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/iscsi_ibft.h>
43 #include <linux/nodemask.h>
44 #include <linux/kexec.h>
45 #include <linux/dmi.h>
46 #include <linux/pfn.h>
47 #include <linux/pci.h>
48 #include <asm/pci-direct.h>
49 #include <linux/init_ohci1394_dma.h>
50 #include <linux/kvm_para.h>
52 #include <linux/errno.h>
53 #include <linux/kernel.h>
54 #include <linux/stddef.h>
55 #include <linux/unistd.h>
56 #include <linux/ptrace.h>
57 #include <linux/slab.h>
58 #include <linux/user.h>
59 #include <linux/delay.h>
60 #include <linux/highmem.h>
62 #include <linux/kallsyms.h>
63 #include <linux/edd.h>
64 #include <linux/iscsi_ibft.h>
65 #include <linux/kexec.h>
66 #include <linux/cpufreq.h>
67 #include <linux/dma-mapping.h>
68 #include <linux/ctype.h>
69 #include <linux/uaccess.h>
71 #include <linux/percpu.h>
72 #include <linux/crash_dump.h>
74 #include <video/edid.h>
79 #include <asm/mpspec.h>
80 #include <asm/setup.h>
81 #include <asm/arch_hooks.h>
83 #include <asm/sections.h>
85 #include <asm/io_apic.h>
88 #include <setup_arch.h>
89 #include <asm/bios_ebda.h>
90 #include <asm/cacheflush.h>
91 #include <asm/processor.h>
94 #include <asm/system.h>
95 #include <asm/vsyscall.h>
100 #include <asm/mmu_context.h>
101 #include <asm/proto.h>
103 #include <mach_apic.h>
104 #ifdef CONFIG_PARAVIRT
105 #include <asm/paravirt.h>
110 #include <asm/percpu.h>
111 #include <asm/sections.h>
112 #include <asm/topology.h>
113 #include <asm/apicdef.h>
115 #include <asm/numa_64.h>
118 #include <asm/highmem.h>
121 #ifndef CONFIG_DEBUG_BOOT_PARAMS
122 struct boot_params __initdata boot_params;
124 struct boot_params boot_params;
130 static struct resource data_resource = {
131 .name = "Kernel data",
134 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
137 static struct resource code_resource = {
138 .name = "Kernel code",
141 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
144 static struct resource bss_resource = {
145 .name = "Kernel bss",
148 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
153 /* This value is set up by the early boot code to point to the value
154 immediately after the boot time page tables. It contains a *physical*
155 address, and must not be in the .bss segment! */
156 unsigned long init_pg_tables_start __initdata = ~0UL;
157 unsigned long init_pg_tables_end __initdata = ~0UL;
159 static struct resource video_ram_resource = {
160 .name = "Video RAM area",
163 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
166 /* cpu data as detected by the assembly code in head.S */
167 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
168 /* common cpu data for all cpus */
169 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
170 EXPORT_SYMBOL(boot_cpu_data);
171 static void set_mca_bus(int x)
178 unsigned int def_to_bigsmp;
180 /* for MCA, but anyone else can use it if they want */
181 unsigned int machine_id;
182 unsigned int machine_submodel_id;
183 unsigned int BIOS_revision;
185 struct apm_info apm_info;
186 EXPORT_SYMBOL(apm_info);
188 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
189 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
190 struct ist_info ist_info;
191 EXPORT_SYMBOL(ist_info);
193 struct ist_info ist_info;
197 struct cpuinfo_x86 boot_cpu_data __read_mostly;
198 EXPORT_SYMBOL(boot_cpu_data);
202 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
203 unsigned long mmu_cr4_features;
205 unsigned long mmu_cr4_features = X86_CR4_PAE;
208 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
215 char dmi_alloc_data[DMI_MAX_DATA];
220 struct screen_info screen_info;
221 EXPORT_SYMBOL(screen_info);
222 struct edid_info edid_info;
223 EXPORT_SYMBOL_GPL(edid_info);
225 extern int root_mountflags;
227 unsigned long saved_video_mode;
229 #define RAMDISK_IMAGE_START_MASK 0x07FF
230 #define RAMDISK_PROMPT_FLAG 0x8000
231 #define RAMDISK_LOAD_FLAG 0x4000
233 static char __initdata command_line[COMMAND_LINE_SIZE];
235 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
237 #ifdef CONFIG_EDD_MODULE
241 * copy_edd() - Copy the BIOS EDD information
242 * from boot_params into a safe place.
