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>
61 #include <linux/kallsyms.h>
62 #include <linux/cpufreq.h>
63 #include <linux/dma-mapping.h>
64 #include <linux/ctype.h>
65 #include <linux/uaccess.h>
67 #include <linux/percpu.h>
68 #include <linux/crash_dump.h>
70 #include <video/edid.h>
75 #include <asm/mpspec.h>
76 #include <asm/setup.h>
77 #include <asm/arch_hooks.h>
79 #include <asm/sections.h>
81 #include <asm/io_apic.h>
84 #include <setup_arch.h>
85 #include <asm/bios_ebda.h>
86 #include <asm/cacheflush.h>
87 #include <asm/processor.h>
90 #include <asm/system.h>
91 #include <asm/vsyscall.h>
95 #include <asm/iommu.h>
96 #include <asm/mmu_context.h>
97 #include <asm/proto.h>
99 #include <mach_apic.h>
100 #include <asm/paravirt.h>
102 #include <asm/percpu.h>
103 #include <asm/topology.h>
104 #include <asm/apicdef.h>
106 #include <asm/numa_64.h>
113 #ifndef CONFIG_DEBUG_BOOT_PARAMS
114 struct boot_params __initdata boot_params;
116 struct boot_params boot_params;
122 static struct resource data_resource = {
123 .name = "Kernel data",
126 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
129 static struct resource code_resource = {
130 .name = "Kernel code",
133 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
136 static struct resource bss_resource = {
137 .name = "Kernel bss",
140 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
145 /* This value is set up by the early boot code to point to the value
146 immediately after the boot time page tables. It contains a *physical*
147 address, and must not be in the .bss segment! */
148 unsigned long init_pg_tables_start __initdata = ~0UL;
149 unsigned long init_pg_tables_end __initdata = ~0UL;
151 static struct resource video_ram_resource = {
152 .name = "Video RAM area",
155 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
158 /* cpu data as detected by the assembly code in head.S */
159 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
160 /* common cpu data for all cpus */
161 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
162 EXPORT_SYMBOL(boot_cpu_data);
163 static void set_mca_bus(int x)
170 unsigned int def_to_bigsmp;
172 /* for MCA, but anyone else can use it if they want */
173 unsigned int machine_id;
174 unsigned int machine_submodel_id;
175 unsigned int BIOS_revision;
177 struct apm_info apm_info;
178 EXPORT_SYMBOL(apm_info);
180 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
181 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
182 struct ist_info ist_info;
183 EXPORT_SYMBOL(ist_info);
185 struct ist_info ist_info;
189 struct cpuinfo_x86 boot_cpu_data __read_mostly;
190 EXPORT_SYMBOL(boot_cpu_data);
194 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
195 unsigned long mmu_cr4_features;
197 unsigned long mmu_cr4_features = X86_CR4_PAE;
200 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
207 char dmi_alloc_data[DMI_MAX_DATA];
212 struct screen_info screen_info;
213 EXPORT_SYMBOL(screen_info);
214 struct edid_info edid_info;
215 EXPORT_SYMBOL_GPL(edid_info);
217 extern int root_mountflags;
219 unsigned long saved_video_mode;
221 #define RAMDISK_IMAGE_START_MASK 0x07FF
222 #define RAMDISK_PROMPT_FLAG 0x8000
223 #define RAMDISK_LOAD_FLAG 0x4000
225 static char __initdata command_line[COMMAND_LINE_SIZE];
227 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
229 #ifdef CONFIG_EDD_MODULE
233 * copy_edd() - Copy the BIOS EDD information
234 * from boot_params into a safe place.
