arch/x86_64/kernel/e820.c

   1 /*
   2  * Handle the memory map.
   3  * The functions here do the job until bootmem takes over.
   4  * $Id: e820.c,v 1.4 2002/09/19 19:25:32 ak Exp $
   5  *
   6  *  Getting sanitize_e820_map() in sync with i386 version by applying change:
   7  *  -  Provisions for empty E820 memory regions (reported by certain BIOSes).
   8  *     Alex Achenbach <xela@slit.de>, December 2002.
   9  *  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  10  *
  11  */
  12 #include <linux/config.h>
  13 #include <linux/kernel.h>
  14 #include <linux/types.h>
  15 #include <linux/init.h>
  16 #include <linux/bootmem.h>
  17 #include <linux/ioport.h>
  18 #include <linux/string.h>
  19 #include <linux/kexec.h>
  20 #include <linux/module.h>
  21
  22 #include <asm/page.h>
  23 #include <asm/e820.h>
  24 #include <asm/proto.h>
  25 #include <asm/bootsetup.h>
  26
  27 extern char _end[];
  28
  29 /*
  30  * PFN of last memory page.
  31  */
  32 unsigned long end_pfn;
  33 EXPORT_SYMBOL(end_pfn);
  34
  35 /*
  36  * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
  37  * The direct mapping extends to end_pfn_map, so that we can directly access
  38  * apertures, ACPI and other tables without having to play with fixmaps.
  39  */
  40 unsigned long end_pfn_map;
  41
  42 /*
  43  * Last pfn which the user wants to use.
  44  */
  45 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
  46
  47 extern struct resource code_resource, data_resource;
  48
  49 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
  50 static inline int bad_addr(unsigned long *addrp, unsigned long size)
  51 {
  52         unsigned long addr = *addrp, last = addr + size;
  53
  54         /* various gunk below that needed for SMP startup */
  55         if (addr < 0x8000) {
  56                 *addrp = 0x8000;
  57                 return 1;
  58         }
  59
  60         /* direct mapping tables of the kernel */
  61         if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
  62                 *addrp = table_end << PAGE_SHIFT;
  63                 return 1;
  64         }
  65
  66         /* initrd */
  67 #ifdef CONFIG_BLK_DEV_INITRD
  68         if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
  69             addr < INITRD_START+INITRD_SIZE) {
  70                 *addrp = INITRD_START + INITRD_SIZE;
  71                 return 1;
  72         }
  73 #endif
  74         /* kernel code + 640k memory hole (later should not be needed, but
  75            be paranoid for now) */
  76         if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
  77                 *addrp = __pa_symbol(&_end);
  78                 return 1;
  79         }
  80         /* XXX ramdisk image here? */
  81         return 0;
  82 }
  83
  84 int __init e820_mapped(unsigned long start, unsigned long end, unsigned type)
  85 {
  86         int i;
  87         for (i = 0; i < e820.nr_map; i++) {
  88                 struct e820entry *ei = &e820.map[i];
  89                 if (type && ei->type != type)
  90                         continue;
  91                 if (ei->addr >= end || ei->addr + ei->size <= start)
  92                         continue;
  93                 return 1;
  94         }
  95         return 0;
  96 }
  97
  98 /*
  99  * Find a free area in a specific range.
 100  */
 101 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
 102 {
 103         int i;
 104         for (i = 0; i < e820.nr_map; i++) {
 105                 struct e820entry *ei = &e820.map[i];
 106                 unsigned long addr = ei->addr, last;
 107                 if (ei->type != E820_RAM)
 108                         continue;
 109                 if (addr < start)
 110                         addr = start;
 111                 if (addr > ei->addr + ei->size)
 112                         continue;
 113                 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
 114                         ;
 115                 last = addr + size;
 116                 if (last > ei->addr + ei->size)
 117                         continue;
 118                 if (last > end)
 119                         continue;
 120                 return addr;
 121         }
 122         return -1UL;
 123 }
 124
 125 /*
 126  * Free bootmem based on the e820 table for a node.
