| 1 | /* |
| 2 | * Handle the memory map. |
| 3 | * The functions here do the job until bootmem takes over. |
| 4 | * |
| 5 | * Getting sanitize_e820_map() in sync with i386 version by applying change: |
| 6 | * - Provisions for empty E820 memory regions (reported by certain BIOSes). |
| 7 | * Alex Achenbach <xela@slit.de>, December 2002. |
| 8 | * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> |
| 9 | * |
| 10 | */ |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/types.h> |
| 13 | #include <linux/init.h> |
| 14 | #include <linux/crash_dump.h> |
| 15 | #include <linux/export.h> |
| 16 | #include <linux/bootmem.h> |
| 17 | #include <linux/pfn.h> |
| 18 | #include <linux/suspend.h> |
| 19 | #include <linux/acpi.h> |
| 20 | #include <linux/firmware-map.h> |
| 21 | #include <linux/memblock.h> |
| 22 | #include <linux/sort.h> |
| 23 | |
| 24 | #include <asm/e820.h> |
| 25 | #include <asm/proto.h> |
| 26 | #include <asm/setup.h> |
| 27 | |
| 28 | /* |
| 29 | * The e820 map is the map that gets modified e.g. with command line parameters |
| 30 | * and that is also registered with modifications in the kernel resource tree |
| 31 | * with the iomem_resource as parent. |
| 32 | * |
| 33 | * The e820_saved is directly saved after the BIOS-provided memory map is |
| 34 | * copied. It doesn't get modified afterwards. It's registered for the |
| 35 | * /sys/firmware/memmap interface. |
| 36 | * |
| 37 | * That memory map is not modified and is used as base for kexec. The kexec'd |
| 38 | * kernel should get the same memory map as the firmware provides. Then the |
| 39 | * user can e.g. boot the original kernel with mem=1G while still booting the |
| 40 | * next kernel with full memory. |
| 41 | */ |
| 42 | struct e820map e820; |
| 43 | struct e820map e820_saved; |
| 44 | |
| 45 | /* For PCI or other memory-mapped resources */ |
| 46 | unsigned long pci_mem_start = 0xaeedbabe; |
| 47 | #ifdef CONFIG_PCI |
| 48 | EXPORT_SYMBOL(pci_mem_start); |
| 49 | #endif |
| 50 | |
| 51 | /* |
| 52 | * This function checks if any part of the range <start,end> is mapped |
| 53 | * with type. |
| 54 | */ |
| 55 | int |
| 56 | e820_any_mapped(u64 start, u64 end, unsigned type) |
| 57 | { |
| 58 | int i; |
| 59 | |
| 60 | for (i = 0; i < e820.nr_map; i++) { |
| 61 | struct e820entry *ei = &e820.map[i]; |
| 62 | |
| 63 | if (type && ei->type != type) |
| 64 | continue; |
| 65 | if (ei->addr >= end || ei->addr + ei->size <= start) |
| 66 | continue; |
| 67 | return 1; |
| 68 | } |
| 69 | return 0; |
| 70 | } |
| 71 | EXPORT_SYMBOL_GPL(e820_any_mapped); |
| 72 | |
| 73 | /* |
| 74 | * This function checks if the entire range <start,end> is mapped with type. |
| 75 | * |
| 76 | * Note: this function only works correct if the e820 table is sorted and |
| 77 | * not-overlapping, which is the case |
| 78 | */ |
| 79 | int __init e820_all_mapped(u64 start, u64 end, unsigned type) |
| 80 | { |
| 81 | int i; |
| 82 | |
| 83 | for (i = 0; i < e820.nr_map; i++) { |
| 84 | struct e820entry *ei = &e820.map[i]; |
| 85 | |
| 86 | if (type && ei->type != type) |
| 87 | continue; |
| 88 | /* is the region (part) in overlap with the current region ?*/ |
| 89 | if (ei->addr >= end || ei->addr + ei->size <= start) |
| 90 | continue; |
| 91 | |
| 92 | /* if the region is at the beginning of <start,end> we move |
| 93 | * start to the end of the region since it's ok until there |
| 94 | */ |
| 95 | if (ei->addr <= start) |
| 96 | start = ei->addr + ei->size; |
| 97 | /* |
| 98 | * if start is now at or beyond end, we're done, full |
| 99 | * coverage |
| 100 | */ |
| 101 | if (start >= end) |
| 102 | return 1; |
| 103 | } |
| 104 | return 0; |
| 105 | } |
| 106 | |
| 107 | /* |
| 108 | * Add a memory region to the kernel e820 map. |
| 109 | */ |
| 110 | static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size, |
| 111 | int type) |
| 112 | { |
| 113 | int x = e820x->nr_map; |
| 114 | |
| 115 | if (x >= ARRAY_SIZE(e820x->map)) { |
| 116 | printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n", |
| 117 | (unsigned long long) start, |
| 118 | (unsigned long long) (start + size - 1)); |
| 119 | return; |
| 120 | } |
| 121 | |
| 122 | e820x->map[x].addr = start; |
| 123 | e820x->map[x].size = size; |
| 124 | e820x->map[x].type = type; |
| 125 | e820x->nr_map++; |
| 126 | } |
| 127 | |
| 128 | void __init e820_add_region(u64 start, u64 size, int type) |
| 129 | { |
| 130 | __e820_add_region(&e820, start, size, type); |
| 131 | } |
| 132 | |
| 133 | static void __init e820_print_type(u32 type) |
| 134 | { |
| 135 | switch (type) { |
| 136 | case E820_RAM: |
| 137 | case E820_RESERVED_KERN: |
| 138 | printk(KERN_CONT "usable"); |
| 139 | break; |
| 140 | case E820_RESERVED: |
| 141 | printk(KERN_CONT "reserved"); |
| 142 | break; |
| 143 | case E820_ACPI: |
