arch/powerpc/kernel/fadump.c

   1 /*
   2  * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
   3  * dump with assistance from firmware. This approach does not use kexec,
   4  * instead firmware assists in booting the kdump kernel while preserving
   5  * memory contents. The most of the code implementation has been adapted
   6  * from phyp assisted dump implementation written by Linas Vepstas and
   7  * Manish Ahuja
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  22  *
  23  * Copyright 2011 IBM Corporation
  24  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  25  */
  26
  27 #undef DEBUG
  28 #define pr_fmt(fmt) "fadump: " fmt
  29
  30 #include <linux/string.h>
  31 #include <linux/memblock.h>
  32 #include <linux/delay.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/seq_file.h>
  35 #include <linux/crash_dump.h>
  36 #include <linux/kobject.h>
  37 #include <linux/sysfs.h>
  38
  39 #include <asm/page.h>
  40 #include <asm/prom.h>
  41 #include <asm/rtas.h>
  42 #include <asm/fadump.h>
  43 #include <asm/debug.h>
  44 #include <asm/setup.h>
  45
  46 static struct fw_dump fw_dump;
  47 static struct fadump_mem_struct fdm;
  48 static const struct fadump_mem_struct *fdm_active;
  49
  50 static DEFINE_MUTEX(fadump_mutex);
  51 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
  52 int crash_mem_ranges;
  53
  54 /* Scan the Firmware Assisted dump configuration details. */
  55 int __init early_init_dt_scan_fw_dump(unsigned long node,
  56                         const char *uname, int depth, void *data)
  57 {
  58         const __be32 *sections;
  59         int i, num_sections;
  60         int size;
  61         const __be32 *token;
  62
  63         if (depth != 1 || strcmp(uname, "rtas") != 0)
  64                 return 0;
  65
  66         /*
  67          * Check if Firmware Assisted dump is supported. if yes, check
  68          * if dump has been initiated on last reboot.
  69          */
  70         token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
  71         if (!token)
  72                 return 1;
  73
  74         fw_dump.fadump_supported = 1;
  75         fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
  76
  77         /*
  78          * The 'ibm,kernel-dump' rtas node is present only if there is
  79          * dump data waiting for us.
  80          */
  81         fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
  82         if (fdm_active)
  83                 fw_dump.dump_active = 1;
  84
  85         /* Get the sizes required to store dump data for the firmware provided
  86          * dump sections.
  87          * For each dump section type supported, a 32bit cell which defines
  88          * the ID of a supported section followed by two 32 bit cells which
  89          * gives teh size of the section in bytes.
  90          */
  91         sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
  92                                         &size);
  93
  94         if (!sections)
  95                 return 1;
  96
  97         num_sections = size / (3 * sizeof(u32));
  98
  99         for (i = 0; i < num_sections; i++, sections += 3) {
 100                 u32 type = (u32)of_read_number(sections, 1);
 101
 102                 switch (type) {
 103                 case FADUMP_CPU_STATE_DATA:
 104                         fw_dump.cpu_state_data_size =
 105                                         of_read_ulong(&sections[1], 2);
 106                         break;
 107                 case FADUMP_HPTE_REGION:
 108                         fw_dump.hpte_region_size =
 109                                         of_read_ulong(&sections[1], 2);
 110                         break;
 111                 }
 112         }
 113
 114         return 1;
 115 }
 116
 117 int is_fadump_active(void)
 118 {
 119         return fw_dump.dump_active;
 120 }
 121
 122 /* Print firmware assisted dump configurations for debugging purpose. */
 123 static void fadump_show_config(void)
 124 {
 125         pr_debug("Support for firmware-assisted dump (fadump): %s\n",
 126                         (fw_dump.fadump_supported ? "present" : "no support"));
 127
 128         if (!fw_dump.fadump_supported)
 129                 return;
 130
 131         pr_debug("Fadump enabled    : %s\n",
 132                                 (fw_dump.fadump_enabled ? "yes" : "no"));
 133         pr_debug("Dump Active       : %s\n",
 134                                 (fw_dump.dump_active ? "yes" : "no"));
 135         pr_debug("Dump section sizes:\n");
 136         pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
 137         pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
 138         pr_debug("Boot memory size  : %lx\n", fw_dump.boot_memory_size);
 139 }
 140
 141 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
 142                                 unsigned long addr)
 143 {
 144         if (!fdm)
 145                 return 0;
 146
 147         memset(fdm, 0, sizeof(struct fadump_mem_struct));
 148         addr = addr & PAGE_MASK;
 149
 150         fdm->header.dump_format_version = cpu_to_be32(0x00000001);
 151         fdm->header.dump_num_sections = cpu_to_be16(3);
 152         fdm->header.dump_status_flag = 0;
 153         fdm->header.offset_first_dump_section =
 154                 cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
 155
 156         /*
 157          * Fields for disk dump option.
 158          * We are not using disk dump option, hence set these fields to 0.
