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/console.h>
37 #include <linux/mca.h>
38 #include <linux/root_dev.h>
39 #include <linux/highmem.h>
40 #include <linux/module.h>
41 #include <linux/efi.h>
42 #include <linux/init.h>
43 #include <linux/edd.h>
44 #include <linux/nodemask.h>
45 #include <linux/kexec.h>
46 #include <linux/crash_dump.h>
47 #include <linux/dmi.h>
48 #include <linux/pfn.h>
50 #include <video/edid.h>
54 #include <asm/mpspec.h>
55 #include <asm/mmzone.h>
56 #include <asm/setup.h>
57 #include <asm/arch_hooks.h>
58 #include <asm/sections.h>
59 #include <asm/io_apic.h>
63 #include <setup_arch.h>
64 #include <bios_ebda.h>
66 /* This value is set up by the early boot code to point to the value
67 immediately after the boot time page tables. It contains a *physical*
68 address, and must not be in the .bss segment! */
69 unsigned long init_pg_tables_end __initdata = ~0UL;
71 int disable_pse __devinitdata = 0;
76 extern struct resource code_resource;
77 extern struct resource data_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 drive_info_struct { char dummy[32]; } drive_info;
106 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
107 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
108 EXPORT_SYMBOL(drive_info);
110 struct screen_info screen_info;
111 EXPORT_SYMBOL(screen_info);
112 struct apm_info apm_info;
113 EXPORT_SYMBOL(apm_info);
114 struct sys_desc_table_struct {
115 unsigned short length;
116 unsigned char table[0];
118 struct edid_info edid_info;
119 EXPORT_SYMBOL_GPL(edid_info);
120 struct ist_info ist_info;
121 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
122 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
123 EXPORT_SYMBOL(ist_info);
126 extern void early_cpu_init(void);
127 extern int root_mountflags;
129 unsigned long saved_videomode;
131 #define RAMDISK_IMAGE_START_MASK 0x07FF
132 #define RAMDISK_PROMPT_FLAG 0x8000
133 #define RAMDISK_LOAD_FLAG 0x4000
135 static char __initdata command_line[COMMAND_LINE_SIZE];
137 unsigned char __initdata boot_params[PARAM_SIZE];
139 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
141 #ifdef CONFIG_EDD_MODULE
145 * copy_edd() - Copy the BIOS EDD information
146 * from boot_params into a safe place.
149 static inline void copy_edd(void)
151 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
152 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
153 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
154 edd.edd_info_nr = EDD_NR;
157 static inline void copy_edd(void)
162 int __initdata user_defined_memmap = 0;
165 * "mem=nopentium" disables the 4MB page tables.
166 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
167 * to <mem>, overriding the bios size.
168 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
169 * <start> to <start>+<mem>, overriding the bios size.
171 * HPA tells me bootloaders need to parse mem=, so no new
172 * option should be mem= [also see Documentation/i386/boot.txt]
174 static int __init parse_mem(char *arg)
179 if (strcmp(arg, "nopentium") == 0) {
180 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
183 /* If the user specifies memory size, we
184 * limit the BIOS-provided memory map to
185 * that size. exactmap can be used to specify
186 * the exact map. mem=number can be used to
187 * trim the existing memory map.
189 unsigned long long mem_size;
191 mem_size = memparse(arg, &arg);
192 limit_regions(mem_size);
193 user_defined_memmap = 1;
197 early_param("mem", parse_mem);
199 #ifdef CONFIG_PROC_VMCORE
200 /* elfcorehdr= specifies the location of elf core header
201 * stored by the crashed kernel.
203 static int __init parse_elfcorehdr(char *arg)
208 elfcorehdr_addr = memparse(arg, &arg);
211 early_param("elfcorehdr", parse_elfcorehdr);
212 #endif /* CONFIG_PROC_VMCORE */
215 * highmem=size forces highmem to be exactly 'size' bytes.
216 * This works even on boxes that have no highmem otherwise.
217 * This also works to reduce highmem size on bigger boxes.
219 static int __init parse_highmem(char *arg)
224 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
227 early_param("highmem", parse_highmem);
230 * vmalloc=size forces the vmalloc area to be exactly 'size'
231 * bytes. This can be used to increase (or decrease) the
232 * vmalloc area - the default is 128m.
234 static int __init parse_vmalloc(char *arg)
239 __VMALLOC_RESERVE = memparse(arg, &arg);
242 early_param("vmalloc", parse_vmalloc);
245 * reservetop=size reserves a hole at the top of the kernel address space which
246 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
247 * so relocating the fixmap can be done before paging initialization.
