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 1.13 2004/03/22 00:31:08 ak Exp $
   5  */
   6 #include <linux/config.h>
   7 #include <linux/kernel.h>
   8 #include <linux/types.h>
   9 #include <linux/init.h>
  10 #include <linux/acpi.h>
  11 #include <linux/bootmem.h>
  12 #include <linux/ioport.h>
  13 #include <asm/page.h>
  14 #include <asm/e820.h>
  15 #include <asm/proto.h>
  16 #include <asm/acpi.h>
  17 #include <asm/apic.h>
  18 #include <asm/bootsetup.h>
  19 #include <asm/mpspec.h>
  20 #include <asm/io_apic.h>
  21
  22 extern unsigned long table_start, table_end;
  23 extern char _end[];
  24
  25 #ifdef  CONFIG_ACPI_BOOT
  26 extern acpi_interrupt_flags     acpi_sci_flags;
  27 #endif
  28
  29 extern struct resource code_resource, data_resource, vram_resource;
  30
  31 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
  32 static inline int bad_addr(unsigned long *addrp, unsigned long size)
  33 {
  34         unsigned long addr = *addrp, last = addr + size;
  35
  36         /* various gunk below that needed for SMP startup */
  37         if (addr < 7*PAGE_SIZE) {
  38                 *addrp = 7*PAGE_SIZE;
  39                 return 1;
  40         }
  41         /* direct mapping tables of the kernel */
  42         if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
  43                 *addrp = table_end << PAGE_SHIFT;
  44                 return 1;
  45         }
  46         /* initrd */
  47 #ifdef CONFIG_BLK_DEV_INITRD
  48         if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
  49             addr < INITRD_START+INITRD_SIZE) {
  50                 *addrp = INITRD_START + INITRD_SIZE;
  51                 return 1;
  52         }
  53 #endif
  54         /* kernel code + 640k memory hole (later should not be needed, but
  55            be paranoid for now) */
  56         if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
  57                 *addrp = __pa_symbol(&_end);
  58                 return 1;
  59         }
  60         /* XXX ramdisk image here? */
  61         return 0;
  62 }
  63
  64 int __init e820_mapped(unsigned long start, unsigned long end, int type)
  65 {
  66         int i;
  67         for (i = 0; i < e820.nr_map; i++) {
  68                 struct e820entry *ei = &e820.map[i];
  69                 if (type && ei->type != type)
  70                         continue;
  71                 if (ei->addr >= end || ei->addr + ei->size < start)
  72                         continue;
  73                 return 1;
  74         }
  75         return 0;
  76 }
  77
  78 /*
  79  * Find a free area in a specific range.
  80  */
  81 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
  82 {
  83         int i;
  84         for (i = 0; i < e820.nr_map; i++) {
  85                 struct e820entry *ei = &e820.map[i];
  86                 unsigned long addr = ei->addr, last;
  87                 if (ei->type != E820_RAM)
  88                         continue;
  89                 if (addr < start)
  90                         addr = start;
  91                 if (addr > ei->addr + ei->size)
  92                         continue;
  93                 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
  94                         ;
  95                 last = addr + size;
  96                 if (last > ei->addr + ei->size)
  97                         continue;
  98                 if (last > end)
  99                         continue;
 100                 return addr;
 101         }
 102         return -1UL;
 103 }
 104
 105 /*
 106  * Free bootmem based on the e820 table for a node.
 107  */
 108 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
 109 {
 110         int i;
 111         for (i = 0; i < e820.nr_map; i++) {
 112                 struct e820entry *ei = &e820.map[i];
 113                 unsigned long last, addr;
 114
 115                 if (ei->type != E820_RAM ||
 116                     ei->addr+ei->size <= start ||
 117                     ei->addr > end)
 118                         continue;
 119
 120                 addr = round_up(ei->addr, PAGE_SIZE);
 121                 if (addr < start)
 122                         addr = start;
 123
 124                 last = round_down(ei->addr + ei->size, PAGE_SIZE);
 125                 if (last >= end)
 126                         last = end;
 127
 128                 if (last > addr && last-addr >= PAGE_SIZE)
 129                         free_bootmem_node(pgdat, addr, last-addr);
 130         }
 131 }
 132
 133 /*
 134  * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
 135  * The direct mapping extends to end_pfn_map, so that we can directly access
 136  * ACPI and other tables without having to play with fixmaps.
