arch/x86/kernel/e820_64.c

   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/bootmem.h>
  15 #include <linux/ioport.h>
  16 #include <linux/string.h>
  17 #include <linux/kexec.h>
  18 #include <linux/module.h>
  19 #include <linux/mm.h>
  20 #include <linux/suspend.h>
  21 #include <linux/pfn.h>
  22
  23 #include <asm/pgtable.h>
  24 #include <asm/page.h>
  25 #include <asm/e820.h>
  26 #include <asm/proto.h>
  27 #include <asm/setup.h>
  28 #include <asm/sections.h>
  29 #include <asm/kdebug.h>
  30
  31 struct e820map e820;
  32
  33 /*
  34  * PFN of last memory page.
  35  */
  36 unsigned long end_pfn;
  37
  38 /*
  39  * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
  40  * The direct mapping extends to end_pfn_map, so that we can directly access
  41  * apertures, ACPI and other tables without having to play with fixmaps.
  42  */
  43 unsigned long end_pfn_map;
  44
  45 /*
  46  * Last pfn which the user wants to use.
  47  */
  48 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
  49
  50 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
  51 static inline int bad_addr(unsigned long *addrp, unsigned long size)
  52 {
  53         unsigned long addr = *addrp, last = addr + size;
  54
  55         /* various gunk below that needed for SMP startup */
  56         if (addr < 0x8000) {
  57                 *addrp = PAGE_ALIGN(0x8000);
  58                 return 1;
  59         }
  60
  61         /* direct mapping tables of the kernel */
  62         if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
  63                 *addrp = PAGE_ALIGN(table_end << PAGE_SHIFT);
  64                 return 1;
  65         }
  66
  67         /* initrd */
  68 #ifdef CONFIG_BLK_DEV_INITRD
  69         if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
  70                 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
  71                 unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
  72                 unsigned long ramdisk_end   = ramdisk_image+ramdisk_size;
  73
  74                 if (last >= ramdisk_image && addr < ramdisk_end) {
  75                         *addrp = PAGE_ALIGN(ramdisk_end);
  76                         return 1;
  77                 }
  78         }
  79 #endif
  80         /* kernel code */
  81         if (last >= __pa_symbol(&_text) && addr < __pa_symbol(&_end)) {
  82                 *addrp = PAGE_ALIGN(__pa_symbol(&_end));
  83                 return 1;
  84         }
  85
  86         if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
  87                 *addrp = PAGE_ALIGN(ebda_addr + ebda_size);
  88                 return 1;
  89         }
  90
  91 #ifdef CONFIG_NUMA
  92         /* NUMA memory to node map */
  93         if (last >= nodemap_addr && addr < nodemap_addr + nodemap_size) {
  94                 *addrp = nodemap_addr + nodemap_size;
  95                 return 1;
  96         }
  97 #endif
  98         /* XXX ramdisk image here? */
  99         return 0;
 100 }
 101
 102 /*
 103  * This function checks if any part of the range <start,end> is mapped
 104  * with type.
 105  */
 106 int
 107 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
 108 {
 109         int i;
 110
 111         for (i = 0; i < e820.nr_map; i++) {
 112                 struct e820entry *ei = &e820.map[i];
 113
 114                 if (type && ei->type != type)
 115                         continue;
 116                 if (ei->addr >= end || ei->addr + ei->size <= start)
 117                         continue;
 118                 return 1;
 119         }
 120         return 0;
 121 }
 122 EXPORT_SYMBOL_GPL(e820_any_mapped);
 123
 124 /*
 125  * This function checks if the entire range <start,end> is mapped with type.
