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 #include <asm/paravirt.h>
106 #include <asm/percpu.h>
107 #include <asm/sections.h>
108 #include <asm/topology.h>
109 #include <asm/apicdef.h>
111 #include <asm/numa_64.h>
118 #ifndef CONFIG_DEBUG_BOOT_PARAMS
119 struct boot_params __initdata boot_params;
121 struct boot_params boot_params;
127 static struct resource data_resource = {
128 .name = "Kernel data",
131 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
134 static struct resource code_resource = {
135 .name = "Kernel code",
138 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
141 static struct resource bss_resource = {
142 .name = "Kernel bss",
145 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
150 /* This value is set up by the early boot code to point to the value
151 immediately after the boot time page tables. It contains a *physical*
152 address, and must not be in the .bss segment! */
153 unsigned long init_pg_tables_start __initdata = ~0UL;
154 unsigned long init_pg_tables_end __initdata = ~0UL;
156 static struct resource video_ram_resource = {
157 .name = "Video RAM area",
160 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
163 /* cpu data as detected by the assembly code in head.S */
164 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
165 /* common cpu data for all cpus */
166 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
167 EXPORT_SYMBOL(boot_cpu_data);
168 static void set_mca_bus(int x)
175 unsigned int def_to_bigsmp;
177 /* for MCA, but anyone else can use it if they want */
178 unsigned int machine_id;
179 unsigned int machine_submodel_id;
180 unsigned int BIOS_revision;
182 struct apm_info apm_info;
183 EXPORT_SYMBOL(apm_info);
185 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
186 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
187 struct ist_info ist_info;
188 EXPORT_SYMBOL(ist_info);
190 struct ist_info ist_info;
194 struct cpuinfo_x86 boot_cpu_data __read_mostly;
195 EXPORT_SYMBOL(boot_cpu_data);
199 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
200 unsigned long mmu_cr4_features;
202 unsigned long mmu_cr4_features = X86_CR4_PAE;
205 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
212 char dmi_alloc_data[DMI_MAX_DATA];
217 struct screen_info screen_info;
218 EXPORT_SYMBOL(screen_info);
219 struct edid_info edid_info;
220 EXPORT_SYMBOL_GPL(edid_info);
222 extern int root_mountflags;
224 unsigned long saved_video_mode;
226 #define RAMDISK_IMAGE_START_MASK 0x07FF
227 #define RAMDISK_PROMPT_FLAG 0x8000
228 #define RAMDISK_LOAD_FLAG 0x4000
230 static char __initdata command_line[COMMAND_LINE_SIZE];
232 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
234 #ifdef CONFIG_EDD_MODULE
238 * copy_edd() - Copy the BIOS EDD information
239 * from boot_params into a safe place.
242 static inline void copy_edd(void)
244 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
245 sizeof(edd.mbr_signature));
246 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
247 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
248 edd.edd_info_nr = boot_params.eddbuf_entries;
251 static inline void copy_edd(void)
256 #ifdef CONFIG_BLK_DEV_INITRD
260 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
261 static void __init relocate_initrd(void)
264 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
265 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
266 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
268 unsigned long slop, clen, mapaddr;
271 /* We need to move the initrd down into lowmem */
272 ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
275 if (ramdisk_here == -1ULL)
276 panic("Cannot find place for new RAMDISK of size %lld\n",
279 /* Note: this includes all the lowmem currently occupied by
280 the initrd, we rely on that fact to keep the data intact. */
281 reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