 159          */
 160         fdm->header.dd_block_size = 0;
 161         fdm->header.dd_block_offset = 0;
 162         fdm->header.dd_num_blocks = 0;
 163         fdm->header.dd_offset_disk_path = 0;
 164
 165         /* set 0 to disable an automatic dump-reboot. */
 166         fdm->header.max_time_auto = 0;
 167
 168         /* Kernel dump sections */
 169         /* cpu state data section. */
 170         fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 171         fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
 172         fdm->cpu_state_data.source_address = 0;
 173         fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
 174         fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
 175         addr += fw_dump.cpu_state_data_size;
 176
 177         /* hpte region section */
 178         fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 179         fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
 180         fdm->hpte_region.source_address = 0;
 181         fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
 182         fdm->hpte_region.destination_address = cpu_to_be64(addr);
 183         addr += fw_dump.hpte_region_size;
 184
 185         /* RMA region section */
 186         fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
 187         fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
 188         fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
 189         fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
 190         fdm->rmr_region.destination_address = cpu_to_be64(addr);
 191         addr += fw_dump.boot_memory_size;
 192
 193         return addr;
 194 }
 195
 196 /**
 197  * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
 198  *
 199  * Function to find the largest memory size we need to reserve during early
 200  * boot process. This will be the size of the memory that is required for a
 201  * kernel to boot successfully.
 202  *
 203  * This function has been taken from phyp-assisted dump feature implementation.
 204  *
 205  * returns larger of 256MB or 5% rounded down to multiples of 256MB.
 206  *
 207  * TODO: Come up with better approach to find out more accurate memory size
 208  * that is required for a kernel to boot successfully.
 209  *
 210  */
 211 static inline unsigned long fadump_calculate_reserve_size(void)
 212 {
 213         unsigned long size;
 214
 215         /*
 216          * Check if the size is specified through fadump_reserve_mem= cmdline
 217          * option. If yes, then use that.
 218          */
 219         if (fw_dump.reserve_bootvar)
 220                 return fw_dump.reserve_bootvar;
 221
 222         /* divide by 20 to get 5% of value */
 223         size = memblock_end_of_DRAM() / 20;
 224
 225         /* round it down in multiples of 256 */
 226         size = size & ~0x0FFFFFFFUL;
 227
 228         /* Truncate to memory_limit. We don't want to over reserve the memory.*/
 229         if (memory_limit && size > memory_limit)
 230                 size = memory_limit;
 231
 232         return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
 233 }
 234
 235 /*
 236  * Calculate the total memory size required to be reserved for
 237  * firmware-assisted dump registration.
 238  */
 239 static unsigned long get_fadump_area_size(void)
 240 {
 241         unsigned long size = 0;
 242
 243         size += fw_dump.cpu_state_data_size;
 244         size += fw_dump.hpte_region_size;
 245         size += fw_dump.boot_memory_size;
 246         size += sizeof(struct fadump_crash_info_header);
 247         size += sizeof(struct elfhdr); /* ELF core header.*/
 248         size += sizeof(struct elf_phdr); /* place holder for cpu notes */
 249         /* Program headers for crash memory regions. */
 250         size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
 251
 252         size = PAGE_ALIGN(size);
 253         return size;
 254 }
 255
 256 int __init fadump_reserve_mem(void)
 257 {
 258         unsigned long base, size, memory_boundary;
 259
 260         if (!fw_dump.fadump_enabled)
 261                 return 0;
 262
 263         if (!fw_dump.fadump_supported) {
 264                 printk(KERN_INFO "Firmware-assisted dump is not supported on"
 265                                 " this hardware\n");
 266                 fw_dump.fadump_enabled = 0;
 267                 return 0;
 268         }
 269         /*
 270          * Initialize boot memory size
 271          * If dump is active then we have already calculated the size during
 272          * first kernel.
 273          */
 274         if (fdm_active)
 275                 fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
 276         else
 277                 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
 278
 279         /*
 280          * Calculate the memory boundary.
 281          * If memory_limit is less than actual memory boundary then reserve
 282          * the memory for fadump beyond the memory_limit and adjust the
 283          * memory_limit accordingly, so that the running kernel can run with
 284          * specified memory_limit.
 285          */
 286         if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
 287                 size = get_fadump_area_size();
 288                 if ((memory_limit + size) < memblock_end_of_DRAM())
 289                         memory_limit += size;
 290                 else
 291                         memory_limit = memblock_end_of_DRAM();
 292                 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
 293                                 " dump, now %#016llx\n", memory_limit);
 294         }
 295         if (memory_limit)
 296                 memory_boundary = memory_limit;
 297         else
 298                 memory_boundary = memblock_end_of_DRAM();
 299
 300         if (fw_dump.dump_active) {
 301                 printk(KERN_INFO "Firmware-assisted dump is active.\n");
 302                 /*
 303                  * If last boot has crashed then reserve all the memory
 304                  * above boot_memory_size so that we don't touch it until
 305                  * dump is written to disk by userspace tool. This memory
 306                  * will be released for general use once the dump is saved.