 126  *
 127  * Note: this function only works correct if the e820 table is sorted and
 128  * not-overlapping, which is the case
 129  */
 130 int __init e820_all_mapped(unsigned long start, unsigned long end,
 131                            unsigned type)
 132 {
 133         int i;
 134
 135         for (i = 0; i < e820.nr_map; i++) {
 136                 struct e820entry *ei = &e820.map[i];
 137
 138                 if (type && ei->type != type)
 139                         continue;
 140                 /* is the region (part) in overlap with the current region ?*/
 141                 if (ei->addr >= end || ei->addr + ei->size <= start)
 142                         continue;
 143
 144                 /* if the region is at the beginning of <start,end> we move
 145                  * start to the end of the region since it's ok until there
 146                  */
 147                 if (ei->addr <= start)
 148                         start = ei->addr + ei->size;
 149                 /*
 150                  * if start is now at or beyond end, we're done, full
 151                  * coverage
 152                  */
 153                 if (start >= end)
 154                         return 1;
 155         }
 156         return 0;
 157 }
 158
 159 /*
 160  * Find a free area in a specific range.
 161  */
 162 unsigned long __init find_e820_area(unsigned long start, unsigned long end,
 163                                     unsigned size)
 164 {
 165         int i;
 166
 167         for (i = 0; i < e820.nr_map; i++) {
 168                 struct e820entry *ei = &e820.map[i];
 169                 unsigned long addr = ei->addr, last;
 170
 171                 if (ei->type != E820_RAM)
 172                         continue;
 173                 if (addr < start)
 174                         addr = start;
 175                 if (addr > ei->addr + ei->size)
 176                         continue;
 177                 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
 178                         ;
 179                 last = PAGE_ALIGN(addr) + size;
 180                 if (last > ei->addr + ei->size)
 181                         continue;
 182                 if (last > end)
 183                         continue;
 184                 return addr;
 185         }
 186         return -1UL;
 187 }
 188
 189 /*
 190  * Find the highest page frame number we have available
 191  */
 192 unsigned long __init e820_end_of_ram(void)
 193 {
 194         unsigned long end_pfn;
 195
 196         end_pfn = find_max_pfn_with_active_regions();
 197
 198         if (end_pfn > end_pfn_map)
 199                 end_pfn_map = end_pfn;
 200         if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
 201                 end_pfn_map = MAXMEM>>PAGE_SHIFT;
 202         if (end_pfn > end_user_pfn)
 203                 end_pfn = end_user_pfn;
 204         if (end_pfn > end_pfn_map)
 205                 end_pfn = end_pfn_map;
 206
 207         printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map);
 208         return end_pfn;
 209 }
 210
 211 /*
 212  * Mark e820 reserved areas as busy for the resource manager.
 213  */
 214 void __init e820_reserve_resources(struct resource *code_resource,
 215                 struct resource *data_resource, struct resource *bss_resource)
 216 {
 217         int i;
 218         for (i = 0; i < e820.nr_map; i++) {
 219                 struct resource *res;
 220                 res = alloc_bootmem_low(sizeof(struct resource));
 221                 switch (e820.map[i].type) {
 222                 case E820_RAM:  res->name = "System RAM"; break;
 223                 case E820_ACPI: res->name = "ACPI Tables"; break;
 224                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
 225                 default:        res->name = "reserved";
 226                 }
 227                 res->start = e820.map[i].addr;
 228                 res->end = res->start + e820.map[i].size - 1;
 229                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 230                 request_resource(&iomem_resource, res);
 231                 if (e820.map[i].type == E820_RAM) {
 232                         /*
 233                          * We don't know which RAM region contains kernel data,
 234                          * so we try it repeatedly and let the resource manager
 235                          * test it.
 236                          */
 237                         request_resource(res, code_resource);
 238                         request_resource(res, data_resource);
 239                         request_resource(res, bss_resource);
 240 #ifdef CONFIG_KEXEC
 241                         if (crashk_res.start != crashk_res.end)
 242                                 request_resource(res, &crashk_res);
 243 #endif
 244                 }
 245         }
 246 }
 247
 248 /*
 249  * Find the ranges of physical addresses that do not correspond to
 250  * e820 RAM areas and mark the corresponding pages as nosave for software
 251  * suspend and suspend to RAM.
