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 #include <asm/trampoline.h>
  31
  32 struct e820map e820;
  33
  34 /*
  35  * PFN of last memory page.
  36  */
  37 unsigned long end_pfn;
  38
  39 /*
  40  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
  41  * The direct mapping extends to max_pfn_mapped, so that we can directly access
  42  * apertures, ACPI and other tables without having to play with fixmaps.
  43  */
  44 unsigned long max_pfn_mapped;
  45
  46 /*
  47  * Last pfn which the user wants to use.
  48  */
  49 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
  50
  51 /*
  52  * Early reserved memory areas.
  53  */
  54 #define MAX_EARLY_RES 20
  55
  56 struct early_res {
  57         unsigned long start, end;
  58         char name[16];
  59 };
  60 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
  61         { 0, PAGE_SIZE, "BIOS data page" },                     /* BIOS data page */
  62 #ifdef CONFIG_X86_TRAMPOLINE
  63         { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
  64 #endif
  65         {}
  66 };
  67
  68 void __init reserve_early(unsigned long start, unsigned long end, char *name)
  69 {
  70         int i;
  71         struct early_res *r;
  72         for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
  73                 r = &early_res[i];
  74                 if (end > r->start && start < r->end)
  75                         panic("Overlapping early reservations %lx-%lx %s to %lx-%lx %s\n",
  76                               start, end - 1, name?name:"", r->start, r->end - 1, r->name);
  77         }
  78         if (i >= MAX_EARLY_RES)
  79                 panic("Too many early reservations");
  80         r = &early_res[i];
  81         r->start = start;
  82         r->end = end;
  83         if (name)
  84                 strncpy(r->name, name, sizeof(r->name) - 1);
  85 }
  86
  87 void __init early_res_to_bootmem(void)
  88 {
  89         int i;
  90         for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
  91                 struct early_res *r = &early_res[i];
  92                 printk(KERN_INFO "early res: %d [%lx-%lx] %s\n", i,
  93                         r->start, r->end - 1, r->name);
  94                 reserve_bootmem_generic(r->start, r->end - r->start);
  95         }
  96 }
  97
  98 /* Check for already reserved areas */
  99 static inline int __init
 100 bad_addr(unsigned long *addrp, unsigned long size, unsigned long align)
 101 {
 102         int i;
 103         unsigned long addr = *addrp, last;
 104         int changed = 0;
 105 again:
 106         last = addr + size;
 107         for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
 108                 struct early_res *r = &early_res[i];
 109                 if (last >= r->start && addr < r->end) {
 110                         *addrp = addr = round_up(r->end, align);
 111                         changed = 1;
 112                         goto again;
 113                 }
 114         }
 115         return changed;
 116 }
 117
 118 /* Check for already reserved areas */
 119 static inline int __init
 120 bad_addr_size(unsigned long *addrp, unsigned long *sizep, unsigned long align)
 121 {
 122         int i;
 123         unsigned long addr = *addrp, last;
 124         unsigned long size = *sizep;
 125         int changed = 0;
 126 again:
 127         last = addr + size;
 128         for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
 129                 struct early_res *r = &early_res[i];
 130                 if (last > r->start && addr < r->start) {
 131                         size = r->start - addr;
 132                         changed = 1;
 133                         goto again;
 134                 }
 135                 if (last > r->end && addr < r->end) {
 136                         addr = round_up(r->end, align);
 137                         size = last - addr;
 138                         changed = 1;
 139                         goto again;
 140                 }
 141                 if (last <= r->end && addr >= r->start) {
 142                         (*sizep)++;
 143                         return 0;
 144                 }
 145         }
 146         if (changed) {
 147                 *addrp = addr;
 148                 *sizep = size;
 149         }
 150         return changed;
 151 }
 152 /*
 153  * This function checks if any part of the range <start,end> is mapped
 154  * with type.
 155  */
 156 int
 157 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
 158 {
 159         int i;
 160
 161         for (i = 0; i < e820.nr_map; i++) {
 162                 struct e820entry *ei = &e820.map[i];
 163
 164                 if (type && ei->type != type)
 165                         continue;
 166                 if (ei->addr >= end || ei->addr + ei->size <= start)
 167                         continue;
 168                 return 1;
 169         }
 170         return 0;
 171 }
 172 EXPORT_SYMBOL_GPL(e820_any_mapped);
 173
 174 /*
 175  * This function checks if the entire range <start,end> is mapped with type.
