2 * linux/arch/i386/kernel/setup.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
8 * Memory region support
9 * David Parsons <orc@pell.chi.il.us>, July-August 1999
11 * Added E820 sanitization routine (removes overlapping memory regions);
12 * Brian Moyle <bmoyle@mvista.com>, February 2001
14 * Moved CPU detection code to cpu/${cpu}.c
15 * Patrick Mochel <mochel@osdl.org>, March 2002
17 * Provisions for empty E820 memory regions (reported by certain BIOSes).
18 * Alex Achenbach <xela@slit.de>, December 2002.
23 * This file handles the architecture-dependent parts of initialization
26 #include <linux/sched.h>
28 #include <linux/tty.h>
29 #include <linux/ioport.h>
30 #include <linux/acpi.h>
31 #include <linux/apm_bios.h>
32 #include <linux/initrd.h>
33 #include <linux/bootmem.h>
34 #include <linux/seq_file.h>
35 #include <linux/console.h>
36 #include <linux/mca.h>
37 #include <linux/root_dev.h>
38 #include <linux/highmem.h>
39 #include <linux/module.h>
40 #include <linux/efi.h>
41 #include <linux/init.h>
42 #include <linux/edd.h>
43 #include <linux/nodemask.h>
44 #include <video/edid.h>
46 #include <asm/mpspec.h>
47 #include <asm/setup.h>
48 #include <asm/arch_hooks.h>
49 #include <asm/sections.h>
50 #include <asm/io_apic.h>
53 #include <asm/crash_dump.h>
54 #include "setup_arch_pre.h"
55 #include <bios_ebda.h>
57 /* This value is set up by the early boot code to point to the value
58 immediately after the boot time page tables. It contains a *physical*
59 address, and must not be in the .bss segment! */
60 unsigned long init_pg_tables_end __initdata = ~0UL;
62 int disable_pse __initdata = 0;
63 unsigned int dump_enabled;
71 EXPORT_SYMBOL(efi_enabled);
74 /* cpu data as detected by the assembly code in head.S */
75 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
76 /* common cpu data for all cpus */
77 struct cpuinfo_x86 boot_cpu_data = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
79 unsigned long mmu_cr4_features;
81 #ifdef CONFIG_ACPI_INTERPRETER
82 int acpi_disabled = 0;
84 int acpi_disabled = 1;
86 EXPORT_SYMBOL(acpi_disabled);
88 #ifdef CONFIG_ACPI_BOOT
89 int __initdata acpi_force = 0;
90 extern acpi_interrupt_flags acpi_sci_flags;
93 /* for MCA, but anyone else can use it if they want */
94 unsigned int machine_id;
95 unsigned int machine_submodel_id;
96 unsigned int BIOS_revision;
97 unsigned int mca_pentium_flag;
99 /* For PCI or other memory-mapped resources */
100 unsigned long pci_mem_start = 0x10000000;
102 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
105 /* user-defined highmem size */
106 static unsigned int highmem_pages = -1;
111 struct drive_info_struct { char dummy[32]; } drive_info;
112 struct screen_info screen_info;
113 struct apm_info apm_info;
114 struct sys_desc_table_struct {
115 unsigned short length;
116 unsigned char table[0];
118 struct edid_info edid_info;
119 struct ist_info ist_info;
122 extern void early_cpu_init(void);
123 extern void dmi_scan_machine(void);
124 extern void generic_apic_probe(char *);
125 extern int root_mountflags;
127 unsigned long saved_videomode;
129 #define RAMDISK_IMAGE_START_MASK 0x07FF
130 #define RAMDISK_PROMPT_FLAG 0x8000
131 #define RAMDISK_LOAD_FLAG 0x4000
133 static char command_line[COMMAND_LINE_SIZE];
135 unsigned char __initdata boot_params[PARAM_SIZE];
137 static struct resource data_resource = {
138 .name = "Kernel data",
141 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
144 static struct resource code_resource = {
145 .name = "Kernel code",
148 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
151 static struct resource system_rom_resource = {
152 .name = "System ROM",
155 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
158 static struct resource extension_rom_resource = {
159 .name = "Extension ROM",
162 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
165 static struct resource adapter_rom_resources[] = { {
166 .name = "Adapter ROM",
169 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
171 .name = "Adapter ROM",
174 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
176 .name = "Adapter ROM",
179 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
181 .name = "Adapter ROM",
184 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
186 .name = "Adapter ROM",
189 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
191 .name = "Adapter ROM",
