Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 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> |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 32 | #include <linux/delay.h> |
| 33 | #include <linux/debugfs.h> |
| 34 | #include <linux/seq_file.h> |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 35 | #include <linux/crash_dump.h> |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 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; |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 43 | static struct fadump_mem_struct fdm; |
| 44 | static const struct fadump_mem_struct *fdm_active; |
| 45 | |
| 46 | static DEFINE_MUTEX(fadump_mutex); |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 47 | struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES]; |
| 48 | int crash_mem_ranges; |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 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 | */ |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 77 | fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL); |
| 78 | if (fdm_active) |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 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(§ions[1], 2); |
| 102 | break; |
| 103 | case FADUMP_HPTE_REGION: |
| 104 | fw_dump.hpte_region_size = |
| 105 | of_read_ulong(§ions[1], 2); |
| 106 | break; |
| 107 | } |
| 108 | } |
| 109 | return 1; |
| 110 | } |
| 111 | |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 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 | |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 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; |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 241 | size += sizeof(struct fadump_crash_info_header); |
| 242 | size += sizeof(struct elfhdr); /* ELF core header.*/ |
| 243 | /* Program headers for crash memory regions. */ |
| 244 | size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2); |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 245 | |
| 246 | size = PAGE_ALIGN(size); |
| 247 | return size; |
| 248 | } |
| 249 | |
| 250 | int __init fadump_reserve_mem(void) |
| 251 | { |
| 252 | unsigned long base, size, memory_boundary; |
| 253 | |
| 254 | if (!fw_dump.fadump_enabled) |
| 255 | return 0; |
| 256 | |
| 257 | if (!fw_dump.fadump_supported) { |
| 258 | printk(KERN_INFO "Firmware-assisted dump is not supported on" |
| 259 | " this hardware\n"); |
| 260 | fw_dump.fadump_enabled = 0; |
| 261 | return 0; |
| 262 | } |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 263 | /* |
| 264 | * Initialize boot memory size |
| 265 | * If dump is active then we have already calculated the size during |
| 266 | * first kernel. |
| 267 | */ |
| 268 | if (fdm_active) |
| 269 | fw_dump.boot_memory_size = fdm_active->rmr_region.source_len; |
| 270 | else |
| 271 | fw_dump.boot_memory_size = fadump_calculate_reserve_size(); |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 272 | |
| 273 | /* |
| 274 | * Calculate the memory boundary. |
| 275 | * If memory_limit is less than actual memory boundary then reserve |
| 276 | * the memory for fadump beyond the memory_limit and adjust the |
| 277 | * memory_limit accordingly, so that the running kernel can run with |
| 278 | * specified memory_limit. |
| 279 | */ |