 137  */
 138 unsigned long end_pfn_map;
 139
 140 /*
 141  * Last pfn which the user wants to use.
 142  */
 143 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
 144
 145 /*
 146  * Find the highest page frame number we have available
 147  */
 148
 149 void __init e820_end_of_ram(void)
 150 {
 151         int i;
 152         end_pfn = 0;
 153
 154         for (i = 0; i < e820.nr_map; i++) {
 155                 struct e820entry *ei = &e820.map[i];
 156                 unsigned long start, end;
 157
 158                 start = round_up(ei->addr, PAGE_SIZE);
 159                 end = round_down(ei->addr + ei->size, PAGE_SIZE);
 160                 if (start >= end)
 161                         continue;
 162                 if (ei->type == E820_RAM) {
 163                 if (end > end_pfn<<PAGE_SHIFT)
 164                         end_pfn = end>>PAGE_SHIFT;
 165                 } else {
 166                         if (end > end_pfn_map<<PAGE_SHIFT)
 167                                 end_pfn_map = end>>PAGE_SHIFT;
 168                 }
 169         }
 170
 171         if (end_pfn > end_pfn_map)
 172                 end_pfn_map = end_pfn;
 173         if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
 174                 end_pfn_map = MAXMEM>>PAGE_SHIFT;
 175         if (end_pfn > end_user_pfn)
 176                 end_pfn = end_user_pfn;
 177         if (end_pfn > end_pfn_map)
 178                 end_pfn = end_pfn_map;
 179 }
 180
 181 /*
 182  * Mark e820 reserved areas as busy for the resource manager.
 183  */
 184 void __init e820_reserve_resources(void)
 185 {
 186         int i;
 187         for (i = 0; i < e820.nr_map; i++) {
 188                 struct resource *res;
 189                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
 190                         continue;
 191                 res = alloc_bootmem_low(sizeof(struct resource));
 192                 switch (e820.map[i].type) {
 193                 case E820_RAM:  res->name = "System RAM"; break;
 194                 case E820_ACPI: res->name = "ACPI Tables"; break;
 195                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
 196                 default:        res->name = "reserved";
 197                 }
 198                 res->start = e820.map[i].addr;
 199                 res->end = res->start + e820.map[i].size - 1;
 200                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 201                 request_resource(&iomem_resource, res);
 202                 if (e820.map[i].type == E820_RAM) {
 203                         /*
 204                          *  We dont't know which RAM region contains kernel data,
 205                          *  so we try it repeatedly and let the resource manager
 206                          *  test it.
 207                          */
 208                         request_resource(res, &code_resource);
 209                         request_resource(res, &data_resource);
 210                 }
 211         }
 212 }
 213
 214 /*
 215  * Add a memory region to the kernel e820 map.
 216  */
 217 void __init add_memory_region(unsigned long start, unsigned long size, int type)
 218 {
 219         int x = e820.nr_map;
 220
 221         if (x == E820MAX) {
 222                 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 223                 return;
 224         }
 225
 226         e820.map[x].addr = start;
 227         e820.map[x].size = size;
 228         e820.map[x].type = type;
 229         e820.nr_map++;
 230 }
 231
 232 void __init e820_print_map(char *who)
 233 {
 234         int i;
 235
 236         for (i = 0; i < e820.nr_map; i++) {
 237                 printk(" %s: %016Lx - %016Lx ", who,
 238                         (unsigned long long) e820.map[i].addr,
 239                         (unsigned long long) (e820.map[i].addr + e820.map[i].size));
 240                 switch (e820.map[i].type) {
 241                 case E820_RAM:  printk("(usable)\n");
 242                                 break;
 243                 case E820_RESERVED:
 244                                 printk("(reserved)\n");
 245                                 break;
 246                 case E820_ACPI:
 247                                 printk("(ACPI data)\n");
 248                                 break;
 249                 case E820_NVS:
 250                                 printk("(ACPI NVS)\n");
 251                                 break;
 252                 default:        printk("type %u\n", e820.map[i].type);
 253                                 break;
 254                 }
 255         }
 256 }
 257
 258 /*
 259  * Sanitize the BIOS e820 map.
