x86: Unify i386 and x86-64 early quirks
[linux-2.6/btrfs-unstable.git] / arch / x86 / kernel / setup_32.c
blobd357d60431f02d304414429b486c323c28d17352
1 /*
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>
25 #include <linux/mm.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>
50 #include <asm/apic.h>
51 #include <asm/e820.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>
58 #include <asm/ist.h>
59 #include <asm/io.h>
60 #include <asm/vmi.h>
61 #include <setup_arch.h>
62 #include <bios_ebda.h>
64 /* This value is set up by the early boot code to point to the value
65 immediately after the boot time page tables. It contains a *physical*
66 address, and must not be in the .bss segment! */
67 unsigned long init_pg_tables_end __initdata = ~0UL;
69 int disable_pse __devinitdata = 0;
72 * Machine setup..
74 extern struct resource code_resource;
75 extern struct resource data_resource;
77 /* cpu data as detected by the assembly code in head.S */
78 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
79 /* common cpu data for all cpus */
80 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
81 EXPORT_SYMBOL(boot_cpu_data);
83 unsigned long mmu_cr4_features;
85 /* for MCA, but anyone else can use it if they want */
86 unsigned int machine_id;
87 #ifdef CONFIG_MCA
88 EXPORT_SYMBOL(machine_id);
89 #endif
90 unsigned int machine_submodel_id;
91 unsigned int BIOS_revision;
92 unsigned int mca_pentium_flag;
94 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
95 int bootloader_type;
97 /* user-defined highmem size */
98 static unsigned int highmem_pages = -1;
101 * Setup options
103 struct screen_info screen_info;
104 EXPORT_SYMBOL(screen_info);
105 struct apm_info apm_info;
106 EXPORT_SYMBOL(apm_info);
107 struct edid_info edid_info;
108 EXPORT_SYMBOL_GPL(edid_info);
109 struct ist_info ist_info;
110 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
111 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
112 EXPORT_SYMBOL(ist_info);
113 #endif
115 extern void early_cpu_init(void);
116 extern int root_mountflags;
118 unsigned long saved_videomode;
120 #define RAMDISK_IMAGE_START_MASK 0x07FF
121 #define RAMDISK_PROMPT_FLAG 0x8000
122 #define RAMDISK_LOAD_FLAG 0x4000
124 static char __initdata command_line[COMMAND_LINE_SIZE];
126 struct boot_params __initdata boot_params;
128 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
129 struct edd edd;
130 #ifdef CONFIG_EDD_MODULE
131 EXPORT_SYMBOL(edd);
132 #endif
134 * copy_edd() - Copy the BIOS EDD information
135 * from boot_params into a safe place.
138 static inline void copy_edd(void)
140 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
141 sizeof(edd.mbr_signature));
142 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
143 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
144 edd.edd_info_nr = boot_params.eddbuf_entries;
146 #else
147 static inline void copy_edd(void)
150 #endif
152 int __initdata user_defined_memmap = 0;
155 * "mem=nopentium" disables the 4MB page tables.
156 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
157 * to <mem>, overriding the bios size.
158 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
159 * <start> to <start>+<mem>, overriding the bios size.
161 * HPA tells me bootloaders need to parse mem=, so no new
162 * option should be mem= [also see Documentation/i386/boot.txt]
164 static int __init parse_mem(char *arg)
166 if (!arg)
167 return -EINVAL;
169 if (strcmp(arg, "nopentium") == 0) {
170 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
171 disable_pse = 1;
172 } else {
173 /* If the user specifies memory size, we
174 * limit the BIOS-provided memory map to
175 * that size. exactmap can be used to specify
176 * the exact map. mem=number can be used to
177 * trim the existing memory map.
179 unsigned long long mem_size;
181 mem_size = memparse(arg, &arg);
182 limit_regions(mem_size);
183 user_defined_memmap = 1;
185 return 0;
187 early_param("mem", parse_mem);
189 #ifdef CONFIG_PROC_VMCORE
190 /* elfcorehdr= specifies the location of elf core header
191 * stored by the crashed kernel.
193 static int __init parse_elfcorehdr(char *arg)
195 if (!arg)
196 return -EINVAL;
198 elfcorehdr_addr = memparse(arg, &arg);
199 return 0;
201 early_param("elfcorehdr", parse_elfcorehdr);
202 #endif /* CONFIG_PROC_VMCORE */
205 * highmem=size forces highmem to be exactly 'size' bytes.
