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/mmzone.h>
57 #include <asm/setup.h>
58 #include <asm/arch_hooks.h>
59 #include <asm/sections.h>
60 #include <asm/io_apic.h>
63 #include <setup_arch.h>
64 #include <bios_ebda.h>
66 /* Forward Declaration. */
67 void __init
find_max_pfn(void);
69 /* This value is set up by the early boot code to point to the value
70 immediately after the boot time page tables. It contains a *physical*
71 address, and must not be in the .bss segment! */
72 unsigned long init_pg_tables_end __initdata
= ~0UL;
74 int disable_pse __devinitdata
= 0;
82 EXPORT_SYMBOL(efi_enabled
);
85 /* cpu data as detected by the assembly code in head.S */
86 struct cpuinfo_x86 new_cpu_data __initdata
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
87 /* common cpu data for all cpus */
88 struct cpuinfo_x86 boot_cpu_data __read_mostly
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
89 EXPORT_SYMBOL(boot_cpu_data
);
91 unsigned long mmu_cr4_features
;
93 /* for MCA, but anyone else can use it if they want */
94 unsigned int machine_id
;
96 EXPORT_SYMBOL(machine_id
);
98 unsigned int machine_submodel_id
;
99 unsigned int BIOS_revision
;
100 unsigned int mca_pentium_flag
;
102 /* For PCI or other memory-mapped resources */
103 unsigned long pci_mem_start
= 0x10000000;
105 EXPORT_SYMBOL(pci_mem_start
);
108 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
111 /* user-defined highmem size */
112 static unsigned int highmem_pages
= -1;
117 struct drive_info_struct
{ char dummy
[32]; } drive_info
;
118 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
119 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
120 EXPORT_SYMBOL(drive_info
);
122 struct screen_info screen_info
;
123 EXPORT_SYMBOL(screen_info
);
124 struct apm_info apm_info
;
125 EXPORT_SYMBOL(apm_info
);
126 struct sys_desc_table_struct
{
127 unsigned short length
;
128 unsigned char table
[0];
130 struct edid_info edid_info
;
131 EXPORT_SYMBOL_GPL(edid_info
);
132 struct ist_info ist_info
;
133 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
134 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
135 EXPORT_SYMBOL(ist_info
);
139 extern void early_cpu_init(void);
140 extern int root_mountflags
;
142 unsigned long saved_videomode
;
144 #define RAMDISK_IMAGE_START_MASK 0x07FF
145 #define RAMDISK_PROMPT_FLAG 0x8000
146 #define RAMDISK_LOAD_FLAG 0x4000
148 static char command_line
[COMMAND_LINE_SIZE
];
150 unsigned char __initdata boot_params
[PARAM_SIZE
];
152 static struct resource data_resource
= {
153 .name
= "Kernel data",
156 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
159 static struct resource code_resource
= {
160 .name
= "Kernel code",
163 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
166 static struct resource system_rom_resource
= {
167 .name
= "System ROM",
170 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
173 static struct resource extension_rom_resource
= {
174 .name
= "Extension ROM",
177 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
180 static struct resource adapter_rom_resources
[] = { {
181 .name
= "Adapter ROM",
184 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
186 .name
= "Adapter ROM",
189 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
191 .name
= "Adapter ROM",
194 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
196 .name
= "Adapter ROM",
199 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
201 .name
= "Adapter ROM",
204 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
206 .name
= "Adapter ROM",
209 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
212 static struct resource video_rom_resource
= {
216 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
219 static struct resource video_ram_resource
= {
220 .name
= "Video RAM area",
223 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
226 static struct resource standard_io_resources
[] = { {
230 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
235 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
240 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
245 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
250 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
252 .name
= "dma page reg",
255 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
260 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