 176  *
 177  * Note: this function only works correct if the e820 table is sorted and
 178  * not-overlapping, which is the case
 179  */
 180 int __init e820_all_mapped(unsigned long start, unsigned long end,
 181                            unsigned type)
 182 {
 183         int i;
 184
 185         for (i = 0; i < e820.nr_map; i++) {
 186                 struct e820entry *ei = &e820.map[i];
 187
 188                 if (type && ei->type != type)
 189                         continue;
 190                 /* is the region (part) in overlap with the current region ?*/
 191                 if (ei->addr >= end || ei->addr + ei->size <= start)
 192                         continue;
 193
 194                 /* if the region is at the beginning of <start,end> we move
 195                  * start to the end of the region since it's ok until there
 196                  */
 197                 if (ei->addr <= start)
 198                         start = ei->addr + ei->size;
 199                 /*
 200                  * if start is now at or beyond end, we're done, full
 201                  * coverage
 202                  */
 203                 if (start >= end)
 204                         return 1;
 205         }
 206         return 0;
 207 }
 208
 209 /*
 210  * Find a free area with specified alignment in a specific range.
 211  */
 212 unsigned long __init find_e820_area(unsigned long start, unsigned long end,
 213                                     unsigned long size, unsigned long align)
 214 {
 215         int i;
 216
 217         for (i = 0; i < e820.nr_map; i++) {
 218                 struct e820entry *ei = &e820.map[i];
 219                 unsigned long addr, last;
 220                 unsigned long ei_last;
 221
 222                 if (ei->type != E820_RAM)
 223                         continue;
 224                 addr = round_up(ei->addr, align);
 225                 ei_last = ei->addr + ei->size;
 226                 if (addr < start)
 227                         addr = round_up(start, align);
 228                 if (addr >= ei_last)
 229                         continue;
 230                 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
 231                         ;
 232                 last = addr + size;
 233                 if (last > ei_last)
 234                         continue;
 235                 if (last > end)
 236                         continue;
 237                 return addr;
 238         }
 239         return -1UL;
 240 }
 241
 242 /*
 243  * Find next free range after *start
 244  */
 245 unsigned long __init find_e820_area_size(unsigned long start,
 246                                          unsigned long *sizep,
 247                                          unsigned long align)
 248 {
 249         int i;
 250
 251         for (i = 0; i < e820.nr_map; i++) {
 252                 struct e820entry *ei = &e820.map[i];
 253                 unsigned long addr, last;
 254                 unsigned long ei_last;
 255
 256                 if (ei->type != E820_RAM)
 257                         continue;
 258                 addr = round_up(ei->addr, align);
 259                 ei_last = ei->addr + ei->size;
 260                 if (addr < start)
 261                         addr = round_up(start, align);
 262                 if (addr >= ei_last)
 263                         continue;
 264                 *sizep = ei_last - addr;
 265                 while (bad_addr_size(&addr, sizep, align) &&
 266                         addr + *sizep <= ei_last)
 267                         ;
 268                 last = addr + *sizep;
 269                 if (last > ei_last)
 270                         continue;
 271                 return addr;
 272         }
 273         return -1UL;
 274
 275 }
 276 /*
 277  * Find the highest page frame number we have available
 278  */
 279 unsigned long __init e820_end_of_ram(void)
 280 {
 281         unsigned long end_pfn;
 282
 283         end_pfn = find_max_pfn_with_active_regions();
 284
 285         if (end_pfn > max_pfn_mapped)
 286                 max_pfn_mapped = end_pfn;
 287         if (max_pfn_mapped > MAXMEM>>PAGE_SHIFT)
 288                 max_pfn_mapped = MAXMEM>>PAGE_SHIFT;
 289         if (end_pfn > end_user_pfn)
 290                 end_pfn = end_user_pfn;
 291         if (end_pfn > max_pfn_mapped)
 292                 end_pfn = max_pfn_mapped;
 293
 294         printk(KERN_INFO "max_pfn_mapped = %lu\n", max_pfn_mapped);
 295         return end_pfn;
 296 }
 297
 298 /*
 299  * Mark e820 reserved areas as busy for the resource manager.
