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 <asm/page.h>
  20 #include <asm/e820.h>
  21 #include <asm/proto.h>
  22 #include <asm/bootsetup.h>
  23
  24 extern char _end[];
  25
  26 /*
  27  * PFN of last memory page.
  28  */
  29 unsigned long end_pfn;
  30
  31 /*
  32  * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
  33  * The direct mapping extends to end_pfn_map, so that we can directly access
  34  * apertures, ACPI and other tables without having to play with fixmaps.
  35  */
  36 unsigned long end_pfn_map;
  37
  38 /*
  39  * Last pfn which the user wants to use.
  40  */
  41 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
  42
  43 extern struct resource code_resource, data_resource;
  44
  45 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
  46 static inline int bad_addr(unsigned long *addrp, unsigned long size)
  47 {
  48         unsigned long addr = *addrp, last = addr + size;
  49
  50         /* various gunk below that needed for SMP startup */
  51         if (addr < 0x8000) {
  52                 *addrp = 0x8000;
  53                 return 1;
  54         }
  55
  56         /* direct mapping tables of the kernel */
  57         if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
  58                 *addrp = table_end << PAGE_SHIFT;
  59                 return 1;
  60         }
  61
  62         /* initrd */
  63 #ifdef CONFIG_BLK_DEV_INITRD
  64         if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
  65             addr < INITRD_START+INITRD_SIZE) {
  66                 *addrp = INITRD_START + INITRD_SIZE;
  67                 return 1;
  68         }
  69 #endif
  70         /* kernel code + 640k memory hole (later should not be needed, but
  71            be paranoid for now) */
  72         if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
  73                 *addrp = __pa_symbol(&_end);
  74                 return 1;
  75         }
  76         /* XXX ramdisk image here? */
  77         return 0;
  78 }
  79
  80 int __init e820_mapped(unsigned long start, unsigned long end, unsigned type)
  81 {
  82         int i;
  83         for (i = 0; i < e820.nr_map; i++) {
  84                 struct e820entry *ei = &e820.map[i];
  85                 if (type && ei->type != type)
  86                         continue;
  87                 if (ei->addr >= end || ei->addr + ei->size < start)
  88                         continue;
  89                 return 1;
  90         }
  91         return 0;
  92 }
  93
  94 /*
  95  * Find a free area in a specific range.
  96  */
  97 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
  98 {
  99         int i;
 100         for (i = 0; i < e820.nr_map; i++) {
 101                 struct e820entry *ei = &e820.map[i];
 102                 unsigned long addr = ei->addr, last;
 103                 if (ei->type != E820_RAM)
 104                         continue;
 105                 if (addr < start)
 106                         addr = start;
 107                 if (addr > ei->addr + ei->size)
 108                         continue;
 109                 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
 110                         ;
 111                 last = addr + size;
 112                 if (last > ei->addr + ei->size)
 113                         continue;
 114                 if (last > end)
 115                         continue;
 116                 return addr;
 117         }
 118         return -1UL;
 119 }
 120
 121 /*
 122  * Free bootmem based on the e820 table for a node.
 123  */
 124 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
 125 {
 126         int i;
 127         for (i = 0; i < e820.nr_map; i++) {
 128                 struct e820entry *ei = &e820.map[i];
 129                 unsigned long last, addr;
 130
 131                 if (ei->type != E820_RAM ||
 132                     ei->addr+ei->size <= start ||
 133                     ei->addr > end)
 134                         continue;
 135
 136                 addr = round_up(ei->addr, PAGE_SIZE);
 137                 if (addr < start)
 138                         addr = start;
 139
 140                 last = round_down(ei->addr + ei->size, PAGE_SIZE);
 141                 if (last >= end)
 142                         last = end;
 143
 144                 if (last > addr && last-addr >= PAGE_SIZE)
 145                         free_bootmem_node(pgdat, addr, last-addr);
 146         }
 147 }
 148
 149 /*
 150  * Find the highest page frame number we have available
 151  */
 152 unsigned long __init e820_end_of_ram(void)
 153 {
 154         int i;
 155         unsigned long end_pfn = 0;
 156
 157         for (i = 0; i < e820.nr_map; i++) {
 158                 struct e820entry *ei = &e820.map[i];
 159                 unsigned long start, end;
 160
 161                 start = round_up(ei->addr, PAGE_SIZE);
 162                 end = round_down(ei->addr + ei->size, PAGE_SIZE);
 163                 if (start >= end)
 164                         continue;
 165                 if (ei->type == E820_RAM) {
 166                 if (end > end_pfn<<PAGE_SHIFT)
 167                         end_pfn = end>>PAGE_SHIFT;
 168                 } else {
 169                         if (end > end_pfn_map<<PAGE_SHIFT)
 170                                 end_pfn_map = end>>PAGE_SHIFT;
 171                 }
 172         }
 173
 174         if (end_pfn > end_pfn_map)
 175                 end_pfn_map = end_pfn;
 176         if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
 177                 end_pfn_map = MAXMEM>>PAGE_SHIFT;
 178         if (end_pfn > end_user_pfn)
 179                 end_pfn = end_user_pfn;
 180         if (end_pfn > end_pfn_map)
 181                 end_pfn = end_pfn_map;
 182
 183         return end_pfn;
 184 }
 185
 186 /*
 187  * Mark e820 reserved areas as busy for the resource manager.