245 static inline void copy_edd(void)
247 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
248 sizeof(edd.mbr_signature));
249 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
250 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
251 edd.edd_info_nr = boot_params.eddbuf_entries;
254 static inline void copy_edd(void)
259 #ifdef CONFIG_BLK_DEV_INITRD
263 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
264 static void __init relocate_initrd(void)
267 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
268 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
269 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
271 unsigned long slop, clen, mapaddr;
274 /* We need to move the initrd down into lowmem */
275 ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
278 if (ramdisk_here == -1ULL)
279 panic("Cannot find place for new RAMDISK of size %lld\n",
282 /* Note: this includes all the lowmem currently occupied by
283 the initrd, we rely on that fact to keep the data intact. */
284 reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
286 initrd_start = ramdisk_here + PAGE_OFFSET;
287 initrd_end = initrd_start + ramdisk_size;
288 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
289 ramdisk_here, ramdisk_here + ramdisk_size);
291 q = (char *)initrd_start;
293 /* Copy any lowmem portion of the initrd */
294 if (ramdisk_image < end_of_lowmem) {
295 clen = end_of_lowmem - ramdisk_image;
296 p = (char *)__va(ramdisk_image);
299 ramdisk_image += clen;
300 ramdisk_size -= clen;
303 /* Copy the highmem portion of the initrd */
304 while (ramdisk_size) {
305 slop = ramdisk_image & ~PAGE_MASK;
307 if (clen > MAX_MAP_CHUNK-slop)
308 clen = MAX_MAP_CHUNK-slop;
309 mapaddr = ramdisk_image & PAGE_MASK;
310 p = early_ioremap(mapaddr, clen+slop);
311 memcpy(q, p+slop, clen);
312 early_iounmap(p, clen+slop);
314 ramdisk_image += clen;
315 ramdisk_size -= clen;
317 /* high pages is not converted by early_res_to_bootmem */
318 ramdisk_image = boot_params.hdr.ramdisk_image;
319 ramdisk_size = boot_params.hdr.ramdisk_size;
320 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
321 " %08llx - %08llx\n",
322 ramdisk_image, ramdisk_image + ramdisk_size - 1,
323 ramdisk_here, ramdisk_here + ramdisk_size - 1);
327 static void __init reserve_initrd(void)
329 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
330 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
331 u64 ramdisk_end = ramdisk_image + ramdisk_size;
332 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
334 if (!boot_params.hdr.type_of_loader ||
335 !ramdisk_image || !ramdisk_size)
336 return; /* No initrd provided by bootloader */
340 if (ramdisk_size >= (end_of_lowmem>>1)) {
341 free_early(ramdisk_image, ramdisk_end);
342 printk(KERN_ERR "initrd too large to handle, "
343 "disabling initrd\n");
347 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
351 if (ramdisk_end <= end_of_lowmem) {
352 /* All in lowmem, easy case */
354 * don't need to reserve again, already reserved early
355 * in i386_start_kernel
357 initrd_start = ramdisk_image + PAGE_OFFSET;
358 initrd_end = initrd_start + ramdisk_size;
365 printk(KERN_ERR "initrd extends beyond end of memory "
366 "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
367 ramdisk_end, end_of_lowmem);
370 free_early(ramdisk_image, ramdisk_end);
373 static void __init reserve_initrd(void)
376 #endif /* CONFIG_BLK_DEV_INITRD */
378 static void __init parse_setup_data(void)
380 struct setup_data *data;
383 if (boot_params.hdr.version < 0x0209)
385 pa_data = boot_params.hdr.setup_data;
387 data = early_ioremap(pa_data, PAGE_SIZE);
388 switch (data->type) {
390 parse_e820_ext(data, pa_data);
395 #ifndef CONFIG_DEBUG_BOOT_PARAMS