265 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
270 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
273 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
275 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
277 unsigned char *p
, sum
= 0;
279 for (p
= rom
; p
< rom
+ length
; p
++)
284 static void __init
probe_roms(void)
286 unsigned long start
, length
, upper
;
291 upper
= adapter_rom_resources
[0].start
;
292 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
293 rom
= isa_bus_to_virt(start
);
294 if (!romsignature(rom
))
297 video_rom_resource
.start
= start
;
299 /* 0 < length <= 0x7f * 512, historically */
300 length
= rom
[2] * 512;
302 /* if checksum okay, trust length byte */
303 if (length
&& romchecksum(rom
, length
))
304 video_rom_resource
.end
= start
+ length
- 1;
306 request_resource(&iomem_resource
, &video_rom_resource
);
310 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
315 request_resource(&iomem_resource
, &system_rom_resource
);
316 upper
= system_rom_resource
.start
;
318 /* check for extension rom (ignore length byte!) */
319 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
320 if (romsignature(rom
)) {
321 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
322 if (romchecksum(rom
, length
)) {
323 request_resource(&iomem_resource
, &extension_rom_resource
);
324 upper
= extension_rom_resource
.start
;
328 /* check for adapter roms on 2k boundaries */
329 for (i
= 0; i
< ARRAY_SIZE(adapter_rom_resources
) && start
< upper
; start
+= 2048) {
330 rom
= isa_bus_to_virt(start
);
331 if (!romsignature(rom
))
334 /* 0 < length <= 0x7f * 512, historically */
335 length
= rom
[2] * 512;
337 /* but accept any length that fits if checksum okay */
338 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
341 adapter_rom_resources
[i
].start
= start
;
342 adapter_rom_resources
[i
].end
= start
+ length
- 1;
343 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
345 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
349 static void __init
limit_regions(unsigned long long size
)
351 unsigned long long current_addr
= 0;
355 efi_memory_desc_t
*md
;
358 for (p
= memmap
.map
, i
= 0; p
< memmap
.map_end
;
359 p
+= memmap
.desc_size
, i
++) {
361 current_addr
= md
->phys_addr
+ (md
->num_pages
<< 12);
362 if (md
->type
== EFI_CONVENTIONAL_MEMORY
) {
363 if (current_addr
>= size
) {
365 (((current_addr
-size
) + PAGE_SIZE
-1) >> PAGE_SHIFT
);
366 memmap
.nr_map
= i
+ 1;
372 for (i
= 0; i
< e820
.nr_map
; i
++) {
373 current_addr
= e820
.map
[i
].addr
+ e820
.map
[i
].size
;
374 if (current_addr
< size
)
377 if (e820
.map
[i
].type
!= E820_RAM
)
380 if (e820
.map
[i
].addr
>= size
) {
382 * This region starts past the end of the
383 * requested size, skip it completely.
388 e820
.map
[i
].size
-= current_addr
- size
;
394 void __init
add_memory_region(unsigned long long start
,
395 unsigned long long size
, int type
)
403 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
407 e820
.map
[x
].addr
= start
;
408 e820
.map
[x
].size
= size
;
409 e820
.map
[x
].type
= type
;
412 } /* add_memory_region */
416 static void __init
print_memory_map(char *who
)
420 for (i
= 0; i
< e820
.nr_map
; i
++) {
421 printk(" %s: %016Lx - %016Lx ", who
,
423 e820
.map
[i
].addr
+ e820
.map
[i
].size
);
424 switch (e820
.map
[i
].type
) {
425 case E820_RAM
: printk("(usable)\n");
428 printk("(reserved)\n");
431 printk("(ACPI data)\n");
434 printk("(ACPI NVS)\n");
436 default: printk("type %lu\n", e820
.map
[i
].type
);
443 * Sanitize the BIOS e820 map.
445 * Some e820 responses include overlapping entries. The following
446 * replaces the original e820 map with a new one, removing overlaps.
449 struct change_member
{
450 struct e820entry
*pbios
; /* pointer to original bios entry */
451 unsigned long long addr
; /* address for this change point */
453 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
454 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
455 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
456 static struct e820entry new_bios
[E820MAX
] __initdata
;
458 int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
460 struct change_member
*change_tmp
;
461 unsigned long current_type
, last_type
;
462 unsigned long long last_addr
;
463 int chgidx
, still_changing
;
466 int old_nr
, new_nr
, chg_nr
;
