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/config.h>
27 #include <linux/sched.h>
29 #include <linux/mmzone.h>
30 #include <linux/tty.h>
31 #include <linux/ioport.h>
32 #include <linux/acpi.h>
33 #include <linux/apm_bios.h>
34 #include <linux/initrd.h>
35 #include <linux/bootmem.h>
36 #include <linux/seq_file.h>
37 #include <linux/console.h>
38 #include <linux/mca.h>
39 #include <linux/root_dev.h>
40 #include <linux/highmem.h>
41 #include <linux/module.h>
42 #include <linux/efi.h>
43 #include <linux/init.h>
44 #include <linux/edd.h>
45 #include <linux/nodemask.h>
46 #include <linux/kexec.h>
47 #include <linux/crash_dump.h>
49 #include <video/edid.h>
53 #include <asm/mpspec.h>
54 #include <asm/setup.h>
55 #include <asm/arch_hooks.h>
56 #include <asm/sections.h>
57 #include <asm/io_apic.h>
60 #include "setup_arch_pre.h"
61 #include <bios_ebda.h>
63 /* Forward Declaration. */
64 void __init find_max_pfn(void);
66 /* This value is set up by the early boot code to point to the value
67 immediately after the boot time page tables. It contains a *physical*
68 address, and must not be in the .bss segment! */
69 unsigned long init_pg_tables_end __initdata = ~0UL;
71 int disable_pse __devinitdata = 0;
79 EXPORT_SYMBOL(efi_enabled);
82 /* cpu data as detected by the assembly code in head.S */
83 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
84 /* common cpu data for all cpus */
85 struct cpuinfo_x86 boot_cpu_data = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
86 EXPORT_SYMBOL(boot_cpu_data);
88 unsigned long mmu_cr4_features;
90 #ifdef CONFIG_ACPI_INTERPRETER
91 int acpi_disabled = 0;
93 int acpi_disabled = 1;
95 EXPORT_SYMBOL(acpi_disabled);
97 #ifdef CONFIG_ACPI_BOOT
98 int __initdata acpi_force = 0;
99 extern acpi_interrupt_flags acpi_sci_flags;
102 /* for MCA, but anyone else can use it if they want */
103 unsigned int machine_id;
105 EXPORT_SYMBOL(machine_id);
107 unsigned int machine_submodel_id;
108 unsigned int BIOS_revision;
109 unsigned int mca_pentium_flag;
111 /* For PCI or other memory-mapped resources */
112 unsigned long pci_mem_start = 0x10000000;
114 EXPORT_SYMBOL(pci_mem_start);
117 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
120 /* user-defined highmem size */
121 static unsigned int highmem_pages = -1;
126 struct drive_info_struct { char dummy[32]; } drive_info;
127 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
128 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
129 EXPORT_SYMBOL(drive_info);
131 struct screen_info screen_info;
133 EXPORT_SYMBOL(screen_info);
135 struct apm_info apm_info;
136 EXPORT_SYMBOL(apm_info);
137 struct sys_desc_table_struct {
138 unsigned short length;
139 unsigned char table[0];
141 struct edid_info edid_info;
142 struct ist_info ist_info;
143 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
144 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
145 EXPORT_SYMBOL(ist_info);
149 extern void early_cpu_init(void);
150 extern void dmi_scan_machine(void);
151 extern void generic_apic_probe(char *);
152 extern int root_mountflags;
154 unsigned long saved_videomode;
156 #define RAMDISK_IMAGE_START_MASK 0x07FF
157 #define RAMDISK_PROMPT_FLAG 0x8000
158 #define RAMDISK_LOAD_FLAG 0x4000
160 static char command_line[COMMAND_LINE_SIZE];
162 unsigned char __initdata boot_params[PARAM_SIZE];
164 static struct resource data_resource = {
165 .name = "Kernel data",
168 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
171 static struct resource code_resource = {
172 .name = "Kernel code",
175 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
178 static struct resource system_rom_resource = {
179 .name = "System ROM",
182 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
185 static struct resource extension_rom_resource = {
186 .name = "Extension ROM",
189 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
192 static struct resource adapter_rom_resources[] = { {
193 .name = "Adapter ROM",
196 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
198 .name = "Adapter ROM",
201 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
203 .name = "Adapter ROM",
206 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
208 .name = "Adapter ROM",
211 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
213 .name = "Adapter ROM",
216 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
218 .name = "Adapter ROM",