194 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
197 #define ADAPTER_ROM_RESOURCES \
198 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
200 static struct resource video_rom_resource = {
204 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
207 static struct resource video_ram_resource = {
208 .name = "Video RAM area",
211 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
214 static struct resource standard_io_resources[] = { {
218 .flags = IORESOURCE_BUSY | IORESOURCE_IO
223 .flags = IORESOURCE_BUSY | IORESOURCE_IO
228 .flags = IORESOURCE_BUSY | IORESOURCE_IO
233 .flags = IORESOURCE_BUSY | IORESOURCE_IO
238 .flags = IORESOURCE_BUSY | IORESOURCE_IO
240 .name = "dma page reg",
243 .flags = IORESOURCE_BUSY | IORESOURCE_IO
248 .flags = IORESOURCE_BUSY | IORESOURCE_IO
253 .flags = IORESOURCE_BUSY | IORESOURCE_IO
258 .flags = IORESOURCE_BUSY | IORESOURCE_IO
261 #define STANDARD_IO_RESOURCES \
262 (sizeof standard_io_resources / sizeof standard_io_resources[0])
264 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
266 static int __init romchecksum(unsigned char *rom, unsigned long length)
268 unsigned char *p, sum = 0;
270 for (p = rom; p < rom + length; p++)
275 static void __init probe_roms(void)
277 unsigned long start, length, upper;
282 upper = adapter_rom_resources[0].start;
283 for (start = video_rom_resource.start; start < upper; start += 2048) {
284 rom = isa_bus_to_virt(start);
285 if (!romsignature(rom))
288 video_rom_resource.start = start;
290 /* 0 < length <= 0x7f * 512, historically */
291 length = rom[2] * 512;
293 /* if checksum okay, trust length byte */
294 if (length && romchecksum(rom, length))
295 video_rom_resource.end = start + length - 1;
297 request_resource(&iomem_resource, &video_rom_resource);
301 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
306 request_resource(&iomem_resource, &system_rom_resource);
307 upper = system_rom_resource.start;
309 /* check for extension rom (ignore length byte!) */
310 rom = isa_bus_to_virt(extension_rom_resource.start);
311 if (romsignature(rom)) {
312 length = extension_rom_resource.end - extension_rom_resource.start + 1;
313 if (romchecksum(rom, length)) {
314 request_resource(&iomem_resource, &extension_rom_resource);
315 upper = extension_rom_resource.start;
319 /* check for adapter roms on 2k boundaries */
320 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
321 rom = isa_bus_to_virt(start);
322 if (!romsignature(rom))
325 /* 0 < length <= 0x7f * 512, historically */
326 length = rom[2] * 512;
328 /* but accept any length that fits if checksum okay */
329 if (!length || start + length > upper || !romchecksum(rom, length))
332 adapter_rom_resources[i].start = start;
333 adapter_rom_resources[i].end = start + length - 1;
334 request_resource(&iomem_resource, &adapter_rom_resources[i]);
336 start = adapter_rom_resources[i++].end & ~2047UL;
340 static void __init limit_regions(unsigned long long size)
342 unsigned long long current_addr = 0;
346 for (i = 0; i < memmap.nr_map; i++) {
347 current_addr = memmap.map[i].phys_addr +
348 (memmap.map[i].num_pages << 12);
349 if (memmap.map[i].type == EFI_CONVENTIONAL_MEMORY) {
350 if (current_addr >= size) {
351 memmap.map[i].num_pages -=
352 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
353 memmap.nr_map = i + 1;
359 for (i = 0; i < e820.nr_map; i++) {
360 if (e820.map[i].type == E820_RAM) {
361 current_addr = e820.map[i].addr + e820.map[i].size;
362 if (current_addr >= size) {
363 e820.map[i].size -= current_addr-size;
371 static void __init add_memory_region(unsigned long long start,
372 unsigned long long size, int type)
380 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
384 e820.map[x].addr = start;
385 e820.map[x].size = size;
386 e820.map[x].type = type;
389 } /* add_memory_region */
393 static void __init print_memory_map(char *who)
397 for (i = 0; i < e820.nr_map; i++) {
398 printk(" %s: %016Lx - %016Lx ", who,
400 e820.map[i].addr + e820.map[i].size);
401 switch (e820.map[i].type) {
402 case E820_RAM: printk("(usable)\n");
405 printk("(reserved)\n");
408 printk("(ACPI data)\n");
411 printk("(ACPI NVS)\n");
413 default: printk("type %lu\n", e820.map[i].type);
420 * Sanitize the BIOS e820 map.
422 * Some e820 responses include overlapping entries. The following
423 * replaces the original e820 map with a new one, removing overlaps.
