blob: dc39ca6a7eca82f546c649cb6b2e1b3c6a20b4a2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/arch/i386/kernel/setup.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 *
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
7 *
8 * Memory region support
9 * David Parsons <orc@pell.chi.il.us>, July-August 1999
10 *
11 * Added E820 sanitization routine (removes overlapping memory regions);
12 * Brian Moyle <bmoyle@mvista.com>, February 2001
13 *
14 * Moved CPU detection code to cpu/${cpu}.c
15 * Patrick Mochel <mochel@osdl.org>, March 2002
16 *
17 * Provisions for empty E820 memory regions (reported by certain BIOSes).
18 * Alex Achenbach <xela@slit.de>, December 2002.
19 *
20 */
21
22/*
23 * This file handles the architecture-dependent parts of initialization
24 */
25
Alexey Dobriyan129f6942005-06-23 00:08:33 -070026#include <linux/config.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/sched.h>
28#include <linux/mm.h>
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070029#include <linux/mmzone.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/tty.h>
31#include <linux/ioport.h>
32#include <linux/acpi.h>
33#include <linux/apm_bios.h>
34#include <linux/initrd.h>
35#include <linux/bootmem.h>
36#include <linux/seq_file.h>
37#include <linux/console.h>
38#include <linux/mca.h>
39#include <linux/root_dev.h>
40#include <linux/highmem.h>
41#include <linux/module.h>
42#include <linux/efi.h>
43#include <linux/init.h>
44#include <linux/edd.h>
45#include <linux/nodemask.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070046#include <linux/kexec.h>
Vivek Goyal2030eae2005-06-25 14:58:20 -070047#include <linux/crash_dump.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070048
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <video/edid.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070050
Eric W. Biederman9635b472005-06-25 14:57:41 -070051#include <asm/apic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <asm/e820.h>
53#include <asm/mpspec.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 "setup_arch_pre.h"
61#include <bios_ebda.h>
62
Vivek Goyal92aa63a2005-06-25 14:58:18 -070063/* Forward Declaration. */
64void __init find_max_pfn(void);
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066/* This value is set up by the early boot code to point to the value
67 immediately after the boot time page tables. It contains a *physical*
68 address, and must not be in the .bss segment! */
69unsigned long init_pg_tables_end __initdata = ~0UL;
70
Li Shaohua0bb31842005-06-25 14:54:55 -070071int disable_pse __devinitdata = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73/*
74 * Machine setup..
75 */
76
77#ifdef CONFIG_EFI
78int efi_enabled = 0;
79EXPORT_SYMBOL(efi_enabled);
80#endif
81
82/* cpu data as detected by the assembly code in head.S */
83struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
84/* common cpu data for all cpus */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070085struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
Alexey Dobriyan129f6942005-06-23 00:08:33 -070086EXPORT_SYMBOL(boot_cpu_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88unsigned long mmu_cr4_features;
89
Len Brown84663612005-08-24 12:09:07 -040090#ifdef CONFIG_ACPI
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 int acpi_disabled = 0;
92#else
93 int acpi_disabled = 1;
94#endif
95EXPORT_SYMBOL(acpi_disabled);
96
Len Brown888ba6c2005-08-24 12:07:20 -040097#ifdef CONFIG_ACPI
Linus Torvalds1da177e2005-04-16 15:20:36 -070098int __initdata acpi_force = 0;
99extern acpi_interrupt_flags acpi_sci_flags;
100#endif
101
102/* for MCA, but anyone else can use it if they want */
103unsigned int machine_id;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700104#ifdef CONFIG_MCA
105EXPORT_SYMBOL(machine_id);
106#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107unsigned int machine_submodel_id;
108unsigned int BIOS_revision;
109unsigned int mca_pentium_flag;
110
111/* For PCI or other memory-mapped resources */
112unsigned long pci_mem_start = 0x10000000;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700113#ifdef CONFIG_PCI
114EXPORT_SYMBOL(pci_mem_start);
115#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117/* Boot loader ID as an integer, for the benefit of proc_dointvec */
118int bootloader_type;
119
120/* user-defined highmem size */
121static unsigned int highmem_pages = -1;
122
123/*
124 * Setup options
125 */
126struct drive_info_struct { char dummy[32]; } drive_info;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700127#if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
128 defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
129EXPORT_SYMBOL(drive_info);
130#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131struct screen_info screen_info;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700132#ifdef CONFIG_VT
133EXPORT_SYMBOL(screen_info);
134#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135struct apm_info apm_info;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700136EXPORT_SYMBOL(apm_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137struct sys_desc_table_struct {
138 unsigned short length;
139 unsigned char table[0];
140};
141struct edid_info edid_info;
Antonino A. Daplas5e518d72005-09-09 13:04:34 -0700142EXPORT_SYMBOL_GPL(edid_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143struct ist_info ist_info;
Alexey Dobriyan129f6942005-06-23 00:08:33 -0700144#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
145 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
146EXPORT_SYMBOL(ist_info);
147#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148struct e820map e820;
149
150extern void early_cpu_init(void);
151extern void dmi_scan_machine(void);
152extern void generic_apic_probe(char *);
153extern int root_mountflags;
154
155unsigned long saved_videomode;
156
157#define RAMDISK_IMAGE_START_MASK 0x07FF
158#define RAMDISK_PROMPT_FLAG 0x8000
159#define RAMDISK_LOAD_FLAG 0x4000
160
161static char command_line[COMMAND_LINE_SIZE];
162
163unsigned char __initdata boot_params[PARAM_SIZE];
164
165static struct resource data_resource = {
166 .name = "Kernel data",
167 .start = 0,
168 .end = 0,
169 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
170};
171
172static struct resource code_resource = {
173 .name = "Kernel code",
174 .start = 0,
175 .end = 0,
176 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
177};
178
179static struct resource system_rom_resource = {
180 .name = "System ROM",
181 .start = 0xf0000,
182 .end = 0xfffff,
183 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
184};
185
186static struct resource extension_rom_resource = {
187 .name = "Extension ROM",
188 .start = 0xe0000,
189 .end = 0xeffff,
