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>
48 #include <linux/dmi.h>
50 #include <video/edid.h>
54 #include <asm/mpspec.h>
55 #include <asm/setup.h>
56 #include <asm/arch_hooks.h>
57 #include <asm/sections.h>
58 #include <asm/io_apic.h>
61 #include "setup_arch_pre.h"
62 #include <bios_ebda.h>
64 /* Forward Declaration. */
65 void __init
find_max_pfn(void);
67 /* This value is set up by the early boot code to point to the value
68 immediately after the boot time page tables. It contains a *physical*
69 address, and must not be in the .bss segment! */
70 unsigned long init_pg_tables_end __initdata
= ~0UL;
72 int disable_pse __devinitdata
= 0;
80 EXPORT_SYMBOL(efi_enabled
);
83 /* cpu data as detected by the assembly code in head.S */
84 struct cpuinfo_x86 new_cpu_data __initdata
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
85 /* common cpu data for all cpus */
86 struct cpuinfo_x86 boot_cpu_data __read_mostly
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
87 EXPORT_SYMBOL(boot_cpu_data
);
89 unsigned long mmu_cr4_features
;
92 int acpi_disabled
= 0;
94 int acpi_disabled
= 1;
96 EXPORT_SYMBOL(acpi_disabled
);
99 int __initdata acpi_force
= 0;
100 extern acpi_interrupt_flags acpi_sci_flags
;
103 /* for MCA, but anyone else can use it if they want */
104 unsigned int machine_id
;
106 EXPORT_SYMBOL(machine_id
);
108 unsigned int machine_submodel_id
;
109 unsigned int BIOS_revision
;
110 unsigned int mca_pentium_flag
;
112 /* For PCI or other memory-mapped resources */
113 unsigned long pci_mem_start
= 0x10000000;
115 EXPORT_SYMBOL(pci_mem_start
);
118 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
121 /* user-defined highmem size */
122 static unsigned int highmem_pages
= -1;
127 struct drive_info_struct
{ char dummy
[32]; } drive_info
;
128 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
129 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
130 EXPORT_SYMBOL(drive_info
);
132 struct screen_info screen_info
;
133 EXPORT_SYMBOL(screen_info
);
134 struct apm_info apm_info
;
135 EXPORT_SYMBOL(apm_info
);
136 struct sys_desc_table_struct
{
137 unsigned short length
;
138 unsigned char table
[0];
140 struct edid_info edid_info
;
141 EXPORT_SYMBOL_GPL(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 generic_apic_probe(char *);
151 extern int root_mountflags
;
153 unsigned long saved_videomode
;
155 #define RAMDISK_IMAGE_START_MASK 0x07FF
156 #define RAMDISK_PROMPT_FLAG 0x8000
157 #define RAMDISK_LOAD_FLAG 0x4000
159 static char command_line
[COMMAND_LINE_SIZE
];
161 unsigned char __initdata boot_params
[PARAM_SIZE
];
163 static struct resource data_resource
= {
164 .name
= "Kernel data",
167 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
170 static struct resource code_resource
= {
171 .name
= "Kernel code",
174 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
177 static struct resource system_rom_resource
= {
178 .name
= "System ROM",
181 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
184 static struct resource extension_rom_resource
= {
185 .name
= "Extension ROM",
188 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
191 static struct resource adapter_rom_resources
[] = { {
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
217 .name
= "Adapter ROM",
220 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
223 #define ADAPTER_ROM_RESOURCES \
224 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
226 static struct resource video_rom_resource
= {
230 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
233 static struct resource video_ram_resource
= {
234 .name
= "Video RAM area",
237 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
240 static struct resource standard_io_resources
[] = { {
244 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
249 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
254 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
259 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
264 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
266 .name
= "dma page reg",
269 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
274 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
279 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
284 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