426 struct change_member {
427 struct e820entry *pbios; /* pointer to original bios entry */
428 unsigned long long addr; /* address for this change point */
430 static struct change_member change_point_list[2*E820MAX] __initdata;
431 static struct change_member *change_point[2*E820MAX] __initdata;
432 static struct e820entry *overlap_list[E820MAX] __initdata;
433 static struct e820entry new_bios[E820MAX] __initdata;
435 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
437 struct change_member *change_tmp;
438 unsigned long current_type, last_type;
439 unsigned long long last_addr;
440 int chgidx, still_changing;
443 int old_nr, new_nr, chg_nr;
447 Visually we're performing the following (1,2,3,4 = memory types)...
449 Sample memory map (w/overlaps):
450 ____22__________________
451 ______________________4_
452 ____1111________________
453 _44_____________________
454 11111111________________
455 ____________________33__
456 ___________44___________
457 __________33333_________
458 ______________22________
459 ___________________2222_
460 _________111111111______
461 _____________________11_
462 _________________4______
464 Sanitized equivalent (no overlap):
465 1_______________________
466 _44_____________________
467 ___1____________________
468 ____22__________________
469 ______11________________
470 _________1______________
471 __________3_____________
472 ___________44___________
473 _____________33_________
474 _______________2________
475 ________________1_______
476 _________________4______
477 ___________________2____
478 ____________________33__
479 ______________________4_
482 /* if there's only one memory region, don't bother */
488 /* bail out if we find any unreasonable addresses in bios map */
489 for (i=0; i<old_nr; i++)
490 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
493 /* create pointers for initial change-point information (for sorting) */
494 for (i=0; i < 2*old_nr; i++)
495 change_point[i] = &change_point_list[i];
497 /* record all known change-points (starting and ending addresses),
498 omitting those that are for empty memory regions */
500 for (i=0; i < old_nr; i++) {
501 if (biosmap[i].size != 0) {
502 change_point[chgidx]->addr = biosmap[i].addr;
503 change_point[chgidx++]->pbios = &biosmap[i];
504 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
505 change_point[chgidx++]->pbios = &biosmap[i];
508 chg_nr = chgidx; /* true number of change-points */
510 /* sort change-point list by memory addresses (low -> high) */
512 while (still_changing) {
514 for (i=1; i < chg_nr; i++) {
515 /* if <current_addr> > <last_addr>, swap */
516 /* or, if current=<start_addr> & last=<end_addr>, swap */
517 if ((change_point[i]->addr < change_point[i-1]->addr) ||
518 ((change_point[i]->addr == change_point[i-1]->addr) &&
519 (change_point[i]->addr == change_point[i]->pbios->addr) &&
520 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
523 change_tmp = change_point[i];
524 change_point[i] = change_point[i-1];
525 change_point[i-1] = change_tmp;
531 /* create a new bios memory map, removing overlaps */
532 overlap_entries=0; /* number of entries in the overlap table */
533 new_bios_entry=0; /* index for creating new bios map entries */
534 last_type = 0; /* start with undefined memory type */
535 last_addr = 0; /* start with 0 as last starting address */
536 /* loop through change-points, determining affect on the new bios map */
537 for (chgidx=0; chgidx < chg_nr; chgidx++)
539 /* keep track of all overlapping bios entries */
540 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
542 /* add map entry to overlap list (> 1 entry implies an overlap) */
543 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
547 /* remove entry from list (order independent, so swap with last) */
548 for (i=0; i<overlap_entries; i++)
550 if (overlap_list[i] == change_point[chgidx]->pbios)
551 overlap_list[i] = overlap_list[overlap_entries-1];
555 /* if there are overlapping entries, decide which "type" to use */
556 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
558 for (i=0; i<overlap_entries; i++)
559 if (overlap_list[i]->type > current_type)
560 current_type = overlap_list[i]->type;
561 /* continue building up new bios map based on this information */
562 if (current_type != last_type) {
563 if (last_type != 0) {
564 new_bios[new_bios_entry].size =
565 change_point[chgidx]->addr - last_addr;
566 /* move forward only if the new size was non-zero */
567 if (new_bios[new_bios_entry].size != 0)
568 if (++new_bios_entry >= E820MAX)
569 break; /* no more space left for new bios entries */
571 if (current_type != 0) {
572 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
573 new_bios[new_bios_entry].type = current_type;
574 last_addr=change_point[chgidx]->addr;
576 last_type = current_type;
579 new_nr = new_bios_entry; /* retain count for new bios entries */
581 /* copy new bios mapping into original location */
582 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
589 * Copy the BIOS e820 map into a safe place.
591 * Sanity-check it while we're at it..