190 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
191};
192
193static struct resource adapter_rom_resources[] = { {
194 .name = "Adapter ROM",
195 .start = 0xc8000,
196 .end = 0,
197 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
198}, {
199 .name = "Adapter ROM",
200 .start = 0,
201 .end = 0,
202 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
203}, {
204 .name = "Adapter ROM",
205 .start = 0,
206 .end = 0,
207 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
208}, {
209 .name = "Adapter ROM",
210 .start = 0,
211 .end = 0,
212 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
213}, {
214 .name = "Adapter ROM",
215 .start = 0,
216 .end = 0,
217 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
218}, {
219 .name = "Adapter ROM",
220 .start = 0,
221 .end = 0,
222 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
223} };
224
225#define ADAPTER_ROM_RESOURCES \
226 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
227
228static struct resource video_rom_resource = {
229 .name = "Video ROM",
230 .start = 0xc0000,
231 .end = 0xc7fff,
232 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
233};
234
235static struct resource video_ram_resource = {
236 .name = "Video RAM area",
237 .start = 0xa0000,
238 .end = 0xbffff,
239 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
240};
241
242static struct resource standard_io_resources[] = { {
243 .name = "dma1",
244 .start = 0x0000,
245 .end = 0x001f,
246 .flags = IORESOURCE_BUSY | IORESOURCE_IO
247}, {
248 .name = "pic1",
249 .start = 0x0020,
250 .end = 0x0021,
251 .flags = IORESOURCE_BUSY | IORESOURCE_IO
252}, {
253 .name = "timer0",
254 .start = 0x0040,
255 .end = 0x0043,
256 .flags = IORESOURCE_BUSY | IORESOURCE_IO
257}, {
258 .name = "timer1",
259 .start = 0x0050,
260 .end = 0x0053,
261 .flags = IORESOURCE_BUSY | IORESOURCE_IO
262}, {
263 .name = "keyboard",
264 .start = 0x0060,
265 .end = 0x006f,
266 .flags = IORESOURCE_BUSY | IORESOURCE_IO
267}, {
268 .name = "dma page reg",
269 .start = 0x0080,
270 .end = 0x008f,
271 .flags = IORESOURCE_BUSY | IORESOURCE_IO
272}, {
273 .name = "pic2",
274 .start = 0x00a0,
275 .end = 0x00a1,
276 .flags = IORESOURCE_BUSY | IORESOURCE_IO
277}, {
278 .name = "dma2",
279 .start = 0x00c0,
280 .end = 0x00df,
281 .flags = IORESOURCE_BUSY | IORESOURCE_IO
282}, {
283 .name = "fpu",
284 .start = 0x00f0,
285 .end = 0x00ff,
286 .flags = IORESOURCE_BUSY | IORESOURCE_IO
287} };
288
289#define STANDARD_IO_RESOURCES \
290 (sizeof standard_io_resources / sizeof standard_io_resources[0])
291
292#define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
293
294static int __init romchecksum(unsigned char *rom, unsigned long length)
295{
296 unsigned char *p, sum = 0;
297
298 for (p = rom; p < rom + length; p++)
299 sum += *p;
300 return sum == 0;
301}
302
303static void __init probe_roms(void)
304{
305 unsigned long start, length, upper;
306 unsigned char *rom;
307 int i;
308
309 /* video rom */
310 upper = adapter_rom_resources[0].start;
311 for (start = video_rom_resource.start; start < upper; start += 2048) {
312 rom = isa_bus_to_virt(start);
313 if (!romsignature(rom))
314 continue;
315
316 video_rom_resource.start = start;
317
318 /* 0 < length <= 0x7f * 512, historically */
319 length = rom[2] * 512;
320
321 /* if checksum okay, trust length byte */
322 if (length && romchecksum(rom, length))
323 video_rom_resource.end = start + length - 1;
324
325 request_resource(&iomem_resource, &video_rom_resource);
326 break;
327 }
328
329 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
330 if (start < upper)
331 start = upper;
332
333 /* system rom */
334 request_resource(&iomem_resource, &system_rom_resource);
335 upper = system_rom_resource.start;
336
337 /* check for extension rom (ignore length byte!) */
338 rom = isa_bus_to_virt(extension_rom_resource.start);
339 if (romsignature(rom)) {
340 length = extension_rom_resource.end - extension_rom_resource.start + 1;
341 if (romchecksum(rom, length)) {
342 request_resource(&iomem_resource, &extension_rom_resource);
343 upper = extension_rom_resource.start;
344 }
345 }
346
347 /* check for adapter roms on 2k boundaries */
348 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
349 rom = isa_bus_to_virt(start);
350 if (!romsignature(rom))
351 continue;
352
353 /* 0 < length <= 0x7f * 512, historically */
354 length = rom[2] * 512;
355
356 /* but accept any length that fits if checksum okay */
357 if (!length || start + length > upper || !romchecksum(rom, length))
358 continue;
359
360 adapter_rom_resources[i].start = start;
361 adapter_rom_resources[i].end = start + length - 1;
362 request_resource(&iomem_resource, &adapter_rom_resources[i]);
363
364 start = adapter_rom_resources[i++].end & ~2047UL;
365 }
366}
367
368static void __init limit_regions(unsigned long long size)
369{
370 unsigned long long current_addr = 0;
371 int i;
372
373 if (efi_enabled) {
Matt Tolentino7ae65fd2005-09-03 15:56:27 -0700374 efi_memory_desc_t *md;
375 void *p;
376
377 for (p = memmap.map, i = 0; p < memmap.map_end;
378 p += memmap.desc_size, i++) {
379 md = p;
380 current_addr = md->phys_addr + (md->num_pages << 12);
381 if (md->type == EFI_CONVENTIONAL_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (current_addr >= size) {
Matt Tolentino7ae65fd2005-09-03 15:56:27 -0700383 md->num_pages -=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
385 memmap.nr_map = i + 1;
386 return;
387 }
388 }
389 }
390 }
391 for (i = 0; i < e820.nr_map; i++) {
392 if (e820.map[i].type == E820_RAM) {
393 current_addr = e820.map[i].addr + e820.map[i].size;
394 if (current_addr >= size) {
395 e820.map[i].size -= current_addr-size;
396 e820.nr_map = i + 1;
397 return;
398 }
399 }
400 }
401}
402
403static void __init add_memory_region(unsigned long long start,
404 unsigned long long size, int type)
405{
406 int x;
407
408 if (!efi_enabled) {
409 x = e820.nr_map;
410
411 if (x == E820MAX) {
412 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
413 return;
414 }
415
416 e820.map[x].addr = start;
417 e820.map[x].size = size;
418 e820.map[x].type = type;
419 e820.nr_map++;
420 }
421} /* add_memory_region */
422
423#define E820_DEBUG 1
424
425static void __init print_memory_map(char *who)
426{
427 int i;
428
429 for (i = 0; i < e820.nr_map; i++) {
430 printk(" %s: %016Lx - %016Lx ", who,
431 e820.map[i].addr,
432 e820.map[i].addr + e820.map[i].size);
433 switch (e820.map[i].type) {
434 case E820_RAM: printk("(usable)\n");
435 break;
436 case E820_RESERVED:
437 printk("(reserved)\n");
438 break;
439 case E820_ACPI:
440 printk("(ACPI data)\n");
441 break;
442 case E820_NVS:
443 printk("(ACPI NVS)\n");
444 break;
445 default: printk("type %lu\n", e820.map[i].type);
446 break;
447 }
448 }
449}
450
451/*
452 * Sanitize the BIOS e820 map.