287 #define STANDARD_IO_RESOURCES \
288 (sizeof standard_io_resources / sizeof standard_io_resources[0])
290 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
292 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
294 unsigned char *p
, sum
= 0;
296 for (p
= rom
; p
< rom
+ length
; p
++)
301 static void __init
probe_roms(void)
303 unsigned long start
, length
, upper
;
308 upper
= adapter_rom_resources
[0].start
;
309 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
310 rom
= isa_bus_to_virt(start
);
311 if (!romsignature(rom
))
314 video_rom_resource
.start
= start
;
316 /* 0 < length <= 0x7f * 512, historically */
317 length
= rom
[2] * 512;
319 /* if checksum okay, trust length byte */
320 if (length
&& romchecksum(rom
, length
))
321 video_rom_resource
.end
= start
+ length
- 1;
323 request_resource(&iomem_resource
, &video_rom_resource
);
327 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
332 request_resource(&iomem_resource
, &system_rom_resource
);
333 upper
= system_rom_resource
.start
;
335 /* check for extension rom (ignore length byte!) */
336 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
337 if (romsignature(rom
)) {
338 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
339 if (romchecksum(rom
, length
)) {
340 request_resource(&iomem_resource
, &extension_rom_resource
);
341 upper
= extension_rom_resource
.start
;
345 /* check for adapter roms on 2k boundaries */
346 for (i
= 0; i
< ADAPTER_ROM_RESOURCES
&& start
< upper
; start
+= 2048) {
347 rom
= isa_bus_to_virt(start
);
348 if (!romsignature(rom
))
351 /* 0 < length <= 0x7f * 512, historically */
352 length
= rom
[2] * 512;
354 /* but accept any length that fits if checksum okay */
355 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
358 adapter_rom_resources
[i
].start
= start
;
359 adapter_rom_resources
[i
].end
= start
+ length
- 1;
360 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
362 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
366 static void __init
limit_regions(unsigned long long size
)
368 unsigned long long current_addr
= 0;
372 efi_memory_desc_t
*md
;
375 for (p
= memmap
.map
, i
= 0; p
< memmap
.map_end
;
376 p
+= memmap
.desc_size
, i
++) {
378 current_addr
= md
->phys_addr
+ (md
->num_pages
<< 12);
379 if (md
->type
== EFI_CONVENTIONAL_MEMORY
) {
380 if (current_addr
>= size
) {
382 (((current_addr
-size
) + PAGE_SIZE
-1) >> PAGE_SHIFT
);
383 memmap
.nr_map
= i
+ 1;
389 for (i
= 0; i
< e820
.nr_map
; i
++) {
390 current_addr
= e820
.map
[i
].addr
+ e820
.map
[i
].size
;
391 if (current_addr
< size
)
394 if (e820
.map
[i
].type
!= E820_RAM
)
397 if (e820
.map
[i
].addr
>= size
) {
399 * This region starts past the end of the
400 * requested size, skip it completely.
405 e820
.map
[i
].size
-= current_addr
- size
;
411 static void __init
add_memory_region(unsigned long long start
,
412 unsigned long long size
, int type
)
420 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
424 e820
.map
[x
].addr
= start
;
425 e820
.map
[x
].size
= size
;
426 e820
.map
[x
].type
= type
;
429 } /* add_memory_region */
433 static void __init
print_memory_map(char *who
)
437 for (i
= 0; i
< e820
.nr_map
; i
++) {
438 printk(" %s: %016Lx - %016Lx ", who
,
440 e820
.map
[i
].addr
+ e820
.map
[i
].size
);
441 switch (e820
.map
[i
].type
) {
442 case E820_RAM
: printk("(usable)\n");
445 printk("(reserved)\n");
448 printk("(ACPI data)\n");
451 printk("(ACPI NVS)\n");
453 default: printk("type %lu\n", e820
.map
[i
].type
);
460 * Sanitize the BIOS e820 map.
462 * Some e820 responses include overlapping entries. The following
463 * replaces the original e820 map with a new one, removing overlaps.
466 struct change_member
{
467 struct e820entry
*pbios
; /* pointer to original bios entry */
468 unsigned long long addr
; /* address for this change point */
470 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
471 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
472 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
473 static struct e820entry new_bios
[E820MAX
] __initdata
;
475 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
477 struct change_member
*change_tmp
;
478 unsigned long current_type
, last_type
;
479 unsigned long long last_addr
;
480 int chgidx
, still_changing
;
483 int old_nr
, new_nr
, chg_nr
;
