 0;
1161 }
1162
1163 /*
1164  * If the given physical address falls within the boot memory region then
1165  * return the relocated address that points to the dump region reserved
1166  * for saving initial boot memory contents.
1167  */
1168 static inline unsigned long fadump_relocate(unsigned long paddr)
1169 {
1170         if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
1171                 return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
1172         else
1173                 return paddr;
1174 }
1175
1176 static int fadump_create_elfcore_headers(char *bufp)
1177 {
1178         struct elfhdr *elf;
1179         struct elf_phdr *phdr;
1180         int i;
1181
1182         fadump_init_elfcore_header(bufp);
1183         elf = (struct elfhdr *)bufp;
1184         bufp += sizeof(struct elfhdr);
1185
1186         /*
1187          * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1188          * will be populated during second kernel boot after crash. Hence
1189          * this PT_NOTE will always be the first elf note.
1190          *
1191          * NOTE: Any new ELF note addition should be placed after this note.
1192          */
1193         phdr = (struct elf_phdr *)bufp;
1194         bufp += sizeof(struct elf_phdr);
1195         phdr->p_type = PT_NOTE;
1196         phdr->p_flags = 0;
1197         phdr->p_vaddr = 0;
1198         phdr->p_align = 0;
1199
1200         phdr->p_offset = 0;
1201         phdr->p_paddr = 0;
1202         phdr->p_filesz = 0;
1203         phdr->p_memsz = 0;
1204
1205         (elf->e_phnum)++;
1206
1207         /* setup ELF PT_NOTE for vmcoreinfo */
1208         phdr = (struct elf_phdr *)bufp;
1209         bufp += sizeof(struct elf_phdr);
1210         phdr->p_type    = PT_NOTE;
1211         phdr->p_flags   = 0;
1212         phdr->p_vaddr   = 0;
1213         phdr->p_align   = 0;
1214
1215         phdr->p_paddr   = fadump_relocate(paddr_vmcoreinfo_note());
1216         phdr->p_offset  = phdr->p_paddr;
1217         phdr->p_memsz   = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1218
1219         /* Increment number of program headers. */
1220         (elf->e_phnum)++;
1221
1222         /* setup PT_LOAD sections. */
1223
1224         for (i = 0; i < crash_mem_ranges; i++) {
1225                 unsigned long long mbase, msize;
1226                 mbase = crash_memory_ranges[i].base;
1227                 msize = crash_memory_ranges[i].size;
1228
1229                 if (!msize)
1230                         continue;
1231
1232                 phdr = (struct elf_phdr *)bufp;
1233                 bufp += sizeof(struct elf_phdr);
1234                 phdr->p_type    = PT_LOAD;
1235                 phdr->p_flags   = PF_R|PF_W|PF_X;
1236                 phdr->p_offset  = mbase;
1237
1238                 if (mbase == RMA_START) {
1239                         /*
1240                          * The entire RMA region will be moved by firmware
1241                          * to the specified destination_address. Hence set
1242                          * the correct offset.
1243                          */
1244                         phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
1245                 }
1246
1247                 phdr->p_paddr = mbase;
1248                 phdr->p_vaddr = (unsigned long)__va(mbase);
1249                 phdr->p_filesz = msize;
1250                 phdr->p_memsz = msize;
1251                 phdr->p_align = 0;
1252
1253                 /* Increment number of program headers. */
1254                 (elf->e_phnum)++;
1255         }
1256         return 0;
1257 }
1258
1259 static unsigned long init_fadump_header(unsigned long addr)
1260 {
1261         struct fadump_crash_info_header *fdh;
1262
1263         if (!addr)
1264                 return 0;
1265
1266         fw_dump.fadumphdr_addr = addr;
1267         fdh = __va(addr);
1268         addr += sizeof(struct fadump_crash_info_header);
1269
1270         memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1271         fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1272         fdh->elfcorehdr_addr = addr;
1273         /* We will set the crashing cpu id in crash_fadump() during crash. */
1274         fdh->crashing_cpu = CPU_UNKNOWN;
1275
1276         return addr;
1277 }
1278
1279 static int register_fadump(void)
1280 {
1281         unsigned long addr;
1282         void *vaddr;
1283         int ret;
1284
1285         /*
1286          * If no memory is reserved then we can not register for firmware-
1287          * assisted dump.
