projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
memblock, x86: Replace memblock_x86_reserve/free_range() with generic ones
[linux-2.6-block.git] / arch / x86 / kernel / setup.c
CommitLineData
1da177e4 1/*
1da177e4
LT		 2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
05b79bdc 26#include <linux/mmzone.h>
894673ee 27#include <linux/screen_info.h>
1da177e4
LT		 28#include <linux/ioport.h>
29#include <linux/acpi.h>
efafc8b2 30#include <linux/sfi.h>
1da177e4
LT		 31#include <linux/apm_bios.h>
32#include <linux/initrd.h>
33#include <linux/bootmem.h>
72d7c3b3 34#include <linux/memblock.h>
1da177e4
LT		 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>
138fe4e0 44#include <linux/iscsi_ibft.h>
1da177e4 45#include <linux/nodemask.h>
1bc3b91a 46#include <linux/kexec.h>
e9928674 47#include <linux/dmi.h>
22a9835c 48#include <linux/pfn.h>
376ff035 49#include <linux/pci.h>
46d671b5 50#include <asm/pci-direct.h>
f212ec4b 51#include <linux/init_ohci1394_dma.h>
790c73f6 52#include <linux/kvm_para.h>
1bc3b91a 53
46d671b5
YL		 54#include <linux/errno.h>
55#include <linux/kernel.h>
56#include <linux/stddef.h>
57#include <linux/unistd.h>
58#include <linux/ptrace.h>
46d671b5
YL		 59#include <linux/user.h>
60#include <linux/delay.h>
46d671b5
YL		 61
62#include <linux/kallsyms.h>
46d671b5
YL		 63#include <linux/cpufreq.h>
64#include <linux/dma-mapping.h>
65#include <linux/ctype.h>
66#include <linux/uaccess.h>
67
68#include <linux/percpu.h>
69#include <linux/crash_dump.h>
69575d38 70#include <linux/tboot.h>
46d671b5 71
1da177e4 72#include <video/edid.h>
1bc3b91a 73
093af8d7 74#include <asm/mtrr.h>
9635b47d 75#include <asm/apic.h>
893f38d1 76#include <asm/trampoline.h>
1da177e4
LT		 77#include <asm/e820.h>
78#include <asm/mpspec.h>
79#include <asm/setup.h>
55f26239 80#include <asm/efi.h>
8e6dafd6
IM		 81#include <asm/timer.h>
82#include <asm/i8259.h>
1da177e4 83#include <asm/sections.h>
1c6e5503 84#include <asm/dmi.h>
1da177e4
LT		 85#include <asm/io_apic.h>
86#include <asm/ist.h>
1164dd00 87#include <asm/setup_arch.h>
ce3fe6b2 88#include <asm/bios_ebda.h>
00bf4098 89#include <asm/cacheflush.h>
2fde61fd 90#include <asm/processor.h>
cc9f7a0c 91#include <asm/bugs.h>
1da177e4 92
46d671b5
YL		 93#include <asm/system.h>
94#include <asm/vsyscall.h>
6e5385d4 95#include <asm/cpu.h>
46d671b5
YL		 96#include <asm/desc.h>
97#include <asm/dma.h>
46a7fa27 98#include <asm/iommu.h>
1d9b16d1 99#include <asm/gart.h>
46d671b5
YL		 100#include <asm/mmu_context.h>
101#include <asm/proto.h>
102
46d671b5 103#include <asm/paravirt.h>
88b094fb 104#include <asm/hypervisor.h>
fd699c76 105#include <asm/olpc_ofw.h>
46d671b5
YL		 106
107#include <asm/percpu.h>
46d671b5
YL		 108#include <asm/topology.h>
109#include <asm/apicdef.h>
23ac4ae8 110#include <asm/amd_nb.h>
55f26239
YL		 111#ifdef CONFIG_X86_64
112#include <asm/numa_64.h>
113#endif
a2202aa2 114#include <asm/mce.h>
f49aa448 115#include <asm/alternative.h>
da6b737b 116#include <asm/prom.h>
46d671b5 117
2b72394e
PE		 118/*
119 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
120 * The direct mapping extends to max_pfn_mapped, so that we can directly access
121 * apertures, ACPI and other tables without having to play with fixmaps.
122 */
123unsigned long max_low_pfn_mapped;
124unsigned long max_pfn_mapped;
125
e808bae2 126#ifdef CONFIG_DMI
796216a5 127RESERVE_BRK(dmi_alloc, 65536);
e808bae2 128#endif
796216a5 129
c0b5842a 130
93dbda7c
JF		 131static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
132unsigned long _brk_end = (unsigned long)__brk_base;
133
c0b5842a
IM		 134#ifdef CONFIG_X86_64
135int default_cpu_present_to_apicid(int mps_cpu)
136{
137 return __default_cpu_present_to_apicid(mps_cpu);
138}
139
e11dadab 140int default_check_phys_apicid_present(int phys_apicid)
c0b5842a 141{
e11dadab 142 return __default_check_phys_apicid_present(phys_apicid);
c0b5842a
IM		 143}
144#endif
145
217b8ce8
YL		 146#ifndef CONFIG_DEBUG_BOOT_PARAMS
147struct boot_params __initdata boot_params;
148#else
149struct boot_params boot_params;
150#endif
151
1da177e4
LT		 152/*
153 * Machine setup..