487 Visually we're performing the following (1,2,3,4 = memory types)...
489 Sample memory map (w/overlaps):
490 ____22__________________
491 ______________________4_
492 ____1111________________
493 _44_____________________
494 11111111________________
495 ____________________33__
496 ___________44___________
497 __________33333_________
498 ______________22________
499 ___________________2222_
500 _________111111111______
501 _____________________11_
502 _________________4______
504 Sanitized equivalent (no overlap):
505 1_______________________
506 _44_____________________
507 ___1____________________
508 ____22__________________
509 ______11________________
510 _________1______________
511 __________3_____________
512 ___________44___________
513 _____________33_________
514 _______________2________
515 ________________1_______
516 _________________4______
517 ___________________2____
518 ____________________33__
519 ______________________4_
522 /* if there's only one memory region, don't bother */
528 /* bail out if we find any unreasonable addresses in bios map */
529 for (i
=0; i
<old_nr
; i
++)
530 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
533 /* create pointers for initial change-point information (for sorting) */
534 for (i
=0; i
< 2*old_nr
; i
++)
535 change_point
[i
] = &change_point_list
[i
];
537 /* record all known change-points (starting and ending addresses),
538 omitting those that are for empty memory regions */
540 for (i
=0; i
< old_nr
; i
++) {
541 if (biosmap
[i
].size
!= 0) {
542 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
543 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
544 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
545 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
548 chg_nr
= chgidx
; /* true number of change-points */
550 /* sort change-point list by memory addresses (low -> high) */
552 while (still_changing
) {
554 for (i
=1; i
< chg_nr
; i
++) {
555 /* if <current_addr> > <last_addr>, swap */
556 /* or, if current=<start_addr> & last=<end_addr>, swap */
557 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
558 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
559 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
560 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
563 change_tmp
= change_point
[i
];
564 change_point
[i
] = change_point
[i
-1];
565 change_point
[i
-1] = change_tmp
;
571 /* create a new bios memory map, removing overlaps */
572 overlap_entries
=0; /* number of entries in the overlap table */
573 new_bios_entry
=0; /* index for creating new bios map entries */
574 last_type
= 0; /* start with undefined memory type */
575 last_addr
= 0; /* start with 0 as last starting address */
576 /* loop through change-points, determining affect on the new bios map */
577 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
579 /* keep track of all overlapping bios entries */
580 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
582 /* add map entry to overlap list (> 1 entry implies an overlap) */
583 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
587 /* remove entry from list (order independent, so swap with last) */
588 for (i
=0; i
<overlap_entries
; i
++)
590 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
591 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
595 /* if there are overlapping entries, decide which "type" to use */
596 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
598 for (i
=0; i
<overlap_entries
; i
++)
599 if (overlap_list
[i
]->type
> current_type
)
600 current_type
= overlap_list
[i
]->type
;
601 /* continue building up new bios map based on this information */
602 if (current_type
!= last_type
) {
603 if (last_type
!= 0) {
604 new_bios
[new_bios_entry
].size
=
605 change_point
[chgidx
]->addr
- last_addr
;
606 /* move forward only if the new size was non-zero */
607 if (new_bios
[new_bios_entry
].size
!= 0)
608 if (++new_bios_entry
>= E820MAX
)
609 break; /* no more space left for new bios entries */
611 if (current_type
!= 0) {
612 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
613 new_bios
[new_bios_entry
].type
= current_type
;
614 last_addr
=change_point
[chgidx
]->addr
;
616 last_type
= current_type
;
619 new_nr
= new_bios_entry
; /* retain count for new bios entries */
621 /* copy new bios mapping into original location */
622 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
629 * Copy the BIOS e820 map into a safe place.
631 * Sanity-check it while we're at it..
633 * If we're lucky and live on a modern system, the setup code
634 * will have given us a memory map that we can use to properly
635 * set up memory. If we aren't, we'll fake a memory map.
637 * We check to see that the memory map contains at least 2 elements
638 * before we'll use it, because the detection code in setup.S may
639 * not be perfect and most every PC known to man has two memory
640 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
641 * thinkpad 560x, for example, does not cooperate with the memory
644 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
646 /* Only one memory region (or negative)? Ignore it */
651 unsigned long long start
= biosmap
->addr
;
652 unsigned long long size
= biosmap
->size
;
653 unsigned long long end
= start
+ size
;
654 unsigned long type
= biosmap
->type
;
656 /* Overflow in 64 bits? Ignore the memory map. */
661 * Some BIOSes claim RAM in the 640k - 1M region.
662 * Not right. Fix it up.
664 if (type
== E820_RAM
) {
665 if (start
< 0x100000ULL
&& end
> 0xA0000ULL
) {
666 if (start
< 0xA0000ULL
)
667 add_memory_region(start
, 0xA0000ULL
-start
, type
);
668 if (end
<= 0x100000ULL
)
674 add_memory_region(start
, size
, type
);
675 } while (biosmap
++,--nr_map
);
679 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
681 #ifdef CONFIG_EDD_MODULE
685 * copy_edd() - Copy the BIOS EDD information
686 * from boot_params into a safe place.
689 static inline void copy_edd(void)
691 memcpy(edd
.mbr_signature
, EDD_MBR_SIGNATURE
, sizeof(edd
.mbr_signature
));
692 memcpy(edd
.edd_info
, EDD_BUF
, sizeof(edd
.edd_info
));
693 edd
.mbr_signature_nr
= EDD_MBR_SIG_NR
;
694 edd
.edd_info_nr
= EDD_NR
;
697 static inline void copy_edd(void)
703 * Do NOT EVER look at the BIOS memory size location.
704 * It does not work on many machines.
706 #define LOWMEMSIZE() (0x9f000)
708 static void __init
parse_cmdline_early (char ** cmdline_p
)
710 char c
= ' ', *to
= command_line
, *from
= saved_command_line
;
714 /* Save unparsed command line copy for /proc/cmdline */
715 saved_command_line
[COMMAND_LINE_SIZE
-1] = '\0';
721 * "mem=nopentium" disables the 4MB page tables.