| 280 | if (memory_limit && memory_limit < memblock_end_of_DRAM()) { |
| 281 | size = get_fadump_area_size(); |
| 282 | if ((memory_limit + size) < memblock_end_of_DRAM()) |
| 283 | memory_limit += size; |
| 284 | else |
| 285 | memory_limit = memblock_end_of_DRAM(); |
| 286 | printk(KERN_INFO "Adjusted memory_limit for firmware-assisted" |
| 287 | " dump, now %#016llx\n", |
| 288 | (unsigned long long)memory_limit); |
| 289 | } |
| 290 | if (memory_limit) |
| 291 | memory_boundary = memory_limit; |
| 292 | else |
| 293 | memory_boundary = memblock_end_of_DRAM(); |
| 294 | |
| 295 | if (fw_dump.dump_active) { |
| 296 | printk(KERN_INFO "Firmware-assisted dump is active.\n"); |
| 297 | /* |
| 298 | * If last boot has crashed then reserve all the memory |
| 299 | * above boot_memory_size so that we don't touch it until |
| 300 | * dump is written to disk by userspace tool. This memory |
| 301 | * will be released for general use once the dump is saved. |
| 302 | */ |
| 303 | base = fw_dump.boot_memory_size; |
| 304 | size = memory_boundary - base; |
| 305 | memblock_reserve(base, size); |
| 306 | printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " |
| 307 | "for saving crash dump\n", |
| 308 | (unsigned long)(size >> 20), |
| 309 | (unsigned long)(base >> 20)); |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 310 | |
| 311 | fw_dump.fadumphdr_addr = |
| 312 | fdm_active->rmr_region.destination_address + |
| 313 | fdm_active->rmr_region.source_len; |
| 314 | pr_debug("fadumphdr_addr = %p\n", |
| 315 | (void *) fw_dump.fadumphdr_addr); |
Mahesh Salgaonkar | eb39c88 | 2012-02-16 01:14:22 +0000 | [diff] [blame] | 316 | } else { |
| 317 | /* Reserve the memory at the top of memory. */ |
| 318 | size = get_fadump_area_size(); |
| 319 | base = memory_boundary - size; |
| 320 | memblock_reserve(base, size); |
| 321 | printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " |
| 322 | "for firmware-assisted dump\n", |
| 323 | (unsigned long)(size >> 20), |
| 324 | (unsigned long)(base >> 20)); |
| 325 | } |
| 326 | fw_dump.reserve_dump_area_start = base; |
| 327 | fw_dump.reserve_dump_area_size = size; |
| 328 | return 1; |
| 329 | } |
| 330 | |
| 331 | /* Look for fadump= cmdline option. */ |
| 332 | static int __init early_fadump_param(char *p) |
| 333 | { |
| 334 | if (!p) |
| 335 | return 1; |
| 336 | |
| 337 | if (strncmp(p, "on", 2) == 0) |
| 338 | fw_dump.fadump_enabled = 1; |
| 339 | else if (strncmp(p, "off", 3) == 0) |
| 340 | fw_dump.fadump_enabled = 0; |
| 341 | |
| 342 | return 0; |
| 343 | } |
| 344 | early_param("fadump", early_fadump_param); |
| 345 | |
| 346 | /* Look for fadump_reserve_mem= cmdline option */ |
| 347 | static int __init early_fadump_reserve_mem(char *p) |
| 348 | { |
| 349 | if (p) |
| 350 | fw_dump.reserve_bootvar = memparse(p, &p); |
| 351 | return 0; |
| 352 | } |
| 353 | early_param("fadump_reserve_mem", early_fadump_reserve_mem); |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 354 | |
| 355 | static void register_fw_dump(struct fadump_mem_struct *fdm) |
| 356 | { |
| 357 | int rc; |
| 358 | unsigned int wait_time; |
| 359 | |
| 360 | pr_debug("Registering for firmware-assisted kernel dump...\n"); |
| 361 | |
| 362 | /* TODO: Add upper time limit for the delay */ |
| 363 | do { |