| 144 | printk(KERN_CONT "ACPI data"); |
| 145 | break; |
| 146 | case E820_NVS: |
| 147 | printk(KERN_CONT "ACPI NVS"); |
| 148 | break; |
| 149 | case E820_UNUSABLE: |
| 150 | printk(KERN_CONT "unusable"); |
| 151 | break; |
| 152 | case E820_PRAM: |
| 153 | printk(KERN_CONT "persistent (type %u)", type); |
| 154 | break; |
| 155 | default: |
| 156 | printk(KERN_CONT "type %u", type); |
| 157 | break; |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | void __init e820_print_map(char *who) |
| 162 | { |
| 163 | int i; |
| 164 | |
| 165 | for (i = 0; i < e820.nr_map; i++) { |
| 166 | printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who, |
| 167 | (unsigned long long) e820.map[i].addr, |
| 168 | (unsigned long long) |
| 169 | (e820.map[i].addr + e820.map[i].size - 1)); |
| 170 | e820_print_type(e820.map[i].type); |
| 171 | printk(KERN_CONT "\n"); |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | /* |
| 176 | * Sanitize the BIOS e820 map. |
| 177 | * |
| 178 | * Some e820 responses include overlapping entries. The following |
| 179 | * replaces the original e820 map with a new one, removing overlaps, |
| 180 | * and resolving conflicting memory types in favor of highest |
| 181 | * numbered type. |
| 182 | * |
| 183 | * The input parameter biosmap points to an array of 'struct |
| 184 | * e820entry' which on entry has elements in the range [0, *pnr_map) |
| 185 | * valid, and which has space for up to max_nr_map entries. |
| 186 | * On return, the resulting sanitized e820 map entries will be in |
| 187 | * overwritten in the same location, starting at biosmap. |
| 188 | * |
| 189 | * The integer pointed to by pnr_map must be valid on entry (the |
| 190 | * current number of valid entries located at biosmap). If the |
| 191 | * sanitizing succeeds the *pnr_map will be updated with the new |
| 192 | * number of valid entries (something no more than max_nr_map). |
| 193 | * |
| 194 | * The return value from sanitize_e820_map() is zero if it |
| 195 | * successfully 'sanitized' the map entries passed in, and is -1 |
| 196 | * if it did nothing, which can happen if either of (1) it was |
| 197 | * only passed one map entry, or (2) any of the input map entries |
| 198 | * were invalid (start + size < start, meaning that the size was |
| 199 | * so big the described memory range wrapped around through zero.) |
| 200 | * |
| 201 | * Visually we're performing the following |
| 202 | * (1,2,3,4 = memory types)... |
| 203 | * |
| 204 | * Sample memory map (w/overlaps): |
| 205 | * ____22__________________ |
| 206 | * ______________________4_ |
| 207 | * ____1111________________ |
| 208 | * _44_____________________ |
| 209 | * 11111111________________ |
| 210 | * ____________________33__ |
| 211 | * ___________44___________ |
| 212 | * __________33333_________ |
| 213 | * ______________22________ |
| 214 | * ___________________2222_ |
| 215 | * _________111111111______ |
| 216 | * _____________________11_ |
| 217 | * _________________4______ |
| 218 | * |
| 219 | * Sanitized equivalent (no overlap): |
| 220 | * 1_______________________ |
| 221 | * _44_____________________ |
| 222 | * ___1____________________ |
| 223 | * ____22__________________ |
| 224 | * ______11________________ |
| 225 | * _________1______________ |
| 226 | * __________3_____________ |
| 227 | * ___________44___________ |
| 228 | * _____________33_________ |
| 229 | * _______________2________ |
| 230 | * ________________1_______ |
| 231 | * _________________4______ |
| 232 | * ___________________2____ |
| 233 | * ____________________33__ |
| 234 | * ______________________4_ |
| 235 | */ |
| 236 | struct change_member { |
| 237 | struct e820entry *pbios; /* pointer to original bios entry */ |
| 238 | unsigned long long addr; /* address for this change point */ |
| 239 | }; |
| 240 | |
| 241 | static int __init cpcompare(const void *a, const void *b) |
| 242 | { |
| 243 | struct change_member * const *app = a, * const *bpp = b; |
| 244 | const struct change_member *ap = *app, *bp = *bpp; |
| 245 | |
| 246 | /* |
| 247 | * Inputs are pointers to two elements of change_point[]. If their |
| 248 | * addresses are unequal, their difference dominates. If the addresses |
| 249 | * are equal, then consider one that represents the end of its region |
| 250 | * to be greater than one that does not. |
| 251 | */ |
| 252 | if (ap->addr != bp->addr) |
| 253 | return ap->addr > bp->addr ? 1 : -1; |
| 254 | |
| 255 | return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr); |
| 256 | } |
| 257 | |
| 258 | int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map, |
| 259 | u32 *pnr_map) |
| 260 | { |
| 261 | static struct change_member change_point_list[2*E820_X_MAX] __initdata; |
| 262 | static struct change_member *change_point[2*E820_X_MAX] __initdata; |