722 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
723 * to <mem>, overriding the bios size.
724 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
725 * <start> to <start>+<mem>, overriding the bios size.
727 * HPA tells me bootloaders need to parse mem=, so no new
728 * option should be mem= [also see Documentation/i386/boot.txt]
730 if (!memcmp(from
, "mem=", 4)) {
731 if (to
!= command_line
)
733 if (!memcmp(from
+4, "nopentium", 9)) {
735 clear_bit(X86_FEATURE_PSE
, boot_cpu_data
.x86_capability
);
738 /* If the user specifies memory size, we
739 * limit the BIOS-provided memory map to
740 * that size. exactmap can be used to specify
741 * the exact map. mem=number can be used to
742 * trim the existing memory map.
744 unsigned long long mem_size
;
746 mem_size
= memparse(from
+4, &from
);
747 limit_regions(mem_size
);
752 else if (!memcmp(from
, "memmap=", 7)) {
753 if (to
!= command_line
)
755 if (!memcmp(from
+7, "exactmap", 8)) {
756 #ifdef CONFIG_CRASH_DUMP
757 /* If we are doing a crash dump, we
758 * still need to know the real mem
759 * size before original memory map is
763 saved_max_pfn
= max_pfn
;
769 /* If the user specifies memory size, we
770 * limit the BIOS-provided memory map to
771 * that size. exactmap can be used to specify
772 * the exact map. mem=number can be used to
773 * trim the existing memory map.
775 unsigned long long start_at
, mem_size
;
777 mem_size
= memparse(from
+7, &from
);
779 start_at
= memparse(from
+1, &from
);
780 add_memory_region(start_at
, mem_size
, E820_RAM
);
781 } else if (*from
== '#') {
782 start_at
= memparse(from
+1, &from
);
783 add_memory_region(start_at
, mem_size
, E820_ACPI
);
784 } else if (*from
== '$') {
785 start_at
= memparse(from
+1, &from
);
786 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
788 limit_regions(mem_size
);
794 else if (!memcmp(from
, "noexec=", 7))
795 noexec_setup(from
+ 7);
798 #ifdef CONFIG_X86_SMP
800 * If the BIOS enumerates physical processors before logical,
801 * maxcpus=N at enumeration-time can be used to disable HT.
803 else if (!memcmp(from
, "maxcpus=", 8)) {
804 extern unsigned int maxcpus
;
806 maxcpus
= simple_strtoul(from
+ 8, NULL
, 0);
811 /* "acpi=off" disables both ACPI table parsing and interpreter */
812 else if (!memcmp(from
, "acpi=off", 8)) {
816 /* acpi=force to over-ride black-list */
817 else if (!memcmp(from
, "acpi=force", 10)) {
823 /* acpi=strict disables out-of-spec workarounds */
824 else if (!memcmp(from
, "acpi=strict", 11)) {
828 /* Limit ACPI just to boot-time to enable HT */
829 else if (!memcmp(from
, "acpi=ht", 7)) {
835 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
836 else if (!memcmp(from
, "pci=noacpi", 10)) {
839 /* "acpi=noirq" disables ACPI interrupt routing */
840 else if (!memcmp(from
, "acpi=noirq", 10)) {
844 else if (!memcmp(from
, "acpi_sci=edge", 13))
845 acpi_sci_flags
.trigger
= 1;
847 else if (!memcmp(from
, "acpi_sci=level", 14))
848 acpi_sci_flags
.trigger
= 3;
850 else if (!memcmp(from
, "acpi_sci=high", 13))
851 acpi_sci_flags
.polarity
= 1;
853 else if (!memcmp(from
, "acpi_sci=low", 12))
854 acpi_sci_flags
.polarity
= 3;
856 #ifdef CONFIG_X86_IO_APIC
857 else if (!memcmp(from
, "acpi_skip_timer_override", 24))
858 acpi_skip_timer_override
= 1;
860 if (!memcmp(from
, "disable_timer_pin_1", 19))
861 disable_timer_pin_1
= 1;
862 if (!memcmp(from
, "enable_timer_pin_1", 18))
863 disable_timer_pin_1
= -1;
865 /* disable IO-APIC */
866 else if (!memcmp(from
, "noapic", 6))
867 disable_ioapic_setup();
868 #endif /* CONFIG_X86_IO_APIC */
869 #endif /* CONFIG_ACPI */
871 #ifdef CONFIG_X86_LOCAL_APIC
872 /* enable local APIC */
873 else if (!memcmp(from
, "lapic", 5))
876 /* disable local APIC */
877 else if (!memcmp(from
, "nolapic", 6))
879 #endif /* CONFIG_X86_LOCAL_APIC */
882 /* crashkernel=size@addr specifies the location to reserve for
883 * a crash kernel. By reserving this memory we guarantee
884 * that linux never set's it up as a DMA target.
