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