| 263 | static struct e820entry *overlap_list[E820_X_MAX] __initdata; |
| 264 | static struct e820entry new_bios[E820_X_MAX] __initdata; |
| 265 | unsigned long current_type, last_type; |
| 266 | unsigned long long last_addr; |
| 267 | int chgidx; |
| 268 | int overlap_entries; |
| 269 | int new_bios_entry; |
| 270 | int old_nr, new_nr, chg_nr; |
| 271 | int i; |
| 272 | |
| 273 | /* if there's only one memory region, don't bother */ |
| 274 | if (*pnr_map < 2) |
| 275 | return -1; |
| 276 | |
| 277 | old_nr = *pnr_map; |
| 278 | BUG_ON(old_nr > max_nr_map); |
| 279 | |
| 280 | /* bail out if we find any unreasonable addresses in bios map */ |
| 281 | for (i = 0; i < old_nr; i++) |
| 282 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) |
| 283 | return -1; |
| 284 | |
| 285 | /* create pointers for initial change-point information (for sorting) */ |
| 286 | for (i = 0; i < 2 * old_nr; i++) |
| 287 | change_point[i] = &change_point_list[i]; |
| 288 | |
| 289 | /* record all known change-points (starting and ending addresses), |
| 290 | omitting those that are for empty memory regions */ |
| 291 | chgidx = 0; |
| 292 | for (i = 0; i < old_nr; i++) { |
| 293 | if (biosmap[i].size != 0) { |
| 294 | change_point[chgidx]->addr = biosmap[i].addr; |
| 295 | change_point[chgidx++]->pbios = &biosmap[i]; |
| 296 | change_point[chgidx]->addr = biosmap[i].addr + |
| 297 | biosmap[i].size; |
| 298 | change_point[chgidx++]->pbios = &biosmap[i]; |
| 299 | } |
| 300 | } |
| 301 | chg_nr = chgidx; |
| 302 | |
| 303 | /* sort change-point list by memory addresses (low -> high) */ |
| 304 | sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL); |
| 305 | |
| 306 | /* create a new bios memory map, removing overlaps */ |
| 307 | overlap_entries = 0; /* number of entries in the overlap table */ |
| 308 | new_bios_entry = 0; /* index for creating new bios map entries */ |
| 309 | last_type = 0; /* start with undefined memory type */ |
| 310 | last_addr = 0; /* start with 0 as last starting address */ |
| 311 | |
| 312 | /* loop through change-points, determining affect on the new bios map */ |
| 313 | for (chgidx = 0; chgidx < chg_nr; chgidx++) { |
| 314 | /* keep track of all overlapping bios entries */ |
| 315 | if (change_point[chgidx]->addr == |
| 316 | change_point[chgidx]->pbios->addr) { |
| 317 | /* |
| 318 | * add map entry to overlap list (> 1 entry |
| 319 | * implies an overlap) |
| 320 | */ |
| 321 | overlap_list[overlap_entries++] = |
| 322 | change_point[chgidx]->pbios; |
| 323 | } else { |
| 324 | /* |
| 325 | * remove entry from list (order independent, |
| 326 | * so swap with last) |
| 327 | */ |
| 328 | for (i = 0; i < overlap_entries; i++) { |
| 329 | if (overlap_list[i] == |
| 330 | change_point[chgidx]->pbios) |
| 331 | overlap_list[i] = |
| 332 | overlap_list[overlap_entries-1]; |
| 333 | } |
| 334 | overlap_entries--; |
| 335 | } |
| 336 | /* |
| 337 | * if there are overlapping entries, decide which |
| 338 | * "type" to use (larger value takes precedence -- |
| 339 | * 1=usable, 2,3,4,4+=unusable) |
| 340 | */ |
| 341 | current_type = 0; |
| 342 | for (i = 0; i < overlap_entries; i++) |
| 343 | if (overlap_list[i]->type > current_type) |
| 344 | current_type = overlap_list[i]->type; |
| 345 | /* |
| 346 | * continue building up new bios map based on this |
| 347 | * information |
| 348 | */ |
| 349 | if (current_type != last_type || current_type == E820_PRAM) { |
| 350 | if (last_type != 0) { |
| 351 | new_bios[new_bios_entry].size = |
| 352 | change_point[chgidx]->addr - last_addr; |
| 353 | /* |
| 354 | * move forward only if the new size |
| 355 | * was non-zero |
| 356 | */ |
| 357 | if (new_bios[new_bios_entry].size != 0) |
| 358 | /* |
| 359 | * no more space left for new |
| 360 | * bios entries ? |
| 361 | */ |
| 362 | if (++new_bios_entry >= max_nr_map) |
| 363 | break; |
| 364 | } |
| 365 | if (current_type != 0) { |
| 366 | new_bios[new_bios_entry].addr = |
| 367 | change_point[chgidx]->addr; |
| 368 | new_bios[new_bios_entry].type = current_type; |
| 369 | last_addr = change_point[chgidx]->addr; |
| 370 | } |
| 371 | last_type = current_type; |
| 372 | } |
| 373 | } |
| 374 | /* retain count for new bios entries */ |
| 375 | new_nr = new_bios_entry; |
| 376 | |
| 377 | /* copy new bios mapping into original location */ |
| 378 | memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry)); |
| 379 | *pnr_map = new_nr; |
| 380 | |
| 381 | return 0; |
| 382 | } |
| 383 | |
| 384 | static int __init __append_e820_map(struct e820entry *biosmap, int nr_map) |
| 385 | { |
| 386 | while (nr_map) { |
| 387 | u64 start = biosmap->addr; |
| 388 | u64 size = biosmap->size; |
| 389 | u64 end = start + size; |
| 390 | u32 type = biosmap->type; |
| 391 | |
| 392 | /* Overflow in 64 bits? Ignore the memory map. */ |