221 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
224 #define ADAPTER_ROM_RESOURCES \
225 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
227 static struct resource video_rom_resource = {
231 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
234 static struct resource video_ram_resource = {
235 .name = "Video RAM area",
238 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
241 static struct resource standard_io_resources[] = { {
245 .flags = IORESOURCE_BUSY | IORESOURCE_IO
250 .flags = IORESOURCE_BUSY | IORESOURCE_IO
255 .flags = IORESOURCE_BUSY | IORESOURCE_IO
260 .flags = IORESOURCE_BUSY | IORESOURCE_IO
265 .flags = IORESOURCE_BUSY | IORESOURCE_IO
267 .name = "dma page reg",
270 .flags = IORESOURCE_BUSY | IORESOURCE_IO
275 .flags = IORESOURCE_BUSY | IORESOURCE_IO
280 .flags = IORESOURCE_BUSY | IORESOURCE_IO
285 .flags = IORESOURCE_BUSY | IORESOURCE_IO
288 #define STANDARD_IO_RESOURCES \
289 (sizeof standard_io_resources / sizeof standard_io_resources[0])
291 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
293 static int __init romchecksum(unsigned char *rom, unsigned long length)
295 unsigned char *p, sum = 0;
297 for (p = rom; p < rom + length; p++)
302 static void __init probe_roms(void)
304 unsigned long start, length, upper;
309 upper = adapter_rom_resources[0].start;
310 for (start = video_rom_resource.start; start < upper; start += 2048) {
311 rom = isa_bus_to_virt(start);
312 if (!romsignature(rom))
315 video_rom_resource.start = start;
317 /* 0 < length <= 0x7f * 512, historically */
318 length = rom[2] * 512;
320 /* if checksum okay, trust length byte */
321 if (length && romchecksum(rom, length))
322 video_rom_resource.end = start + length - 1;
324 request_resource(&iomem_resource, &video_rom_resource);
328 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
333 request_resource(&iomem_resource, &system_rom_resource);
334 upper = system_rom_resource.start;
336 /* check for extension rom (ignore length byte!) */
337 rom = isa_bus_to_virt(extension_rom_resource.start);
338 if (romsignature(rom)) {
339 length = extension_rom_resource.end - extension_rom_resource.start + 1;
340 if (romchecksum(rom, length)) {
341 request_resource(&iomem_resource, &extension_rom_resource);
342 upper = extension_rom_resource.start;
346 /* check for adapter roms on 2k boundaries */
347 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
348 rom = isa_bus_to_virt(start);
349 if (!romsignature(rom))
352 /* 0 < length <= 0x7f * 512, historically */
353 length = rom[2] * 512;
355 /* but accept any length that fits if checksum okay */
356 if (!length || start + length > upper || !romchecksum(rom, length))
359 adapter_rom_resources[i].start = start;
360 adapter_rom_resources[i].end = start + length - 1;
361 request_resource(&iomem_resource, &adapter_rom_resources[i]);
363 start = adapter_rom_resources[i++].end & ~2047UL;
367 static void __init limit_regions(unsigned long long size)
369 unsigned long long current_addr = 0;
373 for (i = 0; i < memmap.nr_map; i++) {
374 current_addr = memmap.map[i].phys_addr +
375 (memmap.map[i].num_pages << 12);
376 if (memmap.map[i].type == EFI_CONVENTIONAL_MEMORY) {
377 if (current_addr >= size) {
378 memmap.map[i].num_pages -=
379 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
380 memmap.nr_map = i + 1;
386 for (i = 0; i < e820.nr_map; i++) {
387 if (e820.map[i].type == E820_RAM) {
388 current_addr = e820.map[i].addr + e820.map[i].size;
389 if (current_addr >= size) {
390 e820.map[i].size -= current_addr-size;
398 static void __init add_memory_region(unsigned long long start,
399 unsigned long long size, int type)
407 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
411 e820.map[x].addr = start;
412 e820.map[x].size = size;
413 e820.map[x].type = type;
416 } /* add_memory_region */
420 static void __init print_memory_map(char *who)
424 for (i = 0; i < e820.nr_map; i++) {
425 printk(" %s: %016Lx - %016Lx ", who,
427 e820.map[i].addr + e820.map[i].size);
428 switch (e820.map[i].type) {
429 case E820_RAM: printk("(usable)\n");
432 printk("(reserved)\n");
435 printk("(ACPI data)\n");
438 printk("(ACPI NVS)\n");
440 default: printk("type %lu\n", e820.map[i].type);
447 * Sanitize the BIOS e820 map.
449 * Some e820 responses include overlapping entries. The following
450 * replaces the original e820 map with a new one, removing overlaps.