| 364 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, |
| 365 | FADUMP_REGISTER, fdm, |
| 366 | sizeof(struct fadump_mem_struct)); |
| 367 | |
| 368 | wait_time = rtas_busy_delay_time(rc); |
| 369 | if (wait_time) |
| 370 | mdelay(wait_time); |
| 371 | |
| 372 | } while (wait_time); |
| 373 | |
| 374 | switch (rc) { |
| 375 | case -1: |
| 376 | printk(KERN_ERR "Failed to register firmware-assisted kernel" |
| 377 | " dump. Hardware Error(%d).\n", rc); |
| 378 | break; |
| 379 | case -3: |
| 380 | printk(KERN_ERR "Failed to register firmware-assisted kernel" |
| 381 | " dump. Parameter Error(%d).\n", rc); |
| 382 | break; |
| 383 | case -9: |
| 384 | printk(KERN_ERR "firmware-assisted kernel dump is already " |
| 385 | " registered."); |
| 386 | fw_dump.dump_registered = 1; |
| 387 | break; |
| 388 | case 0: |
| 389 | printk(KERN_INFO "firmware-assisted kernel dump registration" |
| 390 | " is successful\n"); |
| 391 | fw_dump.dump_registered = 1; |
| 392 | break; |
| 393 | } |
| 394 | } |
| 395 | |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 396 | /* |
| 397 | * Validate and process the dump data stored by firmware before exporting |
| 398 | * it through '/proc/vmcore'. |
| 399 | */ |
| 400 | static int __init process_fadump(const struct fadump_mem_struct *fdm_active) |
| 401 | { |
| 402 | struct fadump_crash_info_header *fdh; |
| 403 | |
| 404 | if (!fdm_active || !fw_dump.fadumphdr_addr) |
| 405 | return -EINVAL; |
| 406 | |
| 407 | /* Check if the dump data is valid. */ |
| 408 | if ((fdm_active->header.dump_status_flag == FADUMP_ERROR_FLAG) || |
| 409 | (fdm_active->rmr_region.error_flags != 0)) { |
| 410 | printk(KERN_ERR "Dump taken by platform is not valid\n"); |
| 411 | return -EINVAL; |
| 412 | } |
| 413 | if (fdm_active->rmr_region.bytes_dumped != |
| 414 | fdm_active->rmr_region.source_len) { |
| 415 | printk(KERN_ERR "Dump taken by platform is incomplete\n"); |
| 416 | return -EINVAL; |
| 417 | } |
| 418 | |
| 419 | /* Validate the fadump crash info header */ |
| 420 | fdh = __va(fw_dump.fadumphdr_addr); |
| 421 | if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) { |
| 422 | printk(KERN_ERR "Crash info header is not valid.\n"); |
| 423 | return -EINVAL; |
| 424 | } |
| 425 | |
| 426 | /* |
| 427 | * We are done validating dump info and elfcore header is now ready |
| 428 | * to be exported. set elfcorehdr_addr so that vmcore module will |
| 429 | * export the elfcore header through '/proc/vmcore'. |
| 430 | */ |
| 431 | elfcorehdr_addr = fdh->elfcorehdr_addr; |
| 432 | |
| 433 | return 0; |
| 434 | } |
| 435 | |
| 436 | static inline void fadump_add_crash_memory(unsigned long long base, |
| 437 | unsigned long long end) |
| 438 | { |
| 439 | if (base == end) |
| 440 | return; |
| 441 | |
| 442 | pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n", |
| 443 | crash_mem_ranges, base, end - 1, (end - base)); |
| 444 | crash_memory_ranges[crash_mem_ranges].base = base; |
| 445 | crash_memory_ranges[crash_mem_ranges].size = end - base; |
| 446 | crash_mem_ranges++; |
| 447 | } |
| 448 | |
| 449 | static void fadump_exclude_reserved_area(unsigned long long start, |
| 450 | unsigned long long end) |
| 451 | { |
| 452 | unsigned long long ra_start, ra_end; |
| 453 | |