470 Visually we're performing the following (1,2,3,4 = memory types)...
472 Sample memory map (w/overlaps):
473 ____22__________________
474 ______________________4_
475 ____1111________________
476 _44_____________________
477 11111111________________
478 ____________________33__
479 ___________44___________
480 __________33333_________
481 ______________22________
482 ___________________2222_
483 _________111111111______
484 _____________________11_
485 _________________4______
487 Sanitized equivalent (no overlap):
488 1_______________________
489 _44_____________________
490 ___1____________________
491 ____22__________________
492 ______11________________
493 _________1______________
494 __________3_____________
495 ___________44___________
496 _____________33_________
497 _______________2________
498 ________________1_______
499 _________________4______
500 ___________________2____
501 ____________________33__
502 ______________________4_
505 /* if there's only one memory region, don't bother */
511 /* bail out if we find any unreasonable addresses in bios map */
512 for (i
=0; i
<old_nr
; i
++)
513 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
516 /* create pointers for initial change-point information (for sorting) */
517 for (i
=0; i
< 2*old_nr
; i
++)
518 change_point
[i
] = &change_point_list
[i
];
520 /* record all known change-points (starting and ending addresses),
521 omitting those that are for empty memory regions */
523 for (i
=0; i
< old_nr
; i
++) {
524 if (biosmap
[i
].size
!= 0) {
525 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
526 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
527 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
528 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
531 chg_nr
= chgidx
; /* true number of change-points */
533 /* sort change-point list by memory addresses (low -> high) */
535 while (still_changing
) {
537 for (i
=1; i
< chg_nr
; i
++) {
538 /* if <current_addr> > <last_addr>, swap */
539 /* or, if current=<start_addr> & last=<end_addr>, swap */
540 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
541 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
542 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
543 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
546 change_tmp
= change_point
[i
];
547 change_point
[i
] = change_point
[i
-1];
548 change_point
[i
-1] = change_tmp
;
554 /* create a new bios memory map, removing overlaps */
555 overlap_entries
=0; /* number of entries in the overlap table */
556 new_bios_entry
=0; /* index for creating new bios map entries */
557 last_type
= 0; /* start with undefined memory type */
558 last_addr
= 0; /* start with 0 as last starting address */
559 /* loop through change-points, determining affect on the new bios map */
560 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
562 /* keep track of all overlapping bios entries */
563 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
565 /* add map entry to overlap list (> 1 entry implies an overlap) */
566 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
570 /* remove entry from list (order independent, so swap with last) */
571 for (i
=0; i
<overlap_entries
; i
++)
573 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
574 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
578 /* if there are overlapping entries, decide which "type" to use */
579 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
581 for (i
=0; i
<overlap_entries
; i
++)
582 if (overlap_list
[i
]->type
> current_type
)
583 current_type
= overlap_list
[i
]->type
;
584 /* continue building up new bios map based on this information */
585 if (current_type
!= last_type
) {
586 if (last_type
!= 0) {
587 new_bios
[new_bios_entry
].size
=
588 change_point
[chgidx
]->addr
- last_addr
;
589 /* move forward only if the new size was non-zero */
590 if (new_bios
[new_bios_entry
].size
!= 0)
591 if (++new_bios_entry
>= E820MAX
)
592 break; /* no more space left for new bios entries */
594 if (current_type
!= 0) {
595 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
596 new_bios
[new_bios_entry
].type
= current_type
;
597 last_addr
=change_point
[chgidx
]->addr
;
599 last_type
= current_type
;
602 new_nr
= new_bios_entry
; /* retain count for new bios entries */
604 /* copy new bios mapping into original location */
605 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
612 * Copy the BIOS e820 map into a safe place.
614 * Sanity-check it while we're at it..
616 * If we're lucky and live on a modern system, the setup code
617 * will have given us a memory map that we can use to properly
618 * set up memory. If we aren't, we'll fake a memory map.
620 * We check to see that the memory map contains at least 2 elements
621 * before we'll use it, because the detection code in setup.S may
622 * not be perfect and most every PC known to man has two memory
623 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
624 * thinkpad 560x, for example, does not cooperate with the memory
627 int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
629 /* Only one memory region (or negative)? Ignore it */
634 unsigned long long start
= biosmap
->addr
;
635 unsigned long long size
= biosmap
->size
;
636 unsigned long long end
= start
+ size
;
637 unsigned long type
= biosmap
->type
;
639 /* Overflow in 64 bits? Ignore the memory map. */
644 * Some BIOSes claim RAM in the 640k - 1M region.
645 * Not right. Fix it up.
647 if (type
== E820_RAM
) {
648 if (start
< 0x100000ULL
&& end
> 0xA0000ULL
) {
649 if (start
< 0xA0000ULL
)
650 add_memory_region(start
, 0xA0000ULL
-start
, type
);
651 if (end
<= 0x100000ULL
)
657 add_memory_region(start
, size
, type
);
658 } while (biosmap
++,--nr_map
);
662 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
664 #ifdef CONFIG_EDD_MODULE
668 * copy_edd() - Copy the BIOS EDD information
669 * from boot_params into a safe place.
