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 __cpuinitdata
= 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
;
89 unsigned int machine_submodel_id
;
90 unsigned int BIOS_revision
;
91 unsigned int mca_pentium_flag
;
93 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
96 /* user-defined highmem size */
97 static unsigned int highmem_pages
= -1;
102 struct screen_info screen_info
;
103 EXPORT_SYMBOL(screen_info
);
104 struct apm_info apm_info
;
105 EXPORT_SYMBOL(apm_info
);
106 struct edid_info edid_info
;
107 EXPORT_SYMBOL_GPL(edid_info
);
108 struct ist_info ist_info
;
109 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
110 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
111 EXPORT_SYMBOL(ist_info
);
114 extern void early_cpu_init(void);
115 extern int root_mountflags
;
117 unsigned long saved_videomode
;
119 #define RAMDISK_IMAGE_START_MASK 0x07FF
120 #define RAMDISK_PROMPT_FLAG 0x8000
121 #define RAMDISK_LOAD_FLAG 0x4000
123 static char __initdata command_line
[COMMAND_LINE_SIZE
];
125 struct boot_params __initdata boot_params
;
127 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
129 #ifdef CONFIG_EDD_MODULE
133 * copy_edd() - Copy the BIOS EDD information
134 * from boot_params into a safe place.
137 static inline void copy_edd(void)
139 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
140 sizeof(edd
.mbr_signature
));
141 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
142 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
143 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
146 static inline void copy_edd(void)
151 int __initdata user_defined_memmap
= 0;
154 * "mem=nopentium" disables the 4MB page tables.
155 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
156 * to <mem>, overriding the bios size.
157 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
158 * <start> to <start>+<mem>, overriding the bios size.
160 * HPA tells me bootloaders need to parse mem=, so no new
161 * option should be mem= [also see Documentation/i386/boot.txt]
163 static int __init
parse_mem(char *arg
)
168 if (strcmp(arg
, "nopentium") == 0) {
169 clear_bit(X86_FEATURE_PSE
, boot_cpu_data
.x86_capability
);
172 /* If the user specifies memory size, we
173 * limit the BIOS-provided memory map to
174 * that size. exactmap can be used to specify
175 * the exact map. mem=number can be used to
176 * trim the existing memory map.
178 unsigned long long mem_size
;
180 mem_size
= memparse(arg
, &arg
);
181 limit_regions(mem_size
);
182 user_defined_memmap
= 1;
186 early_param("mem", parse_mem
);
188 #ifdef CONFIG_PROC_VMCORE
189 /* elfcorehdr= specifies the location of elf core header
190 * stored by the crashed kernel.
192 static int __init
parse_elfcorehdr(char *arg
)
197 elfcorehdr_addr
= memparse(arg
, &arg
);
200 early_param("elfcorehdr", parse_elfcorehdr
);
201 #endif /* CONFIG_PROC_VMCORE */
204 * highmem=size forces highmem to be exactly 'size' bytes.
205 * This works even on boxes that have no highmem otherwise.
206 * This also works to reduce highmem size on bigger boxes.
208 static int __init
parse_highmem(char *arg
)
213 highmem_pages
= memparse(arg
, &arg
) >> PAGE_SHIFT
;
216 early_param("highmem", parse_highmem
);
219 * vmalloc=size forces the vmalloc area to be exactly 'size'
220 * bytes. This can be used to increase (or decrease) the
221 * vmalloc area - the default is 128m.
223 static int __init
parse_vmalloc(char *arg
)
228 __VMALLOC_RESERVE
= memparse(arg
, &arg
);
231 early_param("vmalloc", parse_vmalloc
);
234 * reservetop=size reserves a hole at the top of the kernel address space which
235 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
236 * so relocating the fixmap can be done before paging initialization.
