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/mmzone.h>
29 #include <linux/screen_info.h>
30 #include <linux/ioport.h>
31 #include <linux/acpi.h>
32 #include <linux/apm_bios.h>
33 #include <linux/initrd.h>
34 #include <linux/bootmem.h>
35 #include <linux/seq_file.h>
36 #include <linux/platform_device.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>
48 #include <linux/dmi.h>
49 #include <linux/pfn.h>
51 #include <video/edid.h>
55 #include <asm/mpspec.h>
56 #include <asm/setup.h>
57 #include <asm/arch_hooks.h>
58 #include <asm/sections.h>
59 #include <asm/io_apic.h>
62 #include <setup_arch.h>
63 #include <bios_ebda.h>
65 /* Forward Declaration. */
66 void __init find_max_pfn(void);
68 /* This value is set up by the early boot code to point to the value
69 immediately after the boot time page tables. It contains a *physical*
70 address, and must not be in the .bss segment! */
71 unsigned long init_pg_tables_end __initdata = ~0UL;
73 int disable_pse __devinitdata = 0;
81 EXPORT_SYMBOL(efi_enabled);
84 /* cpu data as detected by the assembly code in head.S */
85 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
86 /* common cpu data for all cpus */
87 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
88 EXPORT_SYMBOL(boot_cpu_data);
90 unsigned long mmu_cr4_features;
93 int acpi_disabled = 0;
95 int acpi_disabled = 1;
97 EXPORT_SYMBOL(acpi_disabled);
99 /* for MCA, but anyone else can use it if they want */
100 unsigned int machine_id;
102 EXPORT_SYMBOL(machine_id);
104 unsigned int machine_submodel_id;
105 unsigned int BIOS_revision;
106 unsigned int mca_pentium_flag;
108 /* For PCI or other memory-mapped resources */
109 unsigned long pci_mem_start = 0x10000000;
111 EXPORT_SYMBOL(pci_mem_start);
114 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
117 /* user-defined highmem size */
118 static unsigned int highmem_pages = -1;
123 struct drive_info_struct { char dummy[32]; } drive_info;
124 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
125 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
126 EXPORT_SYMBOL(drive_info);
128 struct screen_info screen_info;
129 EXPORT_SYMBOL(screen_info);
130 struct apm_info apm_info;
131 EXPORT_SYMBOL(apm_info);
132 struct sys_desc_table_struct {
133 unsigned short length;
134 unsigned char table[0];
136 struct edid_info edid_info;
137 EXPORT_SYMBOL_GPL(edid_info);
138 struct ist_info ist_info;
139 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
140 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
141 EXPORT_SYMBOL(ist_info);
145 extern void early_cpu_init(void);
146 extern int root_mountflags;
148 unsigned long saved_videomode;
150 #define RAMDISK_IMAGE_START_MASK 0x07FF
151 #define RAMDISK_PROMPT_FLAG 0x8000
152 #define RAMDISK_LOAD_FLAG 0x4000
154 static char command_line[COMMAND_LINE_SIZE];
156 unsigned char __initdata boot_params[PARAM_SIZE];
158 static struct resource data_resource = {
159 .name = "Kernel data",
162 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
165 static struct resource code_resource = {
166 .name = "Kernel code",
169 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
172 static struct resource system_rom_resource = {
173 .name = "System ROM",
176 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
179 static struct resource extension_rom_resource = {
180 .name = "Extension ROM",
183 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
186 static struct resource adapter_rom_resources[] = { {
187 .name = "Adapter ROM",
190 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
192 .name = "Adapter ROM",
195 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
197 .name = "Adapter ROM",
200 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
202 .name = "Adapter ROM",
205 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
207 .name = "Adapter ROM",
210 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
212 .name = "Adapter ROM",
215 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
218 #define ADAPTER_ROM_RESOURCES \
219 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
221 static struct resource video_rom_resource = {