 127  */
 128 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
 129 {
 130         int i;
 131         for (i = 0; i < e820.nr_map; i++) {
 132                 struct e820entry *ei = &e820.map[i];
 133                 unsigned long last, addr;
 134
 135                 if (ei->type != E820_RAM ||
 136                     ei->addr+ei->size <= start ||
 137                     ei->addr >= end)
 138                         continue;
 139
 140                 addr = round_up(ei->addr, PAGE_SIZE);
 141                 if (addr < start)
 142                         addr = start;
 143
 144                 last = round_down(ei->addr + ei->size, PAGE_SIZE);
 145                 if (last >= end)
 146                         last = end;
 147
 148                 if (last > addr && last-addr >= PAGE_SIZE)
 149                         free_bootmem_node(pgdat, addr, last-addr);
 150         }
 151 }
 152
 153 /*
 154  * Find the highest page frame number we have available
 155  */
 156 unsigned long __init e820_end_of_ram(void)
 157 {
 158         int i;
 159         unsigned long end_pfn = 0;
 160
 161         for (i = 0; i < e820.nr_map; i++) {
 162                 struct e820entry *ei = &e820.map[i];
 163                 unsigned long start, end;
 164
 165                 start = round_up(ei->addr, PAGE_SIZE);
 166                 end = round_down(ei->addr + ei->size, PAGE_SIZE);
 167                 if (start >= end)
 168                         continue;
 169                 if (ei->type == E820_RAM) {
 170                 if (end > end_pfn<<PAGE_SHIFT)
 171                         end_pfn = end>>PAGE_SHIFT;
 172                 } else {
 173                         if (end > end_pfn_map<<PAGE_SHIFT)
 174                                 end_pfn_map = end>>PAGE_SHIFT;
 175                 }
 176         }
 177
 178         if (end_pfn > end_pfn_map)
 179                 end_pfn_map = end_pfn;
 180         if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
 181                 end_pfn_map = MAXMEM>>PAGE_SHIFT;
 182         if (end_pfn > end_user_pfn)
 183                 end_pfn = end_user_pfn;
 184         if (end_pfn > end_pfn_map)
 185                 end_pfn = end_pfn_map;
 186
 187         return end_pfn;
 188 }
 189
 190 /*
 191  * Compute how much memory is missing in a range.
 192  * Unlike the other functions in this file the arguments are in page numbers.
 193  */
 194 unsigned long __init
 195 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
 196 {
 197         unsigned long ram = 0;
 198         unsigned long start = start_pfn << PAGE_SHIFT;
 199         unsigned long end = end_pfn << PAGE_SHIFT;
 200         int i;
 201         for (i = 0; i < e820.nr_map; i++) {
 202                 struct e820entry *ei = &e820.map[i];
 203                 unsigned long last, addr;
 204
 205                 if (ei->type != E820_RAM ||
 206                     ei->addr+ei->size <= start ||
 207                     ei->addr >= end)
 208                         continue;
 209
 210                 addr = round_up(ei->addr, PAGE_SIZE);
 211                 if (addr < start)
 212                         addr = start;
 213
 214                 last = round_down(ei->addr + ei->size, PAGE_SIZE);
 215                 if (last >= end)
 216                         last = end;
 217
 218                 if (last > addr)
 219                         ram += last - addr;
 220         }
 221         return ((end - start) - ram) >> PAGE_SHIFT;
 222 }
 223
 224 /*
 225  * Mark e820 reserved areas as busy for the resource manager.
 226  */
 227 void __init e820_reserve_resources(void)
 228 {
 229         int i;
 230         for (i = 0; i < e820.nr_map; i++) {
 231                 struct resource *res;
 232                 res = alloc_bootmem_low(sizeof(struct resource));
 233                 switch (e820.map[i].type) {
 234                 case E820_RAM:  res->name = "System RAM"; break;
 235                 case E820_ACPI: res->name = "ACPI Tables"; break;
 236                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
 237                 default:        res->name = "reserved";
 238                 }
 239                 res->start = e820.map[i].addr;
 240                 res->end = res->start + e820.map[i].size - 1;
 241                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 242                 request_resource(&iomem_resource, res);
 243                 if (e820.map[i].type == E820_RAM) {
 244                         /*
 245                          *  We don't know which RAM region contains kernel data,
 246                          *  so we try it repeatedly and let the resource manager
 247                          *  test it.