453 struct change_member {
454 struct e820entry *pbios; /* pointer to original bios entry */
455 unsigned long long addr; /* address for this change point */
457 static struct change_member change_point_list[2*E820MAX] __initdata;
458 static struct change_member *change_point[2*E820MAX] __initdata;
459 static struct e820entry *overlap_list[E820MAX] __initdata;
460 static struct e820entry new_bios[E820MAX] __initdata;
462 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
464 struct change_member *change_tmp;
465 unsigned long current_type, last_type;
466 unsigned long long last_addr;
467 int chgidx, still_changing;
470 int old_nr, new_nr, chg_nr;
474 Visually we're performing the following (1,2,3,4 = memory types)...
476 Sample memory map (w/overlaps):
477 ____22__________________
478 ______________________4_
479 ____1111________________
480 _44_____________________
481 11111111________________
482 ____________________33__
483 ___________44___________
484 __________33333_________
485 ______________22________
486 ___________________2222_
487 _________111111111______
488 _____________________11_
489 _________________4______
491 Sanitized equivalent (no overlap):
492 1_______________________
493 _44_____________________
494 ___1____________________
495 ____22__________________
496 ______11________________
497 _________1______________
498 __________3_____________
499 ___________44___________
500 _____________33_________
501 _______________2________
502 ________________1_______
503 _________________4______
504 ___________________2____
505 ____________________33__
506 ______________________4_
509 /* if there's only one memory region, don't bother */
515 /* bail out if we find any unreasonable addresses in bios map */
516 for (i=0; i<old_nr; i++)
517 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
520 /* create pointers for initial change-point information (for sorting) */
521 for (i=0; i < 2*old_nr; i++)
522 change_point[i] = &change_point_list[i];
524 /* record all known change-points (starting and ending addresses),
525 omitting those that are for empty memory regions */
527 for (i=0; i < old_nr; i++) {
528 if (biosmap[i].size != 0) {
529 change_point[chgidx]->addr = biosmap[i].addr;
530 change_point[chgidx++]->pbios = &biosmap[i];
531 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
532 change_point[chgidx++]->pbios = &biosmap[i];
535 chg_nr = chgidx; /* true number of change-points */
537 /* sort change-point list by memory addresses (low -> high) */
539 while (still_changing) {
541 for (i=1; i < chg_nr; i++) {
542 /* if <current_addr> > <last_addr>, swap */
543 /* or, if current=<start_addr> & last=<end_addr>, swap */
544 if ((change_point[i]->addr < change_point[i-1]->addr) ||
545 ((change_point[i]->addr == change_point[i-1]->addr) &&
546 (change_point[i]->addr == change_point[i]->pbios->addr) &&
547 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
550 change_tmp = change_point[i];
551 change_point[i] = change_point[i-1];
552 change_point[i-1] = change_tmp;
558 /* create a new bios memory map, removing overlaps */
559 overlap_entries=0; /* number of entries in the overlap table */
560 new_bios_entry=0; /* index for creating new bios map entries */
561 last_type = 0; /* start with undefined memory type */
562 last_addr = 0; /* start with 0 as last starting address */
563 /* loop through change-points, determining affect on the new bios map */
564 for (chgidx=0; chgidx < chg_nr; chgidx++)
566 /* keep track of all overlapping bios entries */
567 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
569 /* add map entry to overlap list (> 1 entry implies an overlap) */
570 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
574 /* remove entry from list (order independent, so swap with last) */
575 for (i=0; i<overlap_entries; i++)
577 if (overlap_list[i] == change_point[chgidx]->pbios)
578 overlap_list[i] = overlap_list[overlap_entries-1];
582 /* if there are overlapping entries, decide which "type" to use */
583 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
585 for (i=0; i<overlap_entries; i++)
586 if (overlap_list[i]->type > current_type)
587 current_type = overlap_list[i]->type;
588 /* continue building up new bios map based on this information */
589 if (current_type != last_type) {
590 if (last_type != 0) {
591 new_bios[new_bios_entry].size =
592 change_point[chgidx]->addr - last_addr;
593 /* move forward only if the new size was non-zero */
594 if (new_bios[new_bios_entry].size != 0)
595 if (++new_bios_entry >= E820MAX)
596 break; /* no more space left for new bios entries */
598 if (current_type != 0) {
599 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
600 new_bios[new_bios_entry].type = current_type;
601 last_addr=change_point[chgidx]->addr;
603 last_type = current_type;
606 new_nr = new_bios_entry; /* retain count for new bios entries */
608 /* copy new bios mapping into original location */
609 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
616 * Copy the BIOS e820 map into a safe place.
618 * Sanity-check it while we're at it..
620 * If we're lucky and live on a modern system, the setup code
621 * will have given us a memory map that we can use to properly
622 * set up memory. If we aren't, we'll fake a memory map.