225 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
228 static struct resource video_ram_resource = {
229 .name = "Video RAM area",
232 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
235 static struct resource standard_io_resources[] = { {
239 .flags = IORESOURCE_BUSY | IORESOURCE_IO
244 .flags = IORESOURCE_BUSY | IORESOURCE_IO
249 .flags = IORESOURCE_BUSY | IORESOURCE_IO
254 .flags = IORESOURCE_BUSY | IORESOURCE_IO
259 .flags = IORESOURCE_BUSY | IORESOURCE_IO
261 .name = "dma page reg",
264 .flags = IORESOURCE_BUSY | IORESOURCE_IO
269 .flags = IORESOURCE_BUSY | IORESOURCE_IO
274 .flags = IORESOURCE_BUSY | IORESOURCE_IO
279 .flags = IORESOURCE_BUSY | IORESOURCE_IO
282 #define STANDARD_IO_RESOURCES \
283 (sizeof standard_io_resources / sizeof standard_io_resources[0])
285 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
287 static int __init romchecksum(unsigned char *rom, unsigned long length)
289 unsigned char *p, sum = 0;
291 for (p = rom; p < rom + length; p++)
296 static void __init probe_roms(void)
298 unsigned long start, length, upper;
303 upper = adapter_rom_resources[0].start;
304 for (start = video_rom_resource.start; start < upper; start += 2048) {
305 rom = isa_bus_to_virt(start);
306 if (!romsignature(rom))
309 video_rom_resource.start = start;
311 /* 0 < length <= 0x7f * 512, historically */
312 length = rom[2] * 512;
314 /* if checksum okay, trust length byte */
315 if (length && romchecksum(rom, length))
316 video_rom_resource.end = start + length - 1;
318 request_resource(&iomem_resource, &video_rom_resource);
322 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
327 request_resource(&iomem_resource, &system_rom_resource);
328 upper = system_rom_resource.start;
330 /* check for extension rom (ignore length byte!) */
331 rom = isa_bus_to_virt(extension_rom_resource.start);
332 if (romsignature(rom)) {
333 length = extension_rom_resource.end - extension_rom_resource.start + 1;
334 if (romchecksum(rom, length)) {
335 request_resource(&iomem_resource, &extension_rom_resource);
336 upper = extension_rom_resource.start;
340 /* check for adapter roms on 2k boundaries */
341 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
342 rom = isa_bus_to_virt(start);
343 if (!romsignature(rom))
346 /* 0 < length <= 0x7f * 512, historically */
347 length = rom[2] * 512;
349 /* but accept any length that fits if checksum okay */
350 if (!length || start + length > upper || !romchecksum(rom, length))
353 adapter_rom_resources[i].start = start;
354 adapter_rom_resources[i].end = start + length - 1;
355 request_resource(&iomem_resource, &adapter_rom_resources[i]);
357 start = adapter_rom_resources[i++].end & ~2047UL;
361 static void __init limit_regions(unsigned long long size)
363 unsigned long long current_addr = 0;
367 efi_memory_desc_t *md;
370 for (p = memmap.map, i = 0; p < memmap.map_end;
371 p += memmap.desc_size, i++) {
373 current_addr = md->phys_addr + (md->num_pages << 12);
374 if (md->type == EFI_CONVENTIONAL_MEMORY) {
375 if (current_addr >= size) {
377 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
378 memmap.nr_map = i + 1;
384 for (i = 0; i < e820.nr_map; i++) {
385 current_addr = e820.map[i].addr + e820.map[i].size;
386 if (current_addr < size)
389 if (e820.map[i].type != E820_RAM)
392 if (e820.map[i].addr >= size) {
394 * This region starts past the end of the
395 * requested size, skip it completely.
400 e820.map[i].size -= current_addr - size;
406 void __init add_memory_region(unsigned long long start,
407 unsigned long long size, int type)
415 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
419 e820.map[x].addr = start;
420 e820.map[x].size = size;
421 e820.map[x].type = type;
424 } /* add_memory_region */
428 static void __init print_memory_map(char *who)
432 for (i = 0; i < e820.nr_map; i++) {
433 printk(" %s: %016Lx - %016Lx ", who,
435 e820.map[i].addr + e820.map[i].size);
436 switch (e820.map[i].type) {
437 case E820_RAM: printk("(usable)\n");
440 printk("(reserved)\n");
443 printk("(ACPI data)\n");
446 printk("(ACPI NVS)\n");
448 default: printk("type %lu\n", e820.map[i].type);
455 * Sanitize the BIOS e820 map.