238 static int __init
parse_reservetop(char *arg
)
240 unsigned long address
;
245 address
= memparse(arg
, &arg
);
246 reserve_top_address(address
);
249 early_param("reservetop", parse_reservetop
);
252 * Determine low and high memory ranges:
254 unsigned long __init
find_max_low_pfn(void)
256 unsigned long max_low_pfn
;
258 max_low_pfn
= max_pfn
;
259 if (max_low_pfn
> MAXMEM_PFN
) {
260 if (highmem_pages
== -1)
261 highmem_pages
= max_pfn
- MAXMEM_PFN
;
262 if (highmem_pages
+ MAXMEM_PFN
< max_pfn
)
263 max_pfn
= MAXMEM_PFN
+ highmem_pages
;
264 if (highmem_pages
+ MAXMEM_PFN
> max_pfn
) {
265 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn
- MAXMEM_PFN
), pages_to_mb(highmem_pages
));
268 max_low_pfn
= MAXMEM_PFN
;
269 #ifndef CONFIG_HIGHMEM
270 /* Maximum memory usable is what is directly addressable */
271 printk(KERN_WARNING
"Warning only %ldMB will be used.\n",
273 if (max_pfn
> MAX_NONPAE_PFN
)
274 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
276 printk(KERN_WARNING
"Use a HIGHMEM enabled kernel.\n");
277 max_pfn
= MAXMEM_PFN
;
278 #else /* !CONFIG_HIGHMEM */
279 #ifndef CONFIG_HIGHMEM64G
280 if (max_pfn
> MAX_NONPAE_PFN
) {
281 max_pfn
= MAX_NONPAE_PFN
;
282 printk(KERN_WARNING
"Warning only 4GB will be used.\n");
283 printk(KERN_WARNING
"Use a HIGHMEM64G enabled kernel.\n");
285 #endif /* !CONFIG_HIGHMEM64G */
286 #endif /* !CONFIG_HIGHMEM */
288 if (highmem_pages
== -1)
290 #ifdef CONFIG_HIGHMEM
291 if (highmem_pages
>= max_pfn
) {
292 printk(KERN_ERR
"highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages
), pages_to_mb(max_pfn
));
296 if (max_low_pfn
-highmem_pages
< 64*1024*1024/PAGE_SIZE
){
297 printk(KERN_ERR
"highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages
));
300 max_low_pfn
-= highmem_pages
;
304 printk(KERN_ERR
"ignoring highmem size on non-highmem kernel!\n");
311 * workaround for Dell systems that neglect to reserve EBDA
313 static void __init
reserve_ebda_region(void)
316 addr
= get_bios_ebda();
318 reserve_bootmem(addr
, PAGE_SIZE
);
321 #ifndef CONFIG_NEED_MULTIPLE_NODES
322 void __init
setup_bootmem_allocator(void);
323 static unsigned long __init
setup_memory(void)
326 * partially used pages are not usable - thus
327 * we are rounding upwards:
329 min_low_pfn
= PFN_UP(init_pg_tables_end
);
333 max_low_pfn
= find_max_low_pfn();
335 #ifdef CONFIG_HIGHMEM
336 highstart_pfn
= highend_pfn
= max_pfn
;
337 if (max_pfn
> max_low_pfn
) {
338 highstart_pfn
= max_low_pfn
;
340 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
341 pages_to_mb(highend_pfn
- highstart_pfn
));
342 num_physpages
= highend_pfn
;
343 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
345 num_physpages
= max_low_pfn
;
346 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
348 #ifdef CONFIG_FLATMEM
349 max_mapnr
= num_physpages
;
351 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
352 pages_to_mb(max_low_pfn
));
354 setup_bootmem_allocator();
359 void __init
zone_sizes_init(void)
361 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
362 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
363 max_zone_pfns
[ZONE_DMA
] =
364 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
365 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
366 #ifdef CONFIG_HIGHMEM
367 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
368 add_active_range(0, 0, highend_pfn
);
370 add_active_range(0, 0, max_low_pfn
);
373 free_area_init_nodes(max_zone_pfns
);
376 extern unsigned long __init
setup_memory(void);
377 extern void zone_sizes_init(void);
378 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
380 static inline unsigned long long get_total_mem(void)
382 unsigned long long total
;
384 total
= max_low_pfn
- min_low_pfn
;
385 #ifdef CONFIG_HIGHMEM
386 total
+= highend_pfn
- highstart_pfn
;
389 return total
<< PAGE_SHIFT
;
393 static void __init
reserve_crashkernel(void)
395 unsigned long long total_mem
;
396 unsigned long long crash_size
, crash_base
;
399 total_mem
= get_total_mem();
401 ret