| 454 | ra_start = fw_dump.reserve_dump_area_start; |
| 455 | ra_end = ra_start + fw_dump.reserve_dump_area_size; |
| 456 | |
| 457 | if ((ra_start < end) && (ra_end > start)) { |
| 458 | if ((start < ra_start) && (end > ra_end)) { |
| 459 | fadump_add_crash_memory(start, ra_start); |
| 460 | fadump_add_crash_memory(ra_end, end); |
| 461 | } else if (start < ra_start) { |
| 462 | fadump_add_crash_memory(start, ra_start); |
| 463 | } else if (ra_end < end) { |
| 464 | fadump_add_crash_memory(ra_end, end); |
| 465 | } |
| 466 | } else |
| 467 | fadump_add_crash_memory(start, end); |
| 468 | } |
| 469 | |
| 470 | static int fadump_init_elfcore_header(char *bufp) |
| 471 | { |
| 472 | struct elfhdr *elf; |
| 473 | |
| 474 | elf = (struct elfhdr *) bufp; |
| 475 | bufp += sizeof(struct elfhdr); |
| 476 | memcpy(elf->e_ident, ELFMAG, SELFMAG); |
| 477 | elf->e_ident[EI_CLASS] = ELF_CLASS; |
| 478 | elf->e_ident[EI_DATA] = ELF_DATA; |
| 479 | elf->e_ident[EI_VERSION] = EV_CURRENT; |
| 480 | elf->e_ident[EI_OSABI] = ELF_OSABI; |
| 481 | memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); |
| 482 | elf->e_type = ET_CORE; |
| 483 | elf->e_machine = ELF_ARCH; |
| 484 | elf->e_version = EV_CURRENT; |
| 485 | elf->e_entry = 0; |
| 486 | elf->e_phoff = sizeof(struct elfhdr); |
| 487 | elf->e_shoff = 0; |
| 488 | elf->e_flags = ELF_CORE_EFLAGS; |
| 489 | elf->e_ehsize = sizeof(struct elfhdr); |
| 490 | elf->e_phentsize = sizeof(struct elf_phdr); |
| 491 | elf->e_phnum = 0; |
| 492 | elf->e_shentsize = 0; |
| 493 | elf->e_shnum = 0; |
| 494 | elf->e_shstrndx = 0; |
| 495 | |
| 496 | return 0; |
| 497 | } |
| 498 | |
| 499 | /* |
| 500 | * Traverse through memblock structure and setup crash memory ranges. These |
| 501 | * ranges will be used create PT_LOAD program headers in elfcore header. |
| 502 | */ |
| 503 | static void fadump_setup_crash_memory_ranges(void) |
| 504 | { |
| 505 | struct memblock_region *reg; |
| 506 | unsigned long long start, end; |
| 507 | |
| 508 | pr_debug("Setup crash memory ranges.\n"); |
| 509 | crash_mem_ranges = 0; |
| 510 | /* |
| 511 | * add the first memory chunk (RMA_START through boot_memory_size) as |
| 512 | * a separate memory chunk. The reason is, at the time crash firmware |
| 513 | * will move the content of this memory chunk to different location |
| 514 | * specified during fadump registration. We need to create a separate |
| 515 | * program header for this chunk with the correct offset. |
| 516 | */ |
| 517 | fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size); |
| 518 | |
| 519 | for_each_memblock(memory, reg) { |
| 520 | start = (unsigned long long)reg->base; |
| 521 | end = start + (unsigned long long)reg->size; |
| 522 | if (start == RMA_START && end >= fw_dump.boot_memory_size) |
| 523 | start = fw_dump.boot_memory_size; |
| 524 | |
| 525 | /* add this range excluding the reserved dump area. */ |
| 526 | fadump_exclude_reserved_area(start, end); |
| 527 | } |
| 528 | } |
| 529 | |
| 530 | static int fadump_create_elfcore_headers(char *bufp) |
| 531 | { |
| 532 | struct elfhdr *elf; |
| 533 | struct elf_phdr *phdr; |
| 534 | int i; |
| 535 | |
| 536 | fadump_init_elfcore_header(bufp); |
| 537 | elf = (struct elfhdr *)bufp; |
| 538 | bufp += sizeof(struct elfhdr); |
| 539 | |
| 540 | /* setup PT_LOAD sections. */ |