624 * We check to see that the memory map contains at least 2 elements
625 * before we'll use it, because the detection code in setup.S may
626 * not be perfect and most every PC known to man has two memory
627 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
628 * thinkpad 560x, for example, does not cooperate with the memory
631 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
633 /* Only one memory region (or negative)? Ignore it */
638 unsigned long long start = biosmap->addr;
639 unsigned long long size = biosmap->size;
640 unsigned long long end = start + size;
641 unsigned long type = biosmap->type;
643 /* Overflow in 64 bits? Ignore the memory map. */
648 * Some BIOSes claim RAM in the 640k - 1M region.
649 * Not right. Fix it up.
651 if (type == E820_RAM) {
652 if (start < 0x100000ULL && end > 0xA0000ULL) {
653 if (start < 0xA0000ULL)
654 add_memory_region(start, 0xA0000ULL-start, type);
655 if (end <= 0x100000ULL)
661 add_memory_region(start, size, type);
662 } while (biosmap++,--nr_map);
666 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
668 #ifdef CONFIG_EDD_MODULE
672 * copy_edd() - Copy the BIOS EDD information
673 * from boot_params into a safe place.
676 static inline void copy_edd(void)
678 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
679 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
680 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
681 edd.edd_info_nr = EDD_NR;
684 static inline void copy_edd(void)
690 * Do NOT EVER look at the BIOS memory size location.
691 * It does not work on many machines.
693 #define LOWMEMSIZE() (0x9f000)
695 static void __init parse_cmdline_early (char ** cmdline_p)
697 char c = ' ', *to = command_line, *from = saved_command_line;
701 /* Save unparsed command line copy for /proc/cmdline */
702 saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
708 * "mem=nopentium" disables the 4MB page tables.
709 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
710 * to <mem>, overriding the bios size.
711 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
712 * <start> to <start>+<mem>, overriding the bios size.
714 * HPA tells me bootloaders need to parse mem=, so no new
715 * option should be mem= [also see Documentation/i386/boot.txt]
717 if (!memcmp(from, "mem=", 4)) {
718 if (to != command_line)
720 if (!memcmp(from+4, "nopentium", 9)) {
722 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
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 mem_size;
733 mem_size = memparse(from+4, &from);
734 limit_regions(mem_size);
739 else if (!memcmp(from, "memmap=", 7)) {
740 if (to != command_line)
742 if (!memcmp(from+7, "exactmap", 8)) {
743 #ifdef CONFIG_CRASH_DUMP
744 /* If we are doing a crash dump, we
745 * still need to know the real mem
746 * size before original memory map is
750 saved_max_pfn = max_pfn;
756 /* If the user specifies memory size, we
757 * limit the BIOS-provided memory map to
758 * that size. exactmap can be used to specify
759 * the exact map. mem=number can be used to
760 * trim the existing memory map.
762 unsigned long long start_at, mem_size;
764 mem_size = memparse(from+7, &from);
766 start_at = memparse(from+1, &from);
767 add_memory_region(start_at, mem_size, E820_RAM);
768 } else if (*from == '#') {
769 start_at = memparse(from+1, &from);
770 add_memory_region(start_at, mem_size, E820_ACPI);
771 } else if (*from == '$') {
772 start_at = memparse(from+1, &from);
773 add_memory_region(start_at, mem_size, E820_RESERVED);
775 limit_regions(mem_size);
781 else if (!memcmp(from, "noexec=", 7))
782 noexec_setup(from + 7);
785 #ifdef CONFIG_X86_SMP
787 * If the BIOS enumerates physical processors before logical,
788 * maxcpus=N at enumeration-time can be used to disable HT.
790 else if (!memcmp(from, "maxcpus=", 8)) {
791 extern unsigned int maxcpus;
793 maxcpus = simple_strtoul(from + 8, NULL, 0);
797 #ifdef CONFIG_ACPI_BOOT
798 /* "acpi=off" disables both ACPI table parsing and interpreter */
799 else if (!memcmp(from, "acpi=off", 8)) {
803 /* acpi=force to over-ride black-list */
804 else if (!memcmp(from, "acpi=force", 10)) {
810 /* acpi=strict disables out-of-spec workarounds */
811 else if (!memcmp(from, "acpi=strict", 11)) {
815 /* Limit ACPI just to boot-time to enable HT */
816 else if (!memcmp(from, "acpi=ht", 7)) {
822 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
823 else if (!memcmp(from, "pci=noacpi", 10)) {
826 /* "acpi=noirq" disables ACPI interrupt routing */
827 else if (!memcmp(from, "acpi=noirq", 10)) {
831 else if (!memcmp(from, "acpi_sci=edge", 13))
832 acpi_sci_flags.trigger = 1;
834 else if (!memcmp(from, "acpi_sci=level", 14))
835 acpi_sci_flags.trigger = 3;
837 else if (!memcmp(from, "acpi_sci=high", 13))
838 acpi_sci_flags.polarity = 1;
840 else if (!memcmp(from, "acpi_sci=low", 12))
841 acpi_sci_flags.polarity = 3;
843 #ifdef CONFIG_X86_IO_APIC
844 else if (!memcmp(from, "acpi_skip_timer_override", 24))
845 acpi_skip_timer_override = 1;
848 #ifdef CONFIG_X86_LOCAL_APIC
849 /* disable IO-APIC */
850 else if (!memcmp(from, "noapic", 6))
851 disable_ioapic_setup();
852 #endif /* CONFIG_X86_LOCAL_APIC */
853 #endif /* CONFIG_ACPI_BOOT */
855 #ifdef CONFIG_X86_LOCAL_APIC
856 /* enable local APIC */
857 else if (!memcmp(from, "lapic", 5))
860 /* disable local APIC */
861 else if (!memcmp(from, "nolapic", 6))
863 #endif /* CONFIG_X86_LOCAL_APIC */
866 /* crashkernel=size@addr specifies the location to reserve for
867 * a crash kernel. By reserving this memory we guarantee
868 * that linux never set's it up as a DMA target.