672 static inline void copy_edd(void)
674 memcpy(edd
.mbr_signature
, EDD_MBR_SIGNATURE
, sizeof(edd
.mbr_signature
));
675 memcpy(edd
.edd_info
, EDD_BUF
, sizeof(edd
.edd_info
));
676 edd
.mbr_signature_nr
= EDD_MBR_SIG_NR
;
677 edd
.edd_info_nr
= EDD_NR
;
680 static inline void copy_edd(void)
685 static int __initdata user_defined_memmap
= 0;
688 * "mem=nopentium" disables the 4MB page tables.
689 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
690 * to <mem>, overriding the bios size.
691 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
692 * <start> to <start>+<mem>, overriding the bios size.
694 * HPA tells me bootloaders need to parse mem=, so no new
695 * option should be mem= [also see Documentation/i386/boot.txt]
697 static int __init
parse_mem(char *arg
)
702 if (strcmp(arg
, "nopentium") == 0) {
703 clear_bit(X86_FEATURE_PSE
, boot_cpu_data
.x86_capability
);
706 /* If the user specifies memory size, we
707 * limit the BIOS-provided memory map to
708 * that size. exactmap can be used to specify
709 * the exact map. mem=number can be used to
710 * trim the existing memory map.
712 unsigned long long mem_size
;
714 mem_size
= memparse(arg
, &arg
);
715 limit_regions(mem_size
);
716 user_defined_memmap
= 1;
720 early_param("mem", parse_mem
);
722 static int __init
parse_memmap(char *arg
)
727 if (strcmp(arg
, "exactmap") == 0) {
728 #ifdef CONFIG_CRASH_DUMP
729 /* If we are doing a crash dump, we
730 * still need to know the real mem
731 * size before original memory map is
735 saved_max_pfn
= max_pfn
;
738 user_defined_memmap
= 1;
740 /* If the user specifies memory size, we
741 * limit the BIOS-provided memory map to
742 * that size. exactmap can be used to specify
743 * the exact map. mem=number can be used to
744 * trim the existing memory map.
746 unsigned long long start_at
, mem_size
;
748 mem_size
= memparse(arg
, &arg
);
750 start_at
= memparse(arg
+1, &arg
);
751 add_memory_region(start_at
, mem_size
, E820_RAM
);
752 } else if (*arg
== '#') {
753 start_at
= memparse(arg
+1, &arg
);
754 add_memory_region(start_at
, mem_size
, E820_ACPI
);
755 } else if (*arg
== '$') {
756 start_at
= memparse(arg
+1, &arg
);
757 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
759 limit_regions(mem_size
);
760 user_defined_memmap
= 1;
765 early_param("memmap", parse_memmap
);
767 #ifdef CONFIG_PROC_VMCORE
768 /* elfcorehdr= specifies the location of elf core header
769 * stored by the crashed kernel.
771 static int __init
parse_elfcorehdr(char *arg
)
776 elfcorehdr_addr
= memparse(arg
, &arg
);
779 early_param("elfcorehdr", parse_elfcorehdr
);
780 #endif /* CONFIG_PROC_VMCORE */
783 * highmem=size forces highmem to be exactly 'size' bytes.
784 * This works even on boxes that have no highmem otherwise.
785 * This also works to reduce highmem size on bigger boxes.
787 static int __init
parse_highmem(char *arg
)
792 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
795 early_param("highmem", parse_highmem
);
798 * vmalloc=size forces the vmalloc area to be exactly 'size'
799 * bytes. This can be used to increase (or decrease) the
800 * vmalloc area - the default is 128m.
802 static int __init
parse_vmalloc(char *arg
)
807 __VMALLOC_RESERVE
= memparse(arg
, &arg
);
810 early_param("vmalloc", parse_vmalloc
);
813 * reservetop=size reserves a hole at the top of the kernel address space which
814 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
815 * so relocating the fixmap can be done before paging initialization.
817 static int __init
parse_reservetop(char *arg
)
819 unsigned long address
;
824 address
= memparse(arg
, &arg
);
825 reserve_top_address(address
);
828 early_param("reservetop", parse_reservetop
);
831 * Callback for efi_memory_walk.
834 efi_find_max_pfn(unsigned long start
, unsigned long end
, void *arg
)
836 unsigned long *max_pfn
= arg
, pfn
;
839 pfn
= PFN_UP(end
-1);
847 efi_memory_present_wrapper(unsigned long start
, unsigned long end
, void *arg
)
849 memory_present(0, start
, end
);
854 * This function checks if the entire range <start,end> is mapped with type.