885 * Useful for holding code to do something appropriate
886 * after a kernel panic.
888 else if (!memcmp(from
, "crashkernel=", 12)) {
889 unsigned long size
, base
;
890 size
= memparse(from
+12, &from
);
892 base
= memparse(from
+1, &from
);
893 /* FIXME: Do I want a sanity check
894 * to validate the memory range?
896 crashk_res
.start
= base
;
897 crashk_res
.end
= base
+ size
- 1;
901 #ifdef CONFIG_PROC_VMCORE
902 /* elfcorehdr= specifies the location of elf core header
903 * stored by the crashed kernel.
905 else if (!memcmp(from
, "elfcorehdr=", 11))
906 elfcorehdr_addr
= memparse(from
+11, &from
);
910 * highmem=size forces highmem to be exactly 'size' bytes.
911 * This works even on boxes that have no highmem otherwise.
912 * This also works to reduce highmem size on bigger boxes.
914 else if (!memcmp(from
, "highmem=", 8))
915 highmem_pages
= memparse(from
+8, &from
) >> PAGE_SHIFT
;
918 * vmalloc=size forces the vmalloc area to be exactly 'size'
919 * bytes. This can be used to increase (or decrease) the
920 * vmalloc area - the default is 128m.
922 else if (!memcmp(from
, "vmalloc=", 8))
923 __VMALLOC_RESERVE
= memparse(from
+8, &from
);
929 if (COMMAND_LINE_SIZE
<= ++len
)
934 *cmdline_p
= command_line
;
936 printk(KERN_INFO
"user-defined physical RAM map:\n");
937 print_memory_map("user");
942 * Callback for efi_memory_walk.
945 efi_find_max_pfn(unsigned long start
, unsigned long end
, void *arg
)
947 unsigned long *max_pfn
= arg
, pfn
;
950 pfn
= PFN_UP(end
-1);
958 efi_memory_present_wrapper(unsigned long start
, unsigned long end
, void *arg
)
960 memory_present(0, start
, end
);
965 * Find the highest page frame number we have available
967 void __init
find_max_pfn(void)
973 efi_memmap_walk(efi_find_max_pfn
, &max_pfn
);
974 efi_memmap_walk(efi_memory_present_wrapper
, NULL
);
978 for (i
= 0; i
< e820
.nr_map
; i
++) {
979 unsigned long start
, end
;
981 if (e820
.map
[i
].type
!= E820_RAM
)
983 start
= PFN_UP(e820
.map
[i
].addr
);
984 end
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
989 memory_present(0, start
, end
);
994 * Determine low and high memory ranges:
996 unsigned long __init
find_max_low_pfn(void)
998 unsigned long max_low_pfn
;
1000 max_low_pfn
= max_pfn
;
1001 if (max_low_pfn
> MAXMEM_PFN
) {
1002 if (highmem_pages
== -1)
1003 highmem_pages
= max_pfn
- MAXMEM_PFN
;
1004 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
1005 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
1006 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
1007 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn
- MAXMEM_PFN
), pages_to_mb(highmem_pages
));
1010 max_low_pfn
= MAXMEM_PFN
;
1011 #ifndef CONFIG_HIGHMEM
1012 /* Maximum memory usable is what is directly addressable */
1013 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
1015 if (max_pfn
> MAX_NONPAE_PFN
)
1016 printk(KERN_WARNING
"Use a PAE enabled kernel.\n");
1018 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
1019 max_pfn
= MAXMEM_PFN
;
1020 #else /* !CONFIG_HIGHMEM */
1021 #ifndef CONFIG_X86_PAE
1022 if (max_pfn
> MAX_NONPAE_PFN
) {
1023 max_pfn
= MAX_NONPAE_PFN
;
1024 printk(KERN_WARNING
"Warning only 4GB will be used.\n");
1025 printk(KERN_WARNING
"Use a PAE enabled kernel.\n");
1027 #endif /* !CONFIG_X86_PAE */
1028 #endif /* !CONFIG_HIGHMEM */
1030 if (highmem_pages
== -1)
1032 #ifdef CONFIG_HIGHMEM
1033 if (highmem_pages
>= max_pfn
) {
1034 printk(KERN_ERR
"highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
1037 if (highmem_pages
) {
1038 if (max_low_pfn
-highmem_pages
< 64*1024*1024/PAGE_SIZE
){
1039 printk(KERN_ERR
"highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages
));
1042 max_low_pfn
-= highmem_pages
;
1046 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
1053 * Free all available memory for boot time allocation. Used
1054 * as a callback function by efi_memory_walk()
1058 free_available_memory(unsigned long start
, unsigned long end
, void *arg
)
1060 /* check max_low_pfn */
1061 if (start
>= ((max_low_pfn
+ 1) << PAGE_SHIFT
))
1063 if (end
>= ((max_low_pfn
+ 1) << PAGE_SHIFT
))
1064 end
= (max_low_pfn
+ 1) << PAGE_SHIFT
;
1066 free_bootmem(start
, end
- start
);
1071 * Register fully available low RAM pages with the bootmem allocator.