869 * Useful for holding code to do something appropriate
870 * after a kernel panic.
872 else if (!memcmp(from, "crashkernel=", 12)) {
873 unsigned long size, base;
874 size = memparse(from+12, &from);
876 base = memparse(from+1, &from);
877 /* FIXME: Do I want a sanity check
878 * to validate the memory range?
880 crashk_res.start = base;
881 crashk_res.end = base + size - 1;
885 #ifdef CONFIG_CRASH_DUMP
886 /* elfcorehdr= specifies the location of elf core header
887 * stored by the crashed kernel.
889 else if (!memcmp(from, "elfcorehdr=", 11))
890 elfcorehdr_addr = memparse(from+11, &from);
894 * highmem=size forces highmem to be exactly 'size' bytes.
895 * This works even on boxes that have no highmem otherwise.
896 * This also works to reduce highmem size on bigger boxes.
898 else if (!memcmp(from, "highmem=", 8))
899 highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
902 * vmalloc=size forces the vmalloc area to be exactly 'size'
903 * bytes. This can be used to increase (or decrease) the
904 * vmalloc area - the default is 128m.
906 else if (!memcmp(from, "vmalloc=", 8))
907 __VMALLOC_RESERVE = memparse(from+8, &from);
913 if (COMMAND_LINE_SIZE <= ++len)
918 *cmdline_p = command_line;
920 printk(KERN_INFO "user-defined physical RAM map:\n");
921 print_memory_map("user");
926 * Callback for efi_memory_walk.
929 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
931 unsigned long *max_pfn = arg, pfn;
934 pfn = PFN_UP(end -1);
943 * Find the highest page frame number we have available
945 void __init find_max_pfn(void)
951 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
955 for (i = 0; i < e820.nr_map; i++) {
956 unsigned long start, end;
958 if (e820.map[i].type != E820_RAM)
960 start = PFN_UP(e820.map[i].addr);
961 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
970 * Determine low and high memory ranges:
972 unsigned long __init find_max_low_pfn(void)
974 unsigned long max_low_pfn;
976 max_low_pfn = max_pfn;
977 if (max_low_pfn > MAXMEM_PFN) {
978 if (highmem_pages == -1)
979 highmem_pages = max_pfn - MAXMEM_PFN;
980 if (highmem_pages + MAXMEM_PFN < max_pfn)
981 max_pfn = MAXMEM_PFN + highmem_pages;
982 if (highmem_pages + MAXMEM_PFN > max_pfn) {
983 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
986 max_low_pfn = MAXMEM_PFN;
987 #ifndef CONFIG_HIGHMEM
988 /* Maximum memory usable is what is directly addressable */
989 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
991 if (max_pfn > MAX_NONPAE_PFN)
992 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
994 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
995 max_pfn = MAXMEM_PFN;
996 #else /* !CONFIG_HIGHMEM */
997 #ifndef CONFIG_X86_PAE
998 if (max_pfn > MAX_NONPAE_PFN) {
999 max_pfn = MAX_NONPAE_PFN;
1000 printk(KERN_WARNING "Warning only 4GB will be used.\n");
1001 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1003 #endif /* !CONFIG_X86_PAE */
1004 #endif /* !CONFIG_HIGHMEM */
1006 if (highmem_pages == -1)
1008 #ifdef CONFIG_HIGHMEM
1009 if (highmem_pages >= max_pfn) {
1010 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1013 if (highmem_pages) {
1014 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1015 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1018 max_low_pfn -= highmem_pages;
1022 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1029 * Free all available memory for boot time allocation. Used
1030 * as a callback function by efi_memory_walk()
1034 free_available_memory(unsigned long start, unsigned long end, void *arg)
1036 /* check max_low_pfn */
1037 if (start >= ((max_low_pfn + 1) << PAGE_SHIFT))
1039 if (end >= ((max_low_pfn + 1) << PAGE_SHIFT))
1040 end = (max_low_pfn + 1) << PAGE_SHIFT;
1042 free_bootmem(start, end - start);
1047 * Register fully available low RAM pages with the bootmem allocator.