 252  *
 253  * This function requires the e820 map to be sorted and without any
 254  * overlapping entries and assumes the first e820 area to be RAM.
 255  */
 256 void __init e820_mark_nosave_regions(void)
 257 {
 258         int i;
 259         unsigned long paddr;
 260
 261         paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
 262         for (i = 1; i < e820.nr_map; i++) {
 263                 struct e820entry *ei = &e820.map[i];
 264
 265                 if (paddr < ei->addr)
 266                         register_nosave_region(PFN_DOWN(paddr),
 267                                                 PFN_UP(ei->addr));
 268
 269                 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
 270                 if (ei->type != E820_RAM)
 271                         register_nosave_region(PFN_UP(ei->addr),
 272                                                 PFN_DOWN(paddr));
 273
 274                 if (paddr >= (end_pfn << PAGE_SHIFT))
 275                         break;
 276         }
 277 }
 278
 279 /*
 280  * Finds an active region in the address range from start_pfn to end_pfn and
 281  * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
 282  */
 283 static int __init e820_find_active_region(const struct e820entry *ei,
 284                                           unsigned long start_pfn,
 285                                           unsigned long end_pfn,
 286                                           unsigned long *ei_startpfn,
 287                                           unsigned long *ei_endpfn)
 288 {
 289         *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
 290         *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
 291
 292         /* Skip map entries smaller than a page */
 293         if (*ei_startpfn >= *ei_endpfn)
 294                 return 0;
 295
 296         /* Check if end_pfn_map should be updated */
 297         if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
 298                 end_pfn_map = *ei_endpfn;
 299
 300         /* Skip if map is outside the node */
 301         if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
 302                                     *ei_startpfn >= end_pfn)
 303                 return 0;
 304
 305         /* Check for overlaps */
 306         if (*ei_startpfn < start_pfn)
 307                 *ei_startpfn = start_pfn;
 308         if (*ei_endpfn > end_pfn)
 309                 *ei_endpfn = end_pfn;
 310
 311         /* Obey end_user_pfn to save on memmap */
 312         if (*ei_startpfn >= end_user_pfn)
 313                 return 0;
 314         if (*ei_endpfn > end_user_pfn)
 315                 *ei_endpfn = end_user_pfn;
 316
 317         return 1;
 318 }
 319
 320 /* Walk the e820 map and register active regions within a node */
 321 void __init
 322 e820_register_active_regions(int nid, unsigned long start_pfn,
 323                                                         unsigned long end_pfn)
 324 {
 325         unsigned long ei_startpfn;
 326         unsigned long ei_endpfn;
 327         int i;
 328
 329         for (i = 0; i < e820.nr_map; i++)
 330                 if (e820_find_active_region(&e820.map[i],
 331                                             start_pfn, end_pfn,
 332                                             &ei_startpfn, &ei_endpfn))
 333                         add_active_range(nid, ei_startpfn, ei_endpfn);
 334 }
 335
 336 /*
 337  * Add a memory region to the kernel e820 map.
 338  */
 339 void __init add_memory_region(unsigned long start, unsigned long size, int type)
 340 {
 341         int x = e820.nr_map;
 342
 343         if (x == E820MAX) {
 344                 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 345                 return;
 346         }
 347
 348         e820.map[x].addr = start;
 349         e820.map[x].size = size;
 350         e820.map[x].type = type;
 351         e820.nr_map++;
 352 }
 353
 354 /*
 355  * Find the hole size (in bytes) in the memory range.
