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>
63 #include <asm/cacheflush.h>
65 /* This value is set up by the early boot code to point to the value
66 immediately after the boot time page tables. It contains a *physical*
67 address, and must not be in the .bss segment! */
68 unsigned long init_pg_tables_end __initdata
= ~0UL;
70 int disable_pse __devinitdata
= 0;
75 extern struct resource code_resource
;
76 extern struct resource data_resource
;
77 extern struct resource bss_resource
;
79 /* cpu data as detected by the assembly code in head.S */
80 struct cpuinfo_x86 new_cpu_data __cpuinitdata
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
81 /* common cpu data for all cpus */
82 struct cpuinfo_x86 boot_cpu_data __read_mostly
= { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
83 EXPORT_SYMBOL(boot_cpu_data
);
85 unsigned long mmu_cr4_features
;
87 /* for MCA, but anyone else can use it if they want */
88 unsigned int machine_id
;
90 EXPORT_SYMBOL(machine_id
);
92 unsigned int machine_submodel_id
;
93 unsigned int BIOS_revision
;
94 unsigned int mca_pentium_flag
;
96 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
99 /* user-defined highmem size */
100 static unsigned int highmem_pages
= -1;
105 struct screen_info screen_info
;
106 EXPORT_SYMBOL(screen_info
);
107 struct apm_info apm_info
;
108 EXPORT_SYMBOL(apm_info
);
109 struct edid_info edid_info
;
110 EXPORT_SYMBOL_GPL(edid_info
);
111 struct ist_info ist_info
;
112 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
113 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
114 EXPORT_SYMBOL(ist_info
);
117 extern void early_cpu_init(void);
118 extern int root_mountflags
;
120 unsigned long saved_videomode
;
122 #define RAMDISK_IMAGE_START_MASK 0x07FF
123 #define RAMDISK_PROMPT_FLAG 0x8000
124 #define RAMDISK_LOAD_FLAG 0x4000
126 static char __initdata command_line
[COMMAND_LINE_SIZE
];
128 struct boot_params __initdata boot_params
;
130 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
132 #ifdef CONFIG_EDD_MODULE
136 * copy_edd() - Copy the BIOS EDD information
137 * from boot_params into a safe place.
140 static inline void copy_edd(void)
142 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
143 sizeof(edd
.mbr_signature
));
144 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
145 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
146 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
149 static inline void copy_edd(void)
154 int __initdata user_defined_memmap
= 0;
157 * "mem=nopentium" disables the 4MB page tables.
158 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
159 * to <mem>, overriding the bios size.
160 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
161 * <start> to <start>+<mem>, overriding the bios size.
163 * HPA tells me bootloaders need to parse mem=, so no new
164 * option should be mem= [also see Documentation/i386/boot.txt]
166 static int __init
parse_mem(char *arg
)
171 if (strcmp(arg
, "nopentium") == 0) {
172 clear_bit(X86_FEATURE_PSE
, boot_cpu_data
.x86_capability
);
175 /* If the user specifies memory size, we
176 * limit the BIOS-provided memory map to
177 * that size. exactmap can be used to specify
178 * the exact map. mem=number can be used to
179 * trim the existing memory map.
181 unsigned long long mem_size
;
183 mem_size
= memparse(arg
, &arg
);
184 limit_regions(mem_size
);
185 user_defined_memmap
= 1;
189 early_param("mem", parse_mem
);
191 #ifdef CONFIG_PROC_VMCORE
192 /* elfcorehdr= specifies the location of elf core header
193 * stored by the crashed kernel.
195 static int __init
parse_elfcorehdr(char *arg
)
200 elfcorehdr_addr
= memparse(arg
, &arg
);
203 early_param("elfcorehdr", parse_elfcorehdr
);
204 #endif /* CONFIG_PROC_VMCORE */
207 * highmem=size forces highmem to be exactly 'size' bytes.
208 * This works even on boxes that have no highmem otherwise.
209 * This also works to reduce highmem size on bigger boxes.
211 static int __init
parse_highmem(char *arg
)
216 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
219 early_param("highmem", parse_highmem
);
222 * vmalloc=size forces the vmalloc area to be exactly 'size'
223 * bytes. This can be used to increase (or decrease) the
224 * vmalloc area - the default is 128m.