1049 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1054 efi_memmap_walk(free_available_memory, NULL);
1057 for (i = 0; i < e820.nr_map; i++) {
1058 unsigned long curr_pfn, last_pfn, size;
1060 * Reserve usable low memory
1062 if (e820.map[i].type != E820_RAM)
1065 * We are rounding up the start address of usable memory:
1067 curr_pfn = PFN_UP(e820.map[i].addr);
1068 if (curr_pfn >= max_low_pfn)
1071 * ... and at the end of the usable range downwards:
1073 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1075 if (last_pfn > max_low_pfn)
1076 last_pfn = max_low_pfn;
1079 * .. finally, did all the rounding and playing
1080 * around just make the area go away?
1082 if (last_pfn <= curr_pfn)
1085 size = last_pfn - curr_pfn;
1086 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1091 * workaround for Dell systems that neglect to reserve EBDA
1093 static void __init reserve_ebda_region(void)
1096 addr = get_bios_ebda();
1098 reserve_bootmem(addr, PAGE_SIZE);
1101 #ifndef CONFIG_NEED_MULTIPLE_NODES
1102 void __init setup_bootmem_allocator(void);
1103 static unsigned long __init setup_memory(void)
1106 * partially used pages are not usable - thus
1107 * we are rounding upwards:
1109 min_low_pfn = PFN_UP(init_pg_tables_end);
1113 max_low_pfn = find_max_low_pfn();
1115 #ifdef CONFIG_HIGHMEM
1116 highstart_pfn = highend_pfn = max_pfn;
1117 if (max_pfn > max_low_pfn) {
1118 highstart_pfn = max_low_pfn;
1120 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1121 pages_to_mb(highend_pfn - highstart_pfn));
1123 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1124 pages_to_mb(max_low_pfn));
1126 setup_bootmem_allocator();
1131 void __init zone_sizes_init(void)
1133 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1134 unsigned int max_dma, low;
1136 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1140 zones_size[ZONE_DMA] = low;
1142 zones_size[ZONE_DMA] = max_dma;
1143 zones_size[ZONE_NORMAL] = low - max_dma;
1144 #ifdef CONFIG_HIGHMEM
1145 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1148 free_area_init(zones_size);
1151 extern unsigned long __init setup_memory(void);
1152 extern void zone_sizes_init(void);
1153 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1155 void __init setup_bootmem_allocator(void)
1157 unsigned long bootmap_size;
1159 * Initialize the boot-time allocator (with low memory only):
1161 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1163 register_bootmem_low_pages(max_low_pfn);
1166 * Reserve the bootmem bitmap itself as well. We do this in two
1167 * steps (first step was init_bootmem()) because this catches
1168 * the (very unlikely) case of us accidentally initializing the
1169 * bootmem allocator with an invalid RAM area.
1171 reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1172 bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1175 * reserve physical page 0 - it's a special BIOS page on many boxes,
1176 * enabling clean reboots, SMP operation, laptop functions.
1178 reserve_bootmem(0, PAGE_SIZE);
1180 /* reserve EBDA region, it's a 4K region */
1181 reserve_ebda_region();
1183 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1184 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1185 unless you have no PS/2 mouse plugged in. */
1186 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1187 boot_cpu_data.x86 == 6)
1188 reserve_bootmem(0xa0000 - 4096, 4096);
1192 * But first pinch a few for the stack/trampoline stuff
1193 * FIXME: Don't need the extra page at 4K, but need to fix
1194 * trampoline before removing it. (see the GDT stuff)
1196 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1198 #ifdef CONFIG_ACPI_SLEEP
1200 * Reserve low memory region for sleep support.
1202 acpi_reserve_bootmem();
1204 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1206 * Find and reserve possible boot-time SMP configuration:
1211 #ifdef CONFIG_BLK_DEV_INITRD
1212 if (LOADER_TYPE && INITRD_START) {
1213 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1214 reserve_bootmem(INITRD_START, INITRD_SIZE);
1216 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1217 initrd_end = initrd_start+INITRD_SIZE;
1220 printk(KERN_ERR "initrd extends beyond end of memory "
1221 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1222 INITRD_START + INITRD_SIZE,
1223 max_low_pfn << PAGE_SHIFT);
1229 if (crashk_res.start != crashk_res.end)
1230 reserve_bootmem(crashk_res.start,
1231 crashk_res.end - crashk_res.start + 1);
1236 * The node 0 pgdat is initialized before all of these because
1237 * it's needed for bootmem. node>0 pgdats have their virtual
1238 * space allocated before the pagetables are in place to access
1239 * them, so they can't be cleared then.