| 393 | if (start > end) |
| 394 | return -1; |
| 395 | |
| 396 | e820_add_region(start, size, type); |
| 397 | |
| 398 | biosmap++; |
| 399 | nr_map--; |
| 400 | } |
| 401 | return 0; |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * Copy the BIOS e820 map into a safe place. |
| 406 | * |
| 407 | * Sanity-check it while we're at it.. |
| 408 | * |
| 409 | * If we're lucky and live on a modern system, the setup code |
| 410 | * will have given us a memory map that we can use to properly |
| 411 | * set up memory. If we aren't, we'll fake a memory map. |
| 412 | */ |
| 413 | static int __init append_e820_map(struct e820entry *biosmap, int nr_map) |
| 414 | { |
| 415 | /* Only one memory region (or negative)? Ignore it */ |
| 416 | if (nr_map < 2) |
| 417 | return -1; |
| 418 | |
| 419 | return __append_e820_map(biosmap, nr_map); |
| 420 | } |
| 421 | |
| 422 | static u64 __init __e820_update_range(struct e820map *e820x, u64 start, |
| 423 | u64 size, unsigned old_type, |
| 424 | unsigned new_type) |
| 425 | { |
| 426 | u64 end; |
| 427 | unsigned int i; |
| 428 | u64 real_updated_size = 0; |
| 429 | |
| 430 | BUG_ON(old_type == new_type); |
| 431 | |
| 432 | if (size > (ULLONG_MAX - start)) |
| 433 | size = ULLONG_MAX - start; |
| 434 | |
| 435 | end = start + size; |
| 436 | printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", |
| 437 | (unsigned long long) start, (unsigned long long) (end - 1)); |
| 438 | e820_print_type(old_type); |
| 439 | printk(KERN_CONT " ==> "); |
| 440 | e820_print_type(new_type); |
| 441 | printk(KERN_CONT "\n"); |
| 442 | |
| 443 | for (i = 0; i < e820x->nr_map; i++) { |
| 444 | struct e820entry *ei = &e820x->map[i]; |
| 445 | u64 final_start, final_end; |
| 446 | u64 ei_end; |
| 447 | |
| 448 | if (ei->type != old_type) |
| 449 | continue; |
| 450 | |
| 451 | ei_end = ei->addr + ei->size; |
| 452 | /* totally covered by new range? */ |
| 453 | if (ei->addr >= start && ei_end <= end) { |
| 454 | ei->type = new_type; |
| 455 | real_updated_size += ei->size; |
| 456 | continue; |
| 457 | } |
| 458 | |
| 459 | /* new range is totally covered? */ |
| 460 | if (ei->addr < start && ei_end > end) { |
| 461 | __e820_add_region(e820x, start, size, new_type); |
| 462 | __e820_add_region(e820x, end, ei_end - end, ei->type); |
| 463 | ei->size = start - ei->addr; |
| 464 | real_updated_size += size; |
| 465 | continue; |
| 466 | } |
| 467 | |
| 468 | /* partially covered */ |
| 469 | final_start = max(start, ei->addr); |
| 470 | final_end = min(end, ei_end); |
| 471 | if (final_start >= final_end) |
| 472 | continue; |
| 473 | |
| 474 | __e820_add_region(e820x, final_start, final_end - final_start, |
| 475 | new_type); |
| 476 | |
| 477 | real_updated_size += final_end - final_start; |
| 478 | |
| 479 | /* |
| 480 | * left range could be head or tail, so need to update |
| 481 | * size at first. |
| 482 | */ |
| 483 | ei->size -= final_end - final_start; |
| 484 | if (ei->addr < final_start) |
| 485 | continue; |
| 486 | ei->addr = final_end; |
| 487 | } |
| 488 | return real_updated_size; |
| 489 | } |
| 490 | |
| 491 | u64 __init e820_update_range(u64 start, u64 size, unsigned old_type, |
| 492 | unsigned new_type) |
| 493 | { |
| 494 | return __e820_update_range(&e820, start, size, old_type, new_type); |
| 495 | } |
| 496 | |
| 497 | static u64 __init e820_update_range_saved(u64 start, u64 size, |
| 498 | unsigned old_type, unsigned new_type) |
| 499 | { |
| 500 | return __e820_update_range(&e820_saved, start, size, old_type, |
| 501 | new_type); |
| 502 | } |
| 503 | |
| 504 | /* make e820 not cover the range */ |
| 505 | u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type, |
| 506 | int checktype) |
| 507 | { |
| 508 | int i; |
| 509 | u64 end; |
| 510 | u64 real_removed_size = 0; |
| 511 | |
| 512 | if (size > (ULLONG_MAX - start)) |
| 513 | size = ULLONG_MAX - start; |
| 514 | |
| 515 | end = start + size; |
| 516 | printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", |
| 517 | (unsigned long long) start, (unsigned long long) (end - 1)); |
| 518 | if (checktype) |
| 519 | e820_print_type(old_type); |
| 520 | printk(KERN_CONT "\n"); |
| 521 | |
| 522 | for (i = 0; i < e820.nr_map; i++) { |
| 523 | struct e820entry *ei = &e820.map[i]; |
| 524 | u64 final_start, final_end; |
| 525 | u64 ei_end; |
| 526 | |
| 527 | if (checktype && ei->type != old_type) |
| 528 | continue; |
| 529 | |
| 530 | ei_end = ei->addr + ei->size; |
| 531 | /* totally covered? */ |
| 532 | if (ei->addr >= start && ei_end <= end) { |
| 533 | real_removed_size += ei->size; |
| 534 | memset(ei, 0, sizeof(struct e820entry)); |
| 535 | continue; |
| 536 | } |
| 537 | |
| 538 | /* new range is totally covered? */ |
| 539 | if (ei->addr < start && ei_end > end) { |
| 540 | e820_add_region(end, ei_end - end, ei->type); |