396 free_early(pa_data, pa_data+sizeof(*data)+data->len);
398 pa_data = data->next;
399 early_iounmap(data, PAGE_SIZE);
404 * --------- Crashkernel reservation ------------------------------
408 static inline unsigned long long get_total_mem(void)
410 unsigned long long total;
412 total = max_low_pfn - min_low_pfn;
413 #ifdef CONFIG_HIGHMEM
414 total += highend_pfn - highstart_pfn;
417 return total << PAGE_SHIFT;
420 static void __init reserve_crashkernel(void)
422 unsigned long long total_mem;
423 unsigned long long crash_size, crash_base;
426 total_mem = get_total_mem();
428 ret = parse_crashkernel(boot_command_line, total_mem,
429 &crash_size, &crash_base);
430 if (ret == 0 && crash_size > 0) {
431 if (crash_base <= 0) {
432 printk(KERN_INFO "crashkernel reservation failed - "
433 "you have to specify a base address\n");
437 if (reserve_bootmem_generic(crash_base, crash_size,
438 BOOTMEM_EXCLUSIVE) < 0) {
439 printk(KERN_INFO "crashkernel reservation failed - "
440 "memory is in use\n");
444 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
445 "for crashkernel (System RAM: %ldMB)\n",
446 (unsigned long)(crash_size >> 20),
447 (unsigned long)(crash_base >> 20),
448 (unsigned long)(total_mem >> 20));
450 crashk_res.start = crash_base;
451 crashk_res.end = crash_base + crash_size - 1;
452 insert_resource(&iomem_resource, &crashk_res);
456 static void __init reserve_crashkernel(void)
461 static struct resource standard_io_resources[] = {
462 { .name = "dma1", .start = 0x00, .end = 0x1f,
463 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
464 { .name = "pic1", .start = 0x20, .end = 0x21,
465 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
466 { .name = "timer0", .start = 0x40, .end = 0x43,
467 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
468 { .name = "timer1", .start = 0x50, .end = 0x53,
469 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
470 { .name = "keyboard", .start = 0x60, .end = 0x60,
471 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
472 { .name = "keyboard", .start = 0x64, .end = 0x64,
473 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
474 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
475 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
476 { .name = "pic2", .start = 0xa0, .end = 0xa1,
477 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
478 { .name = "dma2", .start = 0xc0, .end = 0xdf,
479 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
480 { .name = "fpu", .start = 0xf0, .end = 0xff,
481 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
484 static void __init reserve_standard_io_resources(void)
488 /* request I/O space for devices used on all i[345]86 PCs */
489 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
490 request_resource(&ioport_resource, &standard_io_resources[i]);
494 #ifdef CONFIG_PROC_VMCORE
495 /* elfcorehdr= specifies the location of elf core header
496 * stored by the crashed kernel. This option will be passed
497 * by kexec loader to the capture kernel.
499 static int __init setup_elfcorehdr(char *arg)
504 elfcorehdr_addr = memparse(arg, &end);
505 return end > arg ? 0 : -EINVAL;
507 early_param("elfcorehdr", setup_elfcorehdr);
511 * Determine if we were loaded by an EFI loader. If so, then we have also been
512 * passed the efi memmap, systab, etc., so we should use these data structures
513 * for initialization. Note, the efi init code path is determined by the
514 * global efi_enabled. This allows the same kernel image to be used on existing
515 * systems (with a traditional BIOS) as well as on EFI systems.
518 * setup_arch - architecture-specific boot-time initializations
520 * Note: On x86_64, fixmaps are ready for use even before this is called.
