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