1241 * This should all compile down to nothing when NUMA is off.
1243 void __init remapped_pgdat_init(void)
1247 for_each_online_node(nid) {
1249 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1254 * Request address space for all standard RAM and ROM resources
1255 * and also for regions reported as reserved by the e820.
1258 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1263 for (i = 0; i < e820.nr_map; i++) {
1264 struct resource *res;
1265 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1267 res = alloc_bootmem_low(sizeof(struct resource));
1268 switch (e820.map[i].type) {
1269 case E820_RAM: res->name = "System RAM"; break;
1270 case E820_ACPI: res->name = "ACPI Tables"; break;
1271 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1272 default: res->name = "reserved";
1274 res->start = e820.map[i].addr;
1275 res->end = res->start + e820.map[i].size - 1;
1276 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1277 request_resource(&iomem_resource, res);
1278 if (e820.map[i].type == E820_RAM) {
1280 * We don't know which RAM region contains kernel data,
1281 * so we try it repeatedly and let the resource manager
1284 request_resource(res, code_resource);
1285 request_resource(res, data_resource);
1287 request_resource(res, &crashk_res);
1294 * Request address space for all standard resources
1296 static void __init register_memory(void)
1298 unsigned long gapstart, gapsize;
1299 unsigned long long last;
1303 efi_initialize_iomem_resources(&code_resource, &data_resource);
1305 legacy_init_iomem_resources(&code_resource, &data_resource);
1307 /* EFI systems may still have VGA */
1308 request_resource(&iomem_resource, &video_ram_resource);
1310 /* request I/O space for devices used on all i[345]86 PCs */
1311 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1312 request_resource(&ioport_resource, &standard_io_resources[i]);
1315 * Search for the bigest gap in the low 32 bits of the e820
1318 last = 0x100000000ull;
1319 gapstart = 0x10000000;
1323 unsigned long long start = e820.map[i].addr;
1324 unsigned long long end = start + e820.map[i].size;
1327 * Since "last" is at most 4GB, we know we'll
1328 * fit in 32 bits if this condition is true
1331 unsigned long gap = last - end;
1333 if (gap > gapsize) {
1343 * Start allocating dynamic PCI memory a bit into the gap,
1344 * aligned up to the nearest megabyte.
1346 * Question: should we try to pad it up a bit (do something
1347 * like " + (gapsize >> 3)" in there too?). We now have the
1350 pci_mem_start = (gapstart + 0xfffff) & ~0xfffff;
1352 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1353 pci_mem_start, gapstart, gapsize);
1356 /* Use inline assembly to define this because the nops are defined
1357 as inline assembly strings in the include files and we cannot
1358 get them easily into strings. */
1359 asm("\t.data\nintelnops: "
1360 GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
1361 GENERIC_NOP7 GENERIC_NOP8);
1362 asm("\t.data\nk8nops: "
1363 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
1365 asm("\t.data\nk7nops: "
1366 K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
1369 extern unsigned char intelnops[], k8nops[], k7nops[];
1370 static unsigned char *intel_nops[ASM_NOP_MAX+1] = {
1375 intelnops + 1 + 2 + 3,
1376 intelnops + 1 + 2 + 3 + 4,
1377 intelnops + 1 + 2 + 3 + 4 + 5,
1378 intelnops + 1 + 2 + 3 + 4 + 5 + 6,
1379 intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1381 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
1387 k8nops + 1 + 2 + 3 + 4,
1388 k8nops + 1 + 2 + 3 + 4 + 5,
1389 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
1390 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1392 static unsigned char *k7_nops[ASM_NOP_MAX+1] = {
1398 k7nops + 1 + 2 + 3 + 4,
1399 k7nops + 1 + 2 + 3 + 4 + 5,
1400 k7nops + 1 + 2 + 3 + 4 + 5 + 6,
1401 k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1405 unsigned char **noptable;
1407 { X86_FEATURE_K8, k8_nops },
1408 { X86_FEATURE_K7, k7_nops },
1412 /* Replace instructions with better alternatives for this CPU type.