 188  */
 189 void __init e820_reserve_resources(void)
 190 {
 191         int i;
 192         for (i = 0; i < e820.nr_map; i++) {
 193                 struct resource *res;
 194                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
 195                         continue;
 196                 res = alloc_bootmem_low(sizeof(struct resource));
 197                 switch (e820.map[i].type) {
 198                 case E820_RAM:  res->name = "System RAM"; break;
 199                 case E820_ACPI: res->name = "ACPI Tables"; break;
 200                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
 201                 default:        res->name = "reserved";
 202                 }
 203                 res->start = e820.map[i].addr;
 204                 res->end = res->start + e820.map[i].size - 1;
 205                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 206                 request_resource(&iomem_resource, res);
 207                 if (e820.map[i].type == E820_RAM) {
 208                         /*
 209                          *  We don't know which RAM region contains kernel data,
 210                          *  so we try it repeatedly and let the resource manager
 211                          *  test it.
 212                          */
 213                         request_resource(res, &code_resource);
 214                         request_resource(res, &data_resource);
 215                 }
 216         }
 217 }
 218
 219 /*
 220  * Add a memory region to the kernel e820 map.
 221  */
 222 void __init add_memory_region(unsigned long start, unsigned long size, int type)
 223 {
 224         int x = e820.nr_map;
 225
 226         if (x == E820MAX) {
 227                 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 228                 return;
 229         }
 230
 231         e820.map[x].addr = start;
 232         e820.map[x].size = size;
 233         e820.map[x].type = type;
 234         e820.nr_map++;
 235 }
 236
 237 void __init e820_print_map(char *who)
 238 {
 239         int i;
 240
 241         for (i = 0; i < e820.nr_map; i++) {
 242                 printk(" %s: %016Lx - %016Lx ", who,
 243                         (unsigned long long) e820.map[i].addr,
 244                         (unsigned long long) (e820.map[i].addr + e820.map[i].size));
 245                 switch (e820.map[i].type) {
 246                 case E820_RAM:  printk("(usable)\n");
 247                                 break;
 248                 case E820_RESERVED:
 249                                 printk("(reserved)\n");
 250                                 break;
 251                 case E820_ACPI:
 252                                 printk("(ACPI data)\n");
 253                                 break;
 254                 case E820_NVS:
 255                                 printk("(ACPI NVS)\n");
 256                                 break;
 257                 default:        printk("type %u\n", e820.map[i].type);
 258                                 break;
 259                 }
 260         }
 261 }
 262
 263 /*
 264  * Sanitize the BIOS e820 map.
 265  *
 266  * Some e820 responses include overlapping entries.  The following
 267  * replaces the original e820 map with a new one, removing overlaps.
 268  *
 269  */
 270 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 271 {
 272         struct change_member {
 273                 struct e820entry *pbios; /* pointer to original bios entry */
 274                 unsigned long long addr; /* address for this change point */
 275         };
 276         static struct change_member change_point_list[2*E820MAX] __initdata;
 277         static struct change_member *change_point[2*E820MAX] __initdata;
 278         static struct e820entry *overlap_list[E820MAX] __initdata;
 279         static struct e820entry new_bios[E820MAX] __initdata;
 280         struct change_member *change_tmp;
 281         unsigned long current_type, last_type;
 282         unsigned long long last_addr;
 283         int chgidx, still_changing;
 284         int overlap_entries;
 285         int new_bios_entry;
 286         int old_nr, new_nr, chg_nr;
 287         int i;