856 * Note: this function only works correct if the e820 table is sorted and
857 * not-overlapping, which is the case
860 e820_all_mapped(unsigned long s
, unsigned long e
, unsigned type
)
865 for (i
= 0; i
< e820
.nr_map
; i
++) {
866 struct e820entry
*ei
= &e820
.map
[i
];
867 if (type
&& ei
->type
!= type
)
869 /* is the region (part) in overlap with the current region ?*/
870 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
872 /* if the region is at the beginning of <start,end> we move
873 * start to the end of the region since it's ok until there
875 if (ei
->addr
<= start
)
876 start
= ei
->addr
+ ei
->size
;
877 /* if start is now at or beyond end, we're done, full
880 return 1; /* we're done */
886 * Find the highest page frame number we have available
888 void __init
find_max_pfn(void)
894 efi_memmap_walk(efi_find_max_pfn
, &max_pfn
);
895 efi_memmap_walk(efi_memory_present_wrapper
, NULL
);
899 for (i
= 0; i
< e820
.nr_map
; i
++) {
900 unsigned long start
, end
;
902 if (e820
.map
[i
].type
!= E820_RAM
)
904 start
= PFN_UP(e820
.map
[i
].addr
);
905 end
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
910 memory_present(0, start
, end
);
915 * Determine low and high memory ranges:
917 unsigned long __init
find_max_low_pfn(void)
919 unsigned long max_low_pfn
;
921 max_low_pfn
= max_pfn
;
922 if (max_low_pfn
> MAXMEM_PFN
) {
923 if (highmem_pages
== -1)
924 highmem_pages
= max_pfn
- MAXMEM_PFN
;
925 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
926 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
927 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
928 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn
- MAXMEM_PFN
), pages_to_mb(highmem_pages
));
931 max_low_pfn
= MAXMEM_PFN
;
932 #ifndef CONFIG_HIGHMEM
933 /* Maximum memory usable is what is directly addressable */
934 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
936 if (max_pfn
> MAX_NONPAE_PFN
)
937 printk(KERN_WARNING
"Use a PAE enabled kernel.\n");
939 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
940 max_pfn
= MAXMEM_PFN
;
941 #else /* !CONFIG_HIGHMEM */
942 #ifndef CONFIG_X86_PAE
943 if (max_pfn
> MAX_NONPAE_PFN
) {
944 max_pfn
= MAX_NONPAE_PFN
;
945 printk(KERN_WARNING
"Warning only 4GB will be used.\n");
946 printk(KERN_WARNING
"Use a PAE enabled kernel.\n");
948 #endif /* !CONFIG_X86_PAE */
949 #endif /* !CONFIG_HIGHMEM */
951 if (highmem_pages
== -1)
953 #ifdef CONFIG_HIGHMEM
954 if (highmem_pages
>= max_pfn
) {
955 printk(KERN_ERR
"highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
959 if (max_low_pfn
-highmem_pages
< 64*1024*1024/PAGE_SIZE
){
960 printk(KERN_ERR
"highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages
));
963 max_low_pfn
-= highmem_pages
;
967 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
974 * Free all available memory for boot time allocation. Used
975 * as a callback function by efi_memory_walk()
979 free_available_memory(unsigned long start
, unsigned long end
, void *arg
)
981 /* check max_low_pfn */
982 if (start
>= (max_low_pfn
<< PAGE_SHIFT
))
984 if (end
>= (max_low_pfn
<< PAGE_SHIFT
))
985 end
= max_low_pfn
<< PAGE_SHIFT
;
987 free_bootmem(start
, end
- start
);
992 * Register fully available low RAM pages with the bootmem allocator.
994 static void __init
register_bootmem_low_pages(unsigned long max_low_pfn
)
999 efi_memmap_walk(free_available_memory
, NULL
);
1002 for (i
= 0; i
< e820
.nr_map
; i
++) {
1003 unsigned long curr_pfn
, last_pfn
, size
;
1005 * Reserve usable low memory
1007 if (e820
.map
[i
].type
!= E820_RAM
)
1010 * We are rounding up the start address of usable memory:
1012 curr_pfn
= PFN_UP(e820
.map
[i
].addr
);
1013 if (curr_pfn
>= max_low_pfn
)
1016 * ... and at the end of the usable range downwards:
1018 last_pfn
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
1020 if (last_pfn
> max_low_pfn
)
1021 last_pfn
= max_low_pfn
;
1024 * .. finally, did all the rounding and playing
1025 * around just make the area go away?