237 static inline void copy_edd(void)
239 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
240 sizeof(edd.mbr_signature));
241 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
242 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
243 edd.edd_info_nr = boot_params.eddbuf_entries;
246 static inline void copy_edd(void)
251 #ifdef CONFIG_BLK_DEV_INITRD
255 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
256 static void __init relocate_initrd(void)
259 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
260 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
261 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
263 unsigned long slop, clen, mapaddr;
266 /* We need to move the initrd down into lowmem */
267 ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
270 if (ramdisk_here == -1ULL)
271 panic("Cannot find place for new RAMDISK of size %lld\n",
274 /* Note: this includes all the lowmem currently occupied by
275 the initrd, we rely on that fact to keep the data intact. */
276 reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
278 initrd_start = ramdisk_here + PAGE_OFFSET;
279 initrd_end = initrd_start + ramdisk_size;
280 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
281 ramdisk_here, ramdisk_here + ramdisk_size);
283 q = (char *)initrd_start;
285 /* Copy any lowmem portion of the initrd */
286 if (ramdisk_image < end_of_lowmem) {
287 clen = end_of_lowmem - ramdisk_image;
288 p = (char *)__va(ramdisk_image);
291 ramdisk_image += clen;
292 ramdisk_size -= clen;
295 /* Copy the highmem portion of the initrd */
296 while (ramdisk_size) {
297 slop = ramdisk_image & ~PAGE_MASK;
299 if (clen > MAX_MAP_CHUNK-slop)
300 clen = MAX_MAP_CHUNK-slop;
301 mapaddr = ramdisk_image & PAGE_MASK;
302 p = early_ioremap(mapaddr, clen+slop);
303 memcpy(q, p+slop, clen);
304 early_iounmap(p, clen+slop);
306 ramdisk_image += clen;
307 ramdisk_size -= clen;
309 /* high pages is not converted by early_res_to_bootmem */
310 ramdisk_image = boot_params.hdr.ramdisk_image;
311 ramdisk_size = boot_params.hdr.ramdisk_size;
312 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
313 " %08llx - %08llx\n",
314 ramdisk_image, ramdisk_image + ramdisk_size - 1,
315 ramdisk_here, ramdisk_here + ramdisk_size - 1);
319 static void __init reserve_initrd(void)
321 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
322 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
323 u64 ramdisk_end = ramdisk_image + ramdisk_size;
324 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
326 if (!boot_params.hdr.type_of_loader ||
327 !ramdisk_image || !ramdisk_size)
328 return; /* No initrd provided by bootloader */
332 if (ramdisk_size >= (end_of_lowmem>>1)) {
333 free_early(ramdisk_image, ramdisk_end);
334 printk(KERN_ERR "initrd too large to handle, "
335 "disabling initrd\n");
339 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
343 if (ramdisk_end <= end_of_lowmem) {
344 /* All in lowmem, easy case */
346 * don't need to reserve again, already reserved early
347 * in i386_start_kernel
349 initrd_start = ramdisk_image + PAGE_OFFSET;
350 initrd_end = initrd_start + ramdisk_size;
357 printk(KERN_ERR "initrd extends beyond end of memory "
358 "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
359 ramdisk_end, end_of_lowmem);
362 free_early(ramdisk_image, ramdisk_end);
365 static void __init reserve_initrd(void)
368 #endif /* CONFIG_BLK_DEV_INITRD */
370 static void __init parse_setup_data(void)
372 struct setup_data *data;
375 if (boot_params.hdr.version < 0x0209)
377 pa_data = boot_params.hdr.setup_data;
379 data = early_ioremap(pa_data, PAGE_SIZE);
380 switch (data->type) {
382 parse_e820_ext(data, pa_data);
387 pa_data = data->next;
388 early_iounmap(data, PAGE_SIZE);
392 static void __init e820_reserve_setup_data(void)
394 struct setup_data *data;
398 if (boot_params.hdr.version < 0x0209)
400 pa_data = boot_params.hdr.setup_data;
402 data = early_ioremap(pa_data, sizeof(*data));
403 e820_update_range(pa_data, sizeof(*data)+data->len,
404 E820_RAM, E820_RESERVED_KERN);
406 pa_data = data->next;
407 early_iounmap(data, sizeof(*data));
412 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
413 memcpy(&e820_saved, &e820, sizeof(struct e820map));
414 printk(KERN_INFO "extended physical RAM map:\n");
415 e820_print_map("reserve setup_data");
418 static void __init reserve_early_setup_data(void)
420 struct setup_data *data;
424 if (boot_params.hdr.version < 0x0209)
426 pa_data = boot_params.hdr.setup_data;
428 data = early_ioremap(pa_data, sizeof(*data));
429 sprintf(buf, "setup data %x", data->type);
430 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
431 pa_data = data->next;
432 early_iounmap(data, sizeof(*data));
437 * --------- Crashkernel reservation ------------------------------
443 * Reserve @size bytes of crashkernel memory at any suitable offset.
445 * @size: Size of the crashkernel memory to reserve.
446 * Returns the base address on success, and -1ULL on failure.