453 *
454 * Some e820 responses include overlapping entries. The following
455 * replaces the original e820 map with a new one, removing overlaps.
456 *
457 */
458struct change_member {
459 struct e820entry *pbios; /* pointer to original bios entry */
460 unsigned long long addr; /* address for this change point */
461};
462static struct change_member change_point_list[2*E820MAX] __initdata;
463static struct change_member *change_point[2*E820MAX] __initdata;
464static struct e820entry *overlap_list[E820MAX] __initdata;
465static struct e820entry new_bios[E820MAX] __initdata;
466
467static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
468{
469 struct change_member *change_tmp;
470 unsigned long current_type, last_type;
471 unsigned long long last_addr;
472 int chgidx, still_changing;
473 int overlap_entries;
474 int new_bios_entry;
475 int old_nr, new_nr, chg_nr;
476 int i;
477
478 /*
479 Visually we're performing the following (1,2,3,4 = memory types)...
480
481 Sample memory map (w/overlaps):
482 ____22__________________
483 ______________________4_
484 ____1111________________
485 _44_____________________
486 11111111________________
487 ____________________33__
488 ___________44___________
489 __________33333_________
490 ______________22________
491 ___________________2222_
492 _________111111111______
493 _____________________11_
494 _________________4______
495
496 Sanitized equivalent (no overlap):
497 1_______________________
498 _44_____________________
499 ___1____________________
500 ____22__________________
501 ______11________________
502 _________1______________
503 __________3_____________
504 ___________44___________
505 _____________33_________
506 _______________2________
507 ________________1_______
508 _________________4______
509 ___________________2____
510 ____________________33__
511 ______________________4_
512 */
513
514 /* if there's only one memory region, don't bother */
515 if (*pnr_map < 2)
516 return -1;
517
518 old_nr = *pnr_map;
519
520 /* bail out if we find any unreasonable addresses in bios map */
521 for (i=0; i<old_nr; i++)
522 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
523 return -1;
524
525 /* create pointers for initial change-point information (for sorting) */
526 for (i=0; i < 2*old_nr; i++)
527 change_point[i] = &change_point_list[i];
528
529 /* record all known change-points (starting and ending addresses),
530 omitting those that are for empty memory regions */
531 chgidx = 0;
532 for (i=0; i < old_nr; i++) {
533 if (biosmap[i].size != 0) {
534 change_point[chgidx]->addr = biosmap[i].addr;
535 change_point[chgidx++]->pbios = &biosmap[i];
536 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
537 change_point[chgidx++]->pbios = &biosmap[i];
538 }
539 }
540 chg_nr = chgidx; /* true number of change-points */
541
542 /* sort change-point list by memory addresses (low -> high) */
543 still_changing = 1;
544 while (still_changing) {
545 still_changing = 0;
546 for (i=1; i < chg_nr; i++) {
547 /* if <current_addr> > <last_addr>, swap */
548 /* or, if current=<start_addr> & last=<end_addr>, swap */
549 if ((change_point[i]->addr < change_point[i-1]->addr) ||
550 ((change_point[i]->addr == change_point[i-1]->addr) &&
551 (change_point[i]->addr == change_point[i]->pbios->addr) &&
552 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
553 )
554 {
555 change_tmp = change_point[i];
556 change_point[i] = change_point[i-1];
557 change_point[i-1] = change_tmp;
558 still_changing=1;
559 }
560 }
561 }
562
563 /* create a new bios memory map, removing overlaps */
564 overlap_entries=0; /* number of entries in the overlap table */
565 new_bios_entry=0; /* index for creating new bios map entries */
566 last_type = 0; /* start with undefined memory type */
567 last_addr = 0; /* start with 0 as last starting address */
568 /* loop through change-points, determining affect on the new bios map */
569 for (chgidx=0; chgidx < chg_nr; chgidx++)
570 {
571 /* keep track of all overlapping bios entries */
572 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
573 {
574 /* add map entry to overlap list (> 1 entry implies an overlap) */
575 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
576 }
577 else
578 {
579 /* remove entry from list (order independent, so swap with last) */
580 for (i=0; i<overlap_entries; i++)
581 {
582 if (overlap_list[i] == change_point[chgidx]->pbios)
583 overlap_list[i] = overlap_list[overlap_entries-1];
584 }
585 overlap_entries--;
586 }
587 /* if there are overlapping entries, decide which "type" to use */
588 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
589 current_type = 0;
590 for (i=0; i<overlap_entries; i++)
591 if (overlap_list[i]->type > current_type)
592 current_type = overlap_list[i]->type;
593 /* continue building up new bios map based on this information */
594 if (current_type != last_type) {
595 if (last_type != 0) {
596 new_bios[new_bios_entry].size =
597 change_point[chgidx]->addr - last_addr;
598 /* move forward only if the new size was non-zero */
599 if (new_bios[new_bios_entry].size != 0)
600 if (++new_bios_entry >= E820MAX)
601 break; /* no more space left for new bios entries */
602 }
603 if (current_type != 0) {
604 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
605 new_bios[new_bios_entry].type = current_type;
606 last_addr=change_point[chgidx]->addr;
607 }
608 last_type = current_type;
609 }
610 }
611 new_nr = new_bios_entry; /* retain count for new bios entries */
612
613 /* copy new bios mapping into original location */
614 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
615 *pnr_map = new_nr;
616
617 return 0;
618}
619
620/*
621 * Copy the BIOS e820 map into a safe place.
622 *
623 * Sanity-check it while we're at it..
624 *
625 * If we're lucky and live on a modern system, the setup code
626 * will have given us a memory map that we can use to properly
627 * set up memory. If we aren't, we'll fake a memory map.
628 *
629 * We check to see that the memory map contains at least 2 elements
630 * before we'll use it, because the detection code in setup.S may
631 * not be perfect and most every PC known to man has two memory
632 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
633 * thinkpad 560x, for example, does not cooperate with the memory
634 * detection code.)