593 * If we're lucky and live on a modern system, the setup code
594 * will have given us a memory map that we can use to properly
595 * set up memory. If we aren't, we'll fake a memory map.
597 * We check to see that the memory map contains at least 2 elements
598 * before we'll use it, because the detection code in setup.S may
599 * not be perfect and most every PC known to man has two memory
600 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
601 * thinkpad 560x, for example, does not cooperate with the memory
604 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
606 /* Only one memory region (or negative)? Ignore it */
611 unsigned long long start = biosmap->addr;
612 unsigned long long size = biosmap->size;
613 unsigned long long end = start + size;
614 unsigned long type = biosmap->type;
616 /* Overflow in 64 bits? Ignore the memory map. */
621 * Some BIOSes claim RAM in the 640k - 1M region.
622 * Not right. Fix it up.
624 if (type == E820_RAM) {
625 if (start < 0x100000ULL && end > 0xA0000ULL) {
626 if (start < 0xA0000ULL)
627 add_memory_region(start, 0xA0000ULL-start, type);
628 if (end <= 0x100000ULL)
634 add_memory_region(start, size, type);
635 } while (biosmap++,--nr_map);
639 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
641 #ifdef CONFIG_EDD_MODULE
645 * copy_edd() - Copy the BIOS EDD information
646 * from boot_params into a safe place.
649 static inline void copy_edd(void)
651 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
652 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
653 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
654 edd.edd_info_nr = EDD_NR;
657 static inline void copy_edd(void)
663 * Do NOT EVER look at the BIOS memory size location.
664 * It does not work on many machines.
666 #define LOWMEMSIZE() (0x9f000)
668 static void __init parse_cmdline_early (char ** cmdline_p)
670 char c = ' ', *to = command_line, *from = saved_command_line;
674 /* Save unparsed command line copy for /proc/cmdline */
675 saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
681 * "mem=nopentium" disables the 4MB page tables.
682 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
683 * to <mem>, overriding the bios size.
684 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
685 * <start> to <start>+<mem>, overriding the bios size.
687 * HPA tells me bootloaders need to parse mem=, so no new
688 * option should be mem= [also see Documentation/i386/boot.txt]
690 if (!memcmp(from, "mem=", 4)) {
691 if (to != command_line)
693 if (!memcmp(from+4, "nopentium", 9)) {
695 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
698 /* If the user specifies memory size, we
699 * limit the BIOS-provided memory map to
700 * that size. exactmap can be used to specify
701 * the exact map. mem=number can be used to
702 * trim the existing memory map.
704 unsigned long long mem_size;
706 mem_size = memparse(from+4, &from);
707 limit_regions(mem_size);
712 else if (!memcmp(from, "memmap=", 7)) {
713 if (to != command_line)
715 if (!memcmp(from+7, "exactmap", 8)) {
716 /* If we are doing a crash dump, we
717 * still need to know the real mem
725 /* If the user specifies memory size, we
726 * limit the BIOS-provided memory map to
727 * that size. exactmap can be used to specify
728 * the exact map. mem=number can be used to
729 * trim the existing memory map.
731 unsigned long long start_at, mem_size;
733 mem_size = memparse(from+7, &from);
735 start_at = memparse(from+1, &from);
736 add_memory_region(start_at, mem_size, E820_RAM);
737 } else if (*from == '#') {
738 start_at = memparse(from+1, &from);
739 add_memory_region(start_at, mem_size, E820_ACPI);
740 } else if (*from == '$') {
741 start_at = memparse(from+1, &from);
742 add_memory_region(start_at, mem_size, E820_RESERVED);
744 limit_regions(mem_size);
750 else if (!memcmp(from, "noexec=", 7))
751 noexec_setup(from + 7);
754 #ifdef CONFIG_X86_SMP
756 * If the BIOS enumerates physical processors before logical,
757 * maxcpus=N at enumeration-time can be used to disable HT.
