Pileus Git - ~andy/linux/blob - arch/powerpc/kernel/fadump.c

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