457 * Some e820 responses include overlapping entries. The following
458 * replaces the original e820 map with a new one, removing overlaps.
461 struct change_member {
462 struct e820entry *pbios; /* pointer to original bios entry */
463 unsigned long long addr; /* address for this change point */
465 static struct change_member change_point_list[2*E820MAX] __initdata;
466 static struct change_member *change_point[2*E820MAX] __initdata;
467 static struct e820entry *overlap_list[E820MAX] __initdata;
468 static struct e820entry new_bios[E820MAX] __initdata;
470 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
472 struct change_member *change_tmp;
473 unsigned long current_type, last_type;
474 unsigned long long last_addr;
475 int chgidx, still_changing;
478 int old_nr, new_nr, chg_nr;
482 Visually we're performing the following (1,2,3,4 = memory types)...
484 Sample memory map (w/overlaps):
485 ____22__________________
486 ______________________4_
487 ____1111________________
488 _44_____________________
489 11111111________________
490 ____________________33__
491 ___________44___________
492 __________33333_________
493 ______________22________
494 ___________________2222_
495 _________111111111______
496 _____________________11_
497 _________________4______
499 Sanitized equivalent (no overlap):
500 1_______________________
501 _44_____________________
502 ___1____________________
503 ____22__________________
504 ______11________________
505 _________1______________
506 __________3_____________
507 ___________44___________
508 _____________33_________
509 _______________2________
510 ________________1_______
511 _________________4______
512 ___________________2____
513 ____________________33__
514 ______________________4_
517 /* if there's only one memory region, don't bother */
523 /* bail out if we find any unreasonable addresses in bios map */
524 for (i=0; i<old_nr; i++)
525 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
528 /* create pointers for initial change-point information (for sorting) */
529 for (i=0; i < 2*old_nr; i++)
530 change_point[i] = &change_point_list[i];
532 /* record all known change-points (starting and ending addresses),
533 omitting those that are for empty memory regions */
535 for (i=0; i < old_nr; i++) {
536 if (biosmap[i].size != 0) {
537 change_point[chgidx]->addr = biosmap[i].addr;
538 change_point[chgidx++]->pbios = &biosmap[i];
539 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
540 change_point[chgidx++]->pbios = &biosmap[i];
543 chg_nr = chgidx; /* true number of change-points */
545 /* sort change-point list by memory addresses (low -> high) */
547 while (still_changing) {
549 for (i=1; i < chg_nr; i++) {
550 /* if <current_addr> > <last_addr>, swap */
551 /* or, if current=<start_addr> & last=<end_addr>, swap */
552 if ((change_point[i]->addr < change_point[i-1]->addr) ||
553 ((change_point[i]->addr == change_point[i-1]->addr) &&
554 (change_point[i]->addr == change_point[i]->pbios->addr) &&
555 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
558 change_tmp = change_point[i];
559 change_point[i] = change_point[i-1];
560 change_point[i-1] = change_tmp;
566 /* create a new bios memory map, removing overlaps */
567 overlap_entries=0; /* number of entries in the overlap table */
568 new_bios_entry=0; /* index for creating new bios map entries */
569 last_type = 0; /* start with undefined memory type */
570 last_addr = 0; /* start with 0 as last starting address */
571 /* loop through change-points, determining affect on the new bios map */
572 for (chgidx=0; chgidx < chg_nr; chgidx++)
574 /* keep track of all overlapping bios entries */
575 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
577 /* add map entry to overlap list (> 1 entry implies an overlap) */
578 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
582 /* remove entry from list (order independent, so swap with last) */
583 for (i=0; i<overlap_entries; i++)
585 if (overlap_list[i] == change_point[chgidx]->pbios)
586 overlap_list[i] = overlap_list[overlap_entries-1];
590 /* if there are overlapping entries, decide which "type" to use */
591 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
593 for (i=0; i<overlap_entries; i++)
594 if (overlap_list[i]->type > current_type)
595 current_type = overlap_list[i]->type;
596 /* continue building up new bios map based on this information */
597 if (current_type != last_type) {
598 if (last_type != 0) {
599 new_bios[new_bios_entry].size =
600 change_point[chgidx]->addr - last_addr;
601 /* move forward only if the new size was non-zero */
602 if (new_bios[new_bios_entry].size != 0)
603 if (++new_bios_entry >= E820MAX)
604 break; /* no more space left for new bios entries */
606 if (current_type != 0) {
607 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
608 new_bios[new_bios_entry].type = current_type;
609 last_addr=change_point[chgidx]->addr;
611 last_type = current_type;
614 new_nr = new_bios_entry; /* retain count for new bios entries */
616 /* copy new bios mapping into original location */
617 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
624 * Copy the BIOS e820 map into a safe place.