759 else if (!memcmp(from, "maxcpus=", 8)) {
760 extern unsigned int maxcpus;
762 maxcpus = simple_strtoul(from + 8, NULL, 0);
766 #ifdef CONFIG_ACPI_BOOT
767 /* "acpi=off" disables both ACPI table parsing and interpreter */
768 else if (!memcmp(from, "acpi=off", 8)) {
772 /* acpi=force to over-ride black-list */
773 else if (!memcmp(from, "acpi=force", 10)) {
779 /* acpi=strict disables out-of-spec workarounds */
780 else if (!memcmp(from, "acpi=strict", 11)) {
784 /* Limit ACPI just to boot-time to enable HT */
785 else if (!memcmp(from, "acpi=ht", 7)) {
791 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
792 else if (!memcmp(from, "pci=noacpi", 10)) {
795 /* "acpi=noirq" disables ACPI interrupt routing */
796 else if (!memcmp(from, "acpi=noirq", 10)) {
800 else if (!memcmp(from, "acpi_sci=edge", 13))
801 acpi_sci_flags.trigger = 1;
803 else if (!memcmp(from, "acpi_sci=level", 14))
804 acpi_sci_flags.trigger = 3;
806 else if (!memcmp(from, "acpi_sci=high", 13))
807 acpi_sci_flags.polarity = 1;
809 else if (!memcmp(from, "acpi_sci=low", 12))
810 acpi_sci_flags.polarity = 3;
812 #ifdef CONFIG_X86_IO_APIC
813 else if (!memcmp(from, "acpi_skip_timer_override", 24))
814 acpi_skip_timer_override = 1;
817 #ifdef CONFIG_X86_LOCAL_APIC
818 /* disable IO-APIC */
819 else if (!memcmp(from, "noapic", 6))
820 disable_ioapic_setup();
821 #endif /* CONFIG_X86_LOCAL_APIC */
822 #endif /* CONFIG_ACPI_BOOT */
825 * highmem=size forces highmem to be exactly 'size' bytes.
826 * This works even on boxes that have no highmem otherwise.
827 * This also works to reduce highmem size on bigger boxes.
829 else if (!memcmp(from, "highmem=", 8))
830 highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
832 if (!memcmp(from, "dump", 4))
836 * vmalloc=size forces the vmalloc area to be exactly 'size'
837 * bytes. This can be used to increase (or decrease) the
838 * vmalloc area - the default is 128m.
840 else if (!memcmp(from, "vmalloc=", 8))
841 __VMALLOC_RESERVE = memparse(from+8, &from);
847 if (COMMAND_LINE_SIZE <= ++len)
852 *cmdline_p = command_line;
854 printk(KERN_INFO "user-defined physical RAM map:\n");
855 print_memory_map("user");
860 * Callback for efi_memory_walk.
863 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
865 unsigned long *max_pfn = arg, pfn;
868 pfn = PFN_UP(end -1);
877 * Find the highest page frame number we have available
879 void __init find_max_pfn(void)
885 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
889 for (i = 0; i < e820.nr_map; i++) {
890 unsigned long start, end;
892 if (e820.map[i].type != E820_RAM)
894 start = PFN_UP(e820.map[i].addr);
895 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
904 * Determine low and high memory ranges:
906 unsigned long __init find_max_low_pfn(void)
908 unsigned long max_low_pfn;
910 max_low_pfn = max_pfn;
911 if (max_low_pfn > MAXMEM_PFN) {
912 if (highmem_pages == -1)
913 highmem_pages = max_pfn - MAXMEM_PFN;
914 if (highmem_pages + MAXMEM_PFN < max_pfn)
915 max_pfn = MAXMEM_PFN + highmem_pages;
916 if (highmem_pages + MAXMEM_PFN > max_pfn) {
917 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
920 max_low_pfn = MAXMEM_PFN;
921 #ifndef CONFIG_HIGHMEM
922 /* Maximum memory usable is what is directly addressable */
923 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
925 if (max_pfn > MAX_NONPAE_PFN)
926 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
928 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
929 max_pfn = MAXMEM_PFN;
930 #else /* !CONFIG_HIGHMEM */
931 #ifndef CONFIG_X86_PAE
932 if (max_pfn > MAX_NONPAE_PFN) {
933 max_pfn = MAX_NONPAE_PFN;
934 printk(KERN_WARNING "Warning only 4GB will be used.\n");
935 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
937 #endif /* !CONFIG_X86_PAE */
938 #endif /* !CONFIG_HIGHMEM */
940 if (highmem_pages == -1)
942 #ifdef CONFIG_HIGHMEM
943 if (highmem_pages >= max_pfn) {
944 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
948 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
949 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
952 max_low_pfn -= highmem_pages;
956 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
963 * Free all available memory for boot time allocation. Used
964 * as a callback function by efi_memory_walk()
968 free_available_memory(unsigned long start, unsigned long end, void *arg)
970 /* check max_low_pfn */
971 if (start >= ((max_low_pfn + 1) << PAGE_SHIFT))
973 if (end >= ((max_low_pfn + 1) << PAGE_SHIFT))
974 end = (max_low_pfn + 1) << PAGE_SHIFT;
976 free_bootmem(start, end - start);
981 * Register fully available low RAM pages with the bootmem allocator.