| 541 | ei->size = start - ei->addr; |
| 542 | real_removed_size += size; |
| 543 | continue; |
| 544 | } |
| 545 | |
| 546 | /* partially covered */ |
| 547 | final_start = max(start, ei->addr); |
| 548 | final_end = min(end, ei_end); |
| 549 | if (final_start >= final_end) |
| 550 | continue; |
| 551 | real_removed_size += final_end - final_start; |
| 552 | |
| 553 | /* |
| 554 | * left range could be head or tail, so need to update |
| 555 | * size at first. |
| 556 | */ |
| 557 | ei->size -= final_end - final_start; |
| 558 | if (ei->addr < final_start) |
| 559 | continue; |
| 560 | ei->addr = final_end; |
| 561 | } |
| 562 | return real_removed_size; |
| 563 | } |
| 564 | |
| 565 | void __init update_e820(void) |
| 566 | { |
| 567 | if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map)) |
| 568 | return; |
| 569 | printk(KERN_INFO "e820: modified physical RAM map:\n"); |
| 570 | e820_print_map("modified"); |
| 571 | } |
| 572 | static void __init update_e820_saved(void) |
| 573 | { |
| 574 | sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), |
| 575 | &e820_saved.nr_map); |
| 576 | } |
| 577 | #define MAX_GAP_END 0x100000000ull |
| 578 | /* |
| 579 | * Search for a gap in the e820 memory space from start_addr to end_addr. |
| 580 | */ |
| 581 | __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize, |
| 582 | unsigned long start_addr, unsigned long long end_addr) |
| 583 | { |
| 584 | unsigned long long last; |
| 585 | int i = e820.nr_map; |
| 586 | int found = 0; |
| 587 | |
| 588 | last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END; |
| 589 | |
| 590 | while (--i >= 0) { |
| 591 | unsigned long long start = e820.map[i].addr; |
| 592 | unsigned long long end = start + e820.map[i].size; |
| 593 | |
| 594 | if (end < start_addr) |
| 595 | continue; |
| 596 | |
| 597 | /* |
| 598 | * Since "last" is at most 4GB, we know we'll |
| 599 | * fit in 32 bits if this condition is true |
| 600 | */ |
| 601 | if (last > end) { |
| 602 | unsigned long gap = last - end; |
| 603 | |
| 604 | if (gap >= *gapsize) { |
| 605 | *gapsize = gap; |
| 606 | *gapstart = end; |
| 607 | found = 1; |
| 608 | } |
| 609 | } |
| 610 | if (start < last) |
| 611 | last = start; |
| 612 | } |
| 613 | return found; |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * Search for the biggest gap in the low 32 bits of the e820 |
| 618 | * memory space. We pass this space to PCI to assign MMIO resources |
| 619 | * for hotplug or unconfigured devices in. |
| 620 | * Hopefully the BIOS let enough space left. |
| 621 | */ |
| 622 | __init void e820_setup_gap(void) |
| 623 | { |
| 624 | unsigned long gapstart, gapsize; |
| 625 | int found; |
| 626 | |
| 627 | gapstart = 0x10000000; |
| 628 | gapsize = 0x400000; |
| 629 | found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END); |
| 630 | |
| 631 | #ifdef CONFIG_X86_64 |
| 632 | if (!found) { |
| 633 | gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024; |
| 634 | printk(KERN_ERR |
| 635 | "e820: cannot find a gap in the 32bit address range\n" |
| 636 | "e820: PCI devices with unassigned 32bit BARs may break!\n"); |
| 637 | } |
| 638 | #endif |
| 639 | |
| 640 | /* |
| 641 | * e820_reserve_resources_late protect stolen RAM already |
| 642 | */ |
| 643 | pci_mem_start = gapstart; |
| 644 | |
| 645 | printk(KERN_INFO |
| 646 | "e820: [mem %#010lx-%#010lx] available for PCI devices\n", |
| 647 | gapstart, gapstart + gapsize - 1); |
| 648 | } |
| 649 | |
| 650 | /** |
| 651 | * Because of the size limitation of struct boot_params, only first |
| 652 | * 128 E820 memory entries are passed to kernel via |
| 653 | * boot_params.e820_map, others are passed via SETUP_E820_EXT node of |
| 654 | * linked list of struct setup_data, which is parsed here. |
| 655 | */ |
| 656 | void __init parse_e820_ext(u64 phys_addr, u32 data_len) |
| 657 | { |
| 658 | int entries; |
| 659 | struct e820entry *extmap; |
| 660 | struct setup_data *sdata; |
| 661 | |
| 662 | sdata = early_memremap(phys_addr, data_len); |
| 663 | entries = sdata->len / sizeof(struct e820entry); |
| 664 | extmap = (struct e820entry *)(sdata->data); |
| 665 | __append_e820_map(extmap, entries); |
| 666 | sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); |
| 667 | early_iounmap(sdata, data_len); |
| 668 | printk(KERN_INFO "e820: extended physical RAM map:\n"); |
| 669 | e820_print_map("extended"); |
| 670 | } |
| 671 | |
| 672 | #if defined(CONFIG_X86_64) || \ |
| 673 | (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION)) |
| 674 | /** |
| 675 | * Find the ranges of physical addresses that do not correspond to |
| 676 | * e820 RAM areas and mark the corresponding pages as nosave for |
| 677 | * hibernation (32 bit) or software suspend and suspend to RAM (64 bit). |
| 678 | * |