448 unsigned long long find_and_reserve_crashkernel(unsigned long long size)
450 const unsigned long long alignment = 16<<20; /* 16M */
451 unsigned long long start = 0LL;
456 start = find_e820_area(start, ULONG_MAX, size, alignment);
460 /* try to reserve it */
461 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
469 static inline unsigned long long get_total_mem(void)
471 unsigned long long total;
473 total = max_low_pfn - min_low_pfn;
474 #ifdef CONFIG_HIGHMEM
475 total += highend_pfn - highstart_pfn;
478 return total << PAGE_SHIFT;
481 static void __init reserve_crashkernel(void)
483 unsigned long long total_mem;
484 unsigned long long crash_size, crash_base;
487 total_mem = get_total_mem();
489 ret = parse_crashkernel(boot_command_line, total_mem,
490 &crash_size, &crash_base);
491 if (ret != 0 || crash_size <= 0)
494 /* 0 means: find the address automatically */
495 if (crash_base <= 0) {
496 crash_base = find_and_reserve_crashkernel(crash_size);
497 if (crash_base == -1ULL) {
498 pr_info("crashkernel reservation failed. "
499 "No suitable area found.\n");
503 ret = reserve_bootmem_generic(crash_base, crash_size,
506 pr_info("crashkernel reservation failed - "
507 "memory is in use\n");
512 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
513 "for crashkernel (System RAM: %ldMB)\n",
514 (unsigned long)(crash_size >> 20),
515 (unsigned long)(crash_base >> 20),
516 (unsigned long)(total_mem >> 20));
518 crashk_res.start = crash_base;
519 crashk_res.end = crash_base + crash_size - 1;
520 insert_resource(&iomem_resource, &crashk_res);
523 static void __init reserve_crashkernel(void)
528 static struct resource standard_io_resources[] = {
529 { .name = "dma1", .start = 0x00, .end = 0x1f,
530 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
531 { .name = "pic1", .start = 0x20, .end = 0x21,
532 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
533 { .name = "timer0", .start = 0x40, .end = 0x43,
534 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
535 { .name = "timer1", .start = 0x50, .end = 0x53,
536 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
537 { .name = "keyboard", .start = 0x60, .end = 0x60,
538 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
539 { .name = "keyboard", .start = 0x64, .end = 0x64,
540 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
541 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
542 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
543 { .name = "pic2", .start = 0xa0, .end = 0xa1,
544 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
545 { .name = "dma2", .start = 0xc0, .end = 0xdf,
546 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
547 { .name = "fpu", .start = 0xf0, .end = 0xff,
548 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
551 static void __init reserve_standard_io_resources(void)
555 /* request I/O space for devices used on all i[345]86 PCs */
556 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
557 request_resource(&ioport_resource, &standard_io_resources[i]);
561 #ifdef CONFIG_PROC_VMCORE
562 /* elfcorehdr= specifies the location of elf core header
563 * stored by the crashed kernel. This option will be passed
564 * by kexec loader to the capture kernel.
566 static int __init setup_elfcorehdr(char *arg)
571 elfcorehdr_addr = memparse(arg, &end);
572 return end > arg ? 0 : -EINVAL;
574 early_param("elfcorehdr", setup_elfcorehdr);
577 static struct x86_quirks default_x86_quirks __initdata;
579 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
582 * Determine if we were loaded by an EFI loader. If so, then we have also been
583 * passed the efi memmap, systab, etc., so we should use these data structures
584 * for initialization. Note, the efi init code path is determined by the
585 * global efi_enabled. This allows the same kernel image to be used on existing
586 * systems (with a traditional BIOS) as well as on EFI systems.
589 * setup_arch - architecture-specific boot-time initializations
591 * Note: On x86_64, fixmaps are ready for use even before this is called.