 300  */
 301 void __init e820_reserve_resources(void)
 302 {
 303         int i;
 304         struct resource *res;
 305
 306         res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
 307         for (i = 0; i < e820.nr_map; i++) {
 308                 switch (e820.map[i].type) {
 309                 case E820_RAM:  res->name = "System RAM"; break;
 310                 case E820_ACPI: res->name = "ACPI Tables"; break;
 311                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
 312                 default:        res->name = "reserved";
 313                 }
 314                 res->start = e820.map[i].addr;
 315                 res->end = res->start + e820.map[i].size - 1;
 316                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 317                 insert_resource(&iomem_resource, res);
 318                 res++;
 319         }
 320 }
 321
 322 /*
 323  * Find the ranges of physical addresses that do not correspond to
 324  * e820 RAM areas and mark the corresponding pages as nosave for software
 325  * suspend and suspend to RAM.
 326  *
 327  * This function requires the e820 map to be sorted and without any
 328  * overlapping entries and assumes the first e820 area to be RAM.
 329  */
 330 void __init e820_mark_nosave_regions(void)
 331 {
 332         int i;
 333         unsigned long paddr;
 334
 335         paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
 336         for (i = 1; i < e820.nr_map; i++) {
 337                 struct e820entry *ei = &e820.map[i];
 338
 339                 if (paddr < ei->addr)
 340                         register_nosave_region(PFN_DOWN(paddr),
 341                                                 PFN_UP(ei->addr));
 342
 343                 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
 344                 if (ei->type != E820_RAM)
 345                         register_nosave_region(PFN_UP(ei->addr),
 346                                                 PFN_DOWN(paddr));
 347
 348                 if (paddr >= (end_pfn << PAGE_SHIFT))
 349                         break;
 350         }
 351 }
 352
 353 /*
 354  * Finds an active region in the address range from start_pfn to end_pfn and
 355  * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
 356  */
 357 static int __init e820_find_active_region(const struct e820entry *ei,
 358                                           unsigned long start_pfn,
 359                                           unsigned long end_pfn,
 360                                           unsigned long *ei_startpfn,
 361                                           unsigned long *ei_endpfn)
 362 {
 363         *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
 364         *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
 365
 366         /* Skip map entries smaller than a page */
 367         if (*ei_startpfn >= *ei_endpfn)
 368                 return 0;
 369
 370         /* Check if max_pfn_mapped should be updated */
 371         if (ei->type != E820_RAM && *ei_endpfn > max_pfn_mapped)
 372                 max_pfn_mapped = *ei_endpfn;
 373
 374         /* Skip if map is outside the node */
 375         if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
 376                                     *ei_startpfn >= end_pfn)
 377                 return 0;
 378
 379         /* Check for overlaps */
 380         if (*ei_startpfn < start_pfn)
 381                 *ei_startpfn = start_pfn;
 382         if (*ei_endpfn > end_pfn)
 383                 *ei_endpfn = end_pfn;
 384
 385         /* Obey end_user_pfn to save on memmap */
 386         if (*ei_startpfn >= end_user_pfn)
 387                 return 0;
 388         if (*ei_endpfn > end_user_pfn)
 389                 *ei_endpfn = end_user_pfn;
 390
 391         return 1;
 392 }
 393
 394 /* Walk the e820 map and register active regions within a node */
 395 void __init
 396 e820_register_active_regions(int nid, unsigned long start_pfn,
 397                                                         unsigned long end_pfn)
 398 {
 399         unsigned long ei_startpfn;
 400         unsigned long ei_endpfn;
 401         int i;
 402
 403         for (i = 0; i < e820.nr_map; i++)
 404                 if (e820_find_active_region(&e820.map[i],
 405                                             start_pfn, end_pfn,
 406                                             &ei_startpfn, &ei_endpfn))
 407                         add_active_range(nid, ei_startpfn, ei_endpfn);
 408 }
 409
 410 /*
 411  * Add a memory region to the kernel e820 map.
 412  */
 413 void __init add_memory_region(unsigned long start, unsigned long size, int type)
 414 {
 415         int x = e820.nr_map;
 416
 417         if (x == E820MAX) {
 418                 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 419                 return;
 420         }
 421
 422         e820.map[x].addr = start;
 423         e820.map[x].size = size;
 424         e820.map[x].type = type;
 425         e820.nr_map++;
 426 }
 427
 428 /*
 429  * Find the hole size (in bytes) in the memory range.