626 * Sanity-check it while we're at it..
628 * If we're lucky and live on a modern system, the setup code
629 * will have given us a memory map that we can use to properly
630 * set up memory. If we aren't, we'll fake a memory map.
632 * We check to see that the memory map contains at least 2 elements
633 * before we'll use it, because the detection code in setup.S may
634 * not be perfect and most every PC known to man has two memory
635 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
636 * thinkpad 560x, for example, does not cooperate with the memory
639 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
641 /* Only one memory region (or negative)? Ignore it */
646 unsigned long long start = biosmap->addr;
647 unsigned long long size = biosmap->size;
648 unsigned long long end = start + size;
649 unsigned long type = biosmap->type;
651 /* Overflow in 64 bits? Ignore the memory map. */
656 * Some BIOSes claim RAM in the 640k - 1M region.
657 * Not right. Fix it up.
659 if (type == E820_RAM) {
660 if (start < 0x100000ULL && end > 0xA0000ULL) {
661 if (start < 0xA0000ULL)
662 add_memory_region(start, 0xA0000ULL-start, type);
663 if (end <= 0x100000ULL)
669 add_memory_region(start, size, type);
670 } while (biosmap++,--nr_map);
674 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
676 #ifdef CONFIG_EDD_MODULE
680 * copy_edd() - Copy the BIOS EDD information
681 * from boot_params into a safe place.
684 static inline void copy_edd(void)
686 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
687 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
688 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
689 edd.edd_info_nr = EDD_NR;
692 static inline void copy_edd(void)
697 static int __initdata user_defined_memmap = 0;
700 * "mem=nopentium" disables the 4MB page tables.
701 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
702 * to <mem>, overriding the bios size.
703 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
704 * <start> to <start>+<mem>, overriding the bios size.
706 * HPA tells me bootloaders need to parse mem=, so no new
707 * option should be mem= [also see Documentation/i386/boot.txt]
709 static int __init parse_mem(char *arg)
714 if (strcmp(arg, "nopentium") == 0) {
715 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
718 /* If the user specifies memory size, we
719 * limit the BIOS-provided memory map to
720 * that size. exactmap can be used to specify
721 * the exact map. mem=number can be used to
722 * trim the existing memory map.
724 unsigned long long mem_size;
726 mem_size = memparse(arg, &arg);
727 limit_regions(mem_size);
728 user_defined_memmap = 1;
732 early_param("mem", parse_mem);
734 static int __init parse_memmap(char *arg)
739 if (strcmp(arg, "exactmap") == 0) {
740 #ifdef CONFIG_CRASH_DUMP
741 /* If we are doing a crash dump, we
742 * still need to know the real mem
743 * size before original memory map is
747 saved_max_pfn = max_pfn;
750 user_defined_memmap = 1;
752 /* If the user specifies memory size, we
753 * limit the BIOS-provided memory map to
754 * that size. exactmap can be used to specify
755 * the exact map. mem=number can be used to
756 * trim the existing memory map.