 307                  */
 308                 base = fw_dump.boot_memory_size;
 309                 size = memory_boundary - base;
 310                 memblock_reserve(base, size);
 311                 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
 312                                 "for saving crash dump\n",
 313                                 (unsigned long)(size >> 20),
 314                                 (unsigned long)(base >> 20));
 315
 316                 fw_dump.fadumphdr_addr =
 317                                 be64_to_cpu(fdm_active->rmr_region.destination_address) +
 318                                 be64_to_cpu(fdm_active->rmr_region.source_len);
 319                 pr_debug("fadumphdr_addr = %p\n",
 320                                 (void *) fw_dump.fadumphdr_addr);
 321         } else {
 322                 /* Reserve the memory at the top of memory. */
 323                 size = get_fadump_area_size();
 324                 base = memory_boundary - size;
 325                 memblock_reserve(base, size);
 326                 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
 327                                 "for firmware-assisted dump\n",
 328                                 (unsigned long)(size >> 20),
 329                                 (unsigned long)(base >> 20));
 330         }
 331         fw_dump.reserve_dump_area_start = base;
 332         fw_dump.reserve_dump_area_size = size;
 333         return 1;
 334 }
 335
 336 unsigned long __init arch_reserved_kernel_pages(void)
 337 {
 338         return memblock_reserved_size() / PAGE_SIZE;
 339 }
 340
 341 /* Look for fadump= cmdline option. */
 342 static int __init early_fadump_param(char *p)
 343 {
 344         if (!p)
 345                 return 1;
 346
 347         if (strncmp(p, "on", 2) == 0)
 348                 fw_dump.fadump_enabled = 1;
 349         else if (strncmp(p, "off", 3) == 0)
 350                 fw_dump.fadump_enabled = 0;
 351
 352         return 0;
 353 }
 354 early_param("fadump", early_fadump_param);
 355
 356 /* Look for fadump_reserve_mem= cmdline option */
 357 static int __init early_fadump_reserve_mem(char *p)
 358 {
 359         if (p)
 360                 fw_dump.reserve_bootvar = memparse(p, &p);
 361         return 0;
 362 }
 363 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
 364
 365 static void register_fw_dump(struct fadump_mem_struct *fdm)
 366 {
 367         int rc;
 368         unsigned int wait_time;
 369
 370         pr_debug("Registering for firmware-assisted kernel dump...\n");
 371
 372         /* TODO: Add upper time limit for the delay */
 373         do {
 374                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
 375                         FADUMP_REGISTER, fdm,
 376                         sizeof(struct fadump_mem_struct));
 377
 378                 wait_time = rtas_busy_delay_time(rc);
 379                 if (wait_time)
 380                         mdelay(wait_time);
 381
 382         } while (wait_time);
 383
 384         switch (rc) {
 385         case -1:
 386                 printk(KERN_ERR "Failed to register firmware-assisted kernel"
 387                         " dump. Hardware Error(%d).\n", rc);
 388                 break;
 389         case -3:
 390                 printk(KERN_ERR "Failed to register firmware-assisted kernel"
 391                         " dump. Parameter Error(%d).\n", rc);
 392                 break;
 393         case -9:
 394                 printk(KERN_ERR "firmware-assisted kernel dump is already "
 395                         " registered.");
 396                 fw_dump.dump_registered = 1;
 397                 break;
 398         case 0:
 399                 printk(KERN_INFO "firmware-assisted kernel dump registration"
 400                         " is successful\n");
 401                 fw_dump.dump_registered = 1;
 402                 break;
 403         }
 404 }
 405
 406 void crash_fadump(struct pt_regs *regs, const char *str)
 407 {
 408         struct fadump_crash_info_header *fdh = NULL;
 409
 410         if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
 411                 return;
 412
 413         fdh = __va(fw_dump.fadumphdr_addr);
 414         crashing_cpu = smp_processor_id();
 415         fdh->crashing_cpu = crashing_cpu;
 416         crash_save_vmcoreinfo();
 417
 418         if (regs)
 419                 fdh->regs = *regs;
 420         else
 421                 ppc_save_regs(&fdh->regs);
 422
 423         fdh->online_mask = *cpu_online_mask;
 424
 425         /* Call ibm,os-term rtas call to trigger firmware assisted dump */
 426         rtas_os_term((char *)str);
 427 }
 428
 429 #define GPR_MASK        0xffffff0000000000
 430 static inline int fadump_gpr_index(u64 id)
 431 {
 432         int i = -1;
 433         char str[3];
 434
 435         if ((id & GPR_MASK) == REG_ID("GPR")) {
 436                 /* get the digits at the end */
 437                 id &= ~GPR_MASK;
 438                 id >>= 24;
 439                 str[2] = '\0';
 440                 str[1] = id & 0xff;
 441                 str[0] = (id >> 8) & 0xff;
 442                 sscanf(str, "%d", &i);
 443                 if (i > 31)
 444                         i = -1;
 445         }
 446         return i;
 447 }
 448
 449 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
 450                                                                 u64 reg_val)
 451 {
 452         int i;
 453
 454         i = fadump_gpr_index(reg_id);
 455         if (i >= 0)
 456                 regs->gpr[i] = (unsigned long)reg_val;
 457         else if (reg_id == REG_ID("NIA"))
 458                 regs->nip = (unsigned long)reg_val;
 459         else if (reg_id == REG_ID("MSR"))
 460                 regs->msr = (unsigned long)reg_val;
 461         else if (reg_id == REG_ID("CTR"))
 462                 regs->ctr = (unsigned long)reg_val;
 463         else if (reg_id == REG_ID("LR"))
 464                 regs->link = (unsigned long)reg_val;
 465         else if (reg_id == REG_ID("XER"))
 466                 regs->xer = (unsigned long)reg_val;
 467         else if (reg_id == REG_ID("CR"))
 468                 regs->ccr = (unsigned long)reg_val;
 469         else if (reg_id == REG_ID("DAR"))
 470                 regs->dar = (unsigned long)reg_val;
 471         else if (reg_id == REG_ID("DSISR"))
 472                 regs->dsisr = (unsigned long)reg_val;
 473 }
 474
 475 static struct fadump_reg_entry*
 476 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
 477 {
 478         memset(regs, 0, sizeof(struct pt_regs));
 479
 480         while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
 481                 fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
 482                                         be64_to_cpu(reg_entry->reg_value));
 483                 reg_entry++;
 484         }
 485         reg_entry++;
 486         return reg_entry;
 487 }
 488
 489 static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
 490                                                 void *data, size_t data_len)
 491 {
 492         struct elf_note note;
 493
 494         note.n_namesz = strlen(name) + 1;
 495         note.n_descsz = data_len;
 496         note.n_type   = type;
 497         memcpy(buf, &note, sizeof(note));
 498         buf += (sizeof(note) + 3)/4;
 499         memcpy(buf, name, note.n_namesz);
 500         buf += (note.n_namesz + 3)/4;
 501         memcpy(buf, data, note.n_descsz);
 502         buf += (note.n_descsz + 3)/4;
 503
 504         return buf;
 505 }
 506
 507 static void fadump_final_note(u32 *buf)
 508 {
 509         struct elf_note note;
 510
 511         note.n_namesz = 0;
 512         note.n_descsz = 0;
 513         note.n_type   = 0;
 514         memcpy(buf, &note, sizeof(note));
 515 }
 516
 517 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
 518 {
 519         struct elf_prstatus prstatus;
 520
 521         memset(&prstatus, 0, sizeof(prstatus));
 522         /*
 523          * FIXME: How do i get PID? Do I really need it?