249 static int __init parse_reservetop(char *arg)
251 unsigned long address;
256 address = memparse(arg, &arg);
257 reserve_top_address(address);
260 early_param("reservetop", parse_reservetop);
263 * Determine low and high memory ranges:
265 unsigned long __init find_max_low_pfn(void)
267 unsigned long max_low_pfn;
269 max_low_pfn = max_pfn;
270 if (max_low_pfn > MAXMEM_PFN) {
271 if (highmem_pages == -1)
272 highmem_pages = max_pfn - MAXMEM_PFN;
273 if (highmem_pages + MAXMEM_PFN < max_pfn)
274 max_pfn = MAXMEM_PFN + highmem_pages;
275 if (highmem_pages + MAXMEM_PFN > max_pfn) {
276 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
279 max_low_pfn = MAXMEM_PFN;
280 #ifndef CONFIG_HIGHMEM
281 /* Maximum memory usable is what is directly addressable */
282 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
284 if (max_pfn > MAX_NONPAE_PFN)
285 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
287 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
288 max_pfn = MAXMEM_PFN;
289 #else /* !CONFIG_HIGHMEM */
290 #ifndef CONFIG_X86_PAE
291 if (max_pfn > MAX_NONPAE_PFN) {
292 max_pfn = MAX_NONPAE_PFN;
293 printk(KERN_WARNING "Warning only 4GB will be used.\n");
294 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
296 #endif /* !CONFIG_X86_PAE */
297 #endif /* !CONFIG_HIGHMEM */
299 if (highmem_pages == -1)
301 #ifdef CONFIG_HIGHMEM
302 if (highmem_pages >= max_pfn) {
303 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
307 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
308 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
311 max_low_pfn -= highmem_pages;
315 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
322 * workaround for Dell systems that neglect to reserve EBDA
324 static void __init reserve_ebda_region(void)
327 addr = get_bios_ebda();
329 reserve_bootmem(addr, PAGE_SIZE);
332 #ifndef CONFIG_NEED_MULTIPLE_NODES
333 void __init setup_bootmem_allocator(void);
334 static unsigned long __init setup_memory(void)
337 * partially used pages are not usable - thus
338 * we are rounding upwards:
340 min_low_pfn = PFN_UP(init_pg_tables_end);
344 max_low_pfn = find_max_low_pfn();
346 #ifdef CONFIG_HIGHMEM
347 highstart_pfn = highend_pfn = max_pfn;
348 if (max_pfn > max_low_pfn) {
349 highstart_pfn = max_low_pfn;
351 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
352 pages_to_mb(highend_pfn - highstart_pfn));
353 num_physpages = highend_pfn;
354 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
356 num_physpages = max_low_pfn;
357 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
359 #ifdef CONFIG_FLATMEM
360 max_mapnr = num_physpages;
362 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
363 pages_to_mb(max_low_pfn));
365 setup_bootmem_allocator();
370 void __init zone_sizes_init(void)
372 unsigned long max_zone_pfns[MAX_NR_ZONES];
373 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
374 max_zone_pfns[ZONE_DMA] =
375 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
376 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
377 #ifdef CONFIG_HIGHMEM
378 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
379 add_active_range(0, 0, highend_pfn);
381 add_active_range(0, 0, max_low_pfn);
384 free_area_init_nodes(max_zone_pfns);
387 extern unsigned long __init setup_memory(void);
388 extern void zone_sizes_init(void);
389 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
391 void __init setup_bootmem_allocator(void)
393 unsigned long bootmap_size;
395 * Initialize the boot-time allocator (with low memory only):
397 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
399 register_bootmem_low_pages(max_low_pfn);
402 * Reserve the bootmem bitmap itself as well. We do this in two
403 * steps (first step was init_bootmem()) because this catches
404 * the (very unlikely) case of us accidentally initializing the
405 * bootmem allocator with an invalid RAM area.
407 reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
408 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text));
411 * reserve physical page 0 - it's a special BIOS page on many boxes,
412 * enabling clean reboots, SMP operation, laptop functions.
414 reserve_bootmem(0, PAGE_SIZE);
416 /* reserve EBDA region, it's a 4K region */
417 reserve_ebda_region();
419 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
420 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
421 unless you have no PS/2 mouse plugged in. */
422 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
423 boot_cpu_data.x86 == 6)
424 reserve_bootmem(0xa0000 - 4096, 4096);
428 * But first pinch a few for the stack/trampoline stuff
429 * FIXME: Don't need the extra page at 4K, but need to fix
430 * trampoline before removing it. (see the GDT stuff)
432 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
434 #ifdef CONFIG_ACPI_SLEEP
436 * Reserve low memory region for sleep support.
