2 * Copyright (C) 1995 Linus Torvalds
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
21 * This file handles the architecture-dependent parts of initialization
24 #include <linux/sched.h>
26 #include <linux/mmzone.h>
27 #include <linux/screen_info.h>
28 #include <linux/ioport.h>
29 #include <linux/acpi.h>
30 #include <linux/apm_bios.h>
31 #include <linux/initrd.h>
32 #include <linux/bootmem.h>
33 #include <linux/seq_file.h>
34 #include <linux/console.h>
35 #include <linux/mca.h>
36 #include <linux/root_dev.h>
37 #include <linux/highmem.h>
38 #include <linux/module.h>
39 #include <linux/efi.h>
40 #include <linux/init.h>
41 #include <linux/edd.h>
42 #include <linux/nodemask.h>
43 #include <linux/kexec.h>
44 #include <linux/crash_dump.h>
45 #include <linux/dmi.h>
46 #include <linux/pfn.h>
48 #include <video/edid.h>
52 #include <asm/mpspec.h>
53 #include <asm/mmzone.h>
54 #include <asm/setup.h>
55 #include <asm/arch_hooks.h>
56 #include <asm/sections.h>
57 #include <asm/io_apic.h>
61 #include <setup_arch.h>
62 #include <bios_ebda.h>
64 /* This value is set up by the early boot code to point to the value
65 immediately after the boot time page tables. It contains a *physical*
66 address, and must not be in the .bss segment! */
67 unsigned long init_pg_tables_end __initdata
= ~0UL;
69 int disable_pse __devinitdata
= 0;
74 extern struct resource code_resource
;
75 extern struct resource data_resource
;
77 /* cpu data as detected by the assembly code in head.S */
78 struct cpuinfo_x86 new_cpu_data __cpuinitdata
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
79 /* common cpu data for all cpus */
80 struct cpuinfo_x86 boot_cpu_data __read_mostly
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
81 EXPORT_SYMBOL(boot_cpu_data
);
83 unsigned long mmu_cr4_features
;
85 /* for MCA, but anyone else can use it if they want */
86 unsigned int machine_id
;
88 EXPORT_SYMBOL(machine_id
);
90 unsigned int machine_submodel_id
;
91 unsigned int BIOS_revision
;
92 unsigned int mca_pentium_flag
;
94 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
97 /* user-defined highmem size */
98 static unsigned int highmem_pages
= -1;
103 struct screen_info screen_info
;
104 EXPORT_SYMBOL(screen_info
);
105 struct apm_info apm_info
;
106 EXPORT_SYMBOL(apm_info
);
107 struct edid_info edid_info
;
108 EXPORT_SYMBOL_GPL(edid_info
);
109 struct ist_info ist_info
;
110 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
111 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
112 EXPORT_SYMBOL(ist_info
);
115 extern void early_cpu_init(void);
116 extern int root_mountflags
;
118 unsigned long saved_videomode
;
120 #define RAMDISK_IMAGE_START_MASK 0x07FF
121 #define RAMDISK_PROMPT_FLAG 0x8000
122 #define RAMDISK_LOAD_FLAG 0x4000
124 static char __initdata command_line
[COMMAND_LINE_SIZE
];
126 struct boot_params __initdata boot_params
;
128 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
130 #ifdef CONFIG_EDD_MODULE
134 * copy_edd() - Copy the BIOS EDD information
135 * from boot_params into a safe place.
138 static inline void copy_edd(void)
140 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
141 sizeof(edd
.mbr_signature
));
142 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
143 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
144 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
147 static inline void copy_edd(void)
152 int __initdata user_defined_memmap
= 0;
155 * "mem=nopentium" disables the 4MB page tables.
156 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
157 * to <mem>, overriding the bios size.
158 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
159 * <start> to <start>+<mem>, overriding the bios size.
161 * HPA tells me bootloaders need to parse mem=, so no new
162 * option should be mem= [also see Documentation/i386/boot.txt]
164 static int __init
parse_mem(char *arg
)
169 if (strcmp(arg
, "nopentium") == 0) {
170 clear_bit(X86_FEATURE_PSE
, boot_cpu_data
.x86_capability
);
173 /* If the user specifies memory size, we
174 * limit the BIOS-provided memory map to
175 * that size. exactmap can be used to specify
176 * the exact map. mem=number can be used to
177 * trim the existing memory map.