 248                          */
 249                         request_resource(res, &code_resource);
 250                         request_resource(res, &data_resource);
 251 #ifdef CONFIG_KEXEC
 252                         request_resource(res, &crashk_res);
 253 #endif
 254                 }
 255         }
 256 }
 257
 258 /*
 259  * Add a memory region to the kernel e820 map.
 260  */
 261 void __init add_memory_region(unsigned long start, unsigned long size, int type)
 262 {
 263         int x = e820.nr_map;
 264
 265         if (x == E820MAX) {
 266                 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 267                 return;
 268         }
 269
 270         e820.map[x].addr = start;
 271         e820.map[x].size = size;
 272         e820.map[x].type = type;
 273         e820.nr_map++;
 274 }
 275
 276 void __init e820_print_map(char *who)
 277 {
 278         int i;
 279
 280         for (i = 0; i < e820.nr_map; i++) {
 281                 printk(" %s: %016Lx - %016Lx ", who,
 282                         (unsigned long long) e820.map[i].addr,
 283                         (unsigned long long) (e820.map[i].addr + e820.map[i].size));
 284                 switch (e820.map[i].type) {
 285                 case E820_RAM:  printk("(usable)\n");
 286                                 break;
 287                 case E820_RESERVED:
 288                                 printk("(reserved)\n");
 289                                 break;
 290                 case E820_ACPI:
 291                                 printk("(ACPI data)\n");
 292                                 break;
 293                 case E820_NVS:
 294                                 printk("(ACPI NVS)\n");
 295                                 break;
 296                 default:        printk("type %u\n", e820.map[i].type);
 297                                 break;
 298                 }
 299         }
 300 }
 301
 302 /*
 303  * Sanitize the BIOS e820 map.
 304  *
 305  * Some e820 responses include overlapping entries.  The following
 306  * replaces the original e820 map with a new one, removing overlaps.
 307  *
 308  */
 309 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 310 {
 311         struct change_member {
 312                 struct e820entry *pbios; /* pointer to original bios entry */
 313                 unsigned long long addr; /* address for this change point */
 314         };
 315         static struct change_member change_point_list[2*E820MAX] __initdata;
 316         static struct change_member *change_point[2*E820MAX] __initdata;
 317         static struct e820entry *overlap_list[E820MAX] __initdata;
 318         static struct e820entry new_bios[E820MAX] __initdata;
 319         struct change_member *change_tmp;
 320         unsigned long current_type, last_type;
 321         unsigned long long last_addr;
 322         int chgidx, still_changing;
 323         int overlap_entries;
 324         int new_bios_entry;
 325         int old_nr, new_nr, chg_nr;
 326         int i;
 327
 328         /*
 329                 Visually we're performing the following (1,2,3,4 = memory types)...
 330
 331                 Sample memory map (w/overlaps):
 332                    ____22__________________
 333                    ______________________4_
 334                    ____1111________________
 335                    _44_____________________
 336                    11111111________________
 337                    ____________________33__
 338                    ___________44___________
 339                    __________33333_________
 340                    ______________22________
 341                    ___________________2222_
 342                    _________111111111______
 343                    _____________________11_
 344                    _________________4______
 345
 346                 Sanitized equivalent (no overlap):
 347                    1_______________________
 348                    _44_____________________
 349                    ___1____________________
 350                    ____22__________________
 351                    ______11________________
 352                    _________1______________
 353                    __________3_____________
 354                    ___________44___________
 355                    _____________33_________
 356                    _______________2________
 357                    ________________1_______
 358                    _________________4______
 359                    ___________________2____
 360                    ____________________33__
 361                    ______________________4_
 362         */
 363
 364         /* if there's only one memory region, don't bother */
 365         if (*pnr_map < 2)
 366                 return -1;
 367
 368         old_nr = *pnr_map;
 369
 370         /* bail out if we find any unreasonable addresses in bios map */
 371         for (i=0; i<old_nr; i++)
 372                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 373                         return -1;
 374
 375         /* create pointers for initial change-point information (for sorting) */
 376         for (i=0; i < 2*old_nr; i++)
 377                 change_point[i] = &change_point_list[i];
 378
 379         /* record all known change-points (starting and ending addresses),
 380            omitting those that are for empty memory regions */
 381         chgidx = 0;