154 */
c9cce83d
BW		 155static struct resource data_resource = {
156 .name = "Kernel data",
157 .start = 0,
158 .end = 0,
159 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
160};
161
162static struct resource code_resource = {
163 .name = "Kernel code",
164 .start = 0,
165 .end = 0,
166 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
167};
168
169static struct resource bss_resource = {
170 .name = "Kernel bss",
171 .start = 0,
172 .end = 0,
173 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
174};
175
7dea23ec
YL		 176
177#ifdef CONFIG_X86_32
1da177e4 178/* cpu data as detected by the assembly code in head.S */
7dea23ec 179struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
1da177e4 180/* common cpu data for all cpus */
7dea23ec 181struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
129f6946 182EXPORT_SYMBOL(boot_cpu_data);
7dea23ec
YL		 183static void set_mca_bus(int x)
184{
185#ifdef CONFIG_MCA
186 MCA_bus = x;
187#endif
188}
1da177e4 189
0c254e38
AS		 190unsigned int def_to_bigsmp;
191
1da177e4
LT		 192/* for MCA, but anyone else can use it if they want */
193unsigned int machine_id;
194unsigned int machine_submodel_id;
195unsigned int BIOS_revision;
1da177e4 196
7dea23ec
YL		 197struct apm_info apm_info;
198EXPORT_SYMBOL(apm_info);
199
200#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
201 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
202struct ist_info ist_info;
203EXPORT_SYMBOL(ist_info);
204#else
205struct ist_info ist_info;
206#endif
207
208#else
ed26dbe5
JF		 209struct cpuinfo_x86 boot_cpu_data __read_mostly = {
210 .x86_phys_bits = MAX_PHYSMEM_BITS,
211};
7dea23ec
YL		 212EXPORT_SYMBOL(boot_cpu_data);
213#endif
214
215
216#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
217unsigned long mmu_cr4_features;
218#else
219unsigned long mmu_cr4_features = X86_CR4_PAE;
220#endif
221
5031296c
PA		 222/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
223int bootloader_type, bootloader_version;
1da177e4 224
1da177e4
LT		 225/*
226 * Setup options
227 */
1da177e4 228struct screen_info screen_info;
129f6946 229EXPORT_SYMBOL(screen_info);
1da177e4 230struct edid_info edid_info;
5e518d76 231EXPORT_SYMBOL_GPL(edid_info);
1da177e4 232
1da177e4
LT		 233extern int root_mountflags;
234
e44b7b75 235unsigned long saved_video_mode;
1da177e4 236
cf8fa920 237#define RAMDISK_IMAGE_START_MASK 0x07FF
1da177e4 238#define RAMDISK_PROMPT_FLAG 0x8000
cf8fa920 239#define RAMDISK_LOAD_FLAG 0x4000
1da177e4 240
4e498b66 241static char __initdata command_line[COMMAND_LINE_SIZE];
516cbf37
TB		 242#ifdef CONFIG_CMDLINE_BOOL
243static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
244#endif
1da177e4 245
1da177e4
LT		 246#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
247struct edd edd;
248#ifdef CONFIG_EDD_MODULE
249EXPORT_SYMBOL(edd);
250#endif
251/**
252 * copy_edd() - Copy the BIOS EDD information
253 * from boot_params into a safe place.