 260  *
 261  * Some e820 responses include overlapping entries.  The following
 262  * replaces the original e820 map with a new one, removing overlaps.
 263  *
 264  */
 265 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 266 {
 267         struct change_member {
 268                 struct e820entry *pbios; /* pointer to original bios entry */
 269                 unsigned long long addr; /* address for this change point */
 270         };
 271         static struct change_member change_point_list[2*E820MAX] __initdata;
 272         static struct change_member *change_point[2*E820MAX] __initdata;
 273         static struct e820entry *overlap_list[E820MAX] __initdata;
 274         static struct e820entry new_bios[E820MAX] __initdata;
 275         struct change_member *change_tmp;
 276         unsigned long current_type, last_type;
 277         unsigned long long last_addr;
 278         int chgidx, still_changing;
 279         int overlap_entries;
 280         int new_bios_entry;
 281         int old_nr, new_nr;
 282         int i;
 283
 284         /*
 285                 Visually we're performing the following (1,2,3,4 = memory types)...
 286
 287                 Sample memory map (w/overlaps):
 288                    ____22__________________
 289                    ______________________4_
 290                    ____1111________________
 291                    _44_____________________
 292                    11111111________________
 293                    ____________________33__
 294                    ___________44___________
 295                    __________33333_________
 296                    ______________22________
 297                    ___________________2222_
 298                    _________111111111______
 299                    _____________________11_
 300                    _________________4______
 301
 302                 Sanitized equivalent (no overlap):
 303                    1_______________________
 304                    _44_____________________
 305                    ___1____________________
 306                    ____22__________________
 307                    ______11________________
 308                    _________1______________
 309                    __________3_____________
 310                    ___________44___________
 311                    _____________33_________
 312                    _______________2________
 313                    ________________1_______
 314                    _________________4______
 315                    ___________________2____
 316                    ____________________33__
 317                    ______________________4_
 318         */
 319
 320         /* if there's only one memory region, don't bother */
 321         if (*pnr_map < 2)
 322                 return -1;
 323
 324         old_nr = *pnr_map;
 325
 326         /* bail out if we find any unreasonable addresses in bios map */
 327         for (i=0; i<old_nr; i++)
 328                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 329                         return -1;
 330
 331         /* create pointers for initial change-point information (for sorting) */
 332         for (i=0; i < 2*old_nr; i++)
 333                 change_point[i] = &change_point_list[i];
 334
 335         /* record all known change-points (starting and ending addresses) */
 336         chgidx = 0;
 337         for (i=0; i < old_nr; i++)      {
 338                 change_point[chgidx]->addr = biosmap[i].addr;
 339                 change_point[chgidx++]->pbios = &biosmap[i];
 340                 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 341                 change_point[chgidx++]->pbios = &biosmap[i];
 342         }
 343
 344         /* sort change-point list by memory addresses (low -> high) */
 345         still_changing = 1;
 346         while (still_changing)  {
 347                 still_changing = 0;
 348                 for (i=1; i < 2*old_nr; i++)  {
 349                         /* if <current_addr> > <last_addr>, swap */
 350                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 351                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 352                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 353                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 354                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 355                            )
 356                         {
 357                                 change_tmp = change_point[i];
 358                                 change_point[i] = change_point[i-1];
 359                                 change_point[i-1] = change_tmp;
 360                                 still_changing=1;
 361                         }
 362                 }
 363         }
 364
 365         /* create a new bios memory map, removing overlaps */
 366         overlap_entries=0;       /* number of entries in the overlap table */
 367         new_bios_entry=0;        /* index for creating new bios map entries */
 368         last_type = 0;           /* start with undefined memory type */
 369         last_addr = 0;           /* start with 0 as last starting address */
 370         /* loop through change-points, determining affect on the new bios map */