 356  * @start: starting address of the memory range to scan
 357  * @end: ending address of the memory range to scan
 358  */
 359 unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
 360 {
 361         unsigned long start_pfn = start >> PAGE_SHIFT;
 362         unsigned long end_pfn = end >> PAGE_SHIFT;
 363         unsigned long ei_startpfn, ei_endpfn, ram = 0;
 364         int i;
 365
 366         for (i = 0; i < e820.nr_map; i++) {
 367                 if (e820_find_active_region(&e820.map[i],
 368                                             start_pfn, end_pfn,
 369                                             &ei_startpfn, &ei_endpfn))
 370                         ram += ei_endpfn - ei_startpfn;
 371         }
 372         return end - start - (ram << PAGE_SHIFT);
 373 }
 374
 375 static void __init e820_print_map(char *who)
 376 {
 377         int i;
 378
 379         for (i = 0; i < e820.nr_map; i++) {
 380                 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
 381                        (unsigned long long) e820.map[i].addr,
 382                        (unsigned long long)
 383                        (e820.map[i].addr + e820.map[i].size));
 384                 switch (e820.map[i].type) {
 385                 case E820_RAM:
 386                         printk(KERN_CONT "(usable)\n");
 387                         break;
 388                 case E820_RESERVED:
 389                         printk(KERN_CONT "(reserved)\n");
 390                         break;
 391                 case E820_ACPI:
 392                         printk(KERN_CONT "(ACPI data)\n");
 393                         break;
 394                 case E820_NVS:
 395                         printk(KERN_CONT "(ACPI NVS)\n");
 396                         break;
 397                 default:
 398                         printk(KERN_CONT "type %u\n", e820.map[i].type);
 399                         break;
 400                 }
 401         }
 402 }
 403
 404 /*
 405  * Sanitize the BIOS e820 map.
 406  *
 407  * Some e820 responses include overlapping entries. The following
 408  * replaces the original e820 map with a new one, removing overlaps.
 409  *
 410  */
 411 static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
 412 {
 413         struct change_member {
 414                 struct e820entry *pbios; /* pointer to original bios entry */
 415                 unsigned long long addr; /* address for this change point */
 416         };
 417         static struct change_member change_point_list[2*E820MAX] __initdata;
 418         static struct change_member *change_point[2*E820MAX] __initdata;
 419         static struct e820entry *overlap_list[E820MAX] __initdata;
 420         static struct e820entry new_bios[E820MAX] __initdata;
 421         struct change_member *change_tmp;
 422         unsigned long current_type, last_type;
 423         unsigned long long last_addr;
 424         int chgidx, still_changing;
 425         int overlap_entries;
 426         int new_bios_entry;
 427         int old_nr, new_nr, chg_nr;
 428         int i;
 429
 430         /*
 431                 Visually we're performing the following
 432                 (1,2,3,4 = memory types)...
 433
 434                 Sample memory map (w/overlaps):
 435                    ____22__________________
 436                    ______________________4_
 437                    ____1111________________
 438                    _44_____________________
 439                    11111111________________
 440                    ____________________33__
 441                    ___________44___________
 442                    __________33333_________
 443                    ______________22________
 444                    ___________________2222_
 445                    _________111111111______
 446                    _____________________11_
 447                    _________________4______
 448
 449                 Sanitized equivalent (no overlap):
 450                    1_______________________
 451                    _44_____________________
 452                    ___1____________________
 453                    ____22__________________
 454                    ______11________________
 455                    _________1______________
 456                    __________3_____________
 457                    ___________44___________
 458                    _____________33_________
 459                    _______________2________
 460                    ________________1_______
 461                    _________________4______
 462                    ___________________2____
 463                    ____________________33__
 464                    ______________________4_
 465         */
 466
 467         /* if there's only one memory region, don't bother */
 468         if (*pnr_map < 2)
 469                 return -1;
 470
 471         old_nr = *pnr_map;
 472
 473         /* bail out if we find any unreasonable addresses in bios map */
 474         for (i = 0; i < old_nr; i++)
 475                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 476                         return -1;
 477
 478         /* create pointers for initial change-point information (for sorting) */
 479         for (i = 0; i < 2 * old_nr; i++)
 480                 change_point[i] = &change_point_list[i];
 481