179 unsigned long long mem_size
;
181 mem_size
= memparse(arg
, &arg
);
182 limit_regions(mem_size
);
183 user_defined_memmap
= 1;
187 early_param("mem", parse_mem
);
189 #ifdef CONFIG_PROC_VMCORE
190 /* elfcorehdr= specifies the location of elf core header
191 * stored by the crashed kernel.
193 static int __init
parse_elfcorehdr(char *arg
)
198 elfcorehdr_addr
= memparse(arg
, &arg
);
201 early_param("elfcorehdr", parse_elfcorehdr
);
202 #endif /* CONFIG_PROC_VMCORE */
205 * highmem=size forces highmem to be exactly 'size' bytes.
206 * This works even on boxes that have no highmem otherwise.
207 * This also works to reduce highmem size on bigger boxes.
209 static int __init
parse_highmem(char *arg
)
214 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
217 early_param("highmem", parse_highmem
);
220 * vmalloc=size forces the vmalloc area to be exactly 'size'
221 * bytes. This can be used to increase (or decrease) the
222 * vmalloc area - the default is 128m.
224 static int __init
parse_vmalloc(char *arg
)
229 __VMALLOC_RESERVE
= memparse(arg
, &arg
);
232 early_param("vmalloc", parse_vmalloc
);
235 * reservetop=size reserves a hole at the top of the kernel address space which
236 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
237 * so relocating the fixmap can be done before paging initialization.
239 static int __init
parse_reservetop(char *arg
)
241 unsigned long address
;
246 address
= memparse(arg
, &arg
);
247 reserve_top_address(address
);
250 early_param("reservetop", parse_reservetop
);
253 * Determine low and high memory ranges:
255 unsigned long __init
find_max_low_pfn(void)
257 unsigned long max_low_pfn
;
259 max_low_pfn
= max_pfn
;
260 if (max_low_pfn
> MAXMEM_PFN
) {
261 if (highmem_pages
== -1)
262 highmem_pages
= max_pfn
- MAXMEM_PFN
;
263 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
264 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
265 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
266 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn
- MAXMEM_PFN
), pages_to_mb(highmem_pages
));
269 max_low_pfn
= MAXMEM_PFN
;
270 #ifndef CONFIG_HIGHMEM
271 /* Maximum memory usable is what is directly addressable */
272 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
274 if (max_pfn
> MAX_NONPAE_PFN
)
275 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
277 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
278 max_pfn
= MAXMEM_PFN
;
279 #else /* !CONFIG_HIGHMEM */
280 #ifndef CONFIG_HIGHMEM64G
281 if (max_pfn
> MAX_NONPAE_PFN
) {
282 max_pfn
= MAX_NONPAE_PFN
;
283 printk(KERN_WARNING
"Warning only 4GB will be used.\n");
284 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
286 #endif /* !CONFIG_HIGHMEM64G */
287 #endif /* !CONFIG_HIGHMEM */
289 if (highmem_pages
== -1)
291 #ifdef CONFIG_HIGHMEM
292 if (highmem_pages
>= max_pfn
) {
293 printk(KERN_ERR
"highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
297 if (max_low_pfn
-highmem_pages
< 64*1024*1024/PAGE_SIZE
){
298 printk(KERN_ERR
"highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages
));
301 max_low_pfn
-= highmem_pages
;
305 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
312 * workaround for Dell systems that neglect to reserve EBDA
314 static void __init
reserve_ebda_region(void)
317 addr
= get_bios_ebda();
319 reserve_bootmem(addr
, PAGE_SIZE
);
322 #ifndef CONFIG_NEED_MULTIPLE_NODES
323 void __init
setup_bootmem_allocator(void);
324 static unsigned long __init
setup_memory(void)
327 * partially used pages are not usable - thus
328 * we are rounding upwards:
330 min_low_pfn
= PFN_UP(init_pg_tables_end
);
334 max_low_pfn
= find_max_low_pfn();
336 #ifdef CONFIG_HIGHMEM
337 highstart_pfn
= highend_pfn
= max_pfn
;
338 if (max_pfn
> max_low_pfn
) {
339 highstart_pfn
= max_low_pfn
;
341 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
342 pages_to_mb(highend_pfn
- highstart_pfn
));
343 num_physpages
= highend_pfn
;
344 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
346 num_physpages
= max_low_pfn
;
347 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
349 #ifdef CONFIG_FLATMEM
350 max_mapnr
= num_physpages
;
352 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
353 pages_to_mb(max_low_pfn
));
355 setup_bootmem_allocator();
360 void __init
zone_sizes_init(void)