226 static int __init
parse_vmalloc(char *arg
)
231 __VMALLOC_RESERVE
= memparse(arg
, &arg
);
234 early_param("vmalloc", parse_vmalloc
);
237 * reservetop=size reserves a hole at the top of the kernel address space which
238 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
239 * so relocating the fixmap can be done before paging initialization.
241 static int __init
parse_reservetop(char *arg
)
243 unsigned long address
;
248 address
= memparse(arg
, &arg
);
249 reserve_top_address(address
);
252 early_param("reservetop", parse_reservetop
);
255 * Determine low and high memory ranges:
257 unsigned long __init
find_max_low_pfn(void)
259 unsigned long max_low_pfn
;
261 max_low_pfn
= max_pfn
;
262 if (max_low_pfn
> MAXMEM_PFN
) {
263 if (highmem_pages
== -1)
264 highmem_pages
= max_pfn
- MAXMEM_PFN
;
265 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
266 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
267 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
268 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn
- MAXMEM_PFN
), pages_to_mb(highmem_pages
));
271 max_low_pfn
= MAXMEM_PFN
;
272 #ifndef CONFIG_HIGHMEM
273 /* Maximum memory usable is what is directly addressable */
274 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
276 if (max_pfn
> MAX_NONPAE_PFN
)
277 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
279 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
280 max_pfn
= MAXMEM_PFN
;
281 #else /* !CONFIG_HIGHMEM */
282 #ifndef CONFIG_HIGHMEM64G
283 if (max_pfn
> MAX_NONPAE_PFN
) {
284 max_pfn
= MAX_NONPAE_PFN
;
285 printk(KERN_WARNING
"Warning only 4GB will be used.\n");
286 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
288 #endif /* !CONFIG_HIGHMEM64G */
289 #endif /* !CONFIG_HIGHMEM */
291 if (highmem_pages
== -1)
293 #ifdef CONFIG_HIGHMEM
294 if (highmem_pages
>= max_pfn
) {
295 printk(KERN_ERR
"highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
299 if (max_low_pfn
-highmem_pages
< 64*1024*1024/PAGE_SIZE
){
300 printk(KERN_ERR
"highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages
));
303 max_low_pfn
-= highmem_pages
;
307 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
314 * workaround for Dell systems that neglect to reserve EBDA
316 static void __init
reserve_ebda_region(void)
319 addr
= get_bios_ebda();
321 reserve_bootmem(addr
, PAGE_SIZE
);
324 #ifndef CONFIG_NEED_MULTIPLE_NODES
325 void __init
setup_bootmem_allocator(void);
326 static unsigned long __init
setup_memory(void)
329 * partially used pages are not usable - thus
330 * we are rounding upwards:
332 min_low_pfn
= PFN_UP(init_pg_tables_end
);
336 max_low_pfn
= find_max_low_pfn();
338 #ifdef CONFIG_HIGHMEM
339 highstart_pfn
= highend_pfn
= max_pfn
;
340 if (max_pfn
> max_low_pfn
) {
341 highstart_pfn
= max_low_pfn
;
343 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
344 pages_to_mb(highend_pfn
- highstart_pfn
));
345 num_physpages
= highend_pfn
;
346 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
348 num_physpages
= max_low_pfn
;
349 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
351 #ifdef CONFIG_FLATMEM
352 max_mapnr
= num_physpages
;
354 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
355 pages_to_mb(max_low_pfn
));
357 setup_bootmem_allocator();
362 void __init
zone_sizes_init(void)
364 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
365 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
366 max_zone_pfns
[ZONE_DMA
] =
367 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
368 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
369 #ifdef CONFIG_HIGHMEM
370 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
371 add_active_range(0, 0, highend_pfn
);
373 add_active_range(0, 0, max_low_pfn
);
376 free_area_init_nodes(max_zone_pfns
);
379 extern unsigned long __init
setup_memory(void);
380 extern void zone_sizes_init(void);
381 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
383 static inline unsigned long long get_total_mem(void)
385 unsigned long long total
;
387 total
= max_low_pfn
- min_low_pfn
;
388 #ifdef CONFIG_HIGHMEM
389 total
+= highend_pfn
- highstart_pfn
;
392 return total
<< PAGE_SHIFT
;
396 static void __init