1414 This runs before SMP is initialized to avoid SMP problems with
1415 self modifying code. This implies that assymetric systems where
1416 APs have less capabilities than the boot processor are not handled.
1417 In this case boot with "noreplacement". */
1418 void apply_alternatives(void *start, void *end)
1420 struct alt_instr *a;
1422 unsigned char **noptable = intel_nops;
1423 for (i = 0; noptypes[i].cpuid >= 0; i++) {
1424 if (boot_cpu_has(noptypes[i].cpuid)) {
1425 noptable = noptypes[i].noptable;
1429 for (a = start; (void *)a < end; a++) {
1430 if (!boot_cpu_has(a->cpuid))
1432 BUG_ON(a->replacementlen > a->instrlen);
1433 memcpy(a->instr, a->replacement, a->replacementlen);
1434 diff = a->instrlen - a->replacementlen;
1435 /* Pad the rest with nops */
1436 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
1438 if (k > ASM_NOP_MAX)
1440 memcpy(a->instr + i, noptable[k], k);
1445 static int no_replacement __initdata = 0;
1447 void __init alternative_instructions(void)
1449 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
1452 apply_alternatives(__alt_instructions, __alt_instructions_end);
1455 static int __init noreplacement_setup(char *s)
1461 __setup("noreplacement", noreplacement_setup);
1463 static char * __init machine_specific_memory_setup(void);
1466 static void set_mca_bus(int x)
1471 static void set_mca_bus(int x) { }
1475 * Determine if we were loaded by an EFI loader. If so, then we have also been
1476 * passed the efi memmap, systab, etc., so we should use these data structures
1477 * for initialization. Note, the efi init code path is determined by the
1478 * global efi_enabled. This allows the same kernel image to be used on existing
1479 * systems (with a traditional BIOS) as well as on EFI systems.
1481 void __init setup_arch(char **cmdline_p)
1483 unsigned long max_low_pfn;
1485 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1486 pre_setup_arch_hook();
1490 * FIXME: This isn't an official loader_type right
1491 * now but does currently work with elilo.
1492 * If we were configured as an EFI kernel, check to make
1493 * sure that we were loaded correctly from elilo and that
1494 * the system table is valid. If not, then initialize normally.
1497 if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1501 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1502 drive_info = DRIVE_INFO;
1503 screen_info = SCREEN_INFO;
1504 edid_info = EDID_INFO;
1505 apm_info.bios = APM_BIOS_INFO;
1506 ist_info = IST_INFO;
1507 saved_videomode = VIDEO_MODE;
1508 if( SYS_DESC_TABLE.length != 0 ) {
1509 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1510 machine_id = SYS_DESC_TABLE.table[0];
1511 machine_submodel_id = SYS_DESC_TABLE.table[1];
1512 BIOS_revision = SYS_DESC_TABLE.table[2];
1514 bootloader_type = LOADER_TYPE;
1516 #ifdef CONFIG_BLK_DEV_RAM
1517 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1518 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1519 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1525 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1526 print_memory_map(machine_specific_memory_setup());
1531 if (!MOUNT_ROOT_RDONLY)
1532 root_mountflags &= ~MS_RDONLY;
1533 init_mm.start_code = (unsigned long) _text;
1534 init_mm.end_code = (unsigned long) _etext;
1535 init_mm.end_data = (unsigned long) _edata;
1536 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1538 code_resource.start = virt_to_phys(_text);
1539 code_resource.end = virt_to_phys(_etext)-1;
1540 data_resource.start = virt_to_phys(_etext);
1541 data_resource.end = virt_to_phys(_edata)-1;
1543 parse_cmdline_early(cmdline_p);
1545 max_low_pfn = setup_memory();
1548 * NOTE: before this point _nobody_ is allowed to allocate
1549 * any memory using the bootmem allocator. Although the
1550 * alloctor is now initialised only the first 8Mb of the kernel
1551 * virtual address space has been mapped. All allocations before
1552 * paging_init() has completed must use the alloc_bootmem_low_pages()
1553 * variant (which allocates DMA'able memory) and care must be taken
1554 * not to exceed the 8Mb limit.
1558 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1561 remapped_pgdat_init();
1566 * NOTE: at this point the bootmem allocator is fully available.
1569 #ifdef CONFIG_EARLY_PRINTK
1571 char *s = strstr(*cmdline_p, "earlyprintk=");
1573 extern void setup_early_printk(char *);
1575 setup_early_printk(s);
1576 printk("early console enabled\n");
1584 #ifdef CONFIG_X86_GENERICARCH
1585 generic_apic_probe(*cmdline_p);
1590 #ifdef CONFIG_ACPI_BOOT
1592 * Parse the ACPI tables for possible boot-time SMP configuration.
1594 acpi_boot_table_init();
1598 #ifdef CONFIG_X86_LOCAL_APIC
1599 if (smp_found_config)
1606 #if defined(CONFIG_VGA_CONSOLE)
1607 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1608 conswitchp = &vga_con;
1609 #elif defined(CONFIG_DUMMY_CONSOLE)
1610 conswitchp = &dummy_con;
1615 #include "setup_arch_post.h"
1619 * c-file-style:"k&r"