283 initrd_start = ramdisk_here + PAGE_OFFSET;
284 initrd_end = initrd_start + ramdisk_size;
285 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
286 ramdisk_here, ramdisk_here + ramdisk_size);
288 q = (char *)initrd_start;
290 /* Copy any lowmem portion of the initrd */
291 if (ramdisk_image < end_of_lowmem) {
292 clen = end_of_lowmem - ramdisk_image;
293 p = (char *)__va(ramdisk_image);
296 ramdisk_image += clen;
297 ramdisk_size -= clen;
300 /* Copy the highmem portion of the initrd */
301 while (ramdisk_size) {
302 slop = ramdisk_image & ~PAGE_MASK;
304 if (clen > MAX_MAP_CHUNK-slop)
305 clen = MAX_MAP_CHUNK-slop;
306 mapaddr = ramdisk_image & PAGE_MASK;
307 p = early_ioremap(mapaddr, clen+slop);
308 memcpy(q, p+slop, clen);
309 early_iounmap(p, clen+slop);
311 ramdisk_image += clen;
312 ramdisk_size -= clen;
314 /* high pages is not converted by early_res_to_bootmem */
315 ramdisk_image = boot_params.hdr.ramdisk_image;
316 ramdisk_size = boot_params.hdr.ramdisk_size;
317 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
318 " %08llx - %08llx\n",
319 ramdisk_image, ramdisk_image + ramdisk_size - 1,
320 ramdisk_here, ramdisk_here + ramdisk_size - 1);
324 static void __init reserve_initrd(void)
326 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
327 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
328 u64 ramdisk_end = ramdisk_image + ramdisk_size;
329 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
331 if (!boot_params.hdr.type_of_loader ||
332 !ramdisk_image || !ramdisk_size)
333 return; /* No initrd provided by bootloader */
337 if (ramdisk_size >= (end_of_lowmem>>1)) {
338 free_early(ramdisk_image, ramdisk_end);
339 printk(KERN_ERR "initrd too large to handle, "
340 "disabling initrd\n");
344 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
348 if (ramdisk_end <= end_of_lowmem) {
349 /* All in lowmem, easy case */
351 * don't need to reserve again, already reserved early
352 * in i386_start_kernel
354 initrd_start = ramdisk_image + PAGE_OFFSET;
355 initrd_end = initrd_start + ramdisk_size;
362 printk(KERN_ERR "initrd extends beyond end of memory "
363 "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
364 ramdisk_end, end_of_lowmem);
367 free_early(ramdisk_image, ramdisk_end);
370 static void __init reserve_initrd(void)
373 #endif /* CONFIG_BLK_DEV_INITRD */
375 static void __init parse_setup_data(void)
377 struct setup_data *data;
380 if (boot_params.hdr.version < 0x0209)
382 pa_data = boot_params.hdr.setup_data;
384 data = early_ioremap(pa_data, PAGE_SIZE);
385 switch (data->type) {
387 parse_e820_ext(data, pa_data);
392 pa_data = data->next;
393 early_iounmap(data, PAGE_SIZE);
397 static void __init e820_reserve_setup_data(void)
399 struct setup_data *data;
403 if (boot_params.hdr.version < 0x0209)
405 pa_data = boot_params.hdr.setup_data;
407 data = early_ioremap(pa_data, sizeof(*data));
408 e820_update_range(pa_data, sizeof(*data)+data->len,
409 E820_RAM, E820_RESERVED_KERN);
411 pa_data = data->next;
412 early_iounmap(data, sizeof(*data));
417 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
418 memcpy(&e820_saved, &e820, sizeof(struct e820map));
419 printk(KERN_INFO "extended physical RAM map:\n");
420 e820_print_map("reserve setup_data");
423 static void __init reserve_early_setup_data(void)
425 struct setup_data *data;
429 if (boot_params.hdr.version < 0x0209)
431 pa_data = boot_params.hdr.setup_data;
433 data = early_ioremap(pa_data, sizeof(*data));
434 sprintf(buf, "setup data %x", data->type);
435 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
436 pa_data = data->next;
437 early_iounmap(data, sizeof(*data));
442 * --------- Crashkernel reservation ------------------------------
448 * Reserve @size bytes of crashkernel memory at any suitable offset.
450 * @size: Size of the crashkernel memory to reserve.
451 * Returns the base address on success, and -1ULL on failure.