635 */
636static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
637{
638 /* Only one memory region (or negative)? Ignore it */
639 if (nr_map < 2)
640 return -1;
641
642 do {
643 unsigned long long start = biosmap->addr;
644 unsigned long long size = biosmap->size;
645 unsigned long long end = start + size;
646 unsigned long type = biosmap->type;
647
648 /* Overflow in 64 bits? Ignore the memory map. */
649 if (start > end)
650 return -1;
651
652 /*
653 * Some BIOSes claim RAM in the 640k - 1M region.
654 * Not right. Fix it up.
655 */
656 if (type == E820_RAM) {
657 if (start < 0x100000ULL && end > 0xA0000ULL) {
658 if (start < 0xA0000ULL)
659 add_memory_region(start, 0xA0000ULL-start, type);
660 if (end <= 0x100000ULL)
661 continue;
662 start = 0x100000ULL;
663 size = end - start;
664 }
665 }
666 add_memory_region(start, size, type);
667 } while (biosmap++,--nr_map);
668 return 0;
669}
670
671#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
672struct edd edd;
673#ifdef CONFIG_EDD_MODULE
674EXPORT_SYMBOL(edd);
675#endif
676/**
677 * copy_edd() - Copy the BIOS EDD information
678 * from boot_params into a safe place.
679 *
680 */
681static inline void copy_edd(void)
682{
683 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
684 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
685 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
686 edd.edd_info_nr = EDD_NR;
687}
688#else
689static inline void copy_edd(void)
690{
691}
692#endif
693
694/*
695 * Do NOT EVER look at the BIOS memory size location.
696 * It does not work on many machines.
697 */
698#define LOWMEMSIZE() (0x9f000)
699
700static void __init parse_cmdline_early (char ** cmdline_p)
701{
702 char c = ' ', *to = command_line, *from = saved_command_line;
703 int len = 0;
704 int userdef = 0;
705
706 /* Save unparsed command line copy for /proc/cmdline */
707 saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
708
709 for (;;) {
710 if (c != ' ')
711 goto next_char;
712 /*
713 * "mem=nopentium" disables the 4MB page tables.
714 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
715 * to <mem>, overriding the bios size.
716 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
717 * <start> to <start>+<mem>, overriding the bios size.
718 *
719 * HPA tells me bootloaders need to parse mem=, so no new
720 * option should be mem= [also see Documentation/i386/boot.txt]
721 */
722 if (!memcmp(from, "mem=", 4)) {
723 if (to != command_line)
724 to--;
725 if (!memcmp(from+4, "nopentium", 9)) {
726 from += 9+4;
727 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
728 disable_pse = 1;
729 } else {
730 /* If the user specifies memory size, we
731 * limit the BIOS-provided memory map to
732 * that size. exactmap can be used to specify
733 * the exact map. mem=number can be used to
734 * trim the existing memory map.
735 */
736 unsigned long long mem_size;
737
738 mem_size = memparse(from+4, &from);
739 limit_regions(mem_size);
740 userdef=1;
741 }
742 }
743
744 else if (!memcmp(from, "memmap=", 7)) {
745 if (to != command_line)
746 to--;
747 if (!memcmp(from+7, "exactmap", 8)) {
Vivek Goyal92aa63a2005-06-25 14:58:18 -0700748#ifdef CONFIG_CRASH_DUMP
749 /* If we are doing a crash dump, we
750 * still need to know the real mem
751 * size before original memory map is
752 * reset.
753 */
754 find_max_pfn();
755 saved_max_pfn = max_pfn;
756#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 from += 8+7;
758 e820.nr_map = 0;
759 userdef = 1;
760 } else {
761 /* If the user specifies memory size, we
762 * limit the BIOS-provided memory map to
763 * that size. exactmap can be used to specify
764 * the exact map. mem=number can be used to
765 * trim the existing memory map.
766 */
767 unsigned long long start_at, mem_size;
768
769 mem_size = memparse(from+7, &from);
770 if (*from == '@') {
771 start_at = memparse(from+1, &from);
772 add_memory_region(start_at, mem_size, E820_RAM);
773 } else if (*from == '#') {
774 start_at = memparse(from+1, &from);
775 add_memory_region(start_at, mem_size, E820_ACPI);
776 } else if (*from == '$') {
777 start_at = memparse(from+1, &from);
778 add_memory_region(start_at, mem_size, E820_RESERVED);
779 } else {
780 limit_regions(mem_size);
781 userdef=1;
782 }
783 }
784 }
785
786 else if (!memcmp(from, "noexec=", 7))
787 noexec_setup(from + 7);
788
789
790#ifdef CONFIG_X86_SMP
791 /*
792 * If the BIOS enumerates physical processors before logical,
793 * maxcpus=N at enumeration-time can be used to disable HT.
794 */
795 else if (!memcmp(from, "maxcpus=", 8)) {
796 extern unsigned int maxcpus;
797
798 maxcpus = simple_strtoul(from + 8, NULL, 0);
799 }
800#endif
801
Len Brown888ba6c2005-08-24 12:07:20 -0400802#ifdef CONFIG_ACPI
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 /* "acpi=off" disables both ACPI table parsing and interpreter */
804 else if (!memcmp(from, "acpi=off", 8)) {
805 disable_acpi();
806 }
807
808 /* acpi=force to over-ride black-list */
809 else if (!memcmp(from, "acpi=force", 10)) {
810 acpi_force = 1;
811 acpi_ht = 1;
812 acpi_disabled = 0;
813 }
814
815 /* acpi=strict disables out-of-spec workarounds */
816 else if (!memcmp(from, "acpi=strict", 11)) {
817 acpi_strict = 1;
818 }
819
820 /* Limit ACPI just to boot-time to enable HT */
821 else if (!memcmp(from, "acpi=ht", 7)) {
822 if (!acpi_force)
823 disable_acpi();
824 acpi_ht = 1;
825 }
826
827 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
828 else if (!memcmp(from, "pci=noacpi", 10)) {
829 acpi_disable_pci();
830 }
831 /* "acpi=noirq" disables ACPI interrupt routing */
832 else if (!memcmp(from, "acpi=noirq", 10)) {
833 acpi_noirq_set();
834 }
835
836 else if (!memcmp(from, "acpi_sci=edge", 13))
837 acpi_sci_flags.trigger = 1;
838
839 else if (!memcmp(from, "acpi_sci=level", 14))
840 acpi_sci_flags.trigger = 3;
841
842 else if (!memcmp(from, "acpi_sci=high", 13))
843 acpi_sci_flags.polarity = 1;
844
845 else if (!memcmp(from, "acpi_sci=low", 12))
846 acpi_sci_flags.polarity = 3;
847
848#ifdef CONFIG_X86_IO_APIC
849 else if (!memcmp(from, "acpi_skip_timer_override", 24))
850 acpi_skip_timer_override = 1;
851#endif
852
853#ifdef CONFIG_X86_LOCAL_APIC
Chuck Ebbert66759a02005-09-12 18:49:25 +0200854 if (!memcmp(from, "disable_timer_pin_1", 19))
855 disable_timer_pin_1 = 1;
856 if (!memcmp(from, "enable_timer_pin_1", 18))
857 disable_timer_pin_1 = -1;
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 /* disable IO-APIC */
860 else if (!memcmp(from, "noapic", 6))
861 disable_ioapic_setup();
862#endif /* CONFIG_X86_LOCAL_APIC */
Len Brown888ba6c2005-08-24 12:07:20 -0400863#endif /* CONFIG_ACPI */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Eric W. Biederman9635b472005-06-25 14:57:41 -0700865#ifdef CONFIG_X86_LOCAL_APIC
866 /* enable local APIC */
867 else if (!memcmp(from, "lapic", 5))
868 lapic_enable();
869
870 /* disable local APIC */
871 else if (!memcmp(from, "nolapic", 6))
872 lapic_disable();
873#endif /* CONFIG_X86_LOCAL_APIC */
874
Eric W. Biederman1bc3b912005-06-25 14:58:01 -0700875#ifdef CONFIG_KEXEC
876 /* crashkernel=size@addr specifies the location to reserve for
877 * a crash kernel. By reserving this memory we guarantee
878 * that linux never set's it up as a DMA target.