 430  * @start: starting address of the memory range to scan
 431  * @end: ending address of the memory range to scan
 432  */
 433 unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
 434 {
 435         unsigned long start_pfn = start >> PAGE_SHIFT;
 436         unsigned long end_pfn = end >> PAGE_SHIFT;
 437         unsigned long ei_startpfn, ei_endpfn, ram = 0;
 438         int i;
 439
 440         for (i = 0; i < e820.nr_map; i++) {
 441                 if (e820_find_active_region(&e820.map[i],
 442                                             start_pfn, end_pfn,
 443                                             &ei_startpfn, &ei_endpfn))
 444                         ram += ei_endpfn - ei_startpfn;
 445         }
 446         return end - start - (ram << PAGE_SHIFT);
 447 }
 448
 449 static void __init e820_print_map(char *who)
 450 {
 451         int i;
 452
 453         for (i = 0; i < e820.nr_map; i++) {
 454                 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
 455                        (unsigned long long) e820.map[i].addr,
 456                        (unsigned long long)
 457                        (e820.map[i].addr + e820.map[i].size));
 458                 switch (e820.map[i].type) {
 459                 case E820_RAM:
 460                         printk(KERN_CONT "(usable)\n");
 461                         break;
 462                 case E820_RESERVED:
 463                         printk(KERN_CONT "(reserved)\n");
 464                         break;
 465                 case E820_ACPI:
 466                         printk(KERN_CONT "(ACPI data)\n");
 467                         break;
 468                 case E820_NVS:
 469                         printk(KERN_CONT "(ACPI NVS)\n");
 470                         break;
 471                 default:
 472                         printk(KERN_CONT "type %u\n", e820.map[i].type);
 473                         break;
 474                 }
 475         }
 476 }
 477
 478 /*
 479  * Sanitize the BIOS e820 map.
 480  *
 481  * Some e820 responses include overlapping entries. The following
 482  * replaces the original e820 map with a new one, removing overlaps.
 483  *
 484  */
 485 static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
 486 {
 487         struct change_member {
 488                 struct e820entry *pbios; /* pointer to original bios entry */
 489                 unsigned long long addr; /* address for this change point */
 490         };
 491         static struct change_member change_point_list[2*E820MAX] __initdata;
 492         static struct change_member *change_point[2*E820MAX] __initdata;
 493         static struct e820entry *overlap_list[E820MAX] __initdata;
 494         static struct e820entry new_bios[E820MAX] __initdata;
 495         struct change_member *change_tmp;
 496         unsigned long current_type, last_type;
 497         unsigned long long last_addr;
 498         int chgidx, still_changing;
 499         int overlap_entries;
 500         int new_bios_entry;
 501         int old_nr, new_nr, chg_nr;
 502         int i;
 503
 504         /*
 505                 Visually we're performing the following
 506                 (1,2,3,4 = memory types)...
 507
 508                 Sample memory map (w/overlaps):
 509                    ____22__________________
 510                    ______________________4_
 511                    ____1111________________
 512                    _44_____________________
 513                    11111111________________
 514                    ____________________33__
 515                    ___________44___________
 516                    __________33333_________
 517                    ______________22________
 518                    ___________________2222_
 519                    _________111111111______
 520                    _____________________11_
 521                    _________________4______
 522
 523                 Sanitized equivalent (no overlap):
 524                    1_______________________
 525                    _44_____________________
 526                    ___1____________________
 527                    ____22__________________
 528                    ______11________________
 529                    _________1______________
 530                    __________3_____________
 531                    ___________44___________
 532                    _____________33_________
 533                    _______________2________
 534                    ________________1_______
 535                    _________________4______
 536                    ___________________2____
 537                    ____________________33__
 538                    ______________________4_
 539         */
 540
 541         /* if there's only one memory region, don't bother */
 542         if (*pnr_map < 2)
 543                 return -1;
 544
 545         old_nr = *pnr_map;
 546
 547         /* bail out if we find any unreasonable addresses in bios map */
 548         for (i = 0; i < old_nr; i++)
 549                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 550                         return -1;
 551
 552         /* create pointers for initial change-point information (for sorting) */
 553         for (i = 0; i < 2 * old_nr; i++)
 554                 change_point[i] = &change_point_list[i];
 555