362 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
363 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
364 max_zone_pfns
[ZONE_DMA
] =
365 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
366 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
367 #ifdef CONFIG_HIGHMEM
368 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
369 add_active_range(0, 0, highend_pfn
);
371 add_active_range(0, 0, max_low_pfn
);
374 free_area_init_nodes(max_zone_pfns
);
377 extern unsigned long __init
setup_memory(void);
378 extern void zone_sizes_init(void);
379 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
381 static inline unsigned long long get_total_mem(void)
383 unsigned long long total
;
385 total
= max_low_pfn
- min_low_pfn
;
386 #ifdef CONFIG_HIGHMEM
387 total
+= highend_pfn
- highstart_pfn
;
390 return total
<< PAGE_SHIFT
;
394 static void __init
reserve_crashkernel(void)
396 unsigned long long total_mem
;
397 unsigned long long crash_size
, crash_base
;
400 total_mem
= get_total_mem();
402 ret
= parse_crashkernel(boot_command_line
, total_mem
,
403 &crash_size
, &crash_base
);
404 if (ret
== 0 && crash_size
> 0) {
405 if (crash_base
> 0) {
406 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
407 "for crashkernel (System RAM: %ldMB)\n",
408 (unsigned long)(crash_size
>> 20),
409 (unsigned long)(crash_base
>> 20),
410 (unsigned long)(total_mem
>> 20));
411 crashk_res
.start
= crash_base
;
412 crashk_res
.end
= crash_base
+ crash_size
- 1;
413 reserve_bootmem(crash_base
, crash_size
);
415 printk(KERN_INFO
"crashkernel reservation failed - "
416 "you have to specify a base address\n");
420 static inline void __init
reserve_crashkernel(void)
424 void __init
setup_bootmem_allocator(void)
426 unsigned long bootmap_size
;
428 * Initialize the boot-time allocator (with low memory only):
430 bootmap_size
= init_bootmem(min_low_pfn
, max_low_pfn
);
432 register_bootmem_low_pages(max_low_pfn
);
435 * Reserve the bootmem bitmap itself as well. We do this in two
436 * steps (first step was init_bootmem()) because this catches
437 * the (very unlikely) case of us accidentally initializing the
438 * bootmem allocator with an invalid RAM area.
440 reserve_bootmem(__pa_symbol(_text
), (PFN_PHYS(min_low_pfn
) +
441 bootmap_size
+ PAGE_SIZE
-1) - __pa_symbol(_text
));
444 * reserve physical page 0 - it's a special BIOS page on many boxes,
445 * enabling clean reboots, SMP operation, laptop functions.
447 reserve_bootmem(0, PAGE_SIZE
);
449 /* reserve EBDA region, it's a 4K region */
450 reserve_ebda_region();
452 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
453 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
454 unless you have no PS/2 mouse plugged in. */
455 if (boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
456 boot_cpu_data
.x86
== 6)
457 reserve_bootmem(0xa0000 - 4096, 4096);
461 * But first pinch a few for the stack/trampoline stuff
462 * FIXME: Don't need the extra page at 4K, but need to fix
463 * trampoline before removing it. (see the GDT stuff)
465 reserve_bootmem(PAGE_SIZE
, PAGE_SIZE
);
467 #ifdef CONFIG_ACPI_SLEEP
469 * Reserve low memory region for sleep support.
471 acpi_reserve_bootmem();
473 #ifdef CONFIG_X86_FIND_SMP_CONFIG
475 * Find and reserve possible boot-time SMP configuration:
480 #ifdef CONFIG_BLK_DEV_INITRD
481 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
482 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
483 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
484 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
485 unsigned long end_of_lowmem
= max_low_pfn
<< PAGE_SHIFT
;
487 if (ramdisk_end
<= end_of_lowmem
) {
488 reserve_bootmem(ramdisk_image
, ramdisk_size
);
489 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
490 initrd_end
= initrd_start
+ramdisk_size
;
492 printk(KERN_ERR
"initrd extends beyond end of memory "
493 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
494 ramdisk_end
, end_of_lowmem
);
499 reserve_crashkernel();
503 * The node 0 pgdat is initialized before all of these because
504 * it's needed for bootmem. node>0 pgdats have their virtual
505 * space allocated before the pagetables are in place to access
506 * them, so they can't be cleared then.
508 * This should all compile down to nothing when NUMA is off.