879 * Useful for holding code to do something appropriate
880 * after a kernel panic.
881 */
882 else if (!memcmp(from, "crashkernel=", 12)) {
883 unsigned long size, base;
884 size = memparse(from+12, &from);
885 if (*from == '@') {
886 base = memparse(from+1, &from);
887 /* FIXME: Do I want a sanity check
888 * to validate the memory range?
889 */
890 crashk_res.start = base;
891 crashk_res.end = base + size - 1;
892 }
893 }
894#endif
Vivek Goyal2030eae2005-06-25 14:58:20 -0700895#ifdef CONFIG_CRASH_DUMP
896 /* elfcorehdr= specifies the location of elf core header
897 * stored by the crashed kernel.
898 */
899 else if (!memcmp(from, "elfcorehdr=", 11))
900 elfcorehdr_addr = memparse(from+11, &from);
901#endif
Eric W. Biederman1bc3b912005-06-25 14:58:01 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 /*
904 * highmem=size forces highmem to be exactly 'size' bytes.
905 * This works even on boxes that have no highmem otherwise.
906 * This also works to reduce highmem size on bigger boxes.
907 */
908 else if (!memcmp(from, "highmem=", 8))
909 highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
910
911 /*
912 * vmalloc=size forces the vmalloc area to be exactly 'size'
913 * bytes. This can be used to increase (or decrease) the
914 * vmalloc area - the default is 128m.
915 */
916 else if (!memcmp(from, "vmalloc=", 8))
917 __VMALLOC_RESERVE = memparse(from+8, &from);
918
919 next_char:
920 c = *(from++);
921 if (!c)
922 break;
923 if (COMMAND_LINE_SIZE <= ++len)
924 break;
925 *(to++) = c;
926 }
927 *to = '\0';
928 *cmdline_p = command_line;
929 if (userdef) {
930 printk(KERN_INFO "user-defined physical RAM map:\n");
931 print_memory_map("user");
932 }
933}
934
935/*
936 * Callback for efi_memory_walk.
937 */
938static int __init
939efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
940{
941 unsigned long *max_pfn = arg, pfn;
942
943 if (start < end) {
944 pfn = PFN_UP(end -1);
945 if (pfn > *max_pfn)
946 *max_pfn = pfn;
947 }
948 return 0;
949}
950
951
952/*
953 * Find the highest page frame number we have available
954 */
955void __init find_max_pfn(void)
956{
957 int i;
958
959 max_pfn = 0;
960 if (efi_enabled) {
961 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
962 return;
963 }
964
965 for (i = 0; i < e820.nr_map; i++) {
966 unsigned long start, end;
967 /* RAM? */
968 if (e820.map[i].type != E820_RAM)
969 continue;
970 start = PFN_UP(e820.map[i].addr);
971 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
972 if (start >= end)
973 continue;
974 if (end > max_pfn)
975 max_pfn = end;
976 }
977}
978
979/*
980 * Determine low and high memory ranges:
981 */
982unsigned long __init find_max_low_pfn(void)
983{
984 unsigned long max_low_pfn;
985
986 max_low_pfn = max_pfn;
987 if (max_low_pfn > MAXMEM_PFN) {
988 if (highmem_pages == -1)
989 highmem_pages = max_pfn - MAXMEM_PFN;
990 if (highmem_pages + MAXMEM_PFN < max_pfn)
991 max_pfn = MAXMEM_PFN + highmem_pages;
992 if (highmem_pages + MAXMEM_PFN > max_pfn) {
993 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
994 highmem_pages = 0;
995 }
996 max_low_pfn = MAXMEM_PFN;
997#ifndef CONFIG_HIGHMEM
998 /* Maximum memory usable is what is directly addressable */
999 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
1000 MAXMEM>>20);
1001 if (max_pfn > MAX_NONPAE_PFN)
1002 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1003 else
1004 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
1005 max_pfn = MAXMEM_PFN;
1006#else /* !CONFIG_HIGHMEM */
1007#ifndef CONFIG_X86_PAE
1008 if (max_pfn > MAX_NONPAE_PFN) {
1009 max_pfn = MAX_NONPAE_PFN;
1010 printk(KERN_WARNING "Warning only 4GB will be used.\n");
1011 printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1012 }
1013#endif /* !CONFIG_X86_PAE */
1014#endif /* !CONFIG_HIGHMEM */
1015 } else {
1016 if (highmem_pages == -1)
1017 highmem_pages = 0;
1018#ifdef CONFIG_HIGHMEM
1019 if (highmem_pages >= max_pfn) {
1020 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1021 highmem_pages = 0;
1022 }
1023 if (highmem_pages) {
1024 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1025 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1026 highmem_pages = 0;
1027 }
1028 max_low_pfn -= highmem_pages;
1029 }
1030#else
1031 if (highmem_pages)
1032 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1033#endif
1034 }
1035 return max_low_pfn;
1036}
1037
1038/*
1039 * Free all available memory for boot time allocation. Used
1040 * as a callback function by efi_memory_walk()
1041 */
1042
1043static int __init
1044free_available_memory(unsigned long start, unsigned long end, void *arg)
1045{
1046 /* check max_low_pfn */
1047 if (start >= ((max_low_pfn + 1) << PAGE_SHIFT))
1048 return 0;
1049 if (end >= ((max_low_pfn + 1) << PAGE_SHIFT))
1050 end = (max_low_pfn + 1) << PAGE_SHIFT;
1051 if (start < end)
1052 free_bootmem(start, end - start);
1053
1054 return 0;
1055}
1056/*
1057 * Register fully available low RAM pages with the bootmem allocator.