1288          */
1289         if (!fw_dump.reserve_dump_area_size)
1290                 return -ENODEV;
1291
1292         ret = fadump_setup_crash_memory_ranges();
1293         if (ret)
1294                 return ret;
1295
1296         addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
1297         /* Initialize fadump crash info header. */
1298         addr = init_fadump_header(addr);
1299         vaddr = __va(addr);
1300
1301         pr_debug("Creating ELF core headers at %#016lx\n", addr);
1302         fadump_create_elfcore_headers(vaddr);
1303
1304         /* register the future kernel dump with firmware. */
1305         return register_fw_dump(&fdm);
1306 }
1307
1308 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1309 {
1310         int rc = 0;
1311         unsigned int wait_time;
1312
1313         pr_debug("Un-register firmware-assisted dump\n");
1314
1315         /* TODO: Add upper time limit for the delay */
1316         do {
1317                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1318                         FADUMP_UNREGISTER, fdm,
1319                         sizeof(struct fadump_mem_struct));
1320
1321                 wait_time = rtas_busy_delay_time(rc);
1322                 if (wait_time)
1323                         mdelay(wait_time);
1324         } while (wait_time);
1325
1326         if (rc) {
1327                 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1328                         " unexpected error(%d).\n", rc);
1329                 return rc;
1330         }
1331         fw_dump.dump_registered = 0;
1332         return 0;
1333 }
1334
1335 static int fadump_invalidate_dump(const struct fadump_mem_struct *fdm)
1336 {
1337         int rc = 0;
1338         unsigned int wait_time;
1339
1340         pr_debug("Invalidating firmware-assisted dump registration\n");
1341
1342         /* TODO: Add upper time limit for the delay */
1343         do {
1344                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1345                         FADUMP_INVALIDATE, fdm,
1346                         sizeof(struct fadump_mem_struct));
1347
1348                 wait_time = rtas_busy_delay_time(rc);
1349                 if (wait_time)
1350                         mdelay(wait_time);
1351         } while (wait_time);
1352
1353         if (rc) {
1354                 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1355                 return rc;
1356         }
1357         fw_dump.dump_active = 0;
1358         fdm_active = NULL;
1359         return 0;
1360 }
1361
1362 void fadump_cleanup(void)
1363 {
1364         /* Invalidate the registration only if dump is active. */
1365         if (fw_dump.dump_active) {
1366                 /* pass the same memory dump structure provided by platform */
1367                 fadump_invalidate_dump(fdm_active);
1368         } else if (fw_dump.dump_registered) {
1369                 /* Un-register Firmware-assisted dump if it was registered. */
1370                 fadump_unregister_dump(&fdm);
1371                 free_crash_memory_ranges();
1372         }
1373 }
1374
1375 static void fadump_free_reserved_memory(unsigned long start_pfn,
1376                                         unsigned long end_pfn)
1377 {
1378         unsigned long pfn;
1379         unsigned long time_limit = jiffies + HZ;
1380
1381         pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1382                 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1383
1384         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1385                 free_reserved_page(pfn_to_page(pfn));
1386
1387                 if (time_after(jiffies, time_limit)) {
1388                         cond_resched();
1389                         time_limit = jiffies + HZ;
1390                 }
1391         }
1392 }
1393
1394 /*
1395  * Skip memory holes and free memory that was actually reserved.
1396  */
1397 static void fadump_release_reserved_area(unsigned long start, unsigned long end)
1398 {
1399         struct memblock_region *reg;
1400         unsigned long tstart, tend;
1401         unsigned long start_pfn = PHYS_PFN(start);
1402         unsigned long end_pfn = PHYS_PFN(end);
1403
1404         for_each_memblock(memory, reg) {
1405                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
1406                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
1407                 if (tstart < tend) {
1408                         fadump_free_reserved_memory(tstart, tend);
1409
1410                         if (tend == end_pfn)
1411                                 break;
1412
1413                         start_pfn = tend + 1;
1414                 }
1415         }
1416 }
1417
1418 /*
1419  * Release the memory that was reserved in early boot to preserve the memory
1420  * contents. The released memory will be available for general use.
1421  */
1422 static void fadump_release_memory(unsigned long begin, unsigned long end)
1423 {
1424         unsigned long ra_start, ra_end;
1425
1426         ra_start = fw_dump.reserve_dump_area_start;
1427         ra_end = ra_start + fw_dump.reserve_dump_area_size;
1428
1429         /*
1430          * exclude the dump reserve area. Will reuse it for next
1431          * fadump registration.