| 679 | * This function requires the e820 map to be sorted and without any |
| 680 | * overlapping entries. |
| 681 | */ |
| 682 | void __init e820_mark_nosave_regions(unsigned long limit_pfn) |
| 683 | { |
| 684 | int i; |
| 685 | unsigned long pfn = 0; |
| 686 | |
| 687 | for (i = 0; i < e820.nr_map; i++) { |
| 688 | struct e820entry *ei = &e820.map[i]; |
| 689 | |
| 690 | if (pfn < PFN_UP(ei->addr)) |
| 691 | register_nosave_region(pfn, PFN_UP(ei->addr)); |
| 692 | |
| 693 | pfn = PFN_DOWN(ei->addr + ei->size); |
| 694 | |
| 695 | if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN) |
| 696 | register_nosave_region(PFN_UP(ei->addr), pfn); |
| 697 | |
| 698 | if (pfn >= limit_pfn) |
| 699 | break; |
| 700 | } |
| 701 | } |
| 702 | #endif |
| 703 | |
| 704 | #ifdef CONFIG_ACPI |
| 705 | /** |
| 706 | * Mark ACPI NVS memory region, so that we can save/restore it during |
| 707 | * hibernation and the subsequent resume. |
| 708 | */ |
| 709 | static int __init e820_mark_nvs_memory(void) |
| 710 | { |
| 711 | int i; |
| 712 | |
| 713 | for (i = 0; i < e820.nr_map; i++) { |
| 714 | struct e820entry *ei = &e820.map[i]; |
| 715 | |
| 716 | if (ei->type == E820_NVS) |
| 717 | acpi_nvs_register(ei->addr, ei->size); |
| 718 | } |
| 719 | |
| 720 | return 0; |
| 721 | } |
| 722 | core_initcall(e820_mark_nvs_memory); |
| 723 | #endif |
| 724 | |
| 725 | /* |
| 726 | * pre allocated 4k and reserved it in memblock and e820_saved |
| 727 | */ |
| 728 | u64 __init early_reserve_e820(u64 size, u64 align) |
| 729 | { |
| 730 | u64 addr; |
| 731 | |
| 732 | addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
| 733 | if (addr) { |
| 734 | e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED); |
| 735 | printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n"); |
| 736 | update_e820_saved(); |
| 737 | } |
| 738 | |
| 739 | return addr; |
| 740 | } |
| 741 | |
| 742 | #ifdef CONFIG_X86_32 |
| 743 | # ifdef CONFIG_X86_PAE |
| 744 | # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT)) |
| 745 | # else |
| 746 | # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT)) |
| 747 | # endif |
| 748 | #else /* CONFIG_X86_32 */ |
| 749 | # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT |
| 750 | #endif |
| 751 | |
| 752 | /* |
| 753 | * Find the highest page frame number we have available |
| 754 | */ |
| 755 | static unsigned long __init e820_end_pfn(unsigned long limit_pfn) |
| 756 | { |
| 757 | int i; |
| 758 | unsigned long last_pfn = 0; |
| 759 | unsigned long max_arch_pfn = MAX_ARCH_PFN; |
| 760 | |
| 761 | for (i = 0; i < e820.nr_map; i++) { |
| 762 | struct e820entry *ei = &e820.map[i]; |
| 763 | unsigned long start_pfn; |
| 764 | unsigned long end_pfn; |
| 765 | |
| 766 | /* |
| 767 | * Persistent memory is accounted as ram for purposes of |
| 768 | * establishing max_pfn and mem_map. |
| 769 | */ |
| 770 | if (ei->type != E820_RAM && ei->type != E820_PRAM) |
| 771 | continue; |
| 772 | |
| 773 | start_pfn = ei->addr >> PAGE_SHIFT; |
| 774 | end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT; |
| 775 | |
| 776 | if (start_pfn >= limit_pfn) |
| 777 | continue; |
| 778 | if (end_pfn > limit_pfn) { |
| 779 | last_pfn = limit_pfn; |
| 780 | break; |
| 781 | } |
| 782 | if (end_pfn > last_pfn) |
| 783 | last_pfn = end_pfn; |
| 784 | } |
| 785 | |
| 786 | if (last_pfn > max_arch_pfn) |
| 787 | last_pfn = max_arch_pfn; |
| 788 | |
| 789 | printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n", |
| 790 | last_pfn, max_arch_pfn); |
| 791 | return last_pfn; |
| 792 | } |
| 793 | unsigned long __init e820_end_of_ram_pfn(void) |
| 794 | { |
| 795 | return e820_end_pfn(MAX_ARCH_PFN); |
| 796 | } |
| 797 | |
| 798 | unsigned long __init e820_end_of_low_ram_pfn(void) |
| 799 | { |
| 800 | return e820_end_pfn(1UL << (32-PAGE_SHIFT)); |
| 801 | } |
| 802 | |
| 803 | static void early_panic(char *msg) |
| 804 | { |
| 805 | early_printk(msg); |
| 806 | panic(msg); |
| 807 | } |
| 808 | |
| 809 | static int userdef __initdata; |
| 810 | |
| 811 | /* "mem=nopentium" disables the 4MB page tables. */ |
| 812 | static int __init parse_memopt(char *p) |
| 813 | { |
| 814 | u64 mem_size; |
| 815 | |
| 816 | if (!p) |
| 817 | return -EINVAL; |
| 818 | |
| 819 | if (!strcmp(p, "nopentium")) { |
| 820 | #ifdef CONFIG_X86_32 |
| 821 | setup_clear_cpu_cap(X86_FEATURE_PSE); |
| 822 | return 0; |
| 823 | #else |
| 824 | printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n"); |
| 825 | return -EINVAL; |
| 826 | #endif |
| 827 | } |
| 828 | |
| 829 | userdef = 1; |
| 830 | mem_size = memparse(p, &p); |
| 831 | /* don't remove all of memory when handling "mem={invalid}" param */ |
| 832 | if (mem_size == 0) |
| 833 | return -EINVAL; |
| 834 | e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1); |
| 835 | |
| 836 | return 0; |
| 837 | } |