254 *
255 */
9eaa192d 256static inline void __init copy_edd(void)
1da177e4 257{
30c82645
PA		 258 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
259 sizeof(edd.mbr_signature));
260 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
261 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
262 edd.edd_info_nr = boot_params.eddbuf_entries;
1da177e4
LT		 263}
264#else
9eaa192d 265static inline void __init copy_edd(void)
1da177e4
LT		 266{
267}
268#endif
269
5368a2be
PA		 270void * __init extend_brk(size_t size, size_t align)
271{
272 size_t mask = align - 1;
273 void *ret;
274
275 BUG_ON(_brk_start == 0);
276 BUG_ON(align & mask);
277
278 _brk_end = (_brk_end + mask) & ~mask;
279 BUG_ON((char *)(_brk_end + size) > __brk_limit);
280
281 ret = (void *)_brk_end;
282 _brk_end += size;
283
284 memset(ret, 0, size);
285
286 return ret;
287}
288
854c879f
PE		 289#ifdef CONFIG_X86_64
290static void __init init_gbpages(void)
291{
292 if (direct_gbpages && cpu_has_gbpages)
293 printk(KERN_INFO "Using GB pages for direct mapping\n");
294 else
295 direct_gbpages = 0;
296}
297#else
298static inline void init_gbpages(void)
299{
300}
e5f15b45 301static void __init cleanup_highmap(void)
f005fe12
YL		 302{
303}
854c879f
PE		 304#endif
305
5368a2be
PA		 306static void __init reserve_brk(void)
307{
308 if (_brk_end > _brk_start)
24aa0788
TH		 309 memblock_reserve(__pa(_brk_start),
310 __pa(_brk_end) - __pa(_brk_start));
5368a2be
PA		 311
312 /* Mark brk area as locked down and no longer taking any
313 new allocations */
314 _brk_start = 0;
315}
316
cf8fa920
PA		 317#ifdef CONFIG_BLK_DEV_INITRD
318
eb1379cb
YL		 319#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
320static void __init relocate_initrd(void)
cf8fa920 321{
c967da6a 322 /* Assume only end is not page aligned */
ba5b14cc
YL		 323 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
324 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
c967da6a 325 u64 area_size = PAGE_ALIGN(ramdisk_size);
8c5dd8f4 326 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
ba5b14cc 327 u64 ramdisk_here;
eb1379cb
YL		 328 unsigned long slop, clen, mapaddr;
329 char *p, *q;
cf8fa920
PA		 330
331 /* We need to move the initrd down into lowmem */
a9ce6bc1 332 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
4e29684c 333 PAGE_SIZE);
cf8fa920 334
1f5026a7 335 if (!ramdisk_here)
3945e2c9
YL		 336 panic("Cannot find place for new RAMDISK of size %lld\n",
337 ramdisk_size);
338
cf8fa920
PA		 339 /* Note: this includes all the lowmem currently occupied by
340 the initrd, we rely on that fact to keep the data intact. */
24aa0788 341 memblock_reserve(ramdisk_here, area_size);
cf8fa920
PA		 342 initrd_start = ramdisk_here + PAGE_OFFSET;
343 initrd_end = initrd_start + ramdisk_size;
f0d43100
YL		 344 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
345 ramdisk_here, ramdisk_here + ramdisk_size);
cf8fa920 346
cf8fa920
PA		 347 q = (char *)initrd_start;
348
349 /* Copy any lowmem portion of the initrd */
350 if (ramdisk_image < end_of_lowmem) {
351 clen = end_of_lowmem - ramdisk_image;
352 p = (char *)__va(ramdisk_image);
353 memcpy(q, p, clen);
354 q += clen;
355 ramdisk_image += clen;
356 ramdisk_size -= clen;
357 }
358
359 /* Copy the highmem portion of the initrd */
360 while (ramdisk_size) {
361 slop = ramdisk_image & ~PAGE_MASK;
362 clen = ramdisk_size;
363 if (clen > MAX_MAP_CHUNK-slop)
364 clen = MAX_MAP_CHUNK-slop;
365 mapaddr = ramdisk_image & PAGE_MASK;
88b4c146 366 p = early_memremap(mapaddr, clen+slop);
cf8fa920 367 memcpy(q, p+slop, clen);
beacfaac 368 early_iounmap(p, clen+slop);
cf8fa920
PA		 369 q += clen;
370 ramdisk_image += clen;
371 ramdisk_size -= clen;
372 }
a4c81cf6 373 /* high pages is not converted by early_res_to_bootmem */
ba5b14cc
YL		 374 ramdisk_image = boot_params.hdr.ramdisk_image;
375 ramdisk_size = boot_params.hdr.ramdisk_size;
eb1379cb
YL		 376 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
377 " %08llx - %08llx\n",
ba5b14cc
YL		 378 ramdisk_image, ramdisk_image + ramdisk_size - 1,
379 ramdisk_here, ramdisk_here + ramdisk_size - 1);
eb1379cb 380}
9a27f5c5 381
eb1379cb
YL		 382static void __init reserve_initrd(void)
383{
c967da6a 384 /* Assume only end is not page aligned */
eb1379cb