1432          */
1433         if (begin < ra_end && end > ra_start) {
1434                 if (begin < ra_start)
1435                         fadump_release_reserved_area(begin, ra_start);
1436                 if (end > ra_end)
1437                         fadump_release_reserved_area(ra_end, end);
1438         } else
1439                 fadump_release_reserved_area(begin, end);
1440 }
1441
1442 static void fadump_invalidate_release_mem(void)
1443 {
1444         unsigned long reserved_area_start, reserved_area_end;
1445         unsigned long destination_address;
1446
1447         mutex_lock(&fadump_mutex);
1448         if (!fw_dump.dump_active) {
1449                 mutex_unlock(&fadump_mutex);
1450                 return;
1451         }
1452
1453         destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1454         fadump_cleanup();
1455         mutex_unlock(&fadump_mutex);
1456
1457         /*
1458          * Save the current reserved memory bounds we will require them
1459          * later for releasing the memory for general use.
1460          */
1461         reserved_area_start = fw_dump.reserve_dump_area_start;
1462         reserved_area_end = reserved_area_start +
1463                         fw_dump.reserve_dump_area_size;
1464         /*
1465          * Setup reserve_dump_area_start and its size so that we can
1466          * reuse this reserved memory for Re-registration.
1467          */
1468         fw_dump.reserve_dump_area_start = destination_address;
1469         fw_dump.reserve_dump_area_size = get_fadump_area_size();
1470
1471         fadump_release_memory(reserved_area_start, reserved_area_end);
1472         if (fw_dump.cpu_notes_buf) {
1473                 fadump_cpu_notes_buf_free(
1474                                 (unsigned long)__va(fw_dump.cpu_notes_buf),
1475                                 fw_dump.cpu_notes_buf_size);
1476                 fw_dump.cpu_notes_buf = 0;
1477                 fw_dump.cpu_notes_buf_size = 0;
1478         }
1479         /* Initialize the kernel dump memory structure for FAD registration. */
1480         init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1481 }
1482
1483 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1484                                         struct kobj_attribute *attr,
1485                                         const char *buf, size_t count)
1486 {
1487         int input = -1;
1488
1489         if (!fw_dump.dump_active)
1490                 return -EPERM;
1491
1492         if (kstrtoint(buf, 0, &input))
1493                 return -EINVAL;
1494
1495         if (input == 1) {
1496                 /*
1497                  * Take away the '/proc/vmcore'. We are releasing the dump
1498                  * memory, hence it will not be valid anymore.
1499                  */
1500 #ifdef CONFIG_PROC_VMCORE
1501                 vmcore_cleanup();
1502 #endif
1503                 fadump_invalidate_release_mem();
1504
1505         } else
1506                 return -EINVAL;
1507         return count;
1508 }
1509
1510 static ssize_t fadump_enabled_show(struct kobject *kobj,
1511                                         struct kobj_attribute *attr,
1512                                         char *buf)
1513 {
1514         return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1515 }
1516
1517 static ssize_t fadump_register_show(struct kobject *kobj,
1518                                         struct kobj_attribute *attr,
1519                                         char *buf)
1520 {
1521         return sprintf(buf, "%d\n", fw_dump.dump_registered);
1522 }
1523
1524 static ssize_t fadump_register_store(struct kobject *kobj,
1525                                         struct kobj_attribute *attr,
1526                                         const char *buf, size_t count)
1527 {
1528         int ret = 0;
1529         int input = -1;
1530
1531         if (!fw_dump.fadump_enabled || fdm_active)
1532                 return -EPERM;
1533
1534         if (kstrtoint(buf, 0, &input))
1535                 return -EINVAL;
1536
1537         mutex_lock(&fadump_mutex);
1538
1539         switch (input) {
1540         case 0:
1541                 if (fw_dump.dump_registered == 0) {
1542                         goto unlock_out;
1543                 }
1544                 /* Un-register Firmware-assisted dump */
1545                 fadump_unregister_dump(&fdm);
1546                 break;
1547         case 1:
1548                 if (fw_dump.dump_registered == 1) {
1549                         /* Un-register Firmware-assisted dump */
1550                         fadump_unregister_dump(&fdm);
1551                 }
1552                 /* Register Firmware-assisted dump */
1553                 ret = register_fadump();
1554                 break;
1555         default:
1556                 ret = -EINVAL;
1557                 break;
1558         }
1559
1560 unlock_out:
1561         mutex_unlock(&fadump_mutex);
1562         return ret < 0 ? ret : count;
1563 }
1564
1565 static int fadump_region_show(struct seq_file *m, void *private)
1566 {
1567         const struct fadump_mem_struct *fdm_ptr;
1568
1569         if (!fw_dump.fadump_enabled)
1570                 return 0;
1571
1572         mutex_lock(&fadump_mutex);
1573         if (fdm_active)
1574                 fdm_ptr = fdm_active;
1575         else {
1576                 mutex_unlock(&fadump_mutex);
1577                 fdm_ptr = &fdm;
1578         }
1579
1580         seq_printf(m,
1581                         "CPU : [%#016llx-%#016llx] %#llx bytes, "
1582                         "Dumped: %#llx\n",