= parse_crashkernel(boot_command_line
, total_mem
,
402 &crash_size
, &crash_base
);
403 if (ret
== 0 && crash_size
> 0) {
404 if (crash_base
> 0) {
405 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
406 "for crashkernel (System RAM: %ldMB)\n",
407 (unsigned long)(crash_size
>> 20),
408 (unsigned long)(crash_base
>> 20),
409 (unsigned long)(total_mem
>> 20));
410 crashk_res
.start
= crash_base
;
411 crashk_res
.end
= crash_base
+ crash_size
- 1;
412 reserve_bootmem(crash_base
, crash_size
);
414 printk(KERN_INFO
"crashkernel reservation failed - "
415 "you have to specify a base address\n");
419 static inline void __init
reserve_crashkernel(void)
423 void __init
setup_bootmem_allocator(void)
425 unsigned long bootmap_size
;
427 * Initialize the boot-time allocator (with low memory only):
429 bootmap_size
= init_bootmem(min_low_pfn
, max_low_pfn
);
431 register_bootmem_low_pages(max_low_pfn
);
434 * Reserve the bootmem bitmap itself as well. We do this in two
435 * steps (first step was init_bootmem()) because this catches
436 * the (very unlikely) case of us accidentally initializing the
437 * bootmem allocator with an invalid RAM area.
439 reserve_bootmem(__pa_symbol(_text
), (PFN_PHYS(min_low_pfn
) +
440 bootmap_size
+ PAGE_SIZE
-1) - __pa_symbol(_text
));
443 * reserve physical page 0 - it's a special BIOS page on many boxes,
444 * enabling clean reboots, SMP operation, laptop functions.
446 reserve_bootmem(0, PAGE_SIZE
);
448 /* reserve EBDA region, it's a 4K region */
449 reserve_ebda_region();
451 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
452 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
453 unless you have no PS/2 mouse plugged in. */
454 if (boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
455 boot_cpu_data
.x86
== 6)
456 reserve_bootmem(0xa0000 - 4096, 4096);
460 * But first pinch a few for the stack/trampoline stuff
461 * FIXME: Don't need the extra page at 4K, but need to fix
462 * trampoline before removing it. (see the GDT stuff)
464 reserve_bootmem(PAGE_SIZE
, PAGE_SIZE
);
466 #ifdef CONFIG_ACPI_SLEEP
468 * Reserve low memory region for sleep support.
470 acpi_reserve_bootmem();
472 #ifdef CONFIG_X86_FIND_SMP_CONFIG
474 * Find and reserve possible boot-time SMP configuration:
479 #ifdef CONFIG_BLK_DEV_INITRD
480 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
481 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
482 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
483 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
484 unsigned long end_of_lowmem
= max_low_pfn
<< PAGE_SHIFT
;
486 if (ramdisk_end
<= end_of_lowmem
) {
487 reserve_bootmem(ramdisk_image
, ramdisk_size
);
488 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
489 initrd_end
= initrd_start
+ramdisk_size
;
491 printk(KERN_ERR
"initrd extends beyond end of memory "
492 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
493 ramdisk_end
, end_of_lowmem
);
498 reserve_crashkernel();
502 * The node 0 pgdat is initialized before all of these because
503 * it's needed for bootmem. node>0 pgdats have their virtual
504 * space allocated before the pagetables are in place to access
505 * them, so they can't be cleared then.
507 * This should all compile down to nothing when NUMA is off.
509 static void __init
remapped_pgdat_init(void)
513 for_each_online_node(nid
) {
515 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
520 static void set_mca_bus(int x
)
525 static void set_mca_bus(int x
) { }
528 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
529 char * __init
__attribute__((weak
)) memory_setup(void)
531 return machine_specific_memory_setup();
535 * Determine if we were loaded by an EFI loader. If so, then we have also been
536 * passed the efi memmap, systab, etc., so we should use these data structures
537 * for initialization. Note, the efi init code path is determined by the
538 * global efi_enabled. This allows the same kernel image to be used on existing
539 * systems (with a traditional BIOS) as well as on EFI systems.