YL		 385 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
386 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
c967da6a 387 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
8c5dd8f4 388 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
eb1379cb
YL		 389
390 if (!boot_params.hdr.type_of_loader ||
391 !ramdisk_image || !ramdisk_size)
392 return; /* No initrd provided by bootloader */
393
394 initrd_start = 0;
395
396 if (ramdisk_size >= (end_of_lowmem>>1)) {
24aa0788 397 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
eb1379cb
YL		 398 printk(KERN_ERR "initrd too large to handle, "
399 "disabling initrd\n");
400 return;
401 }
402
403 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
404 ramdisk_end);
405
406
407 if (ramdisk_end <= end_of_lowmem) {
408 /* All in lowmem, easy case */
409 /*
410 * don't need to reserve again, already reserved early
411 * in i386_start_kernel
412 */
413 initrd_start = ramdisk_image + PAGE_OFFSET;
414 initrd_end = initrd_start + ramdisk_size;
415 return;
416 }
417
eb1379cb 418 relocate_initrd();
8c5dd8f4 419
24aa0788 420 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
cf8fa920 421}
225c37d7 422#else
eb1379cb 423static void __init reserve_initrd(void)
225c37d7
YL		 424{
425}
cf8fa920
PA		 426#endif /* CONFIG_BLK_DEV_INITRD */
427
29f784e3 428static void __init parse_setup_data(void)
257b0fde
YL		 429{
430 struct setup_data *data;
431 u64 pa_data;
432
433 if (boot_params.hdr.version < 0x0209)
434 return;
435 pa_data = boot_params.hdr.setup_data;
436 while (pa_data) {
f1c2b357
SAS		 437 u32 data_len, map_len;
438
439 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
440 (u64)sizeof(struct setup_data));
441 data = early_memremap(pa_data, map_len);
442 data_len = data->len + sizeof(struct setup_data);
443 if (data_len > map_len) {
444 early_iounmap(data, map_len);
445 data = early_memremap(pa_data, data_len);
446 map_len = data_len;
447 }
448
257b0fde
YL		 449 switch (data->type) {
450 case SETUP_E820_EXT:
f1c2b357 451 parse_e820_ext(data);
257b0fde 452 break;
da6b737b
SAS		 453 case SETUP_DTB:
454 add_dtb(pa_data);
257b0fde
YL		 455 break;
456 default:
457 break;
458 }
257b0fde 459 pa_data = data->next;
f1c2b357 460 early_iounmap(data, map_len);
257b0fde
YL		 461 }
462}
463
a0a0becd 464static void __init e820_reserve_setup_data(void)
28bb2237
YL		 465{
466 struct setup_data *data;
467 u64 pa_data;
d9a81b44 468 int found = 0;
28bb2237
YL		 469
470 if (boot_params.hdr.version < 0x0209)
471 return;
472 pa_data = boot_params.hdr.setup_data;
473 while (pa_data) {
88b4c146 474 data = early_memremap(pa_data, sizeof(*data));
28bb2237
YL		 475 e820_update_range(pa_data, sizeof(*data)+data->len,
476 E820_RAM, E820_RESERVED_KERN);
d9a81b44 477 found = 1;
28bb2237
YL		 478 pa_data = data->next;
479 early_iounmap(data, sizeof(*data));
480 }
d9a81b44
YL		 481 if (!found)
482 return;
483
28bb2237 484 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
a0a0becd 485 memcpy(&e820_saved, &e820, sizeof(struct e820map));
28bb2237
YL		 486 printk(KERN_INFO "extended physical RAM map:\n");
487 e820_print_map("reserve setup_data");
488}
489
a9ce6bc1 490static void __init memblock_x86_reserve_range_setup_data(void)
a0a0becd
YL		 491{
492 struct setup_data *data;
493 u64 pa_data;
a0a0becd
YL		 494
495 if (boot_params.hdr.version < 0x0209)
496 return;
497 pa_data = boot_params.hdr.setup_data;
498 while (pa_data) {
88b4c146 499 data = early_memremap(pa_data, sizeof(*data));
24aa0788 500 memblock_reserve(pa_data, sizeof(*data) + data->len);
a0a0becd
YL		 501 pa_data = data->next;
502 early_iounmap(data, sizeof(*data));
503 }
504}
505
ccb4defa
YL		 506/*
507 * --------- Crashkernel reservation ------------------------------
508 */
509
510#ifdef CONFIG_KEXEC
32105f7f 511
ccb4defa
YL		 512static inline unsigned long long get_total_mem(void)
513{
514 unsigned long long total;
515
44280733 516 total = max_pfn - min_low_pfn;
ccb4defa
YL		 517
518 return total << PAGE_SHIFT;
519}
520
7f8595bf
PA		 521/*
522 * Keep the crash kernel below this limit. On 32 bits earlier kernels
523 * would limit the kernel to the low 512 MiB due to mapping restrictions.
524 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
525 * limit once kexec-tools are fixed.