1027 if (last_pfn
<= curr_pfn
)
1030 size
= last_pfn
- curr_pfn
;
1031 free_bootmem(PFN_PHYS(curr_pfn
), PFN_PHYS(size
));
1036 * workaround for Dell systems that neglect to reserve EBDA
1038 static void __init
reserve_ebda_region(void)
1041 addr
= get_bios_ebda();
1043 reserve_bootmem(addr
, PAGE_SIZE
);
1046 #ifndef CONFIG_NEED_MULTIPLE_NODES
1047 void __init
setup_bootmem_allocator(void);
1048 static unsigned long __init
setup_memory(void)
1051 * partially used pages are not usable - thus
1052 * we are rounding upwards:
1054 min_low_pfn
= PFN_UP(init_pg_tables_end
);
1058 max_low_pfn
= find_max_low_pfn();
1060 #ifdef CONFIG_HIGHMEM
1061 highstart_pfn
= highend_pfn
= max_pfn
;
1062 if (max_pfn
> max_low_pfn
) {
1063 highstart_pfn
= max_low_pfn
;
1065 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
1066 pages_to_mb(highend_pfn
- highstart_pfn
));
1067 num_physpages
= highend_pfn
;
1068 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
1070 num_physpages
= max_low_pfn
;
1071 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
1073 #ifdef CONFIG_FLATMEM
1074 max_mapnr
= num_physpages
;
1076 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
1077 pages_to_mb(max_low_pfn
));
1079 setup_bootmem_allocator();
1084 void __init
zone_sizes_init(void)
1086 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
1087 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
1088 max_zone_pfns
[ZONE_DMA
] =
1089 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
1090 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
1091 #ifdef CONFIG_HIGHMEM
1092 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
1093 add_active_range(0, 0, highend_pfn
);
1095 add_active_range(0, 0, max_low_pfn
);
1098 free_area_init_nodes(max_zone_pfns
);
1101 extern unsigned long __init
setup_memory(void);
1102 extern void zone_sizes_init(void);
1103 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1105 void __init
setup_bootmem_allocator(void)
1107 unsigned long bootmap_size
;
1109 * Initialize the boot-time allocator (with low memory only):
1111 bootmap_size
= init_bootmem(min_low_pfn
, max_low_pfn
);
1113 register_bootmem_low_pages(max_low_pfn
);
1116 * Reserve the bootmem bitmap itself as well. We do this in two
1117 * steps (first step was init_bootmem()) because this catches
1118 * the (very unlikely) case of us accidentally initializing the
1119 * bootmem allocator with an invalid RAM area.
1121 reserve_bootmem(__PHYSICAL_START
, (PFN_PHYS(min_low_pfn
) +
1122 bootmap_size
+ PAGE_SIZE
-1) - (__PHYSICAL_START
));
1125 * reserve physical page 0 - it's a special BIOS page on many boxes,
1126 * enabling clean reboots, SMP operation, laptop functions.
1128 reserve_bootmem(0, PAGE_SIZE
);
1130 /* reserve EBDA region, it's a 4K region */
1131 reserve_ebda_region();
1133 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1134 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1135 unless you have no PS/2 mouse plugged in. */
1136 if (boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
1137 boot_cpu_data
.x86
== 6)
1138 reserve_bootmem(0xa0000 - 4096, 4096);
1142 * But first pinch a few for the stack/trampoline stuff
1143 * FIXME: Don't need the extra page at 4K, but need to fix
1144 * trampoline before removing it. (see the GDT stuff)
1146 reserve_bootmem(PAGE_SIZE
, PAGE_SIZE
);
1148 #ifdef CONFIG_ACPI_SLEEP
1150 * Reserve low memory region for sleep support.