1073 static void __init
register_bootmem_low_pages(unsigned long max_low_pfn
)
1078 efi_memmap_walk(free_available_memory
, NULL
);
1081 for (i
= 0; i
< e820
.nr_map
; i
++) {
1082 unsigned long curr_pfn
, last_pfn
, size
;
1084 * Reserve usable low memory
1086 if (e820
.map
[i
].type
!= E820_RAM
)
1089 * We are rounding up the start address of usable memory:
1091 curr_pfn
= PFN_UP(e820
.map
[i
].addr
);
1092 if (curr_pfn
>= max_low_pfn
)
1095 * ... and at the end of the usable range downwards:
1097 last_pfn
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
1099 if (last_pfn
> max_low_pfn
)
1100 last_pfn
= max_low_pfn
;
1103 * .. finally, did all the rounding and playing
1104 * around just make the area go away?
1106 if (last_pfn
<= curr_pfn
)
1109 size
= last_pfn
- curr_pfn
;
1110 free_bootmem(PFN_PHYS(curr_pfn
), PFN_PHYS(size
));
1115 * workaround for Dell systems that neglect to reserve EBDA
1117 static void __init
reserve_ebda_region(void)
1120 addr
= get_bios_ebda();
1122 reserve_bootmem(addr
, PAGE_SIZE
);
1125 #ifndef CONFIG_NEED_MULTIPLE_NODES
1126 void __init
setup_bootmem_allocator(void);
1127 static unsigned long __init
setup_memory(void)
1130 * partially used pages are not usable - thus
1131 * we are rounding upwards:
1133 min_low_pfn
= PFN_UP(init_pg_tables_end
);
1137 max_low_pfn
= find_max_low_pfn();
1139 #ifdef CONFIG_HIGHMEM
1140 highstart_pfn
= highend_pfn
= max_pfn
;
1141 if (max_pfn
> max_low_pfn
) {
1142 highstart_pfn
= max_low_pfn
;
1144 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
1145 pages_to_mb(highend_pfn
- highstart_pfn
));
1147 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
1148 pages_to_mb(max_low_pfn
));
1150 setup_bootmem_allocator();
1155 void __init
zone_sizes_init(void)
1157 unsigned long zones_size
[MAX_NR_ZONES
] = {0, 0, 0};
1158 unsigned int max_dma
, low
;
1160 max_dma
= virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
1164 zones_size
[ZONE_DMA
] = low
;
1166 zones_size
[ZONE_DMA
] = max_dma
;
1167 zones_size
[ZONE_NORMAL
] = low
- max_dma
;
1168 #ifdef CONFIG_HIGHMEM
1169 zones_size
[ZONE_HIGHMEM
] = highend_pfn
- low
;
1172 free_area_init(zones_size
);
1175 extern unsigned long __init
setup_memory(void);
1176 extern void zone_sizes_init(void);
1177 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1179 void __init
setup_bootmem_allocator(void)
1181 unsigned long bootmap_size
;
1183 * Initialize the boot-time allocator (with low memory only):
1185 bootmap_size
= init_bootmem(min_low_pfn
, max_low_pfn
);
1187 register_bootmem_low_pages(max_low_pfn
);
1190 * Reserve the bootmem bitmap itself as well. We do this in two
1191 * steps (first step was init_bootmem()) because this catches
1192 * the (very unlikely) case of us accidentally initializing the
1193 * bootmem allocator with an invalid RAM area.
1195 reserve_bootmem(__PHYSICAL_START
, (PFN_PHYS(min_low_pfn
) +
1196 bootmap_size
+ PAGE_SIZE
-1) - (__PHYSICAL_START
));
1199 * reserve physical page 0 - it's a special BIOS page on many boxes,
1200 * enabling clean reboots, SMP operation, laptop functions.
1202 reserve_bootmem(0, PAGE_SIZE
);
1204 /* reserve EBDA region, it's a 4K region */
1205 reserve_ebda_region();
1207 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1208 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1209 unless you have no PS/2 mouse plugged in. */
1210 if (boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
1211 boot_cpu_data
.x86
== 6)
1212 reserve_bootmem(0xa0000 - 4096, 4096);
1216 * But first pinch a few for the stack/trampoline stuff
1217 * FIXME: Don't need the extra page at 4K, but need to fix
1218 * trampoline before removing it. (see the GDT stuff)
1220 reserve_bootmem(PAGE_SIZE
, PAGE_SIZE
);
1222 #ifdef CONFIG_ACPI_SLEEP
1224 * Reserve low memory region for sleep support.