526 */
527#ifdef CONFIG_X86_32
528# define CRASH_KERNEL_ADDR_MAX (512 << 20)
529#else
530# define CRASH_KERNEL_ADDR_MAX (896 << 20)
531#endif
532
29f784e3 533static void __init reserve_crashkernel(void)
ccb4defa
YL		 534{
535 unsigned long long total_mem;
536 unsigned long long crash_size, crash_base;
537 int ret;
538
539 total_mem = get_total_mem();
540
541 ret = parse_crashkernel(boot_command_line, total_mem,
542 &crash_size, &crash_base);
32105f7f
BW		 543 if (ret != 0 || crash_size <= 0)
544 return;
545
546 /* 0 means: find the address automatically */
547 if (crash_base <= 0) {
44280733
YL		 548 const unsigned long long alignment = 16<<20; /* 16M */
549
9f4c1396 550 /*
7f8595bf 551 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
9f4c1396
YL		 552 */
553 crash_base = memblock_find_in_range(alignment,
7f8595bf 554 CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
9f4c1396 555
1f5026a7 556 if (!crash_base) {
44280733 557 pr_info("crashkernel reservation failed - No suitable area found.\n");
ccb4defa
YL		 558 return;
559 }
32105f7f 560 } else {
44280733
YL		 561 unsigned long long start;
562
9f4c1396
YL		 563 start = memblock_find_in_range(crash_base,
564 crash_base + crash_size, crash_size, 1<<20);
44280733
YL		 565 if (start != crash_base) {
566 pr_info("crashkernel reservation failed - memory is in use.\n");
ccb4defa
YL		 567 return;
568 }
32105f7f 569 }
24aa0788 570 memblock_reserve(crash_base, crash_size);
ccb4defa 571
32105f7f
BW		 572 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
573 "for crashkernel (System RAM: %ldMB)\n",
574 (unsigned long)(crash_size >> 20),
575 (unsigned long)(crash_base >> 20),
576 (unsigned long)(total_mem >> 20));
ccb4defa 577
32105f7f
BW		 578 crashk_res.start = crash_base;
579 crashk_res.end = crash_base + crash_size - 1;
580 insert_resource(&iomem_resource, &crashk_res);
ccb4defa
YL		 581}
582#else
29f784e3 583static void __init reserve_crashkernel(void)
ccb4defa
YL		 584{
585}
586#endif
587
bdba0e70
YL		 588static struct resource standard_io_resources[] = {
589 { .name = "dma1", .start = 0x00, .end = 0x1f,
590 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
591 { .name = "pic1", .start = 0x20, .end = 0x21,
592 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
593 { .name = "timer0", .start = 0x40, .end = 0x43,
594 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
595 { .name = "timer1", .start = 0x50, .end = 0x53,
596 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
597 { .name = "keyboard", .start = 0x60, .end = 0x60,
598 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
599 { .name = "keyboard", .start = 0x64, .end = 0x64,
600 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
601 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
602 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
603 { .name = "pic2", .start = 0xa0, .end = 0xa1,
604 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
605 { .name = "dma2", .start = 0xc0, .end = 0xdf,
606 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
607 { .name = "fpu", .start = 0xf0, .end = 0xff,
608 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
609};
610
8fee697d 611void __init reserve_standard_io_resources(void)
bdba0e70
YL		 612{
613 int i;
614
615 /* request I/O space for devices used on all i[345]86 PCs */
616 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
617 request_resource(&ioport_resource, &standard_io_resources[i]);
618
619}
620
042be38e
YL		 621static __init void reserve_ibft_region(void)
622{
623 unsigned long addr, size = 0;
624
625 addr = find_ibft_region(&size);
626
627 if (size)
24aa0788 628 memblock_reserve(addr, size);
042be38e
YL		 629}
630
9ea77bdb 631static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
5649b7c3 632
1b5576e6
YL		 633static void __init trim_bios_range(void)
634{
635 /*
636 * A special case is the first 4Kb of memory;
637 * This is a BIOS owned area, not kernel ram, but generally
638 * not listed as such in the E820 table.
d0cd7425
PA		 639 *
640 * This typically reserves additional memory (64KiB by default)
641 * since some BIOSes are known to corrupt low memory. See the
9ea77bdb 642 * Kconfig help text for X86_RESERVE_LOW.
1b5576e6 643 */
9ea77bdb 644 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
d0cd7425
PA		 645 E820_RAM, E820_RESERVED);
646
1b5576e6
YL		 647 /*
648 * special case: Some BIOSen report the PC BIOS
649 * area (640->1Mb) as ram even though it is not.
650 * take them out.
651 */
652 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
653 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
654}
655
9ea77bdb
PA		 656static int __init parse_reservelow(char *p)
657{
658 unsigned long long size;
659
660 if (!p)
661 return -EINVAL;
662
663 size = memparse(p, &p);
664
665 if (size < 4096)
666 size = 4096;
667
668 if (size > 640*1024)
669 size = 640*1024;
670
671 reserve_low = size;
672
673 return 0;
674}
675
676early_param("reservelow", parse_reservelow);
677
1da177e4
LT		 678/*
679 * Determine if we were loaded by an EFI loader. If so, then we have also been
680 * passed the efi memmap, systab, etc., so we should use these data structures
681 * for initialization. Note, the efi init code path is determined by the
682 * global efi_enabled. This allows the same kernel image to be used on existing
683 * systems (with a traditional BIOS) as well as on EFI systems.