206 * This works even on boxes that have no highmem otherwise.
207 * This also works to reduce highmem size on bigger boxes.
209 static int __init parse_highmem(char *arg)
211 if (!arg)
212 return -EINVAL;
214 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
215 return 0;
217 early_param("highmem", parse_highmem);
220 * vmalloc=size forces the vmalloc area to be exactly 'size'
221 * bytes. This can be used to increase (or decrease) the
222 * vmalloc area - the default is 128m.
224 static int __init parse_vmalloc(char *arg)
226 if (!arg)
227 return -EINVAL;
229 __VMALLOC_RESERVE = memparse(arg, &arg);
230 return 0;
232 early_param("vmalloc", parse_vmalloc);
235 * reservetop=size reserves a hole at the top of the kernel address space which
236 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
237 * so relocating the fixmap can be done before paging initialization.
239 static int __init parse_reservetop(char *arg)
241 unsigned long address;
243 if (!arg)
244 return -EINVAL;
246 address = memparse(arg, &arg);
247 reserve_top_address(address);
248 return 0;
250 early_param("reservetop", parse_reservetop);
253 * Determine low and high memory ranges:
255 unsigned long __init find_max_low_pfn(void)
257 unsigned long max_low_pfn;
259 max_low_pfn = max_pfn;
260 if (max_low_pfn > MAXMEM_PFN) {
261 if (highmem_pages == -1)
262 highmem_pages = max_pfn - MAXMEM_PFN;
263 if (highmem_pages + MAXMEM_PFN < max_pfn)
264 max_pfn = MAXMEM_PFN + highmem_pages;
265 if (highmem_pages + MAXMEM_PFN > max_pfn) {
266 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
267 highmem_pages = 0;
269 max_low_pfn = MAXMEM_PFN;
270 #ifndef CONFIG_HIGHMEM
271 /* Maximum memory usable is what is directly addressable */
272 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
273 MAXMEM>>20);
274 if (max_pfn > MAX_NONPAE_PFN)
275 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
276 else
277 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
278 max_pfn = MAXMEM_PFN;
279 #else /* !CONFIG_HIGHMEM */
280 #ifndef CONFIG_HIGHMEM64G
281 if (max_pfn > MAX_NONPAE_PFN) {
282 max_pfn = MAX_NONPAE_PFN;
283 printk(KERN_WARNING "Warning only 4GB will be used.\n");
284 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
286 #endif /* !CONFIG_HIGHMEM64G */
287 #endif /* !CONFIG_HIGHMEM */
288 } else {
289 if (highmem_pages == -1)
290 highmem_pages = 0;
291 #ifdef CONFIG_HIGHMEM
292 if (highmem_pages >= max_pfn) {
293 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
294 highmem_pages = 0;
296 if (highmem_pages) {
297 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
298 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
299 highmem_pages = 0;
301 max_low_pfn -= highmem_pages;
303 #else
304 if (highmem_pages)
305 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
306 #endif
308 return max_low_pfn;
312 * workaround for Dell systems that neglect to reserve EBDA
314 static void __init reserve_ebda_region(void)
316 unsigned int addr;
317 addr = get_bios_ebda();
318 if (addr)
319 reserve_bootmem(addr, PAGE_SIZE);
322 #ifndef CONFIG_NEED_MULTIPLE_NODES
323 void __init setup_bootmem_allocator(void);
324 static unsigned long __init setup_memory(void)
327 * partially used pages are not usable - thus
328 * we are rounding upwards:
330 min_low_pfn = PFN_UP(init_pg_tables_end);
332 find_max_pfn();
334 max_low_pfn = find_max_low_pfn();
336 #ifdef CONFIG_HIGHMEM
337 highstart_pfn = highend_pfn = max_pfn;
338 if (max_pfn > max_low_pfn) {
339 highstart_pfn = max_low_pfn;
341 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
342 pages_to_mb(highend_pfn - highstart_pfn));
343 num_physpages = highend_pfn;
344 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
345 #else
346 num_physpages = max_low_pfn;
347 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
348 #endif
349 #ifdef CONFIG_FLATMEM
350 max_mapnr = num_physpages;
351 #endif
352 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
353 pages_to_mb(max_low_pfn));
355 setup_bootmem_allocator();
357 return max_low_pfn;
360 void __init zone_sizes_init(void)
362 unsigned long max_zone_pfns[MAX_NR_ZONES];
363 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
364 max_zone_pfns[ZONE_DMA] =
365 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
366 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
367 #ifdef CONFIG_HIGHMEM
368 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
369 add_active_range(0, 0, highend_pfn);
370 #else
371 add_active_range(0, 0, max_low_pfn);
372 #endif
374 free_area_init_nodes(max_zone_pfns);
376 #else
377 extern unsigned long __init setup_memory(void);
378 extern void zone_sizes_init(void);
379 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
381 void __init setup_bootmem_allocator(void)
383 unsigned long bootmap_size;
385 * Initialize the boot-time allocator (with low memory only):
387 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
389 register_bootmem_low_pages(max_low_pfn);
392 * Reserve the bootmem bitmap itself as well. We do this in two
393 * steps (first step was init_bootmem()) because this catches
394 * the (very unlikely) case of us accidentally initializing the
395 * bootmem allocator with an invalid RAM area.