1226 acpi_reserve_bootmem();
1228 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1230 * Find and reserve possible boot-time SMP configuration:
1235 #ifdef CONFIG_BLK_DEV_INITRD
1236 if (LOADER_TYPE
&& INITRD_START
) {
1237 if (INITRD_START
+ INITRD_SIZE
<= (max_low_pfn
<< PAGE_SHIFT
)) {
1238 reserve_bootmem(INITRD_START
, INITRD_SIZE
);
1240 INITRD_START
? INITRD_START
+ PAGE_OFFSET
: 0;
1241 initrd_end
= initrd_start
+INITRD_SIZE
;
1244 printk(KERN_ERR
"initrd extends beyond end of memory "
1245 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1246 INITRD_START
+ INITRD_SIZE
,
1247 max_low_pfn
<< PAGE_SHIFT
);
1253 if (crashk_res
.start
!= crashk_res
.end
)
1254 reserve_bootmem(crashk_res
.start
,
1255 crashk_res
.end
- crashk_res
.start
+ 1);
1260 * The node 0 pgdat is initialized before all of these because
1261 * it's needed for bootmem. node>0 pgdats have their virtual
1262 * space allocated before the pagetables are in place to access
1263 * them, so they can't be cleared then.
1265 * This should all compile down to nothing when NUMA is off.
1267 void __init
remapped_pgdat_init(void)
1271 for_each_online_node(nid
) {
1273 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
1278 * Request address space for all standard RAM and ROM resources
1279 * and also for regions reported as reserved by the e820.
1282 legacy_init_iomem_resources(struct resource
*code_resource
, struct resource
*data_resource
)
1287 for (i
= 0; i
< e820
.nr_map
; i
++) {
1288 struct resource
*res
;
1289 if (e820
.map
[i
].addr
+ e820
.map
[i
].size
> 0x100000000ULL
)
1291 res
= kzalloc(sizeof(struct resource
), GFP_ATOMIC
);
1292 switch (e820
.map
[i
].type
) {
1293 case E820_RAM
: res
->name
= "System RAM"; break;
1294 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
1295 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
1296 default: res
->name
= "reserved";
1298 res
->start
= e820
.map
[i
].addr
;
1299 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
1300 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
1301 request_resource(&iomem_resource
, res
);
1302 if (e820
.map
[i
].type
== E820_RAM
) {
1304 * We don't know which RAM region contains kernel data,
1305 * so we try it repeatedly and let the resource manager
1308 request_resource(res
, code_resource
);
1309 request_resource(res
, data_resource
);
1311 request_resource(res
, &crashk_res
);
1318 * Request address space for all standard resources
1320 * This is called just before pcibios_assign_resources(), which is also
1321 * an fs_initcall, but is linked in later (in arch/i386/pci/i386.c).
1323 static int __init
request_standard_resources(void)
1327 printk("Setting up standard PCI resources\n");
1329 efi_initialize_iomem_resources(&code_resource
, &data_resource
);
1331 legacy_init_iomem_resources(&code_resource
, &data_resource
);
1333 /* EFI systems may still have VGA */
1334 request_resource(&iomem_resource
, &video_ram_resource
);
1336 /* request I/O space for devices used on all i[345]86 PCs */
1337 for (i
= 0; i
< STANDARD_IO_RESOURCES
; i
++)
1338 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
1342 fs_initcall(request_standard_resources
);
1344 static void __init
register_memory(void)
1346 unsigned long gapstart
, gapsize
, round
;
1347 unsigned long long last
;
1351 * Search for the bigest gap in the low 32 bits of the e820
1354 last
= 0x100000000ull
;
1355 gapstart
= 0x10000000;
1359 unsigned long long start
= e820
.map
[i
].addr
;
1360 unsigned long long end
= start
+ e820
.map
[i
].size
;
1363 * Since "last" is at most 4GB, we know we'll
1364 * fit in 32 bits if this condition is true
1367 unsigned long gap
= last
- end
;
1369 if (gap
> gapsize
) {
1379 * See how much we want to round up: start off with
1380 * rounding to the next 1MB area.
1383 while ((gapsize
>> 4) > round
)
1385 /* Fun with two's complement */
1386 pci_mem_start
= (gapstart
+ round
) & -round
;
1388 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1389 pci_mem_start
, gapstart
, gapsize
);
1392 static char * __init
machine_specific_memory_setup(void);
1395 static void set_mca_bus(int x
)
1400 static void set_mca_bus(int x
) { }
1404 * Determine if we were loaded by an EFI loader. If so, then we have also been
1405 * passed the efi memmap, systab, etc., so we should use these data structures
1406 * for initialization. Note, the efi init code path is determined by the
1407 * global efi_enabled. This allows the same kernel image to be used on existing
1408 * systems (with a traditional BIOS) as well as on EFI systems.