758 unsigned long long start_at, mem_size;
760 mem_size = memparse(arg, &arg);
762 start_at = memparse(arg+1, &arg);
763 add_memory_region(start_at, mem_size, E820_RAM);
764 } else if (*arg == '#') {
765 start_at = memparse(arg+1, &arg);
766 add_memory_region(start_at, mem_size, E820_ACPI);
767 } else if (*arg == '$') {
768 start_at = memparse(arg+1, &arg);
769 add_memory_region(start_at, mem_size, E820_RESERVED);
771 limit_regions(mem_size);
772 user_defined_memmap = 1;
777 early_param("memmap", parse_memmap);
779 #ifdef CONFIG_PROC_VMCORE
780 /* elfcorehdr= specifies the location of elf core header
781 * stored by the crashed kernel.
783 static int __init parse_elfcorehdr(char *arg)
788 elfcorehdr_addr = memparse(arg, &arg);
791 early_param("elfcorehdr", parse_elfcorehdr);
792 #endif /* CONFIG_PROC_VMCORE */
795 * highmem=size forces highmem to be exactly 'size' bytes.
796 * This works even on boxes that have no highmem otherwise.
797 * This also works to reduce highmem size on bigger boxes.
799 static int __init parse_highmem(char *arg)
804 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
807 early_param("highmem", parse_highmem);
810 * vmalloc=size forces the vmalloc area to be exactly 'size'
811 * bytes. This can be used to increase (or decrease) the
812 * vmalloc area - the default is 128m.
814 static int __init parse_vmalloc(char *arg)
819 __VMALLOC_RESERVE = memparse(arg, &arg);
822 early_param("vmalloc", parse_vmalloc);
825 * Callback for efi_memory_walk.
828 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
830 unsigned long *max_pfn = arg, pfn;
833 pfn = PFN_UP(end -1);
841 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
843 memory_present(0, start, end);
848 * This function checks if the entire range <start,end> is mapped with type.
850 * Note: this function only works correct if the e820 table is sorted and
851 * not-overlapping, which is the case
854 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
859 for (i = 0; i < e820.nr_map; i++) {
860 struct e820entry *ei = &e820.map[i];
861 if (type && ei->type != type)
863 /* is the region (part) in overlap with the current region ?*/
864 if (ei->addr >= end || ei->addr + ei->size <= start)
866 /* if the region is at the beginning of <start,end> we move
867 * start to the end of the region since it's ok until there
869 if (ei->addr <= start)
870 start = ei->addr + ei->size;
871 /* if start is now at or beyond end, we're done, full
874 return 1; /* we're done */
880 * Find the highest page frame number we have available
882 void __init find_max_pfn(void)
888 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
889 efi_memmap_walk(efi_memory_present_wrapper, NULL);
893 for (i = 0; i < e820.nr_map; i++) {
894 unsigned long start, end;
896 if (e820.map[i].type != E820_RAM)
898 start = PFN_UP(e820.map[i].addr);
899 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
904 memory_present(0, start, end);
909 * Determine low and high memory ranges:
911 unsigned long __init find_max_low_pfn(void)
913 unsigned long max_low_pfn;
915 max_low_pfn = max_pfn;
916 if (max_low_pfn > MAXMEM_PFN) {
917 if (highmem_pages == -1)
918 highmem_pages = max_pfn - MAXMEM_PFN;
919 if (highmem_pages + MAXMEM_PFN < max_pfn)
920 max_pfn = MAXMEM_PFN + highmem_pages;
921 if (highmem_pages + MAXMEM_PFN > max_pfn) {
922 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
925 max_low_pfn = MAXMEM_PFN;
926 #ifndef CONFIG_HIGHMEM
927 /* Maximum memory usable is what is directly addressable */
928 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
930 if (max_pfn > MAX_NONPAE_PFN)
931 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
933 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
934 max_pfn = MAXMEM_PFN;
935 #else /* !CONFIG_HIGHMEM */
936 #ifndef CONFIG_X86_PAE
937 if (max_pfn > MAX_NONPAE_PFN) {
938 max_pfn = MAX_NONPAE_PFN;
939 printk(KERN_WARNING "Warning only 4GB will be used.\n");
940 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
942 #endif /* !CONFIG_X86_PAE */
943 #endif /* !CONFIG_HIGHMEM */
945 if (highmem_pages == -1)
947 #ifdef CONFIG_HIGHMEM
948 if (highmem_pages >= max_pfn) {
949 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
953 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
954 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
957 max_low_pfn -= highmem_pages;
961 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
968 * Free all available memory for boot time allocation. Used
969 * as a callback function by efi_memory_walk()
973 free_available_memory(unsigned long start, unsigned long end, void *arg)
975 /* check max_low_pfn */
976 if (start >= (max_low_pfn << PAGE_SHIFT))
978 if (end >= (max_low_pfn << PAGE_SHIFT))
979 end = max_low_pfn << PAGE_SHIFT;
981 free_bootmem(start, end - start);
986 * Register fully available low RAM pages with the bootmem allocator.