| 541 | |
| 542 | for (i = 0; i < crash_mem_ranges; i++) { |
| 543 | unsigned long long mbase, msize; |
| 544 | mbase = crash_memory_ranges[i].base; |
| 545 | msize = crash_memory_ranges[i].size; |
| 546 | |
| 547 | if (!msize) |
| 548 | continue; |
| 549 | |
| 550 | phdr = (struct elf_phdr *)bufp; |
| 551 | bufp += sizeof(struct elf_phdr); |
| 552 | phdr->p_type = PT_LOAD; |
| 553 | phdr->p_flags = PF_R|PF_W|PF_X; |
| 554 | phdr->p_offset = mbase; |
| 555 | |
| 556 | if (mbase == RMA_START) { |
| 557 | /* |
| 558 | * The entire RMA region will be moved by firmware |
| 559 | * to the specified destination_address. Hence set |
| 560 | * the correct offset. |
| 561 | */ |
| 562 | phdr->p_offset = fdm.rmr_region.destination_address; |
| 563 | } |
| 564 | |
| 565 | phdr->p_paddr = mbase; |
| 566 | phdr->p_vaddr = (unsigned long)__va(mbase); |
| 567 | phdr->p_filesz = msize; |
| 568 | phdr->p_memsz = msize; |
| 569 | phdr->p_align = 0; |
| 570 | |
| 571 | /* Increment number of program headers. */ |
| 572 | (elf->e_phnum)++; |
| 573 | } |
| 574 | return 0; |
| 575 | } |
| 576 | |
| 577 | static unsigned long init_fadump_header(unsigned long addr) |
| 578 | { |
| 579 | struct fadump_crash_info_header *fdh; |
| 580 | |
| 581 | if (!addr) |
| 582 | return 0; |
| 583 | |
| 584 | fw_dump.fadumphdr_addr = addr; |
| 585 | fdh = __va(addr); |
| 586 | addr += sizeof(struct fadump_crash_info_header); |
| 587 | |
| 588 | memset(fdh, 0, sizeof(struct fadump_crash_info_header)); |
| 589 | fdh->magic_number = FADUMP_CRASH_INFO_MAGIC; |
| 590 | fdh->elfcorehdr_addr = addr; |
| 591 | |
| 592 | return addr; |
| 593 | } |
| 594 | |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 595 | static void register_fadump(void) |
| 596 | { |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 597 | unsigned long addr; |
| 598 | void *vaddr; |
| 599 | |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 600 | /* |
| 601 | * If no memory is reserved then we can not register for firmware- |
| 602 | * assisted dump. |
| 603 | */ |
| 604 | if (!fw_dump.reserve_dump_area_size) |
| 605 | return; |
| 606 | |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 607 | fadump_setup_crash_memory_ranges(); |
| 608 | |
| 609 | addr = fdm.rmr_region.destination_address + fdm.rmr_region.source_len; |
| 610 | /* Initialize fadump crash info header. */ |
| 611 | addr = init_fadump_header(addr); |
| 612 | vaddr = __va(addr); |
| 613 | |
| 614 | pr_debug("Creating ELF core headers at %#016lx\n", addr); |
| 615 | fadump_create_elfcore_headers(vaddr); |
| 616 | |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 617 | /* register the future kernel dump with firmware. */ |
| 618 | register_fw_dump(&fdm); |
| 619 | } |
| 620 | |
| 621 | static int fadump_unregister_dump(struct fadump_mem_struct *fdm) |
| 622 | { |
| 623 | int rc = 0; |
| 624 | unsigned int wait_time; |
| 625 | |
| 626 | pr_debug("Un-register firmware-assisted dump\n"); |
| 627 | |
| 628 | /* TODO: Add upper time limit for the delay */ |
| 629 | do { |
| 630 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, |
| 631 | FADUMP_UNREGISTER, fdm, |
| 632 | sizeof(struct fadump_mem_struct)); |
| 633 | |
| 634 | wait_time = rtas_busy_delay_time(rc); |
| 635 | if (wait_time) |
| 636 | mdelay(wait_time); |