 524          * prstatus.pr_pid = ????
 525          */
 526         elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
 527         buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
 528                                 &prstatus, sizeof(prstatus));
 529         return buf;
 530 }
 531
 532 static void fadump_update_elfcore_header(char *bufp)
 533 {
 534         struct elfhdr *elf;
 535         struct elf_phdr *phdr;
 536
 537         elf = (struct elfhdr *)bufp;
 538         bufp += sizeof(struct elfhdr);
 539
 540         /* First note is a place holder for cpu notes info. */
 541         phdr = (struct elf_phdr *)bufp;
 542
 543         if (phdr->p_type == PT_NOTE) {
 544                 phdr->p_paddr = fw_dump.cpu_notes_buf;
 545                 phdr->p_offset  = phdr->p_paddr;
 546                 phdr->p_filesz  = fw_dump.cpu_notes_buf_size;
 547                 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
 548         }
 549         return;
 550 }
 551
 552 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
 553 {
 554         void *vaddr;
 555         struct page *page;
 556         unsigned long order, count, i;
 557
 558         order = get_order(size);
 559         vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
 560         if (!vaddr)
 561                 return NULL;
 562
 563         count = 1 << order;
 564         page = virt_to_page(vaddr);
 565         for (i = 0; i < count; i++)
 566                 SetPageReserved(page + i);
 567         return vaddr;
 568 }
 569
 570 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
 571 {
 572         struct page *page;
 573         unsigned long order, count, i;
 574
 575         order = get_order(size);
 576         count = 1 << order;
 577         page = virt_to_page(vaddr);
 578         for (i = 0; i < count; i++)
 579                 ClearPageReserved(page + i);
 580         __free_pages(page, order);
 581 }
 582
 583 /*
 584  * Read CPU state dump data and convert it into ELF notes.
 585  * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
 586  * used to access the data to allow for additional fields to be added without
 587  * affecting compatibility. Each list of registers for a CPU starts with
 588  * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
 589  * 8 Byte ASCII identifier and 8 Byte register value. The register entry
 590  * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
 591  * of register value. For more details refer to PAPR document.
 592  *
 593  * Only for the crashing cpu we ignore the CPU dump data and get exact
 594  * state from fadump crash info structure populated by first kernel at the
 595  * time of crash.
 596  */
 597 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
 598 {
 599         struct fadump_reg_save_area_header *reg_header;
 600         struct fadump_reg_entry *reg_entry;
 601         struct fadump_crash_info_header *fdh = NULL;
 602         void *vaddr;
 603         unsigned long addr;
 604         u32 num_cpus, *note_buf;
 605         struct pt_regs regs;
 606         int i, rc = 0, cpu = 0;
 607
 608         if (!fdm->cpu_state_data.bytes_dumped)
 609                 return -EINVAL;
 610
 611         addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
 612         vaddr = __va(addr);
 613
 614         reg_header = vaddr;
 615         if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
 616                 printk(KERN_ERR "Unable to read register save area.\n");
 617                 return -ENOENT;
 618         }
 619         pr_debug("--------CPU State Data------------\n");
 620         pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
 621         pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
 622
 623         vaddr += be32_to_cpu(reg_header->num_cpu_offset);
 624         num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
 625         pr_debug("NumCpus     : %u\n", num_cpus);
 626         vaddr += sizeof(u32);
 627         reg_entry = (struct fadump_reg_entry *)vaddr;
 628
 629         /* Allocate buffer to hold cpu crash notes. */
 630         fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
 631         fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
 632         note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
 633         if (!note_buf) {
 634                 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
 635                         "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
 636                 return -ENOMEM;
 637         }
 638         fw_dump.cpu_notes_buf = __pa(note_buf);
 639
 640         pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
 641                         (num_cpus * sizeof(note_buf_t)), note_buf);
 642
 643         if (fw_dump.fadumphdr_addr)
 644                 fdh = __va(fw_dump.fadumphdr_addr);
 645
 646         for (i = 0; i < num_cpus; i++) {
 647                 if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
 648                         printk(KERN_ERR "Unable to read CPU state data\n");
 649                         rc = -ENOENT;
 650                         goto error_out;
 651                 }
 652                 /* Lower 4 bytes of reg_value contains logical cpu id */
 653                 cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
 654                 if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
 655                         SKIP_TO_NEXT_CPU(reg_entry);
 656                         continue;
 657                 }
 658                 pr_debug("Reading register data for cpu %d...\n", cpu);
 659                 if (fdh && fdh->crashing_cpu == cpu) {
 660                         regs = fdh->regs;
 661                         note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 662                         SKIP_TO_NEXT_CPU(reg_entry);
 663                 } else {
 664                         reg_entry++;
 665                         reg_entry = fadump_read_registers(reg_entry, &regs);
 666                         note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 667                 }
 668         }
 669         fadump_final_note(note_buf);
 670
 671         if (fdh) {
 672                 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
 673                                                         fdh->elfcorehdr_addr);
 674                 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
 675         }
 676         return 0;
 677
 678 error_out:
 679         fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
 680                                         fw_dump.cpu_notes_buf_size);
 681         fw_dump.cpu_notes_buf = 0;
 682         fw_dump.cpu_notes_buf_size = 0;
 683         return rc;
 684
 685 }
 686
 687 /*
 688  * Validate and process the dump data stored by firmware before exporting
 689  * it through '/proc/vmcore'.