453 unsigned long long find_and_reserve_crashkernel(unsigned long long size)
455 const unsigned long long alignment = 16<<20; /* 16M */
456 unsigned long long start = 0LL;
461 start = find_e820_area(start, ULONG_MAX, size, alignment);
465 /* try to reserve it */
466 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
474 static inline unsigned long long get_total_mem(void)
476 unsigned long long total;
478 total = max_low_pfn - min_low_pfn;
479 #ifdef CONFIG_HIGHMEM
480 total += highend_pfn - highstart_pfn;
483 return total << PAGE_SHIFT;
486 static void __init reserve_crashkernel(void)
488 unsigned long long total_mem;
489 unsigned long long crash_size, crash_base;
492 total_mem = get_total_mem();
494 ret = parse_crashkernel(boot_command_line, total_mem,
495 &crash_size, &crash_base);
496 if (ret != 0 || crash_size <= 0)
499 /* 0 means: find the address automatically */
500 if (crash_base <= 0) {
501 crash_base = find_and_reserve_crashkernel(crash_size);
502 if (crash_base == -1ULL) {
503 pr_info("crashkernel reservation failed. "
504 "No suitable area found.\n");
508 ret = reserve_bootmem_generic(crash_base, crash_size,
511 pr_info("crashkernel reservation failed - "
512 "memory is in use\n");
517 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
518 "for crashkernel (System RAM: %ldMB)\n",
519 (unsigned long)(crash_size >> 20),
520 (unsigned long)(crash_base >> 20),
521 (unsigned long)(total_mem >> 20));
523 crashk_res.start = crash_base;
524 crashk_res.end = crash_base + crash_size - 1;
525 insert_resource(&iomem_resource, &crashk_res);
528 static void __init reserve_crashkernel(void)
533 static struct resource standard_io_resources[] = {
534 { .name = "dma1", .start = 0x00, .end = 0x1f,
535 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
536 { .name = "pic1", .start = 0x20, .end = 0x21,
537 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
538 { .name = "timer0", .start = 0x40, .end = 0x43,
539 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
540 { .name = "timer1", .start = 0x50, .end = 0x53,
541 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
542 { .name = "keyboard", .start = 0x60, .end = 0x60,
543 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
544 { .name = "keyboard", .start = 0x64, .end = 0x64,
545 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
546 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
547 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
548 { .name = "pic2", .start = 0xa0, .end = 0xa1,
549 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
550 { .name = "dma2", .start = 0xc0, .end = 0xdf,
551 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
552 { .name = "fpu", .start = 0xf0, .end = 0xff,
553 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
556 static void __init reserve_standard_io_resources(void)
560 /* request I/O space for devices used on all i[345]86 PCs */
561 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
562 request_resource(&ioport_resource, &standard_io_resources[i]);
566 #ifdef CONFIG_PROC_VMCORE
567 /* elfcorehdr= specifies the location of elf core header
568 * stored by the crashed kernel. This option will be passed
569 * by kexec loader to the capture kernel.
571 static int __init setup_elfcorehdr(char *arg)
576 elfcorehdr_addr = memparse(arg, &end);
577 return end > arg ? 0 : -EINVAL;
579 early_param("elfcorehdr", setup_elfcorehdr);
583 * Determine if we were loaded by an EFI loader. If so, then we have also been
584 * passed the efi memmap, systab, etc., so we should use these data structures
585 * for initialization. Note, the efi init code path is determined by the
586 * global efi_enabled. This allows the same kernel image to be used on existing
587 * systems (with a traditional BIOS) as well as on EFI systems.
590 * setup_arch - architecture-specific boot-time initializations
592 * Note: On x86_64, fixmaps are ready for use even before this is called.
595 void __init setup_arch(char **cmdline_p)
598 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
599 visws_early_detect();
600 pre_setup_arch_hook();
603 printk(KERN_INFO "Command line: %s\n", boot_command_line);
606 early_ioremap_init();
608 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
609 screen_info = boot_params.screen_info;
610 edid_info = boot_params.edid_info;
612 apm_info.bios = boot_params.apm_bios_info;
613 ist_info = boot_params.ist_info;
614 if (boot_params.sys_desc_table.length != 0) {
615 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
616 machine_id = boot_params.sys_desc_table.table[0];
617 machine_submodel_id = boot_params.sys_desc_table.table[1];
618 BIOS_revision = boot_params.sys_desc_table.table[2];
621 saved_video_mode = boot_params.hdr.vid_mode;