 482         /* record all known change-points (starting and ending addresses),
 483            omitting those that are for empty memory regions */
 484         chgidx = 0;
 485         for (i = 0; i < old_nr; i++)    {
 486                 if (biosmap[i].size != 0) {
 487                         change_point[chgidx]->addr = biosmap[i].addr;
 488                         change_point[chgidx++]->pbios = &biosmap[i];
 489                         change_point[chgidx]->addr = biosmap[i].addr +
 490                                 biosmap[i].size;
 491                         change_point[chgidx++]->pbios = &biosmap[i];
 492                 }
 493         }
 494         chg_nr = chgidx;
 495
 496         /* sort change-point list by memory addresses (low -> high) */
 497         still_changing = 1;
 498         while (still_changing)  {
 499                 still_changing = 0;
 500                 for (i = 1; i < chg_nr; i++)  {
 501                         unsigned long long curaddr, lastaddr;
 502                         unsigned long long curpbaddr, lastpbaddr;
 503
 504                         curaddr = change_point[i]->addr;
 505                         lastaddr = change_point[i - 1]->addr;
 506                         curpbaddr = change_point[i]->pbios->addr;
 507                         lastpbaddr = change_point[i - 1]->pbios->addr;
 508
 509                         /*
 510                          * swap entries, when:
 511                          *
 512                          * curaddr > lastaddr or
 513                          * curaddr == lastaddr and curaddr == curpbaddr and
 514                          * lastaddr != lastpbaddr
 515                          */
 516                         if (curaddr < lastaddr ||
 517                             (curaddr == lastaddr && curaddr == curpbaddr &&
 518                              lastaddr != lastpbaddr)) {
 519                                 change_tmp = change_point[i];
 520                                 change_point[i] = change_point[i-1];
 521                                 change_point[i-1] = change_tmp;
 522                                 still_changing = 1;
 523                         }
 524                 }
 525         }
 526
 527         /* create a new bios memory map, removing overlaps */
 528         overlap_entries = 0;     /* number of entries in the overlap table */
 529         new_bios_entry = 0;      /* index for creating new bios map entries */
 530         last_type = 0;           /* start with undefined memory type */
 531         last_addr = 0;           /* start with 0 as last starting address */
 532
 533         /* loop through change-points, determining affect on the new bios map */
 534         for (chgidx = 0; chgidx < chg_nr; chgidx++) {
 535                 /* keep track of all overlapping bios entries */
 536                 if (change_point[chgidx]->addr ==
 537                     change_point[chgidx]->pbios->addr) {
 538                         /*
 539                          * add map entry to overlap list (> 1 entry
 540                          * implies an overlap)
 541                          */
 542                         overlap_list[overlap_entries++] =
 543                                 change_point[chgidx]->pbios;
 544                 } else {
 545                         /*
 546                          * remove entry from list (order independent,
 547                          * so swap with last)
 548                          */
 549                         for (i = 0; i < overlap_entries; i++) {
 550                                 if (overlap_list[i] ==
 551                                     change_point[chgidx]->pbios)
 552                                         overlap_list[i] =
 553                                                 overlap_list[overlap_entries-1];
 554                         }
 555                         overlap_entries--;
 556                 }
 557                 /*
 558                  * if there are overlapping entries, decide which
 559                  * "type" to use (larger value takes precedence --
 560                  * 1=usable, 2,3,4,4+=unusable)
 561                  */
 562                 current_type = 0;
 563                 for (i = 0; i < overlap_entries; i++)
 564                         if (overlap_list[i]->type > current_type)
 565                                 current_type = overlap_list[i]->type;
 566                 /*
 567                  * continue building up new bios map based on this
 568                  * information
 569                  */
 570                 if (current_type != last_type)  {
 571                         if (last_type != 0)      {
 572                                 new_bios[new_bios_entry].size =
 573                                         change_point[chgidx]->addr - last_addr;
 574                                 /*
 575                                  * move forward only if the new size
 576                                  * was non-zero
 577                                  */
 578                                 if (new_bios[new_bios_entry].size != 0)
 579                                         /*
 580                                          * no more space left for new
 581                                          * bios entries ?