| 838 | early_param("mem", parse_memopt); |
| 839 | |
| 840 | static int __init parse_memmap_one(char *p) |
| 841 | { |
| 842 | char *oldp; |
| 843 | u64 start_at, mem_size; |
| 844 | |
| 845 | if (!p) |
| 846 | return -EINVAL; |
| 847 | |
| 848 | if (!strncmp(p, "exactmap", 8)) { |
| 849 | #ifdef CONFIG_CRASH_DUMP |
| 850 | /* |
| 851 | * If we are doing a crash dump, we still need to know |
| 852 | * the real mem size before original memory map is |
| 853 | * reset. |
| 854 | */ |
| 855 | saved_max_pfn = e820_end_of_ram_pfn(); |
| 856 | #endif |
| 857 | e820.nr_map = 0; |
| 858 | userdef = 1; |
| 859 | return 0; |
| 860 | } |
| 861 | |
| 862 | oldp = p; |
| 863 | mem_size = memparse(p, &p); |
| 864 | if (p == oldp) |
| 865 | return -EINVAL; |
| 866 | |
| 867 | userdef = 1; |
| 868 | if (*p == '@') { |
| 869 | start_at = memparse(p+1, &p); |
| 870 | e820_add_region(start_at, mem_size, E820_RAM); |
| 871 | } else if (*p == '#') { |
| 872 | start_at = memparse(p+1, &p); |
| 873 | e820_add_region(start_at, mem_size, E820_ACPI); |
| 874 | } else if (*p == '$') { |
| 875 | start_at = memparse(p+1, &p); |
| 876 | e820_add_region(start_at, mem_size, E820_RESERVED); |
| 877 | } else if (*p == '!') { |
| 878 | start_at = memparse(p+1, &p); |
| 879 | e820_add_region(start_at, mem_size, E820_PRAM); |
| 880 | } else |
| 881 | e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1); |
| 882 | |
| 883 | return *p == '\0' ? 0 : -EINVAL; |
| 884 | } |
| 885 | static int __init parse_memmap_opt(char *str) |
| 886 | { |
| 887 | while (str) { |
| 888 | char *k = strchr(str, ','); |
| 889 | |
| 890 | if (k) |
| 891 | *k++ = 0; |
| 892 | |
| 893 | parse_memmap_one(str); |
| 894 | str = k; |
| 895 | } |
| 896 | |
| 897 | return 0; |
| 898 | } |
| 899 | early_param("memmap", parse_memmap_opt); |
| 900 | |
| 901 | void __init finish_e820_parsing(void) |
| 902 | { |
| 903 | if (userdef) { |
| 904 | if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), |
| 905 | &e820.nr_map) < 0) |
| 906 | early_panic("Invalid user supplied memory map"); |
| 907 | |
| 908 | printk(KERN_INFO "e820: user-defined physical RAM map:\n"); |
| 909 | e820_print_map("user"); |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | static inline const char *e820_type_to_string(int e820_type) |
| 914 | { |
| 915 | switch (e820_type) { |
| 916 | case E820_RESERVED_KERN: |
| 917 | case E820_RAM: return "System RAM"; |
| 918 | case E820_ACPI: return "ACPI Tables"; |
| 919 | case E820_NVS: return "ACPI Non-volatile Storage"; |
| 920 | case E820_UNUSABLE: return "Unusable memory"; |
| 921 | case E820_PRAM: return "Persistent RAM"; |
| 922 | default: return "reserved"; |
| 923 | } |
| 924 | } |
| 925 | |
| 926 | /* |
| 927 | * Mark e820 reserved areas as busy for the resource manager. |
| 928 | */ |
| 929 | static struct resource __initdata *e820_res; |
| 930 | void __init e820_reserve_resources(void) |
| 931 | { |
| 932 | int i; |
| 933 | struct resource *res; |
| 934 | u64 end; |
| 935 | |
| 936 | res = alloc_bootmem(sizeof(struct resource) * e820.nr_map); |
| 937 | e820_res = res; |
| 938 | for (i = 0; i < e820.nr_map; i++) { |
| 939 | end = e820.map[i].addr + e820.map[i].size - 1; |
| 940 | if (end != (resource_size_t)end) { |
| 941 | res++; |
| 942 | continue; |
| 943 | } |
| 944 | res->name = e820_type_to_string(e820.map[i].type); |
| 945 | res->start = e820.map[i].addr; |
| 946 | res->end = end; |
| 947 | |
| 948 | res->flags = IORESOURCE_MEM; |
| 949 | |
| 950 | /* |
| 951 | * don't register the region that could be conflicted with |
| 952 | * pci device BAR resource and insert them later in |
| 953 | * pcibios_resource_survey() |
| 954 | */ |
| 955 | if (((e820.map[i].type != E820_RESERVED) && |
| 956 | (e820.map[i].type != E820_PRAM)) || |
| 957 | res->start < (1ULL<<20)) { |
| 958 | res->flags |= IORESOURCE_BUSY; |
| 959 | insert_resource(&iomem_resource, res); |
| 960 | } |
| 961 | res++; |
| 962 | } |
| 963 | |
| 964 | for (i = 0; i < e820_saved.nr_map; i++) { |
| 965 | struct e820entry *entry = &e820_saved.map[i]; |
| 966 | firmware_map_add_early(entry->addr, |
| 967 | entry->addr + entry->size, |
| 968 | e820_type_to_string(entry->type)); |
| 969 | } |
| 970 | } |
| 971 | |
| 972 | /* How much should we pad RAM ending depending on where it is? */ |
| 973 | static unsigned long ram_alignment(resource_size_t pos) |
| 974 | { |
| 975 | unsigned long mb = pos >> 20; |
| 976 | |
| 977 | /* To 64kB in the first megabyte */ |
| 978 | if (!mb) |
| 979 | return 64*1024; |
| 980 | |
| 981 | /* To 1MB in the first 16MB */ |
| 982 | if (mb < 16) |
| 983 | return 1024*1024; |
| 984 | |
| 985 | /* To 64MB for anything above that */ |
| 986 | return 64*1024*1024; |
| 987 | } |
| 988 | |
| 989 | #define MAX_RESOURCE_SIZE ((resource_size_t)-1) |
| 990 | |
| 991 | void __init e820_reserve_resources_late(void) |