684 */
76934ed4
YL		 685/*
686 * setup_arch - architecture-specific boot-time initializations
687 *
688 * Note: On x86_64, fixmaps are ready for use even before this is called.
689 */
690
1da177e4
LT		 691void __init setup_arch(char **cmdline_p)
692{
76934ed4 693#ifdef CONFIG_X86_32
1da177e4 694 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
3b33553b 695 visws_early_detect();
b40827fa
BP		 696
697 /*
698 * copy kernel address range established so far and switch
699 * to the proper swapper page table
700 */
701 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
702 initial_page_table + KERNEL_PGD_BOUNDARY,
703 KERNEL_PGD_PTRS);
704
705 load_cr3(swapper_pg_dir);
706 __flush_tlb_all();
76934ed4
YL		 707#else
708 printk(KERN_INFO "Command line: %s\n", boot_command_line);
709#endif
1da177e4 710
9863c90f
AK		 711 /*
712 * If we have OLPC OFW, we might end up relocating the fixmap due to
713 * reserve_top(), so do this before touching the ioremap area.
714 */
fd699c76
AS		 715 olpc_ofw_detect();
716
29c84391 717 early_trap_init();
9e882c92 718 early_cpu_init();
1a98fd14
JF		 719 early_ioremap_init();
720
fd699c76
AS		 721 setup_olpc_ofw_pgd();
722
30c82645
PA		 723 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
724 screen_info = boot_params.screen_info;
725 edid_info = boot_params.edid_info;
76934ed4 726#ifdef CONFIG_X86_32
30c82645
PA		 727 apm_info.bios = boot_params.apm_bios_info;
728 ist_info = boot_params.ist_info;
76934ed4 729 if (boot_params.sys_desc_table.length != 0) {
30c82645
PA		 730 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
731 machine_id = boot_params.sys_desc_table.table[0];
732 machine_submodel_id = boot_params.sys_desc_table.table[1];
733 BIOS_revision = boot_params.sys_desc_table.table[2];
1da177e4 734 }
76934ed4
YL		 735#endif
736 saved_video_mode = boot_params.hdr.vid_mode;
30c82645 737 bootloader_type = boot_params.hdr.type_of_loader;
5031296c
PA		 738 if ((bootloader_type >> 4) == 0xe) {
739 bootloader_type &= 0xf;
740 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
741 }
742 bootloader_version = bootloader_type & 0xf;
743 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
1da177e4
LT		 744
745#ifdef CONFIG_BLK_DEV_RAM
30c82645
PA		 746 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
747 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
748 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
1da177e4 749#endif
7465252e
YL		 750#ifdef CONFIG_EFI
751 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
76934ed4
YL		 752#ifdef CONFIG_X86_32
753 "EL32",
754#else
755 "EL64",
756#endif
757 4)) {
7465252e 758 efi_enabled = 1;
a9ce6bc1 759 efi_memblock_x86_reserve_range();
7465252e
YL		 760 }
761#endif
762
42bbdb43 763 x86_init.oem.arch_setup();
2215e69d 764
419afdf5 765 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
0dbfafa5 766 setup_memory_map();
28bb2237 767 parse_setup_data();
a0a0becd
YL		 768 /* update the e820_saved too */
769 e820_reserve_setup_data();
28bb2237 770
1da177e4
LT		 771 copy_edd();
772
30c82645 773 if (!boot_params.hdr.root_flags)
1da177e4
LT		 774 root_mountflags &= ~MS_RDONLY;
775 init_mm.start_code = (unsigned long) _text;
776 init_mm.end_code = (unsigned long) _etext;
777 init_mm.end_data = (unsigned long) _edata;
93dbda7c 778 init_mm.brk = _brk_end;
1da177e4
LT		 779
780 code_resource.start = virt_to_phys(_text);
781 code_resource.end = virt_to_phys(_etext)-1;
782 data_resource.start = virt_to_phys(_etext);
783 data_resource.end = virt_to_phys(_edata)-1;
00bf4098
BW		 784 bss_resource.start = virt_to_phys(&__bss_start);
785 bss_resource.end = virt_to_phys(&__bss_stop)-1;
1da177e4 786
516cbf37
TB		 787#ifdef CONFIG_CMDLINE_BOOL
788#ifdef CONFIG_CMDLINE_OVERRIDE
789 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
790#else
791 if (builtin_cmdline[0]) {
792 /* append boot loader cmdline to builtin */
793 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
794 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
795 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
796 }
797#endif
798#endif
799
eda6da92
YL		 800 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
801 *cmdline_p = command_line;
802
eda6da92 803 /*
4b0f3b81
KC		 804 * x86_configure_nx() is called before parse_early_param() to detect
805 * whether hardware doesn't support NX (so that the early EHCI debug
806 * console setup can safely call set_fixmap()). It may then be called
807 * again from within noexec_setup() during parsing early parameters
808 * to honor the respective command line option.