988 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
993 efi_memmap_walk(free_available_memory, NULL);
996 for (i = 0; i < e820.nr_map; i++) {
997 unsigned long curr_pfn, last_pfn, size;
999 * Reserve usable low memory
1001 if (e820.map[i].type != E820_RAM)
1004 * We are rounding up the start address of usable memory:
1006 curr_pfn = PFN_UP(e820.map[i].addr);
1007 if (curr_pfn >= max_low_pfn)
1010 * ... and at the end of the usable range downwards:
1012 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1014 if (last_pfn > max_low_pfn)
1015 last_pfn = max_low_pfn;
1018 * .. finally, did all the rounding and playing
1019 * around just make the area go away?
1021 if (last_pfn <= curr_pfn)
1024 size = last_pfn - curr_pfn;
1025 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1030 * workaround for Dell systems that neglect to reserve EBDA
1032 static void __init reserve_ebda_region(void)
1035 addr = get_bios_ebda();
1037 reserve_bootmem(addr, PAGE_SIZE);
1040 #ifndef CONFIG_NEED_MULTIPLE_NODES
1041 void __init setup_bootmem_allocator(void);
1042 static unsigned long __init setup_memory(void)
1045 * partially used pages are not usable - thus
1046 * we are rounding upwards:
1048 min_low_pfn = PFN_UP(init_pg_tables_end);
1052 max_low_pfn = find_max_low_pfn();
1054 #ifdef CONFIG_HIGHMEM
1055 highstart_pfn = highend_pfn = max_pfn;
1056 if (max_pfn > max_low_pfn) {
1057 highstart_pfn = max_low_pfn;
1059 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1060 pages_to_mb(highend_pfn - highstart_pfn));
1061 num_physpages = highend_pfn;
1062 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
1064 num_physpages = max_low_pfn;
1065 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
1067 #ifdef CONFIG_FLATMEM
1068 max_mapnr = num_physpages;
1070 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1071 pages_to_mb(max_low_pfn));
1073 setup_bootmem_allocator();
1078 void __init zone_sizes_init(void)
1080 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1081 unsigned int max_dma, low;
1083 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1087 zones_size[ZONE_DMA] = low;
1089 zones_size[ZONE_DMA] = max_dma;
1090 zones_size[ZONE_NORMAL] = low - max_dma;
1091 #ifdef CONFIG_HIGHMEM
1092 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1095 free_area_init(zones_size);
1098 extern unsigned long __init setup_memory(void);
1099 extern void zone_sizes_init(void);
1100 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1102 void __init setup_bootmem_allocator(void)
1104 unsigned long bootmap_size;
1106 * Initialize the boot-time allocator (with low memory only):
1108 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1110 register_bootmem_low_pages(max_low_pfn);
1113 * Reserve the bootmem bitmap itself as well. We do this in two
1114 * steps (first step was init_bootmem()) because this catches
1115 * the (very unlikely) case of us accidentally initializing the
1116 * bootmem allocator with an invalid RAM area.
1118 reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1119 bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1122 * reserve physical page 0 - it's a special BIOS page on many boxes,
1123 * enabling clean reboots, SMP operation, laptop functions.
1125 reserve_bootmem(0, PAGE_SIZE);
1127 /* reserve EBDA region, it's a 4K region */
1128 reserve_ebda_region();
1130 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1131 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1132 unless you have no PS/2 mouse plugged in. */
1133 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1134 boot_cpu_data.x86 == 6)
1135 reserve_bootmem(0xa0000 - 4096, 4096);
1139 * But first pinch a few for the stack/trampoline stuff
1140 * FIXME: Don't need the extra page at 4K, but need to fix
1141 * trampoline before removing it. (see the GDT stuff)
1143 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1145 #ifdef CONFIG_ACPI_SLEEP
1147 * Reserve low memory region for sleep support.