523 void __init setup_arch(char **cmdline_p)
526 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
527 pre_setup_arch_hook();
529 early_ioremap_init();
530 reserve_setup_data();
532 printk(KERN_INFO "Command line: %s\n", boot_command_line);
535 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
536 screen_info = boot_params.screen_info;
537 edid_info = boot_params.edid_info;
539 apm_info.bios = boot_params.apm_bios_info;
540 ist_info = boot_params.ist_info;
541 if (boot_params.sys_desc_table.length != 0) {
542 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
543 machine_id = boot_params.sys_desc_table.table[0];
544 machine_submodel_id = boot_params.sys_desc_table.table[1];
545 BIOS_revision = boot_params.sys_desc_table.table[2];
548 saved_video_mode = boot_params.hdr.vid_mode;
549 bootloader_type = boot_params.hdr.type_of_loader;
551 #ifdef CONFIG_BLK_DEV_RAM
552 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
553 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
554 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
557 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
574 if (!boot_params.hdr.root_flags)
575 root_mountflags &= ~MS_RDONLY;
576 init_mm.start_code = (unsigned long) _text;
577 init_mm.end_code = (unsigned long) _etext;
578 init_mm.end_data = (unsigned long) _edata;
580 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
582 init_mm.brk = (unsigned long) &_end;
585 code_resource.start = virt_to_phys(_text);
586 code_resource.end = virt_to_phys(_etext)-1;
587 data_resource.start = virt_to_phys(_etext);
588 data_resource.end = virt_to_phys(_edata)-1;
589 bss_resource.start = virt_to_phys(&__bss_start);
590 bss_resource.end = virt_to_phys(&__bss_stop)-1;
595 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
596 *cmdline_p = command_line;
602 if (acpi_mps_check()) {
603 #ifdef CONFIG_X86_LOCAL_APIC
605 enable_local_apic = -1;
610 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
613 finish_e820_parsing();
618 # ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
619 if (init_ohci1394_dma_early)
620 init_ohci1394_dma_on_all_controllers();
624 /* after parse_early_param, so could debug it */
625 insert_resource(&iomem_resource, &code_resource);
626 insert_resource(&iomem_resource, &data_resource);
627 insert_resource(&iomem_resource, &bss_resource);
633 if (ppro_with_ram_bug()) {
634 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
636 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
637 printk(KERN_INFO "fixed physical RAM map:\n");
638 e820_print_map("bad_ppro");
641 early_gart_iommu_check();
644 e820_register_active_regions(0, 0, -1UL);
646 * partially used pages are not usable - thus
647 * we are rounding upwards:
649 max_pfn = e820_end_of_ram();
651 /* preallocate 4k for mptable mpc */
652 early_reserve_e820_mpc_new();
653 /* update e820 for memory not covered by WB MTRRs */
655 if (mtrr_trim_uncached_memory(max_pfn)) {
656 remove_all_active_ranges();
657 e820_register_active_regions(0, 0, -1UL);
658 max_pfn = e820_end_of_ram();
662 /* max_low_pfn get updated here */
663 find_low_pfn_range();
665 num_physpages = max_pfn;
669 /* How many end-of-memory variables you have, grandma! */
670 /* need this before calling reserve_initrd */
671 max_low_pfn = max_pfn;
672 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
675 /* max_pfn_mapped is updated here */
676 max_pfn_mapped = init_memory_mapping(0, (max_low_pfn << PAGE_SHIFT));
689 * Parse the ACPI tables for possible boot-time SMP configuration.
691 acpi_boot_table_init();
694 /* Remove active ranges so rediscovery with NUMA-awareness happens */
695 remove_all_active_ranges();
698 #ifdef CONFIG_ACPI_NUMA
700 * Parse SRAT to discover nodes.
705 initmem_init(0, max_pfn);
708 dma32_reserve_bootmem();
711 #ifdef CONFIG_ACPI_SLEEP
713 * Reserve low memory region for sleep support.
715 acpi_reserve_bootmem();
717 #ifdef CONFIG_X86_FIND_SMP_CONFIG
719 * Find and reserve possible boot-time SMP configuration:
723 reserve_crashkernel();
725 reserve_ibft_region();
727 #ifdef CONFIG_KVM_CLOCK
731 #if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
733 * Must be after max_low_pfn is determined, and before kernel
734 * pagetables are setup.
746 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
749 #if defined(CONFIG_PROVIDE_OHCI1394_DMA_INIT) && defined(CONFIG_X86_32)
750 if (init_ohci1394_dma_early)
751 init_ohci1394_dma_on_all_controllers();
754 #ifdef CONFIG_X86_GENERICARCH
755 generic_apic_probe();
761 * Read APIC and some other early information from ACPI tables.
769 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
771 * get boot-time SMP configuration:
773 if (smp_found_config)
778 init_apic_mappings();
779 ioapic_init_mappings();
781 # if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
783 printk(KERN_WARNING "More than 8 CPUs detected and "
784 "CONFIG_X86_PC cannot handle it.\nUse "
785 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
790 e820_reserve_resources();
791 e820_mark_nosave_regions(max_low_pfn);
794 request_resource(&iomem_resource, &video_ram_resource);
796 reserve_standard_io_resources();
801 #if defined(CONFIG_VGA_CONSOLE)
802 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
803 conswitchp = &vga_con;
804 #elif defined(CONFIG_DUMMY_CONSOLE)
805 conswitchp = &dummy_con;