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