 371         for (chgidx=0; chgidx < 2*old_nr; chgidx++)
 372         {
 373                 /* keep track of all overlapping bios entries */
 374                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 375                 {
 376                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 377                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 378                 }
 379                 else
 380                 {
 381                         /* remove entry from list (order independent, so swap with last) */
 382                         for (i=0; i<overlap_entries; i++)
 383                         {
 384                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 385                                         overlap_list[i] = overlap_list[overlap_entries-1];
 386                         }
 387                         overlap_entries--;
 388                 }
 389                 /* if there are overlapping entries, decide which "type" to use */
 390                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 391                 current_type = 0;
 392                 for (i=0; i<overlap_entries; i++)
 393                         if (overlap_list[i]->type > current_type)
 394                                 current_type = overlap_list[i]->type;
 395                 /* continue building up new bios map based on this information */
 396                 if (current_type != last_type)  {
 397                         if (last_type != 0)      {
 398                                 new_bios[new_bios_entry].size =
 399                                         change_point[chgidx]->addr - last_addr;
 400                                 /* move forward only if the new size was non-zero */
 401                                 if (new_bios[new_bios_entry].size != 0)
 402                                         if (++new_bios_entry >= E820MAX)
 403                                                 break;  /* no more space left for new bios entries */
 404                         }
 405                         if (current_type != 0)  {
 406                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 407                                 new_bios[new_bios_entry].type = current_type;
 408                                 last_addr=change_point[chgidx]->addr;
 409                         }
 410                         last_type = current_type;
 411                 }
 412         }
 413         new_nr = new_bios_entry;   /* retain count for new bios entries */
 414
 415         /* copy new bios mapping into original location */
 416         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 417         *pnr_map = new_nr;
 418
 419         return 0;
 420 }
 421
 422 /*
 423  * Copy the BIOS e820 map into a safe place.
 424  *
 425  * Sanity-check it while we're at it..
 426  *
 427  * If we're lucky and live on a modern system, the setup code
 428  * will have given us a memory map that we can use to properly
 429  * set up memory.  If we aren't, we'll fake a memory map.
 430  *
 431  * We check to see that the memory map contains at least 2 elements
 432  * before we'll use it, because the detection code in setup.S may
 433  * not be perfect and most every PC known to man has two memory
 434  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 435  * thinkpad 560x, for example, does not cooperate with the memory
 436  * detection code.)
 437  */
 438 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 439 {
 440         /* Only one memory region (or negative)? Ignore it */
 441         if (nr_map < 2)
 442                 return -1;
 443
 444         do {
 445                 unsigned long start = biosmap->addr;
 446                 unsigned long size = biosmap->size;
 447                 unsigned long end = start + size;
 448                 unsigned long type = biosmap->type;
 449
 450                 /* Overflow in 64 bits? Ignore the memory map. */
 451                 if (start > end)
 452                         return -1;
 453
 454                 /*
 455                  * Some BIOSes claim RAM in the 640k - 1M region.
 456                  * Not right. Fix it up.
 457                  *
 458                  * This should be removed on Hammer which is supposed to not
 459                  * have non e820 covered ISA mappings there, but I had some strange
 460                  * problems so it stays for now.  -AK
 461                  */
 462                 if (type == E820_RAM) {
 463                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 464                                 if (start < 0xA0000ULL)
 465                                         add_memory_region(start, 0xA0000ULL-start, type);
 466                                 if (end <= 0x100000ULL)
 467                                         continue;
 468                                 start = 0x100000ULL;
 469                                 size = end - start;
 470                         }
 471                 }
 472
 473                 add_memory_region(start, size, type);
 474         } while (biosmap++,--nr_map);
 475         return 0;
 476 }
 477
 478 void __init setup_memory_region(void)
 479 {
 480         char *who = "BIOS-e820";
 481
 482         /*
 483          * Try to copy the BIOS-supplied E820-map.