1058 */
1059static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1060{
1061 int i;
1062
1063 if (efi_enabled) {
1064 efi_memmap_walk(free_available_memory, NULL);
1065 return;
1066 }
1067 for (i = 0; i < e820.nr_map; i++) {
1068 unsigned long curr_pfn, last_pfn, size;
1069 /*
1070 * Reserve usable low memory
1071 */
1072 if (e820.map[i].type != E820_RAM)
1073 continue;
1074 /*
1075 * We are rounding up the start address of usable memory:
1076 */
1077 curr_pfn = PFN_UP(e820.map[i].addr);
1078 if (curr_pfn >= max_low_pfn)
1079 continue;
1080 /*
1081 * ... and at the end of the usable range downwards:
1082 */
1083 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1084
1085 if (last_pfn > max_low_pfn)
1086 last_pfn = max_low_pfn;
1087
1088 /*
1089 * .. finally, did all the rounding and playing
1090 * around just make the area go away?
1091 */
1092 if (last_pfn <= curr_pfn)
1093 continue;
1094
1095 size = last_pfn - curr_pfn;
1096 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1097 }
1098}
1099
1100/*
1101 * workaround for Dell systems that neglect to reserve EBDA
1102 */
1103static void __init reserve_ebda_region(void)
1104{
1105 unsigned int addr;
1106 addr = get_bios_ebda();
1107 if (addr)
1108 reserve_bootmem(addr, PAGE_SIZE);
1109}
1110
Andy Whitcroft05b79bd2005-06-23 00:07:57 -07001111#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112void __init setup_bootmem_allocator(void);
1113static unsigned long __init setup_memory(void)
1114{
1115 /*
1116 * partially used pages are not usable - thus
1117 * we are rounding upwards:
1118 */
1119 min_low_pfn = PFN_UP(init_pg_tables_end);
1120
1121 find_max_pfn();
1122
1123 max_low_pfn = find_max_low_pfn();
1124
1125#ifdef CONFIG_HIGHMEM
1126 highstart_pfn = highend_pfn = max_pfn;
1127 if (max_pfn > max_low_pfn) {
1128 highstart_pfn = max_low_pfn;
1129 }
1130 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1131 pages_to_mb(highend_pfn - highstart_pfn));
1132#endif
1133 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1134 pages_to_mb(max_low_pfn));
1135
1136 setup_bootmem_allocator();
1137
1138 return max_low_pfn;
1139}
1140
1141void __init zone_sizes_init(void)
1142{
1143 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1144 unsigned int max_dma, low;
1145
1146 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1147 low = max_low_pfn;
1148
1149 if (low < max_dma)
1150 zones_size[ZONE_DMA] = low;
1151 else {
1152 zones_size[ZONE_DMA] = max_dma;
1153 zones_size[ZONE_NORMAL] = low - max_dma;
1154#ifdef CONFIG_HIGHMEM
1155 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1156#endif
1157 }
1158 free_area_init(zones_size);
1159}
1160#else
Andy Whitcroft05b79bd2005-06-23 00:07:57 -07001161extern unsigned long __init setup_memory(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162extern void zone_sizes_init(void);
Andy Whitcroft05b79bd2005-06-23 00:07:57 -07001163#endif /* !CONFIG_NEED_MULTIPLE_NODES */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165void __init setup_bootmem_allocator(void)
1166{
1167 unsigned long bootmap_size;
1168 /*
1169 * Initialize the boot-time allocator (with low memory only):
1170 */
1171 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1172
1173 register_bootmem_low_pages(max_low_pfn);
1174
1175 /*
1176 * Reserve the bootmem bitmap itself as well. We do this in two
1177 * steps (first step was init_bootmem()) because this catches
1178 * the (very unlikely) case of us accidentally initializing the
1179 * bootmem allocator with an invalid RAM area.
1180 */
Vivek Goyal8a919082005-06-25 14:57:51 -07001181 reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1182 bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 /*
1185 * reserve physical page 0 - it's a special BIOS page on many boxes,
1186 * enabling clean reboots, SMP operation, laptop functions.
1187 */
1188 reserve_bootmem(0, PAGE_SIZE);
1189
1190 /* reserve EBDA region, it's a 4K region */
1191 reserve_ebda_region();
1192
1193 /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent
1194 PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1195 unless you have no PS/2 mouse plugged in. */
1196 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1197 boot_cpu_data.x86 == 6)
1198 reserve_bootmem(0xa0000 - 4096, 4096);
1199
1200#ifdef CONFIG_SMP
1201 /*
1202 * But first pinch a few for the stack/trampoline stuff
1203 * FIXME: Don't need the extra page at 4K, but need to fix
1204 * trampoline before removing it. (see the GDT stuff)
1205 */
1206 reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1207#endif
1208#ifdef CONFIG_ACPI_SLEEP
1209 /*
1210 * Reserve low memory region for sleep support.
1211 */
1212 acpi_reserve_bootmem();
1213#endif
1214#ifdef CONFIG_X86_FIND_SMP_CONFIG
1215 /*
1216 * Find and reserve possible boot-time SMP configuration:
1217 */
1218 find_smp_config();
1219#endif
1220
1221#ifdef CONFIG_BLK_DEV_INITRD
1222 if (LOADER_TYPE && INITRD_START) {
1223 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1224 reserve_bootmem(INITRD_START, INITRD_SIZE);
1225 initrd_start =
1226 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1227 initrd_end = initrd_start+INITRD_SIZE;
1228 }
1229 else {
1230 printk(KERN_ERR "initrd extends beyond end of memory "
1231 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1232 INITRD_START + INITRD_SIZE,
1233 max_low_pfn << PAGE_SHIFT);
1234 initrd_start = 0;
1235 }
1236 }
1237#endif
Eric W. Biederman1bc3b912005-06-25 14:58:01 -07001238#ifdef CONFIG_KEXEC
1239 if (crashk_res.start != crashk_res.end)
1240 reserve_bootmem(crashk_res.start,
1241 crashk_res.end - crashk_res.start + 1);
1242#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243}
1244
1245/*
1246 * The node 0 pgdat is initialized before all of these because
1247 * it's needed for bootmem. node>0 pgdats have their virtual
1248 * space allocated before the pagetables are in place to access
1249 * them, so they can't be cleared then.