| 992 | { |
| 993 | int i; |
| 994 | struct resource *res; |
| 995 | |
| 996 | res = e820_res; |
| 997 | for (i = 0; i < e820.nr_map; i++) { |
| 998 | if (!res->parent && res->end) |
| 999 | insert_resource_expand_to_fit(&iomem_resource, res); |
| 1000 | res++; |
| 1001 | } |
| 1002 | |
| 1003 | /* |
| 1004 | * Try to bump up RAM regions to reasonable boundaries to |
| 1005 | * avoid stolen RAM: |
| 1006 | */ |
| 1007 | for (i = 0; i < e820.nr_map; i++) { |
| 1008 | struct e820entry *entry = &e820.map[i]; |
| 1009 | u64 start, end; |
| 1010 | |
| 1011 | if (entry->type != E820_RAM) |
| 1012 | continue; |
| 1013 | start = entry->addr + entry->size; |
| 1014 | end = round_up(start, ram_alignment(start)) - 1; |
| 1015 | if (end > MAX_RESOURCE_SIZE) |
| 1016 | end = MAX_RESOURCE_SIZE; |
| 1017 | if (start >= end) |
| 1018 | continue; |
| 1019 | printk(KERN_DEBUG |
| 1020 | "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", |
| 1021 | start, end); |
| 1022 | reserve_region_with_split(&iomem_resource, start, end, |
| 1023 | "RAM buffer"); |
| 1024 | } |
| 1025 | } |
| 1026 | |
| 1027 | char *__init default_machine_specific_memory_setup(void) |
| 1028 | { |
| 1029 | char *who = "BIOS-e820"; |
| 1030 | u32 new_nr; |
| 1031 | /* |
| 1032 | * Try to copy the BIOS-supplied E820-map. |
| 1033 | * |
| 1034 | * Otherwise fake a memory map; one section from 0k->640k, |
| 1035 | * the next section from 1mb->appropriate_mem_k |
| 1036 | */ |
| 1037 | new_nr = boot_params.e820_entries; |
| 1038 | sanitize_e820_map(boot_params.e820_map, |
| 1039 | ARRAY_SIZE(boot_params.e820_map), |
| 1040 | &new_nr); |
| 1041 | boot_params.e820_entries = new_nr; |
| 1042 | if (append_e820_map(boot_params.e820_map, boot_params.e820_entries) |
| 1043 | < 0) { |
| 1044 | u64 mem_size; |
| 1045 | |
| 1046 | /* compare results from other methods and take the greater */ |
| 1047 | if (boot_params.alt_mem_k |
| 1048 | < boot_params.screen_info.ext_mem_k) { |
| 1049 | mem_size = boot_params.screen_info.ext_mem_k; |
| 1050 | who = "BIOS-88"; |
| 1051 | } else { |
| 1052 | mem_size = boot_params.alt_mem_k; |
| 1053 | who = "BIOS-e801"; |
| 1054 | } |
| 1055 | |
| 1056 | e820.nr_map = 0; |
| 1057 | e820_add_region(0, LOWMEMSIZE(), E820_RAM); |
| 1058 | e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM); |
| 1059 | } |
| 1060 | |
| 1061 | /* In case someone cares... */ |
| 1062 | return who; |
| 1063 | } |
| 1064 | |
| 1065 | void __init setup_memory_map(void) |
| 1066 | { |
| 1067 | char *who; |
| 1068 | |
| 1069 | who = x86_init.resources.memory_setup(); |
| 1070 | memcpy(&e820_saved, &e820, sizeof(struct e820map)); |
| 1071 | printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n"); |
| 1072 | e820_print_map(who); |
| 1073 | } |
| 1074 | |
| 1075 | void __init memblock_x86_fill(void) |
| 1076 | { |
| 1077 | int i; |
| 1078 | u64 end; |
| 1079 | |
| 1080 | /* |
| 1081 | * EFI may have more than 128 entries |
| 1082 | * We are safe to enable resizing, beause memblock_x86_fill() |
| 1083 | * is rather later for x86 |
| 1084 | */ |
| 1085 | memblock_allow_resize(); |
| 1086 | |
| 1087 | for (i = 0; i < e820.nr_map; i++) { |
| 1088 | struct e820entry *ei = &e820.map[i]; |
| 1089 | |
| 1090 | end = ei->addr + ei->size; |
| 1091 | if (end != (resource_size_t)end) |
| 1092 | continue; |
| 1093 | |
| 1094 | if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN) |
| 1095 | continue; |
| 1096 | |
| 1097 | memblock_add(ei->addr, ei->size); |
| 1098 | } |
| 1099 | |
| 1100 | /* throw away partial pages */ |
| 1101 | memblock_trim_memory(PAGE_SIZE); |
| 1102 | |
| 1103 | memblock_dump_all(); |
| 1104 | } |
| 1105 | |
| 1106 | void __init memblock_find_dma_reserve(void) |
| 1107 | { |
| 1108 | #ifdef CONFIG_X86_64 |
| 1109 | u64 nr_pages = 0, nr_free_pages = 0; |
| 1110 | unsigned long start_pfn, end_pfn; |
| 1111 | phys_addr_t start, end; |
| 1112 | int i; |
| 1113 | u64 u; |
| 1114 | |
| 1115 | /* |
| 1116 | * need to find out used area below MAX_DMA_PFN |
| 1117 | * need to use memblock to get free size in [0, MAX_DMA_PFN] |
| 1118 | * at first, and assume boot_mem will not take below MAX_DMA_PFN |
| 1119 | */ |
| 1120 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
| 1121 | start_pfn = min(start_pfn, MAX_DMA_PFN); |
| 1122 | end_pfn = min(end_pfn, MAX_DMA_PFN); |
| 1123 | nr_pages += end_pfn - start_pfn; |
| 1124 | } |
| 1125 | |
| 1126 | for_each_free_mem_range(u, NUMA_NO_NODE, &start, &end, NULL) { |
| 1127 | start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN); |
| 1128 | end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN); |
| 1129 | if (start_pfn < end_pfn) |
| 1130 | nr_free_pages += end_pfn - start_pfn; |
| 1131 | } |
| 1132 | |
| 1133 | set_dma_reserve(nr_pages - nr_free_pages); |
| 1134 | #endif |
| 1135 | } |