| 637 | } while (wait_time); |
| 638 | |
| 639 | if (rc) { |
| 640 | printk(KERN_ERR "Failed to un-register firmware-assisted dump." |
| 641 | " unexpected error(%d).\n", rc); |
| 642 | return rc; |
| 643 | } |
| 644 | fw_dump.dump_registered = 0; |
| 645 | return 0; |
| 646 | } |
| 647 | |
| 648 | static ssize_t fadump_enabled_show(struct kobject *kobj, |
| 649 | struct kobj_attribute *attr, |
| 650 | char *buf) |
| 651 | { |
| 652 | return sprintf(buf, "%d\n", fw_dump.fadump_enabled); |
| 653 | } |
| 654 | |
| 655 | static ssize_t fadump_register_show(struct kobject *kobj, |
| 656 | struct kobj_attribute *attr, |
| 657 | char *buf) |
| 658 | { |
| 659 | return sprintf(buf, "%d\n", fw_dump.dump_registered); |
| 660 | } |
| 661 | |
| 662 | static ssize_t fadump_register_store(struct kobject *kobj, |
| 663 | struct kobj_attribute *attr, |
| 664 | const char *buf, size_t count) |
| 665 | { |
| 666 | int ret = 0; |
| 667 | |
| 668 | if (!fw_dump.fadump_enabled || fdm_active) |
| 669 | return -EPERM; |
| 670 | |
| 671 | mutex_lock(&fadump_mutex); |
| 672 | |
| 673 | switch (buf[0]) { |
| 674 | case '0': |
| 675 | if (fw_dump.dump_registered == 0) { |
| 676 | ret = -EINVAL; |
| 677 | goto unlock_out; |
| 678 | } |
| 679 | /* Un-register Firmware-assisted dump */ |
| 680 | fadump_unregister_dump(&fdm); |
| 681 | break; |
| 682 | case '1': |
| 683 | if (fw_dump.dump_registered == 1) { |
| 684 | ret = -EINVAL; |
| 685 | goto unlock_out; |
| 686 | } |
| 687 | /* Register Firmware-assisted dump */ |
| 688 | register_fadump(); |
| 689 | break; |
| 690 | default: |
| 691 | ret = -EINVAL; |
| 692 | break; |
| 693 | } |
| 694 | |
| 695 | unlock_out: |
| 696 | mutex_unlock(&fadump_mutex); |
| 697 | return ret < 0 ? ret : count; |
| 698 | } |
| 699 | |
| 700 | static int fadump_region_show(struct seq_file *m, void *private) |
| 701 | { |
| 702 | const struct fadump_mem_struct *fdm_ptr; |
| 703 | |
| 704 | if (!fw_dump.fadump_enabled) |
| 705 | return 0; |
| 706 | |
| 707 | if (fdm_active) |
| 708 | fdm_ptr = fdm_active; |
| 709 | else |
| 710 | fdm_ptr = &fdm; |
| 711 | |
| 712 | seq_printf(m, |
| 713 | "CPU : [%#016llx-%#016llx] %#llx bytes, " |
| 714 | "Dumped: %#llx\n", |
| 715 | fdm_ptr->cpu_state_data.destination_address, |
| 716 | fdm_ptr->cpu_state_data.destination_address + |
| 717 | fdm_ptr->cpu_state_data.source_len - 1, |
| 718 | fdm_ptr->cpu_state_data.source_len, |
| 719 | fdm_ptr->cpu_state_data.bytes_dumped); |
| 720 | seq_printf(m, |
| 721 | "HPTE: [%#016llx-%#016llx] %#llx bytes, " |
| 722 | "Dumped: %#llx\n", |
| 723 | fdm_ptr->hpte_region.destination_address, |
| 724 | fdm_ptr->hpte_region.destination_address + |
| 725 | fdm_ptr->hpte_region.source_len - 1, |
| 726 | fdm_ptr->hpte_region.source_len, |
| 727 | fdm_ptr->hpte_region.bytes_dumped); |
| 728 | seq_printf(m, |
| 729 | "DUMP: [%#016llx-%#016llx] %#llx bytes, " |
| 730 | "Dumped: %#llx\n", |
| 731 | fdm_ptr->rmr_region.destination_address, |
| 732 | fdm_ptr->rmr_region.destination_address + |
| 733 | fdm_ptr->rmr_region.source_len - 1, |
| 734 | fdm_ptr->rmr_region.source_len, |
| 735 | fdm_ptr->rmr_region.bytes_dumped); |
| 736 | |
| 737 | if (!fdm_active || |
| 738 | (fw_dump.reserve_dump_area_start == |
| 739 | fdm_ptr->cpu_state_data.destination_address)) |