438 acpi_reserve_bootmem();
440 #ifdef CONFIG_X86_FIND_SMP_CONFIG
442 * Find and reserve possible boot-time SMP configuration:
447 #ifdef CONFIG_BLK_DEV_INITRD
448 if (LOADER_TYPE && INITRD_START) {
449 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
450 reserve_bootmem(INITRD_START, INITRD_SIZE);
451 initrd_start = INITRD_START + PAGE_OFFSET;
452 initrd_end = initrd_start+INITRD_SIZE;
455 printk(KERN_ERR "initrd extends beyond end of memory "
456 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
457 INITRD_START + INITRD_SIZE,
458 max_low_pfn << PAGE_SHIFT);
464 if (crashk_res.start != crashk_res.end)
465 reserve_bootmem(crashk_res.start,
466 crashk_res.end - crashk_res.start + 1);
471 * The node 0 pgdat is initialized before all of these because
472 * it's needed for bootmem. node>0 pgdats have their virtual
473 * space allocated before the pagetables are in place to access
474 * them, so they can't be cleared then.
476 * This should all compile down to nothing when NUMA is off.
478 void __init remapped_pgdat_init(void)
482 for_each_online_node(nid) {
484 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
489 static void set_mca_bus(int x)
494 static void set_mca_bus(int x) { }
497 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
498 char * __init __attribute__((weak)) memory_setup(void)
500 return machine_specific_memory_setup();
504 * Determine if we were loaded by an EFI loader. If so, then we have also been
505 * passed the efi memmap, systab, etc., so we should use these data structures
506 * for initialization. Note, the efi init code path is determined by the
507 * global efi_enabled. This allows the same kernel image to be used on existing
508 * systems (with a traditional BIOS) as well as on EFI systems.
510 void __init setup_arch(char **cmdline_p)
512 unsigned long max_low_pfn;
514 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
515 pre_setup_arch_hook();
519 * FIXME: This isn't an official loader_type right
520 * now but does currently work with elilo.
521 * If we were configured as an EFI kernel, check to make
522 * sure that we were loaded correctly from elilo and that
523 * the system table is valid. If not, then initialize normally.
526 if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
530 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
531 drive_info = DRIVE_INFO;
532 screen_info = SCREEN_INFO;
533 edid_info = EDID_INFO;
534 apm_info.bios = APM_BIOS_INFO;
536 saved_videomode = VIDEO_MODE;
537 if( SYS_DESC_TABLE.length != 0 ) {
538 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
539 machine_id = SYS_DESC_TABLE.table[0];
540 machine_submodel_id = SYS_DESC_TABLE.table[1];
541 BIOS_revision = SYS_DESC_TABLE.table[2];
543 bootloader_type = LOADER_TYPE;
545 #ifdef CONFIG_BLK_DEV_RAM
546 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
547 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
548 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
554 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
555 print_memory_map(memory_setup());
560 if (!MOUNT_ROOT_RDONLY)
561 root_mountflags &= ~MS_RDONLY;
562 init_mm.start_code = (unsigned long) _text;
563 init_mm.end_code = (unsigned long) _etext;
564 init_mm.end_data = (unsigned long) _edata;
565 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
567 code_resource.start = virt_to_phys(_text);
568 code_resource.end = virt_to_phys(_etext)-1;
569 data_resource.start = virt_to_phys(_etext);
570 data_resource.end = virt_to_phys(_edata)-1;
574 if (user_defined_memmap) {
575 printk(KERN_INFO "user-defined physical RAM map:\n");
576 print_memory_map("user");
579 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
580 *cmdline_p = command_line;
582 max_low_pfn = setup_memory();
586 * Must be after max_low_pfn is determined, and before kernel
587 * pagetables are setup.
593 * NOTE: before this point _nobody_ is allowed to allocate
594 * any memory using the bootmem allocator. Although the
595 * alloctor is now initialised only the first 8Mb of the kernel
596 * virtual address space has been mapped. All allocations before
597 * paging_init() has completed must use the alloc_bootmem_low_pages()
598 * variant (which allocates DMA'able memory) and care must be taken
599 * not to exceed the 8Mb limit.
603 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
606 remapped_pgdat_init();
611 * NOTE: at this point the bootmem allocator is fully available.
616 #ifdef CONFIG_X86_GENERICARCH
617 generic_apic_probe();
624 * Parse the ACPI tables for possible boot-time SMP configuration.
626 acpi_boot_table_init();
630 #ifdef CONFIG_X86_IO_APIC
631 check_acpi_pci(); /* Checks more than just ACPI actually */
638 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
640 printk(KERN_WARNING "More than 8 CPUs detected and "
641 "CONFIG_X86_PC cannot handle it.\nUse "
642 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
645 #ifdef CONFIG_X86_LOCAL_APIC
646 if (smp_found_config)
650 e820_register_memory();
653 #if defined(CONFIG_VGA_CONSOLE)
654 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
655 conswitchp = &vga_con;
656 #elif defined(CONFIG_DUMMY_CONSOLE)
657 conswitchp = &dummy_con;