 288
 289         /*
 290                 Visually we're performing the following (1,2,3,4 = memory types)...
 291
 292                 Sample memory map (w/overlaps):
 293                    ____22__________________
 294                    ______________________4_
 295                    ____1111________________
 296                    _44_____________________
 297                    11111111________________
 298                    ____________________33__
 299                    ___________44___________
 300                    __________33333_________
 301                    ______________22________
 302                    ___________________2222_
 303                    _________111111111______
 304                    _____________________11_
 305                    _________________4______
 306
 307                 Sanitized equivalent (no overlap):
 308                    1_______________________
 309                    _44_____________________
 310                    ___1____________________
 311                    ____22__________________
 312                    ______11________________
 313                    _________1______________
 314                    __________3_____________
 315                    ___________44___________
 316                    _____________33_________
 317                    _______________2________
 318                    ________________1_______
 319                    _________________4______
 320                    ___________________2____
 321                    ____________________33__
 322                    ______________________4_
 323         */
 324
 325         /* if there's only one memory region, don't bother */
 326         if (*pnr_map < 2)
 327                 return -1;
 328
 329         old_nr = *pnr_map;
 330
 331         /* bail out if we find any unreasonable addresses in bios map */
 332         for (i=0; i<old_nr; i++)
 333                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 334                         return -1;
 335
 336         /* create pointers for initial change-point information (for sorting) */
 337         for (i=0; i < 2*old_nr; i++)
 338                 change_point[i] = &change_point_list[i];
 339
 340         /* record all known change-points (starting and ending addresses),
 341            omitting those that are for empty memory regions */
 342         chgidx = 0;
 343         for (i=0; i < old_nr; i++)      {
 344                 if (biosmap[i].size != 0) {
 345                         change_point[chgidx]->addr = biosmap[i].addr;
 346                         change_point[chgidx++]->pbios = &biosmap[i];
 347                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 348                         change_point[chgidx++]->pbios = &biosmap[i];
 349                 }
 350         }
 351         chg_nr = chgidx;
 352
 353         /* sort change-point list by memory addresses (low -> high) */
 354         still_changing = 1;
 355         while (still_changing)  {
 356                 still_changing = 0;
 357                 for (i=1; i < chg_nr; i++)  {
 358                         /* if <current_addr> > <last_addr>, swap */
 359                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 360                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 361                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 362                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 363                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 364                            )
 365                         {
 366                                 change_tmp = change_point[i];
 367                                 change_point[i] = change_point[i-1];
 368                                 change_point[i-1] = change_tmp;
 369                                 still_changing=1;
 370                         }
 371                 }
 372         }
 373
 374         /* create a new bios memory map, removing overlaps */
 375         overlap_entries=0;       /* number of entries in the overlap table */
 376         new_bios_entry=0;        /* index for creating new bios map entries */
 377         last_type = 0;           /* start with undefined memory type */
 378         last_addr = 0;           /* start with 0 as last starting address */
 379         /* loop through change-points, determining affect on the new bios map */
 380         for (chgidx=0; chgidx < chg_nr; chgidx++)
 381         {
 382                 /* keep track of all overlapping bios entries */
 383                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 384                 {
 385                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 386                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 387                 }
 388                 else
 389                 {
 390                         /* remove entry from list (order independent, so swap with last) */
 391                         for (i=0; i<overlap_entries; i++)
 392                         {
 393                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 394                                         overlap_list[i] = overlap_list[overlap_entries-1];
 395                         }
 396                         overlap_entries--;
 397                 }
 398                 /* if there are overlapping entries, decide which "type" to use */
 399                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 400                 current_type = 0;
 401                 for (i=0; i<overlap_entries; i++)
 402                         if (overlap_list[i]->type > current_type)
 403                                 current_type = overlap_list[i]->type;
 404                 /* continue building up new bios map based on this information */
 405                 if (current_type != last_type)  {
 406                         if (last_type != 0)      {
 407                                 new_bios[new_bios_entry].size =
 408                                         change_point[chgidx]->addr - last_addr;
 409                                 /* move forward only if the new size was non-zero */
 410                                 if (new_bios[new_bios_entry].size != 0)
 411                                         if (++new_bios_entry >= E820MAX)
 412                                                 break;  /* no more space left for new bios entries */
 413                         }
 414                         if (current_type != 0)  {
 415                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 416                                 new_bios[new_bios_entry].type = current_type;
 417                                 last_addr=change_point[chgidx]->addr;
 418                         }
 419                         last_type = current_type;
 420                 }
 421         }
 422         new_nr = new_bios_entry;   /* retain count for new bios entries */
 423
 424         /* copy new bios mapping into original location */
 425         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 426         *pnr_map = new_nr;
 427
 428         return 0;
 429 }
 430
 431 /*
 432  * Copy the BIOS e820 map into a safe place.
 433  *
 434  * Sanity-check it while we're at it..
 435  *
 436  * If we're lucky and live on a modern system, the setup code
 437  * will have given us a memory map that we can use to properly
 438  * set up memory.  If we aren't, we'll fake a memory map.