594 void __init setup_arch(char **cmdline_p)
597 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
598 visws_early_detect();
599 pre_setup_arch_hook();
602 printk(KERN_INFO "Command line: %s\n", boot_command_line);
605 early_ioremap_init();
607 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
608 screen_info = boot_params.screen_info;
609 edid_info = boot_params.edid_info;
611 apm_info.bios = boot_params.apm_bios_info;
612 ist_info = boot_params.ist_info;
613 if (boot_params.sys_desc_table.length != 0) {
614 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
615 machine_id = boot_params.sys_desc_table.table[0];
616 machine_submodel_id = boot_params.sys_desc_table.table[1];
617 BIOS_revision = boot_params.sys_desc_table.table[2];
620 saved_video_mode = boot_params.hdr.vid_mode;
621 bootloader_type = boot_params.hdr.type_of_loader;
623 #ifdef CONFIG_BLK_DEV_RAM
624 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
625 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
626 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
629 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
645 /* update the e820_saved too */
646 e820_reserve_setup_data();
650 if (!boot_params.hdr.root_flags)
651 root_mountflags &= ~MS_RDONLY;
652 init_mm.start_code = (unsigned long) _text;
653 init_mm.end_code = (unsigned long) _etext;
654 init_mm.end_data = (unsigned long) _edata;
656 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
658 init_mm.brk = (unsigned long) &_end;
661 code_resource.start = virt_to_phys(_text);
662 code_resource.end = virt_to_phys(_etext)-1;
663 data_resource.start = virt_to_phys(_etext);
664 data_resource.end = virt_to_phys(_edata)-1;
665 bss_resource.start = virt_to_phys(&__bss_start);
666 bss_resource.end = virt_to_phys(&__bss_stop)-1;
671 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
672 *cmdline_p = command_line;
676 /* after early param, so could get panic from serial */
677 reserve_early_setup_data();
679 if (acpi_mps_check()) {
680 #ifdef CONFIG_X86_LOCAL_APIC
683 setup_clear_cpu_cap(X86_FEATURE_APIC);
687 if (pci_early_dump_regs)
688 early_dump_pci_devices();
691 finish_e820_parsing();
697 /* after parse_early_param, so could debug it */
698 insert_resource(&iomem_resource, &code_resource);
699 insert_resource(&iomem_resource, &data_resource);
700 insert_resource(&iomem_resource, &bss_resource);
706 if (ppro_with_ram_bug()) {
707 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
709 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
710 printk(KERN_INFO "fixed physical RAM map:\n");
711 e820_print_map("bad_ppro");
714 early_gart_iommu_check();
718 * partially used pages are not usable - thus
719 * we are rounding upwards:
721 max_pfn = e820_end_of_ram_pfn();
723 /* preallocate 4k for mptable mpc */
724 early_reserve_e820_mpc_new();
725 /* update e820 for memory not covered by WB MTRRs */
727 if (mtrr_trim_uncached_memory(max_pfn))
728 max_pfn = e820_end_of_ram_pfn();
731 /* max_low_pfn get updated here */
732 find_low_pfn_range();
734 num_physpages = max_pfn;
738 /* How many end-of-memory variables you have, grandma! */
739 /* need this before calling reserve_initrd */
740 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
741 max_low_pfn = e820_end_of_low_ram_pfn();
743 max_low_pfn = max_pfn;
745 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
748 /* max_pfn_mapped is updated here */
749 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
750 max_pfn_mapped = max_low_pfn_mapped;
753 if (max_pfn > max_low_pfn) {
754 max_pfn_mapped = init_memory_mapping(1UL<<32,
755 max_pfn<<PAGE_SHIFT);
756 /* can we preseve max_low_pfn ?*/
757 max_low_pfn = max_pfn;
762 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
765 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
766 if (init_ohci1394_dma_early)
767 init_ohci1394_dma_on_all_controllers();
781 * Parse the ACPI tables for possible boot-time SMP configuration.
783 acpi_boot_table_init();
785 #ifdef CONFIG_ACPI_NUMA
787 * Parse SRAT to discover nodes.
792 initmem_init(0, max_pfn);
795 dma32_reserve_bootmem();
798 #ifdef CONFIG_ACPI_SLEEP
800 * Reserve low memory region for sleep support.
802 acpi_reserve_bootmem();
804 #ifdef CONFIG_X86_FIND_SMP_CONFIG
806 * Find and reserve possible boot-time SMP configuration:
810 reserve_crashkernel();
812 reserve_ibft_region();
814 #ifdef CONFIG_KVM_CLOCK
818 #if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
820 * Must be after max_low_pfn is determined, and before kernel
821 * pagetables are setup.
826 paravirt_pagetable_setup_start(swapper_pg_dir);
828 paravirt_pagetable_setup_done(swapper_pg_dir);
829 paravirt_post_allocator_init();
835 #ifdef CONFIG_X86_GENERICARCH
836 generic_apic_probe();
842 * Read APIC and some other early information from ACPI tables.
846 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
848 * get boot-time SMP configuration:
850 if (smp_found_config)
854 prefill_possible_map();
859 init_apic_mappings();
860 ioapic_init_mappings();
862 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
864 printk(KERN_WARNING "More than 8 CPUs detected and "
865 "CONFIG_X86_PC cannot handle it.\nUse "
866 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
870 e820_reserve_resources();
871 e820_mark_nosave_regions(max_low_pfn);
874 request_resource(&iomem_resource, &video_ram_resource);
876 reserve_standard_io_resources();
881 #if defined(CONFIG_VGA_CONSOLE)
882 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
883 conswitchp = &vga_con;
884 #elif defined(CONFIG_DUMMY_CONSOLE)
885 conswitchp = &dummy_con;