1149 acpi_reserve_bootmem();
1151 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1153 * Find and reserve possible boot-time SMP configuration:
1158 #ifdef CONFIG_BLK_DEV_INITRD
1159 if (LOADER_TYPE && INITRD_START) {
1160 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1161 reserve_bootmem(INITRD_START, INITRD_SIZE);
1163 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1164 initrd_end = initrd_start+INITRD_SIZE;
1167 printk(KERN_ERR "initrd extends beyond end of memory "
1168 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1169 INITRD_START + INITRD_SIZE,
1170 max_low_pfn << PAGE_SHIFT);
1176 if (crashk_res.start != crashk_res.end)
1177 reserve_bootmem(crashk_res.start,
1178 crashk_res.end - crashk_res.start + 1);
1183 * The node 0 pgdat is initialized before all of these because
1184 * it's needed for bootmem. node>0 pgdats have their virtual
1185 * space allocated before the pagetables are in place to access
1186 * them, so they can't be cleared then.
1188 * This should all compile down to nothing when NUMA is off.
1190 void __init remapped_pgdat_init(void)
1194 for_each_online_node(nid) {
1196 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1201 * Request address space for all standard RAM and ROM resources
1202 * and also for regions reported as reserved by the e820.
1205 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1210 for (i = 0; i < e820.nr_map; i++) {
1211 struct resource *res;
1212 #ifndef CONFIG_RESOURCES_64BIT
1213 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1216 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
1217 switch (e820.map[i].type) {
1218 case E820_RAM: res->name = "System RAM"; break;
1219 case E820_ACPI: res->name = "ACPI Tables"; break;
1220 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1221 default: res->name = "reserved";
1223 res->start = e820.map[i].addr;
1224 res->end = res->start + e820.map[i].size - 1;
1225 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1226 if (request_resource(&iomem_resource, res)) {
1230 if (e820.map[i].type == E820_RAM) {
1232 * We don't know which RAM region contains kernel data,
1233 * so we try it repeatedly and let the resource manager
1236 request_resource(res, code_resource);
1237 request_resource(res, data_resource);
1239 request_resource(res, &crashk_res);
1246 * Request address space for all standard resources
1248 * This is called just before pcibios_init(), which is also a
1249 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1251 static int __init request_standard_resources(void)
1255 printk("Setting up standard PCI resources\n");
1257 efi_initialize_iomem_resources(&code_resource, &data_resource);
1259 legacy_init_iomem_resources(&code_resource, &data_resource);
1261 /* EFI systems may still have VGA */
1262 request_resource(&iomem_resource, &video_ram_resource);
1264 /* request I/O space for devices used on all i[345]86 PCs */
1265 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1266 request_resource(&ioport_resource, &standard_io_resources[i]);
1270 subsys_initcall(request_standard_resources);
1272 static void __init register_memory(void)
1274 unsigned long gapstart, gapsize, round;
1275 unsigned long long last;
1279 * Search for the bigest gap in the low 32 bits of the e820
1282 last = 0x100000000ull;
1283 gapstart = 0x10000000;
1287 unsigned long long start = e820.map[i].addr;
1288 unsigned long long end = start + e820.map[i].size;
1291 * Since "last" is at most 4GB, we know we'll
1292 * fit in 32 bits if this condition is true
1295 unsigned long gap = last - end;
1297 if (gap > gapsize) {
1307 * See how much we want to round up: start off with
1308 * rounding to the next 1MB area.
1311 while ((gapsize >> 4) > round)
1313 /* Fun with two's complement */
1314 pci_mem_start = (gapstart + round) & -round;
1316 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1317 pci_mem_start, gapstart, gapsize);
1321 static void set_mca_bus(int x)
1326 static void set_mca_bus(int x) { }
1330 * Determine if we were loaded by an EFI loader. If so, then we have also been
1331 * passed the efi memmap, systab, etc., so we should use these data structures
1332 * for initialization. Note, the efi init code path is determined by the
1333 * global efi_enabled. This allows the same kernel image to be used on existing
1334 * systems (with a traditional BIOS) as well as on EFI systems.