 484          *
 485          * Otherwise fake a memory map; one section from 0k->640k,
 486          * the next section from 1mb->appropriate_mem_k
 487          */
 488         sanitize_e820_map(E820_MAP, &E820_MAP_NR);
 489         if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
 490                 unsigned long mem_size;
 491
 492                 /* compare results from other methods and take the greater */
 493                 if (ALT_MEM_K < EXT_MEM_K) {
 494                         mem_size = EXT_MEM_K;
 495                         who = "BIOS-88";
 496                 } else {
 497                         mem_size = ALT_MEM_K;
 498                         who = "BIOS-e801";
 499                 }
 500                 e820.nr_map = 0;
 501                 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
 502                 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
 503         }
 504         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 505         e820_print_map(who);
 506 }
 507
 508 extern char command_line[], saved_command_line[];
 509 extern int fallback_aper_order;
 510 extern int iommu_setup(char *opt);
 511
 512 void __init parse_mem_cmdline (char ** cmdline_p)
 513 {
 514         char c = ' ', *to = command_line, *from = COMMAND_LINE;
 515         int len = 0;
 516
 517         /* Save unparsed command line copy for /proc/cmdline */
 518         memcpy(saved_command_line, COMMAND_LINE, COMMAND_LINE_SIZE);
 519         saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
 520
 521         for (;;) {
 522                 if (c != ' ')
 523                         goto next;
 524
 525                 /*
 526                  * mem=XXX[kKmM] limits kernel memory to XXX+1MB
 527                  *
 528                  * It would be more logical to count from 0 instead of from
 529                  * HIGH_MEMORY, but we keep that for now for i386 compatibility. -AK
 530                  */
 531                 if (!memcmp(from, "mem=", 4)) {
 532                         /*
 533                          * No support for custom mapping like i386.
 534                          * The reason is that we need to read the e820 map
 535                          * anyways to handle the ACPI mappings in the
 536                          * direct map.
 537                          * Also on x86-64 there should be always a good e820
 538                          * map. This is only an upper limit, you cannot force
 539                          * usage of memory not in e820.
 540                                          */
 541                         end_user_pfn = memparse(from+4, &from) + HIGH_MEMORY;
 542                         end_user_pfn >>= PAGE_SHIFT;
 543                 }
 544 #ifdef CONFIG_GART_IOMMU
 545                 else if (!memcmp(from,"iommu=",6)) {
 546                         iommu_setup(from+6);
 547                 }
 548 #endif
 549 #ifdef  CONFIG_SMP
 550                 /*
 551                  * If the BIOS enumerates physical processors before logical,
 552                  * maxcpus=N at enumeration-time can be used to disable HT.
 553                  */
 554                 else if (!memcmp(from, "maxcpus=", 8)) {
 555                         extern unsigned int max_cpus;
 556
 557                         max_cpus = simple_strtoul(from + 8, NULL, 0);
 558                 }
 559 #endif
 560
 561 #ifdef  CONFIG_ACPI_BOOT
 562                 else if (!memcmp(from, "acpi=off", 8))
 563                         disable_acpi();
 564
 565                 /* acpi=strict disables out-of-spec workarounds */
 566                 else if (!memcmp(from, "acpi=strict", 11)) {
 567                         acpi_strict = 1;
 568                 }
 569
 570                 else if (!memcmp(from, "pci=noacpi", 10))
 571                         acpi_disable_pci();
 572                 else if (!memcmp(from, "acpi=noirq", 10))
 573                         acpi_noirq_set();
 574                 else if (!memcmp(from, "acpi_sci=edge", 13))
 575                         acpi_sci_flags.trigger =  1;
 576                 else if (!memcmp(from, "acpi_sci=level", 14))
 577                         acpi_sci_flags.trigger = 3;
 578                 else if (!memcmp(from, "acpi_sci=high", 13))
 579                         acpi_sci_flags.polarity = 1;
 580                 else if (!memcmp(from, "acpi_sci=low", 12))
 581                         acpi_sci_flags.polarity = 3;
 582 #endif
 583                 else if (!memcmp(from,"maxcpus=0",9)) {
 584                         disable_ioapic_setup();
 585                         apic_disabled = 1;
 586                 }
 587
 588                 else if (!memcmp(from, "noapic", 6))
 589                         disable_ioapic_setup();
 590                 else if (!memcmp(from, "nolocalapic", 11) || !memcmp(from,"nolapic",7))
 591                         apic_disabled = 1;
 592                 else if (!memcmp(from,"apic",4)) {
 593                         extern int ioapic_force;
 594                         ioapic_force = 1;
 595                         skip_ioapic_setup = 0;
 596                 }
 597                 else if (!memcmp(from, "noexec=", 7)) {
 598                         extern int nonx_setup(char *);
 599                         nonx_setup(from + 7);
 600                 }
 601         next:
 602                 c = *(from++);
 603                 if (!c)
 604                         break;
 605                 if (COMMAND_LINE_SIZE <= ++len)
 606                         break;
 607                 *(to++) = c;
 608         }
 609         *to = '\0';
 610         *cmdline_p = command_line;
 611 }