1250 *
1251 * This should all compile down to nothing when NUMA is off.
1252 */
1253void __init remapped_pgdat_init(void)
1254{
1255 int nid;
1256
1257 for_each_online_node(nid) {
1258 if (nid != 0)
1259 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1260 }
1261}
1262
1263/*
1264 * Request address space for all standard RAM and ROM resources
1265 * and also for regions reported as reserved by the e820.
1266 */
1267static void __init
1268legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1269{
1270 int i;
1271
1272 probe_roms();
1273 for (i = 0; i < e820.nr_map; i++) {
1274 struct resource *res;
1275 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1276 continue;
1277 res = alloc_bootmem_low(sizeof(struct resource));
1278 switch (e820.map[i].type) {
1279 case E820_RAM: res->name = "System RAM"; break;
1280 case E820_ACPI: res->name = "ACPI Tables"; break;
1281 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1282 default: res->name = "reserved";
1283 }
1284 res->start = e820.map[i].addr;
1285 res->end = res->start + e820.map[i].size - 1;
1286 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1287 request_resource(&iomem_resource, res);
1288 if (e820.map[i].type == E820_RAM) {
1289 /*
1290 * We don't know which RAM region contains kernel data,
1291 * so we try it repeatedly and let the resource manager
1292 * test it.
1293 */
1294 request_resource(res, code_resource);
1295 request_resource(res, data_resource);
Eric W. Biederman1bc3b912005-06-25 14:58:01 -07001296#ifdef CONFIG_KEXEC
1297 request_resource(res, &crashk_res);
1298#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 }
1300 }
1301}
1302
1303/*
1304 * Request address space for all standard resources
1305 */
1306static void __init register_memory(void)
1307{
Daniel Ritzf0eca962005-09-09 00:57:14 +02001308 unsigned long gapstart, gapsize, round;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 unsigned long long last;
1310 int i;
1311
1312 if (efi_enabled)
1313 efi_initialize_iomem_resources(&code_resource, &data_resource);
1314 else
1315 legacy_init_iomem_resources(&code_resource, &data_resource);
1316
1317 /* EFI systems may still have VGA */
1318 request_resource(&iomem_resource, &video_ram_resource);
1319
1320 /* request I/O space for devices used on all i[345]86 PCs */
1321 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1322 request_resource(&ioport_resource, &standard_io_resources[i]);
1323
1324 /*
1325 * Search for the bigest gap in the low 32 bits of the e820
1326 * memory space.
1327 */
1328 last = 0x100000000ull;
1329 gapstart = 0x10000000;
1330 gapsize = 0x400000;
1331 i = e820.nr_map;
1332 while (--i >= 0) {
1333 unsigned long long start = e820.map[i].addr;
1334 unsigned long long end = start + e820.map[i].size;
1335
1336 /*
1337 * Since "last" is at most 4GB, we know we'll
1338 * fit in 32 bits if this condition is true
1339 */
1340 if (last > end) {
1341 unsigned long gap = last - end;
1342
1343 if (gap > gapsize) {
1344 gapsize = gap;
1345 gapstart = end;
1346 }
1347 }
1348 if (start < last)
1349 last = start;
1350 }
1351
1352 /*
Daniel Ritzf0eca962005-09-09 00:57:14 +02001353 * See how much we want to round up: start off with
1354 * rounding to the next 1MB area.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 */
Daniel Ritzf0eca962005-09-09 00:57:14 +02001356 round = 0x100000;
1357 while ((gapsize >> 4) > round)
1358 round += round;
1359 /* Fun with two's complement */
1360 pci_mem_start = (gapstart + round) & -round;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1363 pci_mem_start, gapstart, gapsize);
1364}
1365
1366/* Use inline assembly to define this because the nops are defined
1367 as inline assembly strings in the include files and we cannot
1368 get them easily into strings. */
1369asm("\t.data\nintelnops: "
1370 GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
1371 GENERIC_NOP7 GENERIC_NOP8);
1372asm("\t.data\nk8nops: "
1373 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
1374 K8_NOP7 K8_NOP8);
1375asm("\t.data\nk7nops: "
1376 K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
1377 K7_NOP7 K7_NOP8);
1378
1379extern unsigned char intelnops[], k8nops[], k7nops[];
1380static unsigned char *intel_nops[ASM_NOP_MAX+1] = {
1381 NULL,
1382 intelnops,
1383 intelnops + 1,
1384 intelnops + 1 + 2,
1385 intelnops + 1 + 2 + 3,
1386 intelnops + 1 + 2 + 3 + 4,
1387 intelnops + 1 + 2 + 3 + 4 + 5,
1388 intelnops + 1 + 2 + 3 + 4 + 5 + 6,
1389 intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1390};
1391static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
1392 NULL,
1393 k8nops,
1394 k8nops + 1,
1395 k8nops + 1 + 2,
1396 k8nops + 1 + 2 + 3,
1397 k8nops + 1 + 2 + 3 + 4,
1398 k8nops + 1 + 2 + 3 + 4 + 5,
1399 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
1400 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1401};
1402static unsigned char *k7_nops[ASM_NOP_MAX+1] = {
1403 NULL,
1404 k7nops,
1405 k7nops + 1,
1406 k7nops + 1 + 2,
1407 k7nops + 1 + 2 + 3,
1408 k7nops + 1 + 2 + 3 + 4,
1409 k7nops + 1 + 2 + 3 + 4 + 5,
1410 k7nops + 1 + 2 + 3 + 4 + 5 + 6,
1411 k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
1412};
1413static struct nop {
1414 int cpuid;
1415 unsigned char **noptable;
1416} noptypes[] = {
1417 { X86_FEATURE_K8, k8_nops },
1418 { X86_FEATURE_K7, k7_nops },
1419 { -1, NULL }
1420};
1421
1422/* Replace instructions with better alternatives for this CPU type.