1336 void __init setup_arch(char **cmdline_p)
1338 unsigned long max_low_pfn;
1340 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1341 pre_setup_arch_hook();
1345 * FIXME: This isn't an official loader_type right
1346 * now but does currently work with elilo.
1347 * If we were configured as an EFI kernel, check to make
1348 * sure that we were loaded correctly from elilo and that
1349 * the system table is valid. If not, then initialize normally.
1352 if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1356 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1357 drive_info = DRIVE_INFO;
1358 screen_info = SCREEN_INFO;
1359 edid_info = EDID_INFO;
1360 apm_info.bios = APM_BIOS_INFO;
1361 ist_info = IST_INFO;
1362 saved_videomode = VIDEO_MODE;
1363 if( SYS_DESC_TABLE.length != 0 ) {
1364 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1365 machine_id = SYS_DESC_TABLE.table[0];
1366 machine_submodel_id = SYS_DESC_TABLE.table[1];
1367 BIOS_revision = SYS_DESC_TABLE.table[2];
1369 bootloader_type = LOADER_TYPE;
1371 #ifdef CONFIG_BLK_DEV_RAM
1372 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1373 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1374 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1380 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1381 print_memory_map(machine_specific_memory_setup());
1386 if (!MOUNT_ROOT_RDONLY)
1387 root_mountflags &= ~MS_RDONLY;
1388 init_mm.start_code = (unsigned long) _text;
1389 init_mm.end_code = (unsigned long) _etext;
1390 init_mm.end_data = (unsigned long) _edata;
1391 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1393 code_resource.start = virt_to_phys(_text);
1394 code_resource.end = virt_to_phys(_etext)-1;
1395 data_resource.start = virt_to_phys(_etext);
1396 data_resource.end = virt_to_phys(_edata)-1;
1398 parse_early_param();
1400 if (user_defined_memmap) {
1401 printk(KERN_INFO "user-defined physical RAM map:\n");
1402 print_memory_map("user");
1405 strlcpy(command_line, saved_command_line, COMMAND_LINE_SIZE);
1406 *cmdline_p = command_line;
1408 max_low_pfn = setup_memory();
1411 * NOTE: before this point _nobody_ is allowed to allocate
1412 * any memory using the bootmem allocator. Although the
1413 * alloctor is now initialised only the first 8Mb of the kernel
1414 * virtual address space has been mapped. All allocations before
1415 * paging_init() has completed must use the alloc_bootmem_low_pages()
1416 * variant (which allocates DMA'able memory) and care must be taken
1417 * not to exceed the 8Mb limit.
1421 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1424 remapped_pgdat_init();
1429 * NOTE: at this point the bootmem allocator is fully available.
1434 #ifdef CONFIG_X86_GENERICARCH
1435 generic_apic_probe();
1442 * Parse the ACPI tables for possible boot-time SMP configuration.
1444 acpi_boot_table_init();
1447 #ifdef CONFIG_X86_IO_APIC
1448 check_acpi_pci(); /* Checks more than just ACPI actually */
1454 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1456 printk(KERN_WARNING "More than 8 CPUs detected and "
1457 "CONFIG_X86_PC cannot handle it.\nUse "
1458 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1461 #ifdef CONFIG_X86_LOCAL_APIC
1462 if (smp_found_config)
1469 #if defined(CONFIG_VGA_CONSOLE)
1470 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1471 conswitchp = &vga_con;
1472 #elif defined(CONFIG_DUMMY_CONSOLE)
1473 conswitchp = &dummy_con;
1479 static __init int add_pcspkr(void)
1481 struct platform_device *pd;
1484 pd = platform_device_alloc("pcspkr", -1);
1488 ret = platform_device_add(pd);
1490 platform_device_put(pd);
1494 device_initcall(add_pcspkr);
1499 * c-file-style:"k&r"