1152 acpi_reserve_bootmem();
1154 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1156 * Find and reserve possible boot-time SMP configuration:
1161 #ifdef CONFIG_BLK_DEV_INITRD
1162 if (LOADER_TYPE
&& INITRD_START
) {
1163 if (INITRD_START
+ INITRD_SIZE
<= (max_low_pfn
<< PAGE_SHIFT
)) {
1164 reserve_bootmem(INITRD_START
, INITRD_SIZE
);
1166 INITRD_START
? INITRD_START
+ PAGE_OFFSET
: 0;
1167 initrd_end
= initrd_start
+INITRD_SIZE
;
1170 printk(KERN_ERR
"initrd extends beyond end of memory "
1171 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1172 INITRD_START
+ INITRD_SIZE
,
1173 max_low_pfn
<< PAGE_SHIFT
);
1179 if (crashk_res
.start
!= crashk_res
.end
)
1180 reserve_bootmem(crashk_res
.start
,
1181 crashk_res
.end
- crashk_res
.start
+ 1);
1186 * The node 0 pgdat is initialized before all of these because
1187 * it's needed for bootmem. node>0 pgdats have their virtual
1188 * space allocated before the pagetables are in place to access
1189 * them, so they can't be cleared then.
1191 * This should all compile down to nothing when NUMA is off.
1193 void __init
remapped_pgdat_init(void)
1197 for_each_online_node(nid
) {
1199 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
1204 * Request address space for all standard RAM and ROM resources
1205 * and also for regions reported as reserved by the e820.
1208 legacy_init_iomem_resources(struct resource
*code_resource
, struct resource
*data_resource
)
1213 for (i
= 0; i
< e820
.nr_map
; i
++) {
1214 struct resource
*res
;
1215 #ifndef CONFIG_RESOURCES_64BIT
1216 if (e820
.map
[i
].addr
+ e820
.map
[i
].size
> 0x100000000ULL
)
1219 res
= kzalloc(sizeof(struct resource
), GFP_ATOMIC
);
1220 switch (e820
.map
[i
].type
) {
1221 case E820_RAM
: res
->name
= "System RAM"; break;
1222 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
1223 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
1224 default: res
->name
= "reserved";
1226 res
->start
= e820
.map
[i
].addr
;
1227 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
1228 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
1229 if (request_resource(&iomem_resource
, res
)) {
1233 if (e820
.map
[i
].type
== E820_RAM
) {
1235 * We don't know which RAM region contains kernel data,
1236 * so we try it repeatedly and let the resource manager
1239 request_resource(res
, code_resource
);
1240 request_resource(res
, data_resource
);
1242 request_resource(res
, &crashk_res
);
1249 * Request address space for all standard resources
1251 * This is called just before pcibios_init(), which is also a
1252 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1254 static int __init
request_standard_resources(void)
1258 printk("Setting up standard PCI resources\n");
1260 efi_initialize_iomem_resources(&code_resource
, &data_resource
);
1262 legacy_init_iomem_resources(&code_resource
, &data_resource
);
1264 /* EFI systems may still have VGA */
1265 request_resource(&iomem_resource
, &video_ram_resource
);
1267 /* request I/O space for devices used on all i[345]86 PCs */
1268 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
1269 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
1273 subsys_initcall(request_standard_resources
);
1275 static void __init
register_memory(void)
1277 unsigned long gapstart
, gapsize
, round
;
1278 unsigned long long last
;
1282 * Search for the bigest gap in the low 32 bits of the e820
1285 last
= 0x100000000ull
;
1286 gapstart
= 0x10000000;
1290 unsigned long long start
= e820
.map
[i
].addr
;
1291 unsigned long long end
= start
+ e820
.map
[i
].size
;
1294 * Since "last" is at most 4GB, we know we'll
1295 * fit in 32 bits if this condition is true
1298 unsigned long gap
= last
- end
;
1300 if (gap
> gapsize
) {
1310 * See how much we want to round up: start off with
1311 * rounding to the next 1MB area.
1314 while ((gapsize
>> 4) > round
)
1316 /* Fun with two's complement */
1317 pci_mem_start
= (gapstart
+ round
) & -round
;
1319 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1320 pci_mem_start
, gapstart
, gapsize
);
1324 static void set_mca_bus(int x
)
1329 static void set_mca_bus(int x
) { }
1333 * Determine if we were loaded by an EFI loader. If so, then we have also been
1334 * passed the efi memmap, systab, etc., so we should use these data structures
1335 * for initialization. Note, the efi init code path is determined by the
1336 * global efi_enabled. This allows the same kernel image to be used on existing
1337 * systems (with a traditional BIOS) as well as on EFI systems.
1339 void __init
setup_arch(char **cmdline_p
)
1341 unsigned long max_low_pfn
;
1343 memcpy(&boot_cpu_data
, &new_cpu_data
, sizeof(new_cpu_data
));
1344 pre_setup_arch_hook();
1348 * FIXME: This isn't an official loader_type right
1349 * now but does currently work with elilo.