 582                                          */
 583                                         if (++new_bios_entry >= E820MAX)
 584                                                 break;
 585                         }
 586                         if (current_type != 0)  {
 587                                 new_bios[new_bios_entry].addr =
 588                                         change_point[chgidx]->addr;
 589                                 new_bios[new_bios_entry].type = current_type;
 590                                 last_addr = change_point[chgidx]->addr;
 591                         }
 592                         last_type = current_type;
 593                 }
 594         }
 595         /* retain count for new bios entries */
 596         new_nr = new_bios_entry;
 597
 598         /* copy new bios mapping into original location */
 599         memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
 600         *pnr_map = new_nr;
 601
 602         return 0;
 603 }
 604
 605 /*
 606  * Copy the BIOS e820 map into a safe place.
 607  *
 608  * Sanity-check it while we're at it..
 609  *
 610  * If we're lucky and live on a modern system, the setup code
 611  * will have given us a memory map that we can use to properly
 612  * set up memory.  If we aren't, we'll fake a memory map.
 613  */
 614 static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
 615 {
 616         /* Only one memory region (or negative)? Ignore it */
 617         if (nr_map < 2)
 618                 return -1;
 619
 620         do {
 621                 unsigned long start = biosmap->addr;
 622                 unsigned long size = biosmap->size;
 623                 unsigned long end = start + size;
 624                 unsigned long type = biosmap->type;
 625
 626                 /* Overflow in 64 bits? Ignore the memory map. */
 627                 if (start > end)
 628                         return -1;
 629
 630                 add_memory_region(start, size, type);
 631         } while (biosmap++, --nr_map);
 632         return 0;
 633 }
 634
 635 static void early_panic(char *msg)
 636 {
 637         early_printk(msg);
 638         panic(msg);
 639 }
 640
 641 /* We're not void only for x86 32-bit compat */
 642 char * __init machine_specific_memory_setup(void)
 643 {
 644         char *who = "BIOS-e820";
 645         /*
 646          * Try to copy the BIOS-supplied E820-map.
 647          *
 648          * Otherwise fake a memory map; one section from 0k->640k,
 649          * the next section from 1mb->appropriate_mem_k
 650          */
 651         sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
 652         if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
 653                 early_panic("Cannot find a valid memory map");
 654         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 655         e820_print_map(who);
 656
 657         /* In case someone cares... */
 658         return who;
 659 }
 660
 661 static int __init parse_memopt(char *p)
 662 {
 663         if (!p)
 664                 return -EINVAL;
 665         end_user_pfn = memparse(p, &p);
 666         end_user_pfn >>= PAGE_SHIFT;
 667         return 0;
 668 }
 669 early_param("mem", parse_memopt);
 670
 671 static int userdef __initdata;
 672
 673 static int __init parse_memmap_opt(char *p)
 674 {
 675         char *oldp;
 676         unsigned long long start_at, mem_size;
 677
 678         if (!strcmp(p, "exactmap")) {
 679 #ifdef CONFIG_CRASH_DUMP
 680                 /*
 681                  * If we are doing a crash dump, we still need to know
 682                  * the real mem size before original memory map is
 683                  * reset.