622 bootloader_type = boot_params.hdr.type_of_loader;
624 #ifdef CONFIG_BLK_DEV_RAM
625 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
626 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
627 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
630 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
646 /* update the e820_saved too */
647 e820_reserve_setup_data();
651 if (!boot_params.hdr.root_flags)
652 root_mountflags &= ~MS_RDONLY;
653 init_mm.start_code = (unsigned long) _text;
654 init_mm.end_code = (unsigned long) _etext;
655 init_mm.end_data = (unsigned long) _edata;
657 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
659 init_mm.brk = (unsigned long) &_end;
662 code_resource.start = virt_to_phys(_text);
663 code_resource.end = virt_to_phys(_etext)-1;
664 data_resource.start = virt_to_phys(_etext);
665 data_resource.end = virt_to_phys(_edata)-1;
666 bss_resource.start = virt_to_phys(&__bss_start);
667 bss_resource.end = virt_to_phys(&__bss_stop)-1;
672 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
673 *cmdline_p = command_line;
677 /* after early param, so could get panic from serial */
678 reserve_early_setup_data();
680 if (acpi_mps_check()) {
681 #ifdef CONFIG_X86_LOCAL_APIC
684 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
687 finish_e820_parsing();
693 /* after parse_early_param, so could debug it */
694 insert_resource(&iomem_resource, &code_resource);
695 insert_resource(&iomem_resource, &data_resource);
696 insert_resource(&iomem_resource, &bss_resource);
702 if (ppro_with_ram_bug()) {
703 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
705 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
706 printk(KERN_INFO "fixed physical RAM map:\n");
707 e820_print_map("bad_ppro");
710 early_gart_iommu_check();
714 * partially used pages are not usable - thus
715 * we are rounding upwards:
717 max_pfn = e820_end_of_ram_pfn();
719 /* preallocate 4k for mptable mpc */
720 early_reserve_e820_mpc_new();
721 /* update e820 for memory not covered by WB MTRRs */
723 if (mtrr_trim_uncached_memory(max_pfn))
724 max_pfn = e820_end_of_ram_pfn();
727 /* max_low_pfn get updated here */
728 find_low_pfn_range();
730 num_physpages = max_pfn;
734 /* How many end-of-memory variables you have, grandma! */
735 /* need this before calling reserve_initrd */
736 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
737 max_low_pfn = e820_end_of_low_ram_pfn();
739 max_low_pfn = max_pfn;
741 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
744 /* max_pfn_mapped is updated here */
745 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
746 max_pfn_mapped = max_low_pfn_mapped;
749 if (max_pfn > max_low_pfn) {
750 max_pfn_mapped = init_memory_mapping(1UL<<32,
751 max_pfn<<PAGE_SHIFT);
752 /* can we preseve max_low_pfn ?*/
753 max_low_pfn = max_pfn;
758 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
761 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
762 if (init_ohci1394_dma_early)
763 init_ohci1394_dma_on_all_controllers();
777 * Parse the ACPI tables for possible boot-time SMP configuration.
779 acpi_boot_table_init();
781 #ifdef CONFIG_ACPI_NUMA
783 * Parse SRAT to discover nodes.
788 initmem_init(0, max_pfn);
791 dma32_reserve_bootmem();
794 #ifdef CONFIG_ACPI_SLEEP
796 * Reserve low memory region for sleep support.
798 acpi_reserve_bootmem();
800 #ifdef CONFIG_X86_FIND_SMP_CONFIG
802 * Find and reserve possible boot-time SMP configuration:
806 reserve_crashkernel();
808 reserve_ibft_region();
810 #ifdef CONFIG_KVM_CLOCK
814 #if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
816 * Must be after max_low_pfn is determined, and before kernel
817 * pagetables are setup.
828 #ifdef CONFIG_X86_GENERICARCH
829 generic_apic_probe();
835 * Read APIC and some other early information from ACPI tables.
839 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
841 * get boot-time SMP configuration:
843 if (smp_found_config)
847 prefill_possible_map();
852 #ifdef CONFIG_X86_NUMAQ
854 * need to check online nodes num, call it
855 * here before time_init/tsc_init
860 init_apic_mappings();
861 ioapic_init_mappings();
863 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
865 printk(KERN_WARNING "More than 8 CPUs detected and "
866 "CONFIG_X86_PC cannot handle it.\nUse "
867 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
871 e820_reserve_resources();
872 e820_mark_nosave_regions(max_low_pfn);
875 request_resource(&iomem_resource, &video_ram_resource);
877 reserve_standard_io_resources();
882 #if defined(CONFIG_VGA_CONSOLE)
883 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
884 conswitchp = &vga_con;
885 #elif defined(CONFIG_DUMMY_CONSOLE)
886 conswitchp = &dummy_con;