 690  */
 691 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
 692 {
 693         struct fadump_crash_info_header *fdh;
 694         int rc = 0;
 695
 696         if (!fdm_active || !fw_dump.fadumphdr_addr)
 697                 return -EINVAL;
 698
 699         /* Check if the dump data is valid. */
 700         if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
 701                         (fdm_active->cpu_state_data.error_flags != 0) ||
 702                         (fdm_active->rmr_region.error_flags != 0)) {
 703                 printk(KERN_ERR "Dump taken by platform is not valid\n");
 704                 return -EINVAL;
 705         }
 706         if ((fdm_active->rmr_region.bytes_dumped !=
 707                         fdm_active->rmr_region.source_len) ||
 708                         !fdm_active->cpu_state_data.bytes_dumped) {
 709                 printk(KERN_ERR "Dump taken by platform is incomplete\n");
 710                 return -EINVAL;
 711         }
 712
 713         /* Validate the fadump crash info header */
 714         fdh = __va(fw_dump.fadumphdr_addr);
 715         if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
 716                 printk(KERN_ERR "Crash info header is not valid.\n");
 717                 return -EINVAL;
 718         }
 719
 720         rc = fadump_build_cpu_notes(fdm_active);
 721         if (rc)
 722                 return rc;
 723
 724         /*
 725          * We are done validating dump info and elfcore header is now ready
 726          * to be exported. set elfcorehdr_addr so that vmcore module will
 727          * export the elfcore header through '/proc/vmcore'.
 728          */
 729         elfcorehdr_addr = fdh->elfcorehdr_addr;
 730
 731         return 0;
 732 }
 733
 734 static inline void fadump_add_crash_memory(unsigned long long base,
 735                                         unsigned long long end)
 736 {
 737         if (base == end)
 738                 return;
 739
 740         pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
 741                 crash_mem_ranges, base, end - 1, (end - base));
 742         crash_memory_ranges[crash_mem_ranges].base = base;
 743         crash_memory_ranges[crash_mem_ranges].size = end - base;
 744         crash_mem_ranges++;
 745 }
 746
 747 static void fadump_exclude_reserved_area(unsigned long long start,
 748                                         unsigned long long end)
 749 {
 750         unsigned long long ra_start, ra_end;
 751
 752         ra_start = fw_dump.reserve_dump_area_start;
 753         ra_end = ra_start + fw_dump.reserve_dump_area_size;
 754
 755         if ((ra_start < end) && (ra_end > start)) {
 756                 if ((start < ra_start) && (end > ra_end)) {
 757                         fadump_add_crash_memory(start, ra_start);
 758                         fadump_add_crash_memory(ra_end, end);
 759                 } else if (start < ra_start) {
 760                         fadump_add_crash_memory(start, ra_start);
 761                 } else if (ra_end < end) {
 762                         fadump_add_crash_memory(ra_end, end);
 763                 }
 764         } else
 765                 fadump_add_crash_memory(start, end);
 766 }
 767
 768 static int fadump_init_elfcore_header(char *bufp)
 769 {
 770         struct elfhdr *elf;
 771
 772         elf = (struct elfhdr *) bufp;
 773         bufp += sizeof(struct elfhdr);
 774         memcpy(elf->e_ident, ELFMAG, SELFMAG);
 775         elf->e_ident[EI_CLASS] = ELF_CLASS;
 776         elf->e_ident[EI_DATA] = ELF_DATA;
 777         elf->e_ident[EI_VERSION] = EV_CURRENT;
 778         elf->e_ident[EI_OSABI] = ELF_OSABI;
 779         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 780         elf->e_type = ET_CORE;
 781         elf->e_machine = ELF_ARCH;
 782         elf->e_version = EV_CURRENT;
 783         elf->e_entry = 0;
 784         elf->e_phoff = sizeof(struct elfhdr);
 785         elf->e_shoff = 0;
 786         elf->e_flags = ELF_CORE_EFLAGS;
 787         elf->e_ehsize = sizeof(struct elfhdr);
 788         elf->e_phentsize = sizeof(struct elf_phdr);
 789         elf->e_phnum = 0;
 790         elf->e_shentsize = 0;
 791         elf->e_shnum = 0;
 792         elf->e_shstrndx = 0;
 793
 794         return 0;
 795 }
 796
 797 /*
 798  * Traverse through memblock structure and setup crash memory ranges. These
 799  * ranges will be used create PT_LOAD program headers in elfcore header.
 800  */
 801 static void fadump_setup_crash_memory_ranges(void)
 802 {
 803         struct memblock_region *reg;
 804         unsigned long long start, end;
 805
 806         pr_debug("Setup crash memory ranges.\n");
 807         crash_mem_ranges = 0;
 808         /*
 809          * add the first memory chunk (RMA_START through boot_memory_size) as
 810          * a separate memory chunk. The reason is, at the time crash firmware
 811          * will move the content of this memory chunk to different location
 812          * specified during fadump registration. We need to create a separate
 813          * program header for this chunk with the correct offset.