510 static void __init
remapped_pgdat_init(void)
514 for_each_online_node(nid
) {
516 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
521 static void set_mca_bus(int x
)
526 static void set_mca_bus(int x
) { }
529 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
530 char * __init
__attribute__((weak
)) memory_setup(void)
532 return machine_specific_memory_setup();
536 * Determine if we were loaded by an EFI loader. If so, then we have also been
537 * passed the efi memmap, systab, etc., so we should use these data structures
538 * for initialization. Note, the efi init code path is determined by the
539 * global efi_enabled. This allows the same kernel image to be used on existing
540 * systems (with a traditional BIOS) as well as on EFI systems.
542 void __init
setup_arch(char **cmdline_p
)
544 unsigned long max_low_pfn
;
546 memcpy(&boot_cpu_data
, &new_cpu_data
, sizeof(new_cpu_data
));
547 pre_setup_arch_hook();
551 * FIXME: This isn't an official loader_type right
552 * now but does currently work with elilo.
553 * If we were configured as an EFI kernel, check to make
554 * sure that we were loaded correctly from elilo and that
555 * the system table is valid. If not, then initialize normally.
558 if ((boot_params
.hdr
.type_of_loader
== 0x50) &&
559 boot_params
.efi_info
.efi_systab
)
563 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
564 screen_info
= boot_params
.screen_info
;
565 edid_info
= boot_params
.edid_info
;
566 apm_info
.bios
= boot_params
.apm_bios_info
;
567 ist_info
= boot_params
.ist_info
;
568 saved_videomode
= boot_params
.hdr
.vid_mode
;
569 if( boot_params
.sys_desc_table
.length
!= 0 ) {
570 set_mca_bus(boot_params
.sys_desc_table
.table
[3] & 0x2);
571 machine_id
= boot_params
.sys_desc_table
.table
[0];
572 machine_submodel_id
= boot_params
.sys_desc_table
.table
[1];
573 BIOS_revision
= boot_params
.sys_desc_table
.table
[2];
575 bootloader_type
= boot_params
.hdr
.type_of_loader
;
577 #ifdef CONFIG_BLK_DEV_RAM
578 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
579 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
580 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
586 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
587 print_memory_map(memory_setup());
592 if (!boot_params
.hdr
.root_flags
)
593 root_mountflags
&= ~MS_RDONLY
;
594 init_mm
.start_code
= (unsigned long) _text
;
595 init_mm
.end_code
= (unsigned long) _etext
;
596 init_mm
.end_data
= (unsigned long) _edata
;
597 init_mm
.brk
= init_pg_tables_end
+ PAGE_OFFSET
;
599 code_resource
.start
= virt_to_phys(_text
);
600 code_resource
.end
= virt_to_phys(_etext
)-1;
601 data_resource
.start
= virt_to_phys(_etext
);
602 data_resource
.end
= virt_to_phys(_edata
)-1;
606 if (user_defined_memmap
) {
607 printk(KERN_INFO
"user-defined physical RAM map:\n");
608 print_memory_map("user");
611 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
612 *cmdline_p
= command_line
;
614 max_low_pfn
= setup_memory();
618 * Must be after max_low_pfn is determined, and before kernel
619 * pagetables are setup.
625 * NOTE: before this point _nobody_ is allowed to allocate
626 * any memory using the bootmem allocator. Although the
627 * alloctor is now initialised only the first 8Mb of the kernel
628 * virtual address space has been mapped. All allocations before
629 * paging_init() has completed must use the alloc_bootmem_low_pages()
630 * variant (which allocates DMA'able memory) and care must be taken
631 * not to exceed the 8Mb limit.
635 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
638 remapped_pgdat_init();
643 * NOTE: at this point the bootmem allocator is fully available.
646 paravirt_post_allocator_init();
650 #ifdef CONFIG_X86_GENERICARCH
651 generic_apic_probe();
658 * Parse the ACPI tables for possible boot-time SMP configuration.
660 acpi_boot_table_init();
664 #ifdef CONFIG_X86_IO_APIC
665 check_acpi_pci(); /* Checks more than just ACPI actually */
672 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
674 printk(KERN_WARNING
"More than 8 CPUs detected and "
675 "CONFIG_X86_PC cannot handle it.\nUse "
676 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
679 #ifdef CONFIG_X86_LOCAL_APIC
680 if (smp_found_config
)
684 e820_register_memory();
685 e820_mark_nosave_regions();
688 #if defined(CONFIG_VGA_CONSOLE)
689 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
690 conswitchp
= &vga_con
;
691 #elif defined(CONFIG_DUMMY_CONSOLE)
692 conswitchp
= &dummy_con
;