 382         for (i=0; i < old_nr; i++)      {
 383                 if (biosmap[i].size != 0) {
 384                         change_point[chgidx]->addr = biosmap[i].addr;
 385                         change_point[chgidx++]->pbios = &biosmap[i];
 386                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 387                         change_point[chgidx++]->pbios = &biosmap[i];
 388                 }
 389         }
 390         chg_nr = chgidx;
 391
 392         /* sort change-point list by memory addresses (low -> high) */
 393         still_changing = 1;
 394         while (still_changing)  {
 395                 still_changing = 0;
 396                 for (i=1; i < chg_nr; i++)  {
 397                         /* if <current_addr> > <last_addr>, swap */
 398                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 399                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 400                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 401                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 402                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 403                            )
 404                         {
 405                                 change_tmp = change_point[i];
 406                                 change_point[i] = change_point[i-1];
 407                                 change_point[i-1] = change_tmp;
 408                                 still_changing=1;
 409                         }
 410                 }
 411         }
 412
 413         /* create a new bios memory map, removing overlaps */
 414         overlap_entries=0;       /* number of entries in the overlap table */
 415         new_bios_entry=0;        /* index for creating new bios map entries */
 416         last_type = 0;           /* start with undefined memory type */
 417         last_addr = 0;           /* start with 0 as last starting address */
 418         /* loop through change-points, determining affect on the new bios map */
 419         for (chgidx=0; chgidx < chg_nr; chgidx++)
 420         {
 421                 /* keep track of all overlapping bios entries */
 422                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 423                 {
 424                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 425                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 426                 }
 427                 else
 428                 {
 429                         /* remove entry from list (order independent, so swap with last) */
 430                         for (i=0; i<overlap_entries; i++)
 431                         {
 432                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 433                                         overlap_list[i] = overlap_list[overlap_entries-1];
 434                         }
 435                         overlap_entries--;
 436                 }
 437                 /* if there are overlapping entries, decide which "type" to use */
 438                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 439                 current_type = 0;
 440                 for (i=0; i<overlap_entries; i++)
 441                         if (overlap_list[i]->type > current_type)
 442                                 current_type = overlap_list[i]->type;
 443                 /* continue building up new bios map based on this information */
 444                 if (current_type != last_type)  {
 445                         if (last_type != 0)      {
 446                                 new_bios[new_bios_entry].size =
 447                                         change_point[chgidx]->addr - last_addr;
 448                                 /* move forward only if the new size was non-zero */
 449                                 if (new_bios[new_bios_entry].size != 0)
 450                                         if (++new_bios_entry >= E820MAX)
 451                                                 break;  /* no more space left for new bios entries */
 452                         }
 453                         if (current_type != 0)  {
 454                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 455                                 new_bios[new_bios_entry].type = current_type;
 456                                 last_addr=change_point[chgidx]->addr;
 457                         }
 458                         last_type = current_type;
 459                 }
 460         }
 461         new_nr = new_bios_entry;   /* retain count for new bios entries */
 462
 463         /* copy new bios mapping into original location */
 464         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 465         *pnr_map = new_nr;
 466
 467         return 0;
 468 }
 469
 470 /*
 471  * Copy the BIOS e820 map into a safe place.
 472  *
 473  * Sanity-check it while we're at it..
 474  *
 475  * If we're lucky and live on a modern system, the setup code
 476  * will have given us a memory map that we can use to properly
 477  * set up memory.  If we aren't, we'll fake a memory map.
 478  *
 479  * We check to see that the memory map contains at least 2 elements
 480  * before we'll use it, because the detection code in setup.S may
 481  * not be perfect and most every PC known to man has two memory
 482  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 483  * thinkpad 560x, for example, does not cooperate with the memory
 484  * detection code.)