 814          */
 815         fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
 816
 817         for_each_memblock(memory, reg) {
 818                 start = (unsigned long long)reg->base;
 819                 end = start + (unsigned long long)reg->size;
 820                 if (start == RMA_START && end >= fw_dump.boot_memory_size)
 821                         start = fw_dump.boot_memory_size;
 822
 823                 /* add this range excluding the reserved dump area. */
 824                 fadump_exclude_reserved_area(start, end);
 825         }
 826 }
 827
 828 /*
 829  * If the given physical address falls within the boot memory region then
 830  * return the relocated address that points to the dump region reserved
 831  * for saving initial boot memory contents.
 832  */
 833 static inline unsigned long fadump_relocate(unsigned long paddr)
 834 {
 835         if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
 836                 return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
 837         else
 838                 return paddr;
 839 }
 840
 841 static int fadump_create_elfcore_headers(char *bufp)
 842 {
 843         struct elfhdr *elf;
 844         struct elf_phdr *phdr;
 845         int i;
 846
 847         fadump_init_elfcore_header(bufp);
 848         elf = (struct elfhdr *)bufp;
 849         bufp += sizeof(struct elfhdr);
 850
 851         /*
 852          * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
 853          * will be populated during second kernel boot after crash. Hence
 854          * this PT_NOTE will always be the first elf note.
 855          *
 856          * NOTE: Any new ELF note addition should be placed after this note.
 857          */
 858         phdr = (struct elf_phdr *)bufp;
 859         bufp += sizeof(struct elf_phdr);
 860         phdr->p_type = PT_NOTE;
 861         phdr->p_flags = 0;
 862         phdr->p_vaddr = 0;
 863         phdr->p_align = 0;
 864
 865         phdr->p_offset = 0;
 866         phdr->p_paddr = 0;
 867         phdr->p_filesz = 0;
 868         phdr->p_memsz = 0;
 869
 870         (elf->e_phnum)++;
 871
 872         /* setup ELF PT_NOTE for vmcoreinfo */
 873         phdr = (struct elf_phdr *)bufp;
 874         bufp += sizeof(struct elf_phdr);
 875         phdr->p_type    = PT_NOTE;
 876         phdr->p_flags   = 0;
 877         phdr->p_vaddr   = 0;
 878         phdr->p_align   = 0;
 879
 880         phdr->p_paddr   = fadump_relocate(paddr_vmcoreinfo_note());
 881         phdr->p_offset  = phdr->p_paddr;
 882         phdr->p_memsz   = vmcoreinfo_max_size;
 883         phdr->p_filesz  = vmcoreinfo_max_size;
 884
 885         /* Increment number of program headers. */
 886         (elf->e_phnum)++;
 887
 888         /* setup PT_LOAD sections. */
 889
 890         for (i = 0; i < crash_mem_ranges; i++) {
 891                 unsigned long long mbase, msize;
 892                 mbase = crash_memory_ranges[i].base;
 893                 msize = crash_memory_ranges[i].size;
 894
 895                 if (!msize)
 896                         continue;
 897
 898                 phdr = (struct elf_phdr *)bufp;
 899                 bufp += sizeof(struct elf_phdr);
 900                 phdr->p_type    = PT_LOAD;
 901                 phdr->p_flags   = PF_R|PF_W|PF_X;
 902                 phdr->p_offset  = mbase;
 903
 904                 if (mbase == RMA_START) {
 905                         /*
 906                          * The entire RMA region will be moved by firmware
 907                          * to the specified destination_address. Hence set
 908                          * the correct offset.
 909                          */
 910                         phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
 911                 }
 912
 913                 phdr->p_paddr = mbase;
 914                 phdr->p_vaddr = (unsigned long)__va(mbase);
 915                 phdr->p_filesz = msize;
 916                 phdr->p_memsz = msize;
 917                 phdr->p_align = 0;
 918
 919                 /* Increment number of program headers. */
 920                 (elf->e_phnum)++;
 921         }
 922         return 0;
 923 }
 924
 925 static unsigned long init_fadump_header(unsigned long addr)
 926 {
 927         struct fadump_crash_info_header *fdh;
 928
 929         if (!addr)
 930                 return 0;
 931
 932         fw_dump.fadumphdr_addr = addr;
 933         fdh = __va(addr);
 934         addr += sizeof(struct fadump_crash_info_header);
 935
 936         memset(fdh, 0, sizeof(struct fadump_crash_info_header));
 937         fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
 938         fdh->elfcorehdr_addr = addr;
 939         /* We will set the crashing cpu id in crash_fadump() during crash. */
 940         fdh->crashing_cpu = CPU_UNKNOWN;
 941
 942         return addr;
 943 }
 944
 945 static void register_fadump(void)
 946 {
 947         unsigned long addr;
 948         void *vaddr;
 949
 950         /*
 951          * If no memory is reserved then we can not register for firmware-
 952          * assisted dump.