541 void __init
setup_arch(char **cmdline_p
)
543 unsigned long max_low_pfn
;
545 memcpy(&boot_cpu_data
, &new_cpu_data
, sizeof(new_cpu_data
));
546 pre_setup_arch_hook();
550 * FIXME: This isn't an official loader_type right
551 * now but does currently work with elilo.
552 * If we were configured as an EFI kernel, check to make
553 * sure that we were loaded correctly from elilo and that
554 * the system table is valid. If not, then initialize normally.
557 if ((boot_params
.hdr
.type_of_loader
== 0x50) &&
558 boot_params
.efi_info
.efi_systab
)
562 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
563 screen_info
= boot_params
.screen_info
;
564 edid_info
= boot_params
.edid_info
;
565 apm_info
.bios
= boot_params
.apm_bios_info
;
566 ist_info
= boot_params
.ist_info
;
567 saved_videomode
= boot_params
.hdr
.vid_mode
;
568 if( boot_params
.sys_desc_table
.length
!= 0 ) {
569 set_mca_bus(boot_params
.sys_desc_table
.table
[3] & 0x2);
570 machine_id
= boot_params
.sys_desc_table
.table
[0];
571 machine_submodel_id
= boot_params
.sys_desc_table
.table
[1];
572 BIOS_revision
= boot_params
.sys_desc_table
.table
[2];
574 bootloader_type
= boot_params
.hdr
.type_of_loader
;
576 #ifdef CONFIG_BLK_DEV_RAM
577 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
578 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
579 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
585 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
586 print_memory_map(memory_setup());
591 if (!boot_params
.hdr
.root_flags
)
592 root_mountflags
&= ~MS_RDONLY
;
593 init_mm
.start_code
= (unsigned long) _text
;
594 init_mm
.end_code
= (unsigned long) _etext
;
595 init_mm
.end_data
= (unsigned long) _edata
;
596 init_mm
.brk
= init_pg_tables_end
+ PAGE_OFFSET
;
598 code_resource
.start
= virt_to_phys(_text
);
599 code_resource
.end
= virt_to_phys(_etext
)-1;
600 data_resource
.start
= virt_to_phys(_etext
);
601 data_resource
.end
= virt_to_phys(_edata
)-1;
602 bss_resource
.start
= virt_to_phys(&__bss_start
);
603 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
607 if (user_defined_memmap
) {
608 printk(KERN_INFO
"user-defined physical RAM map:\n");
609 print_memory_map("user");
612 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
613 *cmdline_p
= command_line
;
615 max_low_pfn
= setup_memory();
619 * Must be after max_low_pfn is determined, and before kernel
620 * pagetables are setup.
626 * NOTE: before this point _nobody_ is allowed to allocate
627 * any memory using the bootmem allocator. Although the
628 * allocator is now initialised only the first 8Mb of the kernel
629 * virtual address space has been mapped. All allocations before
630 * paging_init() has completed must use the alloc_bootmem_low_pages()
631 * variant (which allocates DMA'able memory) and care must be taken
632 * not to exceed the 8Mb limit.
636 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
639 remapped_pgdat_init();
644 * NOTE: at this point the bootmem allocator is fully available.
647 paravirt_post_allocator_init();
651 #ifdef CONFIG_X86_GENERICARCH
652 generic_apic_probe();
659 * Parse the ACPI tables for possible boot-time SMP configuration.
661 acpi_boot_table_init();
671 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
673 printk(KERN_WARNING
"More than 8 CPUs detected and "
674 "CONFIG_X86_PC cannot handle it.\nUse "
675 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
678 #ifdef CONFIG_X86_LOCAL_APIC
679 if (smp_found_config
)
683 e820_register_memory();
684 e820_mark_nosave_regions();
687 #if defined(CONFIG_VGA_CONSOLE)
688 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
689 conswitchp
= &vga_con
;
690 #elif defined(CONFIG_DUMMY_CONSOLE)
691 conswitchp
= &dummy_con
;