 556         /* record all known change-points (starting and ending addresses),
 557            omitting those that are for empty memory regions */
 558         chgidx = 0;
 559         for (i = 0; i < old_nr; i++)    {
 560                 if (biosmap[i].size != 0) {
 561                         change_point[chgidx]->addr = biosmap[i].addr;
 562                         change_point[chgidx++]->pbios = &biosmap[i];
 563                         change_point[chgidx]->addr = biosmap[i].addr +
 564                                 biosmap[i].size;
 565                         change_point[chgidx++]->pbios = &biosmap[i];
 566                 }
 567         }
 568         chg_nr = chgidx;
 569
 570         /* sort change-point list by memory addresses (low -> high) */
 571         still_changing = 1;
 572         while (still_changing)  {
 573                 still_changing = 0;
 574                 for (i = 1; i < chg_nr; i++)  {
 575                         unsigned long long curaddr, lastaddr;
 576                         unsigned long long curpbaddr, lastpbaddr;
 577
 578                         curaddr = change_point[i]->addr;
 579                         lastaddr = change_point[i - 1]->addr;
 580                         curpbaddr = change_point[i]->pbios->addr;
 581                         lastpbaddr = change_point[i - 1]->pbios->addr;
 582
 583                         /*
 584                          * swap entries, when:
 585                          *
 586                          * curaddr > lastaddr or
 587                          * curaddr == lastaddr and curaddr == curpbaddr and
 588                          * lastaddr != lastpbaddr
 589                          */
 590                         if (curaddr < lastaddr ||
 591                             (curaddr == lastaddr && curaddr == curpbaddr &&
 592                              lastaddr != lastpbaddr)) {
 593                                 change_tmp = change_point[i];
 594                                 change_point[i] = change_point[i-1];
 595                                 change_point[i-1] = change_tmp;
 596                                 still_changing = 1;
 597                         }
 598                 }
 599         }
 600
 601         /* create a new bios memory map, removing overlaps */
 602         overlap_entries = 0;     /* number of entries in the overlap table */
 603         new_bios_entry = 0;      /* index for creating new bios map entries */
 604         last_type = 0;           /* start with undefined memory type */
 605         last_addr = 0;           /* start with 0 as last starting address */
 606
 607         /* loop through change-points, determining affect on the new bios map */
 608         for (chgidx = 0; chgidx < chg_nr; chgidx++) {
 609                 /* keep track of all overlapping bios entries */
 610                 if (change_point[chgidx]->addr ==
 611                     change_point[chgidx]->pbios->addr) {
 612                         /*
 613                          * add map entry to overlap list (> 1 entry
 614                          * implies an overlap)
 615                          */
 616                         overlap_list[overlap_entries++] =
 617                                 change_point[chgidx]->pbios;
 618                 } else {
 619                         /*
 620                          * remove entry from list (order independent,
 621                          * so swap with last)
 622                          */
 623                         for (i = 0; i < overlap_entries; i++) {
 624                                 if (overlap_list[i] ==
 625                                     change_point[chgidx]->pbios)
 626                                         overlap_list[i] =
 627                                                 overlap_list[overlap_entries-1];
 628                         }
 629                         overlap_entries--;
 630                 }
 631                 /*
 632                  * if there are overlapping entries, decide which
 633                  * "type" to use (larger value takes precedence --
 634                  * 1=usable, 2,3,4,4+=unusable)
 635                  */
 636                 current_type = 0;
 637                 for (i = 0; i < overlap_entries; i++)
 638                         if (overlap_list[i]->type > current_type)
 639                                 current_type = overlap_list[i]->type;
 640                 /*
 641                  * continue building up new bios map based on this
 642                  * information
 643                  */
 644                 if (current_type != last_type)  {
 645                         if (last_type != 0)      {
 646                                 new_bios[new_bios_entry].size =
 647                                         change_point[chgidx]->addr - last_addr;
 648                                 /*
 649                                  * move forward only if the new size
 650                                  * was non-zero
 651                                  */
 652                                 if (new_bios[new_bios_entry].size != 0)
 653                                         /*
 654                                          * no more space left for new
 655                                          * bios entries ?