 953          */
 954         if (!fw_dump.reserve_dump_area_size)
 955                 return;
 956
 957         fadump_setup_crash_memory_ranges();
 958
 959         addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
 960         /* Initialize fadump crash info header. */
 961         addr = init_fadump_header(addr);
 962         vaddr = __va(addr);
 963
 964         pr_debug("Creating ELF core headers at %#016lx\n", addr);
 965         fadump_create_elfcore_headers(vaddr);
 966
 967         /* register the future kernel dump with firmware. */
 968         register_fw_dump(&fdm);
 969 }
 970
 971 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
 972 {
 973         int rc = 0;
 974         unsigned int wait_time;
 975
 976         pr_debug("Un-register firmware-assisted dump\n");
 977
 978         /* TODO: Add upper time limit for the delay */
 979         do {
 980                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
 981                         FADUMP_UNREGISTER, fdm,
 982                         sizeof(struct fadump_mem_struct));
 983
 984                 wait_time = rtas_busy_delay_time(rc);
 985                 if (wait_time)
 986                         mdelay(wait_time);
 987         } while (wait_time);
 988
 989         if (rc) {
 990                 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
 991                         " unexpected error(%d).\n", rc);
 992                 return rc;
 993         }
 994         fw_dump.dump_registered = 0;
 995         return 0;
 996 }
 997
 998 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
 999 {
1000         int rc = 0;
1001         unsigned int wait_time;
1002
1003         pr_debug("Invalidating firmware-assisted dump registration\n");
1004
1005         /* TODO: Add upper time limit for the delay */
1006         do {
1007                 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1008                         FADUMP_INVALIDATE, fdm,
1009                         sizeof(struct fadump_mem_struct));
1010
1011                 wait_time = rtas_busy_delay_time(rc);
1012                 if (wait_time)
1013                         mdelay(wait_time);
1014         } while (wait_time);
1015
1016         if (rc) {
1017                 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1018                 return rc;
1019         }
1020         fw_dump.dump_active = 0;
1021         fdm_active = NULL;
1022         return 0;
1023 }
1024
1025 void fadump_cleanup(void)
1026 {
1027         /* Invalidate the registration only if dump is active. */
1028         if (fw_dump.dump_active) {
1029                 init_fadump_mem_struct(&fdm,
1030                         be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1031                 fadump_invalidate_dump(&fdm);
1032         }
1033 }
1034
1035 /*
1036  * Release the memory that was reserved in early boot to preserve the memory
1037  * contents. The released memory will be available for general use.
1038  */
1039 static void fadump_release_memory(unsigned long begin, unsigned long end)
1040 {
1041         unsigned long addr;
1042         unsigned long ra_start, ra_end;
1043
1044         ra_start = fw_dump.reserve_dump_area_start;
1045         ra_end = ra_start + fw_dump.reserve_dump_area_size;
1046
1047         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1048                 /*
1049                  * exclude the dump reserve area. Will reuse it for next
1050                  * fadump registration.
1051                  */
1052                 if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1053                         continue;
1054
1055                 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1056         }
1057 }
1058
1059 static void fadump_invalidate_release_mem(void)
1060 {
1061         unsigned long reserved_area_start, reserved_area_end;
1062         unsigned long destination_address;
1063
1064         mutex_lock(&fadump_mutex);
1065         if (!fw_dump.dump_active) {
1066                 mutex_unlock(&fadump_mutex);
1067                 return;
1068         }
1069
1070         destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1071         fadump_cleanup();
1072         mutex_unlock(&fadump_mutex);
1073
1074         /*
1075          * Save the current reserved memory bounds we will require them
1076          * later for releasing the memory for general use.
1077          */
1078         reserved_area_start = fw_dump.reserve_dump_area_start;
1079         reserved_area_end = reserved_area_start +
1080                         fw_dump.reserve_dump_area_size;
1081         /*
1082          * Setup reserve_dump_area_start and its size so that we can
1083          * reuse this reserved memory for Re-registration.
1084          */
1085         fw_dump.reserve_dump_area_start = destination_address;
1086         fw_dump.reserve_dump_area_size = get_fadump_area_size();
1087
1088         fadump_release_memory(reserved_area_start, reserved_area_end);
1089         if (fw_dump.cpu_notes_buf) {
1090                 fadump_cpu_notes_buf_free(
1091                                 (unsigned long)__va(fw_dump.cpu_notes_buf),
1092                                 fw_dump.cpu_notes_buf_size);
1093                 fw_dump.cpu_notes_buf = 0;
1094                 fw_dump.cpu_notes_buf_size = 0;
1095         }
1096         /* Initialize the kernel dump memory structure for FAD registration. */
1097         init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1098 }
1099
1100 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1101                                         struct kobj_attribute *attr,
1102                                         const char *buf, size_t count)
1103 {
1104         if (!fw_dump.dump_active)
1105                 return -EPERM;
1106
1107         if (buf[0] == '1') {
1108                 /*
1109                  * Take away the '/proc/vmcore'. We are releasing the dump
1110                  * memory, hence it will not be valid anymore.