 684                  */
 685                 e820_register_active_regions(0, 0, -1UL);
 686                 saved_max_pfn = e820_end_of_ram();
 687                 remove_all_active_ranges();
 688 #endif
 689                 end_pfn_map = 0;
 690                 e820.nr_map = 0;
 691                 userdef = 1;
 692                 return 0;
 693         }
 694
 695         oldp = p;
 696         mem_size = memparse(p, &p);
 697         if (p == oldp)
 698                 return -EINVAL;
 699
 700         userdef = 1;
 701         if (*p == '@') {
 702                 start_at = memparse(p+1, &p);
 703                 add_memory_region(start_at, mem_size, E820_RAM);
 704         } else if (*p == '#') {
 705                 start_at = memparse(p+1, &p);
 706                 add_memory_region(start_at, mem_size, E820_ACPI);
 707         } else if (*p == '$') {
 708                 start_at = memparse(p+1, &p);
 709                 add_memory_region(start_at, mem_size, E820_RESERVED);
 710         } else {
 711                 end_user_pfn = (mem_size >> PAGE_SHIFT);
 712         }
 713         return *p == '\0' ? 0 : -EINVAL;
 714 }
 715 early_param("memmap", parse_memmap_opt);
 716
 717 void __init finish_e820_parsing(void)
 718 {
 719         if (userdef) {
 720                 char nr = e820.nr_map;
 721
 722                 if (sanitize_e820_map(e820.map, &nr) < 0)
 723                         early_panic("Invalid user supplied memory map");
 724                 e820.nr_map = nr;
 725
 726                 printk(KERN_INFO "user-defined physical RAM map:\n");
 727                 e820_print_map("user");
 728         }
 729 }
 730
 731 unsigned long pci_mem_start = 0xaeedbabe;
 732 EXPORT_SYMBOL(pci_mem_start);
 733
 734 /*
 735  * Search for the biggest gap in the low 32 bits of the e820
 736  * memory space.  We pass this space to PCI to assign MMIO resources
 737  * for hotplug or unconfigured devices in.
 738  * Hopefully the BIOS let enough space left.
 739  */
 740 __init void e820_setup_gap(void)
 741 {
 742         unsigned long gapstart, gapsize, round;
 743         unsigned long last;
 744         int i;
 745         int found = 0;
 746
 747         last = 0x100000000ull;
 748         gapstart = 0x10000000;
 749         gapsize = 0x400000;
 750         i = e820.nr_map;
 751         while (--i >= 0) {
 752                 unsigned long long start = e820.map[i].addr;
 753                 unsigned long long end = start + e820.map[i].size;
 754
 755                 /*
 756                  * Since "last" is at most 4GB, we know we'll
 757                  * fit in 32 bits if this condition is true
 758                  */
 759                 if (last > end) {
 760                         unsigned long gap = last - end;
 761
 762                         if (gap > gapsize) {
 763                                 gapsize = gap;
 764                                 gapstart = end;
 765                                 found = 1;
 766                         }
 767                 }
 768                 if (start < last)
 769                         last = start;
 770         }
 771
 772         if (!found) {
 773                 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
 774                 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
 775                        "address range\n"
 776                        KERN_ERR "PCI: Unassigned devices with 32bit resource "
 777                        "registers may break!\n");
 778         }
 779
 780         /*
 781          * See how much we want to round up: start off with
 782          * rounding to the next 1MB area.
 783          */
 784         round = 0x100000;
 785         while ((gapsize >> 4) > round)
 786                 round += round;
 787         /* Fun with two's complement */
 788         pci_mem_start = (gapstart + round) & -round;
 789
 790         printk(KERN_INFO
 791                "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
 792                pci_mem_start, gapstart, gapsize);
 793 }
 794
 795 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
 796 {
 797         int i;
 798
 799         if (slot < 0 || slot >= e820.nr_map)
 800                 return -1;
 801         for (i = slot; i < e820.nr_map; i++) {
 802                 if (e820.map[i].type != E820_RAM)
 803                         continue;
 804                 break;
 805         }
 806         if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
 807                 return -1;
 808         *addr = e820.map[i].addr;
 809         *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
 810                 max_pfn << PAGE_SHIFT) - *addr;
 811         return i + 1;
 812 }