983 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
988 efi_memmap_walk(free_available_memory, NULL);
991 for (i = 0; i < e820.nr_map; i++) {
992 unsigned long curr_pfn, last_pfn, size;
994 * Reserve usable low memory
996 if (e820.map[i].type != E820_RAM)
999 * We are rounding up the start address of usable memory:
1001 curr_pfn = PFN_UP(e820.map[i].addr);
1002 if (curr_pfn >= max_low_pfn)
1005 * ... and at the end of the usable range downwards:
1007 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1009 if (last_pfn > max_low_pfn)
1010 last_pfn = max_low_pfn;
1013 * .. finally, did all the rounding and playing
1014 * around just make the area go away?
1016 if (last_pfn <= curr_pfn)
1019 size = last_pfn - curr_pfn;
1020 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1025 * workaround for Dell systems that neglect to reserve EBDA
1027 static void __init reserve_ebda_region(void)
1030 addr = get_bios_ebda();
1032 reserve_bootmem(addr, PAGE_SIZE);
1035 #ifndef CONFIG_DISCONTIGMEM
1036 void __init setup_bootmem_allocator(void);
1037 static unsigned long __init setup_memory(void)
1040 * partially used pages are not usable - thus
1041 * we are rounding upwards:
1043 min_low_pfn = PFN_UP(init_pg_tables_end);
1047 max_low_pfn = find_max_low_pfn();
1049 #ifdef CONFIG_HIGHMEM
1050 highstart_pfn = highend_pfn = max_pfn;
1051 if (max_pfn > max_low_pfn) {
1052 highstart_pfn = max_low_pfn;
1054 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1055 pages_to_mb(highend_pfn - highstart_pfn));
1057 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1058 pages_to_mb(max_low_pfn));
1060 setup_bootmem_allocator();
1065 void __init zone_sizes_init(void)
1067 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1068 unsigned int max_dma, low;
1070 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1074 zones_size[ZONE_DMA] = low;
1076 zones_size[ZONE_DMA] = max_dma;
1077 zones_size[ZONE_NORMAL] = low - max_dma;
1078 #ifdef CONFIG_HIGHMEM
1079 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1082 free_area_init(zones_size);
1085 extern unsigned long setup_memory(void);
1086 extern void zone_sizes_init(void);
1087 #endif /* !CONFIG_DISCONTIGMEM */
1089 void __init setup_bootmem_allocator(void)
1091 unsigned long bootmap_size;
1093 * Initialize the boot-time allocator (with low memory only):
1095 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1097 register_bootmem_low_pages(max_low_pfn);
1100 * Reserve the bootmem bitmap itself as well. We do this in two
1101 * steps (first step was init_bootmem()) because this catches
1102 * the (very unlikely) case of us accidentally initializing the
1103 * bootmem allocator with an invalid RAM area.
1105 reserve_bootmem(HIGH_MEMORY, (PFN_PHYS(min_low_pfn) +
1106 bootmap_size + PAGE_SIZE-1) - (HIGH_MEMORY));
1109 * reserve physical page 0 - it's a special BIOS page on many boxes,
1110 * enabling clean reboots, SMP operation, laptop functions.
1112 reserve_bootmem(0, PAGE_SIZE);
1114 /* reserve EBDA region, it's a 4K region */
1115 reserve_ebda_region();
1117 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1118 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1119 unless you have no PS/2 mouse plugged in. */
1120 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1121 boot_cpu_data.x86 == 6)
1122 reserve_bootmem(0xa0000 - 4096, 4096);
1126 * But first pinch a few for the stack/trampoline stuff
1127 * FIXME: Don't need the extra page at 4K, but need to fix
1128 * trampoline before removing it. (see the GDT stuff)
1130 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1132 #ifdef CONFIG_ACPI_SLEEP
1134 * Reserve low memory region for sleep support.
1136 acpi_reserve_bootmem();
1138 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1140 * Find and reserve possible boot-time SMP configuration:
1145 #ifdef CONFIG_BLK_DEV_INITRD
1146 if (LOADER_TYPE && INITRD_START) {
1147 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1148 reserve_bootmem(INITRD_START, INITRD_SIZE);
1150 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1151 initrd_end = initrd_start+INITRD_SIZE;
1154 printk(KERN_ERR "initrd extends beyond end of memory "
1155 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1156 INITRD_START + INITRD_SIZE,
1157 max_low_pfn << PAGE_SHIFT);
1165 * The node 0 pgdat is initialized before all of these because
1166 * it's needed for bootmem. node>0 pgdats have their virtual
1167 * space allocated before the pagetables are in place to access
1168 * them, so they can't be cleared then.