 439  *
 440  * We check to see that the memory map contains at least 2 elements
 441  * before we'll use it, because the detection code in setup.S may
 442  * not be perfect and most every PC known to man has two memory
 443  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 444  * thinkpad 560x, for example, does not cooperate with the memory
 445  * detection code.)
 446  */
 447 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 448 {
 449         /* Only one memory region (or negative)? Ignore it */
 450         if (nr_map < 2)
 451                 return -1;
 452
 453         do {
 454                 unsigned long start = biosmap->addr;
 455                 unsigned long size = biosmap->size;
 456                 unsigned long end = start + size;
 457                 unsigned long type = biosmap->type;
 458
 459                 /* Overflow in 64 bits? Ignore the memory map. */
 460                 if (start > end)
 461                         return -1;
 462
 463                 /*
 464                  * Some BIOSes claim RAM in the 640k - 1M region.
 465                  * Not right. Fix it up.
 466                  *
 467                  * This should be removed on Hammer which is supposed to not
 468                  * have non e820 covered ISA mappings there, but I had some strange
 469                  * problems so it stays for now.  -AK
 470                  */
 471                 if (type == E820_RAM) {
 472                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 473                                 if (start < 0xA0000ULL)
 474                                         add_memory_region(start, 0xA0000ULL-start, type);
 475                                 if (end <= 0x100000ULL)
 476                                         continue;
 477                                 start = 0x100000ULL;
 478                                 size = end - start;
 479                         }
 480                 }
 481
 482                 add_memory_region(start, size, type);
 483         } while (biosmap++,--nr_map);
 484         return 0;
 485 }
 486
 487 void __init setup_memory_region(void)
 488 {
 489         char *who = "BIOS-e820";
 490
 491         /*
 492          * Try to copy the BIOS-supplied E820-map.
 493          *
 494          * Otherwise fake a memory map; one section from 0k->640k,
 495          * the next section from 1mb->appropriate_mem_k
 496          */
 497         sanitize_e820_map(E820_MAP, &E820_MAP_NR);
 498         if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
 499                 unsigned long mem_size;
 500
 501                 /* compare results from other methods and take the greater */
 502                 if (ALT_MEM_K < EXT_MEM_K) {
 503                         mem_size = EXT_MEM_K;
 504                         who = "BIOS-88";
 505                 } else {
 506                         mem_size = ALT_MEM_K;
 507                         who = "BIOS-e801";
 508                 }
 509
 510                 e820.nr_map = 0;
 511                 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
 512                 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
 513         }
 514         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 515         e820_print_map(who);
 516 }
 517
 518 void __init parse_memopt(char *p, char **from)
 519 {
 520         end_user_pfn = memparse(p, from);
 521         end_user_pfn >>= PAGE_SHIFT;
 522 }
 523
 524 unsigned long pci_mem_start = 0xaeedbabe;
 525
 526 /*
 527  * Search for the biggest gap in the low 32 bits of the e820
 528  * memory space.  We pass this space to PCI to assign MMIO resources
 529  * for hotplug or unconfigured devices in.
 530  * Hopefully the BIOS let enough space left.
 531  */
 532 __init void e820_setup_gap(void)
 533 {
 534         unsigned long gapstart, gapsize;
 535         unsigned long last;
 536         int i;
 537         int found = 0;
 538
 539         last = 0x100000000ull;
 540         gapstart = 0x10000000;
 541         gapsize = 0x400000;
 542         i = e820.nr_map;
 543         while (--i >= 0) {
 544                 unsigned long long start = e820.map[i].addr;
 545                 unsigned long long end = start + e820.map[i].size;
 546
 547                 /*
 548                  * Since "last" is at most 4GB, we know we'll
 549                  * fit in 32 bits if this condition is true
 550                  */
 551                 if (last > end) {
 552                         unsigned long gap = last - end;
 553
 554                         if (gap > gapsize) {
 555                                 gapsize = gap;
 556                                 gapstart = end;
 557                                 found = 1;
 558                         }
 559                 }
 560                 if (start < last)
 561                         last = start;
 562         }
 563
 564         if (!found) {
 565                 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
 566                 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
 567                        KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
 568         }
 569
 570         /*
 571          * Start allocating dynamic PCI memory a bit into the gap,
 572          * aligned up to the nearest megabyte.
 573          *
 574          * Question: should we try to pad it up a bit (do something
 575          * like " + (gapsize >> 3)" in there too?). We now have the
 576          * technology.
 577          */
 578         pci_mem_start = (gapstart + 0xfffff) & ~0xfffff;
 579
 580         printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
 581                 pci_mem_start, gapstart, gapsize);
 582 }