1111                  */
1112                 vmcore_cleanup();
1113                 fadump_invalidate_release_mem();
1114
1115         } else
1116                 return -EINVAL;
1117         return count;
1118 }
1119
1120 static ssize_t fadump_enabled_show(struct kobject *kobj,
1121                                         struct kobj_attribute *attr,
1122                                         char *buf)
1123 {
1124         return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1125 }
1126
1127 static ssize_t fadump_register_show(struct kobject *kobj,
1128                                         struct kobj_attribute *attr,
1129                                         char *buf)
1130 {
1131         return sprintf(buf, "%d\n", fw_dump.dump_registered);
1132 }
1133
1134 static ssize_t fadump_register_store(struct kobject *kobj,
1135                                         struct kobj_attribute *attr,
1136                                         const char *buf, size_t count)
1137 {
1138         int ret = 0;
1139
1140         if (!fw_dump.fadump_enabled || fdm_active)
1141                 return -EPERM;
1142
1143         mutex_lock(&fadump_mutex);
1144
1145         switch (buf[0]) {
1146         case '0':
1147                 if (fw_dump.dump_registered == 0) {
1148                         ret = -EINVAL;
1149                         goto unlock_out;
1150                 }
1151                 /* Un-register Firmware-assisted dump */
1152                 fadump_unregister_dump(&fdm);
1153                 break;
1154         case '1':
1155                 if (fw_dump.dump_registered == 1) {
1156                         ret = -EINVAL;
1157                         goto unlock_out;
1158                 }
1159                 /* Register Firmware-assisted dump */
1160                 register_fadump();
1161                 break;
1162         default:
1163                 ret = -EINVAL;
1164                 break;
1165         }
1166
1167 unlock_out:
1168         mutex_unlock(&fadump_mutex);
1169         return ret < 0 ? ret : count;
1170 }
1171
1172 static int fadump_region_show(struct seq_file *m, void *private)
1173 {
1174         const struct fadump_mem_struct *fdm_ptr;
1175
1176         if (!fw_dump.fadump_enabled)
1177                 return 0;
1178
1179         mutex_lock(&fadump_mutex);
1180         if (fdm_active)
1181                 fdm_ptr = fdm_active;
1182         else {
1183                 mutex_unlock(&fadump_mutex);
1184                 fdm_ptr = &fdm;
1185         }
1186
1187         seq_printf(m,
1188                         "CPU : [%#016llx-%#016llx] %#llx bytes, "
1189                         "Dumped: %#llx\n",
1190                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1191                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1192                         be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1193                         be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1194                         be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1195         seq_printf(m,
1196                         "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1197                         "Dumped: %#llx\n",
1198                         be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1199                         be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1200                         be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1201                         be64_to_cpu(fdm_ptr->hpte_region.source_len),
1202                         be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1203         seq_printf(m,
1204                         "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1205                         "Dumped: %#llx\n",
1206                         be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1207                         be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1208                         be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1209                         be64_to_cpu(fdm_ptr->rmr_region.source_len),
1210                         be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1211
1212         if (!fdm_active ||
1213                 (fw_dump.reserve_dump_area_start ==
1214                 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1215                 goto out;
1216
1217         /* Dump is active. Show reserved memory region. */
1218         seq_printf(m,
1219                         "    : [%#016llx-%#016llx] %#llx bytes, "
1220                         "Dumped: %#llx\n",
1221                         (unsigned long long)fw_dump.reserve_dump_area_start,
1222                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1223                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1224                         fw_dump.reserve_dump_area_start,
1225                         be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1226                         fw_dump.reserve_dump_area_start);
1227 out:
1228         if (fdm_active)
1229                 mutex_unlock(&fadump_mutex);
1230         return 0;
1231 }
1232
1233 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1234                                                 0200, NULL,
1235                                                 fadump_release_memory_store);
1236 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1237                                                 0444, fadump_enabled_show,
1238                                                 NULL);
1239 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1240                                                 0644, fadump_register_show,
1241                                                 fadump_register_store);
1242
1243 static int fadump_region_open(struct inode *inode, struct file *file)
1244 {
1245         return single_open(file, fadump_region_show, inode->i_private);
1246 }
1247
1248 static const struct file_operations fadump_region_fops = {
1249         .open    = fadump_region_open,
1250         .read    = seq_read,
1251         .llseek  = seq_lseek,
1252         .release = single_release,
1253 };
1254
1255 static void fadump_init_files(void)
1256 {
1257         struct dentry *debugfs_file;
1258         int rc = 0;
1259
1260         rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1261         if (rc)
1262                 printk(KERN_ERR "fadump: unable to create sysfs file"
1263                         " fadump_enabled (%d)\n", rc);
1264
1265         rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1266         if (rc)
1267                 printk(KERN_ERR "fadump: unable to create sysfs file"
1268                         " fadump_registered (%d)\n", rc);
1269
1270         debugfs_file = debugfs_create_file("fadump_region", 0444,
1271                                         powerpc_debugfs_root, NULL,
1272                                         &fadump_region_fops);
1273         if (!debugfs_file)
1274                 printk(KERN_ERR "fadump: unable to create debugfs file"
1275                                 " fadump_region\n");
1276
1277         if (fw_dump.dump_active) {
1278                 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1279                 if (rc)
1280                         printk(KERN_ERR "fadump: unable to create sysfs file"
1281                                 " fadump_release_mem (%d)\n", rc);
1282         }
1283         return;
1284 }
1285
1286 /*
1287  * Prepare for firmware-assisted dump.
1288  */
1289 int __init setup_fadump(void)
1290 {
1291         if (!fw_dump.fadump_enabled)
1292                 return 0;
1293
1294         if (!fw_dump.fadump_supported) {
1295                 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1296                         " this hardware\n");
1297                 return 0;
1298         }
1299
1300         fadump_show_config();
1301         /*
1302          * If dump data is available then see if it is valid and prepare for
1303          * saving it to the disk.
1304          */
1305         if (fw_dump.dump_active) {
1306                 /*
1307                  * if dump process fails then invalidate the registration
1308                  * and release memory before proceeding for re-registration.
1309                  */
1310                 if (process_fadump(fdm_active) < 0)
1311                         fadump_invalidate_release_mem();
1312         }
1313         /* Initialize the kernel dump memory structure for FAD registration. */
1314         else if (fw_dump.reserve_dump_area_size)
1315                 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1316         fadump_init_files();
1317
1318         return 1;
1319 }
1320 subsys_initcall(setup_fadump);