1170 * This should all compile down to nothing when NUMA is off.
1172 void __init remapped_pgdat_init(void)
1176 for_each_online_node(nid) {
1178 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1183 * Request address space for all standard RAM and ROM resources
1184 * and also for regions reported as reserved by the e820.
1187 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1192 for (i = 0; i < e820.nr_map; i++) {
1193 struct resource *res;
1194 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1196 res = alloc_bootmem_low(sizeof(struct resource));
1197 switch (e820.map[i].type) {
1198 case E820_RAM: res->name = "System RAM"; break;
1199 case E820_ACPI: res->name = "ACPI Tables"; break;
1200 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1201 default: res->name = "reserved";
1203 res->start = e820.map[i].addr;
1204 res->end = res->start + e820.map[i].size - 1;
1205 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1206 request_resource(&iomem_resource, res);
1207 if (e820.map[i].type == E820_RAM) {
1209 * We don't know which RAM region contains kernel data,
1210 * so we try it repeatedly and let the resource manager
1213 request_resource(res, code_resource);
1214 request_resource(res, data_resource);
1220 * Request address space for all standard resources
1222 static void __init register_memory(void)
1224 unsigned long gapstart, gapsize;
1225 unsigned long long last;
1229 efi_initialize_iomem_resources(&code_resource, &data_resource);
1231 legacy_init_iomem_resources(&code_resource, &data_resource);
1233 /* EFI systems may still have VGA */
1234 request_resource(&iomem_resource, &video_ram_resource);
1236 /* request I/O space for devices used on all i[345]86 PCs */
1237 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1238 request_resource(&ioport_resource, &standard_io_resources[i]);
1241 * Search for the bigest gap in the low 32 bits of the e820
1244 last = 0x100000000ull;
1245 gapstart = 0x10000000;
1249 unsigned long long start = e820.map[i].addr;
1250 unsigned long long end = start + e820.map[i].size;
1253 * Since "last" is at most 4GB, we know we'll
1254 * fit in 32 bits if this condition is true
1257 unsigned long gap = last - end;
1259 if (gap > gapsize) {
1269 * Start allocating dynamic PCI memory a bit into the gap,
1270 * aligned up to the nearest megabyte.
1272 * Question: should we try to pad it up a bit (do something
1273 * like " + (gapsize >> 3)" in there too?). We now have the
1276 pci_mem_start = (gapstart + 0xfffff) & ~0xfffff;
1278 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1279 pci_mem_start, gapstart, gapsize);
1282 /* Use inline assembly to define this because the nops are defined
1283 as inline assembly strings in the include files and we cannot
1284 get them easily into strings. */
1285 asm("\t.data\nintelnops: "
1286 GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
1287 GENERIC_NOP7 GENERIC_NOP8);
1288 asm("\t.data\nk8nops: "
1289 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
1291 asm("\t.data\nk7nops: "
1292 K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
1295 extern unsigned char intelnops[], k8nops[], k7nops[];
1296 static unsigned char *intel_nops[ASM_NOP_MAX+1] = {
1301 intelnops + 1 + 2 + 3,
1302 intelnops + 1 + 2 + 3 + 4,
1303 intelnops + 1 + 2 + 3 + 4 + 5,
1304 intelnops + 1 + 2 + 3 + 4 + 5 + 6,
1305 intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1307 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
1313 k8nops + 1 + 2 + 3 + 4,
1314 k8nops + 1 + 2 + 3 + 4 + 5,
1315 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
1316 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1318 static unsigned char *k7_nops[ASM_NOP_MAX+1] = {
1324 k7nops + 1 + 2 + 3 + 4,
1325 k7nops + 1 + 2 + 3 + 4 + 5,
1326 k7nops + 1 + 2 + 3 + 4 + 5 + 6,
1327 k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1331 unsigned char **noptable;
1333 { X86_FEATURE_K8, k8_nops },
1334 { X86_FEATURE_K7, k7_nops },
1338 /* Replace instructions with better alternatives for this CPU type.