 485  */
 486 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 487 {
 488         /* Only one memory region (or negative)? Ignore it */
 489         if (nr_map < 2)
 490                 return -1;
 491
 492         do {
 493                 unsigned long start = biosmap->addr;
 494                 unsigned long size = biosmap->size;
 495                 unsigned long end = start + size;
 496                 unsigned long type = biosmap->type;
 497
 498                 /* Overflow in 64 bits? Ignore the memory map. */
 499                 if (start > end)
 500                         return -1;
 501
 502                 /*
 503                  * Some BIOSes claim RAM in the 640k - 1M region.
 504                  * Not right. Fix it up.
 505                  *
 506                  * This should be removed on Hammer which is supposed to not
 507                  * have non e820 covered ISA mappings there, but I had some strange
 508                  * problems so it stays for now.  -AK
 509                  */
 510                 if (type == E820_RAM) {
 511                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 512                                 if (start < 0xA0000ULL)
 513                                         add_memory_region(start, 0xA0000ULL-start, type);
 514                                 if (end <= 0x100000ULL)
 515                                         continue;
 516                                 start = 0x100000ULL;
 517                                 size = end - start;
 518                         }
 519                 }
 520
 521                 add_memory_region(start, size, type);
 522         } while (biosmap++,--nr_map);
 523         return 0;
 524 }
 525
 526 void __init setup_memory_region(void)
 527 {
 528         char *who = "BIOS-e820";
 529
 530         /*
 531          * Try to copy the BIOS-supplied E820-map.
 532          *
 533          * Otherwise fake a memory map; one section from 0k->640k,
 534          * the next section from 1mb->appropriate_mem_k
 535          */
 536         sanitize_e820_map(E820_MAP, &E820_MAP_NR);
 537         if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
 538                 unsigned long mem_size;
 539
 540                 /* compare results from other methods and take the greater */
 541                 if (ALT_MEM_K < EXT_MEM_K) {
 542                         mem_size = EXT_MEM_K;
 543                         who = "BIOS-88";
 544                 } else {
 545                         mem_size = ALT_MEM_K;
 546                         who = "BIOS-e801";
 547                 }
 548
 549                 e820.nr_map = 0;
 550                 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
 551                 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
 552         }
 553         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 554         e820_print_map(who);
 555 }
 556
 557 void __init parse_memopt(char *p, char **from)
 558 {
 559         end_user_pfn = memparse(p, from);
 560         end_user_pfn >>= PAGE_SHIFT;
 561 }
 562
 563 unsigned long pci_mem_start = 0xaeedbabe;
 564
 565 /*
 566  * Search for the biggest gap in the low 32 bits of the e820
 567  * memory space.  We pass this space to PCI to assign MMIO resources
 568  * for hotplug or unconfigured devices in.
 569  * Hopefully the BIOS let enough space left.
 570  */
 571 __init void e820_setup_gap(void)
 572 {
 573         unsigned long gapstart, gapsize, round;
 574         unsigned long last;
 575         int i;
 576         int found = 0;
 577
 578         last = 0x100000000ull;
 579         gapstart = 0x10000000;
 580         gapsize = 0x400000;
 581         i = e820.nr_map;
 582         while (--i >= 0) {
 583                 unsigned long long start = e820.map[i].addr;
 584                 unsigned long long end = start + e820.map[i].size;
 585
 586                 /*
 587                  * Since "last" is at most 4GB, we know we'll
 588                  * fit in 32 bits if this condition is true
 589                  */
 590                 if (last > end) {
 591                         unsigned long gap = last - end;
 592
 593                         if (gap > gapsize) {
 594                                 gapsize = gap;
 595                                 gapstart = end;
 596                                 found = 1;
 597                         }
 598                 }
 599                 if (start < last)
 600                         last = start;
 601         }
 602
 603         if (!found) {
 604                 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
 605                 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
 606                        KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
 607         }
 608
 609         /*
 610          * See how much we want to round up: start off with
 611          * rounding to the next 1MB area.
 612          */
 613         round = 0x100000;
 614         while ((gapsize >> 4) > round)
 615                 round += round;
 616         /* Fun with two's complement */
 617         pci_mem_start = (gapstart + round) & -round;
 618
 619         printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
 620                 pci_mem_start, gapstart, gapsize);
 621 }