eda6da92 809 */
4763ed4d 810 x86_configure_nx();
eda6da92
YL		 811
812 parse_early_param();
813
4b0f3b81 814 x86_report_nx();
0ad5bce7 815
28bb2237 816 /* after early param, so could get panic from serial */
a9ce6bc1 817 memblock_x86_reserve_range_setup_data();
28bb2237 818
76934ed4 819 if (acpi_mps_check()) {
3eb11edc 820#ifdef CONFIG_X86_LOCAL_APIC
76934ed4 821 disable_apic = 1;
3eb11edc 822#endif
988781dc 823 setup_clear_cpu_cap(X86_FEATURE_APIC);
3c999f14
YL		 824 }
825
dc7c65db
LT		 826#ifdef CONFIG_PCI
827 if (pci_early_dump_regs)
828 early_dump_pci_devices();
829#endif
830
0dbfafa5 831 finish_e820_parsing();
1a3f239d 832
ff0c0874
BM		 833 if (efi_enabled)
834 efi_init();
835
2216d199
YL		 836 dmi_scan_machine();
837
88b094fb
AK		 838 /*
839 * VMware detection requires dmi to be available, so this
840 * needs to be done after dmi_scan_machine, for the BP.
841 */
2d826404 842 init_hypervisor_platform();
88b094fb 843
f7cf5a5b 844 x86_init.resources.probe_roms();
41c094fd
YL		 845
846 /* after parse_early_param, so could debug it */
847 insert_resource(&iomem_resource, &code_resource);
848 insert_resource(&iomem_resource, &data_resource);
849 insert_resource(&iomem_resource, &bss_resource);
850
1b5576e6 851 trim_bios_range();
76934ed4 852#ifdef CONFIG_X86_32
cc9f7a0c
YL		 853 if (ppro_with_ram_bug()) {
854 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
855 E820_RESERVED);
856 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
857 printk(KERN_INFO "fixed physical RAM map:\n");
858 e820_print_map("bad_ppro");
859 }
76934ed4
YL		 860#else
861 early_gart_iommu_check();
862#endif
cc9f7a0c 863
7b2a0a6c
YL		 864 /*
865 * partially used pages are not usable - thus
866 * we are rounding upwards:
867 */
f361a450 868 max_pfn = e820_end_of_ram_pfn();
7b2a0a6c 869
093af8d7
YL		 870 /* update e820 for memory not covered by WB MTRRs */
871 mtrr_bp_init();
2dc807b3 872 if (mtrr_trim_uncached_memory(max_pfn))
f361a450 873 max_pfn = e820_end_of_ram_pfn();
76c32418 874
76934ed4 875#ifdef CONFIG_X86_32
4e29684c 876 /* max_low_pfn get updated here */
2ec65f8b 877 find_low_pfn_range();
76934ed4
YL		 878#else
879 num_physpages = max_pfn;
880
06cd9a7d 881 check_x2apic();
76934ed4
YL		 882
883 /* How many end-of-memory variables you have, grandma! */
884 /* need this before calling reserve_initrd */
f361a450
YL		 885 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
886 max_low_pfn = e820_end_of_low_ram_pfn();
887 else
888 max_low_pfn = max_pfn;
889
76934ed4 890 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
5394f80f
JF		 891#endif
892
893f38d1
YL		 893 /*
894 * Find and reserve possible boot-time SMP configuration:
895 */
896 find_smp_config();
897
042be38e
YL		 898 reserve_ibft_region();
899
72d7c3b3
YL		 900 /*
901 * Need to conclude brk, before memblock_x86_fill()
902 * it could use memblock_find_in_range, could overlap with
903 * brk area.
904 */
905 reserve_brk();
906
e5f15b45
YL		 907 cleanup_highmap();
908
72d7c3b3
YL		 909 memblock.current_limit = get_max_mapped();
910 memblock_x86_fill();
911
916f676f
MG		 912 /*
913 * The EFI specification says that boot service code won't be called
914 * after ExitBootServices(). This is, in fact, a lie.
