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