reserve_crashkernel(void)
398 unsigned long long total_mem
;
399 unsigned long long crash_size
, crash_base
;
402 total_mem
= get_total_mem();
404 ret
= parse_crashkernel(boot_command_line
, total_mem
,
405 &crash_size
, &crash_base
);
406 if (ret
== 0 && crash_size
> 0) {
407 if (crash_base
> 0) {
408 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
409 "for crashkernel (System RAM: %ldMB)\n",
410 (unsigned long)(crash_size
>> 20),
411 (unsigned long)(crash_base
>> 20),
412 (unsigned long)(total_mem
>> 20));
413 crashk_res
.start
= crash_base
;
414 crashk_res
.end
= crash_base
+ crash_size
- 1;
415 reserve_bootmem(crash_base
, crash_size
);
417 printk(KERN_INFO
"crashkernel reservation failed - "
418 "you have to specify a base address\n");
422 static inline void __init
reserve_crashkernel(void)
426 void __init
setup_bootmem_allocator(void)
428 unsigned long bootmap_size
;
430 * Initialize the boot-time allocator (with low memory only):
432 bootmap_size
= init_bootmem(min_low_pfn
, max_low_pfn
);
434 register_bootmem_low_pages(max_low_pfn
);
437 * Reserve the bootmem bitmap itself as well. We do this in two
438 * steps (first step was init_bootmem()) because this catches
439 * the (very unlikely) case of us accidentally initializing the
440 * bootmem allocator with an invalid RAM area.
442 reserve_bootmem(__pa_symbol(_text
), (PFN_PHYS(min_low_pfn
) +
443 bootmap_size
+ PAGE_SIZE
-1) - __pa_symbol(_text
));
446 * reserve physical page 0 - it's a special BIOS page on many boxes,
447 * enabling clean reboots, SMP operation, laptop functions.
449 reserve_bootmem(0, PAGE_SIZE
);
451 /* reserve EBDA region, it's a 4K region */
452 reserve_ebda_region();
454 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
455 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
456 unless you have no PS/2 mouse plugged in. */
457 if (boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
458 boot_cpu_data
.x86
== 6)
459 reserve_bootmem(0xa0000 - 4096, 4096);
463 * But first pinch a few for the stack/trampoline stuff
464 * FIXME: Don't need the extra page at 4K, but need to fix
465 * trampoline before removing it. (see the GDT stuff)
467 reserve_bootmem(PAGE_SIZE
, PAGE_SIZE
);
469 #ifdef CONFIG_ACPI_SLEEP
471 * Reserve low memory region for sleep support.
473 acpi_reserve_bootmem();
475 #ifdef CONFIG_X86_FIND_SMP_CONFIG
477 * Find and reserve possible boot-time SMP configuration:
482 #ifdef CONFIG_BLK_DEV_INITRD
483 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
484 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
485 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
486 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
487 unsigned long end_of_lowmem
= max_low_pfn
<< PAGE_SHIFT
;
489 if (ramdisk_end
<= end_of_lowmem
) {
490 reserve_bootmem(ramdisk_image
, ramdisk_size
);
491 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
492 initrd_end
= initrd_start
+ramdisk_size
;
494 printk(KERN_ERR
"initrd extends beyond end of memory "
495 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
496 ramdisk_end
, end_of_lowmem
);
501 reserve_crashkernel();
505 * The node 0 pgdat is initialized before all of these because
506 * it's needed for bootmem. node>0 pgdats have their virtual
507 * space allocated before the pagetables are in place to access
508 * them, so they can't be cleared then.
510 * This should all compile down to nothing when NUMA is off.
512 static void __init
remapped_pgdat_init(void)
516 for_each_online_node(nid
) {
518 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
523 static void set_mca_bus(int x
)
528 static void set_mca_bus(int x
) { }
531 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
532 char * __init
__attribute__((weak
)) memory_setup(void)
534 return machine_specific_memory_setup();
538 * Determine if we were loaded by an EFI loader. If so, then we have also been
539 * passed the efi memmap, systab, etc., so we should use these data structures
540 * for initialization. Note, the efi init code path is determined by the
541 * global efi_enabled. This allows the same kernel image to be used on existing
542 * systems (with a traditional BIOS) as well as on EFI systems.