1410 void __init
setup_arch(char **cmdline_p
)
1412 unsigned long max_low_pfn
;
1414 memcpy(&boot_cpu_data
, &new_cpu_data
, sizeof(new_cpu_data
));
1415 pre_setup_arch_hook();
1419 * FIXME: This isn't an official loader_type right
1420 * now but does currently work with elilo.
1421 * If we were configured as an EFI kernel, check to make
1422 * sure that we were loaded correctly from elilo and that
1423 * the system table is valid. If not, then initialize normally.
1426 if ((LOADER_TYPE
== 0x50) && EFI_SYSTAB
)
1430 ROOT_DEV
= old_decode_dev(ORIG_ROOT_DEV
);
1431 drive_info
= DRIVE_INFO
;
1432 screen_info
= SCREEN_INFO
;
1433 edid_info
= EDID_INFO
;
1434 apm_info
.bios
= APM_BIOS_INFO
;
1435 ist_info
= IST_INFO
;
1436 saved_videomode
= VIDEO_MODE
;
1437 if( SYS_DESC_TABLE
.length
!= 0 ) {
1438 set_mca_bus(SYS_DESC_TABLE
.table
[3] & 0x2);
1439 machine_id
= SYS_DESC_TABLE
.table
[0];
1440 machine_submodel_id
= SYS_DESC_TABLE
.table
[1];
1441 BIOS_revision
= SYS_DESC_TABLE
.table
[2];
1443 bootloader_type
= LOADER_TYPE
;
1445 #ifdef CONFIG_BLK_DEV_RAM
1446 rd_image_start
= RAMDISK_FLAGS
& RAMDISK_IMAGE_START_MASK
;
1447 rd_prompt
= ((RAMDISK_FLAGS
& RAMDISK_PROMPT_FLAG
) != 0);
1448 rd_doload
= ((RAMDISK_FLAGS
& RAMDISK_LOAD_FLAG
) != 0);
1454 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
1455 print_memory_map(machine_specific_memory_setup());
1460 if (!MOUNT_ROOT_RDONLY
)
1461 root_mountflags
&= ~MS_RDONLY
;
1462 init_mm
.start_code
= (unsigned long) _text
;
1463 init_mm
.end_code
= (unsigned long) _etext
;
1464 init_mm
.end_data
= (unsigned long) _edata
;
1465 init_mm
.brk
= init_pg_tables_end
+ PAGE_OFFSET
;
1467 code_resource
.start
= virt_to_phys(_text
);
1468 code_resource
.end
= virt_to_phys(_etext
)-1;
1469 data_resource
.start
= virt_to_phys(_etext
);
1470 data_resource
.end
= virt_to_phys(_edata
)-1;
1472 parse_cmdline_early(cmdline_p
);
1474 #ifdef CONFIG_EARLY_PRINTK
1476 char *s
= strstr(*cmdline_p
, "earlyprintk=");
1478 setup_early_printk(strchr(s
, '=') + 1);
1479 printk("early console enabled\n");
1484 max_low_pfn
= setup_memory();
1487 * NOTE: before this point _nobody_ is allowed to allocate
1488 * any memory using the bootmem allocator. Although the
1489 * alloctor is now initialised only the first 8Mb of the kernel
1490 * virtual address space has been mapped. All allocations before
1491 * paging_init() has completed must use the alloc_bootmem_low_pages()
1492 * variant (which allocates DMA'able memory) and care must be taken
1493 * not to exceed the 8Mb limit.
1497 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1500 remapped_pgdat_init();
1505 * NOTE: at this point the bootmem allocator is fully available.
1510 #ifdef CONFIG_X86_GENERICARCH
1511 generic_apic_probe(*cmdline_p
);
1516 #ifdef CONFIG_X86_IO_APIC
1517 check_acpi_pci(); /* Checks more than just ACPI actually */
1522 * Parse the ACPI tables for possible boot-time SMP configuration.
1524 acpi_boot_table_init();
1527 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1529 printk(KERN_WARNING
"More than 8 CPUs detected and "
1530 "CONFIG_X86_PC cannot handle it.\nUse "
1531 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1534 #ifdef CONFIG_X86_LOCAL_APIC
1535 if (smp_found_config
)
1542 #if defined(CONFIG_VGA_CONSOLE)
1543 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
1544 conswitchp
= &vga_con
;
1545 #elif defined(CONFIG_DUMMY_CONSOLE)
1546 conswitchp
= &dummy_con
;
1551 #include "setup_arch_post.h"
1555 * c-file-style:"k&r"