915 */
916 if (efi_enabled)
917 efi_reserve_boot_services();
918
72d7c3b3
YL		 919 /* preallocate 4k for mptable mpc */
920 early_reserve_e820_mpc_new();
921
922#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
923 setup_bios_corruption_check();
924#endif
925
926 printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
927 max_pfn_mapped<<PAGE_SHIFT);
928
4822b7fc 929 setup_trampolines();
893f38d1 930
854c879f
PE		 931 init_gbpages();
932
4e29684c 933 /* max_pfn_mapped is updated here */
f361a450
YL		 934 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
935 max_pfn_mapped = max_low_pfn_mapped;
936
f8911250
TH		 937#ifdef CONFIG_X86_64
938 if (max_pfn > max_low_pfn) {
939 max_pfn_mapped = init_memory_mapping(1UL<<32,
940 max_pfn<<PAGE_SHIFT);
941 /* can we preseve max_low_pfn ?*/
942 max_low_pfn = max_pfn;
943 }
944#endif
72d7c3b3 945 memblock.current_limit = get_max_mapped();
4e29684c 946
e7b37895
YL		 947 /*
948 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
949 */
950
951#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
952 if (init_ohci1394_dma_early)
953 init_ohci1394_dma_on_all_controllers();
954#endif
162a7e75
MT		 955 /* Allocate bigger log buffer */
956 setup_log_buf(1);
e7b37895 957
2ec65f8b
YL		 958 reserve_initrd();
959
44280733
YL		 960 reserve_crashkernel();
961
76934ed4 962 vsmp_init();
76934ed4 963
1c6e5503
YL		 964 io_delay_init();
965
1c6e5503
YL		 966 /*
967 * Parse the ACPI tables for possible boot-time SMP configuration.
968 */
969 acpi_boot_table_init();
1c6e5503 970
2e42060c
JS		 971 early_acpi_boot_init();
972
d8fc3afc 973 initmem_init();
6f2a7536 974 memblock_find_dma_reserve();
91467bdf 975
790c73f6
GOC		 976#ifdef CONFIG_KVM_CLOCK
977 kvmclock_init();
978#endif
979
030cb6c0 980 x86_init.paging.pagetable_setup_start(swapper_pg_dir);
1da177e4 981 paging_init();
030cb6c0 982 x86_init.paging.pagetable_setup_done(swapper_pg_dir);
f212ec4b 983
4da9484b
PA		 984 if (boot_cpu_data.cpuid_level >= 0) {
985 /* A CPU has %cr4 if and only if it has CPUID */
986 mmu_cr4_features = read_cr4();
987 }
988
b40827fa
BP		 989#ifdef CONFIG_X86_32
990 /* sync back kernel address range */
991 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
992 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
993 KERNEL_PGD_PTRS);
994#endif
fd89a137 995
31625340
JC		 996 tboot_probe();
997
76934ed4
YL		 998#ifdef CONFIG_X86_64
999 map_vsyscall();
1000#endif
1001
1a3f239d 1002 generic_apic_probe();
1da177e4 1003
54ef3400 1004 early_quirks();
d44647b0 1005
295deae4
YL		 1006 /*
1007 * Read APIC and some other early information from ACPI tables.
1008 */
1da177e4 1009 acpi_boot_init();
efafc8b2 1010 sfi_init();
a906fdaa 1011 x86_dtb_init();
04606618 1012
295deae4
YL		 1013 /*
1014 * get boot-time SMP configuration:
1015 */
e0da3364
YL		 1016 if (smp_found_config)
1017 get_smp_config();
76934ed4 1018
329513a3 1019 prefill_possible_map();
301e6190 1020
5f4765f9 1021 init_cpu_to_node();
5f4765f9 1022
76934ed4 1023 init_apic_mappings();
23f9b267 1024 ioapic_and_gsi_init();
9d6a4d08 1025
295deae4 1026 kvm_guest_init();
1da177e4 1027
41c094fd 1028 e820_reserve_resources();
bf62f398 1029 e820_mark_nosave_regions(max_low_pfn);
1da177e4 1030
8fee697d 1031 x86_init.resources.reserve_resources();
41c094fd
YL		 1032
1033 e820_setup_gap();
1034
1da177e4
LT		 1035#ifdef CONFIG_VT
1036#if defined(CONFIG_VGA_CONSOLE)
1037 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1038 conswitchp = &vga_con;
1039#elif defined(CONFIG_DUMMY_CONSOLE)
1040 conswitchp = &dummy_con;
1041#endif
1042#endif
6f30c1ac 1043 x86_init.oem.banner();
a2202aa2 1044
6b617e22
FT		 1045 x86_init.timers.wallclock_init();
1046
a2202aa2 1047 mcheck_init();
f49aa448 1048
dc326fca 1049 arch_init_ideal_nops();
1da177e4 1050}
5649b7c3 1051
9be1b56a
IM		 1052#ifdef CONFIG_X86_32
1053
8fee697d
TG		 1054static struct resource video_ram_resource = {
1055 .name = "Video RAM area",
1056 .start = 0xa0000,
1057 .end = 0xbffff,
1058 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
9be1b56a
IM		 1059};
1060
8fee697d 1061void __init i386_reserve_resources(void)
9be1b56a 1062{
8fee697d
TG		 1063 request_resource(&iomem_resource, &video_ram_resource);
1064 reserve_standard_io_resources();
9be1b56a
IM		 1065}
1066
9be1b56a 1067#endif /* CONFIG_X86_32 */
Linux 6.x block layer and io_uring tree(s)