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