1340 This runs before SMP is initialized to avoid SMP problems with
1341 self modifying code. This implies that assymetric systems where
1342 APs have less capabilities than the boot processor are not handled.
1343 In this case boot with "noreplacement". */
1344 void apply_alternatives(void *start, void *end)
1346 struct alt_instr *a;
1348 unsigned char **noptable = intel_nops;
1349 for (i = 0; noptypes[i].cpuid >= 0; i++) {
1350 if (boot_cpu_has(noptypes[i].cpuid)) {
1351 noptable = noptypes[i].noptable;
1355 for (a = start; (void *)a < end; a++) {
1356 if (!boot_cpu_has(a->cpuid))
1358 BUG_ON(a->replacementlen > a->instrlen);
1359 memcpy(a->instr, a->replacement, a->replacementlen);
1360 diff = a->instrlen - a->replacementlen;
1361 /* Pad the rest with nops */
1362 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
1364 if (k > ASM_NOP_MAX)
1366 memcpy(a->instr + i, noptable[k], k);
1371 static int no_replacement __initdata = 0;
1373 void __init alternative_instructions(void)
1375 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
1378 apply_alternatives(__alt_instructions, __alt_instructions_end);
1381 static int __init noreplacement_setup(char *s)
1387 __setup("noreplacement", noreplacement_setup);
1389 static char * __init machine_specific_memory_setup(void);
1392 static void set_mca_bus(int x)
1397 static void set_mca_bus(int x) { }
1401 * Determine if we were loaded by an EFI loader. If so, then we have also been
1402 * passed the efi memmap, systab, etc., so we should use these data structures
1403 * for initialization. Note, the efi init code path is determined by the
1404 * global efi_enabled. This allows the same kernel image to be used on existing
1405 * systems (with a traditional BIOS) as well as on EFI systems.
1407 void __init setup_arch(char **cmdline_p)
1409 unsigned long max_low_pfn;
1411 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1412 pre_setup_arch_hook();
1416 * FIXME: This isn't an official loader_type right
1417 * now but does currently work with elilo.
1418 * If we were configured as an EFI kernel, check to make
1419 * sure that we were loaded correctly from elilo and that
1420 * the system table is valid. If not, then initialize normally.
1423 if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1427 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1428 drive_info = DRIVE_INFO;
1429 screen_info = SCREEN_INFO;
1430 edid_info = EDID_INFO;
1431 apm_info.bios = APM_BIOS_INFO;
1432 ist_info = IST_INFO;
1433 saved_videomode = VIDEO_MODE;
1434 if( SYS_DESC_TABLE.length != 0 ) {
1435 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1436 machine_id = SYS_DESC_TABLE.table[0];
1437 machine_submodel_id = SYS_DESC_TABLE.table[1];
1438 BIOS_revision = SYS_DESC_TABLE.table[2];
1440 bootloader_type = LOADER_TYPE;
1442 #ifdef CONFIG_BLK_DEV_RAM
1443 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1444 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1445 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1451 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1452 print_memory_map(machine_specific_memory_setup());
1457 if (!MOUNT_ROOT_RDONLY)
1458 root_mountflags &= ~MS_RDONLY;
1459 init_mm.start_code = (unsigned long) _text;
1460 init_mm.end_code = (unsigned long) _etext;
1461 init_mm.end_data = (unsigned long) _edata;
1462 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1464 code_resource.start = virt_to_phys(_text);
1465 code_resource.end = virt_to_phys(_etext)-1;
1466 data_resource.start = virt_to_phys(_etext);
1467 data_resource.end = virt_to_phys(_edata)-1;
1469 parse_cmdline_early(cmdline_p);
1471 max_low_pfn = setup_memory();
1474 * NOTE: before this point _nobody_ is allowed to allocate
1475 * any memory using the bootmem allocator. Although the
1476 * alloctor is now initialised only the first 8Mb of the kernel
1477 * virtual address space has been mapped. All allocations before
1478 * paging_init() has completed must use the alloc_bootmem_low_pages()
1479 * variant (which allocates DMA'able memory) and care must be taken
1480 * not to exceed the 8Mb limit.
1484 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1487 remapped_pgdat_init();
1491 * NOTE: at this point the bootmem allocator is fully available.
1494 #ifdef CONFIG_EARLY_PRINTK
1496 char *s = strstr(*cmdline_p, "earlyprintk=");
1498 extern void setup_early_printk(char *);
1500 setup_early_printk(s);
1501 printk("early console enabled\n");
1509 #ifdef CONFIG_X86_GENERICARCH
1510 generic_apic_probe(*cmdline_p);
1515 #ifdef CONFIG_ACPI_BOOT
1517 * Parse the ACPI tables for possible boot-time SMP configuration.
1519 acpi_boot_table_init();
1523 #ifdef CONFIG_X86_LOCAL_APIC
1524 if (smp_found_config)
1531 #if defined(CONFIG_VGA_CONSOLE)
1532 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1533 conswitchp = &vga_con;
1534 #elif defined(CONFIG_DUMMY_CONSOLE)
1535 conswitchp = &dummy_con;
1540 #include "setup_arch_post.h"
1544 * c-file-style:"k&r"