397 reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
398 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text));
401 * reserve physical page 0 - it's a special BIOS page on many boxes,
402 * enabling clean reboots, SMP operation, laptop functions.
404 reserve_bootmem(0, PAGE_SIZE);
406 /* reserve EBDA region, it's a 4K region */
407 reserve_ebda_region();
409 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
410 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
411 unless you have no PS/2 mouse plugged in. */
412 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
413 boot_cpu_data.x86 == 6)
414 reserve_bootmem(0xa0000 - 4096, 4096);
416 #ifdef CONFIG_SMP
418 * But first pinch a few for the stack/trampoline stuff
419 * FIXME: Don't need the extra page at 4K, but need to fix
420 * trampoline before removing it. (see the GDT stuff)
422 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
423 #endif
424 #ifdef CONFIG_ACPI_SLEEP
426 * Reserve low memory region for sleep support.
428 acpi_reserve_bootmem();
429 #endif
430 #ifdef CONFIG_X86_FIND_SMP_CONFIG
432 * Find and reserve possible boot-time SMP configuration:
434 find_smp_config();
435 #endif
436 numa_kva_reserve();
437 #ifdef CONFIG_BLK_DEV_INITRD
438 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
439 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
440 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
441 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
442 unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
444 if (ramdisk_end <= end_of_lowmem) {
445 reserve_bootmem(ramdisk_image, ramdisk_size);
446 initrd_start = ramdisk_image + PAGE_OFFSET;
447 initrd_end = initrd_start+ramdisk_size;
448 } else {
449 printk(KERN_ERR "initrd extends beyond end of memory "
450 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
451 ramdisk_end, end_of_lowmem);
452 initrd_start = 0;
455 #endif
456 #ifdef CONFIG_KEXEC
457 if (crashk_res.start != crashk_res.end)
458 reserve_bootmem(crashk_res.start,
459 crashk_res.end - crashk_res.start + 1);
460 #endif
464 * The node 0 pgdat is initialized before all of these because
465 * it's needed for bootmem. node>0 pgdats have their virtual
466 * space allocated before the pagetables are in place to access
467 * them, so they can't be cleared then.
469 * This should all compile down to nothing when NUMA is off.
471 static void __init remapped_pgdat_init(void)
473 int nid;
475 for_each_online_node(nid) {
476 if (nid != 0)
477 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
481 #ifdef CONFIG_MCA
482 static void set_mca_bus(int x)
484 MCA_bus = x;
486 #else
487 static void set_mca_bus(int x) { }
488 #endif
490 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
491 char * __init __attribute__((weak)) memory_setup(void)
493 return machine_specific_memory_setup();
497 * Determine if we were loaded by an EFI loader. If so, then we have also been
498 * passed the efi memmap, systab, etc., so we should use these data structures
499 * for initialization. Note, the efi init code path is determined by the
500 * global efi_enabled. This allows the same kernel image to be used on existing
501 * systems (with a traditional BIOS) as well as on EFI systems.