 656                                          */
 657                                         if (++new_bios_entry >= E820MAX)
 658                                                 break;
 659                         }
 660                         if (current_type != 0)  {
 661                                 new_bios[new_bios_entry].addr =
 662                                         change_point[chgidx]->addr;
 663                                 new_bios[new_bios_entry].type = current_type;
 664                                 last_addr = change_point[chgidx]->addr;
 665                         }
 666                         last_type = current_type;
 667                 }
 668         }
 669         /* retain count for new bios entries */
 670         new_nr = new_bios_entry;
 671
 672         /* copy new bios mapping into original location */
 673         memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
 674         *pnr_map = new_nr;
 675
 676         return 0;
 677 }
 678
 679 /*
 680  * Copy the BIOS e820 map into a safe place.
 681  *
 682  * Sanity-check it while we're at it..
 683  *
 684  * If we're lucky and live on a modern system, the setup code
 685  * will have given us a memory map that we can use to properly
 686  * set up memory.  If we aren't, we'll fake a memory map.
 687  */
 688 static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
 689 {
 690         /* Only one memory region (or negative)? Ignore it */
 691         if (nr_map < 2)
 692                 return -1;
 693
 694         do {
 695                 u64 start = biosmap->addr;
 696                 u64 size = biosmap->size;
 697                 u64 end = start + size;
 698                 u32 type = biosmap->type;
 699
 700                 /* Overflow in 64 bits? Ignore the memory map. */
 701                 if (start > end)
 702                         return -1;
 703
 704                 add_memory_region(start, size, type);
 705         } while (biosmap++, --nr_map);
 706         return 0;
 707 }
 708
 709 static void early_panic(char *msg)
 710 {
 711         early_printk(msg);
 712         panic(msg);
 713 }
 714
 715 /* We're not void only for x86 32-bit compat */
 716 char * __init machine_specific_memory_setup(void)
 717 {
 718         char *who = "BIOS-e820";
 719         /*
 720          * Try to copy the BIOS-supplied E820-map.
 721          *
 722          * Otherwise fake a memory map; one section from 0k->640k,
 723          * the next section from 1mb->appropriate_mem_k
 724          */
 725         sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
 726         if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
 727                 early_panic("Cannot find a valid memory map");
 728         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 729         e820_print_map(who);
 730
 731         /* In case someone cares... */
 732         return who;
 733 }
 734
 735 static int __init parse_memopt(char *p)
 736 {
 737         if (!p)
 738                 return -EINVAL;
 739         end_user_pfn = memparse(p, &p);
 740         end_user_pfn >>= PAGE_SHIFT;
 741         return 0;
 742 }
 743 early_param("mem", parse_memopt);
 744
 745 static int userdef __initdata;
 746
 747 static int __init parse_memmap_opt(char *p)
 748 {
 749         char *oldp;
 750         unsigned long long start_at, mem_size;
 751
 752         if (!strcmp(p, "exactmap")) {
 753 #ifdef CONFIG_CRASH_DUMP
 754                 /*
 755                  * If we are doing a crash dump, we still need to know
 756                  * the real mem size before original memory map is
 757                  * reset.