1350 * If we were configured as an EFI kernel, check to make
1351 * sure that we were loaded correctly from elilo and that
1352 * the system table is valid. If not, then initialize normally.
1355 if ((LOADER_TYPE
== 0x50) && EFI_SYSTAB
)
1359 ROOT_DEV
= old_decode_dev(ORIG_ROOT_DEV
);
1360 drive_info
= DRIVE_INFO
;
1361 screen_info
= SCREEN_INFO
;
1362 edid_info
= EDID_INFO
;
1363 apm_info
.bios
= APM_BIOS_INFO
;
1364 ist_info
= IST_INFO
;
1365 saved_videomode
= VIDEO_MODE
;
1366 if( SYS_DESC_TABLE
.length
!= 0 ) {
1367 set_mca_bus(SYS_DESC_TABLE
.table
[3] & 0x2);
1368 machine_id
= SYS_DESC_TABLE
.table
[0];
1369 machine_submodel_id
= SYS_DESC_TABLE
.table
[1];
1370 BIOS_revision
= SYS_DESC_TABLE
.table
[2];
1372 bootloader_type
= LOADER_TYPE
;
1374 #ifdef CONFIG_BLK_DEV_RAM
1375 rd_image_start
= RAMDISK_FLAGS
& RAMDISK_IMAGE_START_MASK
;
1376 rd_prompt
= ((RAMDISK_FLAGS
& RAMDISK_PROMPT_FLAG
) != 0);
1377 rd_doload
= ((RAMDISK_FLAGS
& RAMDISK_LOAD_FLAG
) != 0);
1383 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
1384 print_memory_map(machine_specific_memory_setup());
1389 if (!MOUNT_ROOT_RDONLY
)
1390 root_mountflags
&= ~MS_RDONLY
;
1391 init_mm
.start_code
= (unsigned long) _text
;
1392 init_mm
.end_code
= (unsigned long) _etext
;
1393 init_mm
.end_data
= (unsigned long) _edata
;
1394 init_mm
.brk
= init_pg_tables_end
+ PAGE_OFFSET
;
1396 code_resource
.start
= virt_to_phys(_text
);
1397 code_resource
.end
= virt_to_phys(_etext
)-1;
1398 data_resource
.start
= virt_to_phys(_etext
);
1399 data_resource
.end
= virt_to_phys(_edata
)-1;
1401 parse_early_param();
1403 if (user_defined_memmap
) {
1404 printk(KERN_INFO
"user-defined physical RAM map:\n");
1405 print_memory_map("user");
1408 strlcpy(command_line
, saved_command_line
, COMMAND_LINE_SIZE
);
1409 *cmdline_p
= command_line
;
1411 max_low_pfn
= setup_memory();
1414 * NOTE: before this point _nobody_ is allowed to allocate
1415 * any memory using the bootmem allocator. Although the
1416 * alloctor is now initialised only the first 8Mb of the kernel
1417 * virtual address space has been mapped. All allocations before
1418 * paging_init() has completed must use the alloc_bootmem_low_pages()
1419 * variant (which allocates DMA'able memory) and care must be taken
1420 * not to exceed the 8Mb limit.
1424 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1427 remapped_pgdat_init();
1432 * NOTE: at this point the bootmem allocator is fully available.
1437 #ifdef CONFIG_X86_GENERICARCH
1438 generic_apic_probe();
1445 * Parse the ACPI tables for possible boot-time SMP configuration.
1447 acpi_boot_table_init();
1451 #ifdef CONFIG_X86_IO_APIC
1452 check_acpi_pci(); /* Checks more than just ACPI actually */
1459 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1461 printk(KERN_WARNING
"More than 8 CPUs detected and "
1462 "CONFIG_X86_PC cannot handle it.\nUse "
1463 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1466 #ifdef CONFIG_X86_LOCAL_APIC
1467 if (smp_found_config
)
1474 #if defined(CONFIG_VGA_CONSOLE)
1475 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
1476 conswitchp
= &vga_con
;
1477 #elif defined(CONFIG_DUMMY_CONSOLE)
1478 conswitchp
= &dummy_con
;
1484 static __init
int add_pcspkr(void)
1486 struct platform_device
*pd
;
1489 pd
= platform_device_alloc("pcspkr", -1);
1493 ret
= platform_device_add(pd
);
1495 platform_device_put(pd
);
1499 device_initcall(add_pcspkr
);
1504 * c-file-style:"k&r"