544 void __init
setup_arch(char **cmdline_p
)
546 unsigned long max_low_pfn
;
548 memcpy(&boot_cpu_data
, &new_cpu_data
, sizeof(new_cpu_data
));
549 pre_setup_arch_hook();
553 * FIXME: This isn't an official loader_type right
554 * now but does currently work with elilo.
555 * If we were configured as an EFI kernel, check to make
556 * sure that we were loaded correctly from elilo and that
557 * the system table is valid. If not, then initialize normally.
560 if ((boot_params
.hdr
.type_of_loader
== 0x50) &&
561 boot_params
.efi_info
.efi_systab
)
565 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
566 screen_info
= boot_params
.screen_info
;
567 edid_info
= boot_params
.edid_info
;
568 apm_info
.bios
= boot_params
.apm_bios_info
;
569 ist_info
= boot_params
.ist_info
;
570 saved_videomode
= boot_params
.hdr
.vid_mode
;
571 if( boot_params
.sys_desc_table
.length
!= 0 ) {
572 set_mca_bus(boot_params
.sys_desc_table
.table
[3] & 0x2);
573 machine_id
= boot_params
.sys_desc_table
.table
[0];
574 machine_submodel_id
= boot_params
.sys_desc_table
.table
[1];
575 BIOS_revision
= boot_params
.sys_desc_table
.table
[2];
577 bootloader_type
= boot_params
.hdr
.type_of_loader
;
579 #ifdef CONFIG_BLK_DEV_RAM
580 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
581 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
582 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
588 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
589 print_memory_map(memory_setup());
594 if (!boot_params
.hdr
.root_flags
)
595 root_mountflags
&= ~MS_RDONLY
;
596 init_mm
.start_code
= (unsigned long) _text
;
597 init_mm
.end_code
= (unsigned long) _etext
;
598 init_mm
.end_data
= (unsigned long) _edata
;
599 init_mm
.brk
= init_pg_tables_end
+ PAGE_OFFSET
;
601 code_resource
.start
= virt_to_phys(_text
);
602 code_resource
.end
= virt_to_phys(_etext
)-1;
603 data_resource
.start
= virt_to_phys(_etext
);
604 data_resource
.end
= virt_to_phys(_edata
)-1;
605 bss_resource
.start
= virt_to_phys(&__bss_start
);
606 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
610 if (user_defined_memmap
) {
611 printk(KERN_INFO
"user-defined physical RAM map:\n");
612 print_memory_map("user");
615 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
616 *cmdline_p
= command_line
;
618 max_low_pfn
= setup_memory();
622 * Must be after max_low_pfn is determined, and before kernel
623 * pagetables are setup.
629 * NOTE: before this point _nobody_ is allowed to allocate
630 * any memory using the bootmem allocator. Although the
631 * allocator is now initialised only the first 8Mb of the kernel
632 * virtual address space has been mapped. All allocations before
633 * paging_init() has completed must use the alloc_bootmem_low_pages()
634 * variant (which allocates DMA'able memory) and care must be taken
635 * not to exceed the 8Mb limit.
639 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
642 remapped_pgdat_init();
647 * NOTE: at this point the bootmem allocator is fully available.
650 paravirt_post_allocator_init();
654 #ifdef CONFIG_X86_GENERICARCH
655 generic_apic_probe();
662 * Parse the ACPI tables for possible boot-time SMP configuration.
664 acpi_boot_table_init();
674 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
676 printk(KERN_WARNING
"More than 8 CPUs detected and "
677 "CONFIG_X86_PC cannot handle it.\nUse "
678 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
681 #ifdef CONFIG_X86_LOCAL_APIC
682 if (smp_found_config
)
686 e820_register_memory();
687 e820_mark_nosave_regions();
690 #if defined(CONFIG_VGA_CONSOLE)
691 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
692 conswitchp
= &vga_con
;
693 #elif defined(CONFIG_DUMMY_CONSOLE)
694 conswitchp
= &dummy_con
;