| 740 | return 0; |
| 741 | |
| 742 | /* Dump is active. Show reserved memory region. */ |
| 743 | seq_printf(m, |
| 744 | " : [%#016llx-%#016llx] %#llx bytes, " |
| 745 | "Dumped: %#llx\n", |
| 746 | (unsigned long long)fw_dump.reserve_dump_area_start, |
| 747 | fdm_ptr->cpu_state_data.destination_address - 1, |
| 748 | fdm_ptr->cpu_state_data.destination_address - |
| 749 | fw_dump.reserve_dump_area_start, |
| 750 | fdm_ptr->cpu_state_data.destination_address - |
| 751 | fw_dump.reserve_dump_area_start); |
| 752 | return 0; |
| 753 | } |
| 754 | |
| 755 | static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled, |
| 756 | 0444, fadump_enabled_show, |
| 757 | NULL); |
| 758 | static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered, |
| 759 | 0644, fadump_register_show, |
| 760 | fadump_register_store); |
| 761 | |
| 762 | static int fadump_region_open(struct inode *inode, struct file *file) |
| 763 | { |
| 764 | return single_open(file, fadump_region_show, inode->i_private); |
| 765 | } |
| 766 | |
| 767 | static const struct file_operations fadump_region_fops = { |
| 768 | .open = fadump_region_open, |
| 769 | .read = seq_read, |
| 770 | .llseek = seq_lseek, |
| 771 | .release = single_release, |
| 772 | }; |
| 773 | |
| 774 | static void fadump_init_files(void) |
| 775 | { |
| 776 | struct dentry *debugfs_file; |
| 777 | int rc = 0; |
| 778 | |
| 779 | rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr); |
| 780 | if (rc) |
| 781 | printk(KERN_ERR "fadump: unable to create sysfs file" |
| 782 | " fadump_enabled (%d)\n", rc); |
| 783 | |
| 784 | rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr); |
| 785 | if (rc) |
| 786 | printk(KERN_ERR "fadump: unable to create sysfs file" |
| 787 | " fadump_registered (%d)\n", rc); |
| 788 | |
| 789 | debugfs_file = debugfs_create_file("fadump_region", 0444, |
| 790 | powerpc_debugfs_root, NULL, |
| 791 | &fadump_region_fops); |
| 792 | if (!debugfs_file) |
| 793 | printk(KERN_ERR "fadump: unable to create debugfs file" |
| 794 | " fadump_region\n"); |
| 795 | return; |
| 796 | } |
| 797 | |
| 798 | /* |
| 799 | * Prepare for firmware-assisted dump. |
| 800 | */ |
| 801 | int __init setup_fadump(void) |
| 802 | { |
| 803 | if (!fw_dump.fadump_enabled) |
| 804 | return 0; |
| 805 | |
| 806 | if (!fw_dump.fadump_supported) { |
| 807 | printk(KERN_ERR "Firmware-assisted dump is not supported on" |
| 808 | " this hardware\n"); |
| 809 | return 0; |
| 810 | } |
| 811 | |
| 812 | fadump_show_config(); |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 813 | /* |
| 814 | * If dump data is available then see if it is valid and prepare for |
| 815 | * saving it to the disk. |
| 816 | */ |
| 817 | if (fw_dump.dump_active) |
| 818 | process_fadump(fdm_active); |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 819 | /* Initialize the kernel dump memory structure for FAD registration. */ |
Mahesh Salgaonkar | 2df173d | 2012-02-16 01:14:37 +0000 | [diff] [blame^] | 820 | else if (fw_dump.reserve_dump_area_size) |
Mahesh Salgaonkar | 3ccc00a | 2012-02-20 02:15:03 +0000 | [diff] [blame] | 821 | init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start); |
| 822 | fadump_init_files(); |
| 823 | |
| 824 | return 1; |
| 825 | } |
| 826 | subsys_initcall(setup_fadump); |