1583                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1584                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1585                         be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1586                         be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1587                         be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1588         seq_printf(m,
1589                         "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1590                         "Dumped: %#llx\n",
1591                         be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1592                         be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1593                         be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1594                         be64_to_cpu(fdm_ptr->hpte_region.source_len),
1595                         be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1596         seq_printf(m,
1597                         "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1598                         "Dumped: %#llx\n",
1599                         be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1600                         be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1601                         be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1602                         be64_to_cpu(fdm_ptr->rmr_region.source_len),
1603                         be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1604
1605         if (!fdm_active ||
1606                 (fw_dump.reserve_dump_area_start ==
1607                 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1608                 goto out;
1609
1610         /* Dump is active. Show reserved memory region. */
1611         seq_printf(m,
1612                         "    : [%#016llx-%#016llx] %#llx bytes, "
1613                         "Dumped: %#llx\n",
1614                         (unsigned long long)fw_dump.reserve_dump_area_start,
1615                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1616                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1617                         fw_dump.reserve_dump_area_start,
1618                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1619                         fw_dump.reserve_dump_area_start);
1620 out:
1621         if (fdm_active)
1622                 mutex_unlock(&fadump_mutex);
1623         return 0;
1624 }
1625
1626 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1627                                                 0200, NULL,
1628                                                 fadump_release_memory_store);
1629 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1630                                                 0444, fadump_enabled_show,
1631                                                 NULL);
1632 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1633                                                 0644, fadump_register_show,
1634                                                 fadump_register_store);
1635
1636 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1637
1638 static void fadump_init_files(void)
1639 {
1640         struct dentry *debugfs_file;
1641         int rc = 0;
1642
1643         rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1644         if (rc)
1645                 printk(KERN_ERR "fadump: unable to create sysfs file"
1646                         " fadump_enabled (%d)\n", rc);
1647
1648         rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1649         if (rc)
1650                 printk(KERN_ERR "fadump: unable to create sysfs file"
1651                         " fadump_registered (%d)\n", rc);
1652
1653         debugfs_file = debugfs_create_file("fadump_region", 0444,
1654                                         powerpc_debugfs_root, NULL,
1655                                         &fadump_region_fops);
1656         if (!debugfs_file)
1657                 printk(KERN_ERR "fadump: unable to create debugfs file"
1658                                 " fadump_region\n");
1659
1660         if (fw_dump.dump_active) {
1661                 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1662                 if (rc)
1663                         printk(KERN_ERR "fadump: unable to create sysfs file"
1664                                 " fadump_release_mem (%d)\n", rc);
1665         }
1666         return;
1667 }
1668
1669 /*
1670  * Prepare for firmware-assisted dump.
1671  */
1672 int __init setup_fadump(void)
1673 {
1674         if (!fw_dump.fadump_enabled)
1675                 return 0;
1676
1677         if (!fw_dump.fadump_supported) {
1678                 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1679                         " this hardware\n");
1680                 return 0;
1681         }
1682
1683         fadump_show_config();
1684         /*
1685          * If dump data is available then see if it is valid and prepare for
1686          * saving it to the disk.
1687          */
1688         if (fw_dump.dump_active) {
1689                 /*
1690                  * if dump process fails then invalidate the registration
1691                  * and release memory before proceeding for re-registration.
1692                  */
1693                 if (process_fadump(fdm_active) < 0)
1694                         fadump_invalidate_release_mem();
1695         }
1696         /* Initialize the kernel dump memory structure for FAD registration. */
1697         else if (fw_dump.reserve_dump_area_size)
1698                 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1699         fadump_init_files();
1700
1701         return 1;
1702 }
1703 subsys_initcall(setup_fadump);