1423
1424 This runs before SMP is initialized to avoid SMP problems with
1425 self modifying code. This implies that assymetric systems where
1426 APs have less capabilities than the boot processor are not handled.
Linus Torvalds72538d82005-07-22 18:29:40 -04001427 Tough. Make sure you disable such features by hand. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428void apply_alternatives(void *start, void *end)
1429{
1430 struct alt_instr *a;
1431 int diff, i, k;
1432 unsigned char **noptable = intel_nops;
1433 for (i = 0; noptypes[i].cpuid >= 0; i++) {
1434 if (boot_cpu_has(noptypes[i].cpuid)) {
1435 noptable = noptypes[i].noptable;
1436 break;
1437 }
1438 }
1439 for (a = start; (void *)a < end; a++) {
1440 if (!boot_cpu_has(a->cpuid))
1441 continue;
1442 BUG_ON(a->replacementlen > a->instrlen);
1443 memcpy(a->instr, a->replacement, a->replacementlen);
1444 diff = a->instrlen - a->replacementlen;
1445 /* Pad the rest with nops */
1446 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
1447 k = diff;
1448 if (k > ASM_NOP_MAX)
1449 k = ASM_NOP_MAX;
1450 memcpy(a->instr + i, noptable[k], k);
1451 }
1452 }
1453}
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455void __init alternative_instructions(void)
1456{
1457 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 apply_alternatives(__alt_instructions, __alt_instructions_end);
1459}
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461static char * __init machine_specific_memory_setup(void);
1462
1463#ifdef CONFIG_MCA
1464static void set_mca_bus(int x)
1465{
1466 MCA_bus = x;
1467}
1468#else
1469static void set_mca_bus(int x) { }
1470#endif
1471
1472/*
1473 * Determine if we were loaded by an EFI loader. If so, then we have also been
1474 * passed the efi memmap, systab, etc., so we should use these data structures
1475 * for initialization. Note, the efi init code path is determined by the
1476 * global efi_enabled. This allows the same kernel image to be used on existing
1477 * systems (with a traditional BIOS) as well as on EFI systems.
1478 */
1479void __init setup_arch(char **cmdline_p)
1480{
1481 unsigned long max_low_pfn;
1482
1483 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1484 pre_setup_arch_hook();
1485 early_cpu_init();
1486
1487 /*
1488 * FIXME: This isn't an official loader_type right
1489 * now but does currently work with elilo.
1490 * If we were configured as an EFI kernel, check to make
1491 * sure that we were loaded correctly from elilo and that
1492 * the system table is valid. If not, then initialize normally.
1493 */
1494#ifdef CONFIG_EFI
1495 if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1496 efi_enabled = 1;
1497#endif
1498
1499 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1500 drive_info = DRIVE_INFO;
1501 screen_info = SCREEN_INFO;
1502 edid_info = EDID_INFO;
1503 apm_info.bios = APM_BIOS_INFO;
1504 ist_info = IST_INFO;
1505 saved_videomode = VIDEO_MODE;
1506 if( SYS_DESC_TABLE.length != 0 ) {
1507 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1508 machine_id = SYS_DESC_TABLE.table[0];
1509 machine_submodel_id = SYS_DESC_TABLE.table[1];
1510 BIOS_revision = SYS_DESC_TABLE.table[2];
1511 }
1512 bootloader_type = LOADER_TYPE;
1513
1514#ifdef CONFIG_BLK_DEV_RAM
1515 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1516 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1517 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1518#endif
1519 ARCH_SETUP
1520 if (efi_enabled)
1521 efi_init();
1522 else {
1523 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1524 print_memory_map(machine_specific_memory_setup());
1525 }
1526
1527 copy_edd();
1528
1529 if (!MOUNT_ROOT_RDONLY)
1530 root_mountflags &= ~MS_RDONLY;
1531 init_mm.start_code = (unsigned long) _text;
1532 init_mm.end_code = (unsigned long) _etext;
1533 init_mm.end_data = (unsigned long) _edata;
1534 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1535
1536 code_resource.start = virt_to_phys(_text);
1537 code_resource.end = virt_to_phys(_etext)-1;
1538 data_resource.start = virt_to_phys(_etext);
1539 data_resource.end = virt_to_phys(_edata)-1;
1540
1541 parse_cmdline_early(cmdline_p);
1542
1543 max_low_pfn = setup_memory();
1544
1545 /*
1546 * NOTE: before this point _nobody_ is allowed to allocate
1547 * any memory using the bootmem allocator. Although the
1548 * alloctor is now initialised only the first 8Mb of the kernel
1549 * virtual address space has been mapped. All allocations before
1550 * paging_init() has completed must use the alloc_bootmem_low_pages()
1551 * variant (which allocates DMA'able memory) and care must be taken
1552 * not to exceed the 8Mb limit.
1553 */
1554
1555#ifdef CONFIG_SMP
1556 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1557#endif
1558 paging_init();
1559 remapped_pgdat_init();
Andy Whitcroft05b79bd2005-06-23 00:07:57 -07001560 sparse_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 zone_sizes_init();
1562
1563 /*
1564 * NOTE: at this point the bootmem allocator is fully available.
1565 */
1566
1567#ifdef CONFIG_EARLY_PRINTK
1568 {
1569 char *s = strstr(*cmdline_p, "earlyprintk=");
1570 if (s) {
1571 extern void setup_early_printk(char *);
1572
1573 setup_early_printk(s);
1574 printk("early console enabled\n");
1575 }
1576 }
1577#endif
1578
1579
1580 dmi_scan_machine();
1581
1582#ifdef CONFIG_X86_GENERICARCH
1583 generic_apic_probe(*cmdline_p);
1584#endif
1585 if (efi_enabled)
1586 efi_map_memmap();
1587
Len Brown888ba6c2005-08-24 12:07:20 -04001588#ifdef CONFIG_ACPI
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 /*
1590 * Parse the ACPI tables for possible boot-time SMP configuration.
1591 */
1592 acpi_boot_table_init();
1593 acpi_boot_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Venkatesh Pallipadi911a62d2005-09-03 15:56:31 -07001595#if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1596 if (def_to_bigsmp)
1597 printk(KERN_WARNING "More than 8 CPUs detected and "
1598 "CONFIG_X86_PC cannot handle it.\nUse "
1599 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1600#endif
1601#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602#ifdef CONFIG_X86_LOCAL_APIC
1603 if (smp_found_config)
1604 get_smp_config();
1605#endif
1606
1607 register_memory();
1608
1609#ifdef CONFIG_VT
1610#if defined(CONFIG_VGA_CONSOLE)
1611 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1612 conswitchp = &vga_con;
1613#elif defined(CONFIG_DUMMY_CONSOLE)
1614 conswitchp = &dummy_con;
1615#endif
1616#endif
1617}
1618
1619#include "setup_arch_post.h"
1620/*
1621 * Local Variables:
1622 * mode:c
1623 * c-file-style:"k&r"
1624 * c-basic-offset:8
1625 * End:
1626 */