 758                  */
 759                 e820_register_active_regions(0, 0, -1UL);
 760                 saved_max_pfn = e820_end_of_ram();
 761                 remove_all_active_ranges();
 762 #endif
 763                 max_pfn_mapped = 0;
 764                 e820.nr_map = 0;
 765                 userdef = 1;
 766                 return 0;
 767         }
 768
 769         oldp = p;
 770         mem_size = memparse(p, &p);
 771         if (p == oldp)
 772                 return -EINVAL;
 773
 774         userdef = 1;
 775         if (*p == '@') {
 776                 start_at = memparse(p+1, &p);
 777                 add_memory_region(start_at, mem_size, E820_RAM);
 778         } else if (*p == '#') {
 779                 start_at = memparse(p+1, &p);
 780                 add_memory_region(start_at, mem_size, E820_ACPI);
 781         } else if (*p == '$') {
 782                 start_at = memparse(p+1, &p);
 783                 add_memory_region(start_at, mem_size, E820_RESERVED);
 784         } else {
 785                 end_user_pfn = (mem_size >> PAGE_SHIFT);
 786         }
 787         return *p == '\0' ? 0 : -EINVAL;
 788 }
 789 early_param("memmap", parse_memmap_opt);
 790
 791 void __init finish_e820_parsing(void)
 792 {
 793         if (userdef) {
 794                 char nr = e820.nr_map;
 795
 796                 if (sanitize_e820_map(e820.map, &nr) < 0)
 797                         early_panic("Invalid user supplied memory map");
 798                 e820.nr_map = nr;
 799
 800                 printk(KERN_INFO "user-defined physical RAM map:\n");
 801                 e820_print_map("user");
 802         }
 803 }
 804
 805 void __init update_memory_range(u64 start, u64 size, unsigned old_type,
 806                                 unsigned new_type)
 807 {
 808         int i;
 809
 810         BUG_ON(old_type == new_type);
 811
 812         for (i = 0; i < e820.nr_map; i++) {
 813                 struct e820entry *ei = &e820.map[i];
 814                 u64 final_start, final_end;
 815                 if (ei->type != old_type)
 816                         continue;
 817                 /* totally covered? */
 818                 if (ei->addr >= start && ei->size <= size) {
 819                         ei->type = new_type;
 820                         continue;
 821                 }
 822                 /* partially covered */
 823                 final_start = max(start, ei->addr);
 824                 final_end = min(start + size, ei->addr + ei->size);
 825                 if (final_start >= final_end)
 826                         continue;
 827                 add_memory_region(final_start, final_end - final_start,
 828                                          new_type);
 829         }
 830 }
 831
 832 void __init update_e820(void)
 833 {
 834         u8 nr_map;
 835
 836         nr_map = e820.nr_map;
 837         if (sanitize_e820_map(e820.map, &nr_map))
 838                 return;
 839         e820.nr_map = nr_map;
 840         printk(KERN_INFO "modified physical RAM map:\n");
 841         e820_print_map("modified");
 842 }
 843
 844 unsigned long pci_mem_start = 0xaeedbabe;
 845 EXPORT_SYMBOL(pci_mem_start);
 846
 847 /*
 848  * Search for the biggest gap in the low 32 bits of the e820
 849  * memory space.  We pass this space to PCI to assign MMIO resources
 850  * for hotplug or unconfigured devices in.
 851  * Hopefully the BIOS let enough space left.
 852  */
 853 __init void e820_setup_gap(void)
 854 {
 855         unsigned long gapstart, gapsize, round;
 856         unsigned long last;
 857         int i;
 858         int found = 0;
 859
 860         last = 0x100000000ull;
 861         gapstart = 0x10000000;
 862         gapsize = 0x400000;
 863         i = e820.nr_map;
 864         while (--i >= 0) {
 865                 unsigned long long start = e820.map[i].addr;
 866                 unsigned long long end = start + e820.map[i].size;
 867
 868                 /*
 869                  * Since "last" is at most 4GB, we know we'll
 870                  * fit in 32 bits if this condition is true
 871                  */
 872                 if (last > end) {
 873                         unsigned long gap = last - end;
 874
 875                         if (gap > gapsize) {
 876                                 gapsize = gap;
 877                                 gapstart = end;
 878                                 found = 1;
 879                         }
 880                 }
 881                 if (start < last)
 882                         last = start;
 883         }
 884
 885         if (!found) {
 886                 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
 887                 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
 888                        "address range\n"
 889                        KERN_ERR "PCI: Unassigned devices with 32bit resource "
 890                        "registers may break!\n");
 891         }
 892
 893         /*
 894          * See how much we want to round up: start off with
 895          * rounding to the next 1MB area.
 896          */
 897         round = 0x100000;
 898         while ((gapsize >> 4) > round)
 899                 round += round;
 900         /* Fun with two's complement */
 901         pci_mem_start = (gapstart + round) & -round;
 902
 903         printk(KERN_INFO
 904                "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
 905                pci_mem_start, gapstart, gapsize);
 906 }
 907
 908 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
 909 {
 910         int i;
 911
 912         if (slot < 0 || slot >= e820.nr_map)
 913                 return -1;
 914         for (i = slot; i < e820.nr_map; i++) {
 915                 if (e820.map[i].type != E820_RAM)
 916                         continue;
 917                 break;
 918         }
 919         if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
 920                 return -1;
 921         *addr = e820.map[i].addr;
 922         *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
 923                 max_pfn << PAGE_SHIFT) - *addr;
 924         return i + 1;
 925 }