503 void __init setup_arch(char **cmdline_p)
505 unsigned long max_low_pfn;
507 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
508 pre_setup_arch_hook();
509 early_cpu_init();
512 * FIXME: This isn't an official loader_type right
513 * now but does currently work with elilo.
514 * If we were configured as an EFI kernel, check to make
515 * sure that we were loaded correctly from elilo and that
516 * the system table is valid. If not, then initialize normally.
518 #ifdef CONFIG_EFI
519 if ((boot_params.hdr.type_of_loader == 0x50) &&
520 boot_params.efi_info.efi_systab)
521 efi_enabled = 1;
522 #endif
524 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
525 screen_info = boot_params.screen_info;
526 edid_info = boot_params.edid_info;
527 apm_info.bios = boot_params.apm_bios_info;
528 ist_info = boot_params.ist_info;
529 saved_videomode = boot_params.hdr.vid_mode;
530 if( boot_params.sys_desc_table.length != 0 ) {
531 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
532 machine_id = boot_params.sys_desc_table.table[0];
533 machine_submodel_id = boot_params.sys_desc_table.table[1];
534 BIOS_revision = boot_params.sys_desc_table.table[2];
536 bootloader_type = boot_params.hdr.type_of_loader;
538 #ifdef CONFIG_BLK_DEV_RAM
539 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
540 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
541 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
542 #endif
543 ARCH_SETUP
544 if (efi_enabled)
545 efi_init();
546 else {
547 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
548 print_memory_map(memory_setup());
551 copy_edd();
553 if (!boot_params.hdr.root_flags)
554 root_mountflags &= ~MS_RDONLY;
555 init_mm.start_code = (unsigned long) _text;
556 init_mm.end_code = (unsigned long) _etext;
557 init_mm.end_data = (unsigned long) _edata;
558 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
560 code_resource.start = virt_to_phys(_text);
561 code_resource.end = virt_to_phys(_etext)-1;
562 data_resource.start = virt_to_phys(_etext);
563 data_resource.end = virt_to_phys(_edata)-1;
565 parse_early_param();
567 if (user_defined_memmap) {
568 printk(KERN_INFO "user-defined physical RAM map:\n");
569 print_memory_map("user");
572 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
573 *cmdline_p = command_line;
575 max_low_pfn = setup_memory();
577 #ifdef CONFIG_VMI
579 * Must be after max_low_pfn is determined, and before kernel
580 * pagetables are setup.
582 vmi_init();
583 #endif
586 * NOTE: before this point _nobody_ is allowed to allocate
587 * any memory using the bootmem allocator. Although the
588 * alloctor is now initialised only the first 8Mb of the kernel
589 * virtual address space has been mapped. All allocations before
590 * paging_init() has completed must use the alloc_bootmem_low_pages()
591 * variant (which allocates DMA'able memory) and care must be taken
592 * not to exceed the 8Mb limit.
595 #ifdef CONFIG_SMP
596 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
597 #endif
598 paging_init();
599 remapped_pgdat_init();
600 sparse_init();
601 zone_sizes_init();
604 * NOTE: at this point the bootmem allocator is fully available.
607 paravirt_post_allocator_init();
609 dmi_scan_machine();
611 #ifdef CONFIG_X86_GENERICARCH
612 generic_apic_probe();
613 #endif
614 if (efi_enabled)
615 efi_map_memmap();
617 #ifdef CONFIG_ACPI
619 * Parse the ACPI tables for possible boot-time SMP configuration.
621 acpi_boot_table_init();
622 #endif
624 #ifdef CONFIG_PCI
625 early_quirks();
626 #endif
628 #ifdef CONFIG_ACPI
629 acpi_boot_init();
631 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
632 if (def_to_bigsmp)
633 printk(KERN_WARNING "More than 8 CPUs detected and "
634 "CONFIG_X86_PC cannot handle it.\nUse "
635 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
636 #endif
637 #endif
638 #ifdef CONFIG_X86_LOCAL_APIC
639 if (smp_found_config)
640 get_smp_config();
641 #endif
643 e820_register_memory();
644 e820_mark_nosave_regions();
646 #ifdef CONFIG_VT
647 #if defined(CONFIG_VGA_CONSOLE)
648 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
649 conswitchp = &vga_con;
650 #elif defined(CONFIG_DUMMY_CONSOLE)
651 conswitchp = &dummy_con;
652 #endif
653 #endif